1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Zoned block device handling
4 *
5 * Copyright (c) 2015, Hannes Reinecke
6 * Copyright (c) 2015, SUSE Linux GmbH
7 *
8 * Copyright (c) 2016, Damien Le Moal
9 * Copyright (c) 2016, Western Digital
10 */
11
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/mm.h>
18#include <linux/vmalloc.h>
19#include <linux/sched/mm.h>
20
21#include "blk.h"
22
23#define ZONE_COND_NAME(name) [BLK_ZONE_COND_##name] = #name
24static 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),
31 ZONE_COND_NAME(FULL),
32 ZONE_COND_NAME(OFFLINE),
33};
34#undef ZONE_COND_NAME
35
36/**
37 * blk_zone_cond_str - Return string XXX in BLK_ZONE_COND_XXX.
38 * @zone_cond: BLK_ZONE_COND_XXX.
39 *
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".
43 */
44const char *blk_zone_cond_str(enum blk_zone_cond zone_cond)
45{
46 static const char *zone_cond_str = "UNKNOWN";
47
48 if (zone_cond < ARRAY_SIZE(zone_cond_name) && zone_cond_name[zone_cond])
49 zone_cond_str = zone_cond_name[zone_cond];
50
51 return zone_cond_str;
52}
53EXPORT_SYMBOL_GPL(blk_zone_cond_str);
54
55/*
56 * Return true if a request is a write requests that needs zone write locking.
57 */
58bool blk_req_needs_zone_write_lock(struct request *rq)
59{
60 if (!rq->q->disk->seq_zones_wlock)
61 return false;
62
63 return blk_rq_is_seq_zoned_write(rq);
64}
65EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
66
67bool blk_req_zone_write_trylock(struct request *rq)
68{
69 unsigned int zno = blk_rq_zone_no(rq);
70
71 if (test_and_set_bit(nr: zno, addr: rq->q->disk->seq_zones_wlock))
72 return false;
73
74 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
75 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
76
77 return true;
78}
79EXPORT_SYMBOL_GPL(blk_req_zone_write_trylock);
80
81void __blk_req_zone_write_lock(struct request *rq)
82{
83 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
84 rq->q->disk->seq_zones_wlock)))
85 return;
86
87 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
88 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
89}
90EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
91
92void __blk_req_zone_write_unlock(struct request *rq)
93{
94 rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED;
95 if (rq->q->disk->seq_zones_wlock)
96 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq),
97 rq->q->disk->seq_zones_wlock));
98}
99EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
100
101/**
102 * bdev_nr_zones - Get number of zones
103 * @bdev: Target device
104 *
105 * Return the total number of zones of a zoned block device. For a block
106 * device without zone capabilities, the number of zones is always 0.
107 */
108unsigned int bdev_nr_zones(struct block_device *bdev)
109{
110 sector_t zone_sectors = bdev_zone_sectors(bdev);
111
112 if (!bdev_is_zoned(bdev))
113 return 0;
114 return (bdev_nr_sectors(bdev) + zone_sectors - 1) >>
115 ilog2(zone_sectors);
116}
117EXPORT_SYMBOL_GPL(bdev_nr_zones);
118
119/**
120 * blkdev_report_zones - Get zones information
121 * @bdev: Target block device
122 * @sector: Sector from which to report zones
123 * @nr_zones: Maximum number of zones to report
124 * @cb: Callback function called for each reported zone
125 * @data: Private data for the callback
126 *
127 * Description:
128 * Get zone information starting from the zone containing @sector for at most
129 * @nr_zones, and call @cb for each zone reported by the device.
130 * To report all zones in a device starting from @sector, the BLK_ALL_ZONES
131 * constant can be passed to @nr_zones.
132 * Returns the number of zones reported by the device, or a negative errno
133 * value in case of failure.
134 *
135 * Note: The caller must use memalloc_noXX_save/restore() calls to control
136 * memory allocations done within this function.
