linux/fs/smb/client/dfs_cache.c

1297 lines
31 KiB
C
Raw Normal View History

cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
// SPDX-License-Identifier: GPL-2.0
/*
* DFS referral cache routines
*
* Copyright (c) 2018-2019 Paulo Alcantara <palcantara@suse.de>
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
*/
#include <linux/jhash.h>
#include <linux/ktime.h>
#include <linux/slab.h>
#include <linux/proc_fs.h>
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
#include <linux/nls.h>
#include <linux/workqueue.h>
#include <linux/uuid.h>
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
#include "cifsglob.h"
#include "smb2pdu.h"
#include "smb2proto.h"
#include "cifsproto.h"
#include "cifs_debug.h"
#include "cifs_unicode.h"
#include "smb2glob.h"
#include "dns_resolve.h"
#include "dfs.h"
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
#include "dfs_cache.h"
#define CACHE_HTABLE_SIZE 32
#define CACHE_MAX_ENTRIES 64
#define CACHE_MIN_TTL 120 /* 2 minutes */
#define CACHE_DEFAULT_TTL 300 /* 5 minutes */
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
struct cache_dfs_tgt {
char *name;
int path_consumed;
struct list_head list;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
};
struct cache_entry {
struct hlist_node hlist;
const char *path;
int hdr_flags; /* RESP_GET_DFS_REFERRAL.ReferralHeaderFlags */
int ttl; /* DFS_REREFERRAL_V3.TimeToLive */
int srvtype; /* DFS_REREFERRAL_V3.ServerType */
int ref_flags; /* DFS_REREFERRAL_V3.ReferralEntryFlags */
struct timespec64 etime;
int path_consumed; /* RESP_GET_DFS_REFERRAL.PathConsumed */
int numtgts;
struct list_head tlist;
struct cache_dfs_tgt *tgthint;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
};
static struct kmem_cache *cache_slab __read_mostly;
struct workqueue_struct *dfscache_wq;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
atomic_t dfs_cache_ttl;
static struct nls_table *cache_cp;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
/*
* Number of entries in the cache
*/
static atomic_t cache_count;
static struct hlist_head cache_htable[CACHE_HTABLE_SIZE];
static DECLARE_RWSEM(htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
/**
* dfs_cache_canonical_path - get a canonical DFS path
*
* @path: DFS path
* @cp: codepage
* @remap: mapping type
*
* Return canonical path if success, otherwise error.
*/
char *dfs_cache_canonical_path(const char *path, const struct nls_table *cp, int remap)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
char *tmp;
int plen = 0;
char *npath;
if (!path || strlen(path) < 3 || (*path != '\\' && *path != '/'))
return ERR_PTR(-EINVAL);
if (unlikely(strcmp(cp->charset, cache_cp->charset))) {
tmp = (char *)cifs_strndup_to_utf16(path, strlen(path), &plen, cp, remap);
if (!tmp) {
cifs_dbg(VFS, "%s: failed to convert path to utf16\n", __func__);
return ERR_PTR(-EINVAL);
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
npath = cifs_strndup_from_utf16(tmp, plen, true, cache_cp);
kfree(tmp);
if (!npath) {
cifs_dbg(VFS, "%s: failed to convert path from utf16\n", __func__);
return ERR_PTR(-EINVAL);
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
} else {
npath = kstrdup(path, GFP_KERNEL);
if (!npath)
return ERR_PTR(-ENOMEM);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
convert_delimiter(npath, '\\');
return npath;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
static inline bool cache_entry_expired(const struct cache_entry *ce)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct timespec64 ts;
cifs: update for current_kernel_time64() removal Fixes cifs build failure after merge of the y2038 tree After merging the y2038 tree, today's linux-next build (x86_64 allmodconfig) failed like this: fs/cifs/dfs_cache.c: In function 'cache_entry_expired': fs/cifs/dfs_cache.c:106:7: error: implicit declaration of function 'current_kernel_time64'; did you mean 'core_kernel_text'? [-Werror=implicit-function-declaration] ts = current_kernel_time64(); ^~~~~~~~~~~~~~~~~~~~~ core_kernel_text fs/cifs/dfs_cache.c:106:5: error: incompatible types when assigning to type 'struct timespec64' from type 'int' ts = current_kernel_time64(); ^ fs/cifs/dfs_cache.c: In function 'get_expire_time': fs/cifs/dfs_cache.c:342:24: error: incompatible type for argument 1 of 'timespec64_add' return timespec64_add(current_kernel_time64(), ts); ^~~~~~~~~~~~~~~~~~~~~~~ In file included from include/linux/restart_block.h:10, from include/linux/thread_info.h:13, from arch/x86/include/asm/preempt.h:7, from include/linux/preempt.h:78, from include/linux/rcupdate.h:40, from fs/cifs/dfs_cache.c:8: include/linux/time64.h:66:66: note: expected 'struct timespec64' but argument is of type 'int' static inline struct timespec64 timespec64_add(struct timespec64 lhs, ~~~~~~~~~~~~~~~~~~^~~ fs/cifs/dfs_cache.c:343:1: warning: control reaches end of non-void function [-Wreturn-type] } ^ Caused by: commit ccea641b6742 ("timekeeping: remove obsolete time accessors") interacting with: commit 34a44fb160f9 ("cifs: Add DFS cache routines") from the cifs tree. Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Reviewed-by: Paulo Alcantara <palcantara@suse.de> Acked-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-12-17 09:11:46 +00:00
ktime_get_coarse_real_ts64(&ts);
return timespec64_compare(&ts, &ce->etime) >= 0;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
static inline void free_tgts(struct cache_entry *ce)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct cache_dfs_tgt *t, *n;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
list_for_each_entry_safe(t, n, &ce->tlist, list) {
list_del(&t->list);
kfree(t->name);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
kfree(t);
}
}
static inline void flush_cache_ent(struct cache_entry *ce)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
hlist_del_init(&ce->hlist);
kfree(ce->path);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
free_tgts(ce);
atomic_dec(&cache_count);
kmem_cache_free(cache_slab, ce);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
static void flush_cache_ents(void)
{
int i;
for (i = 0; i < CACHE_HTABLE_SIZE; i++) {
struct hlist_head *l = &cache_htable[i];
struct hlist_node *n;
struct cache_entry *ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
hlist_for_each_entry_safe(ce, n, l, hlist) {
if (!hlist_unhashed(&ce->hlist))
flush_cache_ent(ce);
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
}
/*
* dfs cache /proc file
*/
static int dfscache_proc_show(struct seq_file *m, void *v)
{
int i;
struct cache_entry *ce;
struct cache_dfs_tgt *t;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
seq_puts(m, "DFS cache\n---------\n");
down_read(&htable_rw_lock);
for (i = 0; i < CACHE_HTABLE_SIZE; i++) {
struct hlist_head *l = &cache_htable[i];
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
hlist_for_each_entry(ce, l, hlist) {
if (hlist_unhashed(&ce->hlist))
continue;
seq_printf(m,
"cache entry: path=%s,type=%s,ttl=%d,etime=%ld,hdr_flags=0x%x,ref_flags=0x%x,interlink=%s,path_consumed=%d,expired=%s\n",
ce->path, ce->srvtype == DFS_TYPE_ROOT ? "root" : "link",
ce->ttl, ce->etime.tv_nsec, ce->hdr_flags, ce->ref_flags,
