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>
17 #include <linux/sched/mm.h>
21 static inline sector_t blk_zone_start(struct request_queue *q,
24 sector_t zone_mask = blk_queue_zone_sectors(q) - 1;
26 return sector & ~zone_mask;
30 * Return true if a request is a write requests that needs zone write locking.
32 bool blk_req_needs_zone_write_lock(struct request *rq)
34 if (!rq->q->seq_zones_wlock)
37 if (blk_rq_is_passthrough(rq))
41 case REQ_OP_WRITE_ZEROES:
42 case REQ_OP_WRITE_SAME:
44 return blk_rq_zone_is_seq(rq);
49 EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
51 void __blk_req_zone_write_lock(struct request *rq)
53 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
54 rq->q->seq_zones_wlock)))
57 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
58 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
60 EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
62 void __blk_req_zone_write_unlock(struct request *rq)
64 rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED;
65 if (rq->q->seq_zones_wlock)
66 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq),
67 rq->q->seq_zones_wlock));
69 EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
71 static inline unsigned int __blkdev_nr_zones(struct request_queue *q,
74 sector_t zone_sectors = blk_queue_zone_sectors(q);
76 return (nr_sectors + zone_sectors - 1) >> ilog2(zone_sectors);
80 * blkdev_nr_zones - Get number of zones
81 * @bdev: Target block device
84 * Return the total number of zones of a zoned block device.
85 * For a regular block device, the number of zones is always 0.
87 unsigned int blkdev_nr_zones(struct block_device *bdev)
89 struct request_queue *q = bdev_get_queue(bdev);
91 if (!blk_queue_is_zoned(q))
94 return __blkdev_nr_zones(q, bdev->bd_part->nr_sects);
96 EXPORT_SYMBOL_GPL(blkdev_nr_zones);
99 * Check that a zone report belongs to this partition, and if yes, fix its start
100 * sector and write pointer and return true. Return false otherwise.
102 static bool blkdev_report_zone(struct block_device *bdev, struct blk_zone *rep)
104 sector_t offset = get_start_sect(bdev);
106 if (rep->start < offset)
109 rep->start -= offset;
110 if (rep->start + rep->len > bdev->bd_part->nr_sects)
113 if (rep->type == BLK_ZONE_TYPE_CONVENTIONAL)
114 rep->wp = rep->start + rep->len;
120 static int blk_report_zones(struct gendisk *disk, sector_t sector,
121 struct blk_zone *zones, unsigned int *nr_zones)
123 struct request_queue *q = disk->queue;
124 unsigned int z = 0, n, nrz = *nr_zones;
125 sector_t capacity = get_capacity(disk);
128 while (z < nrz && sector < capacity) {
130 ret = disk->fops->report_zones(disk, sector, &zones[z], &n);
135 sector += blk_queue_zone_sectors(q) * n;
139 WARN_ON(z > *nr_zones);
146 * blkdev_report_zones - Get zones information
147 * @bdev: Target block device
148 * @sector: Sector from which to report zones
149 * @zones: Array of zone structures where to return the zones information
150 * @nr_zones: Number of zone structures in the zone array
153 * Get zone information starting from the zone containing @sector.
154 * The number of zone information reported may be less than the number
155 * requested by @nr_zones. The number of zones actually reported is
156 * returned in @nr_zones.
157 * The caller must use memalloc_noXX_save/restore() calls to control
158 * memory allocations done within this function (zone array and command
159 * buffer allocation by the device driver).
161 int blkdev_report_zones(struct block_device *bdev, sector_t sector,
162 struct blk_zone *zones, unsigned int *nr_zones)
164 struct request_queue *q = bdev_get_queue(bdev);
168 if (!blk_queue_is_zoned(q))
172 * A block device that advertized itself as zoned must have a
173 * report_zones method. If it does not have one defined, the device
174 * driver has a bug. So warn about that.
