net/mlx5: Fix possible use-after-free in async command interface
authorTariq Toukan <tariqt@nvidia.com>
Wed, 26 Oct 2022 13:51:45 +0000 (14:51 +0100)
committerJakub Kicinski <kuba@kernel.org>
Thu, 27 Oct 2022 18:06:56 +0000 (11:06 -0700)
commitbacd22df95147ed673bec4692ab2d4d585935241
tree3855d3e936fd1c78ce48592ee0d4402680b7ee4c
parent0f3caaa2c6fbf9f892bd235c9dce9eb551f8d815
net/mlx5: Fix possible use-after-free in async command interface

mlx5_cmd_cleanup_async_ctx should return only after all its callback
handlers were completed. Before this patch, the below race between
mlx5_cmd_cleanup_async_ctx and mlx5_cmd_exec_cb_handler was possible and
lead to a use-after-free:

1. mlx5_cmd_cleanup_async_ctx is called while num_inflight is 2 (i.e.
   elevated by 1, a single inflight callback).
2. mlx5_cmd_cleanup_async_ctx decreases num_inflight to 1.
3. mlx5_cmd_exec_cb_handler is called, decreases num_inflight to 0 and
   is about to call wake_up().
4. mlx5_cmd_cleanup_async_ctx calls wait_event, which returns
   immediately as the condition (num_inflight == 0) holds.
5. mlx5_cmd_cleanup_async_ctx returns.
6. The caller of mlx5_cmd_cleanup_async_ctx frees the mlx5_async_ctx
   object.
7. mlx5_cmd_exec_cb_handler goes on and calls wake_up() on the freed
   object.

Fix it by syncing using a completion object. Mark it completed when
num_inflight reaches 0.

Trace:

BUG: KASAN: use-after-free in do_raw_spin_lock+0x23d/0x270
Read of size 4 at addr ffff888139cd12f4 by task swapper/5/0

CPU: 5 PID: 0 Comm: swapper/5 Not tainted 6.0.0-rc3_for_upstream_debug_2022_08_30_13_10 #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
Call Trace:
 <IRQ>
 dump_stack_lvl+0x57/0x7d
 print_report.cold+0x2d5/0x684
 ? do_raw_spin_lock+0x23d/0x270
 kasan_report+0xb1/0x1a0
 ? do_raw_spin_lock+0x23d/0x270
 do_raw_spin_lock+0x23d/0x270
 ? rwlock_bug.part.0+0x90/0x90
 ? __delete_object+0xb8/0x100
 ? lock_downgrade+0x6e0/0x6e0
 _raw_spin_lock_irqsave+0x43/0x60
 ? __wake_up_common_lock+0xb9/0x140
 __wake_up_common_lock+0xb9/0x140
 ? __wake_up_common+0x650/0x650
 ? destroy_tis_callback+0x53/0x70 [mlx5_core]
 ? kasan_set_track+0x21/0x30
 ? destroy_tis_callback+0x53/0x70 [mlx5_core]
 ? kfree+0x1ba/0x520
 ? do_raw_spin_unlock+0x54/0x220
 mlx5_cmd_exec_cb_handler+0x136/0x1a0 [mlx5_core]
 ? mlx5_cmd_cleanup_async_ctx+0x220/0x220 [mlx5_core]
 ? mlx5_cmd_cleanup_async_ctx+0x220/0x220 [mlx5_core]
 mlx5_cmd_comp_handler+0x65a/0x12b0 [mlx5_core]
 ? dump_command+0xcc0/0xcc0 [mlx5_core]
 ? lockdep_hardirqs_on_prepare+0x400/0x400
 ? cmd_comp_notifier+0x7e/0xb0 [mlx5_core]
 cmd_comp_notifier+0x7e/0xb0 [mlx5_core]
 atomic_notifier_call_chain+0xd7/0x1d0
 mlx5_eq_async_int+0x3ce/0xa20 [mlx5_core]
 atomic_notifier_call_chain+0xd7/0x1d0
 ? irq_release+0x140/0x140 [mlx5_core]
 irq_int_handler+0x19/0x30 [mlx5_core]
 __handle_irq_event_percpu+0x1f2/0x620
 handle_irq_event+0xb2/0x1d0
 handle_edge_irq+0x21e/0xb00
 __common_interrupt+0x79/0x1a0
 common_interrupt+0x78/0xa0
 </IRQ>
 <TASK>
 asm_common_interrupt+0x22/0x40
RIP: 0010:default_idle+0x42/0x60
Code: c1 83 e0 07 48 c1 e9 03 83 c0 03 0f b6 14 11 38 d0 7c 04 84 d2 75 14 8b 05 eb 47 22 02 85 c0 7e 07 0f 00 2d e0 9f 48 00 fb f4 <c3> 48 c7 c7 80 08 7f 85 e8 d1 d3 3e fe eb de 66 66 2e 0f 1f 84 00
RSP: 0018:ffff888100dbfdf0 EFLAGS: 00000242
RAX: 0000000000000001 RBX: ffffffff84ecbd48 RCX: 1ffffffff0afe110
RDX: 0000000000000004 RSI: 0000000000000000 RDI: ffffffff835cc9bc
RBP: 0000000000000005 R08: 0000000000000001 R09: ffff88881dec4ac3
R10: ffffed1103bd8958 R11: 0000017d0ca571c9 R12: 0000000000000005
R13: ffffffff84f024e0 R14: 0000000000000000 R15: dffffc0000000000
 ? default_idle_call+0xcc/0x450
 default_idle_call+0xec/0x450
 do_idle+0x394/0x450
 ? arch_cpu_idle_exit+0x40/0x40
 ? do_idle+0x17/0x450
 cpu_startup_entry+0x19/0x20
 start_secondary+0x221/0x2b0
 ? set_cpu_sibling_map+0x2070/0x2070
 secondary_startup_64_no_verify+0xcd/0xdb
 </TASK>