DFS_INTERLINK(ce->hdr_flags) ? "yes" : "no",
ce->path_consumed, cache_entry_expired(ce) ? "yes" : "no");
list_for_each_entry(t, &ce->tlist, list) {
seq_printf(m, " %s%s\n",
t->name,
READ_ONCE(ce->tgthint) == t ? " (target hint)" : "");
}
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
up_read(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return 0;
}
static ssize_t dfscache_proc_write(struct file *file, const char __user *buffer,
size_t count, loff_t *ppos)
{
char c;
int rc;
rc = get_user(c, buffer);
if (rc)
return rc;
if (c != '0')
return -EINVAL;
cifs_dbg(FYI, "clearing dfs cache\n");
down_write(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
flush_cache_ents();
up_write(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return count;
}
static int dfscache_proc_open(struct inode *inode, struct file *file)
{
return single_open(file, dfscache_proc_show, NULL);
}
const struct proc_ops dfscache_proc_ops = {
.proc_open = dfscache_proc_open,
.proc_read = seq_read,
.proc_lseek = seq_lseek,
.proc_release = single_release,
.proc_write = dfscache_proc_write,
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
};
#ifdef CONFIG_CIFS_DEBUG2
static inline void dump_tgts(const struct cache_entry *ce)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct cache_dfs_tgt *t;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cifs_dbg(FYI, "target list:\n");
list_for_each_entry(t, &ce->tlist, list) {
cifs_dbg(FYI, " %s%s\n", t->name,
READ_ONCE(ce->tgthint) == t ? " (target hint)" : "");
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
}
static inline void dump_ce(const struct cache_entry *ce)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
cifs_dbg(FYI, "cache entry: path=%s,type=%s,ttl=%d,etime=%ld,hdr_flags=0x%x,ref_flags=0x%x,interlink=%s,path_consumed=%d,expired=%s\n",
ce->path,
ce->srvtype == DFS_TYPE_ROOT ? "root" : "link", ce->ttl,
ce->etime.tv_nsec,
ce->hdr_flags, ce->ref_flags,
DFS_INTERLINK(ce->hdr_flags) ? "yes" : "no",
ce->path_consumed,
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cache_entry_expired(ce) ? "yes" : "no");
dump_tgts(ce);
}
static inline void dump_refs(const struct dfs_info3_param *refs, int numrefs)
{
int i;
cifs_dbg(FYI, "DFS referrals returned by the server:\n");
for (i = 0; i < numrefs; i++) {
const struct dfs_info3_param *ref = &refs[i];
cifs_dbg(FYI,
"\n"
"flags: 0x%x\n"
"path_consumed: %d\n"
"server_type: 0x%x\n"
"ref_flag: 0x%x\n"
"path_name: %s\n"
"node_name: %s\n"
"ttl: %d (%dm)\n",
ref->flags, ref->path_consumed, ref->server_type,
ref->ref_flag, ref->path_name, ref->node_name,
ref->ttl, ref->ttl / 60);
}
}
#else
#define dump_tgts(e)
#define dump_ce(e)
#define dump_refs(r, n)
#endif
/**
* dfs_cache_init - Initialize DFS referral cache.
*
* Return zero if initialized successfully, otherwise non-zero.
*/
int dfs_cache_init(void)
{
int rc;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
int i;
dfscache_wq = alloc_workqueue("cifs-dfscache",
WQ_UNBOUND|WQ_FREEZABLE|WQ_MEM_RECLAIM,
0);
if (!dfscache_wq)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return -ENOMEM;
cache_slab = kmem_cache_create("cifs_dfs_cache",
sizeof(struct cache_entry), 0,
SLAB_HWCACHE_ALIGN, NULL);
if (!cache_slab) {
rc = -ENOMEM;
goto out_destroy_wq;
}
for (i = 0; i < CACHE_HTABLE_SIZE; i++)
INIT_HLIST_HEAD(&cache_htable[i]);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
atomic_set(&cache_count, 0);
atomic_set(&dfs_cache_ttl, CACHE_DEFAULT_TTL);
cache_cp = load_nls("utf8");
if (!cache_cp)
cache_cp = load_nls_default();
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cifs_dbg(FYI, "%s: initialized DFS referral cache\n", __func__);
return 0;
out_destroy_wq:
destroy_workqueue(dfscache_wq);
return rc;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
static int cache_entry_hash(const void *data, int size, unsigned int *hash)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
int i, clen;
const unsigned char *s = data;
wchar_t c;
unsigned int h = 0;
for (i = 0; i < size; i += clen) {
clen = cache_cp->char2uni(&s[i], size - i, &c);
if (unlikely(clen < 0)) {
cifs_dbg(VFS, "%s: can't convert char\n", __func__);
return clen;
}
c = cifs_toupper(c);
h = jhash(&c, sizeof(c), h);
}
*hash = h % CACHE_HTABLE_SIZE;
return 0;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
/* Return target hint of a DFS cache entry */
static inline char *get_tgt_name(const struct cache_entry *ce)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct cache_dfs_tgt *t = READ_ONCE(ce->tgthint);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return t ? t->name : ERR_PTR(-ENOENT);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
/* Return expire time out of a new entry's TTL */
static inline struct timespec64 get_expire_time(int ttl)
{
struct timespec64 ts = {
.tv_sec = ttl,
.tv_nsec = 0,
};
cifs: update for current_kernel_time64() removal Fixes cifs build failure after merge of the y2038 tree After merging the y2038 tree, today's linux-next build (x86_64 allmodconfig) failed like this: fs/cifs/dfs_cache.c: In function 'cache_entry_expired': fs/cifs/dfs_cache.c:106:7: error: implicit declaration of function 'current_kernel_time64'; did you mean 'core_kernel_text'? [-Werror=implicit-function-declaration] ts = current_kernel_time64(); ^~~~~~~~~~~~~~~~~~~~~ core_kernel_text fs/cifs/dfs_cache.c:106:5: error: incompatible types when assigning to type 'struct timespec64' from type 'int' ts = current_kernel_time64(); ^ fs/cifs/dfs_cache.c: In function 'get_expire_time': fs/cifs/dfs_cache.c:342:24: error: incompatible type for argument 1 of 'timespec64_add' return timespec64_add(current_kernel_time64(), ts); ^~~~~~~~~~~~~~~~~~~~~~~ In file included from include/linux/restart_block.h:10, from include/linux/thread_info.h:13, from arch/x86/include/asm/preempt.h:7, from include/linux/preempt.h:78, from include/linux/rcupdate.h:40, from fs/cifs/dfs_cache.c:8: include/linux/time64.h:66:66: note: expected 'struct timespec64' but argument is of type 'int' static inline struct timespec64 timespec64_add(struct timespec64 lhs, ~~~~~~~~~~~~~~~~~~^~~ fs/cifs/dfs_cache.c:343:1: warning: control reaches end of non-void function [-Wreturn-type] } ^ Caused by: commit ccea641b6742 ("timekeeping: remove obsolete time accessors") interacting with: commit 34a44fb160f9 ("cifs: Add DFS cache routines") from the cifs tree. Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Reviewed-by: Paulo Alcantara <palcantara@suse.de> Acked-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-12-17 09:11:46 +00:00
