• Selvin Xavier's avatar
    RDMA/bnxt_re: Avoid freeing MR resources if dereg fails · 9a4467a6
    Selvin Xavier authored
    The driver returns an error code for MR dereg, but frees the MR structure.
    When the MR dereg is retried due to previous error, the system crashes as
    the structure is already freed.
    
      BUG: unable to handle kernel NULL pointer dereference at 00000000000001b8
      PGD 0 P4D 0
      Oops: 0000 [#1] SMP PTI
      CPU: 7 PID: 12178 Comm: ib_send_bw Kdump: loaded Not tainted 4.18.0-124.el8.x86_64 #1
      Hardware name: Dell Inc. PowerEdge R430/03XKDV, BIOS 1.1.10 03/10/2015
      RIP: 0010:__dev_printk+0x2a/0x70
      Code: 0f 1f 44 00 00 49 89 d1 48 85 f6 0f 84 f6 2b 00 00 4c 8b 46 70 4d 85 c0 75 04 4c 8b
    46 10 48 8b 86 a8 00 00 00 48 85 c0 74 16 <48> 8b 08 0f be 7f 01 48 c7 c2 13 ac ac 83 83 ef 30 e9 10 fe ff ff
      RSP: 0018:ffffaf7c04607a60 EFLAGS: 00010006
      RAX: 00000000000001b8 RBX: ffffa0010c91c488 RCX: 0000000000000246
      RDX: ffffaf7c04607a68 RSI: ffffa0010c91caa8 RDI: ffffffff83a788eb
      RBP: ffffaf7c04607ac8 R08: 0000000000000000 R09: ffffaf7c04607a68
      R10: 0000000000000000 R11: 0000000000000001 R12: ffffaf7c04607b90
      R13: 000000000000000e R14: 0000000000000000 R15: 00000000ffffa001
      FS:  0000146fa1f1cdc0(0000) GS:ffffa0012fac0000(0000) knlGS:0000000000000000
      CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
      CR2: 00000000000001b8 CR3: 000000007680a003 CR4: 00000000001606e0
      Call Trace:
       dev_err+0x6c/0x90
       ? dev_printk_emit+0x4e/0x70
       bnxt_qplib_rcfw_send_message+0x594/0x660 [bnxt_re]
       ? dev_err+0x6c/0x90
       bnxt_qplib_free_mrw+0x80/0xe0 [bnxt_re]
       bnxt_re_dereg_mr+0x2e/0xd0 [bnxt_re]
       ib_dereg_mr+0x2f/0x50 [ib_core]
       destroy_hw_idr_uobject+0x20/0x70 [ib_uverbs]
       uverbs_destroy_uobject+0x2e/0x170 [ib_uverbs]
       __uverbs_cleanup_ufile+0x6e/0x90 [ib_uverbs]
       uverbs_destroy_ufile_hw+0x61/0x130 [ib_uverbs]
       ib_uverbs_close+0x1f/0x80 [ib_uverbs]
       __fput+0xb7/0x230
       task_work_run+0x8a/0xb0
       do_exit+0x2da/0xb40
    ...
      RIP: 0033:0x146fa113a387
      Code: Bad RIP value.
      RSP: 002b:00007fff945d1478 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff02
      RAX: 0000000000000000 RBX: 000055a248908d70 RCX: 0000000000000000
      RDX: 0000146fa1f2b000 RSI: 0000000000000001 RDI: 000055a248906488
      RBP: 000055a248909630 R08: 0000000000010000 R09: 0000000000000000
      R10: 0000000000000000 R11: 0000000000000000 R12: 000055a248906488
      R13: 0000000000000001 R14: 0000000000000000 R15: 000055a2489095f0
    
    Do not free the MR structures, when driver returns error to the stack.
    
    Fixes: 872f3578 ("RDMA/bnxt_re: Add support for MRs with Huge pages")
    Link: https://lore.kernel.org/r/1574671174-5064-2-git-send-email-selvin.xavier@broadcom.comSigned-off-by: default avatarSelvin Xavier <selvin.xavier@broadcom.com>
    Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
    9a4467a6
ib_verbs.c 101 KB