From: Mike Christie michael.christie@oracle.com
[ Upstream commit 4edd8cd4e86dd3047e5294bbefcc0a08f66a430f ]
This fixes a regression added with:
commit f0f82e2476f6 ("scsi: core: Fix capacity set to zero after offlinining device")
The problem is that after iSCSI recovery, iscsid will call into the kernel to set the dev's state to running, and with that patch we now call scsi_rescan_device() with the state_mutex held. If the SCSI error handler thread is just starting to test the device in scsi_send_eh_cmnd() then it's going to try to grab the state_mutex.
We are then stuck, because when scsi_rescan_device() tries to send its I/O scsi_queue_rq() calls -> scsi_host_queue_ready() -> scsi_host_in_recovery() which will return true (the host state is still in recovery) and I/O will just be requeued. scsi_send_eh_cmnd() will then never be able to grab the state_mutex to finish error handling.
To prevent the deadlock move the rescan-related code to after we drop the state_mutex.
This also adds a check for if we are already in the running state. This prevents extra scans and helps the iscsid case where if the transport class has already onlined the device during its recovery process then we don't need userspace to do it again plus possibly block that daemon.
Link: https://lore.kernel.org/r/20211105221048.6541-3-michael.christie@oracle.com Fixes: f0f82e2476f6 ("scsi: core: Fix capacity set to zero after offlinining device") Cc: Bart Van Assche bvanassche@acm.org Cc: lijinlin lijinlin3@huawei.com Cc: Wu Bo wubo40@huawei.com Reviewed-by: Lee Duncan lduncan@suse.com Reviewed-by: Wu Bo wubo40@huawei.com Signed-off-by: Mike Christie michael.christie@oracle.com Signed-off-by: Martin K. Petersen martin.petersen@oracle.com Signed-off-by: Sasha Levin sashal@kernel.org --- drivers/scsi/scsi_sysfs.c | 30 +++++++++++++++++++----------- 1 file changed, 19 insertions(+), 11 deletions(-)
diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c index 12064ce76777d..16432d42a50aa 100644 --- a/drivers/scsi/scsi_sysfs.c +++ b/drivers/scsi/scsi_sysfs.c @@ -776,6 +776,7 @@ store_state_field(struct device *dev, struct device_attribute *attr, int i, ret; struct scsi_device *sdev = to_scsi_device(dev); enum scsi_device_state state = 0; + bool rescan_dev = false;
for (i = 0; i < ARRAY_SIZE(sdev_states); i++) { const int len = strlen(sdev_states[i].name); @@ -794,20 +795,27 @@ store_state_field(struct device *dev, struct device_attribute *attr, }
mutex_lock(&sdev->state_mutex); - ret = scsi_device_set_state(sdev, state); - /* - * If the device state changes to SDEV_RUNNING, we need to - * run the queue to avoid I/O hang, and rescan the device - * to revalidate it. Running the queue first is necessary - * because another thread may be waiting inside - * blk_mq_freeze_queue_wait() and because that call may be - * waiting for pending I/O to finish. - */ - if (ret == 0 && state == SDEV_RUNNING) { + if (sdev->sdev_state == SDEV_RUNNING && state == SDEV_RUNNING) { + ret = count; + } else { + ret = scsi_device_set_state(sdev, state); + if (ret == 0 && state == SDEV_RUNNING) + rescan_dev = true; + } + mutex_unlock(&sdev->state_mutex); + + if (rescan_dev) { + /* + * If the device state changes to SDEV_RUNNING, we need to + * run the queue to avoid I/O hang, and rescan the device + * to revalidate it. Running the queue first is necessary + * because another thread may be waiting inside + * blk_mq_freeze_queue_wait() and because that call may be + * waiting for pending I/O to finish. + */ blk_mq_run_hw_queues(sdev->request_queue, true); scsi_rescan_device(dev); } - mutex_unlock(&sdev->state_mutex);
return ret == 0 ? count : -EINVAL; }