struct timespec64 now;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cifs: update for current_kernel_time64() removal Fixes cifs build failure after merge of the y2038 tree After merging the y2038 tree, today's linux-next build (x86_64 allmodconfig) failed like this: fs/cifs/dfs_cache.c: In function 'cache_entry_expired': fs/cifs/dfs_cache.c:106:7: error: implicit declaration of function 'current_kernel_time64'; did you mean 'core_kernel_text'? [-Werror=implicit-function-declaration] ts = current_kernel_time64(); ^~~~~~~~~~~~~~~~~~~~~ core_kernel_text fs/cifs/dfs_cache.c:106:5: error: incompatible types when assigning to type 'struct timespec64' from type 'int' ts = current_kernel_time64(); ^ fs/cifs/dfs_cache.c: In function 'get_expire_time': fs/cifs/dfs_cache.c:342:24: error: incompatible type for argument 1 of 'timespec64_add' return timespec64_add(current_kernel_time64(), ts); ^~~~~~~~~~~~~~~~~~~~~~~ In file included from include/linux/restart_block.h:10, from include/linux/thread_info.h:13, from arch/x86/include/asm/preempt.h:7, from include/linux/preempt.h:78, from include/linux/rcupdate.h:40, from fs/cifs/dfs_cache.c:8: include/linux/time64.h:66:66: note: expected 'struct timespec64' but argument is of type 'int' static inline struct timespec64 timespec64_add(struct timespec64 lhs, ~~~~~~~~~~~~~~~~~~^~~ fs/cifs/dfs_cache.c:343:1: warning: control reaches end of non-void function [-Wreturn-type] } ^ Caused by: commit ccea641b6742 ("timekeeping: remove obsolete time accessors") interacting with: commit 34a44fb160f9 ("cifs: Add DFS cache routines") from the cifs tree. Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Reviewed-by: Paulo Alcantara <palcantara@suse.de> Acked-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-12-17 09:11:46 +00:00
ktime_get_coarse_real_ts64(&now);
return timespec64_add(now, ts);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
/* Allocate a new DFS target */
static struct cache_dfs_tgt *alloc_target(const char *name, int path_consumed)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct cache_dfs_tgt *t;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
t = kmalloc(sizeof(*t), GFP_ATOMIC);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!t)
return ERR_PTR(-ENOMEM);
t->name = kstrdup(name, GFP_ATOMIC);
if (!t->name) {
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
kfree(t);
return ERR_PTR(-ENOMEM);
}
t->path_consumed = path_consumed;
INIT_LIST_HEAD(&t->list);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return t;
}
/*
* Copy DFS referral information to a cache entry and conditionally update
* target hint.
*/
static int copy_ref_data(const struct dfs_info3_param *refs, int numrefs,
struct cache_entry *ce, const char *tgthint)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct cache_dfs_tgt *target;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
int i;
ce->ttl = max_t(int, refs[0].ttl, CACHE_MIN_TTL);
ce->etime = get_expire_time(ce->ttl);
ce->srvtype = refs[0].server_type;
ce->hdr_flags = refs[0].flags;
ce->ref_flags = refs[0].ref_flag;
ce->path_consumed = refs[0].path_consumed;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
for (i = 0; i < numrefs; i++) {
struct cache_dfs_tgt *t;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
t = alloc_target(refs[i].node_name, refs[i].path_consumed);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (IS_ERR(t)) {
free_tgts(ce);
return PTR_ERR(t);
}
if (tgthint && !strcasecmp(t->name, tgthint)) {
list_add(&t->list, &ce->tlist);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
tgthint = NULL;
} else {
list_add_tail(&t->list, &ce->tlist);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
ce->numtgts++;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
target = list_first_entry_or_null(&ce->tlist, struct cache_dfs_tgt,
list);
WRITE_ONCE(ce->tgthint, target);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return 0;
}
/* Allocate a new cache entry */
static struct cache_entry *alloc_cache_entry(struct dfs_info3_param *refs, int numrefs)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct cache_entry *ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
int rc;
ce = kmem_cache_zalloc(cache_slab, GFP_KERNEL);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!ce)
return ERR_PTR(-ENOMEM);
ce->path = refs[0].path_name;
refs[0].path_name = NULL;
INIT_HLIST_NODE(&ce->hlist);
INIT_LIST_HEAD(&ce->tlist);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
rc = copy_ref_data(refs, numrefs, ce, NULL);
if (rc) {
kfree(ce->path);
kmem_cache_free(cache_slab, ce);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
ce = ERR_PTR(rc);
}
return ce;
}
static void remove_oldest_entry_locked(void)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
int i;
struct cache_entry *ce;
struct cache_entry *to_del = NULL;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
WARN_ON(!rwsem_is_locked(&htable_rw_lock));
for (i = 0; i < CACHE_HTABLE_SIZE; i++) {
struct hlist_head *l = &cache_htable[i];
hlist_for_each_entry(ce, l, hlist) {
if (hlist_unhashed(&ce->hlist))
continue;
if (!to_del || timespec64_compare(&ce->etime,
&to_del->etime) < 0)
to_del = ce;
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!to_del) {
cifs_dbg(FYI, "%s: no entry to remove\n", __func__);
return;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
cifs_dbg(FYI, "%s: removing entry\n", __func__);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
dump_ce(to_del);
flush_cache_ent(to_del);
}
/* Add a new DFS cache entry */
static struct cache_entry *add_cache_entry_locked(struct dfs_info3_param *refs,
int numrefs)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
int rc;
struct cache_entry *ce;
unsigned int hash;
int ttl;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
WARN_ON(!rwsem_is_locked(&htable_rw_lock));
if (atomic_read(&cache_count) >= CACHE_MAX_ENTRIES) {
cifs_dbg(FYI, "%s: reached max cache size (%d)\n", __func__, CACHE_MAX_ENTRIES);
remove_oldest_entry_locked();
}
rc = cache_entry_hash(refs[0].path_name, strlen(refs[0].path_name), &hash);
if (rc)
return ERR_PTR(rc);
ce = alloc_cache_entry(refs, numrefs);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (IS_ERR(ce))
return ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
ttl = min_t(int, atomic_read(&dfs_cache_ttl), ce->ttl);
atomic_set(&dfs_cache_ttl, ttl);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
hlist_add_head(&ce->hlist, &cache_htable[hash]);
dump_ce(ce);
atomic_inc(&cache_count);
return ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
/* Check if two DFS paths are equal. @s1 and @s2 are expected to be in @cache_cp's charset */
static bool dfs_path_equal(const char *s1, int len1, const char *s2, int len2)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
int i, l1, l2;
wchar_t c1, c2;
if (len1 != len2)
return false;
for (i = 0; i < len1; i += l1) {
l1 = cache_cp->char2uni(&s1[i], len1 - i, &c1);
l2 = cache_cp->char2uni(&s2[i], len2 - i, &c2);
if (unlikely(l1 < 0 && l2 < 0)) {
if (s1[i] != s2[i])
return false;
l1 = 1;
continue;
}
if (l1 != l2)
return false;
if (cifs_toupper(c1) != cifs_toupper(c2))
return false;
}
return true;
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
static struct cache_entry *__lookup_cache_entry(const char *path, unsigned int hash, int len)
{
struct cache_entry *ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
hlist_for_each_entry(ce, &cache_htable[hash], hlist) {
if (dfs_path_equal(ce->path, strlen(ce->path), path, len)) {
dump_ce(ce);
return ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
}
return ERR_PTR(-ENOENT);
}
/*
* Find a DFS cache entry in hash table and optionally check prefix path against normalized @path.
