Re: [PATCH] nfs: decrement nrequests counter before releasing the req

[Benjamin Coddington]

On 28 Sep 2023, at 9:04, Benjamin Coddington wrote:

> On 27 Sep 2023, at 17:16, Scott Mayhew wrote:
>
>> On Wed, 27 Sep 2023, Trond Myklebust wrote:
>>
>>> On Tue, 2023-09-19 at 09:17 -0400, Jeff Layton wrote:
>>>> I hit this panic in testing:
>>>>
>>>> [ 6235.500016] run fstests generic/464 at 2023-09-18 22:51:24
>>>> [ 6288.410761] BUG: kernel NULL pointer dereference, address:
>>>> 0000000000000000
>>>> [ 6288.412174] #PF: supervisor read access in kernel mode
>>>> [ 6288.413160] #PF: error_code(0x0000) - not-present page
>>>> [ 6288.413992] PGD 0 P4D 0
>>>> [ 6288.414603] Oops: 0000 [#1] PREEMPT SMP PTI
>>>> [ 6288.415419] CPU: 0 PID: 340798 Comm: kworker/u18:8 Not tainted
>>>> 6.6.0-rc1-gdcf620ceebac #95
>>>> [ 6288.416538] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009),
>>>> BIOS 1.16.2-1.fc38 04/01/2014
>>>> [ 6288.417701] Workqueue: nfsiod rpc_async_release [sunrpc]
>>>> [ 6288.418676] RIP: 0010:nfs_inode_remove_request+0xc8/0x150 [nfs]
>>>> [ 6288.419836] Code: ff ff 48 8b 43 38 48 8b 7b 10 a8 04 74 5b 48 85
>>>> ff 74 56 48 8b 07 a9 00 00 08 00 74 58 48 8b 07 f6 c4 10 74 50 e8 c8
>>>> 44 b3 d5 <48> 8b 00 f0 48 ff 88 30 ff ff ff 5b 5d 41 5c c3 cc cc cc
>>>> cc 48 8b
>>>> [ 6288.422389] RSP: 0018:ffffbd618353bda8 EFLAGS: 00010246
>>>> [ 6288.423234] RAX: 0000000000000000 RBX: ffff9a29f9a25280 RCX:
>>>> 0000000000000000
>>>> [ 6288.424351] RDX: ffff9a29f9a252b4 RSI: 000000000000000b RDI:
>>>> ffffef41448e3840
>>>> [ 6288.425345] RBP: ffffef41448e3840 R08: 0000000000000038 R09:
>>>> ffffffffffffffff
>>>> [ 6288.426334] R10: 0000000000033f80 R11: ffff9a2a7fffa000 R12:
>>>> ffff9a29093f98c4
>>>> [ 6288.427353] R13: 0000000000000000 R14: ffff9a29230f62e0 R15:
>>>> ffff9a29230f62d0
>>>> [ 6288.428358] FS:  0000000000000000(0000) GS:ffff9a2a77c00000(0000)
>>>> knlGS:0000000000000000
>>>> [ 6288.429513] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>> [ 6288.430427] CR2: 0000000000000000 CR3: 0000000264748002 CR4:
>>>> 0000000000770ef0
>>>> [ 6288.431553] DR0: 0000000000000000 DR1: 0000000000000000 DR2:
>>>> 0000000000000000
>>>> [ 6288.432715] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7:
>>>> 0000000000000400
>>>> [ 6288.433698] PKRU: 55555554
>>>> [ 6288.434196] Call Trace:
>>>> [ 6288.434667]  <TASK>
>>>> [ 6288.435132]  ? __die+0x1f/0x70
>>>> [ 6288.435723]  ? page_fault_oops+0x159/0x450
>>>> [ 6288.436389]  ? try_to_wake_up+0x98/0x5d0
>>>> [ 6288.437044]  ? do_user_addr_fault+0x65/0x660
>>>> [ 6288.437728]  ? exc_page_fault+0x7a/0x180
>>>> [ 6288.438368]  ? asm_exc_page_fault+0x22/0x30
>>>> [ 6288.439137]  ? nfs_inode_remove_request+0xc8/0x150 [nfs]
>>>> [ 6288.440112]  ? nfs_inode_remove_request+0xa0/0x150 [nfs]
>>>> [ 6288.440924]  nfs_commit_release_pages+0x16e/0x340 [nfs]
>>>> [ 6288.441700]  ? __pfx_call_transmit+0x10/0x10 [sunrpc]
>>>> [ 6288.442475]  ? _raw_spin_lock_irqsave+0x23/0x50
>>>> [ 6288.443161]  nfs_commit_release+0x15/0x40 [nfs]
>>>> [ 6288.443926]  rpc_free_task+0x36/0x60 [sunrpc]
>>>> [ 6288.444741]  rpc_async_release+0x29/0x40 [sunrpc]
>>>> [ 6288.445509]  process_one_work+0x171/0x340
>>>> [ 6288.446135]  worker_thread+0x277/0x3a0
>>>> [ 6288.446724]  ? __pfx_worker_thread+0x10/0x10
>>>> [ 6288.447376]  kthread+0xf0/0x120
>>>> [ 6288.447903]  ? __pfx_kthread+0x10/0x10
>>>> [ 6288.448500]  ret_from_fork+0x2d/0x50
>>>> [ 6288.449078]  ? __pfx_kthread+0x10/0x10
>>>> [ 6288.449665]  ret_from_fork_asm+0x1b/0x30
>>>> [ 6288.450283]  </TASK>
>>>> [ 6288.450688] Modules linked in: rpcsec_gss_krb5 auth_rpcgss nfsv4
>>>> dns_resolver nfs lockd grace sunrpc nls_iso8859_1 nls_cp437 vfat fat
