move more work to disk_release v4
Subsystems (3)
Hi all,
this series resurrects and forward ports ports larger parts of the "block: don't drain file system I/O on del_gendisk" series from Ming, but does not remove the draining in del_gendisk, but instead the one in the sd driver, which always was a bit ad-hoc. As part of that sd and sr are switched to use the new ->free_disk method to avoid having to clear disk->private_data and the way to lookup the SCSI ULP is cleaned up as well.
The latest version only has cosmetic changes and plenty of rewiews and should be ready to merge now.
Git branch:
git://git.infradead.org/users/hch/block.git freeze-5.18
Gitweb:
http://git.infradead.org/users/hch/block.git/shortlog/refs/heads/freeze-5.18
Changes since v3:
- make use of existing local variables and helpers
- update a few commit logs
- better document the disk_dev in the sd driver
Changes since v2:
- handle the weird dm-multipath flush sequence corner case when assinging rq->part
Changes since v1:
- fix a refcounting bug in sd
- rename a function
Diffstat:
block/blk-core.c | 7 -- block/blk-mq.c | 15 +++-- block/blk-sysfs.c | 25 -------- block/blk.h | 2 block/elevator.c | 6 +- block/genhd.c | 38 ++++++++++++- drivers/scsi/sd.c | 114 +++++++++------------------------------ drivers/scsi/sd.h | 13 +++- drivers/scsi/sr.c | 129 +++++++++------------------------------------ drivers/scsi/sr.h | 5 - drivers/scsi/st.c | 1 drivers/scsi/st.h | 1 include/scsi/scsi_cmnd.h | 9 --- include/scsi/scsi_driver.h | 9 ++- 14 files changed, 124 insertions(+), 250 deletions(-)
PATCH 1/14 blk-mq: do not include passthrough requests in I/O accounting
I/O accounting buckets I/O into the read/write/discard categories into which passthrough I/O does not fit at all. It also accounts to the block_device, which may not even exist for passthrough I/O.
--- block/blk-mq.c | 11 ++++++++--- block/blk.h | 2 +- 2 files changed, 9 insertions(+), 4 deletions(-)
block/blk-mq.c +8 -3
@@ -883,10 +883,15 @@ static inline void blk_account_io_done(struct request *req, u64 now)static void __blk_account_io_start (struct request *rq) { -/* passthrough requests can hold bios that do not have ->bi_bdev set */ -if (rq->bio && rq->bio->bi_bdev) +/* +* All non-passthrough requests are created from a bio with one +* exception: when a flush command that is part of a flush sequence +* generated by the state machine in blk-flush.c is cloned onto the +* lower device by dm-multipath we can get here without a bio. +*/ +if (rq->bio) rq->part = rq->bio->bi_bdev; -else if (rq->q->disk) +else rq->part = rq->q->disk->part0;part_stat_lock ();
block/blk.h +1 -1
@@ -325,7 +325,7 @@ int blk_dev_init(void);*/ static inline bool blk_do_io_stat (struct request *rq) { -return (rq->rq_flags &RQF_IO_STAT ) && rq->q->disk; +return (rq->rq_flags &RQF_IO_STAT ) && !blk_rq_is_passthrough (rq); }void update_io_ticks (struct block_device *part,unsigned long now,bool end);
Signature · 1 line
2.30.2
PATCH 2/14 blk-mq: handle already freed tags gracefully in blk_mq_free_rqs
In reply to Christoph HellwigFrom: Ming Lei <[email protected]>
To simplify further changes allow for double calling blk_mq_free_rqs on a queue.
--- block/blk-mq.c | 3 +++ 1 file changed, 3 insertions(+)
block/blk-mq.c +3 -0
@@ -3070,6 +3070,9 @@ void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,struct blk_mq_tags *drv_tags;struct page *page; +if (list_empty (&tags->page_list)) +return ; +if (blk_mq_is_shared_tags (set->flags)) drv_tags = set->shared_tags;else
Signature · 1 line
2.30.2
PATCH 3/14 scsi: don't use disk->private_data to find the scsi_driver
In reply to Christoph HellwigRequiring every ULP to have the scsi_drive as first member of the private data is rather fragile and not necessary anyway. Just use the driver hanging off the SCSI device instead.