*
* Use whole path components in the match. Must be called with htable_rw_lock held.
*
* Return cached entry if successful.
* Return ERR_PTR(-ENOENT) if the entry is not found.
* Return error ptr otherwise.
*/
static struct cache_entry *lookup_cache_entry(const char *path)
{
struct cache_entry *ce;
int cnt = 0;
const char *s = path, *e;
char sep = *s;
unsigned int hash;
int rc;
while ((s = strchr(s, sep)) && ++cnt < 3)
s++;
if (cnt < 3) {
rc = cache_entry_hash(path, strlen(path), &hash);
if (rc)
return ERR_PTR(rc);
return __lookup_cache_entry(path, hash, strlen(path));
}
/*
* Handle paths that have more than two path components and are a complete prefix of the DFS
* referral request path (@path).
*
* See MS-DFSC 3.2.5.5 "Receiving a Root Referral Request or Link Referral Request".
*/
e = path + strlen(path) - 1;
while (e > s) {
int len;
/* skip separators */
while (e > s && *e == sep)
e--;
if (e == s)
break;
len = e + 1 - path;
rc = cache_entry_hash(path, len, &hash);
if (rc)
return ERR_PTR(rc);
ce = __lookup_cache_entry(path, hash, len);
if (!IS_ERR(ce))
return ce;
/* backward until separator */
while (e > s && *e != sep)
e--;
}
return ERR_PTR(-ENOENT);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
/**
* dfs_cache_destroy - destroy DFS referral cache
*/
void dfs_cache_destroy(void)
{
unload_nls(cache_cp);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
flush_cache_ents();
kmem_cache_destroy(cache_slab);
destroy_workqueue(dfscache_wq);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cifs_dbg(FYI, "%s: destroyed DFS referral cache\n", __func__);
}
/* Update a cache entry with the new referral in @refs */
static int update_cache_entry_locked(struct cache_entry *ce, const struct dfs_info3_param *refs,
int numrefs)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct cache_dfs_tgt *target;
char *th = NULL;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
int rc;
WARN_ON(!rwsem_is_locked(&htable_rw_lock));
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
target = READ_ONCE(ce->tgthint);
if (target) {
th = kstrdup(target->name, GFP_ATOMIC);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!th)
return -ENOMEM;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
free_tgts(ce);
ce->numtgts = 0;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
rc = copy_ref_data(refs, numrefs, ce, th);
kfree(th);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return rc;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
static int get_dfs_referral(const unsigned int xid, struct cifs_ses *ses, const char *path,
struct dfs_info3_param **refs, int *numrefs)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
int rc;
int i;
*refs = NULL;
*numrefs = 0;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!ses || !ses->server || !ses->server->ops->get_dfs_refer)
return -EOPNOTSUPP;
if (unlikely(!cache_cp))
return -EINVAL;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cifs_dbg(FYI, "%s: ipc=%s referral=%s\n", __func__, ses->tcon_ipc->tree_name, path);
rc = ses->server->ops->get_dfs_refer(xid, ses, path, refs, numrefs, cache_cp,
NO_MAP_UNI_RSVD);
if (!rc) {
struct dfs_info3_param *ref = *refs;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
for (i = 0; i < *numrefs; i++)
convert_delimiter(ref[i].path_name, '\\');
}
return rc;
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
/*
* Find, create or update a DFS cache entry.
*
* If the entry wasn't found, it will create a new one. Or if it was found but
* expired, then it will update the entry accordingly.
*
* For interlinks, cifs_mount() and expand_dfs_referral() are supposed to
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
* handle them properly.
*
* On success, return entry with acquired lock for reading, otherwise error ptr.
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
*/
static struct cache_entry *cache_refresh_path(const unsigned int xid,
struct cifs_ses *ses,
const char *path,
bool force_refresh)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct dfs_info3_param *refs = NULL;
struct cache_entry *ce;
int numrefs = 0;
int rc;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cifs_dbg(FYI, "%s: search path: %s\n", __func__, path);
down_read(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
ce = lookup_cache_entry(path);
if (!IS_ERR(ce)) {
if (!force_refresh && !cache_entry_expired(ce))
return ce;
} else if (PTR_ERR(ce) != -ENOENT) {
up_read(&htable_rw_lock);
return ce;
}
/*
* Unlock shared access as we don't want to hold any locks while getting
* a new referral. The @ses used for performing the I/O could be
* reconnecting and it acquires @htable_rw_lock to look up the dfs cache
* in order to failover -- if necessary.
*/
up_read(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
/*
* Either the entry was not found, or it is expired, or it is a forced
* refresh.
* Request a new DFS referral in order to create or update a cache entry.
