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);
55 static inline sector_t blk_zone_start(struct request_queue *q,
58 sector_t zone_mask = blk_queue_zone_sectors(q) - 1;
60 return sector & ~zone_mask;
64 * Return true if a request is a write requests that needs zone write locking.
66 bool blk_req_needs_zone_write_lock(struct request *rq)
68 if (!rq->q->seq_zones_wlock)
71 if (blk_rq_is_passthrough(rq))
75 case REQ_OP_WRITE_ZEROES:
76 case REQ_OP_WRITE_SAME:
78 return blk_rq_zone_is_seq(rq);
83 EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
85 bool blk_req_zone_write_trylock(struct request *rq)
87 unsigned int zno = blk_rq_zone_no(rq);
89 if (test_and_set_bit(zno, rq->q->seq_zones_wlock))
92 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
93 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
97 EXPORT_SYMBOL_GPL(blk_req_zone_write_trylock);
99 void __blk_req_zone_write_lock(struct request *rq)
101 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
102 rq->q->seq_zones_wlock)))
105 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
106 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
108 EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
110 void __blk_req_zone_write_unlock(struct request *rq)
112 rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED;
113 if (rq->q->seq_zones_wlock)
114 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq),
115 rq->q->seq_zones_wlock));
117 EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
120 * blkdev_nr_zones - Get number of zones
121 * @disk: Target gendisk
123 * Return the total number of zones of a zoned block device. For a block
124 * device without zone capabilities, the number of zones is always 0.
126 unsigned int blkdev_nr_zones(struct gendisk *disk)
128 sector_t zone_sectors = blk_queue_zone_sectors(disk->queue);
130 if (!blk_queue_is_zoned(disk->queue))
132 return (get_capacity(disk) + zone_sectors - 1) >> ilog2(zone_sectors);
134 EXPORT_SYMBOL_GPL(blkdev_nr_zones);
137 * blkdev_report_zones - Get zones information
138 * @bdev: Target block device
139 * @sector: Sector from which to report zones
140 * @nr_zones: Maximum number of zones to report
141 * @cb: Callback function called for each reported zone
142 * @data: Private data for the callback
145 * Get zone information starting from the zone containing @sector for at most
146 * @nr_zones, and call @cb for each zone reported by the device.
147 * To report all zones in a device starting from @sector, the BLK_ALL_ZONES
148 * constant can be passed to @nr_zones.
149 * Returns the number of zones reported by the device, or a negative errno
150 * value in case of failure.
152 * Note: The caller must use memalloc_noXX_save/restore() calls to control
153 * memory allocations done within this function.
155 int blkdev_report_zones(struct block_device *bdev, sector_t sector,
156 unsigned int nr_zones, report_zones_cb cb, void *data)
158 struct gendisk *disk = bdev->bd_disk;
159 sector_t capacity = get_capacity(disk);
161 if (!blk_queue_is_zoned(bdev_get_queue(bdev)) ||
162 WARN_ON_ONCE(!disk->fops->report_zones))
165 if (!nr_zones || sector >= capacity)
168 return disk->fops->report_zones(disk, sector, nr_zones, cb, data);
170 EXPORT_SYMBOL_GPL(blkdev_report_zones);
172 static inline bool blkdev_allow_reset_all_zones(struct block_device *bdev,
176 if (!blk_queue_zone_resetall(bdev_get_queue(bdev)))
180 * REQ_OP_ZONE_RESET_ALL can be executed only if the number of sectors
181 * of the applicable zone range is the entire disk.
183 return !sector && nr_sectors == get_capacity(bdev->bd_disk);
187 * blkdev_zone_mgmt - Execute a zone management operation on a range of zones
188 * @bdev: Target block device
189 * @op: Operation to be performed on the zones
190 * @sector: Start sector of the first zone to operate on
191 * @nr_sectors: Number of sectors, should be at least the length of one zone and
192 * must be zone size aligned.
193 * @gfp_mask: Memory allocation flags (for bio_alloc)
196 * Perform the specified operation on the range of zones specified by
197 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
198 * is valid, but the specified range should not contain conventional zones.
199 * The operation to execute on each zone can be a zone reset, open, close
202 int blkdev_zone_mgmt(struct block_device *bdev, enum req_opf op,
203 sector_t sector, sector_t nr_sectors,
206 struct request_queue *q = bdev_get_queue(bdev);
207 sector_t zone_sectors = blk_queue_zone_sectors(q);
208 sector_t capacity = get_capacity(bdev->bd_disk);
209 sector_t end_sector = sector + nr_sectors;
210 struct bio *bio = NULL;
213 if (!blk_queue_is_zoned(q))
216 if (bdev_read_only(bdev))
219 if (!op_is_zone_mgmt(op))
222 if (end_sector <= sector || end_sector > capacity)
226 /* Check alignment (handle eventual smaller last zone) */
227 if (sector & (zone_sectors - 1))
230 if ((nr_sectors & (zone_sectors - 1)) && end_sector != capacity)
233 while (sector < end_sector) {
234 bio = blk_next_bio(bio, 0, gfp_mask);
235 bio_set_dev(bio, bdev);
238 * Special case for the zone reset operation that reset all
239 * zones, this is useful for applications like mkfs.
