Re: __fput and with silly delegation current_umask()

[Trond Myklebust <trondmy@hammerspace.com>]

Hi Ben,

On Fri, 2019-05-10 at 16:39 -0400, Benjamin Coddington wrote:
> A shiney new version is out, and I just had a report of a crash:
> 
> [ 2853.916121] BUG: unable to handle kernel NULL pointer dereference
> at 
> 000000000000000c
> [ 2853.917625] PGD 0 P4D 0
> [ 2853.918106] Oops: 0000 [#1] SMP PTI
> [ 2853.918761] CPU: 0 PID: 25711 Comm: file_exec.sh Kdump: loaded
> Not 
> tainted 4.18.0-83.el8.x86_64 #1
> [ 2853.920406] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
> [ 2853.921469] RIP: 0010:current_umask+0x15/0x20
> [ 2853.922274] Code: 76 48 e9 3e 54 fe ff 90 90 90 90 90 90 90 90 90
> 90 
> 90 90 90 90 0f 1f 44 00 00 65 48 8b 04 25 80 5c 01 00 48 8b 80 18 0b
> 00 
> 00 <8b> 40 0c c3 0f 1f 80 00 00 00 00 0f 1f 44 00 00 55 48 89 e5 41
> 55
> [ 2853.925675] RSP: 0018:ffffa446014d3ba8 EFLAGS: 00010246
> [ 2853.926644] RAX: 0000000000000000 RBX: ffff93ccc402e000 RCX: 
> 0000000500000000
> [ 2853.927939] RDX: 0000000000000000 RSI: 0000000500000000 RDI: 
> ffff93cd494fa6d8
> [ 2853.929200] RBP: ffff93cd494fa680 R08: ffff93cd7ba28160 R09: 
> ffff93cd47c02380
> [ 2853.930489] R10: ffffe964443aa6c0 R11: ffffffffc0967080 R12: 
> 0000000000000000
> [ 2853.931957] R13: 0000000000000000 R14: ffff93cd54d3e800 R15: 
> ffff93ccc4076400
> [ 2853.933420] FS:  00007f31147b8740(0000) GS:ffff93cd7ba00000(0000) 
> knlGS:0000000000000000
> [ 2853.935071] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 2853.936262] CR2: 000000000000000c CR3: 000000000aa0a004 CR4: 
> 00000000007606f0
> [ 2853.937730] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 
> 0000000000000000
> [ 2853.939186] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 
> 0000000000000400
> [ 2853.940657] PKRU: 55555554
> [ 2853.941226] Call Trace:
> [ 2853.941729]  nfs4_opendata_alloc+0x135/0x460 [nfsv4]
> [ 2853.942655]  nfs4_open_recoverdata_alloc.isra.59+0x21/0x40 [nfsv4]
> [ 2853.943854]  nfs4_open_delegation_recall+0x40/0x140 [nfsv4]
> [ 2853.945000]  ? __filemap_fdatawrite_range+0xc8/0xf0
> [ 2853.946020]  nfs_end_delegation_return+0x15d/0x390 [nfsv4]
> [ 2853.947162]  nfs_complete_unlink+0x15e/0x220 [nfs]
> [ 2853.948160]  nfs_dentry_iput+0x37/0x50 [nfs]
> [ 2853.949046]  __dentry_kill+0xd5/0x170
> [ 2853.949807]  dentry_kill+0x4d/0x190
> [ 2853.950537]  dput.part.32+0xcb/0x110
> [ 2853.951285]  __fput+0x108/0x220
> [ 2853.951942]  task_work_run+0x8a/0xb0
> [ 2853.952665]  do_exit+0x2db/0xad0
> [ 2853.953266]  ? syscall_trace_enter+0x1d3/0x2c0
> [ 2853.954152]  do_group_exit+0x3a/0xa0
> [ 2853.954902]  __x64_sys_exit_group+0x14/0x20
> [ 2853.955742]  do_syscall_64+0x5b/0x1b0
> [ 2853.956422]  entry_SYSCALL_64_after_hwframe+0x65/0xca
> [ 2853.957366] RIP: 0033:0x7f3113e76a86
> [ 2853.958031] Code: Bad RIP value.
> [ 2853.958637] RSP: 002b:00007fff7ab7de18 EFLAGS: 00000246 ORIG_RAX: 
> 00000000000000e7
> [ 2853.960023] RAX: ffffffffffffffda RBX: 00007f3114167740 RCX: 
> 00007f3113e76a86
> [ 2853.961328] RDX: 000000000000007e RSI: 000000000000003c RDI: 
> 000000000000007e
> [ 2853.962636] RBP: 000000000000007e R08: 00000000000000e7 R09: 
> ffffffffffffff80
> [ 2853.963941] R10: 00007fff7ab7dccc R11: 0000000000000246 R12: 
> 00007f3114167740
> [ 2853.965240] R13: 0000000000000001 R14: 00007f3114170448 R15: 
> 0000000000000000
> 
> The problem seems to be that current->fs = NULL at this point, so
> current_umask() in nfs4_opendata_alloc() shouldn't be used in this
> path. 
>   We
> do exit_fs before task_work_run starts us cleaning up the inode, so
> this
> isn't /delayed/ __fput, but it does have things happening in the
> wrong 
> order
> if we want to use current->fs.
> 
> Should we check for this specific case and use a static umask, or
> maybe 
> fix
> current_umask() for all delayed fput() paths?  A umask doesn't
> really 
> make
> any sense if we're doing doing OPEN with CLAIM_DELEG_CUR_FH..
> 
> I'm feeling unhappy with any options that spring to mind.
> 
> Any other suggestions?  Maybe the fix will be obvious after a nap...
> 
The current->mask() should only be needed if we're creating a new file,
so it should not be necessary here. So perhaps you could add a test for
O_CREAT to nfs4_opendata_alloc()?

Other things that should not be needed when we don't set O_CREAT
include p->o_arg.u.attrs, p->o_arg.label and p->o_arg.u.verifier.

Cheers
  Trond
-- 
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@hammerspace.com