>>>> 9p netfs ext4 kvm_intel crc16 mbcache jbd2 joydev kvm xfs irqbypass
>>>> virtio_net pcspkr net_failover psmouse failover 9pnet_virtio cirrus
>>>> drm_shmem_helper virtio_balloon drm_kms_helper button evdev drm loop
>>>> dm_mod zram zsmalloc crct10dif_pclmul crc32_pclmul
>>>> ghash_clmulni_intel sha512_ssse3 sha512_generic virtio_blk nvme
>>>> aesni_intel crypto_simd cryptd nvme_core t10_pi i6300esb
>>>> crc64_rocksoft_generic crc64_rocksoft crc64 virtio_pci virtio
>>>> virtio_pci_legacy_dev virtio_pci_modern_dev virtio_ring serio_raw
>>>> btrfs blake2b_generic libcrc32c crc32c_generic crc32c_intel xor
>>>> raid6_pq autofs4
>>>> [ 6288.460211] CR2: 0000000000000000
>>>> [ 6288.460787] ---[ end trace 0000000000000000 ]---
>>>> [ 6288.461571] RIP: 0010:nfs_inode_remove_request+0xc8/0x150 [nfs]
>>>> [ 6288.462500] Code: ff ff 48 8b 43 38 48 8b 7b 10 a8 04 74 5b 48 85
>>>> ff 74 56 48 8b 07 a9 00 00 08 00 74 58 48 8b 07 f6 c4 10 74 50 e8 c8
>>>> 44 b3 d5 <48> 8b 00 f0 48 ff 88 30 ff ff ff 5b 5d 41 5c c3 cc cc cc
>>>> cc 48 8b
>>>> [ 6288.465136] RSP: 0018:ffffbd618353bda8 EFLAGS: 00010246
>>>> [ 6288.465963] RAX: 0000000000000000 RBX: ffff9a29f9a25280 RCX:
>>>> 0000000000000000
>>>> [ 6288.467035] RDX: ffff9a29f9a252b4 RSI: 000000000000000b RDI:
>>>> ffffef41448e3840
>>>> [ 6288.468093] RBP: ffffef41448e3840 R08: 0000000000000038 R09:
>>>> ffffffffffffffff
>>>> [ 6288.469121] R10: 0000000000033f80 R11: ffff9a2a7fffa000 R12:
>>>> ffff9a29093f98c4
>>>> [ 6288.470109] R13: 0000000000000000 R14: ffff9a29230f62e0 R15:
>>>> ffff9a29230f62d0
>>>> [ 6288.471106] FS:  0000000000000000(0000) GS:ffff9a2a77c00000(0000)
>>>> knlGS:0000000000000000
>>>> [ 6288.472216] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>> [ 6288.473059] CR2: 0000000000000000 CR3: 0000000264748002 CR4:
>>>> 0000000000770ef0
>>>> [ 6288.474096] DR0: 0000000000000000 DR1: 0000000000000000 DR2:
>>>> 0000000000000000
>>>> [ 6288.475097] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7:
>>>> 0000000000000400
>>>> [ 6288.476148] PKRU: 55555554
>>>> [ 6288.476665] note: kworker/u18:8[340798] exited with irqs disabled
>>>>
>>>> Once we've released "req", it's not safe to dereference it anymore.
>>>> Decrement the nrequests counter before dropping the reference.
>>>>
>>>> Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
>>>> ---
>>>> I've only hit this once after a lot of testing, so I can't confirm
>>>> that
>>>> this fixes anything. It seems like the right thing to do, however.
>>>> ---
>>>>  fs/nfs/write.c | 2 +-
>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/fs/nfs/write.c b/fs/nfs/write.c
>>>> index 8c1ee1a1a28f..7720b5e43014 100644
>>>> --- a/fs/nfs/write.c
>>>> +++ b/fs/nfs/write.c
>>>> @@ -802,8 +802,8 @@ static void nfs_inode_remove_request(struct
>>>> nfs_page *req)
>>>>         }
>>>>  
>>>>         if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) {
>>>> -               nfs_release_request(req);
>>>>                 atomic_long_dec(&NFS_I(nfs_page_to_inode(req))-
>>>>> nrequests);
>>>> +               nfs_release_request(req);
>>>>         }
>>>>  }
>>>>
>>>
>>> Isn't this the same issue that Scott Mayhew posted a patch ("NFS: Fix
>>> potential oops in nfs_inode_remove_request()") on July 25th?
>>>
>>> At the time I argued for something like the above patch, but both Scott
>>> and you argued that it was insufficient to fix the problem.
>>>
>>> So which patch is the one we should apply?
>>
>> This one.  If there's a separate bug, I've had zero luck in
>> reproducing it.
>
> Would you say you've had zero luck reproducing this one too?
>
> I think your patch on July 25th fixes a real bug based on that vmcore we
> have, but it's a pretty crazy race.   I'll try again to reproduce it.
>
> Ben