Commit ec16b6bb authored by Mark Zhang's avatar Mark Zhang Committed by Jason Gunthorpe

RDMA/mlx5: Fix the number of hwcounters of a dynamic counter

When we read the global counter and there's any dynamic counter allocated,
the value of a hwcounter is the sum of the default counter and all dynamic
counters. So the number of hwcounters of a dynamically allocated counter
must be same as of the default counter, otherwise there will be read
violations.

This fixes the KASAN slab-out-of-bounds bug:

  BUG: KASAN: slab-out-of-bounds in rdma_counter_get_hwstat_value+0x36d/0x390 [ib_core]
  Read of size 8 at addr ffff8884192a5778 by task rdma/10138

  CPU: 7 PID: 10138 Comm: rdma Not tainted 5.5.0-for-upstream-dbg-2020-02-06_18-30-19-27 #1
  Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.org 04/01/2014
  Call Trace:
   dump_stack+0xb7/0x10b
   print_address_description.constprop.4+0x1e2/0x400
   ? rdma_counter_get_hwstat_value+0x36d/0x390 [ib_core]
   __kasan_report+0x15c/0x1e0
   ? mlx5_ib_query_q_counters+0x13f/0x270 [mlx5_ib]
   ? rdma_counter_get_hwstat_value+0x36d/0x390 [ib_core]
   kasan_report+0xe/0x20
   rdma_counter_get_hwstat_value+0x36d/0x390 [ib_core]
   ? rdma_counter_query_stats+0xd0/0xd0 [ib_core]
   ? memcpy+0x34/0x50
   ? nla_put+0xe2/0x170
   nldev_stat_get_doit+0x9c7/0x14f0 [ib_core]
   ...
   do_syscall_64+0x95/0x490
   entry_SYSCALL_64_after_hwframe+0x49/0xbe
  RIP: 0033:0x7fcc457fe65a
  Code: bb 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 8b 05 fa f1 2b 00 45 89 c9 4c 63 d1 48 63 ff 85 c0 75 15 b8 2c 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 76 f3 c3 0f 1f 40 00 41 55 41 54 4d 89 c5 55
  RSP: 002b:00007ffc0586f868 EFLAGS: 00000246 ORIG_RAX: 000000000000002c
  RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fcc457fe65a
  RDX: 0000000000000020 RSI: 00000000013db920 RDI: 0000000000000003
  RBP: 00007ffc0586fa90 R08: 00007fcc45ac10e0 R09: 000000000000000c
  R10: 0000000000000000 R11: 0000000000000246 R12: 00000000004089c0
  R13: 0000000000000000 R14: 00007ffc0586fab0 R15: 00000000013dc9a0

  Allocated by task 9700:
   save_stack+0x19/0x80
   __kasan_kmalloc.constprop.7+0xa0/0xd0
   mlx5_ib_counter_alloc_stats+0xd1/0x1d0 [mlx5_ib]
   rdma_counter_alloc+0x16d/0x3f0 [ib_core]
   rdma_counter_bind_qpn_alloc+0x216/0x4e0 [ib_core]
   nldev_stat_set_doit+0x8c2/0xb10 [ib_core]
   rdma_nl_rcv_msg+0x3d2/0x730 [ib_core]
   rdma_nl_rcv+0x2a8/0x400 [ib_core]
   netlink_unicast+0x448/0x620
   netlink_sendmsg+0x731/0xd10
   sock_sendmsg+0xb1/0xf0
   __sys_sendto+0x25d/0x2c0
   __x64_sys_sendto+0xdd/0x1b0
   do_syscall_64+0x95/0x490
   entry_SYSCALL_64_after_hwframe+0x49/0xbe

Fixes: 18d422ce ("IB/mlx5: Add counter_alloc_stats() and counter_update_stats() support")
Link: https://lore.kernel.org/r/20200305124052.196688-1-leon@kernel.orgSigned-off-by: default avatarMark Zhang <markz@mellanox.com>
Signed-off-by: default avatarLeon Romanovsky <leonro@mellanox.com>
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
parent d372abf3
......@@ -5722,9 +5722,10 @@ mlx5_ib_counter_alloc_stats(struct rdma_counter *counter)
const struct mlx5_ib_counters *cnts =
get_counters(dev, counter->port - 1);
/* Q counters are in the beginning of all counters */
return rdma_alloc_hw_stats_struct(cnts->names,
cnts->num_q_counters,
cnts->num_q_counters +
cnts->num_cong_counters +
cnts->num_ext_ppcnt_counters,
RDMA_HW_STATS_DEFAULT_LIFESPAN);
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment