1 // SPDX-License-Identifier: GPL-2.0
3 * Zoned block device handling
5 * Copyright (c) 2015, Hannes Reinecke
6 * Copyright (c) 2015, SUSE Linux GmbH
8 * Copyright (c) 2016, Damien Le Moal
9 * Copyright (c) 2016, Western Digital
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/rbtree.h>
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
18 #include <linux/vmalloc.h>
19 #include <linux/sched/mm.h>
23 #define ZONE_COND_NAME(name) [BLK_ZONE_COND_##name] = #name
24 static const char *const zone_cond_name
[] = {
25 ZONE_COND_NAME(NOT_WP
),
26 ZONE_COND_NAME(EMPTY
),
27 ZONE_COND_NAME(IMP_OPEN
),
28 ZONE_COND_NAME(EXP_OPEN
),
29 ZONE_COND_NAME(CLOSED
),
30 ZONE_COND_NAME(READONLY
),
32 ZONE_COND_NAME(OFFLINE
),
37 * blk_zone_cond_str - Return string XXX in BLK_ZONE_COND_XXX.
38 * @zone_cond: BLK_ZONE_COND_XXX.
40 * Description: Centralize block layer function to convert BLK_ZONE_COND_XXX
41 * into string format. Useful in the debugging and tracing zone conditions. For
42 * invalid BLK_ZONE_COND_XXX it returns string "UNKNOWN".
44 const char *blk_zone_cond_str(enum blk_zone_cond zone_cond
)
46 static const char *zone_cond_str
= "UNKNOWN";
48 if (zone_cond
< ARRAY_SIZE(zone_cond_name
) && zone_cond_name
[zone_cond
])
49 zone_cond_str
= zone_cond_name
[zone_cond
];
53 EXPORT_SYMBOL_GPL(blk_zone_cond_str
);
56 * Return true if a request is a write requests that needs zone write locking.
58 bool blk_req_needs_zone_write_lock(struct request
*rq
)
60 if (blk_rq_is_passthrough(rq
))
63 if (!rq
->q
->disk
->seq_zones_wlock
)
67 case REQ_OP_WRITE_ZEROES
:
69 return blk_rq_zone_is_seq(rq
);
74 EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock
);
76 bool blk_req_zone_write_trylock(struct request
*rq
)
78 unsigned int zno
= blk_rq_zone_no(rq
);
80 if (test_and_set_bit(zno
, rq
->q
->disk
->seq_zones_wlock
))
83 WARN_ON_ONCE(rq
->rq_flags
& RQF_ZONE_WRITE_LOCKED
);
84 rq
->rq_flags
|= RQF_ZONE_WRITE_LOCKED
;
88 EXPORT_SYMBOL_GPL(blk_req_zone_write_trylock
);
90 void __blk_req_zone_write_lock(struct request
*rq
)
92 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq
),
93 rq
->q
->disk
->seq_zones_wlock
)))
96 WARN_ON_ONCE(rq
->rq_flags
& RQF_ZONE_WRITE_LOCKED
);
97 rq
->rq_flags
|= RQF_ZONE_WRITE_LOCKED
;
99 EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock
);
101 void __blk_req_zone_write_unlock(struct request
*rq
)
103 rq
->rq_flags
&= ~RQF_ZONE_WRITE_LOCKED
;
104 if (rq
->q
->disk
->seq_zones_wlock
)
105 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq
),
106 rq
->q
->disk
->seq_zones_wlock
));
108 EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock
);
111 * bdev_nr_zones - Get number of zones
112 * @bdev: Target device
114 * Return the total number of zones of a zoned block device. For a block
115 * device without zone capabilities, the number of zones is always 0.