241 if (op == REQ_OP_ZONE_RESET &&
242 blkdev_allow_reset_all_zones(bdev, sector, nr_sectors)) {
243 bio->bi_opf = REQ_OP_ZONE_RESET_ALL | REQ_SYNC;
247 bio->bi_opf = op | REQ_SYNC;
248 bio->bi_iter.bi_sector = sector;
249 sector += zone_sectors;
251 /* This may take a while, so be nice to others */
255 ret = submit_bio_wait(bio);
260 EXPORT_SYMBOL_GPL(blkdev_zone_mgmt);
262 struct zone_report_args {
263 struct blk_zone __user *zones;
266 static int blkdev_copy_zone_to_user(struct blk_zone *zone, unsigned int idx,
269 struct zone_report_args *args = data;
271 if (copy_to_user(&args->zones[idx], zone, sizeof(struct blk_zone)))
277 * BLKREPORTZONE ioctl processing.
278 * Called from blkdev_ioctl.
280 int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
281 unsigned int cmd, unsigned long arg)
283 void __user *argp = (void __user *)arg;
284 struct zone_report_args args;
285 struct request_queue *q;
286 struct blk_zone_report rep;
292 q = bdev_get_queue(bdev);
296 if (!blk_queue_is_zoned(q))
299 if (!capable(CAP_SYS_ADMIN))
302 if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report)))
308 args.zones = argp + sizeof(struct blk_zone_report);
309 ret = blkdev_report_zones(bdev, rep.sector, rep.nr_zones,
310 blkdev_copy_zone_to_user, &args);
315 rep.flags = BLK_ZONE_REP_CAPACITY;
316 if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report)))
321 static int blkdev_truncate_zone_range(struct block_device *bdev, fmode_t mode,
322 const struct blk_zone_range *zrange)
326 if (zrange->sector + zrange->nr_sectors <= zrange->sector ||
327 zrange->sector + zrange->nr_sectors > get_capacity(bdev->bd_disk))
331 start = zrange->sector << SECTOR_SHIFT;
332 end = ((zrange->sector + zrange->nr_sectors) << SECTOR_SHIFT) - 1;
334 return truncate_bdev_range(bdev, mode, start, end);
338 * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
339 * Called from blkdev_ioctl.
341 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
342 unsigned int cmd, unsigned long arg)
344 void __user *argp = (void __user *)arg;
345 struct request_queue *q;
346 struct blk_zone_range zrange;
353 q = bdev_get_queue(bdev);
357 if (!blk_queue_is_zoned(q))
360 if (!capable(CAP_SYS_ADMIN))
363 if (!(mode & FMODE_WRITE))
366 if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range)))
371 op = REQ_OP_ZONE_RESET;
373 /* Invalidate the page cache, including dirty pages. */
374 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
379 op = REQ_OP_ZONE_OPEN;
382 op = REQ_OP_ZONE_CLOSE;
385 op = REQ_OP_ZONE_FINISH;
391 ret = blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors,
395 * Invalidate the page cache again for zone reset: writes can only be
396 * direct for zoned devices so concurrent writes would not add any page
397 * to the page cache after/during reset. The page cache may be filled
398 * again due to concurrent reads though and dropping the pages for
401 if (!ret && cmd == BLKRESETZONE)
402 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
407 static inline unsigned long *blk_alloc_zone_bitmap(int node,
408 unsigned int nr_zones)
410 return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long),
414 void blk_queue_free_zone_bitmaps(struct request_queue *q)
416 kfree(q->conv_zones_bitmap);
417 q->conv_zones_bitmap = NULL;
418 kfree(q->seq_zones_wlock);
419 q->seq_zones_wlock = NULL;
422 struct blk_revalidate_zone_args {
423 struct gendisk *disk;
424 unsigned long *conv_zones_bitmap;
425 unsigned long *seq_zones_wlock;
426 unsigned int nr_zones;
427 sector_t zone_sectors;
432 * Helper function to check the validity of zones of a zoned block device.