--- drivers/scsi/sd.c | 3 +-- drivers/scsi/sd.h | 3 +-- drivers/scsi/sr.c | 5 ++--- drivers/scsi/sr.h | 1 - drivers/scsi/st.c | 1 - drivers/scsi/st.h | 1 - include/scsi/scsi_cmnd.h | 9 --------- include/scsi/scsi_driver.h | 9 +++++++-- 8 files changed, 11 insertions(+), 21 deletions(-)
drivers/scsi/sd.c +1 -2
@@ -3515,7 +3515,6 @@ static int sd_probe(struct device *dev) } sdkp->device = sdp; - sdkp->driver = &sd_template; sdkp->disk = gd; sdkp->index = index; sdkp->max_retries =SD_MAX_RETRIES ;
@@ -3548,7 +3547,7 @@ static int sd_probe(struct device *dev) gd->minors =SD_MINORS ; gd->fops = &sd_fops; - gd->private_data = &sdkp->driver; + gd->private_data = sdkp;/* defaults, until the device tells us otherwise */ sdp->sector_size =512 ;
drivers/scsi/sd.h +1 -2
@@ -68,7 +68,6 @@ enum { };struct scsi_disk { -struct scsi_driver *driver;/* always &sd_template */ struct scsi_device *device;struct device dev;struct gendisk *disk;
@@ -131,7 +130,7 @@ struct scsi_disk {static inline struct scsi_disk *scsi_disk (struct gendisk *disk) { - return container_of(disk->private_data, struct scsi_disk, driver); +return disk->private_data; } #define sd_printk(prefix, sdsk, fmt, a...) \
drivers/scsi/sr.c +2 -3
@@ -147,7 +147,7 @@ static void sr_kref_release(struct kref *kref);static inline struct scsi_cd *scsi_cd (struct gendisk *disk) { - return container_of(disk->private_data, struct scsi_cd, driver); +return disk->private_data; }static int sr_runtime_suspend (struct device *dev)
@@ -692,7 +692,6 @@ static int sr_probe(struct device *dev) cd->device = sdev; cd->disk = disk; - cd->driver = &sr_template; cd->capacity =0x1fffff ; cd->device->changed =1 ;/* force recheck CD type */ cd->media_present =1 ;
@@ -713,7 +712,7 @@ static int sr_probe(struct device *dev)sr_vendor_init (cd);set_capacity (disk, cd->capacity); - disk->private_data = &cd->driver; + disk->private_data = cd;if (register_cdrom (disk, &cd->cdi)) goto fail_minor;
drivers/scsi/sr.h +0 -1
@@ -32,7 +32,6 @@ struct scsi_device; typedef struct scsi_cd { - struct scsi_driver *driver; unsigned capacity; /* size in blocks */ struct scsi_device *device; unsigned int vendor; /* vendor code, see sr_vendor.c */
drivers/scsi/st.c +0 -1
@@ -4276,7 +4276,6 @@ static int st_probe(struct device *dev) goto out_buffer_free; }kref_init (&tpnt->kref); - tpnt->driver = &st_template; tpnt->device = SDp;if (SDp->scsi_level <=2 )
drivers/scsi/st.h +0 -1
@@ -117,7 +117,6 @@ struct scsi_tape_stats {/* The tape drive descriptor */ struct scsi_tape { -struct scsi_driver *driver;struct scsi_device *device;struct mutex lock;/* For serialization */ struct completion wait;/* For SCSI commands */
include/scsi/scsi_cmnd.h +0 -9
@@ -13,7 +13,6 @@#include <scsi/scsi_request.h> struct Scsi_Host ; -struct scsi_driver ;/* * MAX_COMMAND_SIZE is:
@@ -159,14 +158,6 @@ static inline void *scsi_cmd_priv(struct scsi_cmnd *cmd)return cmd +1 ; } -/* make sure not to use it with passthrough commands */ -static inline struct scsi_driver *scsi_cmd_to_driver (struct scsi_cmnd *cmd) -{ -struct request *rq =scsi_cmd_to_rq (cmd); - -return *(struct scsi_driver **)rq->q->disk->private_data; -} -void scsi_done (struct scsi_cmnd *cmd);extern void scsi_finish_command (struct scsi_cmnd *cmd);
include/scsi/scsi_driver.h +7 -2
@@ -4,11 +4,10 @@#include <linux/blk_types.h> #include <linux/device.h> +#include <scsi/scsi_cmnd.h> struct module ;struct request ; -struct scsi_cmnd ; -struct scsi_device ;struct scsi_driver {struct device_driver gendrv;
@@ -31,4 +30,10 @@ extern int scsi_register_interface(struct class_interface *);#define scsi_unregister_interface (intf) \ class_interface_unregister(intf) +/* make sure not to use it with passthrough commands */ +static inline struct scsi_driver *scsi_cmd_to_driver (struct scsi_cmnd *cmd) +{ +return to_scsi_driver (cmd->device->sdev_gendev.driver); +} +#endif /* _SCSI_SCSI_DRIVER_H */
Signature · 1 line
2.30.2
PATCH 4/14 sd: rename the scsi_disk.dev field
In reply to Christoph Hellwigdev is very hard to grep for. Give the field a more descriptive name and documents its purpose.