137 */
138int blkdev_report_zones(struct block_device *bdev, sector_t sector,
139 unsigned int nr_zones, report_zones_cb cb, void *data)
140{
141 struct gendisk *disk = bdev->bd_disk;
142 sector_t capacity = get_capacity(disk);
143
144 if (!bdev_is_zoned(bdev) || WARN_ON_ONCE(!disk->fops->report_zones))
145 return -EOPNOTSUPP;
146
147 if (!nr_zones || sector >= capacity)
148 return 0;
149
150 return disk->fops->report_zones(disk, sector, nr_zones, cb, data);
151}
152EXPORT_SYMBOL_GPL(blkdev_report_zones);
153
154static inline unsigned long *blk_alloc_zone_bitmap(int node,
155 unsigned int nr_zones)
156{
157 return kcalloc_node(BITS_TO_LONGS(nr_zones), size: sizeof(unsigned long),
158 GFP_NOIO, node);
159}
160
161static int blk_zone_need_reset_cb(struct blk_zone *zone, unsigned int idx,
162 void *data)
163{
164 /*
165 * For an all-zones reset, ignore conventional, empty, read-only
166 * and offline zones.
167 */
168 switch (zone->cond) {
169 case BLK_ZONE_COND_NOT_WP:
170 case BLK_ZONE_COND_EMPTY:
171 case BLK_ZONE_COND_READONLY:
172 case BLK_ZONE_COND_OFFLINE:
173 return 0;
174 default:
175 set_bit(nr: idx, addr: (unsigned long *)data);
176 return 0;
177 }
178}
179
180static int blkdev_zone_reset_all_emulated(struct block_device *bdev,
181 gfp_t gfp_mask)
182{
183 struct gendisk *disk = bdev->bd_disk;
184 sector_t capacity = bdev_nr_sectors(bdev);
185 sector_t zone_sectors = bdev_zone_sectors(bdev);
186 unsigned long *need_reset;
187 struct bio *bio = NULL;
188 sector_t sector = 0;
189 int ret;
190
191 need_reset = blk_alloc_zone_bitmap(node: disk->queue->node, nr_zones: disk->nr_zones);
192 if (!need_reset)
193 return -ENOMEM;
194
195 ret = disk->fops->report_zones(disk, 0, disk->nr_zones,
196 blk_zone_need_reset_cb, need_reset);
197 if (ret < 0)
198 goto out_free_need_reset;
199
200 ret = 0;
201 while (sector < capacity) {
202 if (!test_bit(disk_zone_no(disk, sector), need_reset)) {
203 sector += zone_sectors;
204 continue;
205 }
206
207 bio = blk_next_bio(bio, bdev, nr_pages: 0, opf: REQ_OP_ZONE_RESET | REQ_SYNC,
208 gfp: gfp_mask);
209 bio->bi_iter.bi_sector = sector;
210 sector += zone_sectors;
211
212 /* This may take a while, so be nice to others */
213 cond_resched();
214 }
215
216 if (bio) {
217 ret = submit_bio_wait(bio);
218 bio_put(bio);
219 }
220
221out_free_need_reset:
222 kfree(objp: need_reset);
223 return ret;
224}
225
226static int blkdev_zone_reset_all(struct block_device *bdev, gfp_t gfp_mask)
227{
228 struct bio bio;
229
230 bio_init(bio: &bio, bdev, NULL, max_vecs: 0, opf: REQ_OP_ZONE_RESET_ALL | REQ_SYNC);
231 return submit_bio_wait(bio: &bio);
232}
233
234/**
235 * blkdev_zone_mgmt - Execute a zone management operation on a range of zones
236 * @bdev: Target block device
237 * @op: Operation to be performed on the zones
238 * @sector: Start sector of the first zone to operate on
239 * @nr_sectors: Number of sectors, should be at least the length of one zone and
240 * must be zone size aligned.
241 * @gfp_mask: Memory allocation flags (for bio_alloc)
242 *
243 * Description:
244 * Perform the specified operation on the range of zones specified by
245 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
246 * is valid, but the specified range should not contain conventional zones.
247 * The operation to execute on each zone can be a zone reset, open, close
248 * or finish request.