*/
rc = get_dfs_referral(xid, ses, path, &refs, &numrefs);
if (rc) {
ce = ERR_PTR(rc);
goto out;
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
dump_refs(refs, numrefs);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
down_write(&htable_rw_lock);
/* Re-check as another task might have it added or refreshed already */
ce = lookup_cache_entry(path);
if (!IS_ERR(ce)) {
if (force_refresh || cache_entry_expired(ce)) {
rc = update_cache_entry_locked(ce, refs, numrefs);
if (rc)
ce = ERR_PTR(rc);
}
} else if (PTR_ERR(ce) == -ENOENT) {
ce = add_cache_entry_locked(refs, numrefs);
}
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (IS_ERR(ce)) {
up_write(&htable_rw_lock);
goto out;
}
downgrade_write(&htable_rw_lock);
out:
free_dfs_info_array(refs, numrefs);
return ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
/*
* Set up a DFS referral from a given cache entry.
*
* Must be called with htable_rw_lock held.
*/
static int setup_referral(const char *path, struct cache_entry *ce,
struct dfs_info3_param *ref, const char *target)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
int rc;
cifs_dbg(FYI, "%s: set up new ref\n", __func__);
memset(ref, 0, sizeof(*ref));
ref->path_name = kstrdup(path, GFP_ATOMIC);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!ref->path_name)
return -ENOMEM;
ref->node_name = kstrdup(target, GFP_ATOMIC);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!ref->node_name) {
rc = -ENOMEM;
goto err_free_path;
}
ref->path_consumed = ce->path_consumed;
ref->ttl = ce->ttl;
ref->server_type = ce->srvtype;
ref->ref_flag = ce->ref_flags;
ref->flags = ce->hdr_flags;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return 0;
err_free_path:
kfree(ref->path_name);
ref->path_name = NULL;
return rc;
}
/* Return target list of a DFS cache entry */
static int get_targets(struct cache_entry *ce, struct dfs_cache_tgt_list *tl)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
int rc;
struct list_head *head = &tl->tl_list;
struct cache_dfs_tgt *t;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
struct dfs_cache_tgt_iterator *it, *nit;
memset(tl, 0, sizeof(*tl));
INIT_LIST_HEAD(head);
list_for_each_entry(t, &ce->tlist, list) {
it = kzalloc(sizeof(*it), GFP_ATOMIC);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!it) {
rc = -ENOMEM;
goto err_free_it;
}
it->it_name = kstrdup(t->name, GFP_ATOMIC);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!it->it_name) {
kfree(it);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
rc = -ENOMEM;
goto err_free_it;
}
it->it_path_consumed = t->path_consumed;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (READ_ONCE(ce->tgthint) == t)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
list_add(&it->it_list, head);
else
list_add_tail(&it->it_list, head);
}
tl->tl_numtgts = ce->numtgts;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return 0;
err_free_it:
list_for_each_entry_safe(it, nit, head, it_list) {
list_del(&it->it_list);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
kfree(it->it_name);
kfree(it);
}
return rc;
}
/**
* dfs_cache_find - find a DFS cache entry
*
* If it doesn't find the cache entry, then it will get a DFS referral
* for @path and create a new entry.
*
* In case the cache entry exists but expired, it will get a DFS referral
* for @path and then update the respective cache entry.
*
* These parameters are passed down to the get_dfs_refer() call if it
* needs to be issued:
* @xid: syscall xid
* @ses: smb session to issue the request on
* @cp: codepage
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
* @remap: path character remapping type
* @path: path to lookup in DFS referral cache.
*
* @ref: when non-NULL, store single DFS referral result in it.
* @tgt_list: when non-NULL, store complete DFS target list in it.
*
* Return zero if the target was found, otherwise non-zero.
*/
int dfs_cache_find(const unsigned int xid, struct cifs_ses *ses, const struct nls_table *cp,
int remap, const char *path, struct dfs_info3_param *ref,
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
struct dfs_cache_tgt_list *tgt_list)
{
int rc;
const char *npath;
struct cache_entry *ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
npath = dfs_cache_canonical_path(path, cp, remap);
if (IS_ERR(npath))
return PTR_ERR(npath);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
ce = cache_refresh_path(xid, ses, npath, false);
if (IS_ERR(ce)) {
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
rc = PTR_ERR(ce);
goto out_free_path;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
if (ref)
rc = setup_referral(path, ce, ref, get_tgt_name(ce));
else
rc = 0;
if (!rc && tgt_list)
rc = get_targets(ce, tgt_list);
up_read(&htable_rw_lock);
out_free_path:
kfree(npath);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return rc;
}
/**
* dfs_cache_noreq_find - find a DFS cache entry without sending any requests to
* the currently connected server.
*
* NOTE: This function will neither update a cache entry in case it was
* expired, nor create a new cache entry if @path hasn't been found. It heavily
* relies on an existing cache entry.
*
* @path: canonical DFS path to lookup in the DFS referral cache.
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
* @ref: when non-NULL, store single DFS referral result in it.
* @tgt_list: when non-NULL, store complete DFS target list in it.
*
* Return 0 if successful.
* Return -ENOENT if the entry was not found.
* Return non-zero for other errors.
*/
int dfs_cache_noreq_find(const char *path, struct dfs_info3_param *ref,
struct dfs_cache_tgt_list *tgt_list)
{
int rc;
struct cache_entry *ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cifs_dbg(FYI, "%s: path: %s\n", __func__, path);
down_read(&htable_rw_lock);
ce = lookup_cache_entry(path);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (IS_ERR(ce)) {
rc = PTR_ERR(ce);
goto out_unlock;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
if (ref)
rc = setup_referral(path, ce, ref, get_tgt_name(ce));
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
else
rc = 0;
if (!rc && tgt_list)
rc = get_targets(ce, tgt_list);
out_unlock:
up_read(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return rc;
}
/**
* dfs_cache_noreq_update_tgthint - update target hint of a DFS cache entry
* without sending any requests to the currently connected server.
*
* NOTE: This function will neither update a cache entry in case it was
* expired, nor create a new cache entry if @path hasn't been found. It heavily
* relies on an existing cache entry.
*
* @path: canonical DFS path to lookup in DFS referral cache.
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
* @it: target iterator which contains the target hint to update the cache
* entry with.
*
* Return zero if the target hint was updated successfully, otherwise non-zero.
*/
void dfs_cache_noreq_update_tgthint(const char *path, const struct dfs_cache_tgt_iterator *it)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct cache_dfs_tgt *t;
struct cache_entry *ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!path || !it)
return;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cifs_dbg(FYI, "%s: path: %s\n", __func__, path);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
down_read(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
ce = lookup_cache_entry(path);
if (IS_ERR(ce))
goto out_unlock;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
t = READ_ONCE(ce->tgthint);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (unlikely(!strcasecmp(it->it_name, t->name)))
goto out_unlock;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
list_for_each_entry(t, &ce->tlist, list) {
if (!strcasecmp(t->name, it->it_name)) {
WRITE_ONCE(ce->tgthint, t);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
cifs_dbg(FYI, "%s: new target hint: %s\n", __func__,
it->it_name);
break;
}
}
out_unlock:
up_read(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
/**
* dfs_cache_get_tgt_referral - returns a DFS referral (@ref) from a given
* target iterator (@it).
