Minki/linux
1
0
Fork 1
mirror of https://github.com/torvalds/linux.git synced 2024-12-15 15:41:58 +00:00
Code Issues Packages Projects Releases Wiki Activity
12fdba564a
linux/fs/afs/server.c
David Howells 6e0e99d58a afs: Fix mmap coherency vs 3rd-party changes 
Fix the coherency management of mmap'd data such that 3rd-party changes
become visible as soon as possible after the callback notification is
delivered by the fileserver.  This is done by the following means:

 (1) When we break a callback on a vnode specified by the CB.CallBack call
     from the server, we queue a work item (vnode->cb_work) to go and
     clobber all the PTEs mapping to that inode.

     This causes the CPU to trip through the ->map_pages() and
     ->page_mkwrite() handlers if userspace attempts to access the page(s)
     again.

     (Ideally, this would be done in the service handler for CB.CallBack,
     but the server is waiting for our reply before considering, and we
     have a list of vnodes, all of which need breaking - and the process of
     getting the mmap_lock and stripping the PTEs on all CPUs could be
     quite slow.)

 (2) Call afs_validate() from the ->map_pages() handler to check to see if
     the file has changed and to get a new callback promise from the
     server.

Also handle the fileserver telling us that it's dropping all callbacks,
possibly after it's been restarted by sending us a CB.InitCallBackState*
call by the following means:

 (3) Maintain a per-cell list of afs files that are currently mmap'd
     (cell->fs_open_mmaps).

 (4) Add a work item to each server that is invoked if there are any open
     mmaps when CB.InitCallBackState happens.  This work item goes through
     the aforementioned list and invokes the vnode->cb_work work item for
     each one that is currently using this server.

     This causes the PTEs to be cleared, causing ->map_pages() or
     ->page_mkwrite() to be called again, thereby calling afs_validate()
     again.

I've chosen to simply strip the PTEs at the point of notification reception
rather than invalidate all the pages as well because (a) it's faster, (b)
we may get a notification for other reasons than the data being altered (in
which case we don't want to clobber the pagecache) and (c) we need to ask
the server to find out - and I don't want to wait for the reply before
holding up userspace.

This was tested using the attached test program:

	#include <stdbool.h>
	#include <stdio.h>
	#include <stdlib.h>
	#include <unistd.h>
	#include <fcntl.h>
	#include <sys/mman.h>
	int main(int argc, char *argv[])
	{
		size_t size = getpagesize();
		unsigned char *p;
		bool mod = (argc == 3);
		int fd;
		if (argc != 2 && argc != 3) {
			fprintf(stderr, "Format: %s <file> [mod]\n", argv[0]);
			exit(2);
		}
		fd = open(argv[1], mod ? O_RDWR : O_RDONLY);
		if (fd < 0) {
			perror(argv[1]);
			exit(1);
		}

		p = mmap(NULL, size, mod ? PROT_READ|PROT_WRITE : PROT_READ,
			 MAP_SHARED, fd, 0);
		if (p == MAP_FAILED) {
			perror("mmap");
			exit(1);
		}
		for (;;) {
			if (mod) {
				p[0]++;
				msync(p, size, MS_ASYNC);
				fsync(fd);
			}
			printf("%02x", p[0]);
			fflush(stdout);
			sleep(1);
		}
	}

It runs in two modes: in one mode, it mmaps a file, then sits in a loop
reading the first byte, printing it and sleeping for a second; in the
second mode it mmaps a file, then sits in a loop incrementing the first
byte and flushing, then printing and sleeping.

Two instances of this program can be run on different machines, one doing
the reading and one doing the writing.  The reader should see the changes
made by the writer, but without this patch, they aren't because validity
checking is being done lazily - only on entry to the filesystem.

Testing the InitCallBackState change is more complicated.  The server has
to be taken offline, the saved callback state file removed and then the
server restarted whilst the reading-mode program continues to run.  The
client machine then has to poke the server to trigger the InitCallBackState
call.

