Loading...
Note: File does not exist in v3.1.
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->seq_zones_wlock)
61 return false;
62
63 if (blk_rq_is_passthrough(rq))
64 return false;
65
66 switch (req_op(rq)) {
67 case REQ_OP_WRITE_ZEROES:
68 case REQ_OP_WRITE_SAME:
69 case REQ_OP_WRITE:
70 return blk_rq_zone_is_seq(rq);
71 default:
72 return false;
73 }
74}
75EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
76
77bool blk_req_zone_write_trylock(struct request *rq)
78{
79 unsigned int zno = blk_rq_zone_no(rq);
80
81 if (test_and_set_bit(zno, rq->q->seq_zones_wlock))
82 return false;
83
84 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
85 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
86
87 return true;
88}
89EXPORT_SYMBOL_GPL(blk_req_zone_write_trylock);
90
91void __blk_req_zone_write_lock(struct request *rq)
92{
93 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
94 rq->q->seq_zones_wlock)))
95 return;
96
97 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
98 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
99}
100EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
101
102void __blk_req_zone_write_unlock(struct request *rq)
103{
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));
108}
109EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
110
111/**
112 * blkdev_nr_zones - Get number of zones
113 * @disk: Target gendisk
114 *
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.
117 */
118unsigned int blkdev_nr_zones(struct gendisk *disk)
119{
120 sector_t zone_sectors = blk_queue_zone_sectors(disk->queue);
121
122 if (!blk_queue_is_zoned(disk->queue))
123 return 0;
124 return (get_capacity(disk) + zone_sectors - 1) >> ilog2(zone_sectors);
125}
126EXPORT_SYMBOL_GPL(blkdev_nr_zones);
127
128/**
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
135 *
136 * Description:
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.
143 *
144 * Note: The caller must use memalloc_noXX_save/restore() calls to control
145 * memory allocations done within this function.
146 */
147int blkdev_report_zones(struct block_device *bdev, sector_t sector,
148 unsigned int nr_zones, report_zones_cb cb, void *data)
149{
150 struct gendisk *disk = bdev->bd_disk;
151 sector_t capacity = get_capacity(disk);
152
153 if (!blk_queue_is_zoned(bdev_get_queue(bdev)) ||
154 WARN_ON_ONCE(!disk->fops->report_zones))
155 return -EOPNOTSUPP;
156
157 if (!nr_zones || sector >= capacity)
158 return 0;
159
160 return disk->fops->report_zones(disk, sector, nr_zones, cb, data);
161}
162EXPORT_SYMBOL_GPL(blkdev_report_zones);
163
164static inline unsigned long *blk_alloc_zone_bitmap(int node,
165 unsigned int nr_zones)
166{
167 return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long),
168 GFP_NOIO, node);
169}
170
171static int blk_zone_need_reset_cb(struct blk_zone *zone, unsigned int idx,
172 void *data)
173{
174 /*
175 * For an all-zones reset, ignore conventional, empty, read-only
176 * and offline zones.
177 */
178 switch (zone->cond) {
179 case BLK_ZONE_COND_NOT_WP:
180 case BLK_ZONE_COND_EMPTY:
181 case BLK_ZONE_COND_READONLY:
182 case BLK_ZONE_COND_OFFLINE:
183 return 0;
184 default:
185 set_bit(idx, (unsigned long *)data);
186 return 0;
187 }
188}
189
190static int blkdev_zone_reset_all_emulated(struct block_device *bdev,
191 gfp_t gfp_mask)
192{
193 struct request_queue *q = bdev_get_queue(bdev);
194 sector_t capacity = get_capacity(bdev->bd_disk);
195 sector_t zone_sectors = blk_queue_zone_sectors(q);
196 unsigned long *need_reset;
197 struct bio *bio = NULL;
198 sector_t sector = 0;
199 int ret;
200
201 need_reset = blk_alloc_zone_bitmap(q->node, q->nr_zones);
202 if (!need_reset)
203 return -ENOMEM;
204
205 ret = bdev->bd_disk->fops->report_zones(bdev->bd_disk, 0,
206 q->nr_zones, blk_zone_need_reset_cb,
207 need_reset);
208 if (ret < 0)
209 goto out_free_need_reset;
210
211 ret = 0;
212 while (sector < capacity) {
213 if (!test_bit(blk_queue_zone_no(q, sector), need_reset)) {
214 sector += zone_sectors;
215 continue;
216 }
217
218 bio = blk_next_bio(bio, 0, gfp_mask);
219 bio_set_dev(bio, bdev);
220 bio->bi_opf = REQ_OP_ZONE_RESET | REQ_SYNC;
221 bio->bi_iter.bi_sector = sector;
222 sector += zone_sectors;
223
224 /* This may take a while, so be nice to others */
225 cond_resched();
226 }
227
228 if (bio) {
229 ret = submit_bio_wait(bio);
230 bio_put(bio);
231 }
232
233out_free_need_reset:
234 kfree(need_reset);
235 return ret;
236}
237
238static int blkdev_zone_reset_all(struct block_device *bdev, gfp_t gfp_mask)
239{
240 struct bio bio;
241
242 bio_init(&bio, NULL, 0);
243 bio_set_dev(&bio, bdev);
244 bio.bi_opf = REQ_OP_ZONE_RESET_ALL | REQ_SYNC;
245
246 return submit_bio_wait(&bio);
247}
248
249/**
250 * blkdev_zone_mgmt - Execute a zone management operation on a range of zones
251 * @bdev: Target block device
252 * @op: Operation to be performed on the zones
253 * @sector: Start sector of the first zone to operate on
254 * @nr_sectors: Number of sectors, should be at least the length of one zone and
255 * must be zone size aligned.
