RDMA/cma: Don't overwrite sgid_attr after device is released
authorLeon Romanovsky <leonro@nvidia.com>
Sun, 13 Dec 2020 13:29:39 +0000 (15:29 +0200)
committerJason Gunthorpe <jgg@nvidia.com>
Mon, 14 Dec 2020 19:23:06 +0000 (15:23 -0400)
As part of the cma_dev release, that pointer will be set to NULL.  In case
it happens in rdma_bind_addr() (part of an error flow), the next call to
addr_handler() will have a call to cma_acquire_dev_by_src_ip() which will
overwrite sgid_attr without releasing it.

  WARNING: CPU: 2 PID: 108 at drivers/infiniband/core/cma.c:606 cma_bind_sgid_attr drivers/infiniband/core/cma.c:606 [inline]
  WARNING: CPU: 2 PID: 108 at drivers/infiniband/core/cma.c:606 cma_acquire_dev_by_src_ip+0x470/0x4b0 drivers/infiniband/core/cma.c:649
  CPU: 2 PID: 108 Comm: kworker/u8:1 Not tainted 5.10.0-rc6+ #257
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
  Workqueue: ib_addr process_one_req
  RIP: 0010:cma_bind_sgid_attr drivers/infiniband/core/cma.c:606 [inline]
  RIP: 0010:cma_acquire_dev_by_src_ip+0x470/0x4b0 drivers/infiniband/core/cma.c:649
  Code: 66 d9 4a ff 4d 8b 6e 10 49 8d bd 1c 08 00 00 e8 b6 d6 4a ff 45 0f b6 bd 1c 08 00 00 41 83 e7 01 e9 49 fd ff ff e8 90 c5 29 ff <0f> 0b e9 80 fe ff ff e8 84 c5 29 ff 4c 89 f7 e8 2c d9 4a ff 4d 8b
  RSP: 0018:ffff8881047c7b40 EFLAGS: 00010293
  RAX: ffff888104789c80 RBX: 0000000000000001 RCX: ffffffff820b8ef8
  RDX: 0000000000000000 RSI: ffffffff820b9080 RDI: ffff88810cd4c998
  RBP: ffff8881047c7c08 R08: ffff888104789c80 R09: ffffed10209f4036
  R10: ffff888104fa01ab R11: ffffed10209f4035 R12: ffff88810cd4c800
  R13: ffff888105750e28 R14: ffff888108f0a100 R15: ffff88810cd4c998
  FS:  0000000000000000(0000) GS:ffff888119c00000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 0000000000000000 CR3: 0000000104e60005 CR4: 0000000000370ea0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
  Call Trace:
   addr_handler+0x266/0x350 drivers/infiniband/core/cma.c:3190
   process_one_req+0xa3/0x300 drivers/infiniband/core/addr.c:645
   process_one_work+0x54c/0x930 kernel/workqueue.c:2272
   worker_thread+0x82/0x830 kernel/workqueue.c:2418
   kthread+0x1ca/0x220 kernel/kthread.c:292
   ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:296

Fixes: ff11c6cd521f ("RDMA/cma: Introduce and use cma_acquire_dev_by_src_ip()")
Link: https://lore.kernel.org/r/20201213132940.345554-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
drivers/infiniband/core/cma.c

index c06c87a4dc5e7d2ba41578e91a9fd07c999cfd40..c51b84b2d2f3750ef1c9d183640d4119a2a57a20 100644 (file)
@@ -477,6 +477,10 @@ static void cma_release_dev(struct rdma_id_private *id_priv)
        list_del(&id_priv->list);
        cma_dev_put(id_priv->cma_dev);
        id_priv->cma_dev = NULL;
+       if (id_priv->id.route.addr.dev_addr.sgid_attr) {
+               rdma_put_gid_attr(id_priv->id.route.addr.dev_addr.sgid_attr);
+               id_priv->id.route.addr.dev_addr.sgid_attr = NULL;
+       }
        mutex_unlock(&lock);
 }
 
@@ -1861,9 +1865,6 @@ static void _destroy_id(struct rdma_id_private *id_priv,
 
        kfree(id_priv->id.route.path_rec);
 
-       if (id_priv->id.route.addr.dev_addr.sgid_attr)
-               rdma_put_gid_attr(id_priv->id.route.addr.dev_addr.sgid_attr);
-
        put_net(id_priv->id.route.addr.dev_addr.net);
        rdma_restrack_del(&id_priv->res);
        kfree(id_priv);