117 unsigned int bdev_nr_zones(struct block_device
*bdev
)
119 sector_t zone_sectors
= bdev_zone_sectors(bdev
);
121 if (!bdev_is_zoned(bdev
))
123 return (bdev_nr_sectors(bdev
) + zone_sectors
- 1) >>
126 EXPORT_SYMBOL_GPL(bdev_nr_zones
);
129 * blkdev_report_zones - Get zones information
130 * @bdev: Target block device
131 * @sector: Sector from which to report zones
132 * @nr_zones: Maximum number of zones to report
133 * @cb: Callback function called for each reported zone
134 * @data: Private data for the callback
137 * Get zone information starting from the zone containing @sector for at most
138 * @nr_zones, and call @cb for each zone reported by the device.
139 * To report all zones in a device starting from @sector, the BLK_ALL_ZONES
140 * constant can be passed to @nr_zones.
141 * Returns the number of zones reported by the device, or a negative errno
142 * value in case of failure.
144 * Note: The caller must use memalloc_noXX_save/restore() calls to control
145 * memory allocations done within this function.
147 int blkdev_report_zones(struct block_device
*bdev
, sector_t sector
,
148 unsigned int nr_zones
, report_zones_cb cb
, void *data
)
150 struct gendisk
*disk
= bdev
->bd_disk
;
151 sector_t capacity
= get_capacity(disk
);
153 if (!bdev_is_zoned(bdev
) || WARN_ON_ONCE(!disk
->fops
->report_zones
))
156 if (!nr_zones
|| sector
>= capacity
)
159 return disk
->fops
->report_zones(disk
, sector
, nr_zones
, cb
, data
);
161 EXPORT_SYMBOL_GPL(blkdev_report_zones
);
163 static inline unsigned long *blk_alloc_zone_bitmap(int node
,
164 unsigned int nr_zones
)
166 return kcalloc_node(BITS_TO_LONGS(nr_zones
), sizeof(unsigned long),
170 static int blk_zone_need_reset_cb(struct blk_zone
*zone
, unsigned int idx
,
174 * For an all-zones reset, ignore conventional, empty, read-only
177 switch (zone
->cond
) {
178 case BLK_ZONE_COND_NOT_WP
:
179 case BLK_ZONE_COND_EMPTY
:
180 case BLK_ZONE_COND_READONLY
:
181 case BLK_ZONE_COND_OFFLINE
:
184 set_bit(idx
, (unsigned long *)data
);
189 static int blkdev_zone_reset_all_emulated(struct block_device
*bdev
,
192 struct gendisk
*disk
= bdev
->bd_disk
;
193 sector_t capacity
= bdev_nr_sectors(bdev
);
194 sector_t zone_sectors
= bdev_zone_sectors(bdev
);
195 unsigned long *need_reset
;
196 struct bio
*bio
= NULL
;
200 need_reset
= blk_alloc_zone_bitmap(disk
->queue
->node
, disk
->nr_zones
);
204 ret
= disk
->fops
->report_zones(disk
, 0, disk
->nr_zones
,
205 blk_zone_need_reset_cb
, need_reset
);
207 goto out_free_need_reset
;
210 while (sector
< capacity
) {
211 if (!test_bit(disk_zone_no(disk
, sector
), need_reset
)) {
212 sector
+= zone_sectors
;
216 bio
= blk_next_bio(bio
, bdev
, 0, REQ_OP_ZONE_RESET
| REQ_SYNC
,
218 bio
->bi_iter
.bi_sector
= sector
;
219 sector
+= zone_sectors
;
221 /* This may take a while, so be nice to others */
226 ret
= submit_bio_wait(bio
);
235 static int blkdev_zone_reset_all(struct block_device
*bdev
, gfp_t gfp_mask
)
239 bio_init(&bio
, bdev
, NULL
, 0, REQ_OP_ZONE_RESET_ALL
| REQ_SYNC
);
240 return submit_bio_wait(&bio
);
244 * blkdev_zone_mgmt - Execute a zone management operation on a range of zones
245 * @bdev: Target block device
246 * @op: Operation to be performed on the zones
247 * @sector: Start sector of the first zone to operate on
248 * @nr_sectors: Number of sectors, should be at least the length of one zone and
249 * must be zone size aligned.
