net/mlx5e: Fix use-after-free of encap entry in neigh update handler
authorVlad Buslov <vladbu@nvidia.com>
Mon, 31 May 2021 13:28:39 +0000 (16:28 +0300)
committerSaeed Mahameed <saeedm@nvidia.com>
Thu, 10 Jun 2021 00:20:03 +0000 (17:20 -0700)
commitfb1a3132ee1ac968316e45d21a48703a6db0b6c3
treeb8de9e4be293c5fd20037ac4325e5cef1cd34e9b
parent2bf8d2ae3480da06e64dad3b326ebd2e40c0be86
net/mlx5e: Fix use-after-free of encap entry in neigh update handler

Function mlx5e_rep_neigh_update() wasn't updated to accommodate rtnl lock
removal from TC filter update path and properly handle concurrent encap
entry insertion/deletion which can lead to following use-after-free:

 [23827.464923] ==================================================================
 [23827.469446] BUG: KASAN: use-after-free in mlx5e_encap_take+0x72/0x140 [mlx5_core]
 [23827.470971] Read of size 4 at addr ffff8881d132228c by task kworker/u20:6/21635
 [23827.472251]
 [23827.472615] CPU: 9 PID: 21635 Comm: kworker/u20:6 Not tainted 5.13.0-rc3+ #5
 [23827.473788] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
 [23827.475639] Workqueue: mlx5e mlx5e_rep_neigh_update [mlx5_core]
 [23827.476731] Call Trace:
 [23827.477260]  dump_stack+0xbb/0x107
 [23827.477906]  print_address_description.constprop.0+0x18/0x140
 [23827.478896]  ? mlx5e_encap_take+0x72/0x140 [mlx5_core]
 [23827.479879]  ? mlx5e_encap_take+0x72/0x140 [mlx5_core]
 [23827.480905]  kasan_report.cold+0x7c/0xd8
 [23827.481701]  ? mlx5e_encap_take+0x72/0x140 [mlx5_core]
 [23827.482744]  kasan_check_range+0x145/0x1a0
 [23827.493112]  mlx5e_encap_take+0x72/0x140 [mlx5_core]
 [23827.494054]  ? mlx5e_tc_tun_encap_info_equal_generic+0x140/0x140 [mlx5_core]
 [23827.495296]  mlx5e_rep_neigh_update+0x41e/0x5e0 [mlx5_core]
 [23827.496338]  ? mlx5e_rep_neigh_entry_release+0xb80/0xb80 [mlx5_core]
 [23827.497486]  ? read_word_at_a_time+0xe/0x20
 [23827.498250]  ? strscpy+0xa0/0x2a0
 [23827.498889]  process_one_work+0x8ac/0x14e0
 [23827.499638]  ? lockdep_hardirqs_on_prepare+0x400/0x400
 [23827.500537]  ? pwq_dec_nr_in_flight+0x2c0/0x2c0
 [23827.501359]  ? rwlock_bug.part.0+0x90/0x90
 [23827.502116]  worker_thread+0x53b/0x1220
 [23827.502831]  ? process_one_work+0x14e0/0x14e0
 [23827.503627]  kthread+0x328/0x3f0
 [23827.504254]  ? _raw_spin_unlock_irq+0x24/0x40
 [23827.505065]  ? __kthread_bind_mask+0x90/0x90
 [23827.505912]  ret_from_fork+0x1f/0x30
 [23827.506621]
 [23827.506987] Allocated by task 28248:
 [23827.507694]  kasan_save_stack+0x1b/0x40
 [23827.508476]  __kasan_kmalloc+0x7c/0x90
 [23827.509197]  mlx5e_attach_encap+0xde1/0x1d40 [mlx5_core]
 [23827.510194]  mlx5e_tc_add_fdb_flow+0x397/0xc40 [mlx5_core]
 [23827.511218]  __mlx5e_add_fdb_flow+0x519/0xb30 [mlx5_core]
 [23827.512234]  mlx5e_configure_flower+0x191c/0x4870 [mlx5_core]
 [23827.513298]  tc_setup_cb_add+0x1d5/0x420
 [23827.514023]  fl_hw_replace_filter+0x382/0x6a0 [cls_flower]
 [23827.514975]  fl_change+0x2ceb/0x4a51 [cls_flower]
 [23827.515821]  tc_new_tfilter+0x89a/0x2070
 [23827.516548]  rtnetlink_rcv_msg+0x644/0x8c0
 [23827.517300]  netlink_rcv_skb+0x11d/0x340
 [23827.518021]  netlink_unicast+0x42b/0x700
 [23827.518742]  netlink_sendmsg+0x743/0xc20
 [23827.519467]  sock_sendmsg+0xb2/0xe0
 [23827.520131]  ____sys_sendmsg+0x590/0x770
 [23827.520851]  ___sys_sendmsg+0xd8/0x160
 [23827.521552]  __sys_sendmsg+0xb7/0x140
 [23827.522238]  do_syscall_64+0x3a/0x70
 [23827.522907]  entry_SYSCALL_64_after_hwframe+0x44/0xae
 [23827.523797]
 [23827.524163] Freed by task 25948:
 [23827.524780]  kasan_save_stack+0x1b/0x40
 [23827.525488]  kasan_set_track+0x1c/0x30
 [23827.526187]  kasan_set_free_info+0x20/0x30
 [23827.526968]  __kasan_slab_free+0xed/0x130
 [23827.527709]  slab_free_freelist_hook+0xcf/0x1d0
 [23827.528528]  kmem_cache_free_bulk+0x33a/0x6e0
 [23827.529317]  kfree_rcu_work+0x55f/0xb70
 [23827.530024]  process_one_work+0x8ac/0x14e0
 [23827.530770]  worker_thread+0x53b/0x1220
