Hello Sasha!
On Tue, Nov 03, 2026 at 11:39:17AM -0500, Sasha Levin wrote:
quoted hunk ↗ jump to hunk
regex_escape() returns (char *)-ENOMEM when the buffer is too small, but
matches_log_fs_extra() checks the returned pointer with "< 0", which is
never true, and goes on with a bogus cursor. On 32-bit architectures gcc
also warns about the cast of the pointer in the error return:
fs_test.c:7474:24: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
Compare against (char *)-ENOMEM instead.
The size left after the escaped path also subtracts the prefix a second
time, which can make a long path fail with -E2BIG. Compute it from the
cursor instead.
Fixes: 316d06b01130 ("selftests/landlock: Add audit tests for filesystem")
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
Found with 32-bit builds of the selftests (gcc 16 and clang 21 with
-m32, with and without -D_TIME_BITS=64 -D_FILE_OFFSET_BITS=64);
KernelCI's 32-bit arm builds show the same warnings. The changed files
build without them now, and the x86_64 builds don't change.
tools/testing/selftests/landlock/fs_test.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/tools/testing/selftests/landlock/fs_test.c b/tools/testing/selftests/landlock/fs_test.c
index f725882fa4a78..27a78434d1c21 100644
--- a/tools/testing/selftests/landlock/fs_test.c
+++ b/tools/testing/selftests/landlock/fs_test.c
@@ -7470,10 +7470,10 @@ static int matches_log_fs_extra(struct __test_metadata *const _metadata,
log_match_cursor = regex_escape(absolute_path, log_match_cursor,
log_match_remaining);
free(absolute_path);
- if (log_match_cursor < 0)
- return (long long)log_match_cursor;
+ if (log_match_cursor == (char *)-ENOMEM)
+ return -ENOMEM;
- log_match_remaining -= log_match_cursor - log_match;
+ log_match_remaining = sizeof(log_match) - (log_match_cursor - log_match);
chunk_len = snprintf(log_match_cursor, log_match_remaining,
"\" dev=\"[^\"]\\+\" ino=[0-9]\\+%s$",
extra ?: "");
--
2.53.0
Thanks for the contribution! I also stumbled across this one when I
did a wider cleanup, and at the time I came to the conclusion that it
would be wiser to fix the weird way in which regex_escape() reports
errors.
https://lore.kernel.org/all/20261008142604.39107-20-gnoack3000@gmail.com/ (local)
Does that seem reasonable to you?
If you are looking into Landlock selftests, you might want to have a
look at that patch set, it already fixes a wide range of issues.
–Günther
P.S.: Mickaël, gentle nudge! I would still appreciate a review on the
selftest cleanup patch set, even though I know that I need to
start rebasing it soon to make it apply cleanly.