libfc: Fix fc_fcp_cleanup_each_cmd()
authorBart Van Assche <bart.vanassche@sandisk.com>
Fri, 5 Jun 2015 21:20:51 +0000 (14:20 -0700)
committerJames Bottomley <JBottomley@Odin.com>
Wed, 12 Aug 2015 18:24:21 +0000 (11:24 -0700)
Since fc_fcp_cleanup_cmd() can sleep this function must not
be called while holding a spinlock. This patch avoids that
fc_fcp_cleanup_each_cmd() triggers the following bug:

BUG: scheduling while atomic: sg_reset/1512/0x00000202
1 lock held by sg_reset/1512:
 #0:  (&(&fsp->scsi_pkt_lock)->rlock){+.-...}, at: [<ffffffffc0225cd5>] fc_fcp_cleanup_each_cmd.isra.21+0xa5/0x150 [libfc]
Preemption disabled at:[<ffffffffc0225cd5>] fc_fcp_cleanup_each_cmd.isra.21+0xa5/0x150 [libfc]
Call Trace:
 [<ffffffff816c612c>] dump_stack+0x4f/0x7b
 [<ffffffff810828bc>] __schedule_bug+0x6c/0xd0
 [<ffffffff816c87aa>] __schedule+0x71a/0xa10
 [<ffffffff816c8ad2>] schedule+0x32/0x80
 [<ffffffffc0217eac>] fc_seq_set_resp+0xac/0x100 [libfc]
 [<ffffffffc0218b11>] fc_exch_done+0x41/0x60 [libfc]
 [<ffffffffc0225cff>] fc_fcp_cleanup_each_cmd.isra.21+0xcf/0x150 [libfc]
 [<ffffffffc0225f43>] fc_eh_device_reset+0x1c3/0x270 [libfc]
 [<ffffffff814a2cc9>] scsi_try_bus_device_reset+0x29/0x60
 [<ffffffff814a3908>] scsi_ioctl_reset+0x258/0x2d0
 [<ffffffff814a2650>] scsi_ioctl+0x150/0x440
 [<ffffffff814b3a9d>] sd_ioctl+0xad/0x120
 [<ffffffff8132f266>] blkdev_ioctl+0x1b6/0x810
 [<ffffffff811da608>] block_ioctl+0x38/0x40
 [<ffffffff811b4e08>] do_vfs_ioctl+0x2f8/0x530
 [<ffffffff811b50c1>] SyS_ioctl+0x81/0xa0
 [<ffffffff816cf8b2>] system_call_fastpath+0x16/0x7a

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: James Bottomley <JBottomley@Odin.com>
drivers/scsi/libfc/fc_fcp.c

index c6795941b45d98579cd6117cc5b72f8c5c4d0bf9..2d5909c4685ca63375f5041d960effada171f5fc 100644 (file)
@@ -1039,11 +1039,26 @@ restart:
                fc_fcp_pkt_hold(fsp);
                spin_unlock_irqrestore(&si->scsi_queue_lock, flags);
 
-               if (!fc_fcp_lock_pkt(fsp)) {
+               spin_lock_bh(&fsp->scsi_pkt_lock);
+               if (!(fsp->state & FC_SRB_COMPL)) {
+                       fsp->state |= FC_SRB_COMPL;
+                       /*
+                        * TODO: dropping scsi_pkt_lock and then reacquiring
+                        * again around fc_fcp_cleanup_cmd() is required,
+                        * since fc_fcp_cleanup_cmd() calls into
+                        * fc_seq_set_resp() and that func preempts cpu using
+                        * schedule. May be schedule and related code should be
+                        * removed instead of unlocking here to avoid scheduling
+                        * while atomic bug.
+                        */
+                       spin_unlock_bh(&fsp->scsi_pkt_lock);
+
                        fc_fcp_cleanup_cmd(fsp, error);
+
+                       spin_lock_bh(&fsp->scsi_pkt_lock);
                        fc_io_compl(fsp);
-                       fc_fcp_unlock_pkt(fsp);
                }
+               spin_unlock_bh(&fsp->scsi_pkt_lock);
 
                fc_fcp_pkt_release(fsp);
                spin_lock_irqsave(&si->scsi_queue_lock, flags);