*
* @path: canonical DFS path to lookup in DFS referral cache.
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
* @it: DFS target iterator.
* @ref: DFS referral pointer to set up the gathered information.
*
* Return zero if the DFS referral was set up correctly, otherwise non-zero.
*/
int dfs_cache_get_tgt_referral(const char *path, const struct dfs_cache_tgt_iterator *it,
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
struct dfs_info3_param *ref)
{
int rc;
struct cache_entry *ce;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (!it || !ref)
return -EINVAL;
cifs_dbg(FYI, "%s: path: %s\n", __func__, path);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
down_read(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
ce = lookup_cache_entry(path);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
if (IS_ERR(ce)) {
rc = PTR_ERR(ce);
goto out_unlock;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}
cifs_dbg(FYI, "%s: target name: %s\n", __func__, it->it_name);
rc = setup_referral(path, ce, ref, it->it_name);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
out_unlock:
up_read(&htable_rw_lock);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
return rc;
}
cifs: skip trailing separators of prefix paths During DFS failover, prefix paths may change, so make sure to not leave trailing separators when parsing thew in dfs_cache_get_tgt_share(). The separators of prefix paths are already handled by build_path_from_dentry_optional_prefix(). Consider the following DFS link: //dom/dfs/link: [\srv1\share\dir1, \srv2\share\dir1] Before commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1\ mv foo bar ... SMB2 430 Create Request File: dir1\\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\\bar;Close Request SMB2 478 Create Response File: dir1\\foo;SetInfo Response, Error: STATUS_OBJECT_NAME_INVALID;Close Response After commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1 mv foo bar ... SMB2 430 Create Request File: dir1\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\bar;Close Request SMB2 478 Create Response File: dir1\foo;SetInfo Response;Close Response Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com>
2022-06-03 19:13:02 +00:00
/* Extract share from DFS target and return a pointer to prefix path or NULL */
static const char *parse_target_share(const char *target, char **share)
{
const char *s, *seps = "/\\";
size_t len;
s = strpbrk(target + 1, seps);
if (!s)
return ERR_PTR(-EINVAL);
len = strcspn(s + 1, seps);
if (!len)
return ERR_PTR(-EINVAL);
s += len;
len = s - target + 1;
*share = kstrndup(target, len, GFP_KERNEL);
if (!*share)
return ERR_PTR(-ENOMEM);
s = target + len;
return s + strspn(s, seps);
}
/**
* dfs_cache_get_tgt_share - parse a DFS target
*
* @path: DFS full path
* @it: DFS target iterator.
* @share: tree name.
* @prefix: prefix path.
*
* Return zero if target was parsed correctly, otherwise non-zero.
*/
int dfs_cache_get_tgt_share(char *path, const struct dfs_cache_tgt_iterator *it, char **share,
char **prefix)
{
cifs: skip trailing separators of prefix paths During DFS failover, prefix paths may change, so make sure to not leave trailing separators when parsing thew in dfs_cache_get_tgt_share(). The separators of prefix paths are already handled by build_path_from_dentry_optional_prefix(). Consider the following DFS link: //dom/dfs/link: [\srv1\share\dir1, \srv2\share\dir1] Before commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1\ mv foo bar ... SMB2 430 Create Request File: dir1\\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\\bar;Close Request SMB2 478 Create Response File: dir1\\foo;SetInfo Response, Error: STATUS_OBJECT_NAME_INVALID;Close Response After commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1 mv foo bar ... SMB2 430 Create Request File: dir1\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\bar;Close Request SMB2 478 Create Response File: dir1\foo;SetInfo Response;Close Response Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com>
2022-06-03 19:13:02 +00:00
char sep;
char *target_share;
char *ppath = NULL;
cifs: skip trailing separators of prefix paths During DFS failover, prefix paths may change, so make sure to not leave trailing separators when parsing thew in dfs_cache_get_tgt_share(). The separators of prefix paths are already handled by build_path_from_dentry_optional_prefix(). Consider the following DFS link: //dom/dfs/link: [\srv1\share\dir1, \srv2\share\dir1] Before commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1\ mv foo bar ... SMB2 430 Create Request File: dir1\\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\\bar;Close Request SMB2 478 Create Response File: dir1\\foo;SetInfo Response, Error: STATUS_OBJECT_NAME_INVALID;Close Response After commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1 mv foo bar ... SMB2 430 Create Request File: dir1\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\bar;Close Request SMB2 478 Create Response File: dir1\foo;SetInfo Response;Close Response Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com>
2022-06-03 19:13:02 +00:00
const char *target_ppath, *dfsref_ppath;
size_t target_pplen, dfsref_pplen;
size_t len, c;
if (!it || !path || !share || !prefix || strlen(path) < it->it_path_consumed)
return -EINVAL;
sep = it->it_name[0];
if (sep != '\\' && sep != '/')
return -EINVAL;
cifs: skip trailing separators of prefix paths During DFS failover, prefix paths may change, so make sure to not leave trailing separators when parsing thew in dfs_cache_get_tgt_share(). The separators of prefix paths are already handled by build_path_from_dentry_optional_prefix(). Consider the following DFS link: //dom/dfs/link: [\srv1\share\dir1, \srv2\share\dir1] Before commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1\ mv foo bar ... SMB2 430 Create Request File: dir1\\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\\bar;Close Request SMB2 478 Create Response File: dir1\\foo;SetInfo Response, Error: STATUS_OBJECT_NAME_INVALID;Close Response After commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1 mv foo bar ... SMB2 430 Create Request File: dir1\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\bar;Close Request SMB2 478 Create Response File: dir1\foo;SetInfo Response;Close Response Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com>
2022-06-03 19:13:02 +00:00
target_ppath = parse_target_share(it->it_name, &target_share);
if (IS_ERR(target_ppath))
return PTR_ERR(target_ppath);
