linux/lib/rhashtable.c

790 lines
19 KiB
C
Raw Normal View History

/*
* Resizable, Scalable, Concurrent Hash Table
*
* Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au>
* Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
* Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
*
* Code partially derived from nft_hash
* Rewritten with rehash code from br_multicast plus single list
* pointer as suggested by Josh Triplett
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/log2.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/mm.h>
2014-12-10 15:33:11 +00:00
#include <linux/jhash.h>
#include <linux/random.h>
#include <linux/rhashtable.h>
#include <linux/err.h>
#define HASH_DEFAULT_SIZE 64UL
#define HASH_MIN_SIZE 4U
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
#define BUCKET_LOCKS_PER_CPU 128UL
static u32 head_hashfn(struct rhashtable *ht,
const struct bucket_table *tbl,
const struct rhash_head *he)
{
return rht_head_hashfn(ht, tbl, he, ht->p);
}
#ifdef CONFIG_PROVE_LOCKING
#define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
int lockdep_rht_mutex_is_held(struct rhashtable *ht)
{
return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
}
EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
{
spinlock_t *lock = rht_bucket_lock(tbl, hash);
return (debug_locks) ? lockdep_is_held(lock) : 1;
}
EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
#else
#define ASSERT_RHT_MUTEX(HT)
#endif
static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl,
gfp_t gfp)
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
{
unsigned int i, size;
#if defined(CONFIG_PROVE_LOCKING)
unsigned int nr_pcpus = 2;
#else
unsigned int nr_pcpus = num_possible_cpus();
#endif
nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL);
size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul);
/* Never allocate more than 0.5 locks per bucket */
size = min_t(unsigned int, size, tbl->size >> 1);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
if (sizeof(spinlock_t) != 0) {
#ifdef CONFIG_NUMA
if (size * sizeof(spinlock_t) > PAGE_SIZE &&
gfp == GFP_KERNEL)
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
tbl->locks = vmalloc(size * sizeof(spinlock_t));
else
#endif
tbl->locks = kmalloc_array(size, sizeof(spinlock_t),
gfp);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
if (!tbl->locks)
return -ENOMEM;
for (i = 0; i < size; i++)
spin_lock_init(&tbl->locks[i]);
}
tbl->locks_mask = size - 1;
return 0;
}
static void bucket_table_free(const struct bucket_table *tbl)
{
if (tbl)
kvfree(tbl->locks);
kvfree(tbl);
}
static void bucket_table_free_rcu(struct rcu_head *head)
{
bucket_table_free(container_of(head, struct bucket_table, rcu));
}
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
size_t nbuckets,
gfp_t gfp)
{
struct bucket_table *tbl = NULL;
size_t size;
int i;
size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER) ||
gfp != GFP_KERNEL)
tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY);
if (tbl == NULL && gfp == GFP_KERNEL)
tbl = vzalloc(size);
if (tbl == NULL)
return NULL;
tbl->size = nbuckets;
if (alloc_bucket_locks(ht, tbl, gfp) < 0) {
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
bucket_table_free(tbl);
return NULL;
}
INIT_LIST_HEAD(&tbl->walkers);
get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
for (i = 0; i < nbuckets; i++)
INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
return tbl;
}
static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
struct bucket_table *tbl)
{
struct bucket_table *new_tbl;
do {
new_tbl = tbl;
tbl = rht_dereference_rcu(tbl->future_tbl, ht);
} while (tbl);
return new_tbl;
}
static int rhashtable_rehash_one(struct rhashtable *ht, unsigned old_hash)
{
struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
struct bucket_table *new_tbl = rhashtable_last_table(ht,
rht_dereference_rcu(old_tbl->future_tbl, ht));
struct rhash_head __rcu **pprev = &old_tbl->buckets[old_hash];
int err = -ENOENT;
struct rhash_head *head, *next, *entry;
spinlock_t *new_bucket_lock;
