All of lore.kernel.org
 help / color / mirror / Atom feed
From: bugzilla-daemon@kernel.org
To: linux-ext4@vger.kernel.org
Subject: [Bug 222294] New: [REGRESSION] 6.1.191+: NULL deref in ext4_mb_load_buddy_gfp() after a3a7d360252c ("ext4: fix e4b bitmap inconsistency reports")
Date: Thu, 19 Nov 2026 09:51:01 +0000	[thread overview]
Message-ID: <bug-222294-13602@https.bugzilla.kernel.org/> (raw)

https://bugzilla.kernel.org/show_bug.cgi?id=222294

            Bug ID: 222294
           Summary: [REGRESSION] 6.1.191+: NULL deref in
                    ext4_mb_load_buddy_gfp() after a3a7d360252c ("ext4:
                    fix e4b bitmap inconsistency reports")
           Product: File System
           Version: 2.5
          Hardware: Intel
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P3
         Component: ext4
          Assignee: fs_ext4@kernel-bugs.osdl.org
          Reporter: internet.jejkowice@gmail.com
        Regression: No

Created attachment 311518
  --> https://bugzilla.kernel.org/attachment.cgi?id=311518&action=edit
patch

Regression in the 6.1.y stable series, introduced in 6.1.191 and still
present in 6.1.192. 6.1.190 is fine.

REPRODUCER (100% on 6.1.191/6.1.192, 0% on 6.1.190):

    sync; echo 1 > /proc/sys/vm/drop_caches

The machine oopses immediately (kernel.panic_on_oops=1, so it reboots).
Without manual action it also hit during a nightly cron job (rsync doing
unlink()/rename() on an ext4 volume) at 04:45, after the page cache had
been churned by earlier I/O.

SYSTEM
Dell PowerEdge R820, BIOS 2.7.0, x86_64, 40 CPUs, CONFIG_PREEMPT,
ext4 on megaraid_sas. Self-built kernel from kernel.org tarballs.
Tainted (O): out-of-tree vboxdrv/vboxnetflt/vboxnetadp and xt_IPMARK,
none of which appear in the traces.
<.config attached / available on request>

FIRST OOPS (rsync, unlink path)

  Oops: 0000 [#1] PREEMPT SMP
  CPU: 15 PID: 73107 Comm: rsync Tainted: G O 6.1.192 #1
  RIP: 0010:ext4_mb_load_buddy_gfp+0xbc/0x390 [ext4]
  Code: ... 49 89 c7 48 3d 00 f0 ff ff 0f 87 3a 01 00 00 <48> 8b 00 a8 04
        0f 84 24 01 00 00 ...
  RAX: 0000000000000000 RBX: 00000000000003e4 ...
  CR2: 0000000000000000
  Call Trace:
   ext4_free_blocks+0x3a7/0xb30 [ext4]
   ext4_ext_remove_space+0x80f/0x1570 [ext4]
   ext4_ext_truncate+0xb4/0xc0 [ext4]
   ext4_truncate+0x310/0x480 [ext4]
   ext4_evict_inode+0x308/0x720 [ext4]
   evict+0x103/0x2b0
   do_unlinkat+0x143/0x2f0
   __x64_sys_unlink+0x3e/0x60
   do_syscall_64+0x35/0x90

A second rsync hit the same RIP via the allocation path
(ext4_mb_regular_allocator <- ext4_mb_new_blocks <- ext4_ext_map_blocks <-
ext4_writepages <- ext4_rename), for a different group (0x3fa vs 0x3e4).
Everything after that was fallout: BSD process accounting writes from
do_exit() into the already-broken ext4 led to further oopses, a dead
mutex waiter on a freed stack, and finally soft lockups on the
acct mutex's wait_lock until the box was reset.

CAUSE
Diffing 6.1.190 -> 6.1.191, the regression comes from
  a3a7d360252c ("ext4: fix e4b bitmap inconsistency reports")
  [upstream bdc56a9c46b2]
preceded in the same release by the preparatory commits
  e394adafb089, e6b6cbe9f2fb, 1c0f0e2b9b69
which are not themselves the cause.

The patch converts ext4_mb_load_buddy_gfp() and
ext4_mb_get_buddy_page_lock() to __filemap_get_folio() and checks the
result with IS_ERR() only. In 6.1.y, __filemap_get_folio() never returns
ERR_PTR() (mm/filemap.c, no ERR_PTR anywhere in the function). It returns
NULL when:
 - the lookup misses and FGP_CREAT is not set (falls through no_page to
   the final "return folio;" with folio == NULL),
 - FGP_NOWAIT and folio_trylock() fails,
 - filemap_alloc_folio() fails,
 - filemap_add_folio() fails (folio = NULL).

So in the fast path
    folio = __filemap_get_folio(mapping, pnum, FGP_ACCESSED, 0);
    if (IS_ERR(folio) || !folio_test_uptodate(folio) || ...)
a cache miss reaches folio_test_uptodate(NULL) and dereferences address 0.
This matches the oops: "cmp $0xfffff000,%rax; ja" is IS_ERR(), followed
by "mov (%rax),%rax; test $4,%al" (folio->flags, PG_uptodate) with
RAX = 0, CR2 = 0. The FGP_CREAT slow path has the same flaw
(PTR_ERR(NULL) == 0, so ret = 0 and the NULL folio is used), as does
ext4_mb_get_buddy_page_lock() (folio->mapping on NULL).

Upstream (>= v6.3) returns ERR_PTR from __filemap_get_folio(), so only
stable branches that received this backport without adapting it are
affected. 5.15.y and 5.10.y may be affected too if they got the same
backport; I have not checked.

WORKAROUND / FIX
Wrapper that maps NULL to ERR_PTR(-ENOMEM), used for all six
__filemap_get_folio() calls in fs/ext4/mballoc.c:

    static struct folio *ext4_mb_get_folio(struct address_space *mapping,
                    pgoff_t index, int fgp, gfp_t gfp)
    {
            struct folio *folio = __filemap_get_folio(mapping, index, fgp,
gfp);

            return folio ? folio : ERR_PTR(-ENOMEM);
    }

With this applied on 6.1.192, 4x "echo 1" and 7x "echo 3" to drop_caches
no longer crash (previously a single "echo 1" did).
<Result of cron.daily rsync after drop_caches with the patch: ...>
Patch attached; also sent to stable@vger.kernel.org and
linux-ext4@vger.kernel.org: <lore.kernel.org link>

Reverting a3a7d360252c is the other option.

-- 
You are receiving this mail because:
You are watching the assignee of the bug.

             reply	other threads:[~2026-11-19  9:51 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-19  9:51 bugzilla-daemon [this message]
2026-11-19  9:53 ` [Bug 222294] [REGRESSION] 6.1.191+: NULL deref in ext4_mb_load_buddy_gfp() after a3a7d360252c ("ext4: fix e4b bitmap inconsistency reports") bugzilla-daemon
2026-11-19 10:06 ` bugzilla-daemon
2026-11-19 10:13 ` bugzilla-daemon
2026-11-20  7:03 ` bugzilla-daemon

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=bug-222294-13602@https.bugzilla.kernel.org/ \
    --to=bugzilla-daemon@kernel.org \
    --cc=linux-ext4@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.