linux/lib/rhashtable.c

1236 lines
29 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
* 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
*/
#include <linux/atomic.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/log2.h>
#include <linux/sched.h>
#include <linux/rculist.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/mm.h>
2014-12-10 16:33:11 +01:00
#include <linux/jhash.h>
#include <linux/random.h>
#include <linux/rhashtable.h>
#include <linux/err.h>
#include <linux/export.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 23:00:20 +01:00
union nested_table {
union nested_table __rcu *table;
struct rhash_lock_head *bucket;
};
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)
{
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
if (!debug_locks)
return 1;
if (unlikely(tbl->nest))
return 1;
return bit_spin_is_locked(0, (unsigned long *)&tbl->buckets[hash]);
}
EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
#else
#define ASSERT_RHT_MUTEX(HT)
#endif
static void nested_table_free(union nested_table *ntbl, unsigned int size)
{
const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
const unsigned int len = 1 << shift;
unsigned int i;
ntbl = rcu_dereference_raw(ntbl->table);
if (!ntbl)
return;
if (size > len) {
size >>= shift;
for (i = 0; i < len; i++)
nested_table_free(ntbl + i, size);
}
kfree(ntbl);
}
static void nested_bucket_table_free(const struct bucket_table *tbl)
{
unsigned int size = tbl->size >> tbl->nest;
unsigned int len = 1 << tbl->nest;
union nested_table *ntbl;
unsigned int i;
ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
for (i = 0; i < len; i++)
nested_table_free(ntbl + i, size);
kfree(ntbl);
}
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 23:00:20 +01:00
static void bucket_table_free(const struct bucket_table *tbl)
{
if (tbl->nest)
nested_bucket_table_free(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 23:00:20 +01:00
kvfree(tbl);
}
static void bucket_table_free_rcu(struct rcu_head *head)
{
bucket_table_free(container_of(head, struct bucket_table, rcu));
}
static union nested_table *nested_table_alloc(struct rhashtable *ht,
union nested_table __rcu **prev,
bool leaf)
{
union nested_table *ntbl;
int i;
ntbl = rcu_dereference(*prev);
if (ntbl)
return ntbl;
ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC);
if (ntbl && leaf) {
for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0]); i++)
INIT_RHT_NULLS_HEAD(ntbl[i].bucket);
}
if (cmpxchg((union nested_table **)prev, NULL, ntbl) == NULL)
return ntbl;
/* Raced with another thread. */
kfree(ntbl);
return rcu_dereference(*prev);
}
static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht,
size_t nbuckets,
gfp_t gfp)
{
const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
struct bucket_table *tbl;
size_t size;
if (nbuckets < (1 << (shift + 1)))
return NULL;
size = sizeof(*tbl) + sizeof(tbl->buckets[0]);
tbl = kzalloc(size, gfp);
if (!tbl)
return NULL;
if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets,
false)) {
kfree(tbl);
return NULL;
}
tbl->nest = (ilog2(nbuckets) - 1) % shift + 1;
return 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 23:00:20 +01:00
static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
size_t nbuckets,
gfp_t gfp)
{
struct bucket_table *tbl = NULL;
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
size_t size;
int i;
static struct lock_class_key __key;
tbl = kvzalloc(struct_size(tbl, buckets, nbuckets), gfp);
size = nbuckets;
if (tbl == NULL && (gfp & ~__GFP_NOFAIL) != GFP_KERNEL) {
tbl = nested_bucket_table_alloc(ht, nbuckets, gfp);
nbuckets = 0;
}
if (tbl == NULL)
return NULL;
lockdep_init_map(&tbl->dep_map, "rhashtable_bucket", &__key, 0);
tbl->size = size;
rcu_head_init(&tbl->rcu);
INIT_LIST_HEAD(&tbl->walkers);
tbl->hash_rnd = get_random_u32();
for (i = 0; i < nbuckets; i++)
INIT_RHT_NULLS_HEAD(tbl->buckets[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 23:00:20 +01:00
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;
}
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
static int rhashtable_rehash_one(struct rhashtable *ht,
struct rhash_lock_head **bkt,
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
unsigned int old_hash)
{
struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
struct bucket_table *new_tbl = rhashtable_last_table(ht, old_tbl);
int err = -EAGAIN;
struct rhash_head *head, *next, *entry;
struct rhash_head __rcu **pprev = NULL;
unsigned int new_hash;
if (new_tbl->nest)
goto out;
err = -ENOENT;
rht_for_each_from(entry, rht_ptr(bkt, old_tbl, old_hash),
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 23:00:20 +01:00
new_hash = head_hashfn(ht, new_tbl, entry);
rht_lock_nested(new_tbl, &new_tbl->buckets[new_hash], SINGLE_DEPTH_NESTING);
head = rht_ptr(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 23:00:20 +01:00
RCU_INIT_POINTER(entry->next, head);
rht_assign_unlock(new_tbl, &new_tbl->buckets[new_hash], entry);
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 23:00:20 +01:00
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
if (pprev)
rcu_assign_pointer(*pprev, next);
else
/* Need to preserved the bit lock. */
rht_assign_locked(bkt, 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 23:00:20 +01:00
static int rhashtable_rehash_chain(struct rhashtable *ht,
unsigned int old_hash)
{
struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
struct rhash_lock_head **bkt = rht_bucket_var(old_tbl, old_hash);
int err;
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
if (!bkt)
return 0;
rht_lock(old_tbl, bkt);
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
while (!(err = rhashtable_rehash_one(ht, bkt, old_hash)))
;
if (err == -ENOENT)
err = 0;
rht_unlock(old_tbl, bkt);
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 23:00:20 +01:00
}
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 23:00:20 +01:00
{
/* Make insertions go into the new, empty table right away. Deletions
* and lookups will be attempted in both tables until we synchronize.
* As cmpxchg() provides strong barriers, we do not need
* rcu_assign_pointer().
*/
if (cmpxchg((struct bucket_table **)&old_tbl->future_tbl, NULL,
new_tbl) != NULL)
return -EEXIST;
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 int old_hash;
int err;
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++) {
err = rhashtable_rehash_chain(ht, old_hash);
if (err)
return err;
cond_resched();
}
/* 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;
/* Wait for readers. All new readers will see the new
* table, and thus no references to the old table will
* remain.
* We do this inside the locked region so that
* rhashtable_walk_stop() can use rcu_head_after_call_rcu()
* to check if it should not re-link the table.
