On Fri, Jan 8, 2021 at 5:56 PM Martin KaFai Lau [off-list ref] wrote:

On Fri, Jan 08, 2021 at 01:02:22PM -0800, Stanislav Fomichev wrote:

quoted

When we attach a bpf program to cgroup/getsockopt any other getsockopt()
syscall starts incurring kzalloc/kfree cost.

Let add a small buffer on the stack and use it for small (majority)
{s,g}etsockopt values. The buffer is small enough to fit into
the cache line and cover the majority of simple options (most
of them are 4 byte ints).

It seems natural to do the same for setsockopt, but it's a bit more
involved when the BPF program modifies the data (where we have to
kmalloc). The assumption is that for the majority of setsockopt
calls (which are doing pure BPF options or apply policy) this
will bring some benefit as well.

Without this patch (we remove about 1% __kmalloc):
     3.38%     0.07%  tcp_mmap  [kernel.kallsyms]  [k] __cgroup_bpf_run_filter_getsockopt
            |
             --3.30%--__cgroup_bpf_run_filter_getsockopt
                       |
                        --0.81%--__kmalloc

Signed-off-by: Stanislav Fomichev <redacted>
Cc: Martin KaFai Lau <redacted>
Cc: Song Liu <redacted>
---
 include/linux/filter.h |  5 ++++
 kernel/bpf/cgroup.c    | 52 ++++++++++++++++++++++++++++++++++++------
 2 files changed, 50 insertions(+), 7 deletions(-)

diff --git a/include/linux/filter.h b/include/linux/filter.h
index 29c27656165b..8739f1d4cac4 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h

@@ -1281,6 +1281,11 @@ struct bpf_sysctl_kern {
      u64 tmp_reg;
 };

+#define BPF_SOCKOPT_KERN_BUF_SIZE    32
+struct bpf_sockopt_buf {
+     u8              data[BPF_SOCKOPT_KERN_BUF_SIZE];
+};
+
 struct bpf_sockopt_kern {
      struct sock     *sk;
      u8              *optval;

diff --git a/kernel/bpf/cgroup.c b/kernel/bpf/cgroup.c
index c41bb2f34013..a9aad9c419e1 100644
--- a/kernel/bpf/cgroup.c
+++ b/kernel/bpf/cgroup.c

@@ -1298,7 +1298,8 @@ static bool __cgroup_bpf_prog_array_is_empty(struct cgroup *cgrp,
      return empty;
 }

-static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen)
+static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen,
+                          struct bpf_sockopt_buf *buf)
 {
      if (unlikely(max_optlen < 0))
              return -EINVAL;

@@ -1310,6 +1311,15 @@ static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen)
              max_optlen = PAGE_SIZE;
      }

+     if (max_optlen <= sizeof(buf->data)) {
+             /* When the optval fits into BPF_SOCKOPT_KERN_BUF_SIZE
+              * bytes avoid the cost of kzalloc.
+              */
+             ctx->optval = buf->data;
+             ctx->optval_end = ctx->optval + max_optlen;
+             return max_optlen;
+     }
+
      ctx->optval = kzalloc(max_optlen, GFP_USER);
      if (!ctx->optval)
              return -ENOMEM;

@@ -1319,16 +1329,26 @@ static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen)
      return max_optlen;
 }

-static void sockopt_free_buf(struct bpf_sockopt_kern *ctx)
+static void sockopt_free_buf(struct bpf_sockopt_kern *ctx,
+                          struct bpf_sockopt_buf *buf)
 {
+     if (ctx->optval == buf->data)
+             return;
      kfree(ctx->optval);
 }

+static bool sockopt_buf_allocated(struct bpf_sockopt_kern *ctx,
+                               struct bpf_sockopt_buf *buf)
+{
+     return ctx->optval != buf->data;
+}
+
 int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level,
                                     int *optname, char __user *optval,
                                     int *optlen, char **kernel_optval)
 {
      struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
+     struct bpf_sockopt_buf buf = {};
      struct bpf_sockopt_kern ctx = {
              .sk = sk,
              .level = *level,

@@ -1350,7 +1370,7 @@ int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level,
       */
      max_optlen = max_t(int, 16, *optlen);

-     max_optlen = sockopt_alloc_buf(&ctx, max_optlen);
+     max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
      if (max_optlen < 0)
              return max_optlen;

@@ -1390,13 +1410,30 @@ int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level,
               */
              if (ctx.optlen != 0) {

When ctx.optlen == 0, is sockopt_free_buf() called?
Did I miss something?

Ouch, good catch, it looks like we do leak the buf here with optlen == 0.

We should probably change the following below to:
out:
      if (!*kernel_optval)
           sockopt_free_buf(...);

I'll send a bpf patch with a Fixes tag, thanks!