bpf: Format string can't be empty

[PATCH] bpf: Format string can't be empty

Posted by Edward Adam Davis 1 month ago

The user constructed a BPF program containing a bpf_snprintf() call.
The fmt parameter passed to bpf_snprintf() was not assigned a value;
it only executed the BPF_MAP_FREEZE command to freeze the fmt string.
Furthermore, when bpf_check() executed check_reg_const_str() and
check_bpf_snprintf_call() to check the fmt input parameter of the
user-constructed BPF program's bpf_snprintf() call, strnchr() only
checked if fmt was a null-terminated string. This led the BPF verifier
to incorrectly assume the constant format string was valid.
When the BPF program was actually executed, the out-of-bounds (OOB)
issue reported by syzbot occurred [1].

This issue is strongly related to bpf_snprintf(), therefore adding a
check for an empty format string in check_bpf_snprintf_call() would
be beneficial. Since it calls bpf_bprintf_prepare(), only adding a
check on the result of strnchr() is needed to prevent the case where
the format string is empty.

[1]
BUG: KASAN: slab-out-of-bounds in strnchr+0x5e/0x80 lib/string.c:405
Read of size 1 at addr ffff888029e093b0 by task ksoftirqd/1/23
Call Trace:
 strnchr+0x5e/0x80 lib/string.c:405
 bpf_bprintf_prepare+0x167/0x13d0 kernel/bpf/helpers.c:829
 ____bpf_snprintf kernel/bpf/helpers.c:1065 [inline]
 bpf_snprintf+0xd3/0x1b0 kernel/bpf/helpers.c:1049

Allocated by task 6022:
 __bpf_map_area_alloc kernel/bpf/syscall.c:395 [inline]
 bpf_map_area_alloc+0x64/0x180 kernel/bpf/syscall.c:408
 insn_array_alloc+0x52/0x140 kernel/bpf/bpf_insn_array.c:49
 map_create+0xafd/0x16a0 kernel/bpf/syscall.c:1514

The buggy address is located 0 bytes to the right of
 allocated 944-byte region [ffff888029e09000, ffff888029e093b0)

Fixes: d9c9e4db186a ("bpf: Factorize bpf_trace_printk and bpf_seq_printf")
Reported-by: syzbot+2c29addf92581b410079@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=2c29addf92581b410079
Tested-by: syzbot+2c29addf92581b410079@syzkaller.appspotmail.com
Signed-off-by: Edward Adam Davis <eadavis@qq.com>
---
 kernel/bpf/helpers.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
index db72b96f9c8c..88da2d0e634c 100644
--- a/kernel/bpf/helpers.c
+++ b/kernel/bpf/helpers.c
@@ -827,7 +827,7 @@ int bpf_bprintf_prepare(const char *fmt, u32 fmt_size, const u64 *raw_args,
 	char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
 
 	fmt_end = strnchr(fmt, fmt_size, 0);
-	if (!fmt_end)
+	if (!fmt_end || fmt_end == fmt)
 		return -EINVAL;
 	fmt_size = fmt_end - fmt;
 