Signed-off-by: David Howells <dhowells@redhat.com>
Tested-by: Markus Suvanto <markus.suvanto@gmail.com>
cc: linux-afs@lists.infradead.org
Link: https://lore.kernel.org/r/163111668833.283156.382633263709075739.stgit@warthog.procyon.org.uk/
2021-09-13 09:10:39 +01:00

716 lines
18 KiB
C
Raw Blame History

// SPDX-License-Identifier: GPL-2.0-or-later
/* AFS server record management
*
* Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
* Written by David Howells (dhowells@redhat.com)
*/
#include <linux/sched.h>
#include <linux/slab.h>
#include "afs_fs.h"
#include "internal.h"
#include "protocol_yfs.h"
static unsigned afs_server_gc_delay = 10; /* Server record timeout in seconds */
static atomic_t afs_server_debug_id;
static struct afs_server *afs_maybe_use_server(struct afs_server *,
enum afs_server_trace);
static void __afs_put_server(struct afs_net *, struct afs_server *);
/*
* Find a server by one of its addresses.
*/
struct afs_server *afs_find_server(struct afs_net *net,
const struct sockaddr_rxrpc *srx)
{
const struct afs_addr_list *alist;
struct afs_server *server = NULL;
unsigned int i;
int seq = 0, diff;
rcu_read_lock();
do {
if (server)
afs_unuse_server_notime(net, server, afs_server_trace_put_find_rsq);
server = NULL;
read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
if (srx->transport.family == AF_INET6) {
const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
alist = rcu_dereference(server->addresses);
for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
b = &alist->addrs[i].transport.sin6;
diff = ((u16 __force)a->sin6_port -
(u16 __force)b->sin6_port);
if (diff == 0)
diff = memcmp(&a->sin6_addr,
&b->sin6_addr,
sizeof(struct in6_addr));
if (diff == 0)
goto found;
}
}
} else {
const struct sockaddr_in *a = &srx->transport.sin, *b;
hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
alist = rcu_dereference(server->addresses);
for (i = 0; i < alist->nr_ipv4; i++) {
b = &alist->addrs[i].transport.sin;
diff = ((u16 __force)a->sin_port -
(u16 __force)b->sin_port);
if (diff == 0)
diff = ((u32 __force)a->sin_addr.s_addr -
(u32 __force)b->sin_addr.s_addr);
if (diff == 0)
goto found;
}
}
}
server = NULL;
continue;
found:
server = afs_maybe_use_server(server, afs_server_trace_get_by_addr);
} while (need_seqretry(&net->fs_addr_lock, seq));
done_seqretry(&net->fs_addr_lock, seq);
rcu_read_unlock();
return server;
}
/*
* Look up a server by its UUID and mark it active.
*/
struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
{
struct afs_server *server = NULL;
struct rb_node *p;
int diff, seq = 0;
_enter("%pU", uuid);
do {
/* Unfortunately, rbtree walking doesn't give reliable results
* under just the RCU read lock, so we have to check for
* changes.
*/
if (server)
afs_unuse_server(net, server, afs_server_trace_put_uuid_rsq);
server = NULL;
read_seqbegin_or_lock(&net->fs_lock, &seq);
p = net->fs_servers.rb_node;
while (p) {
server = rb_entry(p, struct afs_server, uuid_rb);
diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
if (diff < 0) {
p = p->rb_left;
} else if (diff > 0) {
p = p->rb_right;
} else {
afs_use_server(server, afs_server_trace_get_by_uuid);
break;
}
server = NULL;
}
} while (need_seqretry(&net->fs_lock, seq));
done_seqretry(&net->fs_lock, seq);
_leave(" = %p", server);
return server;
}
/*
* Install a server record in the namespace tree. If there's a clash, we stick
* it into a list anchored on whichever afs_server struct is actually in the
* tree.
*/
static struct afs_server *afs_install_server(struct afs_cell *cell,
struct afs_server *candidate)
{
const struct afs_addr_list *alist;
struct afs_server *server, *next;
struct afs_net *net = cell->net;
struct rb_node **pp, *p;
int diff;
_enter("%p", candidate);
write_seqlock(&net->fs_lock);
/* Firstly install the server in the UUID lookup tree */
pp = &net->fs_servers.rb_node;
p = NULL;
while (*pp) {
p = *pp;
_debug("- consider %p", p);
server = rb_entry(p, struct afs_server, uuid_rb);
diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
if (diff < 0) {
pp = &(*pp)->rb_left;
} else if (diff > 0) {
pp = &(*pp)->rb_right;
} else {
if (server->cell == cell)
goto exists;
/* We have the same UUID representing servers in
* different cells. Append the new server to the list.
*/
for (;;) {
next = rcu_dereference_protected(
