releases/4.19.31/ib-hfi1-close-race-condition-on-user-context-disable-and-close.patch

   1 From bc5add09764c123f58942a37c8335247e683d234 Mon Sep 17 00:00:00 2001
   2 From: "Michael J. Ruhl" <michael.j.ruhl@intel.com>
   3 Date: Tue, 26 Feb 2019 08:45:35 -0800
   4 Subject: IB/hfi1: Close race condition on user context disable and close
   5
   6 From: Michael J. Ruhl <michael.j.ruhl@intel.com>
   7
   8 commit bc5add09764c123f58942a37c8335247e683d234 upstream.
   9
  10 When disabling and removing a receive context, it is possible for an
  11 asynchronous event (i.e IRQ) to occur.  Because of this, there is a race
  12 between cleaning up the context, and the context being used by the
  13 asynchronous event.
  14
  15 cpu 0  (context cleanup)
  16     rc->ref_count-- (ref_count == 0)
  17     hfi1_rcd_free()
  18 cpu 1  (IRQ (with rcd index))
  19         rcd_get_by_index()
  20         lock
  21         ref_count+++     <-- reference count race (WARNING)
  22         return rcd
  23         unlock
  24 cpu 0
  25     hfi1_free_ctxtdata() <-- incorrect free location
  26     lock
  27     remove rcd from array
  28     unlock
  29     free rcd
  30
  31 This race will cause the following WARNING trace:
  32
  33 WARNING: CPU: 0 PID: 175027 at include/linux/kref.h:52 hfi1_rcd_get_by_index+0x84/0xa0 [hfi1]
  34 CPU: 0 PID: 175027 Comm: IMB-MPI1 Kdump: loaded Tainted: G OE ------------ 3.10.0-957.el7.x86_64 #1
  35 Hardware name: Intel Corporation S2600KP/S2600KP, BIOS SE5C610.86B.11.01.0076.C4.111920150602 11/19/2015
  36 Call Trace:
  37   dump_stack+0x19/0x1b
  38   __warn+0xd8/0x100
  39   warn_slowpath_null+0x1d/0x20
  40   hfi1_rcd_get_by_index+0x84/0xa0 [hfi1]
  41   is_rcv_urgent_int+0x24/0x90 [hfi1]
  42   general_interrupt+0x1b6/0x210 [hfi1]
  43   __handle_irq_event_percpu+0x44/0x1c0
  44   handle_irq_event_percpu+0x32/0x80
  45   handle_irq_event+0x3c/0x60
  46   handle_edge_irq+0x7f/0x150
  47   handle_irq+0xe4/0x1a0
  48   do_IRQ+0x4d/0xf0
  49   common_interrupt+0x162/0x162
  50
  51 The race can also lead to a use after free which could be similar to:
  52
  53 general protection fault: 0000 1 SMP
  54 CPU: 71 PID: 177147 Comm: IMB-MPI1 Kdump: loaded Tainted: G W OE ------------ 3.10.0-957.el7.x86_64 #1
  55 Hardware name: Intel Corporation S2600KP/S2600KP, BIOS SE5C610.86B.11.01.0076.C4.111920150602 11/19/2015
  56 task: ffff9962a8098000 ti: ffff99717a508000 task.ti: ffff99717a508000 __kmalloc+0x94/0x230
  57 Call Trace:
  58   ? hfi1_user_sdma_process_request+0x9c8/0x1250 [hfi1]
  59   hfi1_user_sdma_process_request+0x9c8/0x1250 [hfi1]
  60   hfi1_aio_write+0xba/0x110 [hfi1]
  61   do_sync_readv_writev+0x7b/0xd0
  62   do_readv_writev+0xce/0x260
  63   ? handle_mm_fault+0x39d/0x9b0
  64   ? pick_next_task_fair+0x5f/0x1b0
  65   ? sched_clock_cpu+0x85/0xc0
  66   ? __schedule+0x13a/0x890
  67   vfs_writev+0x35/0x60
  68   SyS_writev+0x7f/0x110
  69   system_call_fastpath+0x22/0x27
  70
  71 Use the appropriate kref API to verify access.
  72
  73 Reorder context cleanup to ensure context removal before cleanup occurs
  74 correctly.