249 */
250int blkdev_zone_mgmt(struct block_device *bdev, enum req_op op,
251 sector_t sector, sector_t nr_sectors, gfp_t gfp_mask)
252{
253 struct request_queue *q = bdev_get_queue(bdev);
254 sector_t zone_sectors = bdev_zone_sectors(bdev);
255 sector_t capacity = bdev_nr_sectors(bdev);
256 sector_t end_sector = sector + nr_sectors;
257 struct bio *bio = NULL;
258 int ret = 0;
259
260 if (!bdev_is_zoned(bdev))
261 return -EOPNOTSUPP;
262
263 if (bdev_read_only(bdev))
264 return -EPERM;
265
266 if (!op_is_zone_mgmt(op))
267 return -EOPNOTSUPP;
268
269 if (end_sector <= sector || end_sector > capacity)
270 /* Out of range */
271 return -EINVAL;
272
273 /* Check alignment (handle eventual smaller last zone) */
274 if (!bdev_is_zone_start(bdev, sector))
275 return -EINVAL;
276
277 if (!bdev_is_zone_start(bdev, sector: nr_sectors) && end_sector != capacity)
278 return -EINVAL;
279
280 /*
281 * In the case of a zone reset operation over all zones,
282 * REQ_OP_ZONE_RESET_ALL can be used with devices supporting this
283 * command. For other devices, we emulate this command behavior by
284 * identifying the zones needing a reset.
285 */
286 if (op == REQ_OP_ZONE_RESET && sector == 0 && nr_sectors == capacity) {
287 if (!blk_queue_zone_resetall(q))
288 return blkdev_zone_reset_all_emulated(bdev, gfp_mask);
289 return blkdev_zone_reset_all(bdev, gfp_mask);
290 }
291
292 while (sector < end_sector) {
293 bio = blk_next_bio(bio, bdev, nr_pages: 0, opf: op | REQ_SYNC, gfp: gfp_mask);
294 bio->bi_iter.bi_sector = sector;
295 sector += zone_sectors;
296
297 /* This may take a while, so be nice to others */
298 cond_resched();
299 }
300
301 ret = submit_bio_wait(bio);
302 bio_put(bio);
303
304 return ret;
305}
306EXPORT_SYMBOL_GPL(blkdev_zone_mgmt);
307
308struct zone_report_args {
309 struct blk_zone __user *zones;
310};
311
312static int blkdev_copy_zone_to_user(struct blk_zone *zone, unsigned int idx,
313 void *data)
314{
315 struct zone_report_args *args = data;
316
317 if (copy_to_user(to: &args->zones[idx], from: zone, n: sizeof(struct blk_zone)))
318 return -EFAULT;
319 return 0;
320}
321
322/*
323 * BLKREPORTZONE ioctl processing.
324 * Called from blkdev_ioctl.
325 */
326int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd,
327 unsigned long arg)
328{
329 void __user *argp = (void __user *)arg;
330 struct zone_report_args args;
331 struct blk_zone_report rep;
332 int ret;
333
334 if (!argp)
335 return -EINVAL;
336
337 if (!bdev_is_zoned(bdev))
338 return -ENOTTY;
339
340 if (copy_from_user(to: &rep, from: argp, n: sizeof(struct blk_zone_report)))
341 return -EFAULT;
342
343 if (!rep.nr_zones)
344 return -EINVAL;
345
346 args.zones = argp + sizeof(struct blk_zone_report);
347 ret = blkdev_report_zones(bdev, rep.sector, rep.nr_zones,
348 blkdev_copy_zone_to_user, &args);
349 if (ret < 0)
350 return ret;
351
352 rep.nr_zones = ret;
353 rep.flags = BLK_ZONE_REP_CAPACITY;
354 if (copy_to_user(to: argp, from: &rep, n: sizeof(struct blk_zone_report)))
355 return -EFAULT;
356 return 0;
357}
358
359static int blkdev_truncate_zone_range(struct block_device *bdev,
360 blk_mode_t mode, const struct blk_zone_range *zrange)
361{
362 loff_t start, end;
363
364 if (zrange->sector + zrange->nr_sectors <= zrange->sector ||
365 zrange->sector + zrange->nr_sectors > get_capacity(disk: bdev->bd_disk))
366 /* Out of range */
367 return -EINVAL;
368
369 start = zrange->sector << SECTOR_SHIFT;
370 end = ((zrange->sector + zrange->nr_sectors) << SECTOR_SHIFT) - 1;
371
372 return truncate_bdev_range(bdev, mode, lstart: start, lend: end);
373}
374
375/*
376 * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
377 * Called from blkdev_ioctl.
