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 (!rq->q->seq_zones_wlock)
63 if (blk_rq_is_passthrough(rq))
67 case REQ_OP_WRITE_ZEROES:
68 case REQ_OP_WRITE_SAME:
70 return blk_rq_zone_is_seq(rq);
75 EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
77 bool blk_req_zone_write_trylock(struct request *rq)
79 unsigned int zno = blk_rq_zone_no(rq);
81 if (test_and_set_bit(zno, rq->q->seq_zones_wlock))
84 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
85 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
89 EXPORT_SYMBOL_GPL(blk_req_zone_write_trylock);
91 void __blk_req_zone_write_lock(struct request *rq)
93 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
94 rq->q->seq_zones_wlock)))
97 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
98 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
100 EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
102 void __blk_req_zone_write_unlock(struct request *rq)
104 rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED;
105 if (rq->q->seq_zones_wlock)
106 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq),
107 rq->q->seq_zones_wlock));
109 EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
112 * blkdev_nr_zones - Get number of zones
113 * @disk: Target gendisk
115 * Return the total number of zones of a zoned block device. For a block
116 * device without zone capabilities, the number of zones is always 0.
118 unsigned int blkdev_nr_zones(struct gendisk *disk)
120 sector_t zone_sectors = blk_queue_zone_sectors(disk->queue);
122 if (!blk_queue_is_zoned(disk->queue))
124 return (get_capacity(disk) + zone_sectors - 1) >> ilog2(zone_sectors);
126 EXPORT_SYMBOL_GPL(blkdev_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 (!blk_queue_is_zoned(bdev_get_queue(bdev)) ||
154 WARN_ON_ONCE(!disk->fops->report_zones))
157 if (!nr_zones || sector >= capacity)
160 return disk->fops->report_zones(disk, sector, nr_zones, cb, data);
162 EXPORT_SYMBOL_GPL(blkdev_report_zones);
164 static inline bool blkdev_allow_reset_all_zones(struct block_device *bdev,
168 if (!blk_queue_zone_resetall(bdev_get_queue(bdev)))
172 * REQ_OP_ZONE_RESET_ALL can be executed only if the number of sectors
173 * of the applicable zone range is the entire disk.
175 return !sector && nr_sectors == get_capacity(bdev->bd_disk);
179 * blkdev_zone_mgmt - Execute a zone management operation on a range of zones
180 * @bdev: Target block device
181 * @op: Operation to be performed on the zones
182 * @sector: Start sector of the first zone to operate on
183 * @nr_sectors: Number of sectors, should be at least the length of one zone and
184 * must be zone size aligned.
185 * @gfp_mask: Memory allocation flags (for bio_alloc)
188 * Perform the specified operation on the range of zones specified by
189 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
190 * is valid, but the specified range should not contain conventional zones.
191 * The operation to execute on each zone can be a zone reset, open, close
194 int blkdev_zone_mgmt(struct block_device *bdev, enum req_opf op,
195 sector_t sector, sector_t nr_sectors,
198 struct request_queue *q = bdev_get_queue(bdev);
199 sector_t zone_sectors = blk_queue_zone_sectors(q);
200 sector_t capacity = get_capacity(bdev->bd_disk);
201 sector_t end_sector = sector + nr_sectors;
202 struct bio *bio = NULL;
205 if (!blk_queue_is_zoned(q))
208 if (bdev_read_only(bdev))
211 if (!op_is_zone_mgmt(op))
214 if (end_sector <= sector || end_sector > capacity)
218 /* Check alignment (handle eventual smaller last zone) */
219 if (sector & (zone_sectors - 1))
222 if ((nr_sectors & (zone_sectors - 1)) && end_sector != capacity)
225 while (sector < end_sector) {
226 bio = blk_next_bio(bio, 0, gfp_mask);
227 bio_set_dev(bio, bdev);
230 * Special case for the zone reset operation that reset all
231 * zones, this is useful for applications like mkfs.
