[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <fa3c59e3-da85-aec3-d7bb-6360c842042c@meta.com>
Date: Thu, 15 Dec 2022 23:03:16 -0800
From: Yonghong Song <yhs@...a.com>
To: Hao Sun <sunhao.th@...il.com>, bpf@...r.kernel.org
Cc: ast@...nel.org, daniel@...earbox.net, john.fastabend@...il.com,
andrii@...nel.org, martin.lau@...ux.dev, song@...nel.org,
yhs@...com, kpsingh@...nel.org, sdf@...gle.com, haoluo@...gle.com,
jolsa@...nel.org, davem@...emloft.net, linux-kernel@...r.kernel.org
Subject: Re: [PATCH bpf-next] bpf: dup xlated insns with kvmalloc+memcpy
On 12/15/22 9:54 PM, Hao Sun wrote:
> Currently, kmemdup() is used for allocating and copying xlated insns
> in bpf_insn_prepare_dump(). The following warning can be triggered
> when dup large amount of insns (roughly BPF_COMPLEXITY_LIMIT_INSNS/2)
> because kmemdup() uses kmalloc() which would fail when allocing size
> is too big, leading to failure in dump xlated insns:
>
> WARNING: CPU: 2 PID: 7060 at mm/page_alloc.c:5534
> Call Trace:
> <TASK>
> __alloc_pages_node include/linux/gfp.h:237 [inline]
> alloc_pages_node include/linux/gfp.h:260 [inline]
> __kmalloc_large_node+0x81/0x160 mm/slab_common.c:1096
> __do_kmalloc_node mm/slab_common.c:943 [inline]
> __kmalloc_node_track_caller.cold+0x5/0x5d mm/slab_common.c:975
> kmemdup+0x29/0x60 mm/util.c:129
> kmemdup include/linux/fortify-string.h:585 [inline]
> bpf_insn_prepare_dump kernel/bpf/syscall.c:3820 [inline]
> bpf_prog_get_info_by_fd+0x9a3/0x2cb0 kernel/bpf/syscall.c:3975
> bpf_obj_get_info_by_fd kernel/bpf/syscall.c:4297 [inline]
> __sys_bpf+0x3928/0x56f0 kernel/bpf/syscall.c:5004
> __do_sys_bpf kernel/bpf/syscall.c:5069 [inline]
> __se_sys_bpf kernel/bpf/syscall.c:5067 [inline]
> ...
>
> So use kvmalloc()+memcpy() to fix this, for small size of insns,
> this is same as kmemdup(), but this also support dup large amount
> of xlated insns.
>
> Signed-off-by: Hao Sun <sunhao.th@...il.com>
> ---
> kernel/bpf/syscall.c | 8 ++++----
> 1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> index 35972afb6850..06229fddac0d 100644
> --- a/kernel/bpf/syscall.c
> +++ b/kernel/bpf/syscall.c
> @@ -3831,10 +3831,10 @@ static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
> u8 code;
> int i;
>
> - insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
> - GFP_USER);
Does kmemdup(prog->insnsi, bpf_prog_insn_size(prog), GFP_USER |
__GFP_NOWARN) work?
> - if (!insns)
> + insns = kvmalloc(bpf_prog_insn_size(prog), GFP_USER | __GFP_NOWARN);
> + if (unlikely(!insns))
> return insns;
> + memcpy(insns, prog->insnsi, bpf_prog_insn_size(prog));
>
> for (i = 0; i < prog->len; i++) {
> code = insns[i].code;
> @@ -3992,7 +3992,7 @@ static int bpf_prog_get_info_by_fd(struct file *file,
> uinsns = u64_to_user_ptr(info.xlated_prog_insns);
> ulen = min_t(u32, info.xlated_prog_len, ulen);
> fault = copy_to_user(uinsns, insns_sanitized, ulen);
> - kfree(insns_sanitized);
> + kvfree(insns_sanitized);
> if (fault)
> return -EFAULT;
> }
>
> base-commit: 0e43662e61f2569500ab83b8188c065603530785
Powered by blists - more mailing lists