fcoe: ensure that skb placed on the fip_recv_list are unshared
authorNeil Horman <nhorman@tuxdriver.com>
Sat, 3 Aug 2013 10:45:53 +0000 (10:45 +0000)
committerRobert Love <robert.w.love@intel.com>
Wed, 4 Sep 2013 18:48:19 +0000 (11:48 -0700)
commitc0866286f10964e61ec10c8c605ef86e65fbbd38
tree3f4773cde5aa4b4e05485e61e6743376bcd03ed6
parent6e4664525b1db28f8c4e1130957f70a94c19213e
fcoe: ensure that skb placed on the fip_recv_list are unshared

Recently had this Oops reported to me on the 3.10 kernel:

[  807.554955] BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
[  807.562799] IP: [<ffffffff814e6fc7>] skb_dequeue+0x47/0x70
[  807.568296] PGD 20c889067 PUD 20c8b8067 PMD 0
[  807.572769] Oops: 0002 [#1] SMP
[  807.655597] Hardware name: Dell Inc. PowerEdge R415/0DDT2D, BIOS 1.8.6 12/06/2011
[  807.663079] Workqueue: events fcoe_ctlr_recv_work [libfcoe]
[  807.668656] task: ffff88020b42a160 ti: ffff88020ae6c000 task.ti: ffff88020ae6c000
[  807.676126] RIP: 0010:[<ffffffff814e6fc7>]  [<ffffffff814e6fc7>] skb_dequeue+0x47/0x70
[  807.684046] RSP: 0000:ffff88020ae6dd70  EFLAGS: 00010097
[  807.689349] RAX: 0000000000000246 RBX: ffff8801d04d6700 RCX: 0000000000000000
[  807.696474] RDX: 0000000000000000 RSI: 0000000000000246 RDI: ffff88020df26434
[  807.703598] RBP: ffff88020ae6dd88 R08: 00000000000173e0 R09: ffff880216e173e0
[  807.710723] R10: ffffffff814e5897 R11: ffffea0007413580 R12: ffff88020df26420
[  807.717847] R13: ffff88020df26434 R14: 0000000000000004 R15: ffff8801d04c42ce
[  807.724972] FS:  00007fdaab6048c0(0000) GS:ffff880216e00000(0000) knlGS:0000000000000000
[  807.733049] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[  807.738785] CR2: 0000000000000008 CR3: 000000020cbc9000 CR4: 00000000000006f0
[  807.745910] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  807.753033] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[  807.760156] Stack:
[  807.762162]  ffff8801d04d6700 0000000000000001 ffff88020df26400 ffff88020ae6de20
[  807.769586]  ffffffffa0444409 ffff88020b046a00 ffff88020ae6dde8 ffffffff810105be
[  807.777008]  ffff88020b42a868 0000000000000000 ffff88020df264a8 ffff88020df26348
[  807.784431] Call Trace:
[  807.786885]  [<ffffffffa0444409>] fcoe_ctlr_recv_work+0x59/0x9a0 [libfcoe]
[  807.793755]  [<ffffffff810105be>] ? __switch_to+0x13e/0x4a0
[  807.799324]  [<ffffffff8107d0e6>] process_one_work+0x176/0x420
[  807.805151]  [<ffffffff8107dd0b>] worker_thread+0x11b/0x3a0
[  807.810717]  [<ffffffff8107dbf0>] ? rescuer_thread+0x350/0x350
[  807.816545]  [<ffffffff810842b0>] kthread+0xc0/0xd0
[  807.821416]  [<ffffffff810841f0>] ? insert_kthread_work+0x40/0x40
[  807.827503]  [<ffffffff8160ce2c>] ret_from_fork+0x7c/0xb0
[  807.832897]  [<ffffffff810841f0>] ? insert_kthread_work+0x40/0x40
[  807.858500] RIP  [<ffffffff814e6fc7>] skb_dequeue+0x47/0x70
[  807.864076]  RSP <ffff88020ae6dd70>
[  807.867558] CR2: 0000000000000008

Looks like the root cause is the fact that the packet recieve function
fcoe_ctlr_recv enqueues the skb to a sk_buff_head_list prior to ensuring that
the skb is unshared.  This can happen when multiple packet listeners recieve an
skb, as the deliver_skb function just increments skb->users for each handler.
As a result, having multiple users of a single skb results in multiple
manipulators of its methods, implying list corruption, and the oops recorded
above.

The fix is pretty easy, just make sure that we clone the skb if its got multiple
users with the skb_share_check function, like other protocols do.

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
drivers/scsi/fcoe/fcoe_ctlr.c