Package: src:linux
Version: 6.1.192-1
Severity: important
Tags: regression

Dear Maintainer,

after unattended-upgrades installed linux-image-6.1.0-55-cloud-amd64
(6.1.192-1, built 2026-11-14) on a Debian 12 (bookworm) Google Compute
Engine VM, the machine started to hang roughly two hours after every boot.
The kernel oopses in ext4_mb_load_buddy_gfp() and from then on all I/O to
the root ext4 filesystem blocks, so every service on the VM stops
responding until a hard reset.

Environment:
- Google Compute Engine, e2-small (2 vCPU shared, 2 GiB RAM), europe-west1
- Debian 12 cloud image, Secure Boot enabled
- Root filesystem: ext4 on /dev/sda1 (~40 GiB pd-standard), no quota
- Workload: Docker (containerd) running a small Node.js + SQLite app, plus a
systemd timer running a Python script that writes and rotates backup
files (write to temp file + rename())

What happens:
On 2026-11-17 the VM hung twice. The second time the full oops was captured
on the serial console (uptime 7443 s). The task was the Python backup script
calling rename(); the replaced target inode is evicted and truncated, and
ext4_free_blocks() -> ext4_mb_load_buddy_gfp() dereferences NULL:

[ 7443.456579] BUG: kernel NULL pointer dereference, address:
0000000000000000
[ 7443.463974] #PF: supervisor read access in kernel mode
[ 7443.469358] #PF: error_code(0x0000) - not-present page
[ 7443.474659] PGD 800000007f7ba067 P4D 800000007f7ba067 PUD 5350067 PMD 0
[ 7443.481679] Oops: 0000 [#1] PREEMPT SMP PTI
[ 7443.486160] CPU: 0 PID: 24869 Comm: python3 Not tainted
6.1.0-55-cloud-amd64 #1 Debian 6.1.192-1
[ 7443.495250] Hardware name: Google Google Compute Engine/Google Compute
Engine, BIOS Google 10/07/2026
[ 7443.504604] RIP: 0010:ext4_mb_load_buddy_gfp+0xbc/0x380
[ 7443.510120] Code: 89 44 24 08 f7 3c 24 48 63 d8 89 54 24 0c ba 01 00 00
00 48 89 de e8 33 d8 e3 ff 49 89 c7 48 3d 00 f0 ff ff 0f 87 28 01 00 00
<48> 8b 00 a8 04 0f 84 12 01 00 00 49 8b 07 a8 01 0f 85 07 01 00 00
[ 7443.529191] RSP: 0018:ffffcf2fc393b900 EFLAGS: 00010207
[ 7443.534561] RAX: 0000000000000000 RBX: 00000000000000a8 RCX:
0000000000000003
[ 7443.542002] RDX: 0000000000000028 RSI: ffff8a3d3ed44490 RDI:
ffff8a3cc2b13280
[ 7443.549378] RBP: ffff8a3d3ea7e000 R08: ffff8a3d3efb1c98 R09:
0000000000000000
[ 7443.556717] R10: ffffcf2fc393b998 R11: 0000000000000000 R12:
ffffcf2fc393b998
[ 7443.563972] R13: 0000000000008c40 R14: ffff8a3d3eebcf20 R15:
0000000000000000
[ 7443.571611] FS: 00007f5a1a804040(0000) GS:ffff8a3d3c000000(0000)
knlGS:0000000000000000
[ 7443.579910] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 7443.585791] CR2: 0000000000000000 CR3: 0000000002ac0004 CR4:
00000000003706b0
[ 7443.608029] Call Trace:
[ 7443.610875] <TASK>
[ 7443.613782] ext4_free_blocks+0x410/0xbb0
[ 7443.618172] ext4_ext_remove_space+0x7ff/0x15d0
[ 7443.623060] ext4_ext_truncate+0xb4/0xd0
[ 7443.627642] ext4_truncate+0x314/0x490
[ 7443.632182] ? __ext4_journal_start_sb+0x105/0x120
[ 7443.637353] ext4_evict_inode+0x2f8/0x710
[ 7443.641687] evict+0x112/0x2c0
[ 7443.645209] ? fsnotify_grab_connector+0x43/0x80
[ 7443.650170] ? fsnotify_destroy_marks+0x26/0x180
[ 7443.655002] ? _atomic_dec_and_lock+0x35/0x50
[ 7443.659758] __dentry_kill+0xdc/0x170
[ 7443.663706] dput+0x132/0x310
[ 7443.666797] do_renameat2+0x375/0x5f0
[ 7443.670920] __x64_sys_rename+0x3f/0x50
[ 7443.675239] do_syscall_64+0x5d/0x90
[ 7443.731919] entry_SYSCALL_64_after_hwframe+0x6e/0xd8
[ 7443.806698] </TASK>
[ 7443.809025] Modules linked in: veth nf_conntrack_netlink xt_nat
xt_tcpudp xt_conntrack xt_MASQUERADE bridge stp llc xfrm_user xfrm_algo
xt_set ip_set nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6
nf_defrag_ipv4 xt_addrtype nft_compat nf_tables libcrc32c nfnetlink
vsock_loopback vmw_vsock_virtio_transport_common vmw_vsock_vmci_transport
vsock vmw_vmci nvme_fabrics overlay intel_rapl_msr intel_rapl_common
iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel sha512_ssse3
sha512_generic sha256_ssse3 binfmt_misc sha1_ssse3 aesni_intel crypto_simd
cryptd nls_ascii nls_cp437 vfat rapl fat pvpanic_mmio pvpanic
virtio_balloon evdev button serio_raw sg fuse loop efi_pstore dm_mod
configfs efivarfs virtio_rng ip_tables x_tables autofs4 sd_mod virtio_scsi
scsi_mod virtio_net net_failover failover scsi_common virtio_pci
virtio_pci_legacy_dev virtio_pci_modern_dev crc32c_intel virtio virtio_ring
[ 7443.891059] CR2: 0000000000000000
[ 7443.894512] ---[ end trace 0000000000000000 ]---

The faulting instruction is "mov (%rax),%rax" right after the call that
returns the buddy folio/page, i.e. the returned pointer is NULL and is
dereferenced (looks like a folio/page flags check on a NULL folio).

The first hang earlier the same day had identical symptoms (VM alive, all
I/O stuck after ~2 h of uptime), but its log was not preserved.

Workaround:
Booting the previous ABI, linux-image-6.1.0-54-cloud-amd64, via GRUB_DEFAULT
and holding linux-image-cloud-amd64. A forced fsck of the root filesystem
after the crashes found no errors. The VM is now running on 6.1.0-54 and I
will report back if the issue reappears there.

Possibly related:
During the 6.1.167-rc1 stable review (March 2026) a very similar NULL
dereference in ext4_mb_load_buddy_gfp() was reported and bisected to the
backport of "ext4: convert bd_buddy_page to bd_buddy_folio":
https://www.google.com/url?q=https://lkml.iu.edu/2603.3/01511.html&source=gmail&ust=1795033009329000&sa=E
https://www.google.com/url?q=https://lkml.iu.edu/2603.3/00952.html&source=gmail&ust=1795033009329000&sa=E
Upstream later fixed related error handling in that function with
"ext4: fix error pointer dereference in ext4_mb_load_buddy_gfp()"
(c6a6c9694aad). I have not verified whether 6.1.192 carries a similar
partial backport.

I can provide more information or test a candidate kernel if needed.

Thank you,
Lukáš Molík

Reply via email to