IB/mlx5: Fix use-after-free error while accessing ev_file pointer
authorYishai Hadas <yishaih@mellanox.com>
Thu, 8 Aug 2019 08:15:38 +0000 (11:15 +0300)
committerDoug Ledford <dledford@redhat.com>
Mon, 12 Aug 2019 14:46:30 +0000 (10:46 -0400)
Call to uverbs_close_fd() releases file pointer to 'ev_file' and
mlx5_ib_dev is going to be inaccessible. Cache pointer prior cleaning
resources to solve the KASAN warning below.

BUG: KASAN: use-after-free in devx_async_event_close+0x391/0x480 [mlx5_ib]
Read of size 8 at addr ffff888301e3cec0 by task devx_direct_tes/4631
CPU: 1 PID: 4631 Comm: devx_direct_tes Tainted: G OE 5.3.0-rc1-for-upstream-dbg-2019-07-26_01-19-56-93 #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu2 04/01/2014
Call Trace:
dump_stack+0x9a/0xeb
print_address_description+0x1e2/0x400
? devx_async_event_close+0x391/0x480 [mlx5_ib]
__kasan_report+0x15c/0x1df
? devx_async_event_close+0x391/0x480 [mlx5_ib]
kasan_report+0xe/0x20
devx_async_event_close+0x391/0x480 [mlx5_ib]
__fput+0x26a/0x7b0
task_work_run+0x10d/0x180
exit_to_usermode_loop+0x137/0x160
do_syscall_64+0x3c7/0x490
entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x7f5df907d664
Code: 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b7 0f 1f
80 00 00 00 00 8b 05 6a cd 20 00 48 63 ff 85 c0 75 13 b8
03 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 44 f3 c3 66 90
48 83 ec 18 48 89 7c 24 08 e8
RSP: 002b:00007ffd353cb958 EFLAGS: 00000246 ORIG_RAX: 0000000000000003
RAX: 0000000000000000 RBX: 000056017a88c348 RCX: 00007f5df907d664
RDX: 00007f5df969d400 RSI: 00007f5de8f1ec90 RDI: 0000000000000006
RBP: 00007f5df9681dc0 R08: 00007f5de8736410 R09: 000056017a9d2dd0
R10: 000000000000000b R11: 0000000000000246 R12: 00007f5de899d7d0
R13: 00007f5df96c4248 R14: 00007f5de8f1ecb0 R15: 000056017ae41308

Allocated by task 4631:
save_stack+0x19/0x80
kasan_kmalloc.constprop.3+0xa0/0xd0
alloc_uobj+0x71/0x230 [ib_uverbs]
alloc_begin_fd_uobject+0x2e/0xc0 [ib_uverbs]
rdma_alloc_begin_uobject+0x96/0x140 [ib_uverbs]
ib_uverbs_run_method+0xdf0/0x1940 [ib_uverbs]
ib_uverbs_cmd_verbs+0x57e/0xdb0 [ib_uverbs]
ib_uverbs_ioctl+0x177/0x260 [ib_uverbs]
do_vfs_ioctl+0x18f/0x1010
ksys_ioctl+0x70/0x80
__x64_sys_ioctl+0x6f/0xb0
do_syscall_64+0x95/0x490
entry_SYSCALL_64_after_hwframe+0x49/0xbe

Freed by task 4631:
save_stack+0x19/0x80
__kasan_slab_free+0x11d/0x160
slab_free_freelist_hook+0x67/0x1a0
kfree+0xb9/0x2a0
uverbs_close_fd+0x118/0x1c0 [ib_uverbs]
devx_async_event_close+0x28a/0x480 [mlx5_ib]
__fput+0x26a/0x7b0
task_work_run+0x10d/0x180
exit_to_usermode_loop+0x137/0x160
do_syscall_64+0x3c7/0x490
entry_SYSCALL_64_after_hwframe+0x49/0xbe

The buggy address belongs to the object at ffff888301e3cda8
which belongs to the cache kmalloc-512 of size 512
The buggy address is located 280 bytes inside of 512-byte region
[ffff888301e3cda8ffff888301e3cfa8)
The buggy address belongs to the page:
page:ffffea000c078e00 refcount:1 mapcount:0
mapping:ffff888352811300 index:0x0 compound_mapcount: 0
flags: 0x2fffff80010200(slab|head)
raw: 002fffff80010200 ffffea000d152608 ffffea000c077808 ffff888352811300
raw: 0000000000000000 0000000000250025 00000001ffffffff 0000000000000000
page dumped because: kasan: bad access detected
Memory state around the buggy address:
ffff888301e3cd80: fc fc fc fc fc fb fb fb fb fb fb fb fb fb fb fb
ffff888301e3ce00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
ffff888301e3ce80: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
ffff888301e3cf00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
ffff888301e3cf80: fb fb fb fb fb fc fc fc fc fc fc fc fc fc fc fc
Disabling lock debugging due to kernel taint

Cc: <stable@vger.kernel.org> # 5.2
Fixes: 759738537142 ("IB/mlx5: Enable subscription for device events over DEVX")
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Link: https://lore.kernel.org/r/20190808081538.28772-1-leon@kernel.org
Signed-off-by: Doug Ledford <dledford@redhat.com>
drivers/infiniband/hw/mlx5/devx.c

index 2d1b3d9609d929c49f29e6326f94bfb3c5b00204..af5bbb35c0589364794ca131a1e741392cfe5931 100644 (file)
@@ -2644,12 +2644,13 @@ static int devx_async_event_close(struct inode *inode, struct file *filp)
        struct devx_async_event_file *ev_file = filp->private_data;
        struct devx_event_subscription *event_sub, *event_sub_tmp;
        struct devx_async_event_data *entry, *tmp;
+       struct mlx5_ib_dev *dev = ev_file->dev;
 
-       mutex_lock(&ev_file->dev->devx_event_table.event_xa_lock);
+       mutex_lock(&dev->devx_event_table.event_xa_lock);
        /* delete the subscriptions which are related to this FD */
        list_for_each_entry_safe(event_sub, event_sub_tmp,
                                 &ev_file->subscribed_events_list, file_list) {
-               devx_cleanup_subscription(ev_file->dev, event_sub);
+               devx_cleanup_subscription(dev, event_sub);
                if (event_sub->eventfd)
                        eventfd_ctx_put(event_sub->eventfd);
 
@@ -2658,7 +2659,7 @@ static int devx_async_event_close(struct inode *inode, struct file *filp)
                kfree_rcu(event_sub, rcu);
        }
 
-       mutex_unlock(&ev_file->dev->devx_event_table.event_xa_lock);
+       mutex_unlock(&dev->devx_event_table.event_xa_lock);
 
        /* free the pending events allocation */
        if (!ev_file->omit_data) {
@@ -2670,7 +2671,7 @@ static int devx_async_event_close(struct inode *inode, struct file *filp)
        }
 
        uverbs_close_fd(filp);
-       put_device(&ev_file->dev->ib_dev.dev);
+       put_device(&dev->ib_dev.dev);
        return 0;
 }