| Message ID | 20200915223507.3792641-1-yhs@fb.com |
|---|---|
| State | New |
| Headers | show |
| Series | [bpf-next,v2] bpf: using rcu_read_lock for bpf_sk_storage_map iterator | expand |
On Tue, Sep 15, 2020 at 03:35:07PM -0700, Yonghong Song wrote: > If a bucket contains a lot of sockets, during bpf_iter traversing > a bucket, concurrent userspace bpf_map_update_elem() and > bpf program bpf_sk_storage_{get,delete}() may experience > some undesirable delays as they will compete with bpf_iter > for bucket lock. > > Note that the number of buckets for bpf_sk_storage_map > is roughly the same as the number of cpus. So if there > are lots of sockets in the system, each bucket could > contain lots of sockets. > > Different actual use cases may experience different delays. > Here, using selftest bpf_iter subtest bpf_sk_storage_map, > I hacked the kernel with ktime_get_mono_fast_ns() > to collect the time when a bucket was locked > during bpf_iter prog traversing that bucket. This way, > the maximum incurred delay was measured w.r.t. the > number of elements in a bucket. > # elems in each bucket delay(ns) > 64 17000 > 256 72512 > 2048 875246 > > The potential delays will be further increased if > we have even more elemnts in a bucket. Using rcu_read_lock() > is a reasonable compromise here. It may lose some precision, e.g., > access stale sockets, but it will not hurt performance of > bpf program or user space application which also tries > to get/delete or update map elements. > > Cc: Martin KaFai Lau <kafai@fb.com> > Acked-by: Song Liu <songliubraving@fb.com> > Signed-off-by: Yonghong Song <yhs@fb.com> > --- > net/core/bpf_sk_storage.c | 21 ++++++++------------- > 1 file changed, 8 insertions(+), 13 deletions(-) > > Changelog: > v1 -> v2: > - added some performance number. (Song) > - tried to silence some sparse complains. but still has some left like > context imbalance in "..." - different lock contexts for basic block > which the code is too hard for sparse to analyze. (Jakub) > > diff --git a/net/core/bpf_sk_storage.c b/net/core/bpf_sk_storage.c > index 4a86ea34f29e..4fc6b03d3639 100644 > --- a/net/core/bpf_sk_storage.c > +++ b/net/core/bpf_sk_storage.c > @@ -678,6 +678,7 @@ struct bpf_iter_seq_sk_storage_map_info { > static struct bpf_local_storage_elem * > bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, > struct bpf_local_storage_elem *prev_selem) > + __acquires(RCU) __releases(RCU) > { > struct bpf_local_storage *sk_storage; > struct bpf_local_storage_elem *selem; > @@ -701,7 +702,7 @@ bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, > if (!selem) { > /* not found, unlock and go to the next bucket */ > b = &smap->buckets[bucket_id++]; > - raw_spin_unlock_bh(&b->lock); > + rcu_read_unlock(); > skip_elems = 0; > break; > } > @@ -715,7 +716,7 @@ bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, > > for (i = bucket_id; i < (1U << smap->bucket_log); i++) { > b = &smap->buckets[i]; > - raw_spin_lock_bh(&b->lock); > + rcu_read_lock(); > count = 0; > hlist_for_each_entry(selem, &b->list, map_node) { hlist_for_each_entry_rcu() > sk_storage = rcu_dereference_raw(selem->local_storage); Does lockdep complain without "_raw"?
On 9/15/20 10:37 PM, Martin KaFai Lau wrote: > On Tue, Sep 15, 2020 at 03:35:07PM -0700, Yonghong Song wrote: >> If a bucket contains a lot of sockets, during bpf_iter traversing >> a bucket, concurrent userspace bpf_map_update_elem() and >> bpf program bpf_sk_storage_{get,delete}() may experience >> some undesirable delays as they will compete with bpf_iter >> for bucket lock. >> >> Note that the number of buckets for bpf_sk_storage_map >> is roughly the same as the number of cpus. So if there >> are lots of sockets in the system, each bucket could >> contain lots of sockets. >> >> Different actual use cases may experience different delays. >> Here, using selftest bpf_iter subtest bpf_sk_storage_map, >> I hacked the kernel with ktime_get_mono_fast_ns() >> to collect the time when a bucket was locked >> during bpf_iter prog traversing that bucket. This way, >> the maximum incurred delay was measured w.r.t. the >> number of elements in a bucket. >> # elems in each bucket delay(ns) >> 64 17000 >> 256 72512 >> 2048 875246 >> >> The potential delays will be further increased if >> we have even more elemnts in a bucket. Using rcu_read_lock() >> is a reasonable compromise here. It may lose some precision, e.g., >> access stale sockets, but it will not hurt performance of >> bpf program or user space application which also tries >> to get/delete or update map elements. >> >> Cc: Martin KaFai Lau <kafai@fb.com> >> Acked-by: Song Liu <songliubraving@fb.com> >> Signed-off-by: Yonghong Song <yhs@fb.com> >> --- >> net/core/bpf_sk_storage.c | 21 ++++++++------------- >> 1 file changed, 8 insertions(+), 13 deletions(-) >> >> Changelog: >> v1 -> v2: >> - added some performance number. (Song) >> - tried to silence some sparse complains. but still has some left like >> context imbalance in "..." - different lock contexts for basic block >> which the code is too hard for sparse to analyze. (Jakub) >> >> diff --git a/net/core/bpf_sk_storage.c b/net/core/bpf_sk_storage.c >> index 4a86ea34f29e..4fc6b03d3639 100644 >> --- a/net/core/bpf_sk_storage.c >> +++ b/net/core/bpf_sk_storage.c >> @@ -678,6 +678,7 @@ struct bpf_iter_seq_sk_storage_map_info { >> static struct bpf_local_storage_elem * >> bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, >> struct bpf_local_storage_elem *prev_selem) >> + __acquires(RCU) __releases(RCU) >> { >> struct bpf_local_storage *sk_storage; >> struct bpf_local_storage_elem *selem; >> @@ -701,7 +702,7 @@ bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, >> if (!selem) { >> /* not found, unlock and go to the next bucket */ >> b = &smap->buckets[bucket_id++]; >> - raw_spin_unlock_bh(&b->lock); >> + rcu_read_unlock(); >> skip_elems = 0; >> break; >> } >> @@ -715,7 +716,7 @@ bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, >> >> for (i = bucket_id; i < (1U << smap->bucket_log); i++) { >> b = &smap->buckets[i]; >> - raw_spin_lock_bh(&b->lock); >> + rcu_read_lock(); >> count = 0; >> hlist_for_each_entry(selem, &b->list, map_node) { > hlist_for_each_entry_rcu() Good catch! > >> sk_storage = rcu_dereference_raw(selem->local_storage); > Does lockdep complain without "_raw"? It didn't. But I think using rcu_dereference() is better as it provides extra checking. Will fix and send v3.