*/
call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
spin_unlock(&ht->lock);
return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
}
static int rhashtable_rehash_alloc(struct rhashtable *ht,
struct bucket_table *old_tbl,
unsigned int size)
{
struct bucket_table *new_tbl;
int err;
ASSERT_RHT_MUTEX(ht);
new_tbl = bucket_table_alloc(ht, size, 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 23:00:20 +01:00
* 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 23:00:20 +01:00
*
* 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 *old_tbl = rht_dereference(ht->tbl, ht);
rhashtable: fix shift by 64 when shrinking I got this: ================================================================================ UBSAN: Undefined behaviour in ./include/linux/log2.h:63:13 shift exponent 64 is too large for 64-bit type 'long unsigned int' CPU: 1 PID: 721 Comm: kworker/1:1 Not tainted 4.8.0-rc1+ #87 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.3-0-ge2fc41e-prebuilt.qemu-project.org 04/01/2014 Workqueue: events rht_deferred_worker 0000000000000000 ffff88011661f8d8 ffffffff82344f50 0000000041b58ab3 ffffffff84f98000 ffffffff82344ea4 ffff88011661f900 ffff88011661f8b0 0000000000000001 ffff88011661f6b8 dffffc0000000000 ffffffff867f7640 Call Trace: [<ffffffff82344f50>] dump_stack+0xac/0xfc [<ffffffff82344ea4>] ? _atomic_dec_and_lock+0xc4/0xc4 [<ffffffff8242f5b8>] ubsan_epilogue+0xd/0x8a [<ffffffff82430c41>] __ubsan_handle_shift_out_of_bounds+0x255/0x29a [<ffffffff824309ec>] ? __ubsan_handle_out_of_bounds+0x180/0x180 [<ffffffff84003436>] ? nl80211_req_set_reg+0x256/0x2f0 [<ffffffff812112ba>] ? print_context_stack+0x8a/0x160 [<ffffffff81200031>] ? amd_pmu_reset+0x341/0x380 [<ffffffff823af808>] rht_deferred_worker+0x1618/0x1790 [<ffffffff823af808>] ? rht_deferred_worker+0x1618/0x1790 [<ffffffff823ae1f0>] ? rhashtable_jhash2+0x370/0x370 [<ffffffff8134c12d>] ? process_one_work+0x6fd/0x1970 [<ffffffff8134c1cf>] process_one_work+0x79f/0x1970 [<ffffffff8134c12d>] ? process_one_work+0x6fd/0x1970 [<ffffffff8134ba30>] ? try_to_grab_pending+0x4c0/0x4c0 [<ffffffff8134d564>] ? worker_thread+0x1c4/0x1340 [<ffffffff8134d8ff>] worker_thread+0x55f/0x1340 [<ffffffff845e904f>] ? __schedule+0x4df/0x1d40 [<ffffffff8134d3a0>] ? process_one_work+0x1970/0x1970 [<ffffffff8134d3a0>] ? process_one_work+0x1970/0x1970 [<ffffffff813642f7>] kthread+0x237/0x390 [<ffffffff813640c0>] ? __kthread_parkme+0x280/0x280 [<ffffffff845f8c93>] ? _raw_spin_unlock_irq+0x33/0x50 [<ffffffff845f95df>] ret_from_fork+0x1f/0x40 [<ffffffff813640c0>] ? __kthread_parkme+0x280/0x280 ================================================================================ roundup_pow_of_two() is undefined when called with an argument of 0, so let's avoid the call and just fall back to ht->p.min_size (which should never be smaller than HASH_MIN_SIZE). Cc: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: Vegard Nossum <vegard.nossum@oracle.com> Acked-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-12 20:10:44 +02:00
unsigned int nelems = atomic_read(&ht->nelems);
unsigned int size = 0;
rhashtable: fix shift by 64 when shrinking I got this: ================================================================================ UBSAN: Undefined behaviour in ./include/linux/log2.h:63:13 shift exponent 64 is too large for 64-bit type 'long unsigned int' CPU: 1 PID: 721 Comm: kworker/1:1 Not tainted 4.8.0-rc1+ #87 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.3-0-ge2fc41e-prebuilt.qemu-project.org 04/01/2014 Workqueue: events rht_deferred_worker 0000000000000000 ffff88011661f8d8 ffffffff82344f50 0000000041b58ab3 ffffffff84f98000 ffffffff82344ea4 ffff88011661f900 ffff88011661f8b0 0000000000000001 ffff88011661f6b8 dffffc0000000000 ffffffff867f7640 Call Trace: [<ffffffff82344f50>] dump_stack+0xac/0xfc [<ffffffff82344ea4>] ? _atomic_dec_and_lock+0xc4/0xc4 [<ffffffff8242f5b8>] ubsan_epilogue+0xd/0x8a [<ffffffff82430c41>] __ubsan_handle_shift_out_of_bounds+0x255/0x29a [<ffffffff824309ec>] ? __ubsan_handle_out_of_bounds+0x180/0x180 [<ffffffff84003436>] ? nl80211_req_set_reg+0x256/0x2f0 [<ffffffff812112ba>] ? print_context_stack+0x8a/0x160 [<ffffffff81200031>] ? amd_pmu_reset+0x341/0x380 [<ffffffff823af808>] rht_deferred_worker+0x1618/0x1790 [<ffffffff823af808>] ? rht_deferred_worker+0x1618/0x1790 [<ffffffff823ae1f0>] ? rhashtable_jhash2+0x370/0x370 [<ffffffff8134c12d>] ? process_one_work+0x6fd/0x1970 [<ffffffff8134c1cf>] process_one_work+0x79f/0x1970 [<ffffffff8134c12d>] ? process_one_work+0x6fd/0x1970 [<ffffffff8134ba30>] ? try_to_grab_pending+0x4c0/0x4c0 [<ffffffff8134d564>] ? worker_thread+0x1c4/0x1340 [<ffffffff8134d8ff>] worker_thread+0x55f/0x1340 [<ffffffff845e904f>] ? __schedule+0x4df/0x1d40 [<ffffffff8134d3a0>] ? process_one_work+0x1970/0x1970 [<ffffffff8134d3a0>] ? process_one_work+0x1970/0x1970 [<ffffffff813642f7>] kthread+0x237/0x390 [<ffffffff813640c0>] ? __kthread_parkme+0x280/0x280 [<ffffffff845f8c93>] ? _raw_spin_unlock_irq+0x33/0x50 [<ffffffff845f95df>] ret_from_fork+0x1f/0x40 [<ffffffff813640c0>] ? __kthread_parkme+0x280/0x280 ================================================================================ roundup_pow_of_two() is undefined when called with an argument of 0, so let's avoid the call and just fall back to ht->p.min_size (which should never be smaller than HASH_MIN_SIZE). Cc: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: Vegard Nossum <vegard.nossum@oracle.com> Acked-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-12 20:10:44 +02:00
if (nelems)
size = roundup_pow_of_two(nelems * 3 / 2);
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;
return rhashtable_rehash_alloc(ht, old_tbl, size);
}
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 23:00:20 +01:00
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 23:00:20 +01:00
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 23:00:20 +01:00
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 23:00:20 +01:00
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 23:00:20 +01:00
if (rht_grow_above_75(ht, tbl))
err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2);
else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
err = rhashtable_shrink(ht);
else if (tbl->nest)
err = rhashtable_rehash_alloc(ht, tbl, tbl->size);
if (!err || err == -EEXIST) {
int nerr;
nerr = rhashtable_rehash_table(ht);
err = err ?: nerr;
}
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 23:00:20 +01:00
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 23:00:20 +01:00
}
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
static int rhashtable_insert_rehash(struct rhashtable *ht,
struct bucket_table *tbl)
{
struct bucket_table *old_tbl;
struct bucket_table *new_tbl;
unsigned int size;
int err;
old_tbl = rht_dereference_rcu(ht->tbl, ht);
size = tbl->size;
err = -EBUSY;
if (rht_grow_above_75(ht, tbl))
size *= 2;
/* Do not schedule more than one rehash */
else if (old_tbl != tbl)
goto fail;
err = -ENOMEM;
new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC | __GFP_NOWARN);
if (new_tbl == NULL)
goto fail;
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;
fail:
/* Do not fail the insert if someone else did a rehash. */
if (likely(rcu_access_pointer(tbl->future_tbl)))
return 0;
/* Schedule async rehash to retry allocation in process context. */
if (err == -ENOMEM)
schedule_work(&ht->run_work);
return err;
}
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
static void *rhashtable_lookup_one(struct rhashtable *ht,
struct rhash_lock_head **bkt,
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
struct bucket_table *tbl, unsigned int hash,
const void *key, struct rhash_head *obj)
{
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
struct rhashtable_compare_arg arg = {
.ht = ht,
.key = key,
};
struct rhash_head __rcu **pprev = NULL;
struct rhash_head *head;
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
int elasticity;
elasticity = RHT_ELASTICITY;
rht_for_each_from(head, rht_ptr(bkt, tbl, hash), tbl, hash) {
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
struct rhlist_head *list;
struct rhlist_head *plist;
elasticity--;
if (!key ||
(ht->p.obj_cmpfn ?
ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) :
rhashtable_compare(&arg, rht_obj(ht, head)))) {
pprev = &head->next;
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
continue;
}
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
if (!ht->rhlist)
return rht_obj(ht, head);
list = container_of(obj, struct rhlist_head, rhead);
plist = container_of(head, struct rhlist_head, rhead);
RCU_INIT_POINTER(list->next, plist);
head = rht_dereference_bucket(head->next, tbl, hash);
RCU_INIT_POINTER(list->rhead.next, head);
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
if (pprev)
rcu_assign_pointer(*pprev, obj);
else
/* Need to preserve the bit lock */
rht_assign_locked(bkt, obj);
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
return NULL;
}
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
if (elasticity <= 0)
return ERR_PTR(-EAGAIN);
return ERR_PTR(-ENOENT);
}
static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht,
struct rhash_lock_head **bkt,
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
struct bucket_table *tbl,
unsigned int hash,
struct rhash_head *obj,
void *data)
{
struct bucket_table *new_tbl;
struct rhash_head *head;
if (!IS_ERR_OR_NULL(data))
return ERR_PTR(-EEXIST);
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT)
return ERR_CAST(data);
new_tbl = rht_dereference_rcu(tbl->future_tbl, ht);
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
if (new_tbl)
return new_tbl;
if (PTR_ERR(data) != -ENOENT)
return ERR_CAST(data);
if (unlikely(rht_grow_above_max(ht, tbl)))
return ERR_PTR(-E2BIG);
if (unlikely(rht_grow_above_100(ht, tbl)))
return ERR_PTR(-EAGAIN);
head = rht_ptr(bkt, tbl, hash);
RCU_INIT_POINTER(obj->next, head);
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
if (ht->rhlist) {
struct rhlist_head *list;
list = container_of(obj, struct rhlist_head, rhead);
RCU_INIT_POINTER(list->next, NULL);
}
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
/* bkt is always the head of the list, so it holds
* the lock, which we need to preserve
*/
rht_assign_locked(bkt, obj);
atomic_inc(&ht->nelems);
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
if (rht_grow_above_75(ht, tbl))
schedule_work(&ht->run_work);
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
return NULL;
}
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
static void *rhashtable_try_insert(struct rhashtable *ht, const void *key,
struct rhash_head *obj)
{
struct bucket_table *new_tbl;
struct bucket_table *tbl;
struct rhash_lock_head **bkt;
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
unsigned int hash;
void *data;
new_tbl = rcu_dereference(ht->tbl);
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
do {
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
tbl = new_tbl;
hash = rht_head_hashfn(ht, tbl, obj, ht->p);
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
if (rcu_access_pointer(tbl->future_tbl))
/* Failure is OK */
bkt = rht_bucket_var(tbl, hash);
else
bkt = rht_bucket_insert(ht, tbl, hash);
if (bkt == NULL) {
new_tbl = rht_dereference_rcu(tbl->future_tbl, ht);
data = ERR_PTR(-EAGAIN);
} else {
rht_lock(tbl, bkt);
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
data = rhashtable_lookup_one(ht, bkt, tbl,
hash, key, obj);
new_tbl = rhashtable_insert_one(ht, bkt, tbl,
hash, obj, data);
if (PTR_ERR(new_tbl) != -EEXIST)
data = ERR_CAST(new_tbl);
rht_unlock(tbl, bkt);
rhashtable: use bit_spin_locks to protect hash bucket. This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the bucket pointer to lock the hash chain for that bucket. The benefits of a bit spin_lock are: - no need to allocate a separate array of locks. - no need to have a configuration option to guide the choice of the size of this array - locking cost is often a single test-and-set in a cache line that will have to be loaded anyway. When inserting at, or removing from, the head of the chain, the unlock is free - writing the new address in the bucket head implicitly clears the lock bit. For __rhashtable_insert_fast() we ensure this always happens when adding a new key. - even when lockings costs 2 updates (lock and unlock), they are in a cacheline that needs to be read anyway. The cost of using a bit spin_lock is a little bit of code complexity, which I think is quite manageable. Bit spin_locks are sometimes inappropriate because they are not fair - if multiple CPUs repeatedly contend of the same lock, one CPU can easily be starved. This is not a credible situation with rhashtable. Multiple CPUs may want to repeatedly add or remove objects, but they will typically do so at different buckets, so they will attempt to acquire different locks. As we have more bit-locks than we previously had spinlocks (by at least a factor of two) we can expect slightly less contention to go with the slightly better cache behavior and reduced memory consumption. To enhance type checking, a new struct is introduced to represent the pointer plus lock-bit that is stored in the bucket-table. This is "struct rhash_lock_head" and is empty. A pointer to this needs to be cast to either an unsigned lock, or a "struct rhash_head *" to be useful. Variables of this type are most often called "bkt". Previously "pprev" would sometimes point to a bucket, and sometimes a ->next pointer in an rhash_head. As these are now different types, pprev is NULL when it would have pointed to the bucket. In that case, 'blk' is used, together with correct locking protocol. Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-02 01:07:45 +02:00
}
} while (!IS_ERR_OR_NULL(new_tbl));
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
if (PTR_ERR(data) == -EAGAIN)
data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?:
-EAGAIN);
return data;
}
void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
struct rhash_head *obj)
{
void *data;
do {
rcu_read_lock();
data = rhashtable_try_insert(ht, key, obj);
rcu_read_unlock();
} while (PTR_ERR(data) == -EAGAIN);
return data;
}
EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
/**
* rhashtable_walk_enter - 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 be called from any process context, including
* non-preemptable context, but cannot be called from softirq or
* hardirq context.