256 * @gfp_mask: Memory allocation flags (for bio_alloc)
257 *
258 * Description:
259 * Perform the specified operation on the range of zones specified by
260 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
261 * is valid, but the specified range should not contain conventional zones.
262 * The operation to execute on each zone can be a zone reset, open, close
263 * or finish request.
264 */
265int blkdev_zone_mgmt(struct block_device *bdev, enum req_opf op,
266 sector_t sector, sector_t nr_sectors,
267 gfp_t gfp_mask)
268{
269 struct request_queue *q = bdev_get_queue(bdev);
270 sector_t zone_sectors = blk_queue_zone_sectors(q);
271 sector_t capacity = get_capacity(bdev->bd_disk);
272 sector_t end_sector = sector + nr_sectors;
273 struct bio *bio = NULL;
274 int ret = 0;
275
276 if (!blk_queue_is_zoned(q))
277 return -EOPNOTSUPP;
278
279 if (bdev_read_only(bdev))
280 return -EPERM;
281
282 if (!op_is_zone_mgmt(op))
283 return -EOPNOTSUPP;
284
285 if (end_sector <= sector || end_sector > capacity)
286 /* Out of range */
287 return -EINVAL;
288
289 /* Check alignment (handle eventual smaller last zone) */
290 if (sector & (zone_sectors - 1))
291 return -EINVAL;
292
293 if ((nr_sectors & (zone_sectors - 1)) && end_sector != capacity)
294 return -EINVAL;
295
296 /*
297 * In the case of a zone reset operation over all zones,
298 * REQ_OP_ZONE_RESET_ALL can be used with devices supporting this
299 * command. For other devices, we emulate this command behavior by
300 * identifying the zones needing a reset.
301 */
302 if (op == REQ_OP_ZONE_RESET && sector == 0 && nr_sectors == capacity) {
303 if (!blk_queue_zone_resetall(q))
304 return blkdev_zone_reset_all_emulated(bdev, gfp_mask);
305 return blkdev_zone_reset_all(bdev, gfp_mask);
306 }
307
308 while (sector < end_sector) {
309 bio = blk_next_bio(bio, 0, gfp_mask);
310 bio_set_dev(bio, bdev);
311 bio->bi_opf = op | REQ_SYNC;
312 bio->bi_iter.bi_sector = sector;
313 sector += zone_sectors;
314
315 /* This may take a while, so be nice to others */
316 cond_resched();
317 }
318
319 ret = submit_bio_wait(bio);
320 bio_put(bio);
321
322 return ret;
323}
324EXPORT_SYMBOL_GPL(blkdev_zone_mgmt);
325
326struct zone_report_args {
327 struct blk_zone __user *zones;
328};
329
330static int blkdev_copy_zone_to_user(struct blk_zone *zone, unsigned int idx,
331 void *data)
332{
333 struct zone_report_args *args = data;
334
335 if (copy_to_user(&args->zones[idx], zone, sizeof(struct blk_zone)))
336 return -EFAULT;
337 return 0;
338}
339
340/*
341 * BLKREPORTZONE ioctl processing.
342 * Called from blkdev_ioctl.
