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 static inline sector_t blk_zone_start(struct request_queue *q,
26 sector_t zone_mask = blk_queue_zone_sectors(q) - 1;
28 return sector & ~zone_mask;
32 * Return true if a request is a write requests that needs zone write locking.
34 bool blk_req_needs_zone_write_lock(struct request *rq)
36 if (!rq->q->seq_zones_wlock)
39 if (blk_rq_is_passthrough(rq))
43 case REQ_OP_WRITE_ZEROES:
44 case REQ_OP_WRITE_SAME:
46 return blk_rq_zone_is_seq(rq);
51 EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
53 void __blk_req_zone_write_lock(struct request *rq)
55 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
56 rq->q->seq_zones_wlock)))
59 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
60 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
62 EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
64 void __blk_req_zone_write_unlock(struct request *rq)
66 rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED;
67 if (rq->q->seq_zones_wlock)
68 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq),
69 rq->q->seq_zones_wlock));
71 EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
73 static inline unsigned int __blkdev_nr_zones(struct request_queue *q,
76 sector_t zone_sectors = blk_queue_zone_sectors(q);
78 return (nr_sectors + zone_sectors - 1) >> ilog2(zone_sectors);
82 * blkdev_nr_zones - Get number of zones
83 * @bdev: Target block device
86 * Return the total number of zones of a zoned block device.
87 * For a regular block device, the number of zones is always 0.
89 unsigned int blkdev_nr_zones(struct block_device *bdev)
91 struct request_queue *q = bdev_get_queue(bdev);
93 if (!blk_queue_is_zoned(q))
96 return __blkdev_nr_zones(q, bdev->bd_part->nr_sects);
98 EXPORT_SYMBOL_GPL(blkdev_nr_zones);
101 * Check that a zone report belongs to this partition, and if yes, fix its start
102 * sector and write pointer and return true. Return false otherwise.
104 static bool blkdev_report_zone(struct block_device *bdev, struct blk_zone *rep)
106 sector_t offset = get_start_sect(bdev);
108 if (rep->start < offset)
111 rep->start -= offset;
112 if (rep->start + rep->len > bdev->bd_part->nr_sects)
115 if (rep->type == BLK_ZONE_TYPE_CONVENTIONAL)
116 rep->wp = rep->start + rep->len;
122 static int blk_report_zones(struct gendisk *disk, sector_t sector,
123 struct blk_zone *zones, unsigned int *nr_zones)
125 struct request_queue *q = disk->queue;
126 unsigned int z = 0, n, nrz = *nr_zones;
127 sector_t capacity = get_capacity(disk);
130 while (z < nrz && sector < capacity) {
132 ret = disk->fops->report_zones(disk, sector, &zones[z], &n);
137 sector += blk_queue_zone_sectors(q) * n;
141 WARN_ON(z > *nr_zones);
148 * blkdev_report_zones - Get zones information
149 * @bdev: Target block device
150 * @sector: Sector from which to report zones
151 * @zones: Array of zone structures where to return the zones information
152 * @nr_zones: Number of zone structures in the zone array
155 * Get zone information starting from the zone containing @sector.
156 * The number of zone information reported may be less than the number
157 * requested by @nr_zones. The number of zones actually reported is
158 * returned in @nr_zones.
159 * The caller must use memalloc_noXX_save/restore() calls to control
160 * memory allocations done within this function (zone array and command
161 * buffer allocation by the device driver).
163 int blkdev_report_zones(struct block_device *bdev, sector_t sector,
164 struct blk_zone *zones, unsigned int *nr_zones)
166 struct request_queue *q = bdev_get_queue(bdev);
170 if (!blk_queue_is_zoned(q))
174 * A block device that advertized itself as zoned must have a
175 * report_zones method. If it does not have one defined, the device
176 * driver has a bug. So warn about that.
178 if (WARN_ON_ONCE(!bdev->bd_disk->fops->report_zones))
181 if (!*nr_zones || sector >= bdev->bd_part->nr_sects) {
187 __blkdev_nr_zones(q, bdev->bd_part->nr_sects - sector));
188 ret = blk_report_zones(bdev->bd_disk, get_start_sect(bdev) + sector,
193 for (i = 0; i < nrz; i++) {
194 if (!blkdev_report_zone(bdev, zones))
203 EXPORT_SYMBOL_GPL(blkdev_report_zones);
205 static inline bool blkdev_allow_reset_all_zones(struct block_device *bdev,
209 if (!blk_queue_zone_resetall(bdev_get_queue(bdev)))
212 if (sector || nr_sectors != part_nr_sects_read(bdev->bd_part))
215 * REQ_OP_ZONE_RESET_ALL can be executed only if the block device is
216 * the entire disk, that is, if the blocks device start offset is 0 and
217 * its capacity is the same as the entire disk.
219 return get_start_sect(bdev) == 0 &&
220 part_nr_sects_read(bdev->bd_part) == get_capacity(bdev->bd_disk);
224 * blkdev_zone_mgmt - Execute a zone management operation on a range of zones
225 * @bdev: Target block device
226 * @op: Operation to be performed on the zones
227 * @sector: Start sector of the first zone to operate on
228 * @nr_sectors: Number of sectors, should be at least the length of one zone and
229 * must be zone size aligned.
230 * @gfp_mask: Memory allocation flags (for bio_alloc)
233 * Perform the specified operation on the range of zones specified by
234 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
235 * is valid, but the specified range should not contain conventional zones.
236 * The operation to execute on each zone can be a zone reset, open, close
239 int blkdev_zone_mgmt(struct block_device *bdev, enum req_opf op,
240 sector_t sector, sector_t nr_sectors,
243 struct request_queue *q = bdev_get_queue(bdev);
244 sector_t zone_sectors = blk_queue_zone_sectors(q);
245 sector_t end_sector = sector + nr_sectors;
246 struct bio *bio = NULL;
249 if (!blk_queue_is_zoned(q))
252 if (bdev_read_only(bdev))
255 if (!op_is_zone_mgmt(op))
258 if (!nr_sectors || end_sector > bdev->bd_part->nr_sects)
262 /* Check alignment (handle eventual smaller last zone) */
263 if (sector & (zone_sectors - 1))
266 if ((nr_sectors & (zone_sectors - 1)) &&
267 end_sector != bdev->bd_part->nr_sects)
270 while (sector < end_sector) {
271 bio = blk_next_bio(bio, 0, gfp_mask);
272 bio_set_dev(bio, bdev);
275 * Special case for the zone reset operation that reset all
276 * zones, this is useful for applications like mkfs.
278 if (op == REQ_OP_ZONE_RESET &&
279 blkdev_allow_reset_all_zones(bdev, sector, nr_sectors)) {
280 bio->bi_opf = REQ_OP_ZONE_RESET_ALL;
285 bio->bi_iter.bi_sector = sector;
286 sector += zone_sectors;
288 /* This may take a while, so be nice to others */
292 ret = submit_bio_wait(bio);
297 EXPORT_SYMBOL_GPL(blkdev_zone_mgmt);
300 * BLKREPORTZONE ioctl processing.
301 * Called from blkdev_ioctl.
303 int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
304 unsigned int cmd, unsigned long arg)
306 void __user *argp = (void __user *)arg;
307 struct request_queue *q;
308 struct blk_zone_report rep;
309 struct blk_zone *zones;
315 q = bdev_get_queue(bdev);
319 if (!blk_queue_is_zoned(q))
322 if (!capable(CAP_SYS_ADMIN))
325 if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report)))
331 rep.nr_zones = min(blkdev_nr_zones(bdev), rep.nr_zones);
333 zones = kvmalloc_array(rep.nr_zones, sizeof(struct blk_zone),
334 GFP_KERNEL | __GFP_ZERO);
338 ret = blkdev_report_zones(bdev, rep.sector, zones, &rep.nr_zones);
342 if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report))) {
348 if (copy_to_user(argp + sizeof(struct blk_zone_report), zones,
349 sizeof(struct blk_zone) * rep.nr_zones))
360 * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
361 * Called from blkdev_ioctl.
363 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
364 unsigned int cmd, unsigned long arg)
366 void __user *argp = (void __user *)arg;
367 struct request_queue *q;
368 struct blk_zone_range zrange;
374 q = bdev_get_queue(bdev);
378 if (!blk_queue_is_zoned(q))
381 if (!capable(CAP_SYS_ADMIN))
384 if (!(mode & FMODE_WRITE))
387 if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range)))
392 op = REQ_OP_ZONE_RESET;
395 op = REQ_OP_ZONE_OPEN;
398 op = REQ_OP_ZONE_CLOSE;
401 op = REQ_OP_ZONE_FINISH;
407 return blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors,
411 static inline unsigned long *blk_alloc_zone_bitmap(int node,
412 unsigned int nr_zones)
414 return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long),
419 * Allocate an array of struct blk_zone to get nr_zones zone information.
420 * The allocated array may be smaller than nr_zones.
422 static struct blk_zone *blk_alloc_zones(unsigned int *nr_zones)
424 struct blk_zone *zones;
425 size_t nrz = min(*nr_zones, BLK_ZONED_REPORT_MAX_ZONES);
428 * GFP_KERNEL here is meaningless as the caller task context has
429 * the PF_MEMALLOC_NOIO flag set in blk_revalidate_disk_zones()
430 * with memalloc_noio_save().