*
* You must call rhashtable_walk_exit after this function returns.
*/
void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter)
{
iter->ht = ht;
iter->p = NULL;
iter->slot = 0;
iter->skip = 0;
iter->end_of_table = 0;
spin_lock(&ht->lock);
iter->walker.tbl =
rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock));
list_add(&iter->walker.list, &iter->walker.tbl->walkers);
spin_unlock(&ht->lock);
}
EXPORT_SYMBOL_GPL(rhashtable_walk_enter);
/**
* rhashtable_walk_exit - Free an iterator
* @iter: Hash table Iterator
*
* This function frees resources allocated by rhashtable_walk_enter.
*/
void rhashtable_walk_exit(struct rhashtable_iter *iter)
{
spin_lock(&iter->ht->lock);
if (iter->walker.tbl)
list_del(&iter->walker.list);
spin_unlock(&iter->ht->lock);
}
EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
/**
* rhashtable_walk_start_check - Start a hash table walk
* @iter: Hash table iterator
*
* Start a hash table walk at the current iterator position. 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.
*
* rhashtable_walk_start is defined as an inline variant that returns
* void. This is preferred in cases where the caller would ignore
* resize events and always continue.
*/
int rhashtable_walk_start_check(struct rhashtable_iter *iter)
__acquires(RCU)
{
struct rhashtable *ht = iter->ht;
bool rhlist = ht->rhlist;
rcu_read_lock();
spin_lock(&ht->lock);
if (iter->walker.tbl)
list_del(&iter->walker.list);
spin_unlock(&ht->lock);
if (iter->end_of_table)
return 0;
if (!iter->walker.tbl) {
iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht);
iter->slot = 0;
iter->skip = 0;
return -EAGAIN;
}
if (iter->p && !rhlist) {
/*
* We need to validate that 'p' is still in the table, and
* if so, update 'skip'
*/
struct rhash_head *p;
int skip = 0;
rht_for_each_rcu(p, iter->walker.tbl, iter->slot) {
skip++;
if (p == iter->p) {
iter->skip = skip;
goto found;
}
}
iter->p = NULL;
} else if (iter->p && rhlist) {
/* Need to validate that 'list' is still in the table, and
* if so, update 'skip' and 'p'.
*/
struct rhash_head *p;
struct rhlist_head *list;
int skip = 0;
rht_for_each_rcu(p, iter->walker.tbl, iter->slot) {
for (list = container_of(p, struct rhlist_head, rhead);
list;
list = rcu_dereference(list->next)) {
skip++;
if (list == iter->list) {
iter->p = p;
iter->skip = skip;
goto found;
}
}
}
iter->p = NULL;
}
found:
return 0;
}
EXPORT_SYMBOL_GPL(rhashtable_walk_start_check);
/**
* __rhashtable_walk_find_next - Find the next element in a table (or the first
* one in case of a new walk).
*
* @iter: Hash table iterator
*
* Returns the found object or NULL when the end of the table is reached.
*
* Returns -EAGAIN if resize event occurred.
*/
static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter)
{
struct bucket_table *tbl = iter->walker.tbl;
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
struct rhlist_head *list = iter->list;
struct rhashtable *ht = iter->ht;
struct rhash_head *p = iter->p;
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
bool rhlist = ht->rhlist;
if (!tbl)
return NULL;
for (; iter->slot < tbl->size; iter->slot++) {
int skip = iter->skip;
rht_for_each_rcu(p, tbl, iter->slot) {
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
if (rhlist) {
list = container_of(p, struct rhlist_head,
rhead);
do {
if (!skip)
goto next;
skip--;
list = rcu_dereference(list->next);
} while (list);
continue;
}
if (!skip)
break;
skip--;
}
next:
if (!rht_is_a_nulls(p)) {
iter->skip++;
iter->p = p;
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
iter->list = list;
return rht_obj(ht, rhlist ? &list->rhead : p);
}
iter->skip = 0;
}
iter->p = NULL;
/* 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);
} else {
iter->end_of_table = true;
}
return NULL;
}
/**
* 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 occurred. 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 rhlist_head *list = iter->list;
struct rhashtable *ht = iter->ht;
struct rhash_head *p = iter->p;
bool rhlist = ht->rhlist;
if (p) {
if (!rhlist || !(list = rcu_dereference(list->next))) {
p = rcu_dereference(p->next);
list = container_of(p, struct rhlist_head, rhead);
}
if (!rht_is_a_nulls(p)) {
iter->skip++;
iter->p = p;
iter->list = list;
return rht_obj(ht, rhlist ? &list->rhead : p);
}
/* At the end of this slot, switch to next one and then find
* next entry from that point.
*/
iter->skip = 0;
iter->slot++;
}
return __rhashtable_walk_find_next(iter);
}
EXPORT_SYMBOL_GPL(rhashtable_walk_next);
/**
* rhashtable_walk_peek - Return the next object but don't advance the iterator
* @iter: Hash table iterator
*
* Returns the next object or NULL when the end of the table is reached.