378 */
379int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
380 unsigned int cmd, unsigned long arg)
381{
382 void __user *argp = (void __user *)arg;
383 struct blk_zone_range zrange;
384 enum req_op op;
385 int ret;
386
387 if (!argp)
388 return -EINVAL;
389
390 if (!bdev_is_zoned(bdev))
391 return -ENOTTY;
392
393 if (!(mode & BLK_OPEN_WRITE))
394 return -EBADF;
395
396 if (copy_from_user(to: &zrange, from: argp, n: sizeof(struct blk_zone_range)))
397 return -EFAULT;
398
399 switch (cmd) {
400 case BLKRESETZONE:
401 op = REQ_OP_ZONE_RESET;
402
403 /* Invalidate the page cache, including dirty pages. */
404 filemap_invalidate_lock(mapping: bdev->bd_inode->i_mapping);
405 ret = blkdev_truncate_zone_range(bdev, mode, zrange: &zrange);
406 if (ret)
407 goto fail;
408 break;
409 case BLKOPENZONE:
410 op = REQ_OP_ZONE_OPEN;
411 break;
412 case BLKCLOSEZONE:
413 op = REQ_OP_ZONE_CLOSE;
414 break;
415 case BLKFINISHZONE:
416 op = REQ_OP_ZONE_FINISH;
417 break;
418 default:
419 return -ENOTTY;
420 }
421
422 ret = blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors,
423 GFP_KERNEL);
424
425fail:
426 if (cmd == BLKRESETZONE)
427 filemap_invalidate_unlock(mapping: bdev->bd_inode->i_mapping);
428
429 return ret;
430}
431
432void disk_free_zone_bitmaps(struct gendisk *disk)
433{
434 kfree(objp: disk->conv_zones_bitmap);
435 disk->conv_zones_bitmap = NULL;
436 kfree(objp: disk->seq_zones_wlock);
437 disk->seq_zones_wlock = NULL;
438}
439
440struct blk_revalidate_zone_args {
441 struct gendisk *disk;
442 unsigned long *conv_zones_bitmap;
443 unsigned long *seq_zones_wlock;
444 unsigned int nr_zones;
445 sector_t sector;
446};
447
448/*
449 * Helper function to check the validity of zones of a zoned block device.
450 */
451static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx,
452 void *data)
453{
454 struct blk_revalidate_zone_args *args = data;
455 struct gendisk *disk = args->disk;
456 struct request_queue *q = disk->queue;
457 sector_t capacity = get_capacity(disk);
458 sector_t zone_sectors = q->limits.chunk_sectors;
459
460 /* Check for bad zones and holes in the zone report */
461 if (zone->start != args->sector) {
462 pr_warn("%s: Zone gap at sectors %llu..%llu\n",
463 disk->disk_name, args->sector, zone->start);
464 return -ENODEV;
465 }
466
467 if (zone->start >= capacity || !zone->len) {
468 pr_warn("%s: Invalid zone start %llu, length %llu\n",
469 disk->disk_name, zone->start, zone->len);
470 return -ENODEV;
471 }
472
473 /*
474 * All zones must have the same size, with the exception on an eventual
475 * smaller last zone.
476 */
477 if (zone->start + zone->len < capacity) {
478 if (zone->len != zone_sectors) {
479 pr_warn("%s: Invalid zoned device with non constant zone size\n",
480 disk->disk_name);
481 return -ENODEV;
482 }
483 } else if (zone->len > zone_sectors) {
484 pr_warn("%s: Invalid zoned device with larger last zone size\n",
485 disk->disk_name);
486 return -ENODEV;
487 }
488
489 /* Check zone type */
490 switch (zone->type) {
491 case BLK_ZONE_TYPE_CONVENTIONAL:
492 if (!args->conv_zones_bitmap) {
493 args->conv_zones_bitmap =
494 blk_alloc_zone_bitmap(node: q->node, nr_zones: args->nr_zones);
495 if (!args->conv_zones_bitmap)
496 return -ENOMEM;
497 }
498 set_bit(nr: idx, addr: args->conv_zones_bitmap);
499 break;
500 case BLK_ZONE_TYPE_SEQWRITE_REQ:
501 case BLK_ZONE_TYPE_SEQWRITE_PREF:
502 if (!args->seq_zones_wlock) {
503 args->seq_zones_wlock =
504 blk_alloc_zone_bitmap(node: q->node, nr_zones: args->nr_zones);
505 if (!args->seq_zones_wlock)
506 return -ENOMEM;
507 }
508 break;
509 default:
510 pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n",
511 disk->disk_name, (int)zone->type, zone->start);
512 return -ENODEV;
513 }
514
515 args->sector += zone->len;
516 return 0;
517}
518
519/**
520 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
521 * @disk: Target disk
522 * @update_driver_data: Callback to update driver data on the frozen disk
523 *
524 * Helper function for low-level device drivers to check and (re) allocate and
525 * initialize a disk request queue zone bitmaps. This functions should normally
526 * be called within the disk ->revalidate method for blk-mq based drivers.