--- drivers/scsi/sd.c | 22 +++++++++++----------- drivers/scsi/sd.h | 9 +++++++-- 2 files changed, 18 insertions(+), 13 deletions(-)
drivers/scsi/sd.c +11 -11
@@ -672,7 +672,7 @@ static struct scsi_disk *scsi_disk_get(struct gendisk *disk)if (disk->private_data) { sdkp =scsi_disk (disk);if (scsi_device_get (sdkp->device) ==0 ) -get_device (&sdkp->dev); +get_device (&sdkp->disk_dev);else sdkp =NULL ; }
@@ -685,7 +685,7 @@ static void scsi_disk_put(struct scsi_disk *sdkp)struct scsi_device *sdev = sdkp->device;mutex_lock (&sd_ref_mutex); -put_device (&sdkp->dev); +put_device (&sdkp->disk_dev);scsi_device_put (sdev);mutex_unlock (&sd_ref_mutex); }
@@ -3529,14 +3529,14 @@ static int sd_probe(struct device *dev) SD_MOD_TIMEOUT); } -device_initialize (&sdkp->dev); - sdkp->dev.parent =get_device (dev); - sdkp->dev.class = &sd_disk_class; -dev_set_name (&sdkp->dev,"%s" ,dev_name (dev)); +device_initialize (&sdkp->disk_dev); + sdkp->disk_dev.parent =get_device (dev); + sdkp->disk_dev.class = &sd_disk_class; +dev_set_name (&sdkp->disk_dev,"%s" ,dev_name (dev)); - error =device_add (&sdkp->dev); + error =device_add (&sdkp->disk_dev);if (error) { -put_device (&sdkp->dev); +put_device (&sdkp->disk_dev); goto out; }
@@ -3577,7 +3577,7 @@ static int sd_probe(struct device *dev) error =device_add_disk (dev, gd,NULL );if (error) { -put_device (&sdkp->dev); +put_device (&sdkp->disk_dev); goto out; }
@@ -3628,7 +3628,7 @@ static int sd_remove(struct device *dev) sdkp =dev_get_drvdata (dev);scsi_autopm_get_device (sdkp->device); -device_del (&sdkp->dev); +device_del (&sdkp->disk_dev);del_gendisk (sdkp->disk);sd_shutdown (dev);
@@ -3636,7 +3636,7 @@ static int sd_remove(struct device *dev)mutex_lock (&sd_ref_mutex);dev_set_drvdata (dev,NULL ); -put_device (&sdkp->dev); +put_device (&sdkp->disk_dev);mutex_unlock (&sd_ref_mutex);return 0 ;
drivers/scsi/sd.h +7 -2
@@ -69,7 +69,12 @@ enum {struct scsi_disk {struct scsi_device *device; -struct device dev; + +/* +* disk_dev is used to show attributes in /sys/class/scsi_disk/, +* but otherwise not really needed. Do not use for refcounting. +*/ +struct device disk_dev;struct gendisk *disk;struct opal_dev *opal_dev;#ifdef CONFIG_BLK_DEV_ZONED
@@ -126,7 +131,7 @@ struct scsi_disk { unsigned security : 1; unsigned ignore_medium_access_errors : 1; }; -#define to_scsi_disk (obj) container_of(obj,struct scsi_disk,dev) +#define to_scsi_disk (obj) container_of(obj, struct scsi_disk, disk_dev)static inline struct scsi_disk *scsi_disk (struct gendisk *disk) {
Signature · 1 line
2.30.2
PATCH 5/14 sd: call sd_zbc_release_disk before releasing the scsi_device reference
In reply to Christoph Hellwigsd_zbc_release_disk accesses disk->device, so ensure that actually still has a valid reference.
--- drivers/scsi/sd.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/scsi/sd.c +1 -1
@@ -3672,9 +3672,9 @@ static void scsi_disk_release(struct device *dev) disk->private_data =NULL ;put_disk (disk); -put_device (&sdkp->device->sdev_gendev);sd_zbc_release_disk (sdkp); +put_device (&sdkp->device->sdev_gendev);kfree (sdkp); }
Signature · 1 line
2.30.2
PATCH 6/14 sd: delay calling free_opal_dev
In reply to Christoph HellwigCall free_opal_dev from scsi_disk_release as the opal_dev field is accessed from the ioctl handler, which isn't synchronized vs sd_release and thus can be accessed during or after sd_release was called.
--- drivers/scsi/sd.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/scsi/sd.c +1 -2
@@ -3632,8 +3632,6 @@ static int sd_remove(struct device *dev)del_gendisk (sdkp->disk);sd_shutdown (dev); -free_opal_dev (sdkp->opal_dev); -mutex_lock (&sd_ref_mutex);dev_set_drvdata (dev,NULL );put_device (&sdkp->disk_dev);
@@ -3675,6 +3673,7 @@ static void scsi_disk_release(struct device *dev)sd_zbc_release_disk (sdkp);put_device (&sdkp->device->sdev_gendev); +free_opal_dev (sdkp->opal_dev);kfree (sdkp); }
Signature · 1 line
2.30.2
PATCH 7/14 sd: implement ->free_disk to simplify refcounting
In reply to Christoph HellwigImplement the ->free_disk method to to put struct scsi_disk when the last gendisk reference count goes away. This removes the need to clear ->private_data and thus freeze the queue on unbind.
--- drivers/scsi/sd.c | 90 +++++++++-------------------------------------- 1 file changed, 16 insertions(+), 74 deletions(-)
drivers/scsi/sd.c +16 -74
@@ -121,11 +121,6 @@ static void scsi_disk_release(struct device *cdev);static DEFINE_IDA (sd_index_ida); -/* This semaphore is used to mediate the 0->1 reference get in the -* face of object destruction (i.e. we can't allow a get on an -* object after last put) */ -static DEFINE_MUTEX (sd_ref_mutex); -static struct kmem_cache *sd_cdb_cache;static mempool_t *sd_cdb_pool;static mempool_t *sd_page_pool;
@@ -663,33 +658,6 @@ static int sd_major(int major_idx) } } -static struct scsi_disk *scsi_disk_get (struct gendisk *disk) -{ -struct scsi_disk *sdkp =NULL ; - -mutex_lock (&sd_ref_mutex); - -if (disk->private_data) { - sdkp =scsi_disk (disk); -if (scsi_device_get (sdkp->device) ==0 ) -get_device (&sdkp->disk_dev); -else - sdkp =NULL ; - } -mutex_unlock (&sd_ref_mutex); -return sdkp; -} - -static void scsi_disk_put (struct scsi_disk *sdkp) -{ -struct scsi_device *sdev = sdkp->device; - -mutex_lock (&sd_ref_mutex); -put_device (&sdkp->disk_dev); -scsi_device_put (sdev); -mutex_unlock (&sd_ref_mutex); -} -#ifdef CONFIG_BLK_SED_OPAL static int sd_sec_submit (void *data,u16 spsp,u8 secp,void *buffer,size_t len,bool send)
@@ -1418,17 +1386,15 @@ static bool sd_need_revalidate(struct block_device *bdev,**/ static int sd_open (struct block_device *bdev,fmode_t mode) { -struct scsi_disk *sdkp =scsi_disk_get (bdev->bd_disk); -struct scsi_device *sdev; +struct scsi_disk *sdkp =scsi_disk (bdev->bd_disk); +struct scsi_device *sdev = sdkp->device;int retval; -if (!sdkp) +if (scsi_device_get (sdev))return -ENXIO ;SCSI_LOG_HLQUEUE (3 ,sd_printk (KERN_INFO , sdkp,"sd_open\n" )); - sdev = sdkp->device; -/* * If the device is in error recovery, wait until it is done. * If the device is offline, then disallow any access to it.