*
* Returns -EAGAIN if resize event occurred. Note that the iterator
* will rewind back to the beginning and you may continue to use it.
*/
void *rhashtable_walk_peek(struct rhashtable_iter *iter)
{
struct rhlist_head *list = iter->list;
struct rhashtable *ht = iter->ht;
struct rhash_head *p = iter->p;
if (p)
return rht_obj(ht, ht->rhlist ? &list->rhead : p);
/* No object found in current iter, find next one in the table. */
if (iter->skip) {
/* A nonzero skip value points to the next entry in the table
* beyond that last one that was found. Decrement skip so
* we find the current value. __rhashtable_walk_find_next
* will restore the original value of skip assuming that
* the table hasn't changed.
*/
iter->skip--;
}
return __rhashtable_walk_find_next(iter);
}
EXPORT_SYMBOL_GPL(rhashtable_walk_peek);
/**
* rhashtable_walk_stop - Finish a hash table walk
* @iter: Hash table iterator
*
* Finish a hash table walk. Does not reset the iterator to the start of the
* hash table.
*/
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 (rcu_head_after_call_rcu(&tbl->rcu, bucket_table_free_rcu))
/* This bucket table is being freed, don't re-link it. */
iter->walker.tbl = NULL;
else
list_add(&iter->walker.list, &tbl->walkers);
spin_unlock(&ht->lock);
out:
rcu_read_unlock();
}
EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
static size_t rounded_hashtable_size(const struct rhashtable_params *params)
{
size_t retsize;
if (params->nelem_hint)
retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
(unsigned long)params->min_size);
else
retsize = max(HASH_DEFAULT_SIZE,
(unsigned long)params->min_size);
return retsize;
}
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 16:33:11 +01:00
* .hashfn = jhash,
* };
*
* Configuration Example 2: Variable length keys
* struct test_obj {
* [...]
* struct rhash_head node;
* };
*
* u32 my_hash_fn(const void *data, u32 len, u32 seed)
* {
* struct test_obj *obj = data;
*
* return [... hash ...];
* }
*
* struct rhashtable_params params = {
* .head_offset = offsetof(struct test_obj, node),
2014-12-10 16:33:11 +01:00
* .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;
if ((!params->key_len && !params->obj_hashfn) ||
(params->obj_hashfn && !params->obj_cmpfn))
return -EINVAL;
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 23:00:20 +01:00
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 23:00:20 +01:00
memcpy(&ht->p, params, sizeof(*params));
if (params->min_size)
ht->p.min_size = roundup_pow_of_two(params->min_size);
/* Cap total entries at 2^31 to avoid nelems overflow. */
ht->max_elems = 1u << 31;
if (params->max_size) {
ht->p.max_size = rounddown_pow_of_two(params->max_size);
if (ht->p.max_size < ht->max_elems / 2)
ht->max_elems = ht->p.max_size * 2;
}
ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE);
size = rounded_hashtable_size(&ht->p);
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;
}
}
/*
* This is api initialization and thus we need to guarantee the
* initial rhashtable allocation. Upon failure, retry with the
* smallest possible size with __GFP_NOFAIL semantics.
*/
tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
if (unlikely(tbl == NULL)) {
size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE);
tbl = bucket_table_alloc(ht, size, GFP_KERNEL | __GFP_NOFAIL);
}
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 23:00:20 +01:00
return 0;
}
EXPORT_SYMBOL_GPL(rhashtable_init);
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
/**
* rhltable_init - initialize a new hash list table
* @hlt: hash list table to be initialized
* @params: configuration parameters
*
* Initializes a new hash list table.
*
* See documentation for rhashtable_init.
*/
int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params)
{
int err;
err = rhashtable_init(&hlt->ht, params);
hlt->ht.rhlist = true;
return err;
}
EXPORT_SYMBOL_GPL(rhltable_init);
static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj,
void (*free_fn)(void *ptr, void *arg),
void *arg)
{
struct rhlist_head *list;
if (!ht->rhlist) {
free_fn(rht_obj(ht, obj), arg);
return;
}
list = container_of(obj, struct rhlist_head, rhead);
do {
obj = &list->rhead;
list = rht_dereference(list->next, ht);
free_fn(rht_obj(ht, obj), arg);
} while (list);
}
/**
* rhashtable_free_and_destroy - free elements and destroy hash table
* @ht: the hash table to destroy
* @free_fn: callback to release resources of element
* @arg: pointer passed to free_fn
*
* Stops an eventual async resize. If defined, invokes free_fn for each
* element to releasal resources. Please note that RCU protected
* readers may still be accessing the elements. Releasing of resources
* must occur in a compatible manner. Then frees the bucket array.
*
* This function will eventually sleep to wait for an async resize
* to complete. The caller is responsible that no further write operations
* occurs in parallel.