250 * @gfp_mask: Memory allocation flags (for bio_alloc)
253 * Perform the specified operation on the range of zones specified by
254 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
255 * is valid, but the specified range should not contain conventional zones.
256 * The operation to execute on each zone can be a zone reset, open, close
259 int blkdev_zone_mgmt(struct block_device
*bdev
, enum req_op op
,
260 sector_t sector
, sector_t nr_sectors
, gfp_t gfp_mask
)
262 struct request_queue
*q
= bdev_get_queue(bdev
);
263 sector_t zone_sectors
= bdev_zone_sectors(bdev
);
264 sector_t capacity
= bdev_nr_sectors(bdev
);
265 sector_t end_sector
= sector
+ nr_sectors
;
266 struct bio
*bio
= NULL
;
269 if (!bdev_is_zoned(bdev
))
272 if (bdev_read_only(bdev
))
275 if (!op_is_zone_mgmt(op
))
278 if (end_sector
<= sector
|| end_sector
> capacity
)
282 /* Check alignment (handle eventual smaller last zone) */
283 if (sector
& (zone_sectors
- 1))
286 if ((nr_sectors
& (zone_sectors
- 1)) && end_sector
!= capacity
)
290 * In the case of a zone reset operation over all zones,
291 * REQ_OP_ZONE_RESET_ALL can be used with devices supporting this
292 * command. For other devices, we emulate this command behavior by
293 * identifying the zones needing a reset.
295 if (op
== REQ_OP_ZONE_RESET
&& sector
== 0 && nr_sectors
== capacity
) {
296 if (!blk_queue_zone_resetall(q
))
297 return blkdev_zone_reset_all_emulated(bdev
, gfp_mask
);
298 return blkdev_zone_reset_all(bdev
, gfp_mask
);
301 while (sector
< end_sector
) {
302 bio
= blk_next_bio(bio
, bdev
, 0, op
| REQ_SYNC
, gfp_mask
);
303 bio
->bi_iter
.bi_sector
= sector
;
304 sector
+= zone_sectors
;
306 /* This may take a while, so be nice to others */
310 ret
= submit_bio_wait(bio
);
315 EXPORT_SYMBOL_GPL(blkdev_zone_mgmt
);
317 struct zone_report_args
{
318 struct blk_zone __user
*zones
;
321 static int blkdev_copy_zone_to_user(struct blk_zone
*zone
, unsigned int idx
,
324 struct zone_report_args
*args
= data
;
326 if (copy_to_user(&args
->zones
[idx
], zone
, sizeof(struct blk_zone
)))
332 * BLKREPORTZONE ioctl processing.
333 * Called from blkdev_ioctl.
335 int blkdev_report_zones_ioctl(struct block_device
*bdev
, fmode_t mode
,
336 unsigned int cmd
, unsigned long arg
)
338 void __user
*argp
= (void __user
*)arg
;
339 struct zone_report_args args
;
340 struct request_queue
*q
;
341 struct blk_zone_report rep
;
347 q
= bdev_get_queue(bdev
);
351 if (!bdev_is_zoned(bdev
))
354 if (copy_from_user(&rep
, argp
, sizeof(struct blk_zone_report
)))
360 args
.zones
= argp
+ sizeof(struct blk_zone_report
);
361 ret
= blkdev_report_zones(bdev
, rep
.sector
, rep
.nr_zones
,
362 blkdev_copy_zone_to_user
, &args
);
367 rep
.flags
= BLK_ZONE_REP_CAPACITY
;
368 if (copy_to_user(argp
, &rep
, sizeof(struct blk_zone_report
)))
373 static int blkdev_truncate_zone_range(struct block_device
*bdev
, fmode_t mode
,
374 const struct blk_zone_range
*zrange
)
378 if (zrange
->sector
+ zrange
->nr_sectors
<= zrange
->sector
||
379 zrange
->sector
+ zrange
->nr_sectors
> get_capacity(bdev
->bd_disk
))
383 start
= zrange
->sector
<< SECTOR_SHIFT
;
384 end
= ((zrange
->sector
+ zrange
->nr_sectors
) << SECTOR_SHIFT
) - 1;
386 return truncate_bdev_range(bdev
, mode
, start
, end
);
390 * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
391 * Called from blkdev_ioctl.