233 if (op == REQ_OP_ZONE_RESET &&
234 blkdev_allow_reset_all_zones(bdev, sector, nr_sectors)) {
235 bio->bi_opf = REQ_OP_ZONE_RESET_ALL | REQ_SYNC;
239 bio->bi_opf = op | REQ_SYNC;
240 bio->bi_iter.bi_sector = sector;
241 sector += zone_sectors;
243 /* This may take a while, so be nice to others */
247 ret = submit_bio_wait(bio);
252 EXPORT_SYMBOL_GPL(blkdev_zone_mgmt);
254 struct zone_report_args {
255 struct blk_zone __user *zones;
258 static int blkdev_copy_zone_to_user(struct blk_zone *zone, unsigned int idx,
261 struct zone_report_args *args = data;
263 if (copy_to_user(&args->zones[idx], zone, sizeof(struct blk_zone)))
269 * BLKREPORTZONE ioctl processing.
270 * Called from blkdev_ioctl.
272 int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
273 unsigned int cmd, unsigned long arg)
275 void __user *argp = (void __user *)arg;
276 struct zone_report_args args;
277 struct request_queue *q;
278 struct blk_zone_report rep;
284 q = bdev_get_queue(bdev);
288 if (!blk_queue_is_zoned(q))
291 if (!capable(CAP_SYS_ADMIN))
294 if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report)))
300 args.zones = argp + sizeof(struct blk_zone_report);
301 ret = blkdev_report_zones(bdev, rep.sector, rep.nr_zones,
302 blkdev_copy_zone_to_user, &args);
307 rep.flags = BLK_ZONE_REP_CAPACITY;
308 if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report)))
313 static int blkdev_truncate_zone_range(struct block_device *bdev, fmode_t mode,
314 const struct blk_zone_range *zrange)
318 if (zrange->sector + zrange->nr_sectors <= zrange->sector ||
319 zrange->sector + zrange->nr_sectors > get_capacity(bdev->bd_disk))
323 start = zrange->sector << SECTOR_SHIFT;
324 end = ((zrange->sector + zrange->nr_sectors) << SECTOR_SHIFT) - 1;
326 return truncate_bdev_range(bdev, mode, start, end);
330 * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
331 * Called from blkdev_ioctl.
333 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
334 unsigned int cmd, unsigned long arg)
336 void __user *argp = (void __user *)arg;
337 struct request_queue *q;
338 struct blk_zone_range zrange;
345 q = bdev_get_queue(bdev);
349 if (!blk_queue_is_zoned(q))
352 if (!capable(CAP_SYS_ADMIN))
355 if (!(mode & FMODE_WRITE))
358 if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range)))
363 op = REQ_OP_ZONE_RESET;
365 /* Invalidate the page cache, including dirty pages. */
366 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
371 op = REQ_OP_ZONE_OPEN;
374 op = REQ_OP_ZONE_CLOSE;
377 op = REQ_OP_ZONE_FINISH;
383 ret = blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors,
387 * Invalidate the page cache again for zone reset: writes can only be
388 * direct for zoned devices so concurrent writes would not add any page
389 * to the page cache after/during reset. The page cache may be filled
390 * again due to concurrent reads though and dropping the pages for
393 if (!ret && cmd == BLKRESETZONE)
394 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
399 static inline unsigned long *blk_alloc_zone_bitmap(int node,
400 unsigned int nr_zones)
402 return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long),
406 void blk_queue_free_zone_bitmaps(struct request_queue *q)
408 kfree(q->conv_zones_bitmap);
409 q->conv_zones_bitmap = NULL;
410 kfree(q->seq_zones_wlock);
411 q->seq_zones_wlock = NULL;
414 struct blk_revalidate_zone_args {
415 struct gendisk *disk;
416 unsigned long *conv_zones_bitmap;
417 unsigned long *seq_zones_wlock;
418 unsigned int nr_zones;
419 sector_t zone_sectors;
424 * Helper function to check the validity of zones of a zoned block device.