*/
void rhashtable_free_and_destroy(struct rhashtable *ht,
void (*free_fn)(void *ptr, void *arg),
void *arg)
{
rhashtable: add restart routine in rhashtable_free_and_destroy() rhashtable_free_and_destroy() cancels re-hash deferred work then walks and destroys elements. at this moment, some elements can be still in future_tbl. that elements are not destroyed. test case: nft_rhash_destroy() calls rhashtable_free_and_destroy() to destroy all elements of sets before destroying sets and chains. But rhashtable_free_and_destroy() doesn't destroy elements of future_tbl. so that splat occurred. test script: %cat test.nft table ip aa { map map1 { type ipv4_addr : verdict; elements = { 0 : jump a0, 1 : jump a0, 2 : jump a0, 3 : jump a0, 4 : jump a0, 5 : jump a0, 6 : jump a0, 7 : jump a0, 8 : jump a0, 9 : jump a0, } } chain a0 { } } flush ruleset table ip aa { map map1 { type ipv4_addr : verdict; elements = { 0 : jump a0, 1 : jump a0, 2 : jump a0, 3 : jump a0, 4 : jump a0, 5 : jump a0, 6 : jump a0, 7 : jump a0, 8 : jump a0, 9 : jump a0, } } chain a0 { } } flush ruleset %while :; do nft -f test.nft; done Splat looks like: [ 200.795603] kernel BUG at net/netfilter/nf_tables_api.c:1363! [ 200.806944] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN PTI [ 200.812253] CPU: 1 PID: 1582 Comm: nft Not tainted 4.17.0+ #24 [ 200.820297] Hardware name: To be filled by O.E.M. To be filled by O.E.M./Aptio CRB, BIOS 5.6.5 07/08/2015 [ 200.830309] RIP: 0010:nf_tables_chain_destroy.isra.34+0x62/0x240 [nf_tables] [ 200.838317] Code: 43 50 85 c0 74 26 48 8b 45 00 48 8b 4d 08 ba 54 05 00 00 48 c7 c6 60 6d 29 c0 48 c7 c7 c0 65 29 c0 4c 8b 40 08 e8 58 e5 fd f8 <0f> 0b 48 89 da 48 b8 00 00 00 00 00 fc ff [ 200.860366] RSP: 0000:ffff880118dbf4d0 EFLAGS: 00010282 [ 200.866354] RAX: 0000000000000061 RBX: ffff88010cdeaf08 RCX: 0000000000000000 [ 200.874355] RDX: 0000000000000061 RSI: 0000000000000008 RDI: ffffed00231b7e90 [ 200.882361] RBP: ffff880118dbf4e8 R08: ffffed002373bcfb R09: ffffed002373bcfa [ 200.890354] R10: 0000000000000000 R11: ffffed002373bcfb R12: dead000000000200 [ 200.898356] R13: dead000000000100 R14: ffffffffbb62af38 R15: dffffc0000000000 [ 200.906354] FS: 00007fefc31fd700(0000) GS:ffff88011b800000(0000) knlGS:0000000000000000 [ 200.915533] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 200.922355] CR2: 0000557f1c8e9128 CR3: 0000000106880000 CR4: 00000000001006e0 [ 200.930353] Call Trace: [ 200.932351] ? nf_tables_commit+0x26f6/0x2c60 [nf_tables] [ 200.939525] ? nf_tables_setelem_notify.constprop.49+0x1a0/0x1a0 [nf_tables] [ 200.947525] ? nf_tables_delchain+0x6e0/0x6e0 [nf_tables] [ 200.952383] ? nft_add_set_elem+0x1700/0x1700 [nf_tables] [ 200.959532] ? nla_parse+0xab/0x230 [ 200.963529] ? nfnetlink_rcv_batch+0xd06/0x10d0 [nfnetlink] [ 200.968384] ? nfnetlink_net_init+0x130/0x130 [nfnetlink] [ 200.975525] ? debug_show_all_locks+0x290/0x290 [ 200.980363] ? debug_show_all_locks+0x290/0x290 [ 200.986356] ? sched_clock_cpu+0x132/0x170 [ 200.990352] ? find_held_lock+0x39/0x1b0 [ 200.994355] ? sched_clock_local+0x10d/0x130 [ 200.999531] ? memset+0x1f/0x40 V2: - free all tables requested by Herbert Xu Signed-off-by: Taehee Yoo <ap420073@gmail.com> Acked-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-08 04:55:51 +02:00
struct bucket_table *tbl, *next_tbl;
unsigned int 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 23:00:20 +01:00
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 23:00:20 +01:00
mutex_lock(&ht->mutex);
tbl = rht_dereference(ht->tbl, ht);
rhashtable: add restart routine in rhashtable_free_and_destroy() rhashtable_free_and_destroy() cancels re-hash deferred work then walks and destroys elements. at this moment, some elements can be still in future_tbl. that elements are not destroyed. test case: nft_rhash_destroy() calls rhashtable_free_and_destroy() to destroy all elements of sets before destroying sets and chains. But rhashtable_free_and_destroy() doesn't destroy elements of future_tbl. so that splat occurred. test script: %cat test.nft table ip aa { map map1 { type ipv4_addr : verdict; elements = { 0 : jump a0, 1 : jump a0, 2 : jump a0, 3 : jump a0, 4 : jump a0, 5 : jump a0, 6 : jump a0, 7 : jump a0, 8 : jump a0, 9 : jump a0, } } chain a0 { } } flush ruleset table ip aa { map map1 { type ipv4_addr : verdict; elements = { 0 : jump a0, 1 : jump a0, 2 : jump a0, 3 : jump a0, 4 : jump a0, 5 : jump a0, 6 : jump a0, 7 : jump a0, 8 : jump a0, 9 : jump a0, } } chain a0 { } } flush ruleset %while :; do nft -f test.nft; done Splat looks like: [ 200.795603] kernel BUG at net/netfilter/nf_tables_api.c:1363! [ 200.806944] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN PTI [ 200.812253] CPU: 1 PID: 1582 Comm: nft Not tainted 4.17.0+ #24 [ 200.820297] Hardware name: To be filled by O.E.M. To be filled by O.E.M./Aptio CRB, BIOS 5.6.5 07/08/2015 [ 200.830309] RIP: 0010:nf_tables_chain_destroy.isra.34+0x62/0x240 [nf_tables] [ 200.838317] Code: 43 50 85 c0 74 26 48 8b 45 00 48 8b 4d 08 ba 54 05 00 00 48 c7 c6 60 6d 29 c0 48 c7 c7 c0 65 29 c0 4c 8b 40 08 e8 58 e5 fd f8 <0f> 0b 48 89 da 48 b8 00 00 00 00 00 fc ff [ 200.860366] RSP: 0000:ffff880118dbf4d0 EFLAGS: 00010282 [ 200.866354] RAX: 0000000000000061 RBX: ffff88010cdeaf08 RCX: 0000000000000000 [ 200.874355] RDX: 0000000000000061 RSI: 0000000000000008 RDI: ffffed00231b7e90 [ 200.882361] RBP: ffff880118dbf4e8 R08: ffffed002373bcfb R09: ffffed002373bcfa [ 200.890354] R10: 0000000000000000 R11: ffffed002373bcfb R12: dead000000000200 [ 200.898356] R13: dead000000000100 R14: ffffffffbb62af38 R15: dffffc0000000000 [ 200.906354] FS: 00007fefc31fd700(0000) GS:ffff88011b800000(0000) knlGS:0000000000000000 [ 200.915533] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 200.922355] CR2: 0000557f1c8e9128 CR3: 0000000106880000 CR4: 00000000001006e0 [ 200.930353] Call Trace: [ 200.932351] ? nf_tables_commit+0x26f6/0x2c60 [nf_tables] [ 200.939525] ? nf_tables_setelem_notify.constprop.49+0x1a0/0x1a0 [nf_tables] [ 200.947525] ? nf_tables_delchain+0x6e0/0x6e0 [nf_tables] [ 200.952383] ? nft_add_set_elem+0x1700/0x1700 [nf_tables] [ 200.959532] ? nla_parse+0xab/0x230 [ 200.963529] ? nfnetlink_rcv_batch+0xd06/0x10d0 [nfnetlink] [ 200.968384] ? nfnetlink_net_init+0x130/0x130 [nfnetlink] [ 200.975525] ? debug_show_all_locks+0x290/0x290 [ 200.980363] ? debug_show_all_locks+0x290/0x290 [ 200.986356] ? sched_clock_cpu+0x132/0x170 [ 200.990352] ? find_held_lock+0x39/0x1b0 [ 200.994355] ? sched_clock_local+0x10d/0x130 [ 200.999531] ? memset+0x1f/0x40 V2: - free all tables requested by Herbert Xu Signed-off-by: Taehee Yoo <ap420073@gmail.com> Acked-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-08 04:55:51 +02:00
restart:
if (free_fn) {
for (i = 0; i < tbl->size; i++) {
struct rhash_head *pos, *next;
cond_resched();
for (pos = rht_ptr_exclusive(rht_bucket(tbl, i)),
next = !rht_is_a_nulls(pos) ?
rht_dereference(pos->next, ht) : NULL;
!rht_is_a_nulls(pos);
pos = next,
next = !rht_is_a_nulls(pos) ?
rht_dereference(pos->next, ht) : NULL)
rhashtable: Add rhlist interface The insecure_elasticity setting is an ugly wart brought out by users who need to insert duplicate objects (that is, distinct objects with identical keys) into the same table. In fact, those users have a much bigger problem. Once those duplicate objects are inserted, they don't have an interface to find them (unless you count the walker interface which walks over the entire table). Some users have resorted to doing a manual walk over the hash table which is of course broken because they don't handle the potential existence of multiple hash tables. The result is that they will break sporadically when they encounter a hash table resize/rehash. This patch provides a way out for those users, at the expense of an extra pointer per object. Essentially each object is now a list of objects carrying the same key. The hash table will only see the lists so nothing changes as far as rhashtable is concerned. To use this new interface, you need to insert a struct rhlist_head into your objects instead of struct rhash_head. While the hash table is unchanged, for type-safety you'll need to use struct rhltable instead of struct rhashtable. All the existing interfaces have been duplicated for rhlist, including the hash table walker. One missing feature is nulls marking because AFAIK the only potential user of it does not need duplicate objects. Should anyone need this it shouldn't be too hard to add. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Acked-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-19 13:00:09 +02:00
rhashtable_free_one(ht, pos, free_fn, arg);
}
}
rhashtable: add restart routine in rhashtable_free_and_destroy() rhashtable_free_and_destroy() cancels re-hash deferred work then walks and destroys elements. at this moment, some elements can be still in future_tbl. that elements are not destroyed. test case: nft_rhash_destroy() calls rhashtable_free_and_destroy() to destroy all elements of sets before destroying sets and chains. But rhashtable_free_and_destroy() doesn't destroy elements of future_tbl. so that splat occurred. test script: %cat test.nft table ip aa { map map1 { type ipv4_addr : verdict; elements = { 0 : jump a0, 1 : jump a0, 2 : jump a0, 3 : jump a0, 4 : jump a0, 5 : jump a0, 6 : jump a0, 7 : jump a0, 8 : jump a0, 9 : jump a0, } } chain a0 { } } flush ruleset table ip aa { map map1 { type ipv4_addr : verdict; elements = { 0 : jump a0, 1 : jump a0, 2 : jump a0, 3 : jump a0, 4 : jump a0, 5 : jump a0, 6 : jump a0, 7 : jump a0, 8 : jump a0, 9 : jump a0, } } chain a0 { } } flush ruleset %while :; do nft -f test.nft; done Splat looks like: [ 200.795603] kernel BUG at net/netfilter/nf_tables_api.c:1363! [ 200.806944] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN PTI [ 200.812253] CPU: 1 PID: 1582 Comm: nft Not tainted 4.17.0+ #24 [ 200.820297] Hardware name: To be filled by O.E.M. To be filled by O.E.M./Aptio CRB, BIOS 5.6.5 07/08/2015 [ 200.830309] RIP: 0010:nf_tables_chain_destroy.isra.34+0x62/0x240 [nf_tables] [ 200.838317] Code: 43 50 85 c0 74 26 48 8b 45 00 48 8b 4d 08 ba 54 05 00 00 48 c7 c6 60 6d 29 c0 48 c7 c7 c0 65 29 c0 4c 8b 40 08 e8 58 e5 fd f8 <0f> 0b 48 89 da 48 b8 00 00 00 00 00 fc ff [ 200.860366] RSP: 0000:ffff880118dbf4d0 EFLAGS: 00010282 [ 200.866354] RAX: 0000000000000061 RBX: ffff88010cdeaf08 RCX: 0000000000000000 [ 200.874355] RDX: 0000000000000061 RSI: 0000000000000008 RDI: ffffed00231b7e90 [ 200.882361] RBP: ffff880118dbf4e8 R08: ffffed002373bcfb R09: ffffed002373bcfa [ 200.890354] R10: 0000000000000000 R11: ffffed002373bcfb R12: dead000000000200 [ 200.898356] R13: dead000000000100 R14: ffffffffbb62af38 R15: dffffc0000000000 [ 200.906354] FS: 00007fefc31fd700(0000) GS:ffff88011b800000(0000) knlGS:0000000000000000 [ 200.915533] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 200.922355] CR2: 0000557f1c8e9128 CR3: 0000000106880000 CR4: 00000000001006e0 [ 200.930353] Call Trace: [ 200.932351] ? nf_tables_commit+0x26f6/0x2c60 [nf_tables] [ 200.939525] ? nf_tables_setelem_notify.constprop.49+0x1a0/0x1a0 [nf_tables] [ 200.947525] ? nf_tables_delchain+0x6e0/0x6e0 [nf_tables] [ 200.952383] ? nft_add_set_elem+0x1700/0x1700 [nf_tables] [ 200.959532] ? nla_parse+0xab/0x230 [ 200.963529] ? nfnetlink_rcv_batch+0xd06/0x10d0 [nfnetlink] [ 200.968384] ? nfnetlink_net_init+0x130/0x130 [nfnetlink] [ 200.975525] ? debug_show_all_locks+0x290/0x290 [ 200.980363] ? debug_show_all_locks+0x290/0x290 [ 200.986356] ? sched_clock_cpu+0x132/0x170 [ 200.990352] ? find_held_lock+0x39/0x1b0 [ 200.994355] ? sched_clock_local+0x10d/0x130 [ 200.999531] ? memset+0x1f/0x40 V2: - free all tables requested by Herbert Xu Signed-off-by: Taehee Yoo <ap420073@gmail.com> Acked-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-08 04:55:51 +02:00
next_tbl = rht_dereference(tbl->future_tbl, ht);
bucket_table_free(tbl);