@@ -1473,7 +1439,7 @@ static int sd_open(struct block_device *bdev, fmode_t mode)return 0 ; error_out: -scsi_disk_put (sdkp); +scsi_device_put (sdev);return retval; }
@@ -1502,7 +1468,7 @@ static void sd_release(struct gendisk *disk, fmode_t mode)scsi_set_medium_removal (sdev,SCSI_REMOVAL_ALLOW ); } -scsi_disk_put (sdkp); +scsi_device_put (sdev); }static int sd_getgeo (struct block_device *bdev,struct hd_geometry *geo)
@@ -1616,7 +1582,7 @@ static int media_not_present(struct scsi_disk *sdkp,**/ static unsigned int sd_check_events (struct gendisk *disk,unsigned int clearing) { -struct scsi_disk *sdkp =scsi_disk_get (disk); +struct scsi_disk *sdkp = disk->private_data;struct scsi_device *sdp;int retval;bool disk_changed;
@@ -1679,7 +1645,6 @@ static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)*/ disk_changed = sdp->changed; sdp->changed =0 ; -scsi_disk_put (sdkp);return disk_changed ?DISK_EVENT_MEDIA_CHANGE :0 ; }
@@ -1887,6 +1852,13 @@ static const struct pr_ops sd_pr_ops = { .pr_clear = sd_pr_clear, }; +static void scsi_disk_free_disk (struct gendisk *disk) +{ +struct scsi_disk *sdkp =scsi_disk (disk); + +put_device (&sdkp->disk_dev); +} +static const struct block_device_operations sd_fops = { .owner =THIS_MODULE , .open = sd_open,
@@ -1898,6 +1870,7 @@ static const struct block_device_operations sd_fops = { .unlock_native_capacity = sd_unlock_native_capacity, .report_zones = sd_zbc_report_zones, .get_unique_id = sd_get_unique_id, + .free_disk = scsi_disk_free_disk, .pr_ops = &sd_pr_ops, };
@@ -3623,54 +3596,23 @@ static int sd_probe(struct device *dev)**/ static int sd_remove (struct device *dev) { -struct scsi_disk *sdkp; +struct scsi_disk *sdkp =dev_get_drvdata (dev); - sdkp =dev_get_drvdata (dev);scsi_autopm_get_device (sdkp->device);device_del (&sdkp->disk_dev);del_gendisk (sdkp->disk);sd_shutdown (dev); -mutex_lock (&sd_ref_mutex); -dev_set_drvdata (dev,NULL ); -put_device (&sdkp->disk_dev); -mutex_unlock (&sd_ref_mutex); - +put_disk (sdkp->disk);return 0 ; } -/** -* scsi_disk_release - Called to free the scsi_disk structure -* @dev: pointer to embedded class device -* -* sd_ref_mutex must be held entering this routine. Because it is -* called on last put, you should always use the scsi_disk_get() -* scsi_disk_put() helpers which manipulate the semaphore directly -* and never do a direct put_device. -**/ static void scsi_disk_release (struct device *dev) {struct scsi_disk *sdkp =to_scsi_disk (dev); -struct gendisk *disk = sdkp->disk; -struct request_queue *q = disk->queue;ida_free (&sd_index_ida, sdkp->index); - -/* -* Wait until all requests that are in progress have completed. -* This is necessary to avoid that e.g. scsi_end_request() crashes -* due to clearing the disk->private_data pointer. Wait from inside -* scsi_disk_release() instead of from sd_release() to avoid that -* freezing and unfreezing the request queue affects user space I/O -* in case multiple processes open a /dev/sd... node concurrently. -*/ -blk_mq_freeze_queue (q); -blk_mq_unfreeze_queue (q); - - disk->private_data =NULL ; -put_disk (disk); -sd_zbc_release_disk (sdkp);put_device (&sdkp->device->sdev_gendev);free_opal_dev (sdkp->opal_dev);
Signature · 1 line
2.30.2
Re: [PATCH 07/14] sd: implement ->free_disk to simplify refcounting
Quoted 3 lines · Christoph Hellwig
Implement the ->free_disk method to to put struct
scsi_diskwhen the last gendisk reference count goes away. This removes the need to clear ->private_data and thus freeze the queue on unbind.