343 */
344int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
345 unsigned int cmd, unsigned long arg)
346{
347 void __user *argp = (void __user *)arg;
348 struct zone_report_args args;
349 struct request_queue *q;
350 struct blk_zone_report rep;
351 int ret;
352
353 if (!argp)
354 return -EINVAL;
355
356 q = bdev_get_queue(bdev);
357 if (!q)
358 return -ENXIO;
359
360 if (!blk_queue_is_zoned(q))
361 return -ENOTTY;
362
363 if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report)))
364 return -EFAULT;
365
366 if (!rep.nr_zones)
367 return -EINVAL;
368
369 args.zones = argp + sizeof(struct blk_zone_report);
370 ret = blkdev_report_zones(bdev, rep.sector, rep.nr_zones,
371 blkdev_copy_zone_to_user, &args);
372 if (ret < 0)
373 return ret;
374
375 rep.nr_zones = ret;
376 rep.flags = BLK_ZONE_REP_CAPACITY;
377 if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report)))
378 return -EFAULT;
379 return 0;
380}
381
382static int blkdev_truncate_zone_range(struct block_device *bdev, fmode_t mode,
383 const struct blk_zone_range *zrange)
384{
385 loff_t start, end;
386
387 if (zrange->sector + zrange->nr_sectors <= zrange->sector ||
388 zrange->sector + zrange->nr_sectors > get_capacity(bdev->bd_disk))
389 /* Out of range */
390 return -EINVAL;
391
392 start = zrange->sector << SECTOR_SHIFT;
393 end = ((zrange->sector + zrange->nr_sectors) << SECTOR_SHIFT) - 1;
394
395 return truncate_bdev_range(bdev, mode, start, end);
396}
397
398/*
399 * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
400 * Called from blkdev_ioctl.
401 */
402int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
403 unsigned int cmd, unsigned long arg)
404{
405 void __user *argp = (void __user *)arg;
406 struct request_queue *q;
407 struct blk_zone_range zrange;
408 enum req_opf op;
409 int ret;
410
411 if (!argp)
412 return -EINVAL;
413
414 q = bdev_get_queue(bdev);
415 if (!q)
416 return -ENXIO;
417
418 if (!blk_queue_is_zoned(q))
419 return -ENOTTY;
420
421 if (!(mode & FMODE_WRITE))
422 return -EBADF;
423
424 if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range)))
425 return -EFAULT;
426
427 switch (cmd) {
428 case BLKRESETZONE:
429 op = REQ_OP_ZONE_RESET;
430
431 /* Invalidate the page cache, including dirty pages. */
432 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
433 if (ret)
434 return ret;
435 break;
436 case BLKOPENZONE:
437 op = REQ_OP_ZONE_OPEN;
438 break;
439 case BLKCLOSEZONE:
440 op = REQ_OP_ZONE_CLOSE;
441 break;
442 case BLKFINISHZONE:
443 op = REQ_OP_ZONE_FINISH;
444 break;
445 default:
446 return -ENOTTY;
447 }
448
449 ret = blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors,
450 GFP_KERNEL);
451
452 /*
453 * Invalidate the page cache again for zone reset: writes can only be
454 * direct for zoned devices so concurrent writes would not add any page
455 * to the page cache after/during reset. The page cache may be filled
456 * again due to concurrent reads though and dropping the pages for
457 * these is fine.
458 */
459 if (!ret && cmd == BLKRESETZONE)
460 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
461
462 return ret;
463}
464
465void blk_queue_free_zone_bitmaps(struct request_queue *q)
466{
467 kfree(q->conv_zones_bitmap);
468 q->conv_zones_bitmap = NULL;
469 kfree(q->seq_zones_wlock);
470 q->seq_zones_wlock = NULL;
471}
472
473struct blk_revalidate_zone_args {
474 struct gendisk *disk;
475 unsigned long *conv_zones_bitmap;
476 unsigned long *seq_zones_wlock;
477 unsigned int nr_zones;
478 sector_t zone_sectors;
479 sector_t sector;
480};
481
482/*
483 * Helper function to check the validity of zones of a zoned block device.
484 */
485static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx,
486 void *data)
487{
488 struct blk_revalidate_zone_args *args = data;
489 struct gendisk *disk = args->disk;
490 struct request_queue *q = disk->queue;
491 sector_t capacity = get_capacity(disk);
492
493 /*
494 * All zones must have the same size, with the exception on an eventual
495 * smaller last zone.