rhashtable: add restart routine in rhashtable_free_and_destroy() rhashtable_free_and_destroy() cancels re-hash deferred work then walks and destroys elements. at this moment, some elements can be still in future_tbl. that elements are not destroyed. test case: nft_rhash_destroy() calls rhashtable_free_and_destroy() to destroy all elements of sets before destroying sets and chains. But rhashtable_free_and_destroy() doesn't destroy elements of future_tbl. so that splat occurred. test script: %cat test.nft table ip aa { map map1 { type ipv4_addr : verdict; elements = { 0 : jump a0, 1 : jump a0, 2 : jump a0, 3 : jump a0, 4 : jump a0, 5 : jump a0, 6 : jump a0, 7 : jump a0, 8 : jump a0, 9 : jump a0, } } chain a0 { } } flush ruleset table ip aa { map map1 { type ipv4_addr : verdict; elements = { 0 : jump a0, 1 : jump a0, 2 : jump a0, 3 : jump a0, 4 : jump a0, 5 : jump a0, 6 : jump a0, 7 : jump a0, 8 : jump a0, 9 : jump a0, } } chain a0 { } } flush ruleset %while :; do nft -f test.nft; done Splat looks like: [ 200.795603] kernel BUG at net/netfilter/nf_tables_api.c:1363! [ 200.806944] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN PTI [ 200.812253] CPU: 1 PID: 1582 Comm: nft Not tainted 4.17.0+ #24 [ 200.820297] Hardware name: To be filled by O.E.M. To be filled by O.E.M./Aptio CRB, BIOS 5.6.5 07/08/2015 [ 200.830309] RIP: 0010:nf_tables_chain_destroy.isra.34+0x62/0x240 [nf_tables] [ 200.838317] Code: 43 50 85 c0 74 26 48 8b 45 00 48 8b 4d 08 ba 54 05 00 00 48 c7 c6 60 6d 29 c0 48 c7 c7 c0 65 29 c0 4c 8b 40 08 e8 58 e5 fd f8 <0f> 0b 48 89 da 48 b8 00 00 00 00 00 fc ff [ 200.860366] RSP: 0000:ffff880118dbf4d0 EFLAGS: 00010282 [ 200.866354] RAX: 0000000000000061 RBX: ffff88010cdeaf08 RCX: 0000000000000000 [ 200.874355] RDX: 0000000000000061 RSI: 0000000000000008 RDI: ffffed00231b7e90 [ 200.882361] RBP: ffff880118dbf4e8 R08: ffffed002373bcfb R09: ffffed002373bcfa [ 200.890354] R10: 0000000000000000 R11: ffffed002373bcfb R12: dead000000000200 [ 200.898356] R13: dead000000000100 R14: ffffffffbb62af38 R15: dffffc0000000000 [ 200.906354] FS: 00007fefc31fd700(0000) GS:ffff88011b800000(0000) knlGS:0000000000000000 [ 200.915533] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 200.922355] CR2: 0000557f1c8e9128 CR3: 0000000106880000 CR4: 00000000001006e0 [ 200.930353] Call Trace: [ 200.932351] ? nf_tables_commit+0x26f6/0x2c60 [nf_tables] [ 200.939525] ? nf_tables_setelem_notify.constprop.49+0x1a0/0x1a0 [nf_tables] [ 200.947525] ? nf_tables_delchain+0x6e0/0x6e0 [nf_tables] [ 200.952383] ? nft_add_set_elem+0x1700/0x1700 [nf_tables] [ 200.959532] ? nla_parse+0xab/0x230 [ 200.963529] ? nfnetlink_rcv_batch+0xd06/0x10d0 [nfnetlink] [ 200.968384] ? nfnetlink_net_init+0x130/0x130 [nfnetlink] [ 200.975525] ? debug_show_all_locks+0x290/0x290 [ 200.980363] ? debug_show_all_locks+0x290/0x290 [ 200.986356] ? sched_clock_cpu+0x132/0x170 [ 200.990352] ? find_held_lock+0x39/0x1b0 [ 200.994355] ? sched_clock_local+0x10d/0x130 [ 200.999531] ? memset+0x1f/0x40 V2: - free all tables requested by Herbert Xu Signed-off-by: Taehee Yoo <ap420073@gmail.com> Acked-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-08 04:55:51 +02:00
if (next_tbl) {
tbl = next_tbl;
goto restart;
}
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 23:00:20 +01:00
mutex_unlock(&ht->mutex);
}
EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
void rhashtable_destroy(struct rhashtable *ht)
{
return rhashtable_free_and_destroy(ht, NULL, NULL);
}
EXPORT_SYMBOL_GPL(rhashtable_destroy);
struct rhash_lock_head **__rht_bucket_nested(const struct bucket_table *tbl,
unsigned int hash)
{
const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
unsigned int index = hash & ((1 << tbl->nest) - 1);
unsigned int size = tbl->size >> tbl->nest;
unsigned int subhash = hash;
union nested_table *ntbl;
ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash);
subhash >>= tbl->nest;
while (ntbl && size > (1 << shift)) {
index = subhash & ((1 << shift) - 1);
ntbl = rht_dereference_bucket_rcu(ntbl[index].table,
tbl, hash);
size >>= shift;
subhash >>= shift;
}
if (!ntbl)
return NULL;
return &ntbl[subhash].bucket;
}
EXPORT_SYMBOL_GPL(__rht_bucket_nested);
struct rhash_lock_head **rht_bucket_nested(const struct bucket_table *tbl,
unsigned int hash)
{
static struct rhash_lock_head *rhnull;
if (!rhnull)
INIT_RHT_NULLS_HEAD(rhnull);
return __rht_bucket_nested(tbl, hash) ?: &rhnull;
}
EXPORT_SYMBOL_GPL(rht_bucket_nested);
struct rhash_lock_head **rht_bucket_nested_insert(struct rhashtable *ht,
struct bucket_table *tbl,
unsigned int hash)
{
const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
unsigned int index = hash & ((1 << tbl->nest) - 1);
unsigned int size = tbl->size >> tbl->nest;
union nested_table *ntbl;
ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
hash >>= tbl->nest;
ntbl = nested_table_alloc(ht, &ntbl[index].table,
size <= (1 << shift));
while (ntbl && size > (1 << shift)) {
index = hash & ((1 << shift) - 1);
size >>= shift;
hash >>= shift;
ntbl = nested_table_alloc(ht, &ntbl[index].table,
size <= (1 << shift));
}
if (!ntbl)
return NULL;
return &ntbl[hash].bucket;
}
EXPORT_SYMBOL_GPL(rht_bucket_nested_insert);