Skip to contentSkip to conversation

bpf: Fix mprog detachment for empty mprog entry

PATCH Applied Daniel Borkmann 2 participants 3 messages Aug 4, 2023

Subsystems (4)
lore
Share & tools
lore

Daniel Borkmann

author maintainer
#0 lore raw

PATCH 1/2 bpf: Fix mprog detachment for empty mprog entry

syzbot reported an UBSAN array-index-out-of-bounds access in bpf_mprog_read() upon bpf_mprog_detach(). While it did not have a reproducer, I was able to manually reproduce through an empty mprog entry which just has miniq present.

The latter is important given otherwise we get an ENOENT error as tcx detaches the whole mprog entry. The index 4294967295 was triggered via NULL dtuple.prog which then attempts to detach from the back. bpf_mprog_fetch() in this case did hit the idx == total and therefore tried to grab the entry at idx -1.

Fix it by adding an explicit bpf_mprog_total() check in bpf_mprog_detach() and bail out early with ENOENT.

Fixes 053c8e1f235d ("bpf: Add generic attach/detach/query API for multi-progs")
Reported-by [email protected]
---
 kernel/bpf/mprog.c | 2 ++
 1 file changed, 2 insertions(+)
kernel/bpf/mprog.c +2 -0
index f7816d2bc3e4..32d2c4829eb8 100644
@@ -337,6 +337,8 @@ int bpf_mprog_detach(struct bpf_mprog_entry *entry,
 		return -EINVAL;
 	if (revision && revision != bpf_mprog_revision(entry))
 		return -ESTALE;
+	if (!bpf_mprog_total(entry))
+		return -ENOENT;
 	ret = bpf_mprog_tuple_relative(&rtuple, id_or_fd, flags,
 				       prog ? prog->type :
 				       BPF_PROG_TYPE_UNSPEC);
Signature · 1 line
2.34.1
author maintainer
#1 lore raw

PATCH 2/2 selftests/bpf: Add test for detachment on empty mprog entry

Add a detachment test case with miniq present to assert that with and without the miniq we get the same error.

  # ./test_progs -t tc_opts
  #244     tc_opts_after:OK
  #245     tc_opts_append:OK
  #246     tc_opts_basic:OK
  #247     tc_opts_before:OK
  #248     tc_opts_chain_classic:OK
  #249     tc_opts_delete_empty:OK
  #250     tc_opts_demixed:OK
  #251     tc_opts_detach:OK
  #252     tc_opts_detach_after:OK
  #253     tc_opts_detach_before:OK
  #254     tc_opts_dev_cleanup:OK
  #255     tc_opts_invalid:OK
  #256     tc_opts_mixed:OK
  #257     tc_opts_prepend:OK
  #258     tc_opts_replace:OK
  #259     tc_opts_revision:OK
  Summary: 16/0 PASSED, 0 SKIPPED, 0 FAILED
---
 .../selftests/bpf/prog_tests/tc_opts.c        | 31 +++++++++++++++++++
 1 file changed, 31 insertions(+)
tools/testing/selftests/bpf/prog_tests/tc_opts.c +31 -0
index 7914100f9b46..39bd253e41aa 100644
@@ -2237,3 +2237,34 @@ void serial_test_tc_opts_detach_after(void)
 	test_tc_opts_detach_after_target(BPF_TCX_INGRESS);
 	test_tc_opts_detach_after_target(BPF_TCX_EGRESS);
 }
+
+static void test_tc_opts_delete_empty(int target, bool chain_tc_old)
+{
+	LIBBPF_OPTS(bpf_tc_hook, tc_hook, .ifindex = loopback);
+	LIBBPF_OPTS(bpf_prog_detach_opts, optd);
+	int err;
+
+	assert_mprog_count(target, 0);
+	if (chain_tc_old) {
+		tc_hook.attach_point = target == BPF_TCX_INGRESS ?
+				       BPF_TC_INGRESS : BPF_TC_EGRESS;
+		err = bpf_tc_hook_create(&tc_hook);
+		ASSERT_OK(err, "bpf_tc_hook_create");
+		__assert_mprog_count(target, 0, true, loopback);
+	}
+	err = bpf_prog_detach_opts(0, loopback, target, &optd);
+	ASSERT_EQ(err, -ENOENT, "prog_detach");
+	if (chain_tc_old) {
+		tc_hook.attach_point = BPF_TC_INGRESS | BPF_TC_EGRESS;
+		bpf_tc_hook_destroy(&tc_hook);
+	}
+	assert_mprog_count(target, 0);
+}
+
+void serial_test_tc_opts_delete_empty(void)
+{
+	test_tc_opts_delete_empty(BPF_TCX_INGRESS, false);
+	test_tc_opts_delete_empty(BPF_TCX_EGRESS, false);
+	test_tc_opts_delete_empty(BPF_TCX_INGRESS, true);
+	test_tc_opts_delete_empty(BPF_TCX_EGRESS, true);
+}
Signature · 1 line
2.34.1
#2 lore raw
In reply to Daniel Borkmann
This series was applied to bpf/bpf-next.git (master) by Martin KaFai Lau <[email protected]>:

Hello:

This series was applied to bpf/bpf-next.git (master) by Martin KaFai Lau <[email protected]>:

Quoted 10 lines · On Fri, 4 Aug 2023 15:11:11 +0200 you wrote:

syzbot reported an UBSAN array-index-out-of-bounds access in bpf_mprog_read() upon bpf_mprog_detach(). While it did not have a reproducer, I was able to manually reproduce through an empty mprog entry which just has miniq present.

The latter is important given otherwise we get an ENOENT error as tcx detaches the whole mprog entry. The index 4294967295 was triggered via NULL dtuple.prog which then attempts to detach from the back. bpf_mprog_fetch() in this case did hit the idx == total and therefore tried to grab the entry at idx -1.

[...]

Here is the summary with links:
- [bpf-next,1/2] bpf: Fix mprog detachment for empty mprog entry

    https://git.kernel.org/bpf/bpf-next/c/d210f9735e13
  - [bpf-next,2/2] selftests/bpf: Add test for detachment on empty mprog entry
    https://git.kernel.org/bpf/bpf-next/c/21ce6abe178a

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html