| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CACVXFVNPDAJ1hrx7xbp0z8kmCfzyphO_9qf-CYN0LxWeC9Uekw@mail.gmail.com>
Date: Mon, 28 Dec 2015 20:20:33 +0800
From: Ming Lei <tom.leiming@...il.com>
To: Daniel Borkmann <daniel@...earbox.net>
Cc: Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Alexei Starovoitov <ast@...nel.org>,
"David S. Miller" <davem@...emloft.net>,
Network Development <netdev@...r.kernel.org>
Subject: Re: [PATCH 3/3] bpf: hash: use per-bucket spinlock
On Mon, Dec 28, 2015 at 5:13 PM, Daniel Borkmann <daniel@...earbox.net> wrote:
> On 12/26/2015 10:31 AM, Ming Lei wrote:
>>
>> From: Ming Lei <ming.lei@...onical.com>
>>
>> Both htab_map_update_elem() and htab_map_delete_elem() can be
>> called from eBPF program, and they may be in kernel hot path,
>> so it isn't efficient to use a per-hashtable lock in this two
>> helpers.
>>
>> The per-hashtable spinlock is used just for protecting bucket's
>> hlist, and per-bucket lock should be enough. This patch converts
>> the per-hashtable lock into per-bucket spinlock, so that contention
>> can be decreased a lot.
>>
>> Signed-off-by: Ming Lei <ming.lei@...onical.com>
>> ---
>> kernel/bpf/hashtab.c | 35 +++++++++++++++++++++++++----------
>> 1 file changed, 25 insertions(+), 10 deletions(-)
>>
>> diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
>> index d857fcb..66bad7a 100644
>> --- a/kernel/bpf/hashtab.c
>> +++ b/kernel/bpf/hashtab.c
>> @@ -14,9 +14,14 @@
>> #include <linux/filter.h>
>> #include <linux/vmalloc.h>
>>
>> +struct bucket {
>> + struct hlist_head head;
>> + raw_spinlock_t lock;
>> +};
>> +
>> struct bpf_htab {
>> struct bpf_map map;
>> - struct hlist_head *buckets;
>> + struct bucket *buckets;
>> raw_spinlock_t lock;
>
>
> Shouldn't the lock member be removed from here?
Hammm, looks this one is a bad version, sorry for that, and will resend it.
>
>
>> atomic_t count; /* number of elements in this hashtable */
>> u32 n_buckets; /* number of hash buckets */
>> @@ -88,24 +93,25 @@ static struct bpf_map *htab_map_alloc(union bpf_attr
>> *attr)
>> /* make sure page count doesn't overflow */
>> goto free_htab;
>>
>> - htab->map.pages = round_up(htab->n_buckets * sizeof(struct
>> hlist_head) +
>> + htab->map.pages = round_up(htab->n_buckets * sizeof(struct bucket)
>> +
>> htab->elem_size *
>> htab->map.max_entries,
>> PAGE_SIZE) >> PAGE_SHIFT;
>>
>> err = -ENOMEM;
>> - htab->buckets = kmalloc_array(htab->n_buckets, sizeof(struct
>> hlist_head),
>> + htab->buckets = kmalloc_array(htab->n_buckets, sizeof(struct
>> bucket),
>> GFP_USER | __GFP_NOWARN);
>>
>> if (!htab->buckets) {
>> - htab->buckets = vmalloc(htab->n_buckets * sizeof(struct
>> hlist_head));
>> + htab->buckets = vmalloc(htab->n_buckets * sizeof(struct
>> bucket));
>> if (!htab->buckets)
>> goto free_htab;
>> }
>>
>> - for (i = 0; i < htab->n_buckets; i++)
>> - INIT_HLIST_HEAD(&htab->buckets[i]);
>> + for (i = 0; i < htab->n_buckets; i++) {
>> + INIT_HLIST_HEAD(&htab->buckets[i].head);
>> + raw_spin_lock_init(&htab->buckets[i].lock);
>> + }
>>
>> - raw_spin_lock_init(&htab->lock);
>> atomic_set(&htab->count, 0);
>>
>> return &htab->map;
>> @@ -120,11 +126,16 @@ static inline u32 htab_map_hash(const void *key, u32
>> key_len)
>> return jhash(key, key_len, 0);
>> }
>>
>> -static inline struct hlist_head *select_bucket(struct bpf_htab *htab, u32
>> hash)
>> +static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32
>> hash)
>> {
>> return &htab->buckets[hash & (htab->n_buckets - 1)];
>> }
>>
>> +static inline struct hlist_head *select_bucket(struct bpf_htab *htab, u32
>> hash)
>> +{
>> + return &__select_bucket(htab, hash)->head;
>> +}
>> +
>> static struct htab_elem *lookup_elem_raw(struct hlist_head *head, u32
>> hash,
>> void *key, u32 key_size)
>> {
>> @@ -227,6 +238,7 @@ static int htab_map_update_elem(struct bpf_map *map,
>> void *key, void *value,
>> struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
>> struct htab_elem *l_new, *l_old;
>> struct hlist_head *head;
>> + struct bucket *b;
>> unsigned long flags;
>> u32 key_size;
>> int ret;
>> @@ -248,7 +260,8 @@ static int htab_map_update_elem(struct bpf_map *map,
>> void *key, void *value,
>> memcpy(l_new->key + round_up(key_size, 8), value,
>> map->value_size);
>>
>> l_new->hash = htab_map_hash(l_new->key, key_size);
>> - head = select_bucket(htab, l_new->hash);
>> + b = __select_bucket(htab, l_new->hash);
>> + head = &b->head;
>>
>> /* bpf_map_update_elem() can be called in_irq() */
>> raw_spin_lock_irqsave(&htab->lock, flags);
>
>
> I am a bit confused on this one, though.
>
> The raw spin lock is still per hashtable (htab->lock), and has not been
> converted in
> this patch to the per bucket one (as opposed to what the commit message
> says), so
> this patch seems to go into the right direction, but is a bit incomplete?
> Shouldn't
> the above f.e. take b->lock, etc?
>
>> @@ -299,6 +312,7 @@ static int htab_map_delete_elem(struct bpf_map *map,
>> void *key)
>> {
>> struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
>> struct hlist_head *head;
>> + struct bucket *b;
>> struct htab_elem *l;
>> unsigned long flags;
>> u32 hash, key_size;
>> @@ -309,7 +323,8 @@ static int htab_map_delete_elem(struct bpf_map *map,
>> void *key)
>> key_size = map->key_size;
>>
>> hash = htab_map_hash(key, key_size);
>> - head = select_bucket(htab, hash);
>> + b = __select_bucket(htab, hash);
>> + head = &b->head;
>>
>> raw_spin_lock_irqsave(&htab->lock, flags);
>>
>
> Same here?
>
> Thanks,
> Daniel
--
Ming Lei
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists