From 9077fb2ab78cd76f710893919df6b0b7da79c2c9 Mon Sep 17 00:00:00 2001 From: Hannes Reinecke Date: Wed, 14 Jun 2023 12:36:10 +0200 Subject: [PATCH 1/7] scsi: bsg: Increase number of devices Larger setups may need to allocate more than 32k bsg devices, so increase the number of devices to the full range of minor device numbers. Signed-off-by: Hannes Reinecke Signed-off-by: Martin Wilck Link: https://lore.kernel.org/r/20230614103616.31857-2-mwilck@suse.com Reviewed-by: Christoph Hellwig Reviewed-by: Ming Lei Reviewed-by: Bart Van Assche Signed-off-by: Martin K. Petersen --- block/bsg.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/block/bsg.c b/block/bsg.c index 7eca43f33d7f..c53f24243bf2 100644 --- a/block/bsg.c +++ b/block/bsg.c @@ -36,7 +36,7 @@ static inline struct bsg_device *to_bsg_device(struct inode *inode) } #define BSG_DEFAULT_CMDS 64 -#define BSG_MAX_DEVS 32768 +#define BSG_MAX_DEVS (1 << MINORBITS) static DEFINE_IDA(bsg_minor_ida); static struct class *bsg_class; From 37c918e03ef7d8c8b1d4f3216d6935479d3acbee Mon Sep 17 00:00:00 2001 From: Hannes Reinecke Date: Wed, 14 Jun 2023 12:36:11 +0200 Subject: [PATCH 2/7] scsi: sg: Increase number of devices Larger setups may need to allocate more than 32k sg devices, so increase the number of devices to the full range of minor device numbers. Signed-off-by: Hannes Reinecke Signed-off-by: Martin Wilck Link: https://lore.kernel.org/r/20230614103616.31857-3-mwilck@suse.com Acked-by: Douglas Gilbert Reviewed-by: Christoph Hellwig Reviewed-by: Ming Lei Reviewed-by: Bart Van Assche Signed-off-by: Martin K. Petersen --- drivers/scsi/sg.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index 037f8c98a6d3..6c04cf941dac 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -71,7 +71,7 @@ static int sg_proc_init(void); #define SG_ALLOW_DIO_DEF 0 -#define SG_MAX_DEVS 32768 +#define SG_MAX_DEVS (1 << MINORBITS) /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater From c5e46f7ad43b0519980020378a2b00b339359968 Mon Sep 17 00:00:00 2001 From: Martin Wilck Date: Wed, 14 Jun 2023 12:36:12 +0200 Subject: [PATCH 3/7] scsi: core: Merge scsi_internal_device_block() and device_block() scsi_internal_device_block() is only called from device_block(). Merge the two functions, and call the result scsi_device_block(), as the name device_block() is confusingly generic. Signed-off-by: Martin Wilck Link: https://lore.kernel.org/r/20230614103616.31857-4-mwilck@suse.com Reviewed-by: Christoph Hellwig Reviewed-by: Hannes Reinecke Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 27 ++++++++------------------- 1 file changed, 8 insertions(+), 19 deletions(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index b7c569a42aa4..357eb0ab6f21 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -2775,13 +2775,12 @@ int scsi_internal_device_block_nowait(struct scsi_device *sdev) EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait); /** - * scsi_internal_device_block - try to transition to the SDEV_BLOCK state + * scsi_device_block - try to transition to the SDEV_BLOCK state * @sdev: device to block + * @data: dummy argument, ignored * * Pause SCSI command processing on the specified device and wait until all - * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep. - * - * Returns zero if successful or a negative error code upon failure. + * ongoing scsi_queue_rq() calls have finished. May sleep. * * Note: * This routine transitions the device to the SDEV_BLOCK state (which must be @@ -2789,7 +2788,7 @@ EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait); * is paused until the device leaves the SDEV_BLOCK state. See also * scsi_internal_device_unblock(). */ -static int scsi_internal_device_block(struct scsi_device *sdev) +static void scsi_device_block(struct scsi_device *sdev, void *data) { int err; @@ -2799,7 +2798,8 @@ static int scsi_internal_device_block(struct scsi_device *sdev) scsi_stop_queue(sdev, false); mutex_unlock(&sdev->state_mutex); - return err; + WARN_ONCE(err, "__scsi_internal_device_block_nowait(%s) failed: err = %d\n", + dev_name(&sdev->sdev_gendev), err); } /** @@ -2882,23 +2882,12 @@ static int scsi_internal_device_unblock(struct scsi_device *sdev, return ret; } -static void -device_block(struct scsi_device *sdev, void *data) -{ - int ret; - - ret = scsi_internal_device_block(sdev); - - WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n", - dev_name(&sdev->sdev_gendev), ret); -} - static int target_block(struct device *dev, void *data) { if (scsi_is_target_device(dev)) starget_for_each_device(to_scsi_target(dev), NULL, - device_block); + scsi_device_block); return 0; } @@ -2907,7 +2896,7 @@ scsi_target_block(struct device *dev) { if (scsi_is_target_device(dev)) starget_for_each_device(to_scsi_target(dev), NULL, - device_block); + scsi_device_block); else device_for_each_child(dev, NULL, target_block); } From d7035b73a73a79a1dc991fad0ee5f784559e81ed Mon Sep 17 00:00:00 2001 From: Martin Wilck Date: Wed, 14 Jun 2023 12:36:13 +0200 Subject: [PATCH 4/7] scsi: core: Don't wait for quiesce in scsi_stop_queue() scsi_stop_queue() has just two callers, one with and one without "nowait". As blk_mq_quiesce_queue() comes down to blk_mq_quiesce_queue_nowait() followed by blk_mq_wait_quiesce_done(), we might as well open-code this in scsi_device_block(). Also, add a comment explaining why blk_mq_quiesce_queue_nowait() must be called with the state_mutex held, see https://lore.kernel.org/linux-scsi/3b8b13bf-a458-827a-b916-07d7eee8ae00@acm.org/. Signed-off-by: Martin Wilck Link: https://lore.kernel.org/r/20230614103616.31857-5-mwilck@suse.com Reviewed-by: Hannes Reinecke Reviewed-by: Christoph Hellwig Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 32 ++++++++++++++++---------------- 1 file changed, 16 insertions(+), 16 deletions(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index 357eb0ab6f21..4b46dcba7649 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -2725,24 +2725,16 @@ void scsi_start_queue(struct scsi_device *sdev) blk_mq_unquiesce_queue(sdev->request_queue); } -static void scsi_stop_queue(struct scsi_device *sdev, bool nowait) +static void scsi_stop_queue(struct scsi_device *sdev) { /* * The atomic variable of ->queue_stopped covers that * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue. * - * However, we still need to wait until quiesce is done - * in case that queue has been stopped. + * The caller needs to wait until quiesce is done. */ - if (!cmpxchg(&sdev->queue_stopped, 0, 1)) { - if (nowait) - blk_mq_quiesce_queue_nowait(sdev->request_queue); - else - blk_mq_quiesce_queue(sdev->request_queue); - } else { - if (!nowait) - blk_mq_wait_quiesce_done(sdev->request_queue->tag_set); - } + if (!cmpxchg(&sdev->queue_stopped, 0, 1)) + blk_mq_quiesce_queue_nowait(sdev->request_queue); } /** @@ -2769,7 +2761,7 @@ int scsi_internal_device_block_nowait(struct scsi_device *sdev) * request queue. */ if (!ret) - scsi_stop_queue(sdev, true); + scsi_stop_queue(sdev); return ret; } EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait); @@ -2794,9 +2786,17 @@ static void scsi_device_block(struct scsi_device *sdev, void *data) mutex_lock(&sdev->state_mutex); err = __scsi_internal_device_block_nowait(sdev); - if (err == 0) - scsi_stop_queue(sdev, false); - mutex_unlock(&sdev->state_mutex); + if (err == 0) { + /* + * scsi_stop_queue() must be called with the state_mutex + * held. Otherwise a simultaneous scsi_start_queue() call + * might unquiesce the queue before we quiesce it. + */ + scsi_stop_queue(sdev); + mutex_unlock(&sdev->state_mutex); + blk_mq_wait_quiesce_done(sdev->request_queue->tag_set); + } else + mutex_unlock(&sdev->state_mutex); WARN_ONCE(err, "__scsi_internal_device_block_nowait(%s) failed: err = %d\n", dev_name(&sdev->sdev_gendev), err); From e20fff8a1f4940f46be888bd175412c2e3e64e96 Mon Sep 17 00:00:00 2001 From: Martin Wilck Date: Wed, 14 Jun 2023 12:36:14 +0200 Subject: [PATCH 5/7] scsi: core: Don't wait for quiesce in scsi_device_block() scsi_device_block() is only called from scsi_target_block(), which calls it repeatedly for every child device. For targets with many devices, waiting for every queue to quiesce may cause a substantial delay (we measured more than 100s delay for blocking a FC rport with 2048 LUNs). Just call blk_mq_wait_quiesce_done() once from scsi_target_block() after stopping all queues. Signed-off-by: Martin Wilck Link: https://lore.kernel.org/r/20230614103616.31857-6-mwilck@suse.com Reviewed-by: Hannes Reinecke Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 17 ++++++++++------- 1 file changed, 10 insertions(+), 7 deletions(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index 4b46dcba7649..b98750028044 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -2771,8 +2771,9 @@ EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait); * @sdev: device to block * @data: dummy argument, ignored * - * Pause SCSI command processing on the specified device and wait until all - * ongoing scsi_queue_rq() calls have finished. May sleep. + * Pause SCSI command processing on the specified device. Callers must wait + * until all ongoing scsi_queue_rq() calls have finished after this function + * returns. * * Note: * This routine transitions the device to the SDEV_BLOCK state (which must be @@ -2786,17 +2787,15 @@ static void scsi_device_block(struct scsi_device *sdev, void *data) mutex_lock(&sdev->state_mutex); err = __scsi_internal_device_block_nowait(sdev); - if (err == 0) { + if (err == 0) /* * scsi_stop_queue() must be called with the state_mutex * held. Otherwise a simultaneous scsi_start_queue() call * might unquiesce the queue before we quiesce it. */ scsi_stop_queue(sdev); - mutex_unlock(&sdev->state_mutex); - blk_mq_wait_quiesce_done(sdev->request_queue->tag_set); - } else - mutex_unlock(&sdev->state_mutex); + + mutex_unlock(&sdev->state_mutex); WARN_ONCE(err, "__scsi_internal_device_block_nowait(%s) failed: err = %d\n", dev_name(&sdev->sdev_gendev), err); @@ -2894,11 +2893,15 @@ target_block(struct device *dev, void *data) void scsi_target_block(struct device *dev) { + struct Scsi_Host *shost = dev_to_shost(dev); + if (scsi_is_target_device(dev)) starget_for_each_device(to_scsi_target(dev), NULL, scsi_device_block); else device_for_each_child(dev, NULL, target_block); + + blk_mq_wait_quiesce_done(&shost->tag_set); } EXPORT_SYMBOL_GPL(scsi_target_block); From 31950192d939a969415d0e1da4c62598023b0850 Mon Sep 17 00:00:00 2001 From: Martin Wilck Date: Wed, 14 Jun 2023 12:36:15 +0200 Subject: [PATCH 6/7] scsi: core: Replace scsi_target_block() with scsi_block_targets() All callers (fc_remote_port_delete(), __iscsi_block_session(), __srp_start_tl_fail_timers(), srp_reconnect_rport(), snic_tgt_del()) pass parent devices of scsi_target devices to scsi_target_block(). Rename the function to scsi_block_targets(), and simplify it by assuming that it is always passed a parent device. Also, have callers pass the Scsi_Host pointer to scsi_block_targets(), as every caller has this pointer readily available. Suggested-by: Christoph Hellwig Suggested-by: Bart Van Assche Signed-off-by: Martin Wilck Link: https://lore.kernel.org/r/20230614103616.31857-7-mwilck@suse.com Cc: Karan Tilak Kumar Cc: Sesidhar Baddela Reviewed-by: Hannes Reinecke Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 26 ++++++++++++++++---------- drivers/scsi/scsi_transport_fc.c | 2 +- drivers/scsi/scsi_transport_iscsi.c | 3 ++- drivers/scsi/scsi_transport_srp.c | 6 +++--- drivers/scsi/snic/snic_disc.c | 2 +- include/scsi/scsi_device.h | 2 +- 6 files changed, 24 insertions(+), 17 deletions(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index b98750028044..55a327b2dd8b 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -2890,20 +2890,26 @@ target_block(struct device *dev, void *data) return 0; } +/** + * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state + * @dev: a parent device of one or more scsi_target devices + * @shost: the Scsi_Host to which this device belongs + * + * Iterate over all children of @dev, which should be scsi_target devices, + * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for + * ongoing scsi_queue_rq() calls to finish. May sleep. + * + * Note: + * @dev must not itself be a scsi_target device. + */ void -scsi_target_block(struct device *dev) +scsi_block_targets(struct Scsi_Host *shost, struct device *dev) { - struct Scsi_Host *shost = dev_to_shost(dev); - - if (scsi_is_target_device(dev)) - starget_for_each_device(to_scsi_target(dev), NULL, - scsi_device_block); - else - device_for_each_child(dev, NULL, target_block); - + WARN_ON_ONCE(scsi_is_target_device(dev)); + device_for_each_child(dev, NULL, target_block); blk_mq_wait_quiesce_done(&shost->tag_set); } -EXPORT_SYMBOL_GPL(scsi_target_block); +EXPORT_SYMBOL_GPL(scsi_block_targets); static void device_unblock(struct scsi_device *sdev, void *data) diff --git a/drivers/scsi/scsi_transport_fc.c b/drivers/scsi/scsi_transport_fc.c index 64ff2629eaf9..b04075f19445 100644 --- a/drivers/scsi/scsi_transport_fc.c +++ b/drivers/scsi/scsi_transport_fc.c @@ -3451,7 +3451,7 @@ fc_remote_port_delete(struct fc_rport *rport) spin_unlock_irqrestore(shost->host_lock, flags); - scsi_target_block(&rport->dev); + scsi_block_targets(shost, &rport->dev); /* see if we need to kill io faster than waiting for device loss */ if ((rport->fast_io_fail_tmo != -1) && diff --git a/drivers/scsi/scsi_transport_iscsi.c b/drivers/scsi/scsi_transport_iscsi.c index b9b97300e3b3..e527ece12453 100644 --- a/drivers/scsi/scsi_transport_iscsi.c +++ b/drivers/scsi/scsi_transport_iscsi.c @@ -1943,13 +1943,14 @@ static void __iscsi_block_session(struct work_struct *work) struct iscsi_cls_session *session = container_of(work, struct iscsi_cls_session, block_work); + struct Scsi_Host *shost = iscsi_session_to_shost(session); unsigned long flags; ISCSI_DBG_TRANS_SESSION(session, "Blocking session\n"); spin_lock_irqsave(&session->lock, flags); session->state = ISCSI_SESSION_FAILED; spin_unlock_irqrestore(&session->lock, flags); - scsi_target_block(&session->dev); + scsi_block_targets(shost, &session->dev); ISCSI_DBG_TRANS_SESSION(session, "Completed SCSI target blocking\n"); if (session->recovery_tmo >= 0) queue_delayed_work(session->workq, diff --git a/drivers/scsi/scsi_transport_srp.c b/drivers/scsi/scsi_transport_srp.c index 87d0fb8dc503..64f6b22e8cc0 100644 --- a/drivers/scsi/scsi_transport_srp.c +++ b/drivers/scsi/scsi_transport_srp.c @@ -396,7 +396,7 @@ static void srp_reconnect_work(struct work_struct *work) } /* - * scsi_target_block() must