On 2024/10/10 10:03, Kent Overstreet wrote: > On Wed, Oct 09, 2024 at 10:07:55PM GMT, Zhao Mengmeng wrote: >> From: Zhao Mengmeng <[email protected]> >> >> syzbot report a shift-out-of-bounds issue: >> ------------[ cut here ]------------ >> UBSAN: shift-out-of-bounds in fs/bcachefs/ec.c:147:2 >> shift exponent 108 is too large for 32-bit type 'unsigned int' >> ---- >> Here s.csum_granularity_bits = 108, so shift is impossible for unsigned >> int. To fix, add a check in bch2_stripe_validate() to bail out, it has >> same checking logic with ec_stripe_key_init(). >> >> Reported-by: [email protected] >> Tested-by: [email protected] >> Closes: https://syzkaller.appspot.com/bug?extid=f8c98a50c323635be65d >> Suggested-by: Hongbo Li <[email protected]> >> Signed-off-by: Zhao Mengmeng <[email protected]> >> --- >> fs/bcachefs/ec.c | 6 ++++++ >> fs/bcachefs/errcode.h | 4 +++- >> 2 files changed, 9 insertions(+), 1 deletion(-) >> >> diff --git a/fs/bcachefs/ec.c b/fs/bcachefs/ec.c >> index 564841e5a24b..f6a02123144d 100644 >> --- a/fs/bcachefs/ec.c >> +++ b/fs/bcachefs/ec.c >> @@ -114,6 +114,12 @@ int bch2_stripe_validate(struct bch_fs *c, struct >> bkey_s_c k, >> const struct bch_stripe *s = bkey_s_c_to_stripe(k).v; >> int ret = 0; >> >> + if (s->csum_granularity_bits >= ilog2(le16_to_cpu(s->sectors))) { >> + bch_err_ratelimited(c, "stripe csum gran bits %u too big", >> + s->csum_granularity_bits); >> + return -BCH_ERR_stripe_csum_granularity_bits_too_big; >> + } >> + > > that should be a bkey_fsck_err_on(), and you'll add to the enum in > sb-errors_format.h
Hi Kent, I try to use bkey_fsck_err_on() at the first time, but it still triggered syzbot warning, cause bch2_stripe_to_text() is called before __bch2_fsck_err(), before it throws new errro, it will trigger the shift out of bounds check. Cheers!