Nice work!
PATCH 8/14 sr: implement ->free_disk to simplify refcounting
In reply to Christoph HellwigSimplify the refcounting and remove the need to clear disk->private_data by implementing the ->free_disk method.
--- drivers/scsi/sr.c | 124 ++++++++++------------------------------------ drivers/scsi/sr.h | 4 -- 2 files changed, 26 insertions(+), 102 deletions(-)
drivers/scsi/sr.c +26 -98
@@ -109,11 +109,6 @@ static DEFINE_SPINLOCK(sr_index_lock);static struct lock_class_key sr_bio_compl_lkclass; -/* This semaphore is used to mediate the 0->1 reference get in the -* face of object destruction (i.e. we can't allow a get on an -* object after last put) */ -static DEFINE_MUTEX (sr_ref_mutex); -static int sr_open (struct cdrom_device_info *,int );static void sr_release (struct cdrom_device_info *);
@@ -143,8 +138,6 @@ static const struct cdrom_device_ops sr_dops = { .capability = SR_CAPABILITIES, }; -static void sr_kref_release (struct kref *kref); -static inline struct scsi_cd *scsi_cd (struct gendisk *disk) {return disk->private_data;
@@ -163,38 +156,6 @@ static int sr_runtime_suspend(struct device *dev)return 0 ; } -/* -* The get and put routines for the struct scsi_cd. Note this entity -* has a scsi_device pointer and owns a reference to this. -*/ -static inline struct scsi_cd *scsi_cd_get (struct gendisk *disk) -{ -struct scsi_cd *cd =NULL ; - -mutex_lock (&sr_ref_mutex); -if (disk->private_data ==NULL ) - goto out; - cd =scsi_cd (disk); -kref_get (&cd->kref); -if (scsi_device_get (cd->device)) { -kref_put (&cd->kref, sr_kref_release); - cd =NULL ; - } - out: -mutex_unlock (&sr_ref_mutex); -return cd; -} - -static void scsi_cd_put (struct scsi_cd *cd) -{ -struct scsi_device *sdev = cd->device; - -mutex_lock (&sr_ref_mutex); -kref_put (&cd->kref, sr_kref_release); -scsi_device_put (sdev); -mutex_unlock (&sr_ref_mutex); -} -static unsigned int sr_get_events (struct scsi_device *sdev) {u8 buf[8 ];
@@ -522,15 +483,13 @@ static void sr_revalidate_disk(struct scsi_cd *cd)static int sr_block_open (struct block_device *bdev,fmode_t mode) { -struct scsi_cd *cd; -struct scsi_device *sdev; +struct scsi_cd *cd = cd =scsi_cd (bdev->bd_disk); +struct scsi_device *sdev = cd->device;int ret = -ENXIO ; - cd =scsi_cd_get (bdev->bd_disk); -if (!cd) - goto out; +if (scsi_device_get (cd->device)) +return -ENXIO ; - sdev = cd->device;scsi_autopm_get_device (sdev);if (bdev_check_media_change (bdev))sr_revalidate_disk (cd);
@@ -541,9 +500,7 @@ static int sr_block_open(struct block_device *bdev, fmode_t mode)scsi_autopm_put_device (sdev);if (ret) -scsi_cd_put (cd); - -out: +scsi_device_put (cd->device);return ret; }
@@ -555,7 +512,7 @@ static void sr_block_release(struct gendisk *disk, fmode_t mode)cdrom_release (&cd->cdi, mode);mutex_unlock (&cd->lock); -scsi_cd_put (cd); +scsi_device_put (cd->device); } static int sr_block_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
@@ -595,18 +552,24 @@ static int sr_block_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,static unsigned int sr_block_check_events (struct gendisk *disk,unsigned int clearing) { -unsigned int ret =0 ; -struct scsi_cd *cd; +struct scsi_cd *cd = disk->private_data; - cd =scsi_cd_get (disk); -if (!cd) +if (atomic_read (&cd->device->disk_events_disable_depth))return 0 ; +return cdrom_check_events (&cd->cdi, clearing); +} -if (!atomic_read (&cd->device->disk_events_disable_depth)) - ret =cdrom_check_events (&cd->cdi, clearing); +static void sr_free_disk (struct gendisk *disk) +{ +struct scsi_cd *cd = disk->private_data; -scsi_cd_put (cd); -return ret; +spin_lock (&sr_index_lock); +clear_bit (MINOR (disk_devt (disk)), sr_index_bits); +spin_unlock (&sr_index_lock); + +unregister_cdrom (&cd->cdi); +mutex_destroy (&cd->lock); +kfree (cd); } static const struct block_device_operations sr_bdops =
@@ -617,6 +580,7 @@ static const struct block_device_operations sr_bdops = .ioctl = sr_block_ioctl, .compat_ioctl = blkdev_compat_ptr_ioctl, .check_events = sr_block_check_events, + .free_disk = sr_free_disk, };static int sr_open (struct cdrom_device_info *cdi,int purpose)
@@ -660,8 +624,6 @@ static int sr_probe(struct device *dev)if (!cd) goto fail; -kref_init (&cd->kref); - disk =__alloc_disk_node (sdev->request_queue,NUMA_NO_NODE , &sr_bio_compl_lkclass);if (!disk)
@@ -727,10 +689,8 @@ static int sr_probe(struct device *dev)sr_revalidate_disk (cd); error =device_add_disk (&sdev->sdev_gendev, disk,NULL ); -if (error) { -kref_put (&cd->kref, sr_kref_release); - goto fail; - } +if (error) + goto unregister_cdrom;sdev_printk (KERN_DEBUG , sdev,"Attached scsi CD-ROM %s\n" , cd->cdi.name);