 [23827.531480]  kthread+0x328/0x3f0
 [23827.532114]  ret_from_fork+0x1f/0x30
 [23827.532785]
 [23827.533147] Last potentially related work creation:
 [23827.534007]  kasan_save_stack+0x1b/0x40
 [23827.534710]  kasan_record_aux_stack+0xab/0xc0
 [23827.535492]  kvfree_call_rcu+0x31/0x7b0
 [23827.536206]  mlx5e_tc_del_fdb_flow+0x577/0xef0 [mlx5_core]
 [23827.537305]  mlx5e_flow_put+0x49/0x80 [mlx5_core]
 [23827.538290]  mlx5e_delete_flower+0x6d1/0xe60 [mlx5_core]
 [23827.539300]  tc_setup_cb_destroy+0x18e/0x2f0
 [23827.540144]  fl_hw_destroy_filter+0x1d2/0x310 [cls_flower]
 [23827.541148]  __fl_delete+0x4dc/0x660 [cls_flower]
 [23827.541985]  fl_delete+0x97/0x160 [cls_flower]
 [23827.542782]  tc_del_tfilter+0x7ab/0x13d0
 [23827.543503]  rtnetlink_rcv_msg+0x644/0x8c0
 [23827.544257]  netlink_rcv_skb+0x11d/0x340
 [23827.544981]  netlink_unicast+0x42b/0x700
 [23827.545700]  netlink_sendmsg+0x743/0xc20
 [23827.546424]  sock_sendmsg+0xb2/0xe0
 [23827.547084]  ____sys_sendmsg+0x590/0x770
 [23827.547850]  ___sys_sendmsg+0xd8/0x160
 [23827.548606]  __sys_sendmsg+0xb7/0x140
 [23827.549303]  do_syscall_64+0x3a/0x70
 [23827.549969]  entry_SYSCALL_64_after_hwframe+0x44/0xae
 [23827.550853]
 [23827.551217] The buggy address belongs to the object at ffff8881d1322200
 [23827.551217]  which belongs to the cache kmalloc-256 of size 256
 [23827.553341] The buggy address is located 140 bytes inside of
 [23827.553341]  256-byte region [ffff8881d1322200ffff8881d1322300)
 [23827.555747] The buggy address belongs to the page:
 [23827.556847] page:00000000898762aa refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x1d1320
 [23827.558651] head:00000000898762aa order:2 compound_mapcount:0 compound_pincount:0
 [23827.559961] flags: 0x2ffff800010200(slab|head|node=0|zone=2|lastcpupid=0x1ffff)
 [23827.561243] raw: 002ffff800010200 dead000000000100 dead000000000122 ffff888100042b40
 [23827.562653] raw: 0000000000000000 0000000000200020 00000001ffffffff 0000000000000000
 [23827.564112] page dumped because: kasan: bad access detected
 [23827.565439]
 [23827.565932] Memory state around the buggy address:
 [23827.566917]  ffff8881d1322180: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 [23827.568485]  ffff8881d1322200: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 [23827.569818] >ffff8881d1322280: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 [23827.571143]                       ^
 [23827.571879]  ffff8881d1322300: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 [23827.573283]  ffff8881d1322380: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 [23827.574654] ==================================================================

Most of the necessary logic is already correctly implemented by
mlx5e_get_next_valid_encap() helper that is used in neigh stats update
handler. Make the handler generic by renaming it to
mlx5e_get_next_matching_encap() and use callback to test whether flow is
matching instead of hardcoded check for 'valid' flag value. Implement
mlx5e_get_next_valid_encap() by calling mlx5e_get_next_matching_encap()
with callback that tests encap MLX5_ENCAP_ENTRY_VALID flag. Implement new
mlx5e_get_next_init_encap() helper by calling
mlx5e_get_next_matching_encap() with callback that tests encap completion
result to be non-error and use it in mlx5e_rep_neigh_update() to safely
iterate over nhe->encap_list.

Remove encap completion logic from mlx5e_rep_update_flows() since the encap
entries passed to this function are already guaranteed to be properly
initialized by similar code in mlx5e_get_next_init_encap().

Fixes: 2a1f1768fa17 ("net/mlx5e: Refactor neigh update for concurrent execution")
Signed-off-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
drivers/net/ethernet/mellanox/mlx5/core/en/rep/neigh.c
drivers/net/ethernet/mellanox/mlx5/core/en/rep/tc.c
drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_encap.c
drivers/net/ethernet/mellanox/mlx5/core/en_tc.h