netdev - [PATCH bpf-next] bpf, net: Use bpf mem allocator for sk local storage

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-Id: <20231025224151.385719-1-thinker.li@gmail.com>
Date: Wed, 25 Oct 2023 15:41:51 -0700
From: thinker.li@...il.com
To: bpf@...r.kernel.org,
	ast@...nel.org,
	martin.lau@...ux.dev,
	song@...nel.org,
	kernel-team@...a.com,
	andrii@...nel.org,
	netdev@...r.kernel.org
Cc: sinquersw@...il.com,
	kuifeng@...a.com,
	Kui-Feng Lee <thinker.li@...il.com>
Subject: [PATCH bpf-next] bpf, net: Use bpf mem allocator for sk local storage

From: Kui-Feng Lee <thinker.li@...il.com>

Switching to BPF memory allocator improve the performance
of sk local storage in terms of creating and destroying.

Here is numbers w/ and w/o this changes.

 - ./bench local-storage-create --storage-type=socket
   - w/ change
     - creates  552.474 ± 1.186k/s (552.474k/prod), 2.10 kmallocs/create
   - w/o change
     - creates  469.960 ± 2.501k/s (469.960k/prod), 4.18 kmallocs/create
 - ./bench -p 4 local-storage-create --storage-type=socket
   - w/ change
     - creates 1236.614 ± 2.833k/s (309.153k/prod), 2.09 kmallocs/create
   - w/o change
     - creates 1120.524 ± 1.527k/s (280.131k/prod), 4.16 kmallocs/create
 - ./bench -p 4 local-storage-create --storage-type=socket \
      --batch-size=1024
   - w/ change
     - creates 1416.437 ± 37.679k/s (354.109k/prod), 2.06 kmallocs/create
   - w/o change
     - creates 1302.636 ± 13.649k/s (325.659k/prod), 4.12 kmallocs/create

Overall, with bpf memory allocator, it improves
 - 17% for single thread and with batch size 32,
 - 10% for 4 threads and with batch size 32, and
 - 8% for 4 threads and with batch size 1024.

Signed-off-by: Kui-Feng Lee <thinker.li@...il.com>
---
 net/core/bpf_sk_storage.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/core/bpf_sk_storage.c b/net/core/bpf_sk_storage.c
index cca7594be92e..fdd9ad15cfe9 100644
--- a/net/core/bpf_sk_storage.c
+++ b/net/core/bpf_sk_storage.c
@@ -68,7 +68,7 @@ static void bpf_sk_storage_map_free(struct bpf_map *map)
 
 static struct bpf_map *bpf_sk_storage_map_alloc(union bpf_attr *attr)
 {
-	return bpf_local_storage_map_alloc(attr, &sk_cache, false);
+	return bpf_local_storage_map_alloc(attr, &sk_cache, true);
 }
 
 static int notsupp_get_next_key(struct bpf_map *map, void *key,
-- 
2.34.1