@@ -738,6 +698,8 @@ static int sr_probe(struct device *dev)return 0 ; +unregister_cdrom: +unregister_cdrom (&cd->cdi); fail_minor:spin_lock (&sr_index_lock);clear_bit (minor, sr_index_bits);
@@ -1009,36 +971,6 @@ static int sr_read_cdda_bpc(struct cdrom_device_info *cdi, void __user *ubuf,return ret; } - -/** -* sr_kref_release - Called to free the scsi_cd structure -* @kref: pointer to embedded kref -* -* sr_ref_mutex must be held entering this routine. Because it is -* called on last put, you should always use the scsi_cd_get() -* scsi_cd_put() helpers which manipulate the semaphore directly -* and never do a direct kref_put(). -**/ -static void sr_kref_release (struct kref *kref) -{ - struct scsi_cd *cd = container_of(kref, struct scsi_cd, kref); -struct gendisk *disk = cd->disk; - -spin_lock (&sr_index_lock); -clear_bit (MINOR (disk_devt (disk)), sr_index_bits); -spin_unlock (&sr_index_lock); - -unregister_cdrom (&cd->cdi); - - disk->private_data =NULL ; - -put_disk (disk); - -mutex_destroy (&cd->lock); - -kfree (cd); -} -static int sr_remove (struct device *dev) {struct scsi_cd *cd =dev_get_drvdata (dev);
@@ -1046,11 +978,7 @@ static int sr_remove(struct device *dev)scsi_autopm_get_device (cd->device);del_gendisk (cd->disk); -dev_set_drvdata (dev,NULL ); - -mutex_lock (&sr_ref_mutex); -kref_put (&cd->kref, sr_kref_release); -mutex_unlock (&sr_ref_mutex); +put_disk (cd->disk);return 0 ; }
drivers/scsi/sr.h +0 -4
@@ -18,7 +18,6 @@#ifndef _SR_H#define _SR_H -#include <linux/kref.h> #include <linux/mutex.h> #define MAX_RETRIES 3
@@ -51,9 +50,6 @@ typedef struct scsi_cd {struct cdrom_device_info cdi;struct mutex lock; -/* We hold gendisk and scsi_device references on probe and use -* the refs on this kref to decide when to release them */ -struct kref kref;struct gendisk *disk; } Scsi_CD;
Signature · 1 line
2.30.2
PATCH 9/14 block: move blkcg initialization/destroy into disk allocation/release handler
In reply to Christoph HellwigFrom: Ming Lei <[email protected]>
blkcg works on FS bio level, so it is reasonable to make both blkcg and gendisk sharing same lifetime. Meantime there won't be any FS IO when releasing disk, so safe to move blkcg initialization/destroy into disk allocation/release handler
Long term, we can move blkcg into gendisk completely.
--- block/blk-core.c | 5 ----- block/blk-sysfs.c | 7 ------- block/genhd.c | 8 ++++++++ 3 files changed, 8 insertions(+), 12 deletions(-)
block/blk-core.c +0 -5
@@ -496,17 +496,12 @@ struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu) PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) goto fail_stats; -if (blkcg_init_queue (q)) - goto fail_ref; -blk_queue_dma_alignment (q,511 );blk_set_default_limits (&q->limits); q->nr_requests =BLKDEV_DEFAULT_RQ ;return q; -fail_ref: -percpu_ref_exit (&q->q_usage_counter); fail_stats:blk_free_queue_stats (q->stats); fail_split:
block/blk-sysfs.c +0 -7
@@ -751,13 +751,6 @@ static void blk_exit_queue(struct request_queue *q)ioc_clear_queue (q);elevator_exit (q); } - -/* -* Remove all references to @q from the block cgroup controller before -* restoring @q->queue_lock to avoid that restoring this pointer causes -* e.g. blkcg_print_blkgs() to crash. -*/ -blkcg_exit_queue (q); }/**
block/genhd.c +8 -0
@@ -1120,9 +1120,12 @@ static void disk_release(struct device *dev)blk_mq_cancel_work_sync (disk->queue); +blkcg_exit_queue (disk->queue); +disk_release_events (disk);kfree (disk->random);xa_destroy (&disk->part_tbl); + disk->queue->disk =NULL ;blk_put_queue (disk->queue);
@@ -1328,6 +1331,9 @@ struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,if (xa_insert (&disk->part_tbl,0 , disk->part0,GFP_KERNEL )) goto out_destroy_part_tbl; +if (blkcg_init_queue (q)) + goto out_erase_part0; +rand_initialize_disk (disk);disk_to_dev (disk)->class = &block_class;disk_to_dev (disk)->type = &disk_type;
@@ -1340,6 +1346,8 @@ struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,#endif return disk; +out_erase_part0: +xa_erase (&disk->part_tbl,0 ); out_destroy_part_tbl:xa_destroy (&disk->part_tbl); disk->part0->bd_disk =NULL ;
Signature · 1 line
2.30.2
PATCH 10/14 block: don't remove hctx debugfs dir from blk_mq_exit_queue
In reply to Christoph HellwigFrom: Ming Lei <[email protected]>
The queue's top debugfs dir is removed from blk_release_queue(), so all hctx's debugfs dirs are removed from there. Given blk_mq_exit_queue() is only called from blk_cleanup_queue(), it isn't necessary to remove hctx debugfs from blk_mq_exit_queue().