unsigned new_hash;
rht_for_each(entry, old_tbl, old_hash) {
err = 0;
next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
if (rht_is_a_nulls(next))
break;
pprev = &entry->next;
}
if (err)
goto out;
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
new_hash = head_hashfn(ht, new_tbl, entry);
new_bucket_lock = rht_bucket_lock(new_tbl, new_hash);
spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING);
head = rht_dereference_bucket(new_tbl->buckets[new_hash],
new_tbl, new_hash);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
if (rht_is_a_nulls(head))
INIT_RHT_NULLS_HEAD(entry->next, ht, new_hash);
else
RCU_INIT_POINTER(entry->next, head);
rcu_assign_pointer(new_tbl->buckets[new_hash], entry);
spin_unlock(new_bucket_lock);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
rcu_assign_pointer(*pprev, next);
out:
return err;
}
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
static void rhashtable_rehash_chain(struct rhashtable *ht, unsigned old_hash)
{
struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
spinlock_t *old_bucket_lock;
old_bucket_lock = rht_bucket_lock(old_tbl, old_hash);
spin_lock_bh(old_bucket_lock);
while (!rhashtable_rehash_one(ht, old_hash))
;
old_tbl->rehash++;
spin_unlock_bh(old_bucket_lock);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
}
static int rhashtable_rehash_attach(struct rhashtable *ht,
struct bucket_table *old_tbl,
struct bucket_table *new_tbl)
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
{
/* Protect future_tbl using the first bucket lock. */
spin_lock_bh(old_tbl->locks);
/* Did somebody beat us to it? */
if (rcu_access_pointer(old_tbl->future_tbl)) {
spin_unlock_bh(old_tbl->locks);
return -EEXIST;
}
/* Make insertions go into the new, empty table right away. Deletions
* and lookups will be attempted in both tables until we synchronize.
*/
rcu_assign_pointer(old_tbl->future_tbl, new_tbl);
/* Ensure the new table is visible to readers. */
smp_wmb();
spin_unlock_bh(old_tbl->locks);
return 0;
}
static int rhashtable_rehash_table(struct rhashtable *ht)
{
struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
struct bucket_table *new_tbl;
struct rhashtable_walker *walker;
unsigned old_hash;
new_tbl = rht_dereference(old_tbl->future_tbl, ht);
if (!new_tbl)
return 0;
for (old_hash = 0; old_hash < old_tbl->size; old_hash++)
rhashtable_rehash_chain(ht, old_hash);
/* Publish the new table pointer. */
rcu_assign_pointer(ht->tbl, new_tbl);
spin_lock(&ht->lock);
list_for_each_entry(walker, &old_tbl->walkers, list)
walker->tbl = NULL;
spin_unlock(&ht->lock);
/* Wait for readers. All new readers will see the new
* table, and thus no references to the old table will
* remain.
*/
call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
}
/**
* rhashtable_expand - Expand hash table while allowing concurrent lookups
* @ht: the hash table to expand
*
* A secondary bucket array is allocated and the hash entries are migrated.
*
* This function may only be called in a context where it is safe to call
* synchronize_rcu(), e.g. not within a rcu_read_lock() section.
*
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
* The caller must ensure that no concurrent resizing occurs by holding
* ht->mutex.
*
* It is valid to have concurrent insertions and deletions protected by per
* bucket locks or concurrent RCU protected lookups and traversals.
*/
static int rhashtable_expand(struct rhashtable *ht)
{
struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
int err;
ASSERT_RHT_MUTEX(ht);
old_tbl = rhashtable_last_table(ht, old_tbl);
new_tbl = bucket_table_alloc(ht, old_tbl->size * 2, GFP_KERNEL);
if (new_tbl == NULL)
return -ENOMEM;
err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
if (err)
bucket_table_free(new_tbl);
return err;
}
/**
* rhashtable_shrink - Shrink hash table while allowing concurrent lookups
* @ht: the hash table to shrink
*
* This function shrinks the hash table to fit, i.e., the smallest
* size would not cause it to expand right away automatically.