server->uuid_next,
lockdep_is_held(&net->fs_lock.lock));
if (!next)
break;
server = next;
}
rcu_assign_pointer(server->uuid_next, candidate);
candidate->uuid_prev = server;
server = candidate;
goto added_dup;
}
}
server = candidate;
rb_link_node(&server->uuid_rb, p, pp);
rb_insert_color(&server->uuid_rb, &net->fs_servers);
hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
added_dup:
write_seqlock(&net->fs_addr_lock);
alist = rcu_dereference_protected(server->addresses,
lockdep_is_held(&net->fs_addr_lock.lock));
/* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
* it in the IPv4 and/or IPv6 reverse-map lists.
*
* TODO: For speed we want to use something other than a flat list
* here; even sorting the list in terms of lowest address would help a
* bit, but anything we might want to do gets messy and memory
* intensive.
*/
if (alist->nr_ipv4 > 0)
hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
if (alist->nr_addrs > alist->nr_ipv4)
hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
write_sequnlock(&net->fs_addr_lock);
exists:
afs_get_server(server, afs_server_trace_get_install);
write_sequnlock(&net->fs_lock);
return server;
}
/*
* Allocate a new server record and mark it active.
*/
static struct afs_server *afs_alloc_server(struct afs_cell *cell,
const uuid_t *uuid,
struct afs_addr_list *alist)
{
struct afs_server *server;
struct afs_net *net = cell->net;
_enter("");
server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
if (!server)
goto enomem;
atomic_set(&server->ref, 1);
atomic_set(&server->active, 1);
server->debug_id = atomic_inc_return(&afs_server_debug_id);
RCU_INIT_POINTER(server->addresses, alist);
server->addr_version = alist->version;
server->uuid = *uuid;
rwlock_init(&server->fs_lock);
INIT_WORK(&server->initcb_work, afs_server_init_callback_work);
init_waitqueue_head(&server->probe_wq);
INIT_LIST_HEAD(&server->probe_link);
spin_lock_init(&server->probe_lock);
server->cell = cell;
server->rtt = UINT_MAX;
afs_inc_servers_outstanding(net);
trace_afs_server(server, 1, 1, afs_server_trace_alloc);
_leave(" = %p", server);
return server;
enomem:
_leave(" = NULL [nomem]");
return NULL;
}
/*
* Look up an address record for a server
*/
static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
struct key *key, const uuid_t *uuid)
{
struct afs_vl_cursor vc;
struct afs_addr_list *alist = NULL;
int ret;
ret = -ERESTARTSYS;
if (afs_begin_vlserver_operation(&vc, cell, key)) {
while (afs_select_vlserver(&vc)) {
if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
alist = afs_yfsvl_get_endpoints(&vc, uuid);
else
alist = afs_vl_get_addrs_u(&vc, uuid);
}
ret = afs_end_vlserver_operation(&vc);
}
return ret < 0 ? ERR_PTR(ret) : alist;
}
/*
* Get or create a fileserver record.
*/
struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
const uuid_t *uuid, u32 addr_version)
{
struct afs_addr_list *alist;
struct afs_server *server, *candidate;
_enter("%p,%pU", cell->net, uuid);
server = afs_find_server_by_uuid(cell->net, uuid);
if (server) {
if (server->addr_version != addr_version)
set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
return server;
}
alist = afs_vl_lookup_addrs(cell, key, uuid);
if (IS_ERR(alist))
return ERR_CAST(alist);
candidate = afs_alloc_server(cell, uuid, alist);
if (!candidate) {
afs_put_addrlist(alist);
return ERR_PTR(-ENOMEM);
}
server = afs_install_server(cell, candidate);
if (server != candidate) {
afs_put_addrlist(alist);
kfree(candidate);
} else {
/* Immediately dispatch an asynchronous probe to each interface
* on the fileserver. This will make sure the repeat-probing
* service is started.
*/
afs_fs_probe_fileserver(cell->net, server, key, true);
}
return server;
}
/*
* Set the server timer to fire after a given delay, assuming it's not already
* set for an earlier time.
*/
static void afs_set_server_timer(struct afs_net *net, time64_t delay)
{
if (net->live) {
afs_inc_servers_outstanding(net);
if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
afs_dec_servers_outstanding(net);
}
}
/*
* Server management timer. We have an increment on fs_outstanding that we
* need to pass along to the work item.
*/
void afs_servers_timer(struct timer_list *timer)
{
struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
_enter("");
if (!queue_work(afs_wq, &net->fs_manager))