176 if (WARN_ON_ONCE(!bdev->bd_disk->fops->report_zones))
179 if (!*nr_zones || sector >= bdev->bd_part->nr_sects) {
185 __blkdev_nr_zones(q, bdev->bd_part->nr_sects - sector));
186 ret = blk_report_zones(bdev->bd_disk, get_start_sect(bdev) + sector,
191 for (i = 0; i < nrz; i++) {
192 if (!blkdev_report_zone(bdev, zones))
201 EXPORT_SYMBOL_GPL(blkdev_report_zones);
204 * blkdev_reset_zones - Reset zones write pointer
205 * @bdev: Target block device
206 * @sector: Start sector of the first zone to reset
207 * @nr_sectors: Number of sectors, at least the length of one zone
208 * @gfp_mask: Memory allocation flags (for bio_alloc)
211 * Reset the write pointer of the zones contained in the range
212 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
213 * is valid, but the specified range should not contain conventional zones.
215 int blkdev_reset_zones(struct block_device *bdev,
216 sector_t sector, sector_t nr_sectors,
219 struct request_queue *q = bdev_get_queue(bdev);
220 sector_t zone_sectors;
221 sector_t end_sector = sector + nr_sectors;
222 struct bio *bio = NULL;
223 struct blk_plug plug;
226 if (!blk_queue_is_zoned(q))
229 if (bdev_read_only(bdev))
232 if (!nr_sectors || end_sector > bdev->bd_part->nr_sects)
236 /* Check alignment (handle eventual smaller last zone) */
237 zone_sectors = blk_queue_zone_sectors(q);
238 if (sector & (zone_sectors - 1))
241 if ((nr_sectors & (zone_sectors - 1)) &&
242 end_sector != bdev->bd_part->nr_sects)
245 blk_start_plug(&plug);
246 while (sector < end_sector) {
248 bio = blk_next_bio(bio, 0, gfp_mask);
249 bio->bi_iter.bi_sector = sector;
250 bio_set_dev(bio, bdev);
251 bio_set_op_attrs(bio, REQ_OP_ZONE_RESET, 0);
253 sector += zone_sectors;
255 /* This may take a while, so be nice to others */
260 ret = submit_bio_wait(bio);
263 blk_finish_plug(&plug);
267 EXPORT_SYMBOL_GPL(blkdev_reset_zones);
270 * BLKREPORTZONE ioctl processing.
271 * Called from blkdev_ioctl.
273 int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
274 unsigned int cmd, unsigned long arg)
276 void __user *argp = (void __user *)arg;
277 struct request_queue *q;
278 struct blk_zone_report rep;
279 struct blk_zone *zones;
285 q = bdev_get_queue(bdev);
289 if (!blk_queue_is_zoned(q))
292 if (!capable(CAP_SYS_ADMIN))
295 if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report)))
301 rep.nr_zones = min(blkdev_nr_zones(bdev), rep.nr_zones);
303 zones = kvmalloc_array(rep.nr_zones, sizeof(struct blk_zone),
304 GFP_KERNEL | __GFP_ZERO);
308 ret = blkdev_report_zones(bdev, rep.sector, zones, &rep.nr_zones);
312 if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report))) {
318 if (copy_to_user(argp + sizeof(struct blk_zone_report), zones,
319 sizeof(struct blk_zone) * rep.nr_zones))
330 * BLKRESETZONE ioctl processing.
331 * Called from blkdev_ioctl.
333 int blkdev_reset_zones_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;
343 q = bdev_get_queue(bdev);
347 if (!blk_queue_is_zoned(q))
350 if (!capable(CAP_SYS_ADMIN))
353 if (!(mode & FMODE_WRITE))
356 if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range)))
359 return blkdev_reset_zones(bdev, zrange.sector, zrange.nr_sectors,
363 static inline unsigned long *blk_alloc_zone_bitmap(int node,
364 unsigned int nr_zones)
366 return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long),
371 * Allocate an array of struct blk_zone to get nr_zones zone information.
