Re: [PATCH v3 0/4] Avoid that __scsi_remove_device() hangs
On Tue, 2017-04-18 at 14:58 +0300, Israel Rukshin wrote: > I tested those patches and I got a NULL dereference at sd_sync_cache_done(). > The test is unloading ib_srp while one port is down. > The previous version worked fine. > > From the log: > [ 190.272412] BUG: unable to handle kernel NULL pointer dereference at > 02f0 > [ 190.281102] IP: sd_sync_cache_done+0x1b/0x80 [sd_mod] Hello Israel, Thanks for testing. I assume that this data refers to the sd_printk() statement? That statement was executed properly in my tests. Anyway, I will leave out that statement, retest and repost this patch series. There is a del_gendisk() call in sd just before the sd_shutdown() call so that means it's not safe to access the disk pointer from anywhere in sd_shutdown(). Bart.
Re: [PATCH v3 0/4] Avoid that __scsi_remove_device() hangs
On 4/17/2017 8:34 PM, Bart Van Assche wrote: __scsi_remove_device() hangs if it is waiting for the SYNCHRONIZE CACHE command submitted by the sd driver to finish if the block layer queue is stopped and does not get restarted. This patch series avoids that that hang occurs. Changes compared to v2: - Moved the "stop_disk" assignment after the sdkp check in the sd driver. - Added a completion function for asynchronous SYNCHRONIZE CACHE commands. - Added "disk" and "done" arguments to scsi_execute_async(). Changes compared to v1: - Reworked the approach of this patch series. Bart Van Assche (4): Introduce scsi_start_queue() Introduce scsi_execute_async() sd: Make synchronize cache upon shutdown asynchronous Avoid that __scsi_remove_device() hangs drivers/scsi/scsi_lib.c| 114 ++--- drivers/scsi/scsi_priv.h | 1 + drivers/scsi/scsi_sysfs.c | 9 drivers/scsi/sd.c | 45 -- include/scsi/scsi_device.h | 5 ++ 5 files changed, 142 insertions(+), 32 deletions(-) Hi Bart, I tested those patches and I got a NULL dereference at sd_sync_cache_done(). The test is unloading ib_srp while one port is down. The previous version worked fine. From the log: [ 190.260240] sd 8:0:0:0: [sdc] Synchronizing SCSI cache [ 190.266412] scsi 8:0:0:0: rejecting I/O to dead device [ 190.272412] BUG: unable to handle kernel NULL pointer dereference at 02f0 [ 190.281102] IP: sd_sync_cache_done+0x1b/0x80 [sd_mod] [ 190.482738] Call Trace: [ 190.486052] blk_finish_request+0x73/0x130 [ 190.491007] __blk_end_bidi_request+0x2d/0x40 [ 190.496215] __blk_end_request_all+0x1f/0x40 [ 190.501338] blk_peek_request+0x1c5/0x2b0 [ 190.506203] scsi_request_fn+0x3f/0x6c0 [ 190.510888] ? kobject_put+0x1f/0x60 [ 190.515305] __blk_run_queue+0x33/0x40 [ 190.519877] blk_start_queue+0x29/0x40 [ 190.524438] scsi_start_queue+0x40/0x60 [ 190.529081] __scsi_remove_device+0x4d/0xe0 [ 190.534079] scsi_forget_host+0x60/0x70 [ 190.538738] scsi_remove_host+0x77/0x110 [ 190.543462] srp_remove_work+0x90/0x230 [ib_srp] Regards, Israel.
[PATCH v3 0/4] Avoid that __scsi_remove_device() hangs
__scsi_remove_device() hangs if it is waiting for the SYNCHRONIZE CACHE command submitted by the sd driver to finish if the block layer queue is stopped and does not get restarted. This patch series avoids that that hang occurs. Changes compared to v2: - Moved the "stop_disk" assignment after the sdkp check in the sd driver. - Added a completion function for asynchronous SYNCHRONIZE CACHE commands. - Added "disk" and "done" arguments to scsi_execute_async(). Changes compared to v1: - Reworked the approach of this patch series. Bart Van Assche (4): Introduce scsi_start_queue() Introduce scsi_execute_async() sd: Make synchronize cache upon shutdown asynchronous Avoid that __scsi_remove_device() hangs drivers/scsi/scsi_lib.c| 114 ++--- drivers/scsi/scsi_priv.h | 1 + drivers/scsi/scsi_sysfs.c | 9 drivers/scsi/sd.c | 45 -- include/scsi/scsi_device.h | 5 ++ 5 files changed, 142 insertions(+), 32 deletions(-) -- 2.12.2