432 zones = kvcalloc(nrz, sizeof(struct blk_zone), GFP_KERNEL);
443 void blk_queue_free_zone_bitmaps(struct request_queue *q)
445 kfree(q->seq_zones_bitmap);
446 q->seq_zones_bitmap = NULL;
447 kfree(q->seq_zones_wlock);
448 q->seq_zones_wlock = NULL;
452 * Helper function to check the validity of zones of a zoned block device.
454 static bool blk_zone_valid(struct gendisk *disk, struct blk_zone *zone,
457 struct request_queue *q = disk->queue;
458 sector_t zone_sectors = blk_queue_zone_sectors(q);
459 sector_t capacity = get_capacity(disk);
462 * All zones must have the same size, with the exception on an eventual
465 if (zone->start + zone_sectors < capacity &&
466 zone->len != zone_sectors) {
467 pr_warn("%s: Invalid zoned device with non constant zone size\n",
472 if (zone->start + zone->len >= capacity &&
473 zone->len > zone_sectors) {
474 pr_warn("%s: Invalid zoned device with larger last zone size\n",
479 /* Check for holes in the zone report */
480 if (zone->start != *sector) {
481 pr_warn("%s: Zone gap at sectors %llu..%llu\n",
482 disk->disk_name, *sector, zone->start);
486 /* Check zone type */
487 switch (zone->type) {
488 case BLK_ZONE_TYPE_CONVENTIONAL:
489 case BLK_ZONE_TYPE_SEQWRITE_REQ:
490 case BLK_ZONE_TYPE_SEQWRITE_PREF:
493 pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n",
494 disk->disk_name, (int)zone->type, zone->start);
498 *sector += zone->len;
504 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
507 * Helper function for low-level device drivers to (re) allocate and initialize
508 * a disk request queue zone bitmaps. This functions should normally be called
509 * within the disk ->revalidate method. For BIO based queues, no zone bitmap
512 int blk_revalidate_disk_zones(struct gendisk *disk)
514 struct request_queue *q = disk->queue;
515 unsigned int nr_zones = __blkdev_nr_zones(q, get_capacity(disk));
516 unsigned long *seq_zones_wlock = NULL, *seq_zones_bitmap = NULL;
517 unsigned int i, rep_nr_zones = 0, z = 0, nrz;
518 struct blk_zone *zones = NULL;
519 unsigned int noio_flag;
523 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
527 * BIO based queues do not use a scheduler so only q->nr_zones
528 * needs to be updated so that the sysfs exposed value is correct.
530 if (!queue_is_mq(q)) {
531 q->nr_zones = nr_zones;
536 * Ensure that all memory allocations in this context are done as
537 * if GFP_NOIO was specified.
539 noio_flag = memalloc_noio_save();
544 /* Allocate bitmaps */
546 seq_zones_wlock = blk_alloc_zone_bitmap(q->node, nr_zones);
547 if (!seq_zones_wlock)
549 seq_zones_bitmap = blk_alloc_zone_bitmap(q->node, nr_zones);
550 if (!seq_zones_bitmap)
554 * Get zone information to check the zones and initialize
557 rep_nr_zones = nr_zones;
558 zones = blk_alloc_zones(&rep_nr_zones);
562 while (z < nr_zones) {
563 nrz = min(nr_zones - z, rep_nr_zones);
564 ret = blk_report_zones(disk, sector, zones, &nrz);
569 for (i = 0; i < nrz; i++) {
570 if (!blk_zone_valid(disk, &zones[i], §or)) {
574 if (zones[i].type != BLK_ZONE_TYPE_CONVENTIONAL)
575 set_bit(z, seq_zones_bitmap);
580 if (WARN_ON(z != nr_zones)) {
587 * Install the new bitmaps, making sure the queue is stopped and
588 * all I/Os are completed (i.e. a scheduler is not referencing the
591 blk_mq_freeze_queue(q);
592 q->nr_zones = nr_zones;
593 swap(q->seq_zones_wlock, seq_zones_wlock);
594 swap(q->seq_zones_bitmap, seq_zones_bitmap);
595 blk_mq_unfreeze_queue(q);
598 memalloc_noio_restore(noio_flag);
601 kfree(seq_zones_wlock);
602 kfree(seq_zones_bitmap);
605 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
606 blk_mq_freeze_queue(q);
607 blk_queue_free_zone_bitmaps(q);
608 blk_mq_unfreeze_queue(q);
613 EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);