cifs: skip trailing separators of prefix paths During DFS failover, prefix paths may change, so make sure to not leave trailing separators when parsing thew in dfs_cache_get_tgt_share(). The separators of prefix paths are already handled by build_path_from_dentry_optional_prefix(). Consider the following DFS link: //dom/dfs/link: [\srv1\share\dir1, \srv2\share\dir1] Before commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1\ mv foo bar ... SMB2 430 Create Request File: dir1\\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\\bar;Close Request SMB2 478 Create Response File: dir1\\foo;SetInfo Response, Error: STATUS_OBJECT_NAME_INVALID;Close Response After commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1 mv foo bar ... SMB2 430 Create Request File: dir1\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\bar;Close Request SMB2 478 Create Response File: dir1\foo;SetInfo Response;Close Response Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com>
2022-06-03 19:13:02 +00:00
/* point to prefix in DFS referral path */
dfsref_ppath = path + it->it_path_consumed;
dfsref_ppath += strspn(dfsref_ppath, "/\\");
cifs: skip trailing separators of prefix paths During DFS failover, prefix paths may change, so make sure to not leave trailing separators when parsing thew in dfs_cache_get_tgt_share(). The separators of prefix paths are already handled by build_path_from_dentry_optional_prefix(). Consider the following DFS link: //dom/dfs/link: [\srv1\share\dir1, \srv2\share\dir1] Before commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1\ mv foo bar ... SMB2 430 Create Request File: dir1\\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\\bar;Close Request SMB2 478 Create Response File: dir1\\foo;SetInfo Response, Error: STATUS_OBJECT_NAME_INVALID;Close Response After commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1 mv foo bar ... SMB2 430 Create Request File: dir1\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\bar;Close Request SMB2 478 Create Response File: dir1\foo;SetInfo Response;Close Response Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com>
2022-06-03 19:13:02 +00:00
target_pplen = strlen(target_ppath);
dfsref_pplen = strlen(dfsref_ppath);
cifs: skip trailing separators of prefix paths During DFS failover, prefix paths may change, so make sure to not leave trailing separators when parsing thew in dfs_cache_get_tgt_share(). The separators of prefix paths are already handled by build_path_from_dentry_optional_prefix(). Consider the following DFS link: //dom/dfs/link: [\srv1\share\dir1, \srv2\share\dir1] Before commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1\ mv foo bar ... SMB2 430 Create Request File: dir1\\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\\bar;Close Request SMB2 478 Create Response File: dir1\\foo;SetInfo Response, Error: STATUS_OBJECT_NAME_INVALID;Close Response After commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1 mv foo bar ... SMB2 430 Create Request File: dir1\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\bar;Close Request SMB2 478 Create Response File: dir1\foo;SetInfo Response;Close Response Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com>
2022-06-03 19:13:02 +00:00
/* merge prefix paths from DFS referral path and target node */
if (target_pplen || dfsref_pplen) {
len = target_pplen + dfsref_pplen + 2;
ppath = kzalloc(len, GFP_KERNEL);
if (!ppath) {
kfree(target_share);
return -ENOMEM;
}
cifs: skip trailing separators of prefix paths During DFS failover, prefix paths may change, so make sure to not leave trailing separators when parsing thew in dfs_cache_get_tgt_share(). The separators of prefix paths are already handled by build_path_from_dentry_optional_prefix(). Consider the following DFS link: //dom/dfs/link: [\srv1\share\dir1, \srv2\share\dir1] Before commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1\ mv foo bar ... SMB2 430 Create Request File: dir1\\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\\bar;Close Request SMB2 478 Create Response File: dir1\\foo;SetInfo Response, Error: STATUS_OBJECT_NAME_INVALID;Close Response After commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1 mv foo bar ... SMB2 430 Create Request File: dir1\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\bar;Close Request SMB2 478 Create Response File: dir1\foo;SetInfo Response;Close Response Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com>
2022-06-03 19:13:02 +00:00
c = strscpy(ppath, target_ppath, len);
if (c && dfsref_pplen)
ppath[c] = sep;
strlcat(ppath, dfsref_ppath, len);
}
cifs: skip trailing separators of prefix paths During DFS failover, prefix paths may change, so make sure to not leave trailing separators when parsing thew in dfs_cache_get_tgt_share(). The separators of prefix paths are already handled by build_path_from_dentry_optional_prefix(). Consider the following DFS link: //dom/dfs/link: [\srv1\share\dir1, \srv2\share\dir1] Before commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1\ mv foo bar ... SMB2 430 Create Request File: dir1\\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\\bar;Close Request SMB2 478 Create Response File: dir1\\foo;SetInfo Response, Error: STATUS_OBJECT_NAME_INVALID;Close Response After commit: mount.cifs //dom/dfs/link tree connect to \\srv1\share; prefix_path=dir1 disconnect srv1; failover to srv2 tree connect to \\srv2\share; prefix_path=dir1 mv foo bar ... SMB2 430 Create Request File: dir1\foo;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 582 Create Response File: dir1\foo;GetInfo Response;Close Response SMB2 430 Create Request File: dir1\bar;GetInfo Request FILE_INFO/SMB2_FILE_ALL_INFO;Close Request SMB2 286 Create Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;GetInfo Response, Error: STATUS_OBJECT_NAME_NOT_FOUND;Close Response, Error: STATUS_OBJECT_NAME_NOT_FOUND SMB2 462 Create Request File: dir1\foo;SetInfo Request FILE_INFO/SMB2_FILE_RENAME_INFO NewName:dir1\bar;Close Request SMB2 478 Create Response File: dir1\foo;SetInfo Response;Close Response Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com>
2022-06-03 19:13:02 +00:00
*share = target_share;
*prefix = ppath;
return 0;
}
static bool target_share_equal(struct TCP_Server_Info *server, const char *s1, const char *s2)
{
char unc[sizeof("\\\\") + SERVER_NAME_LENGTH] = {0};
const char *host;
size_t hostlen;
struct sockaddr_storage ss;
bool match;
int rc;
if (strcasecmp(s1, s2))
return false;
/*
* Resolve share's hostname and check if server address matches. Otherwise just ignore it
* as we could not have upcall to resolve hostname or failed to convert ip address.
*/
extract_unc_hostname(s1, &host, &hostlen);
scnprintf(unc, sizeof(unc), "\\\\%.*s", (int)hostlen, host);
rc = dns_resolve_server_name_to_ip(unc, (struct sockaddr *)&ss, NULL);
if (rc < 0) {
cifs_dbg(FYI, "%s: could not resolve %.*s. assuming server address matches.\n",
__func__, (int)hostlen, host);
return true;
}
cifs_server_lock(server);
match = cifs_match_ipaddr((struct sockaddr *)&server->dstaddr, (struct sockaddr *)&ss);
cifs_server_unlock(server);
return match;
}
/*
* Mark dfs tcon for reconnecting when the currently connected tcon does not match any of the new
* target shares in @refs.