*
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
* The caller must ensure that no concurrent resizing occurs by holding
* ht->mutex.
*
* The caller must ensure that no concurrent table mutations take place.
* It is however valid to have concurrent lookups if they are RCU protected.
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
*
* It is valid to have concurrent insertions and deletions protected by per
* bucket locks or concurrent RCU protected lookups and traversals.
*/
static int rhashtable_shrink(struct rhashtable *ht)
{
struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
unsigned size = roundup_pow_of_two(atomic_read(&ht->nelems) * 3 / 2);
int err;
ASSERT_RHT_MUTEX(ht);
if (size < ht->p.min_size)
size = ht->p.min_size;
if (old_tbl->size <= size)
return 0;
if (rht_dereference(old_tbl->future_tbl, ht))
return -EEXIST;
new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
if (new_tbl == NULL)
return -ENOMEM;
err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
if (err)
bucket_table_free(new_tbl);
return err;
}
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
static void rht_deferred_worker(struct work_struct *work)
{
struct rhashtable *ht;
struct bucket_table *tbl;
int err = 0;
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
ht = container_of(work, struct rhashtable, run_work);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
mutex_lock(&ht->mutex);
if (ht->being_destroyed)
goto unlock;
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
tbl = rht_dereference(ht->tbl, ht);
tbl = rhashtable_last_table(ht, tbl);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
if (rht_grow_above_75(ht, tbl))
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
rhashtable_expand(ht);
else if (rht_shrink_below_30(ht, tbl))
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
rhashtable_shrink(ht);
err = rhashtable_rehash_table(ht);
unlock:
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
mutex_unlock(&ht->mutex);
if (err)
schedule_work(&ht->run_work);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
}
static bool rhashtable_check_elasticity(struct rhashtable *ht,
struct bucket_table *tbl,
unsigned hash)
{
unsigned elasticity = ht->elasticity;
struct rhash_head *head;
rht_for_each(head, tbl, hash)
if (!--elasticity)
return true;
return false;
}
int rhashtable_insert_rehash(struct rhashtable *ht)
{
struct bucket_table *old_tbl;
struct bucket_table *new_tbl;
struct bucket_table *tbl;
unsigned int size;
int err;
old_tbl = rht_dereference_rcu(ht->tbl, ht);
tbl = rhashtable_last_table(ht, old_tbl);
size = tbl->size;
if (rht_grow_above_75(ht, tbl))
size *= 2;
/* More than two rehashes (not resizes) detected. */
else if (WARN_ON(old_tbl != tbl && old_tbl->size == size))
return -EBUSY;
new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC);
if (new_tbl == NULL)
return -ENOMEM;
err = rhashtable_rehash_attach(ht, tbl, new_tbl);
if (err) {
bucket_table_free(new_tbl);
if (err == -EEXIST)
err = 0;
} else
schedule_work(&ht->run_work);
return err;
}
EXPORT_SYMBOL_GPL(rhashtable_insert_rehash);
int rhashtable_insert_slow(struct rhashtable *ht, const void *key,
struct rhash_head *obj,
struct bucket_table *tbl)
{
struct rhash_head *head;
unsigned hash;
int err;
tbl = rhashtable_last_table(ht, tbl);
hash = head_hashfn(ht, tbl, obj);
spin_lock_nested(rht_bucket_lock(tbl, hash), SINGLE_DEPTH_NESTING);
err = -EEXIST;
if (key && rhashtable_lookup_fast(ht, key, ht->p))
goto exit;
err = -EAGAIN;
if (rhashtable_check_elasticity(ht, tbl, hash) ||
rht_grow_above_100(ht, tbl))
goto exit;
err = 0;
head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
RCU_INIT_POINTER(obj->next, head);
rcu_assign_pointer(tbl->buckets[hash], obj);
atomic_inc(&ht->nelems);
exit:
spin_unlock(rht_bucket_lock(tbl, hash));
return err;
}
EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
/**
* rhashtable_walk_init - Initialise an iterator
* @ht: Table to walk over
* @iter: Hash table Iterator
*
* This function prepares a hash table walk.
