nfs: take extra reference to fl->fl_file when running a LOCKU operation
authorJeff Layton <jlayton@poochiereds.net>
Tue, 30 Jun 2015 18:12:30 +0000 (14:12 -0400)
committerTrond Myklebust <trond.myklebust@primarydata.com>
Wed, 1 Jul 2015 15:27:27 +0000 (11:27 -0400)
Jean reported another crash, similar to the one fixed by feaff8e5b2cf:

    BUG: unable to handle kernel NULL pointer dereference at 0000000000000148
    IP: [<ffffffff8124ef7f>] locks_get_lock_context+0xf/0xa0
    PGD 0
    Oops: 0000 [#1] SMP
    Modules linked in: nfsv3 nfs_layout_flexfiles rpcsec_gss_krb5 nfsv4 dns_resolver nfs fscache vmw_vsock_vmci_transport vsock cfg80211 rfkill coretemp crct10dif_pclmul ppdev vmw_balloon crc32_pclmul crc32c_intel ghash_clmulni_intel pcspkr vmxnet3 parport_pc i2c_piix4 microcode serio_raw parport nfsd floppy vmw_vmci acpi_cpufreq auth_rpcgss shpchp nfs_acl lockd grace sunrpc vmwgfx drm_kms_helper ttm drm mptspi scsi_transport_spi mptscsih ata_generic mptbase i2c_core pata_acpi
    CPU: 0 PID: 329 Comm: kworker/0:1H Not tainted 4.1.0-rc7+ #2
    Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 07/30/2013
    Workqueue: rpciod rpc_async_schedule [sunrpc]
    30ec000
    RIP: 0010:[<ffffffff8124ef7f>]  [<ffffffff8124ef7f>] locks_get_lock_context+0xf/0xa0
    RSP: 0018:ffff8802330efc08  EFLAGS: 00010296
    RAX: ffff8802330efc58 RBX: ffff880097187c80 RCX: 0000000000000000
    RDX: 0000000000000000 RSI: 0000000000000002 RDI: 0000000000000000
    RBP: ffff8802330efc18 R08: ffff88023fc173d8 R09: 3038b7bf00000000
    R10: 00002f1a02000000 R11: 3038b7bf00000000 R12: 0000000000000000
    R13: 0000000000000000 R14: ffff8802337a2300 R15: 0000000000000020
    FS:  0000000000000000(0000) GS:ffff88023fc00000(0000) knlGS:0000000000000000
    CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
    CR2: 0000000000000148 CR3: 000000003680f000 CR4: 00000000001407f0
    Stack:
     ffff880097187c80 ffff880097187cd8 ffff8802330efc98 ffffffff81250281
     ffff8802330efc68 ffffffffa013e7df ffff8802330efc98 0000000000000246
     ffff8801f6901c00 ffff880233d2b8d8 ffff8802330efc58 ffff8802330efc58
    Call Trace:
     [<ffffffff81250281>] __posix_lock_file+0x31/0x5e0
     [<ffffffffa013e7df>] ? rpc_wake_up_task_queue_locked.part.35+0xcf/0x240 [sunrpc]
     [<ffffffff8125088b>] posix_lock_file_wait+0x3b/0xd0
     [<ffffffffa03890b2>] ? nfs41_wake_and_assign_slot+0x32/0x40 [nfsv4]
     [<ffffffffa0365808>] ? nfs41_sequence_done+0xd8/0x300 [nfsv4]
     [<ffffffffa0367525>] do_vfs_lock+0x35/0x40 [nfsv4]
     [<ffffffffa03690c1>] nfs4_locku_done+0x81/0x120 [nfsv4]
     [<ffffffffa013e310>] ? rpc_destroy_wait_queue+0x20/0x20 [sunrpc]
     [<ffffffffa013e310>] ? rpc_destroy_wait_queue+0x20/0x20 [sunrpc]
     [<ffffffffa013e33c>] rpc_exit_task+0x2c/0x90 [sunrpc]
     [<ffffffffa0134400>] ? call_refreshresult+0x170/0x170 [sunrpc]
     [<ffffffffa013ece4>] __rpc_execute+0x84/0x410 [sunrpc]
     [<ffffffffa013f085>] rpc_async_schedule+0x15/0x20 [sunrpc]
     [<ffffffff810add67>] process_one_work+0x147/0x400
     [<ffffffff810ae42b>] worker_thread+0x11b/0x460
     [<ffffffff810ae310>] ? rescuer_thread+0x2f0/0x2f0
     [<ffffffff810b35d9>] kthread+0xc9/0xe0
     [<ffffffff81010000>] ? perf_trace_xen_mmu_set_pmd+0xa0/0x160
     [<ffffffff810b3510>] ? kthread_create_on_node+0x170/0x170
     [<ffffffff8173c222>] ret_from_fork+0x42/0x70
     [<ffffffff810b3510>] ? kthread_create_on_node+0x170/0x170
    Code: a5 81 e8 85 75 e4 ff c6 05 31 ee aa 00 01 eb 98 66 66 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 55 48 89 e5 41 54 49 89 fc 53 <48> 8b 9f 48 01 00 00 48 85 db 74 08 48 89 d8 5b 41 5c 5d c3 83
    RIP  [<ffffffff8124ef7f>] locks_get_lock_context+0xf/0xa0
     RSP <ffff8802330efc08>
    CR2: 0000000000000148
    ---[ end trace 64484f16250de7ef ]---

The problem is almost exactly the same as the one fixed by feaff8e5b2cf.
We must take a reference to the struct file when running the LOCKU
compound to prevent the final fput from running until the operation is
complete.

Reported-by: Jean Spector <jean@primarydata.com>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Cc: stable@vger.kernel.org
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
fs/nfs/nfs4proc.c

index 605c203de556af3ceec72a1c371c4f17d1f6d0ea..5e7638d8e31bd72e6aaaff6d2012b17bc697f838 100644 (file)
@@ -5484,6 +5484,7 @@ static struct nfs4_unlockdata *nfs4_alloc_unlockdata(struct file_lock *fl,
        atomic_inc(&lsp->ls_count);
        /* Ensure we don't close file until we're done freeing locks! */
        p->ctx = get_nfs_open_context(ctx);
+       get_file(fl->fl_file);
        memcpy(&p->fl, fl, sizeof(p->fl));
        p->server = NFS_SERVER(inode);
        return p;
@@ -5495,6 +5496,7 @@ static void nfs4_locku_release_calldata(void *data)
        nfs_free_seqid(calldata->arg.seqid);
        nfs4_put_lock_state(calldata->lsp);
        put_nfs_open_context(calldata->ctx);
+       fput(calldata->fl.fl_file);
        kfree(calldata);
 }