496 */
497 if (zone->start == 0) {
498 if (zone->len == 0 || !is_power_of_2(zone->len)) {
499 pr_warn("%s: Invalid zoned device with non power of two zone size (%llu)\n",
500 disk->disk_name, zone->len);
501 return -ENODEV;
502 }
503
504 args->zone_sectors = zone->len;
505 args->nr_zones = (capacity + zone->len - 1) >> ilog2(zone->len);
506 } else if (zone->start + args->zone_sectors < capacity) {
507 if (zone->len != args->zone_sectors) {
508 pr_warn("%s: Invalid zoned device with non constant zone size\n",
509 disk->disk_name);
510 return -ENODEV;
511 }
512 } else {
513 if (zone->len > args->zone_sectors) {
514 pr_warn("%s: Invalid zoned device with larger last zone size\n",
515 disk->disk_name);
516 return -ENODEV;
517 }
518 }
519
520 /* Check for holes in the zone report */
521 if (zone->start != args->sector) {
522 pr_warn("%s: Zone gap at sectors %llu..%llu\n",
523 disk->disk_name, args->sector, zone->start);
524 return -ENODEV;
525 }
526
527 /* Check zone type */
528 switch (zone->type) {
529 case BLK_ZONE_TYPE_CONVENTIONAL:
530 if (!args->conv_zones_bitmap) {
531 args->conv_zones_bitmap =
532 blk_alloc_zone_bitmap(q->node, args->nr_zones);
533 if (!args->conv_zones_bitmap)
534 return -ENOMEM;
535 }
536 set_bit(idx, args->conv_zones_bitmap);
537 break;
538 case BLK_ZONE_TYPE_SEQWRITE_REQ:
539 case BLK_ZONE_TYPE_SEQWRITE_PREF:
540 if (!args->seq_zones_wlock) {
541 args->seq_zones_wlock =
542 blk_alloc_zone_bitmap(q->node, args->nr_zones);
543 if (!args->seq_zones_wlock)
544 return -ENOMEM;
545 }
546 break;
547 default:
548 pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n",
549 disk->disk_name, (int)zone->type, zone->start);
550 return -ENODEV;
551 }
552
553 args->sector += zone->len;
554 return 0;
555}
556
557/**
558 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
559 * @disk: Target disk
560 * @update_driver_data: Callback to update driver data on the frozen disk
561 *
562 * Helper function for low-level device drivers to (re) allocate and initialize
563 * a disk request queue zone bitmaps. This functions should normally be called
564 * within the disk ->revalidate method for blk-mq based drivers. For BIO based
565 * drivers only q->nr_zones needs to be updated so that the sysfs exposed value
566 * is correct.
567 * If the @update_driver_data callback function is not NULL, the callback is
568 * executed with the device request queue frozen after all zones have been
569 * checked.
570 */
571int blk_revalidate_disk_zones(struct gendisk *disk,
572 void (*update_driver_data)(struct gendisk *disk))
573{
574 struct request_queue *q = disk->queue;
575 struct blk_revalidate_zone_args args = {
576 .disk = disk,
577 };
578 unsigned int noio_flag;
579 int ret;
580
581 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
582 return -EIO;
583 if (WARN_ON_ONCE(!queue_is_mq(q)))
584 return -EIO;
585
586 if (!get_capacity(disk))
587 return -EIO;
588
589 /*
590 * Ensure that all memory allocations in this context are done as if
591 * GFP_NOIO was specified.
592 */
593 noio_flag = memalloc_noio_save();
594 ret = disk->fops->report_zones(disk, 0, UINT_MAX,
595 blk_revalidate_zone_cb, &args);
596 if (!ret) {
597 pr_warn("%s: No zones reported\n", disk->disk_name);
598 ret = -ENODEV;
599 }
600 memalloc_noio_restore(noio_flag);
601
602 /*
603 * If zones where reported, make sure that the entire disk capacity
604 * has been checked.
605 */
606 if (ret > 0 && args.sector != get_capacity(disk)) {
607 pr_warn("%s: Missing zones from sector %llu\n",
608 disk->disk_name, args.sector);
609 ret = -ENODEV;
610 }
611
612 /*
613 * Install the new bitmaps and update nr_zones only once the queue is
614 * stopped and all I/Os are completed (i.e. a scheduler is not
615 * referencing the bitmaps).
616 */
617 blk_mq_freeze_queue(q);
618 if (ret > 0) {
619 blk_queue_chunk_sectors(q, args.zone_sectors);
620 q->nr_zones = args.nr_zones;
621 swap(q->seq_zones_wlock, args.seq_zones_wlock);
622 swap(q->conv_zones_bitmap, args.conv_zones_bitmap);
623 if (update_driver_data)
624 update_driver_data(disk);
625 ret = 0;
626 } else {
627 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
628 blk_queue_free_zone_bitmaps(q);
629 }
630 blk_mq_unfreeze_queue(q);
631
632 kfree(args.seq_zones_wlock);
633 kfree(args.conv_zones_bitmap);
634 return ret;
635}
636EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);
637
638void blk_queue_clear_zone_settings(struct request_queue *q)
639{
640 blk_mq_freeze_queue(q);
641
642 blk_queue_free_zone_bitmaps(q);
643 blk_queue_flag_clear(QUEUE_FLAG_ZONE_RESETALL, q);
644 q->required_elevator_features &= ~ELEVATOR_F_ZBD_SEQ_WRITE;
645 q->nr_zones = 0;
646 q->max_open_zones = 0;
647 q->max_active_zones = 0;
648 q->limits.chunk_sectors = 0;
649 q->limits.zone_write_granularity = 0;
650 q->limits.max_zone_append_sectors = 0;
651
652 blk_mq_unfreeze_queue(q);
653}