*
* Note that if you restart a walk after rhashtable_walk_stop you
* may see the same object twice. Also, you may miss objects if
* there are removals in between rhashtable_walk_stop and the next
* call to rhashtable_walk_start.
*
* For a completely stable walk you should construct your own data
* structure outside the hash table.
*
* This function may sleep so you must not call it from interrupt
* context or with spin locks held.
*
* You must call rhashtable_walk_exit if this function returns
* successfully.
*/
int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter)
{
iter->ht = ht;
iter->p = NULL;
iter->slot = 0;
iter->skip = 0;
iter->walker = kmalloc(sizeof(*iter->walker), GFP_KERNEL);
if (!iter->walker)
return -ENOMEM;
mutex_lock(&ht->mutex);
iter->walker->tbl = rht_dereference(ht->tbl, ht);
list_add(&iter->walker->list, &iter->walker->tbl->walkers);
mutex_unlock(&ht->mutex);
return 0;
}
EXPORT_SYMBOL_GPL(rhashtable_walk_init);
/**
* rhashtable_walk_exit - Free an iterator
* @iter: Hash table Iterator
*
* This function frees resources allocated by rhashtable_walk_init.
*/
void rhashtable_walk_exit(struct rhashtable_iter *iter)
{
mutex_lock(&iter->ht->mutex);
if (iter->walker->tbl)
list_del(&iter->walker->list);
mutex_unlock(&iter->ht->mutex);
kfree(iter->walker);
}
EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
/**
* rhashtable_walk_start - Start a hash table walk
* @iter: Hash table iterator
*
* Start a hash table walk. Note that we take the RCU lock in all
* cases including when we return an error. So you must always call
* rhashtable_walk_stop to clean up.
*
* Returns zero if successful.
*
* Returns -EAGAIN if resize event occured. Note that the iterator
* will rewind back to the beginning and you may use it immediately
* by calling rhashtable_walk_next.
*/
int rhashtable_walk_start(struct rhashtable_iter *iter)
__acquires(RCU)
{
struct rhashtable *ht = iter->ht;
mutex_lock(&ht->mutex);
if (iter->walker->tbl)
list_del(&iter->walker->list);
rcu_read_lock();
mutex_unlock(&ht->mutex);
if (!iter->walker->tbl) {
iter->walker->tbl = rht_dereference_rcu(ht->tbl, ht);
return -EAGAIN;
}
return 0;
}
EXPORT_SYMBOL_GPL(rhashtable_walk_start);
/**
* rhashtable_walk_next - Return the next object and advance the iterator
* @iter: Hash table iterator
*
* Note that you must call rhashtable_walk_stop when you are finished
* with the walk.
*
* Returns the next object or NULL when the end of the table is reached.
*
* Returns -EAGAIN if resize event occured. Note that the iterator
* will rewind back to the beginning and you may continue to use it.
*/
void *rhashtable_walk_next(struct rhashtable_iter *iter)
{
struct bucket_table *tbl = iter->walker->tbl;
struct rhashtable *ht = iter->ht;
struct rhash_head *p = iter->p;
void *obj = NULL;
if (p) {
p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot);
goto next;
}
for (; iter->slot < tbl->size; iter->slot++) {
int skip = iter->skip;
rht_for_each_rcu(p, tbl, iter->slot) {
if (!skip)
break;
skip--;
}
next:
if (!rht_is_a_nulls(p)) {
iter->skip++;
iter->p = p;
obj = rht_obj(ht, p);
goto out;
}
iter->skip = 0;
}
/* Ensure we see any new tables. */
smp_rmb();
iter->walker->tbl = rht_dereference_rcu(tbl->future_tbl, ht);
if (iter->walker->tbl) {
iter->slot = 0;
iter->skip = 0;
return ERR_PTR(-EAGAIN);
}
iter->p = NULL;
out:
return obj;
}
EXPORT_SYMBOL_GPL(rhashtable_walk_next);
/**
* rhashtable_walk_stop - Finish a hash table walk
* @iter: Hash table iterator
*
* Finish a hash table walk.