afs_dec_servers_outstanding(net);
}
/*
* Get a reference on a server object.
*/
struct afs_server *afs_get_server(struct afs_server *server,
enum afs_server_trace reason)
{
unsigned int u = atomic_inc_return(&server->ref);
trace_afs_server(server, u, atomic_read(&server->active), reason);
return server;
}
/*
* Try to get a reference on a server object.
*/
static struct afs_server *afs_maybe_use_server(struct afs_server *server,
enum afs_server_trace reason)
{
unsigned int r = atomic_fetch_add_unless(&server->ref, 1, 0);
unsigned int a;
if (r == 0)
return NULL;
a = atomic_inc_return(&server->active);
trace_afs_server(server, r, a, reason);
return server;
}
/*
* Get an active count on a server object.
*/
struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
{
unsigned int r = atomic_inc_return(&server->ref);
unsigned int a = atomic_inc_return(&server->active);
trace_afs_server(server, r, a, reason);
return server;
}
/*
* Release a reference on a server record.
*/
void afs_put_server(struct afs_net *net, struct afs_server *server,
enum afs_server_trace reason)
{
unsigned int usage;
if (!server)
return;
usage = atomic_dec_return(&server->ref);
trace_afs_server(server, usage, atomic_read(&server->active), reason);
if (unlikely(usage == 0))
__afs_put_server(net, server);
}
/*
* Drop an active count on a server object without updating the last-unused
* time.
*/
void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
enum afs_server_trace reason)
{
if (server) {
unsigned int active = atomic_dec_return(&server->active);
if (active == 0)
afs_set_server_timer(net, afs_server_gc_delay);
afs_put_server(net, server, reason);
}
}
/*
* Drop an active count on a server object.
*/
void afs_unuse_server(struct afs_net *net, struct afs_server *server,
enum afs_server_trace reason)
{
if (server) {
server->unuse_time = ktime_get_real_seconds();
afs_unuse_server_notime(net, server, reason);
}
}
static void afs_server_rcu(struct rcu_head *rcu)
{
struct afs_server *server = container_of(rcu, struct afs_server, rcu);
trace_afs_server(server, atomic_read(&server->ref),
atomic_read(&server->active), afs_server_trace_free);
afs_put_addrlist(rcu_access_pointer(server->addresses));
kfree(server);
}
static void __afs_put_server(struct afs_net *net, struct afs_server *server)
{
call_rcu(&server->rcu, afs_server_rcu);
afs_dec_servers_outstanding(net);
}
static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
{
struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
struct afs_addr_cursor ac = {
.alist = alist,
.index = alist->preferred,
.error = 0,
};
afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
}
/*
* destroy a dead server
*/
static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
{
if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
afs_give_up_callbacks(net, server);
flush_work(&server->initcb_work);
afs_put_server(net, server, afs_server_trace_destroy);
}
/*
* Garbage collect any expired servers.
*/
static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
{
struct afs_server *server, *next, *prev;
int active;
while ((server = gc_list)) {
gc_list = server->gc_next;
write_seqlock(&net->fs_lock);
active = atomic_read(&server->active);
if (active == 0) {
trace_afs_server(server, atomic_read(&server->ref),
active, afs_server_trace_gc);
next = rcu_dereference_protected(
server->uuid_next, lockdep_is_held(&net->fs_lock.lock));
prev = server->uuid_prev;
if (!prev) {
/* The one at the front is in the tree */
if (!next) {
rb_erase(&server->uuid_rb, &net->fs_servers);
} else {
rb_replace_node_rcu(&server->uuid_rb,
&next->uuid_rb,
&net->fs_servers);
next->uuid_prev = NULL;
}
} else {
/* This server is not at the front */
rcu_assign_pointer(prev->uuid_next, next);
if (next)
next->uuid_prev = prev;
}
list_del(&server->probe_link);
hlist_del_rcu(&server->proc_link);
if (!hlist_unhashed(&server->addr4_link))
hlist_del_rcu(&server->addr4_link);
if (!hlist_unhashed(&server->addr6_link))
hlist_del_rcu(&server->addr6_link);
}
write_sequnlock(&net->fs_lock);
if (active == 0)
afs_destroy_server(net, server);
}
}
/*
* Manage the records of servers known to be within a network namespace. This
* includes garbage collecting unused servers.
*
* Note also that we were given an increment on net->servers_outstanding by