372 * The allocated array may be smaller than nr_zones.
374 static struct blk_zone *blk_alloc_zones(int node, unsigned int *nr_zones)
376 size_t size = *nr_zones * sizeof(struct blk_zone);
380 for (order = get_order(size); order >= 0; order--) {
381 page = alloc_pages_node(node, GFP_NOIO | __GFP_ZERO, order);
383 *nr_zones = min_t(unsigned int, *nr_zones,
384 (PAGE_SIZE << order) / sizeof(struct blk_zone));
385 return page_address(page);
392 void blk_queue_free_zone_bitmaps(struct request_queue *q)
394 kfree(q->seq_zones_bitmap);
395 q->seq_zones_bitmap = NULL;
396 kfree(q->seq_zones_wlock);
397 q->seq_zones_wlock = NULL;
401 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
404 * Helper function for low-level device drivers to (re) allocate and initialize
405 * a disk request queue zone bitmaps. This functions should normally be called
406 * within the disk ->revalidate method. For BIO based queues, no zone bitmap
409 int blk_revalidate_disk_zones(struct gendisk *disk)
411 struct request_queue *q = disk->queue;
412 unsigned int nr_zones = __blkdev_nr_zones(q, get_capacity(disk));
413 unsigned long *seq_zones_wlock = NULL, *seq_zones_bitmap = NULL;
414 unsigned int i, rep_nr_zones = 0, z = 0, nrz;
415 struct blk_zone *zones = NULL;
416 unsigned int noio_flag;
421 * BIO based queues do not use a scheduler so only q->nr_zones
422 * needs to be updated so that the sysfs exposed value is correct.
424 if (!queue_is_mq(q)) {
425 q->nr_zones = nr_zones;
430 * Ensure that all memory allocations in this context are done as
431 * if GFP_NOIO was specified.
433 noio_flag = memalloc_noio_save();
435 if (!blk_queue_is_zoned(q) || !nr_zones) {
440 /* Allocate bitmaps */
442 seq_zones_wlock = blk_alloc_zone_bitmap(q->node, nr_zones);
443 if (!seq_zones_wlock)
445 seq_zones_bitmap = blk_alloc_zone_bitmap(q->node, nr_zones);
446 if (!seq_zones_bitmap)
449 /* Get zone information and initialize seq_zones_bitmap */
450 rep_nr_zones = nr_zones;
451 zones = blk_alloc_zones(q->node, &rep_nr_zones);
455 while (z < nr_zones) {
456 nrz = min(nr_zones - z, rep_nr_zones);
457 ret = blk_report_zones(disk, sector, zones, &nrz);
462 for (i = 0; i < nrz; i++) {
463 if (zones[i].type != BLK_ZONE_TYPE_CONVENTIONAL)
464 set_bit(z, seq_zones_bitmap);
467 sector += nrz * blk_queue_zone_sectors(q);
470 if (WARN_ON(z != nr_zones)) {
477 * Install the new bitmaps, making sure the queue is stopped and
478 * all I/Os are completed (i.e. a scheduler is not referencing the
481 blk_mq_freeze_queue(q);
482 q->nr_zones = nr_zones;
483 swap(q->seq_zones_wlock, seq_zones_wlock);
484 swap(q->seq_zones_bitmap, seq_zones_bitmap);
485 blk_mq_unfreeze_queue(q);
488 memalloc_noio_restore(noio_flag);
490 free_pages((unsigned long)zones,
491 get_order(rep_nr_zones * sizeof(struct blk_zone)));
492 kfree(seq_zones_wlock);
493 kfree(seq_zones_bitmap);
496 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
497 blk_mq_freeze_queue(q);
498 blk_queue_free_zone_bitmaps(q);
499 blk_mq_unfreeze_queue(q);
504 EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);