linux-kernel - Re: [PATCH] net: filter: fix upper BPF instruction limit

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

Message-ID: <CAMEtUuwPug7Bbgi2ccOx4jxd2Q80AnVoiHeeBW2av=_sQMiSfg@mail.gmail.com>
Date:	Wed, 18 Jun 2014 16:19:45 -0700
From:	Alexei Starovoitov <ast@...mgrid.com>
To:	Kees Cook <keescook@...omium.org>
Cc:	LKML <linux-kernel@...r.kernel.org>,
	"David S. Miller" <davem@...emloft.net>,
	Daniel Borkmann <dborkman@...hat.com>,
	Eric Dumazet <edumazet@...gle.com>,
	Chema Gonzalez <chema@...gle.com>,
	Network Development <netdev@...r.kernel.org>
Subject: Re: [PATCH] net: filter: fix upper BPF instruction limit

On Wed, Jun 18, 2014 at 3:55 PM, Kees Cook <keescook@...omium.org> wrote:
> On Wed, Jun 18, 2014 at 3:48 PM, Alexei Starovoitov <ast@...mgrid.com> wrote:
>> On Wed, Jun 18, 2014 at 3:34 PM, Kees Cook <keescook@...omium.org> wrote:
>>> The original checks (via sk_chk_filter) for instruction count uses ">",
>>> not ">=", so changing this in sk_convert_filter has the potential to break
>>> existing seccomp filters that used exactly BPF_MAXINSNS many instructions.
>>>
>>> Fixes: bd4cf0ed331a ("net: filter: rework/optimize internal BPF interpreter's instruction set")
>>> Signed-off-by: Kees Cook <keescook@...omium.org>
>>> Cc: stable@...r.kernel.org # v3.15+
>>
>> Acked-by: Alexei Starovoitov <ast@...mgrid.com>
>>
>> I wonder how did you catch this? :)
>> Just code inspection or seccomp actually generating such programs?
>
> In the process of merging my seccomp thread-sync series back with
> mainline, I got uncomfortable that I was moving filter size validation
> around without actually testing it. When I added it, I was happy that
> my series was correctly checking size limits, but then discovered my
> newly added check actually failed on an earlier kernel (3.2). Tracking
> it down found the corner case under 3.15.
>
> Here's the test I added to the seccomp regression tests, if you're interested:
> https://github.com/kees/seccomp/commit/794d54a340cde70a3bdf7fe0ade1f95d160b2883

Nice. I'm assuming https://github.com/redpig/seccomp is still the main tree
for seccomp testsuite…

btw I've tried to add 'real' test to it (one generated by chrome)

+TEST(chrome_syscalls) {
+       static struct sock_filter filter[] = {
+               { 32, 240, 61, 4 },          /* 0:      ld [4] */
+               { 21, 1, 0, -1073741762 },   /* 1:      jeq #0xc000003e, 3, 2 */
+               { 5, 0, 0, 271 },            /* 2:      ja 274 */
+               { 32, 208, 198, 0 },         /* 3:      ld [0] */
+               { 69, 0, 1, 1073741824 },    /* 4:      jset
#0x40000000, 5, 6 */
+               { 6, 0, 0, 196615 },         /* 5:      ret #0x30007 */
+               { 53, 0, 7, 121 },           /* 6:      jge #0x79, 7, 14 */
+               { 53, 0, 12, 214 },          /* 7:      jge #0xd6, 8, 20 */
…
+               { 6, 0, 0, 2147418112 },     /* 272:    ret #0x7fff0000 */
+               { 6, 0, 0, 327681 },         /* 273:    ret #0x50001 */
+               { 6, 0, 0, 196610 },         /* 274:    ret #0x30002 */
+       };
...
+       for (i = 0; i < MAX_SYSCALLS; i++) {
+               ch_pid = fork();
+               ASSERT_LE(0, ch_pid);
+
+               if (ch_pid == 0) {
+                       ret = prctl(PR_SET_SECCOMP,
+                              SECCOMP_MODE_FILTER, &prog);
+                       ASSERT_EQ(0, ret);
+#define MAGIC (-1ll << 2)
+                       err = syscall(i, MAGIC, MAGIC, MAGIC,
+                              MAGIC, MAGIC, MAGIC);
+                       syscall(__NR_exit, 0);
+               }
+               wait(&status);
+               if (status != expected_status[i])
…

but it's really x64 only and looks ugly. Do you have better ideas
on how to test all possible paths through auto-generated branch tree?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/