* whoever queued us that we need to deal with before returning.
*/
void afs_manage_servers(struct work_struct *work)
{
struct afs_net *net = container_of(work, struct afs_net, fs_manager);
struct afs_server *gc_list = NULL;
struct rb_node *cursor;
time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
bool purging = !net->live;
_enter("");
/* Trawl the server list looking for servers that have expired from
* lack of use.
*/
read_seqlock_excl(&net->fs_lock);
for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
struct afs_server *server =
rb_entry(cursor, struct afs_server, uuid_rb);
int active = atomic_read(&server->active);
_debug("manage %pU %u", &server->uuid, active);
if (purging) {
trace_afs_server(server, atomic_read(&server->ref),
active, afs_server_trace_purging);
if (active != 0)
pr_notice("Can't purge s=%08x\n", server->debug_id);
}
if (active == 0) {
time64_t expire_at = server->unuse_time;
if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
!test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
expire_at += afs_server_gc_delay;
if (purging || expire_at <= now) {
server->gc_next = gc_list;
gc_list = server;
} else if (expire_at < next_manage) {
next_manage = expire_at;
}
}
}
read_sequnlock_excl(&net->fs_lock);
/* Update the timer on the way out. We have to pass an increment on
* servers_outstanding in the namespace that we are in to the timer or
* the work scheduler.
*/
if (!purging && next_manage < TIME64_MAX) {
now = ktime_get_real_seconds();
if (next_manage - now <= 0) {
if (queue_work(afs_wq, &net->fs_manager))
afs_inc_servers_outstanding(net);
} else {
afs_set_server_timer(net, next_manage - now);
}
}
afs_gc_servers(net, gc_list);
afs_dec_servers_outstanding(net);
_leave(" [%d]", atomic_read(&net->servers_outstanding));
}
static void afs_queue_server_manager(struct afs_net *net)
{
afs_inc_servers_outstanding(net);
if (!queue_work(afs_wq, &net->fs_manager))
afs_dec_servers_outstanding(net);
}
/*
* Purge list of servers.
*/
void afs_purge_servers(struct afs_net *net)
{
_enter("");
if (del_timer_sync(&net->fs_timer))
afs_dec_servers_outstanding(net);
afs_queue_server_manager(net);
_debug("wait");
atomic_dec(&net->servers_outstanding);
wait_var_event(&net->servers_outstanding,
!atomic_read(&net->servers_outstanding));
_leave("");
}
/*
* Get an update for a server's address list.
*/
static noinline bool afs_update_server_record(struct afs_operation *op,
struct afs_server *server)
{
struct afs_addr_list *alist, *discard;
_enter("");
trace_afs_server(server, atomic_read(&server->ref), atomic_read(&server->active),
afs_server_trace_update);
alist = afs_vl_lookup_addrs(op->volume->cell, op->key, &server->uuid);
if (IS_ERR(alist)) {
if ((PTR_ERR(alist) == -ERESTARTSYS ||
PTR_ERR(alist) == -EINTR) &&
(op->flags & AFS_OPERATION_UNINTR) &&
server->addresses) {
_leave(" = t [intr]");
return true;
}
op->error = PTR_ERR(alist);
_leave(" = f [%d]", op->error);
return false;
}
discard = alist;
if (server->addr_version != alist->version) {
write_lock(&server->fs_lock);
discard = rcu_dereference_protected(server->addresses,
lockdep_is_held(&server->fs_lock));
rcu_assign_pointer(server->addresses, alist);
server->addr_version = alist->version;
write_unlock(&server->fs_lock);
}
afs_put_addrlist(discard);
_leave(" = t");
return true;
}
/*
* See if a server's address list needs updating.
*/
bool afs_check_server_record(struct afs_operation *op, struct afs_server *server)
{
bool success;
int ret, retries = 0;
_enter("");
ASSERT(server);
retry:
if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
goto wait;
if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
goto update;
_leave(" = t [good]");
return true;
update:
if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
success = afs_update_server_record(op, server);
clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
_leave(" = %d", success);
return success;
}
wait:
ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
(op->flags & AFS_OPERATION_UNINTR) ?
TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
if (ret == -ERESTARTSYS) {
op->error = ret;
_leave(" = f [intr]");
return false;
}
retries++;
if (retries == 4) {
_leave(" = f [stale]");
ret = -ESTALE;
return false;
}
goto retry;
}
Reference in New Issue View Git Blame Copy Permalink