*/
void rhashtable_walk_stop(struct rhashtable_iter *iter)
__releases(RCU)
{
struct rhashtable *ht;
struct bucket_table *tbl = iter->walker->tbl;
if (!tbl)
goto out;
ht = iter->ht;
spin_lock(&ht->lock);
if (tbl->rehash < tbl->size)
list_add(&iter->walker->list, &tbl->walkers);
else
iter->walker->tbl = NULL;
spin_unlock(&ht->lock);
iter->p = NULL;
out:
rcu_read_unlock();
}
EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
static size_t rounded_hashtable_size(const struct rhashtable_params *params)
{
return max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
(unsigned long)params->min_size);
}
static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
{
return jhash2(key, length, seed);
}
/**
* rhashtable_init - initialize a new hash table
* @ht: hash table to be initialized
* @params: configuration parameters
*
* Initializes a new hash table based on the provided configuration
* parameters. A table can be configured either with a variable or
* fixed length key:
*
* Configuration Example 1: Fixed length keys
* struct test_obj {
* int key;
* void * my_member;
* struct rhash_head node;
* };
*
* struct rhashtable_params params = {
* .head_offset = offsetof(struct test_obj, node),
* .key_offset = offsetof(struct test_obj, key),
* .key_len = sizeof(int),
2014-12-10 15:33:11 +00:00
* .hashfn = jhash,
* .nulls_base = (1U << RHT_BASE_SHIFT),
* };
*
* Configuration Example 2: Variable length keys
* struct test_obj {
* [...]
* struct rhash_head node;
* };
*
* u32 my_hash_fn(const void *data, u32 seed)
* {
* struct test_obj *obj = data;
*
* return [... hash ...];
* }
*
* struct rhashtable_params params = {
* .head_offset = offsetof(struct test_obj, node),
2014-12-10 15:33:11 +00:00
* .hashfn = jhash,
* .obj_hashfn = my_hash_fn,
* };
*/
int rhashtable_init(struct rhashtable *ht,
const struct rhashtable_params *params)
{
struct bucket_table *tbl;
size_t size;
size = HASH_DEFAULT_SIZE;
if ((!params->key_len && !params->obj_hashfn) ||
(params->obj_hashfn && !params->obj_cmpfn))
return -EINVAL;
if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
return -EINVAL;
if (params->nelem_hint)
size = rounded_hashtable_size(params);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
memset(ht, 0, sizeof(*ht));
mutex_init(&ht->mutex);
spin_lock_init(&ht->lock);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
memcpy(&ht->p, params, sizeof(*params));
if (params->min_size)
ht->p.min_size = roundup_pow_of_two(params->min_size);
if (params->max_size)
ht->p.max_size = rounddown_pow_of_two(params->max_size);
ht->p.min_size = max(ht->p.min_size, HASH_MIN_SIZE);
if (!params->insecure_elasticity)
ht->elasticity = 16;
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
if (params->locks_mul)
ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
else
ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
ht->key_len = ht->p.key_len;
if (!params->hashfn) {
ht->p.hashfn = jhash;
if (!(ht->key_len & (sizeof(u32) - 1))) {
ht->key_len /= sizeof(u32);
ht->p.hashfn = rhashtable_jhash2;
}
}
tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
if (tbl == NULL)
return -ENOMEM;
atomic_set(&ht->nelems, 0);
RCU_INIT_POINTER(ht->tbl, tbl);
INIT_WORK(&ht->run_work, rht_deferred_worker);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
return 0;
}
EXPORT_SYMBOL_GPL(rhashtable_init);
/**
* rhashtable_destroy - destroy hash table
* @ht: the hash table to destroy
*