have been called before this function is + * scsi_block_targets() must have been called before this function is * called to guarantee that no .queuecommand() calls are in progress. */ static void __rport_fail_io_fast(struct srp_rport *rport) @@ -480,7 +480,7 @@ static void __srp_start_tl_fail_timers(struct srp_rport *rport) srp_rport_set_state(rport, SRP_RPORT_BLOCKED) == 0) { pr_debug("%s new state: %d\n", dev_name(&shost->shost_gendev), rport->state); - scsi_target_block(&shost->shost_gendev); + scsi_block_targets(shost, &shost->shost_gendev); if (fast_io_fail_tmo >= 0) queue_delayed_work(system_long_wq, &rport->fast_io_fail_work, @@ -548,7 +548,7 @@ int srp_reconnect_rport(struct srp_rport *rport) * later is ok though, scsi_internal_device_unblock_nowait() * treats SDEV_TRANSPORT_OFFLINE like SDEV_BLOCK. */ - scsi_target_block(&shost->shost_gendev); + scsi_block_targets(shost, &shost->shost_gendev); res = rport->state != SRP_RPORT_LOST ? i->f->reconnect(rport) : -ENODEV; pr_debug("%s (state %d): transport.reconnect() returned %d\n", dev_name(&shost->shost_gendev), rport->state, res); diff --git a/drivers/scsi/snic/snic_disc.c b/drivers/scsi/snic/snic_disc.c index 8fbf3c1b1311..3e2e5783924d 100644 --- a/drivers/scsi/snic/snic_disc.c +++ b/drivers/scsi/snic/snic_disc.c @@ -214,7 +214,7 @@ snic_tgt_del(struct work_struct *work) scsi_flush_work(shost); /* Block IOs on child devices, stops new IOs */ - scsi_target_block(&tgt->dev); + scsi_block_targets(shost, &tgt->dev); /* Cleanup IOs */ snic_tgt_scsi_abort_io(tgt); diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h index f10a008e5bfa..8bd5b00b33cc 100644 --- a/include/scsi/scsi_device.h +++ b/include/scsi/scsi_device.h @@ -450,7 +450,7 @@ extern void scsi_scan_target(struct device *parent, unsigned int channel, unsigned int id, u64 lun, enum scsi_scan_mode rescan); extern void scsi_target_reap(struct scsi_target *); -extern void scsi_target_block(struct device *); +void scsi_block_targets(struct Scsi_Host *shost, struct device *dev); extern void scsi_target_unblock(struct device *, enum scsi_device_state); extern void scsi_remove_target(struct device *); extern const char *scsi_device_state_name(enum scsi_device_state); From 6d7160c7da6fa3010252910a1680c62ababa6c2f Mon Sep 17 00:00:00 2001 From: Martin Wilck Date: Wed, 14 Jun 2023 12:36:16 +0200 Subject: [PATCH 7/7] scsi: core: Improve warning message in scsi_device_block() If __scsi_internal_device_block() returns an error, it is always -EINVAL because of an invalid state transition. For debugging purposes, it makes more sense to print the device state. Signed-off-by: Martin Wilck Link: https://lore.kernel.org/r/20230614103616.31857-8-mwilck@suse.com Reviewed-by: Christoph Hellwig Reviewed-by: Hannes Reinecke Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index 55a327b2dd8b..ea21e11ed51e 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -2784,9 +2784,11 @@ EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait); static void scsi_device_block(struct scsi_device *sdev, void *data) { int err; + enum scsi_device_state state; mutex_lock(&sdev->state_mutex); err = __scsi_internal_device_block_nowait(sdev); + state = sdev->sdev_state; if (err == 0) /* * scsi_stop_queue() must be called with the state_mutex @@ -2797,8 +2799,8 @@ static void scsi_device_block(struct scsi_device *sdev, void *data) mutex_unlock(&sdev->state_mutex); - WARN_ONCE(err, "__scsi_internal_device_block_nowait(%s) failed: err = %d\n", - dev_name(&sdev->sdev_gendev), err); + WARN_ONCE(err, "%s: failed to block %s in state %d\n", + __func__, dev_name(&sdev->sdev_gendev), state); } /**