393 int blkdev_zone_mgmt_ioctl(struct block_device
*bdev
, fmode_t mode
,
394 unsigned int cmd
, unsigned long arg
)
396 void __user
*argp
= (void __user
*)arg
;
397 struct request_queue
*q
;
398 struct blk_zone_range zrange
;
405 q
= bdev_get_queue(bdev
);
409 if (!bdev_is_zoned(bdev
))
412 if (!(mode
& FMODE_WRITE
))
415 if (copy_from_user(&zrange
, argp
, sizeof(struct blk_zone_range
)))
420 op
= REQ_OP_ZONE_RESET
;
422 /* Invalidate the page cache, including dirty pages. */
423 filemap_invalidate_lock(bdev
->bd_inode
->i_mapping
);
424 ret
= blkdev_truncate_zone_range(bdev
, mode
, &zrange
);
429 op
= REQ_OP_ZONE_OPEN
;
432 op
= REQ_OP_ZONE_CLOSE
;
435 op
= REQ_OP_ZONE_FINISH
;
441 ret
= blkdev_zone_mgmt(bdev
, op
, zrange
.sector
, zrange
.nr_sectors
,
445 if (cmd
== BLKRESETZONE
)
446 filemap_invalidate_unlock(bdev
->bd_inode
->i_mapping
);
451 void disk_free_zone_bitmaps(struct gendisk
*disk
)
453 kfree(disk
->conv_zones_bitmap
);
454 disk
->conv_zones_bitmap
= NULL
;
455 kfree(disk
->seq_zones_wlock
);
456 disk
->seq_zones_wlock
= NULL
;
459 struct blk_revalidate_zone_args
{
460 struct gendisk
*disk
;
461 unsigned long *conv_zones_bitmap
;
462 unsigned long *seq_zones_wlock
;
463 unsigned int nr_zones
;
464 sector_t zone_sectors
;
469 * Helper function to check the validity of zones of a zoned block device.
471 static int blk_revalidate_zone_cb(struct blk_zone
*zone
, unsigned int idx
,
474 struct blk_revalidate_zone_args
*args
= data
;
475 struct gendisk
*disk
= args
->disk
;
476 struct request_queue
*q
= disk
->queue
;
477 sector_t capacity
= get_capacity(disk
);
480 * All zones must have the same size, with the exception on an eventual
483 if (zone
->start
== 0) {
484 if (zone
->len
== 0 || !is_power_of_2(zone
->len
)) {
485 pr_warn("%s: Invalid zoned device with non power of two zone size (%llu)\n",
486 disk
->disk_name
, zone
->len
);
490 args
->zone_sectors
= zone
->len
;
491 args
->nr_zones
= (capacity
+ zone
->len
- 1) >> ilog2(zone
->len
);
492 } else if (zone
->start
+ args
->zone_sectors
< capacity
) {
493 if (zone
->len
!= args
->zone_sectors
) {
494 pr_warn("%s: Invalid zoned device with non constant zone size\n",
499 if (zone
->len
> args
->zone_sectors
) {
500 pr_warn("%s: Invalid zoned device with larger last zone size\n",
506 /* Check for holes in the zone report */
507 if (zone
->start
!= args
->sector
) {
508 pr_warn("%s: Zone gap at sectors %llu..%llu\n",
509 disk
->disk_name
, args
->sector
, zone
->start
);
513 /* Check zone type */
514 switch (zone
->type
) {
515 case BLK_ZONE_TYPE_CONVENTIONAL
:
516 if (!args
->conv_zones_bitmap
) {
517 args
->conv_zones_bitmap
=
518 blk_alloc_zone_bitmap(q
->node
, args
->nr_zones
);
519 if (!args
->conv_zones_bitmap
)
522 set_bit(idx
, args
->conv_zones_bitmap
);
524 case BLK_ZONE_TYPE_SEQWRITE_REQ
:
525 case BLK_ZONE_TYPE_SEQWRITE_PREF
:
526 if (!args
->seq_zones_wlock
) {
527 args
->seq_zones_wlock
=
528 blk_alloc_zone_bitmap(q
->node
, args
->nr_zones