434 static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx,
437 struct blk_revalidate_zone_args *args = data;
438 struct gendisk *disk = args->disk;
439 struct request_queue *q = disk->queue;
440 sector_t capacity = get_capacity(disk);
443 * All zones must have the same size, with the exception on an eventual
446 if (zone->start == 0) {
447 if (zone->len == 0 || !is_power_of_2(zone->len)) {
448 pr_warn("%s: Invalid zoned device with non power of two zone size (%llu)\n",
449 disk->disk_name, zone->len);
453 args->zone_sectors = zone->len;
454 args->nr_zones = (capacity + zone->len - 1) >> ilog2(zone->len);
455 } else if (zone->start + args->zone_sectors < capacity) {
456 if (zone->len != args->zone_sectors) {
457 pr_warn("%s: Invalid zoned device with non constant zone size\n",
462 if (zone->len > args->zone_sectors) {
463 pr_warn("%s: Invalid zoned device with larger last zone size\n",
469 /* Check for holes in the zone report */
470 if (zone->start != args->sector) {
471 pr_warn("%s: Zone gap at sectors %llu..%llu\n",
472 disk->disk_name, args->sector, zone->start);
476 /* Check zone type */
477 switch (zone->type) {
478 case BLK_ZONE_TYPE_CONVENTIONAL:
479 if (!args->conv_zones_bitmap) {
480 args->conv_zones_bitmap =
481 blk_alloc_zone_bitmap(q->node, args->nr_zones);
482 if (!args->conv_zones_bitmap)
485 set_bit(idx, args->conv_zones_bitmap);
487 case BLK_ZONE_TYPE_SEQWRITE_REQ:
488 case BLK_ZONE_TYPE_SEQWRITE_PREF:
489 if (!args->seq_zones_wlock) {
490 args->seq_zones_wlock =
491 blk_alloc_zone_bitmap(q->node, args->nr_zones);
492 if (!args->seq_zones_wlock)
497 pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n",
498 disk->disk_name, (int)zone->type, zone->start);
502 args->sector += zone->len;
507 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
509 * @update_driver_data: Callback to update driver data on the frozen disk
511 * Helper function for low-level device drivers to (re) allocate and initialize
512 * a disk request queue zone bitmaps. This functions should normally be called
513 * within the disk ->revalidate method for blk-mq based drivers. For BIO based
514 * drivers only q->nr_zones needs to be updated so that the sysfs exposed value
516 * If the @update_driver_data callback function is not NULL, the callback is
517 * executed with the device request queue frozen after all zones have been
520 int blk_revalidate_disk_zones(struct gendisk *disk,
521 void (*update_driver_data)(struct gendisk *disk))
523 struct request_queue *q = disk->queue;
524 struct blk_revalidate_zone_args args = {
527 unsigned int noio_flag;
530 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
532 if (WARN_ON_ONCE(!queue_is_mq(q)))
535 if (!get_capacity(disk))
539 * Ensure that all memory allocations in this context are done as if
540 * GFP_NOIO was specified.
542 noio_flag = memalloc_noio_save();
543 ret = disk->fops->report_zones(disk, 0, UINT_MAX,
544 blk_revalidate_zone_cb, &args);
546 pr_warn("%s: No zones reported\n", disk->disk_name);
549 memalloc_noio_restore(noio_flag);
552 * If zones where reported, make sure that the entire disk capacity
555 if (ret > 0 && args.sector != get_capacity(disk)) {
556 pr_warn("%s: Missing zones from sector %llu\n",
557 disk->disk_name, args.sector);
562 * Install the new bitmaps and update nr_zones only once the queue is
563 * stopped and all I/Os are completed (i.e. a scheduler is not
564 * referencing the bitmaps).
566 blk_mq_freeze_queue(q);
568 blk_queue_chunk_sectors(q, args.zone_sectors);
569 q->nr_zones = args.nr_zones;
570 swap(q->seq_zones_wlock, args.seq_zones_wlock);
571 swap(q->conv_zones_bitmap, args.conv_zones_bitmap);
572 if (update_driver_data)
573 update_driver_data(disk);
576 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
577 blk_queue_free_zone_bitmaps(q);
579 blk_mq_unfreeze_queue(q);
581 kfree(args.seq_zones_wlock);
582 kfree(args.conv_zones_bitmap);
585 EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);
587 void blk_queue_clear_zone_settings(struct request_queue *q)
589 blk_mq_freeze_queue(q);
591 blk_queue_free_zone_bitmaps(q);
592 blk_queue_flag_clear(QUEUE_FLAG_ZONE_RESETALL, q);
593 q->required_elevator_features &= ~ELEVATOR_F_ZBD_SEQ_WRITE;
595 q->max_open_zones = 0;
596 q->max_active_zones = 0;
597 q->limits.chunk_sectors = 0;
598 q->limits.zone_write_granularity = 0;
599 q->limits.max_zone_append_sectors = 0;
601 blk_mq_unfreeze_queue(q);