426 static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx,
429 struct blk_revalidate_zone_args *args = data;
430 struct gendisk *disk = args->disk;
431 struct request_queue *q = disk->queue;
432 sector_t capacity = get_capacity(disk);
435 * All zones must have the same size, with the exception on an eventual
438 if (zone->start == 0) {
439 if (zone->len == 0 || !is_power_of_2(zone->len)) {
440 pr_warn("%s: Invalid zoned device with non power of two zone size (%llu)\n",
441 disk->disk_name, zone->len);
445 args->zone_sectors = zone->len;
446 args->nr_zones = (capacity + zone->len - 1) >> ilog2(zone->len);
447 } else if (zone->start + args->zone_sectors < capacity) {
448 if (zone->len != args->zone_sectors) {
449 pr_warn("%s: Invalid zoned device with non constant zone size\n",
454 if (zone->len > args->zone_sectors) {
455 pr_warn("%s: Invalid zoned device with larger last zone size\n",
461 /* Check for holes in the zone report */
462 if (zone->start != args->sector) {
463 pr_warn("%s: Zone gap at sectors %llu..%llu\n",
464 disk->disk_name, args->sector, zone->start);
468 /* Check zone type */
469 switch (zone->type) {
470 case BLK_ZONE_TYPE_CONVENTIONAL:
471 if (!args->conv_zones_bitmap) {
472 args->conv_zones_bitmap =
473 blk_alloc_zone_bitmap(q->node, args->nr_zones);
474 if (!args->conv_zones_bitmap)
477 set_bit(idx, args->conv_zones_bitmap);
479 case BLK_ZONE_TYPE_SEQWRITE_REQ:
480 case BLK_ZONE_TYPE_SEQWRITE_PREF:
481 if (!args->seq_zones_wlock) {
482 args->seq_zones_wlock =
483 blk_alloc_zone_bitmap(q->node, args->nr_zones);
484 if (!args->seq_zones_wlock)
489 pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n",
490 disk->disk_name, (int)zone->type, zone->start);
494 args->sector += zone->len;
499 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
501 * @update_driver_data: Callback to update driver data on the frozen disk
503 * Helper function for low-level device drivers to (re) allocate and initialize
504 * a disk request queue zone bitmaps. This functions should normally be called
505 * within the disk ->revalidate method for blk-mq based drivers. For BIO based
506 * drivers only q->nr_zones needs to be updated so that the sysfs exposed value
508 * If the @update_driver_data callback function is not NULL, the callback is
509 * executed with the device request queue frozen after all zones have been
512 int blk_revalidate_disk_zones(struct gendisk *disk,
513 void (*update_driver_data)(struct gendisk *disk))
515 struct request_queue *q = disk->queue;
516 struct blk_revalidate_zone_args args = {
519 unsigned int noio_flag;
522 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
524 if (WARN_ON_ONCE(!queue_is_mq(q)))
527 if (!get_capacity(disk))
531 * Ensure that all memory allocations in this context are done as if
532 * GFP_NOIO was specified.
534 noio_flag = memalloc_noio_save();
535 ret = disk->fops->report_zones(disk, 0, UINT_MAX,
536 blk_revalidate_zone_cb, &args);
538 pr_warn("%s: No zones reported\n", disk->disk_name);
541 memalloc_noio_restore(noio_flag);
544 * If zones where reported, make sure that the entire disk capacity
547 if (ret > 0 && args.sector != get_capacity(disk)) {
548 pr_warn("%s: Missing zones from sector %llu\n",
549 disk->disk_name, args.sector);
554 * Install the new bitmaps and update nr_zones only once the queue is
555 * stopped and all I/Os are completed (i.e. a scheduler is not
556 * referencing the bitmaps).
558 blk_mq_freeze_queue(q);
560 blk_queue_chunk_sectors(q, args.zone_sectors);
561 q->nr_zones = args.nr_zones;
562 swap(q->seq_zones_wlock, args.seq_zones_wlock);
563 swap(q->conv_zones_bitmap, args.conv_zones_bitmap);
564 if (update_driver_data)
565 update_driver_data(disk);
568 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
569 blk_queue_free_zone_bitmaps(q);
571 blk_mq_unfreeze_queue(q);
573 kfree(args.seq_zones_wlock);
574 kfree(args.conv_zones_bitmap);
577 EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);
579 void blk_queue_clear_zone_settings(struct request_queue *q)
581 blk_mq_freeze_queue(q);
583 blk_queue_free_zone_bitmaps(q);
584 blk_queue_flag_clear(QUEUE_FLAG_ZONE_RESETALL, q);
585 q->required_elevator_features &= ~ELEVATOR_F_ZBD_SEQ_WRITE;
587 q->max_open_zones = 0;
588 q->max_active_zones = 0;
589 q->limits.chunk_sectors = 0;
590 q->limits.zone_write_granularity = 0;
591 q->limits.max_zone_append_sectors = 0;
593 blk_mq_unfreeze_queue(q);