);
529 if (!args
->seq_zones_wlock
)
534 pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n",
535 disk
->disk_name
, (int)zone
->type
, zone
->start
);
539 args
->sector
+= zone
->len
;
544 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
546 * @update_driver_data: Callback to update driver data on the frozen disk
548 * Helper function for low-level device drivers to (re) allocate and initialize
549 * a disk request queue zone bitmaps. This functions should normally be called
550 * within the disk ->revalidate method for blk-mq based drivers. For BIO based
551 * drivers only q->nr_zones needs to be updated so that the sysfs exposed value
553 * If the @update_driver_data callback function is not NULL, the callback is
554 * executed with the device request queue frozen after all zones have been
557 int blk_revalidate_disk_zones(struct gendisk
*disk
,
558 void (*update_driver_data
)(struct gendisk
*disk
))
560 struct request_queue
*q
= disk
->queue
;
561 struct blk_revalidate_zone_args args
= {
564 unsigned int noio_flag
;
567 if (WARN_ON_ONCE(!blk_queue_is_zoned(q
)))
569 if (WARN_ON_ONCE(!queue_is_mq(q
)))
572 if (!get_capacity(disk
))
576 * Ensure that all memory allocations in this context are done as if
577 * GFP_NOIO was specified.
579 noio_flag
= memalloc_noio_save();
580 ret
= disk
->fops
->report_zones(disk
, 0, UINT_MAX
,
581 blk_revalidate_zone_cb
, &args
);
583 pr_warn("%s: No zones reported\n", disk
->disk_name
);
586 memalloc_noio_restore(noio_flag
);
589 * If zones where reported, make sure that the entire disk capacity
592 if (ret
> 0 && args
.sector
!= get_capacity(disk
)) {
593 pr_warn("%s: Missing zones from sector %llu\n",
594 disk
->disk_name
, args
.sector
);
599 * Install the new bitmaps and update nr_zones only once the queue is
600 * stopped and all I/Os are completed (i.e. a scheduler is not
601 * referencing the bitmaps).
603 blk_mq_freeze_queue(q
);
605 blk_queue_chunk_sectors(q
, args
.zone_sectors
);
606 disk
->nr_zones
= args
.nr_zones
;
607 swap(disk
->seq_zones_wlock
, args
.seq_zones_wlock
);
608 swap(disk
->conv_zones_bitmap
, args
.conv_zones_bitmap
);
609 if (update_driver_data
)
610 update_driver_data(disk
);
613 pr_warn("%s: failed to revalidate zones\n", disk
->disk_name
);
614 disk_free_zone_bitmaps(disk
);
616 blk_mq_unfreeze_queue(q
);
618 kfree(args
.seq_zones_wlock
);
619 kfree(args
.conv_zones_bitmap
);
622 EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones
);
624 void disk_clear_zone_settings(struct gendisk
*disk
)
626 struct request_queue
*q
= disk
->queue
;
628 blk_mq_freeze_queue(q
);
630 disk_free_zone_bitmaps(disk
);
631 blk_queue_flag_clear(QUEUE_FLAG_ZONE_RESETALL
, q
);
632 q
->required_elevator_features
&= ~ELEVATOR_F_ZBD_SEQ_WRITE
;
634 disk
->max_open_zones
= 0;
635 disk
->max_active_zones
= 0;
636 q
->limits
.chunk_sectors
= 0;
637 q
->limits
.zone_write_granularity
= 0;
638 q
->limits
.max_zone_append_sectors
= 0;
640 blk_mq_unfreeze_queue(q
);