527 * Before calling this function, the device driver must already have set the
528 * device zone size (chunk_sector limit) and the max zone append limit.
529 * For BIO based drivers, this function cannot be used. BIO based device drivers
530 * only need to set disk->nr_zones so that the sysfs exposed value is correct.
531 * If the @update_driver_data callback function is not NULL, the callback is
532 * executed with the device request queue frozen after all zones have been
533 * checked.
534 */
535int blk_revalidate_disk_zones(struct gendisk *disk,
536 void (*update_driver_data)(struct gendisk *disk))
537{
538 struct request_queue *q = disk->queue;
539 sector_t zone_sectors = q->limits.chunk_sectors;
540 sector_t capacity = get_capacity(disk);
541 struct blk_revalidate_zone_args args = { };
542 unsigned int noio_flag;
543 int ret;
544
545 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
546 return -EIO;
547 if (WARN_ON_ONCE(!queue_is_mq(q)))
548 return -EIO;
549
550 if (!capacity)
551 return -ENODEV;
552
553 /*
554 * Checks that the device driver indicated a valid zone size and that
555 * the max zone append limit is set.
556 */
557 if (!zone_sectors || !is_power_of_2(n: zone_sectors)) {
558 pr_warn("%s: Invalid non power of two zone size (%llu)\n",
559 disk->disk_name, zone_sectors);
560 return -ENODEV;
561 }
562
563 if (!q->limits.max_zone_append_sectors) {
564 pr_warn("%s: Invalid 0 maximum zone append limit\n",
565 disk->disk_name);
566 return -ENODEV;
567 }
568
569 /*
570 * Ensure that all memory allocations in this context are done as if
571 * GFP_NOIO was specified.
572 */
573 args.disk = disk;
574 args.nr_zones = (capacity + zone_sectors - 1) >> ilog2(zone_sectors);
575 noio_flag = memalloc_noio_save();
576 ret = disk->fops->report_zones(disk, 0, UINT_MAX,
577 blk_revalidate_zone_cb, &args);
578 if (!ret) {
579 pr_warn("%s: No zones reported\n", disk->disk_name);
580 ret = -ENODEV;
581 }
582 memalloc_noio_restore(flags: noio_flag);
583
584 /*
585 * If zones where reported, make sure that the entire disk capacity
586 * has been checked.
587 */
588 if (ret > 0 && args.sector != capacity) {
589 pr_warn("%s: Missing zones from sector %llu\n",
590 disk->disk_name, args.sector);
591 ret = -ENODEV;
592 }
593
594 /*
595 * Install the new bitmaps and update nr_zones only once the queue is
596 * stopped and all I/Os are completed (i.e. a scheduler is not
597 * referencing the bitmaps).
598 */
599 blk_mq_freeze_queue(q);
600 if (ret > 0) {
601 disk->nr_zones = args.nr_zones;
602 swap(disk->seq_zones_wlock, args.seq_zones_wlock);
603 swap(disk->conv_zones_bitmap, args.conv_zones_bitmap);
604 if (update_driver_data)
605 update_driver_data(disk);
606 ret = 0;
607 } else {
608 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
609 disk_free_zone_bitmaps(disk);
610 }
611 blk_mq_unfreeze_queue(q);
612
613 kfree(objp: args.seq_zones_wlock);
614 kfree(objp: args.conv_zones_bitmap);
615 return ret;
616}
617EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);
618
619void disk_clear_zone_settings(struct gendisk *disk)
620{
621 struct request_queue *q = disk->queue;
622
623 blk_mq_freeze_queue(q);
624
625 disk_free_zone_bitmaps(disk);
626 blk_queue_flag_clear(QUEUE_FLAG_ZONE_RESETALL, q);
627 q->required_elevator_features &= ~ELEVATOR_F_ZBD_SEQ_WRITE;
628 disk->nr_zones = 0;
629 disk->max_open_zones = 0;
630 disk->max_active_zones = 0;
631 q->limits.chunk_sectors = 0;
632 q->limits.zone_write_granularity = 0;
633 q->limits.max_zone_append_sectors = 0;
634
635 blk_mq_unfreeze_queue(q);
636}
637

source code of linux/block/blk-zoned.c