diff --git a/net/core/bpf_sk_storage.c b/net/core/bpf_sk_storage.c index 4a86ea34f29e..4fc6b03d3639 100644 --- a/net/core/bpf_sk_storage.c +++ b/net/core/bpf_sk_storage.c @@ -678,6 +678,7 @@ struct bpf_iter_seq_sk_storage_map_info { static struct bpf_local_storage_elem * bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, struct bpf_local_storage_elem *prev_selem) + __acquires(RCU) __releases(RCU) { struct bpf_local_storage *sk_storage; struct bpf_local_storage_elem *selem; @@ -701,7 +702,7 @@ bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, if (!selem) { /* not found, unlock and go to the next bucket */ b = &smap->buckets[bucket_id++]; - raw_spin_unlock_bh(&b->lock); + rcu_read_unlock(); skip_elems = 0; break; } @@ -715,7 +716,7 @@ bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, for (i = bucket_id; i < (1U << smap->bucket_log); i++) { b = &smap->buckets[i]; - raw_spin_lock_bh(&b->lock); + rcu_read_lock(); count = 0; hlist_for_each_entry(selem, &b->list, map_node) { sk_storage = rcu_dereference_raw(selem->local_storage); @@ -726,7 +727,7 @@ bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, } count++; } - raw_spin_unlock_bh(&b->lock); + rcu_read_unlock(); skip_elems = 0; } @@ -801,18 +802,12 @@ static int bpf_sk_storage_map_seq_show(struct seq_file *seq, void *v) } static void bpf_sk_storage_map_seq_stop(struct seq_file *seq, void *v) + __releases(RCU) { - struct bpf_iter_seq_sk_storage_map_info *info = seq->private; - struct bpf_local_storage_map *smap; - struct bpf_local_storage_map_bucket *b; - - if (!v) { + if (!v) (void)__bpf_sk_storage_map_seq_show(seq, v); - } else { - smap = (struct bpf_local_storage_map *)info->map; - b = &smap->buckets[info->bucket_id]; - raw_spin_unlock_bh(&b->lock); - } + else + rcu_read_unlock(); } static int bpf_iter_init_sk_storage_map(void *priv_data,
If a bucket contains a lot of sockets, during bpf_iter traversing a bucket, concurrent userspace bpf_map_update_elem() and bpf program bpf_sk_storage_{get,delete}() may experience some undesirable delays as they will compete with bpf_iter for bucket lock. Note that the number of buckets for bpf_sk_storage_map is roughly the same as the number of cpus. So if there are lots of sockets in the system, each bucket could contain lots of sockets. Different actual use cases may experience different delays. Here, using selftest bpf_iter subtest bpf_sk_storage_map, I hacked the kernel with ktime_get_mono_fast_ns() to collect the time when a bucket was locked during bpf_iter prog traversing that bucket. This way, the maximum incurred delay was measured w.r.t. the number of elements in a bucket. # elems in each bucket delay(ns) 64 17000 256 72512 2048 875246 The potential delays will be further increased if we have even more elemnts in a bucket. Using rcu_read_lock() is a reasonable compromise here. It may lose some precision, e.g., access stale sockets, but it will not hurt performance of bpf program or user space application which also tries to get/delete or update map elements. Cc: Martin KaFai Lau <kafai@fb.com> Acked-by: Song Liu <songliubraving@fb.com> Signed-off-by: Yonghong Song <yhs@fb.com> --- net/core/bpf_sk_storage.c | 21 ++++++++------------- 1 file changed, 8 insertions(+), 13 deletions(-) Changelog: v1 -> v2: - added some performance number. (Song) - tried to silence some sparse complains. but still has some left like context imbalance in "..." - different lock contexts for basic block which the code is too hard for sparse to analyze. (Jakub)