*/
static void mark_for_reconnect_if_needed(struct TCP_Server_Info *server,
const char *path,
struct dfs_cache_tgt_list *old_tl,
struct dfs_cache_tgt_list *new_tl)
{
struct dfs_cache_tgt_iterator *oit, *nit;
for (oit = dfs_cache_get_tgt_iterator(old_tl); oit;
oit = dfs_cache_get_next_tgt(old_tl, oit)) {
for (nit = dfs_cache_get_tgt_iterator(new_tl); nit;
nit = dfs_cache_get_next_tgt(new_tl, nit)) {
if (target_share_equal(server,
dfs_cache_get_tgt_name(oit),
dfs_cache_get_tgt_name(nit))) {
dfs_cache_noreq_update_tgthint(path, nit);
return;
}
}
}
cifs_dbg(FYI, "%s: no cached or matched targets. mark dfs share for reconnect.\n", __func__);
cifs_signal_cifsd_for_reconnect(server, true);
}
static bool is_ses_good(struct cifs_ses *ses)
{
struct TCP_Server_Info *server = ses->server;
struct cifs_tcon *tcon = ses->tcon_ipc;
bool ret;
spin_lock(&ses->ses_lock);
spin_lock(&ses->chan_lock);
ret = !cifs_chan_needs_reconnect(ses, server) &&
ses->ses_status == SES_GOOD &&
!tcon->need_reconnect;
spin_unlock(&ses->chan_lock);
spin_unlock(&ses->ses_lock);
return ret;
}
/* Refresh dfs referral of tcon and mark it for reconnect if needed */
static int __refresh_tcon(const char *path, struct cifs_ses *ses, bool force_refresh)
{
struct TCP_Server_Info *server = ses->server;
DFS_CACHE_TGT_LIST(old_tl);
DFS_CACHE_TGT_LIST(new_tl);
bool needs_refresh = false;
struct cache_entry *ce;
unsigned int xid;
int rc = 0;
xid = get_xid();
down_read(&htable_rw_lock);
ce = lookup_cache_entry(path);
needs_refresh = force_refresh || IS_ERR(ce) || cache_entry_expired(ce);
if (!IS_ERR(ce)) {
rc = get_targets(ce, &old_tl);
cifs_dbg(FYI, "%s: get_targets: %d\n", __func__, rc);
}
up_read(&htable_rw_lock);
if (!needs_refresh) {
rc = 0;
goto out;
}
ses = CIFS_DFS_ROOT_SES(ses);
if (!is_ses_good(ses)) {
cifs_dbg(FYI, "%s: skip cache refresh due to disconnected ipc\n",
__func__);
goto out;
}
ce = cache_refresh_path(xid, ses, path, true);
if (!IS_ERR(ce)) {
rc = get_targets(ce, &new_tl);
up_read(&htable_rw_lock);
cifs_dbg(FYI, "%s: get_targets: %d\n", __func__, rc);
mark_for_reconnect_if_needed(server, path, &old_tl, &new_tl);
}
out:
free_xid(xid);
dfs_cache_free_tgts(&old_tl);
dfs_cache_free_tgts(&new_tl);
return rc;
}
static int refresh_tcon(struct cifs_tcon *tcon, bool force_refresh)
{
struct TCP_Server_Info *server = tcon->ses->server;
struct cifs_ses *ses = tcon->ses;
mutex_lock(&server->refpath_lock);
if (server->leaf_fullpath)
__refresh_tcon(server->leaf_fullpath + 1, ses, force_refresh);
mutex_unlock(&server->refpath_lock);
return 0;
}
/**
* dfs_cache_remount_fs - remount a DFS share
*
* Reconfigure dfs mount by forcing a new DFS referral and if the currently cached targets do not
* match any of the new targets, mark it for reconnect.
*
* @cifs_sb: cifs superblock.
*
* Return zero if remounted, otherwise non-zero.
*/
int dfs_cache_remount_fs(struct cifs_sb_info *cifs_sb)
{
struct cifs_tcon *tcon;
if (!cifs_sb || !cifs_sb->master_tlink)
return -EINVAL;
tcon = cifs_sb_master_tcon(cifs_sb);
spin_lock(&tcon->tc_lock);
if (!tcon->origin_fullpath) {
spin_unlock(&tcon->tc_lock);
cifs_dbg(FYI, "%s: not a dfs mount\n", __func__);
return 0;
}
spin_unlock(&tcon->tc_lock);
/*
* After reconnecting to a different server, unique ids won't match anymore, so we disable
* serverino. This prevents dentry revalidation to think the dentry are stale (ESTALE).
*/
cifs_autodisable_serverino(cifs_sb);
/*
* Force the use of prefix path to support failover on DFS paths that resolve to targets
* that have different prefix paths.
*/
cifs_sb->mnt_cifs_flags |= CIFS_MOUNT_USE_PREFIX_PATH;
return refresh_tcon(tcon, true);
}
/* Refresh all DFS referrals related to DFS tcon */
void dfs_cache_refresh(struct work_struct *work)
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
{
struct TCP_Server_Info *server;
struct cifs_tcon *tcon;
struct cifs_ses *ses;
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
tcon = container_of(work, struct cifs_tcon, dfs_cache_work.work);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
for (ses = tcon->ses; ses; ses = ses->dfs_root_ses) {
server = ses->server;
mutex_lock(&server->refpath_lock);
if (server->leaf_fullpath)
__refresh_tcon(server->leaf_fullpath + 1, ses, false);
mutex_unlock(&server->refpath_lock);
}
queue_delayed_work(dfscache_wq, &tcon->dfs_cache_work,
atomic_read(&dfs_cache_ttl) * HZ);
cifs: Add DFS cache routines * Add new dfs_cache.[ch] files * Add new /proc/fs/cifs/dfscache file - dump current cache when read - clear current cache when writing "0" to it * Add delayed_work to periodically refresh cache entries The new interface will be used for caching DFS referrals, as well as supporting client target failover. The DFS cache is a hashtable that maps UNC paths to cache entries. A cache entry contains: - the UNC path it is mapped on - how much the the UNC path the entry consumes - flags - a Time-To-Live after which the entry expires - a list of possible targets (linked lists of UNC paths) - a "hint target" pointing the last known working target or the first target if none were tried. This hint lets cifs.ko remember and try working targets first. * Looking for an entry in the cache is done with dfs_cache_find() - if no valid entries are found, a DFS query is made, stored in the cache and returned - the full target list can be copied and returned to avoid race conditions and looped on with the help with the dfs_cache_tgt_iterator * Updating the target hint to the next target is done with dfs_cache_update_tgthint() These functions have a dfs_cache_noreq_XXX() version that doesn't fetches referrals if no entries are found. These versions don't require the tcp/ses/tcon/cifs_sb parameters as a result. Expired entries cannot be used and since they have a pretty short TTL [1] in order for them to be useful for failover the DFS cache adds a delayed work called periodically to keep them fresh. Since we might not have available connections to issue the referral request when refreshing we need to store volume_info structs with credentials and other needed info to be able to connect to the right server. 1: Windows defaults: 5mn for domain-based referrals, 30mn for regular links Signed-off-by: Paulo Alcantara <palcantara@suse.de> Signed-off-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com>
2018-11-14 18:01:21 +00:00
}