  75
  76 Cc: stable@vger.kernel.org # v4.14.0+
  77 Fixes: f683c80ca68e ("IB/hfi1: Resolve kernel panics by reference counting receive contexts")
  78 Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
  79 Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
  80 Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
  81 Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
  82 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  83
  84 ---
  85  drivers/infiniband/hw/hfi1/hfi.h  |    2 +-
  86  drivers/infiniband/hw/hfi1/init.c |   14 +++++++++-----
  87  2 files changed, 10 insertions(+), 6 deletions(-)
  88
  89 --- a/drivers/infiniband/hw/hfi1/hfi.h
  90 +++ b/drivers/infiniband/hw/hfi1/hfi.h
  91 @@ -1425,7 +1425,7 @@ void hfi1_init_pportdata(struct pci_dev
  92                          struct hfi1_devdata *dd, u8 hw_pidx, u8 port);
  93  void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
  94  int hfi1_rcd_put(struct hfi1_ctxtdata *rcd);
  95 -void hfi1_rcd_get(struct hfi1_ctxtdata *rcd);
  96 +int hfi1_rcd_get(struct hfi1_ctxtdata *rcd);
  97  struct hfi1_ctxtdata *hfi1_rcd_get_by_index_safe(struct hfi1_devdata *dd,
  98                                                  u16 ctxt);
  99  struct hfi1_ctxtdata *hfi1_rcd_get_by_index(struct hfi1_devdata *dd, u16 ctxt);
 100 --- a/drivers/infiniband/hw/hfi1/init.c
 101 +++ b/drivers/infiniband/hw/hfi1/init.c
 102 @@ -213,12 +213,12 @@ static void hfi1_rcd_free(struct kref *k
 103         struct hfi1_ctxtdata *rcd =
 104                 container_of(kref, struct hfi1_ctxtdata, kref);
 105
 106 -       hfi1_free_ctxtdata(rcd->dd, rcd);
 107 -
 108         spin_lock_irqsave(&rcd->dd->uctxt_lock, flags);
 109         rcd->dd->rcd[rcd->ctxt] = NULL;
 110         spin_unlock_irqrestore(&rcd->dd->uctxt_lock, flags);
 111
 112 +       hfi1_free_ctxtdata(rcd->dd, rcd);
 113 +
 114         kfree(rcd);
 115  }
 116
 117 @@ -241,10 +241,13 @@ int hfi1_rcd_put(struct hfi1_ctxtdata *r
 118   * @rcd: pointer to an initialized rcd data structure
 119   *
 120   * Use this to get a reference after the init.
 121 + *
 122 + * Return : reflect kref_get_unless_zero(), which returns non-zero on
 123 + * increment, otherwise 0.
 124   */
 125 -void hfi1_rcd_get(struct hfi1_ctxtdata *rcd)
 126 +int hfi1_rcd_get(struct hfi1_ctxtdata *rcd)
 127  {
 128 -       kref_get(&rcd->kref);
 129 +       return kref_get_unless_zero(&rcd->kref);
 130  }
 131
 132  /**
 133 @@ -324,7 +327,8 @@ struct hfi1_ctxtdata *hfi1_rcd_get_by_in
 134         spin_lock_irqsave(&dd->uctxt_lock, flags);
 135         if (dd->rcd[ctxt]) {
 136                 rcd = dd->rcd[ctxt];
 137 -               hfi1_rcd_get(rcd);
 138 +               if (!hfi1_rcd_get(rcd))
 139 +                       rcd = NULL;
 140         }
 141         spin_unlock_irqrestore(&dd->uctxt_lock, flags);
 142