-- 
2.43.0

Re: [PATCH] bpf: Format string can't be empty

Posted by Alexei Starovoitov 1 month ago

On Wed, Jan 7, 2026 at 1:39 AM Edward Adam Davis <eadavis@qq.com> wrote:
>
> The user constructed a BPF program containing a bpf_snprintf() call.
> The fmt parameter passed to bpf_snprintf() was not assigned a value;
> it only executed the BPF_MAP_FREEZE command to freeze the fmt string.
> Furthermore, when bpf_check() executed check_reg_const_str() and
> check_bpf_snprintf_call() to check the fmt input parameter of the
> user-constructed BPF program's bpf_snprintf() call, strnchr() only
> checked if fmt was a null-terminated string. This led the BPF verifier
> to incorrectly assume the constant format string was valid.
> When the BPF program was actually executed, the out-of-bounds (OOB)
> issue reported by syzbot occurred [1].
>
> This issue is strongly related to bpf_snprintf(), therefore adding a
> check for an empty format string in check_bpf_snprintf_call() would
> be beneficial. Since it calls bpf_bprintf_prepare(), only adding a
> check on the result of strnchr() is needed to prevent the case where
> the format string is empty.
>
> [1]
> BUG: KASAN: slab-out-of-bounds in strnchr+0x5e/0x80 lib/string.c:405
> Read of size 1 at addr ffff888029e093b0 by task ksoftirqd/1/23
> Call Trace:
>  strnchr+0x5e/0x80 lib/string.c:405
>  bpf_bprintf_prepare+0x167/0x13d0 kernel/bpf/helpers.c:829
>  ____bpf_snprintf kernel/bpf/helpers.c:1065 [inline]
>  bpf_snprintf+0xd3/0x1b0 kernel/bpf/helpers.c:1049
>
> Allocated by task 6022:
>  __bpf_map_area_alloc kernel/bpf/syscall.c:395 [inline]
>  bpf_map_area_alloc+0x64/0x180 kernel/bpf/syscall.c:408
>  insn_array_alloc+0x52/0x140 kernel/bpf/bpf_insn_array.c:49
>  map_create+0xafd/0x16a0 kernel/bpf/syscall.c:1514
>
> The buggy address is located 0 bytes to the right of
>  allocated 944-byte region [ffff888029e09000, ffff888029e093b0)
>
> Fixes: d9c9e4db186a ("bpf: Factorize bpf_trace_printk and bpf_seq_printf")
> Reported-by: syzbot+2c29addf92581b410079@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=2c29addf92581b410079
> Tested-by: syzbot+2c29addf92581b410079@syzkaller.appspotmail.com
> Signed-off-by: Edward Adam Davis <eadavis@qq.com>
> ---
>  kernel/bpf/helpers.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> index db72b96f9c8c..88da2d0e634c 100644
> --- a/kernel/bpf/helpers.c
> +++ b/kernel/bpf/helpers.c
> @@ -827,7 +827,7 @@ int bpf_bprintf_prepare(const char *fmt, u32 fmt_size, const u64 *raw_args,
>         char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
>
>         fmt_end = strnchr(fmt, fmt_size, 0);
> -       if (!fmt_end)
> +       if (!fmt_end || fmt_end == fmt)
>                 return -EINVAL;

I don't think you root caused it correctly.
The better fix and analysis:
https://patchwork.kernel.org/project/netdevbpf/patch/20260107021037.289644-1-kartikey406@gmail.com/

pw-bot: cr

Re: [PATCH] bpf: Format string can't be empty

Posted by Edward Adam Davis 1 month ago

On Wed, 7 Jan 2026 19:02:37 -0800, Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
> > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> > index db72b96f9c8c..88da2d0e634c 100644
> > --- a/kernel/bpf/helpers.c
> > +++ b/kernel/bpf/helpers.c
> > @@ -827,7 +827,7 @@ int bpf_bprintf_prepare(const char *fmt, u32 fmt_size, const u64 *raw_args,
> >         char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
> >
> >         fmt_end = strnchr(fmt, fmt_size, 0);
> > -       if (!fmt_end)
> > +       if (!fmt_end || fmt_end == fmt)
> >                 return -EINVAL;
> 
> I don't think you root caused it correctly.
> The better fix and analysis:
I am keeping my analysis and patch.
The root cause of the problem is that the format string does not contain
a null terminator ('\0').
Filtering out map type 0x22 to solve the problem is too hasty, as it
would prevent all instructions from calling functions with constant
string arguments.

Re: [PATCH] bpf: Format string can't be empty

Posted by Alexei Starovoitov 1 month ago

On Wed, Jan 7, 2026 at 7:52 PM Edward Adam Davis <eadavis@qq.com> wrote:
>
> On Wed, 7 Jan 2026 19:02:37 -0800, Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
> > > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> > > index db72b96f9c8c..88da2d0e634c 100644
> > > --- a/kernel/bpf/helpers.c
> > > +++ b/kernel/bpf/helpers.c
> > > @@ -827,7 +827,7 @@ int bpf_bprintf_prepare(const char *fmt, u32 fmt_size, const u64 *raw_args,
> > >         char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
> > >
> > >         fmt_end = strnchr(fmt, fmt_size, 0);
> > > -       if (!fmt_end)
> > > +       if (!fmt_end || fmt_end == fmt)
> > >                 return -EINVAL;
> >
> > I don't think you root caused it correctly.
> > The better fix and analysis:
> I am keeping my analysis and patch.
> The root cause of the problem is that the format string does not contain
> a null terminator ('\0').
> Filtering out map type 0x22 to solve the problem is too hasty, as it
> would prevent all instructions from calling functions with constant
> string arguments.

If you think it's still possible to construct a program that
passes empty const string into this helper then please craft
a selftest that demonstrates that.