So remove it from blk_mq_exit_queue().
--- block/blk-mq.c | 1 - 1 file changed, 1 deletion(-)
block/blk-mq.c +0 -1
@@ -3434,7 +3434,6 @@ static void blk_mq_exit_hw_queues(struct request_queue *q,queue_for_each_hw_ctx (q, hctx, i) {if (i == nr_queue)break ; -blk_mq_debugfs_unregister_hctx (hctx);blk_mq_exit_hctx (q, set, hctx, i); } }
Signature · 1 line
2.30.2
PATCH 11/14 block: move q_usage_counter release into blk_queue_release
In reply to Christoph HellwigFrom: Ming Lei <[email protected]>
After blk_cleanup_queue() returns, disk may not be released yet, so probably bio may still be submitted and ->q_usage_counter may be touched, so far this way seems safe, but not good from API's viewpoint.
Move the release q_usage_counter into blk_queue_release().
--- block/blk-core.c | 2 -- block/blk-sysfs.c | 2 ++ 2 files changed, 2 insertions(+), 2 deletions(-)
block/blk-core.c +0 -2
@@ -342,8 +342,6 @@ void blk_cleanup_queue(struct request_queue *q)blk_mq_sched_free_rqs (q);mutex_unlock (&q->sysfs_lock); -percpu_ref_exit (&q->q_usage_counter); -/* @q is and will stay empty, shutdown and put */ blk_put_queue (q); }
block/blk-sysfs.c +2 -0
@@ -780,6 +780,8 @@ static void blk_release_queue(struct kobject *kobj)might_sleep (); +percpu_ref_exit (&q->q_usage_counter); +if (q->poll_stat)blk_stat_remove_callback (q, q->poll_cb);blk_stat_free_callback (q->poll_cb);
Signature · 1 line
2.30.2
PATCH 12/14 block: move blk_exit_queue into disk_release
In reply to Christoph HellwigFrom: Ming Lei <[email protected]>
There can't be file system I/O in disk_release(), so move the call to blk_exit_queue() there, preparing to have the teardown of file system I/O only functionality in one place, when the gendisk that is needed for it is torn down.
We still need to freeze queue here since the request is freed after the bio is completed and passthrough request rely on scheduler tags as well.
The disk can be released before or after queue is cleaned up, and we have to free the scheduler request pool before blk_cleanup_queue returns, while the static request pool has to be freed before exiting the I/O scheduler.
--- block/blk-sysfs.c | 16 ---------------- block/genhd.c | 32 +++++++++++++++++++++++++++++++- 2 files changed, 31 insertions(+), 17 deletions(-)
block/blk-sysfs.c +0 -16
@@ -739,20 +739,6 @@ static void blk_free_queue_rcu(struct rcu_head *rcu_head)kmem_cache_free (blk_get_queue_kmem_cache (blk_queue_has_srcu (q)), q); } -/* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */ -static void blk_exit_queue (struct request_queue *q) -{ -/* -* Since the I/O scheduler exit code may access cgroup information, -* perform I/O scheduler exit before disassociating from the block -* cgroup controller. -*/ -if (q->elevator) { -ioc_clear_queue (q); -elevator_exit (q); - } -} -/** * blk_release_queue - releases all allocated resources of the request_queue * @kobj: pointer to a kobject, whose container is a request_queue
@@ -786,8 +772,6 @@ static void blk_release_queue(struct kobject *kobj)blk_stat_remove_callback (q, q->poll_cb);blk_stat_free_callback (q->poll_cb); -blk_exit_queue (q); -blk_free_queue_stats (q->stats);kfree (q->poll_stat);
block/genhd.c +31 -1
@@ -29,6 +29,7 @@#include "blk.h" #include "blk-mq-sched.h" #include "blk-rq-qos.h" +#include "blk-cgroup.h" static struct kobject *block_depr;
@@ -1097,6 +1098,34 @@ static const struct attribute_group *disk_attr_groups[] = { NULL }; +static void disk_release_mq (struct request_queue *q) +{ +blk_mq_cancel_work_sync (q); + +/* +* There can't be any non non-passthrough bios in flight here, but +* requests stay around longer, including passthrough ones so we +* still need to freeze the queue here. +*/ +blk_mq_freeze_queue (q); + +/* +* Since the I/O scheduler exit code may access cgroup information, +* perform I/O scheduler exit before disassociating from the block +* cgroup controller. +*/ +if (q->elevator) { +ioc_clear_queue (q); + +mutex_lock (&q->sysfs_lock); +blk_mq_sched_free_rqs (q); +elevator_exit (q); +mutex_unlock (&q->sysfs_lock); + } + +__blk_mq_unfreeze_queue (q, true); +} +/** * disk_release - releases all allocated resources of the gendisk * @dev: the device representing this disk
@@ -1118,7 +1147,8 @@ static void disk_release(struct device *dev)might_sleep ();WARN_ON_ONCE (disk_live (disk)); -blk_mq_cancel_work_sync (disk->queue); +if (queue_is_mq (disk->queue)) +disk_release_mq (disk->queue);blkcg_exit_queue (disk->queue);
Signature · 1 line
2.30.2
PATCH 13/14 block: do more work in elevator_exit
In reply to Christoph HellwigMove the calls to ioc_clear_queue and blk_mq_sched_free_rqs into elevator_exit. Except for one call where we know we can't have io_cq structures yet these always go together, and that extra call in an error path is harmless.