rhashtable: fix lockdep splat in rhashtable_destroy() No need for rht_dereference() from rhashtable_destroy() since the existing callers don't hold the mutex when invoking this function from: 1) Netlink, this is called in case of memory allocation errors in the initialization path, no nl_sk_hash_lock is held. 2) Netfilter, this is called from the rcu callback, no nfnl_lock is held either. I think it's reasonable to assume that the caller has to make sure that no hash resizing may happen before releasing the bucket array. Therefore, the caller should be responsible for releasing this in a safe way, document this to make people aware of it. This resolves a rcu lockdep splat in nft_hash: =============================== [ INFO: suspicious RCU usage. ] 3.16.0+ #178 Not tainted ------------------------------- lib/rhashtable.c:596 suspicious rcu_dereference_protected() usage! other info that might help us debug this: rcu_scheduler_active = 1, debug_locks = 1 1 lock held by ksoftirqd/2/18: #0: (rcu_callback){......}, at: [<ffffffff810918fd>] rcu_process_callbacks+0x27e/0x4c7 stack backtrace: CPU: 2 PID: 18 Comm: ksoftirqd/2 Not tainted 3.16.0+ #178 Hardware name: LENOVO 23259H1/23259H1, BIOS G2ET32WW (1.12 ) 05/30/2012 0000000000000001 ffff88011706bb68 ffffffff8143debc 0000000000000000 ffff880117062610 ffff88011706bb98 ffffffff81077515 ffff8800ca041a50 0000000000000004 ffff8800ca386480 ffff8800ca041a00 ffff88011706bbb8 Call Trace: [<ffffffff8143debc>] dump_stack+0x4e/0x68 [<ffffffff81077515>] lockdep_rcu_suspicious+0xfa/0x103 [<ffffffff81228b1b>] rhashtable_destroy+0x46/0x52 [<ffffffffa06f21a7>] nft_hash_destroy+0x73/0x82 [nft_hash] Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Acked-by: Thomas Graf <tgraf@suug.ch>
2014-09-01 22:26:05 +00:00
* Frees the bucket array. This function is not rcu safe, therefore the caller
* has to make sure that no resizing may happen by unpublishing the hashtable
* and waiting for the quiescent cycle before releasing the bucket array.
*/
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
void rhashtable_destroy(struct rhashtable *ht)
{
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
ht->being_destroyed = true;
cancel_work_sync(&ht->run_work);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
mutex_lock(&ht->mutex);
rhashtable: Per bucket locks & deferred expansion/shrinking Introduces an array of spinlocks to protect bucket mutations. The number of spinlocks per CPU is configurable and selected based on the hash of the bucket. This allows for parallel insertions and removals of entries which do not share a lock. The patch also defers expansion and shrinking to a worker queue which allows insertion and removal from atomic context. Insertions and deletions may occur in parallel to it and are only held up briefly while the particular bucket is linked or unzipped. Mutations of the bucket table pointer is protected by a new mutex, read access is RCU protected. In the event of an expansion or shrinking, the new bucket table allocated is exposed as a so called future table as soon as the resize process starts. Lookups, deletions, and insertions will briefly use both tables. The future table becomes the main table after an RCU grace period and initial linking of the old to the new table was performed. Optimization of the chains to make use of the new number of buckets follows only the new table is in use. The side effect of this is that during that RCU grace period, a bucket traversal using any rht_for_each() variant on the main table will not see any insertions performed during the RCU grace period which would at that point land in the future table. The lookup will see them as it searches both tables if needed. Having multiple insertions and removals occur in parallel requires nelems to become an atomic counter. Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-02 22:00:20 +00:00
bucket_table_free(rht_dereference(ht->tbl, ht));
mutex_unlock(&ht->mutex);
}
EXPORT_SYMBOL_GPL(rhashtable_destroy);