Allocated by task 49502:
 kasan_save_stack+0x1e/0x40
 __kasan_kmalloc+0x81/0xa0
 kvmalloc_node+0x48/0xe0
 mlx5e_bulk_async_init+0x35/0x110 [mlx5_core]
 mlx5e_tls_priv_tx_list_cleanup+0x84/0x3e0 [mlx5_core]
 mlx5e_ktls_cleanup_tx+0x38f/0x760 [mlx5_core]
 mlx5e_cleanup_nic_tx+0xa7/0x100 [mlx5_core]
 mlx5e_detach_netdev+0x1ca/0x2b0 [mlx5_core]
 mlx5e_suspend+0xdb/0x140 [mlx5_core]
 mlx5e_remove+0x89/0x190 [mlx5_core]
 auxiliary_bus_remove+0x52/0x70
 device_release_driver_internal+0x40f/0x650
 driver_detach+0xc1/0x180
 bus_remove_driver+0x125/0x2f0
 auxiliary_driver_unregister+0x16/0x50
 mlx5e_cleanup+0x26/0x30 [mlx5_core]
 cleanup+0xc/0x4e [mlx5_core]
 __x64_sys_delete_module+0x2b5/0x450
 do_syscall_64+0x3d/0x90
 entry_SYSCALL_64_after_hwframe+0x46/0xb0

Freed by task 49502:
 kasan_save_stack+0x1e/0x40
 kasan_set_track+0x21/0x30
 kasan_set_free_info+0x20/0x30
 ____kasan_slab_free+0x11d/0x1b0
 kfree+0x1ba/0x520
 mlx5e_tls_priv_tx_list_cleanup+0x2e7/0x3e0 [mlx5_core]
 mlx5e_ktls_cleanup_tx+0x38f/0x760 [mlx5_core]
 mlx5e_cleanup_nic_tx+0xa7/0x100 [mlx5_core]
 mlx5e_detach_netdev+0x1ca/0x2b0 [mlx5_core]
 mlx5e_suspend+0xdb/0x140 [mlx5_core]
 mlx5e_remove+0x89/0x190 [mlx5_core]
 auxiliary_bus_remove+0x52/0x70
 device_release_driver_internal+0x40f/0x650
 driver_detach+0xc1/0x180
 bus_remove_driver+0x125/0x2f0
 auxiliary_driver_unregister+0x16/0x50
 mlx5e_cleanup+0x26/0x30 [mlx5_core]
 cleanup+0xc/0x4e [mlx5_core]
 __x64_sys_delete_module+0x2b5/0x450
 do_syscall_64+0x3d/0x90
 entry_SYSCALL_64_after_hwframe+0x46/0xb0

Fixes: e355477ed9e4 ("net/mlx5: Make mlx5_cmd_exec_cb() a safe API")
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-8-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
drivers/net/ethernet/mellanox/mlx5/core/cmd.c
include/linux/mlx5/driver.h