--- block/elevator.c | 6 +++--- block/genhd.c | 3 --- 2 files changed, 3 insertions(+), 6 deletions(-)
block/elevator.c +3 -3
@@ -192,6 +192,9 @@ void elevator_exit(struct request_queue *q) {struct elevator_queue *e = q->elevator; +ioc_clear_queue (q); +blk_mq_sched_free_rqs (q); +mutex_lock (&e->sysfs_lock);blk_mq_exit_sched (q, e);mutex_unlock (&e->sysfs_lock);
@@ -596,8 +599,6 @@ int elevator_switch_mq(struct request_queue *q,if (q->elevator) {elv_unregister_queue (q); -ioc_clear_queue (q); -blk_mq_sched_free_rqs (q);elevator_exit (q); }
@@ -608,7 +609,6 @@ int elevator_switch_mq(struct request_queue *q,if (new_e) { ret =elv_register_queue (q, true);if (ret) { -blk_mq_sched_free_rqs (q);elevator_exit (q); goto out; }
block/genhd.c +0 -3
@@ -1115,10 +1115,7 @@ static void disk_release_mq(struct request_queue *q)* cgroup controller. */ if (q->elevator) { -ioc_clear_queue (q); -mutex_lock (&q->sysfs_lock); -blk_mq_sched_free_rqs (q);elevator_exit (q);mutex_unlock (&q->sysfs_lock); }
Signature · 1 line
2.30.2
PATCH 14/14 block: move rq_qos_exit() into disk_release()
In reply to Christoph HellwigFrom: Ming Lei <[email protected]>
Keep all teardown of file system I/O related functionality in one place. There can't be file system I/O in disk_release(), so it is safe to move rq_qos_exit() there.
--- block/genhd.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
block/genhd.c +1 -2
@@ -627,7 +627,6 @@ void del_gendisk(struct gendisk *disk)blk_mq_freeze_queue_wait (q); -rq_qos_exit (q);blk_sync_queue (q);blk_flush_integrity ();/*
@@ -1119,7 +1118,7 @@ static void disk_release_mq(struct request_queue *q)elevator_exit (q);mutex_unlock (&q->sysfs_lock); } - +rq_qos_exit (q);__blk_mq_unfreeze_queue (q, true); }
Signature · 1 line
2.30.2
Quoted 9 lines · Christoph Hellwig
this series resurrects and forward ports ports larger parts of the "block: don't drain file system I/O on
del_gendisk" series from Ming, but does not remove the draining indel_gendisk, but instead the one in the sd driver, which always was a bit ad-hoc. As part of that sd and sr are switched to use the new ->free_disk method to avoid having to cleardisk->private_dataand the way to lookup the SCSI ULP is cleaned up as well.[...]
Applied, thanks!
[01/14] blk-mq: do not include passthrough requests in I/O accounting
commit: 00baefa025651b33a67fefcf8f6bf527af7b085f
[02/14] blk-mq: handle already freed tags gracefully in blk_mq_free_rqs
commit: abcc148ab9236abce466d5e5070f43d290cf72c7
[03/14] scsi: don't use disk->private_data to find the scsi_driver
commit: b55ac66299345b24a81444d9222bf0362920b829
[04/14] sd: rename the scsi_disk.dev field
commit: 695a5b27e06242cfc89988d342016697c5f5ab5f
[05/14] sd: call sd_zbc_release_disk before releasing the scsi_device reference
commit: a22da9716c1375b8c0563b607cacdb422680233b
[06/14] sd: delay calling free_opal_dev
commit: f3ca592e32b3b7a74a57b77bf24cdefff1ea53fc
[07/14] sd: implement ->free_disk to simplify refcounting
commit: d31ece5f112968d7efe70fdffa4d39ed6b876f40
[08/14] sr: implement ->free_disk to simplify refcounting
commit: 719468c6d02fdc1a17d7b40ec73cec155cf05945
[09/14] block: move blkcg initialization/destroy into disk allocation/release handler
commit: 754ddb47bbc48c987cd3c718a1ee9f1d83c001f7
[10/14] block: don't remove hctx debugfs dir from blk_mq_exit_queue
commit: f58dfe31e0cd242f1de93d1aad0fbe8af7a09dc2
[11/14] block: move q_usage_counter release into blk_queue_release
commit: 5407dfa49e67160724567e565fa58086a3ec1b8c
[12/14] block: move blk_exit_queue into disk_release
commit: 0f426c2c2c7ea134562e64d7b6ce0acaef67146e
[13/14] block: do more work in elevator_exit
commit: 3e8f2f0ef1000649dbc7d0dd41c714617a1dbc9c
[14/14] block: move rq_qos_exit() into disk_release()
commit: 641fde02765e74b8b5ae8426a02c21abf0ce3a29Best regards,
Signature · 1 line
Jens Axboe