Since commit 80c49563e250 ("scsi: scsi_debug: implement IMMED bit") there are
long delays in F_SYNC_DELAY and F_SSU_DELAY. This can cause a memory leak in
schedule_resp(), which can be invoked while unloading the scsi_debug module:
free_all_queued() had already freed all sd_dp and schedule_resp will alloc a new
one, which will never get freed. Here's the kmemleak report while running
xfstests generic/350:
unreferenced object 0xffff88007d752b00 (size 128):
comm "rmmod", pid 26940, jiffies 4295816945 (age 7.588s)
hex dump (first 32 bytes):
00 2b 75 7d 00 88 ff ff 00 00 00 00 00 00 00 00 .+u}............
00 00 00 00 00 00 00 00 8e 31 a2 34 5f 03 00 00 .........1.4_...
backtrace:
[<000000002abd83d0>] 0xffffffffa000705e
[<000000004c063fda>] scsi_dispatch_cmd+0xc7/0x1a0
[<000000000c119a00>] scsi_request_fn+0x251/0x550
[<000000009de0c736>] __blk_run_queue+0x3f/0x60
[<000000001c4453c8>] blk_execute_rq_nowait+0x98/0xd0
[<00000000d17ec79f>] blk_execute_rq+0x3a/0x50
[<00000000a7654b6e>] scsi_execute+0x113/0x250
[<00000000fd78f7cd>] sd_sync_cache+0x95/0x160
[<0000000024dacb14>] sd_shutdown+0x9b/0xd0
[<00000000e9101710>] sd_remove+0x5f/0xb0
[<00000000c43f0d63>] device_release_driver_internal+0x13c/0x1f0
[<00000000e8ad57b6>] bus_remove_device+0xe9/0x160
[<00000000713a7b8a>] device_del+0x120/0x320
[<00000000e5db670c>] __scsi_remove_device+0x115/0x150
[<00000000eccbef30>] scsi_forget_host+0x20/0x60
[<00000000cd5a0738>] scsi_remove_host+0x6d/0x120
Cc: [email protected]
Signed-off-by: Luis Henriques <[email protected]>
---
drivers/scsi/scsi_debug.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/scsi/scsi_debug.c b/drivers/scsi/scsi_debug.c
index 24d7496cd9e2..364e71861bfd 100644
--- a/drivers/scsi/scsi_debug.c
+++ b/drivers/scsi/scsi_debug.c
@@ -5507,9 +5507,9 @@ static void __exit scsi_debug_exit(void)
int k = sdebug_add_host;
stop_all_queued();
- free_all_queued();
for (; k; k--)
sdebug_remove_adapter();
+ free_all_queued();
driver_unregister(&sdebug_driverfs_driver);
bus_unregister(&pseudo_lld_bus);
root_device_unregister(pseudo_primary);
On 2018-06-18 12:08 PM, Luis Henriques wrote:
> Since commit 80c49563e250 ("scsi: scsi_debug: implement IMMED bit") there are
> long delays in F_SYNC_DELAY and F_SSU_DELAY. This can cause a memory leak in
> schedule_resp(), which can be invoked while unloading the scsi_debug module:
> free_all_queued() had already freed all sd_dp and schedule_resp will alloc a new
> one, which will never get freed. Here's the kmemleak report while running
> xfstests generic/350:
>
> unreferenced object 0xffff88007d752b00 (size 128):
> comm "rmmod", pid 26940, jiffies 4295816945 (age 7.588s)
> hex dump (first 32 bytes):
> 00 2b 75 7d 00 88 ff ff 00 00 00 00 00 00 00 00 .+u}............
> 00 00 00 00 00 00 00 00 8e 31 a2 34 5f 03 00 00 .........1.4_...
> backtrace:
> [<000000002abd83d0>] 0xffffffffa000705e
> [<000000004c063fda>] scsi_dispatch_cmd+0xc7/0x1a0
> [<000000000c119a00>] scsi_request_fn+0x251/0x550
> [<000000009de0c736>] __blk_run_queue+0x3f/0x60
> [<000000001c4453c8>] blk_execute_rq_nowait+0x98/0xd0
> [<00000000d17ec79f>] blk_execute_rq+0x3a/0x50
> [<00000000a7654b6e>] scsi_execute+0x113/0x250
> [<00000000fd78f7cd>] sd_sync_cache+0x95/0x160
> [<0000000024dacb14>] sd_shutdown+0x9b/0xd0
> [<00000000e9101710>] sd_remove+0x5f/0xb0
> [<00000000c43f0d63>] device_release_driver_internal+0x13c/0x1f0
> [<00000000e8ad57b6>] bus_remove_device+0xe9/0x160
> [<00000000713a7b8a>] device_del+0x120/0x320
> [<00000000e5db670c>] __scsi_remove_device+0x115/0x150
> [<00000000eccbef30>] scsi_forget_host+0x20/0x60
> [<00000000cd5a0738>] scsi_remove_host+0x6d/0x120
>
> Cc: [email protected]
> Signed-off-by: Luis Henriques <[email protected]>
Acked-by: Douglas Gilbert <[email protected]>
Thanks.
> ---
> drivers/scsi/scsi_debug.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/scsi/scsi_debug.c b/drivers/scsi/scsi_debug.c
> index 24d7496cd9e2..364e71861bfd 100644
> --- a/drivers/scsi/scsi_debug.c
> +++ b/drivers/scsi/scsi_debug.c
> @@ -5507,9 +5507,9 @@ static void __exit scsi_debug_exit(void)
> int k = sdebug_add_host;
>
> stop_all_queued();
> - free_all_queued();
> for (; k; k--)
> sdebug_remove_adapter();
> + free_all_queued();
> driver_unregister(&sdebug_driverfs_driver);
> bus_unregister(&pseudo_lld_bus);
> root_device_unregister(pseudo_primary);
>
Luis,
> Since commit 80c49563e250 ("scsi: scsi_debug: implement IMMED bit")
> there are long delays in F_SYNC_DELAY and F_SSU_DELAY. This can cause
> a memory leak in schedule_resp(), which can be invoked while unloading
> the scsi_debug module: free_all_queued() had already freed all sd_dp
> and schedule_resp will alloc a new one, which will never get freed.
> Here's the kmemleak report while running xfstests generic/350:
Applied to 4.18/scsi-fixes. Thank you!
--
Martin K. Petersen Oracle Linux Engineering