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 (blk_rq_is_passthrough(rq))
61 return false;
62
63 if (!rq->q->disk->seq_zones_wlock)
64 return false;
65
66 if (bdev_op_is_zoned_write(rq->q->disk->part0, req_op(rq)))
67 return blk_rq_zone_is_seq(rq);
68
69 return false;
70}
71EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
72
73bool blk_req_zone_write_trylock(struct request *rq)
74{
75 unsigned int zno = blk_rq_zone_no(rq);
76
77 if (test_and_set_bit(zno, rq->q->disk->seq_zones_wlock))
78 return false;
79
80 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
81 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
82
83 return true;
84}
85EXPORT_SYMBOL_GPL(blk_req_zone_write_trylock);
86
87void __blk_req_zone_write_lock(struct request *rq)
88{
89 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
90 rq->q->disk->seq_zones_wlock)))
91 return;
92
93 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
94 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
95}
96EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
97
98void __blk_req_zone_write_unlock(struct request *rq)
99{
100 rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED;
101 if (rq->q->disk->seq_zones_wlock)
102 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq),
103 rq->q->disk->seq_zones_wlock));
104}
105EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
106
107/**
108 * bdev_nr_zones - Get number of zones
109 * @bdev: Target device
110 *
111 * Return the total number of zones of a zoned block device. For a block
112 * device without zone capabilities, the number of zones is always 0.
113 */
114unsigned int bdev_nr_zones(struct block_device *bdev)
115{
116 sector_t zone_sectors = bdev_zone_sectors(bdev);
117
118 if (!bdev_is_zoned(bdev))
119 return 0;
120 return (bdev_nr_sectors(bdev) + zone_sectors - 1) >>
121 ilog2(zone_sectors);
122}
123EXPORT_SYMBOL_GPL(bdev_nr_zones);
124
125/**
126 * blkdev_report_zones - Get zones information
127 * @bdev: Target block device
128 * @sector: Sector from which to report zones
129 * @nr_zones: Maximum number of zones to report
130 * @cb: Callback function called for each reported zone
131 * @data: Private data for the callback
132 *
133 * Description:
134 * Get zone information starting from the zone containing @sector for at most
135 * @nr_zones, and call @cb for each zone reported by the device.
136 * To report all zones in a device starting from @sector, the BLK_ALL_ZONES
137 * constant can be passed to @nr_zones.
138 * Returns the number of zones reported by the device, or a negative errno
139 * value in case of failure.
140 *
141 * Note: The caller must use memalloc_noXX_save/restore() calls to control
142 * memory allocations done within this function.
143 */
144int blkdev_report_zones(struct block_device *bdev, sector_t sector,
145 unsigned int nr_zones, report_zones_cb cb, void *data)
146{
147 struct gendisk *disk = bdev->bd_disk;
148 sector_t capacity = get_capacity(disk);
149
150 if (!bdev_is_zoned(bdev) || WARN_ON_ONCE(!disk->fops->report_zones))
151 return -EOPNOTSUPP;
152
153 if (!nr_zones || sector >= capacity)
154 return 0;
155
156 return disk->fops->report_zones(disk, sector, nr_zones, cb, data);
157}
158EXPORT_SYMBOL_GPL(blkdev_report_zones);
159
160static inline unsigned long *blk_alloc_zone_bitmap(int node,
161 unsigned int nr_zones)
162{
163 return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long),
164 GFP_NOIO, node);
165}
166
167static int blk_zone_need_reset_cb(struct blk_zone *zone, unsigned int idx,
168 void *data)
169{
170 /*
171 * For an all-zones reset, ignore conventional, empty, read-only
172 * and offline zones.
173 */
174 switch (zone->cond) {
175 case BLK_ZONE_COND_NOT_WP:
176 case BLK_ZONE_COND_EMPTY:
177 case BLK_ZONE_COND_READONLY:
178 case BLK_ZONE_COND_OFFLINE:
179 return 0;
180 default:
181 set_bit(idx, (unsigned long *)data);
182 return 0;
183 }
184}
185
186static int blkdev_zone_reset_all_emulated(struct block_device *bdev,
187 gfp_t gfp_mask)
188{
189 struct gendisk *disk = bdev->bd_disk;
190 sector_t capacity = bdev_nr_sectors(bdev);
191 sector_t zone_sectors = bdev_zone_sectors(bdev);
192 unsigned long *need_reset;
193 struct bio *bio = NULL;
194 sector_t sector = 0;
195 int ret;
196
197 need_reset = blk_alloc_zone_bitmap(disk->queue->node, disk->nr_zones);
198 if (!need_reset)
199 return -ENOMEM;
200
201 ret = disk->fops->report_zones(disk, 0, disk->nr_zones,
202 blk_zone_need_reset_cb, need_reset);
203 if (ret < 0)
204 goto out_free_need_reset;
205
206 ret = 0;
207 while (sector < capacity) {
208 if (!test_bit(disk_zone_no(disk, sector), need_reset)) {
209 sector += zone_sectors;
210 continue;
211 }
212
213 bio = blk_next_bio(bio, bdev, 0, REQ_OP_ZONE_RESET | REQ_SYNC,
214 gfp_mask);
215 bio->bi_iter.bi_sector = sector;
216 sector += zone_sectors;
217
218 /* This may take a while, so be nice to others */
219 cond_resched();
220 }
221
222 if (bio) {
223 ret = submit_bio_wait(bio);
224 bio_put(bio);
225 }
226
227out_free_need_reset:
228 kfree(need_reset);
229 return ret;
230}
231
232static int blkdev_zone_reset_all(struct block_device *bdev, gfp_t gfp_mask)
233{
234 struct bio bio;
235
236 bio_init(&bio, bdev, NULL, 0, REQ_OP_ZONE_RESET_ALL | REQ_SYNC);
237 return submit_bio_wait(&bio);
238}
239
240/**
241 * blkdev_zone_mgmt - Execute a zone management operation on a range of zones
242 * @bdev: Target block device
243 * @op: Operation to be performed on the zones
244 * @sector: Start sector of the first zone to operate on
245 * @nr_sectors: Number of sectors, should be at least the length of one zone and
246 * must be zone size aligned.
247 * @gfp_mask: Memory allocation flags (for bio_alloc)
248 *
249 * Description:
250 * Perform the specified operation on the range of zones specified by
251 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
252 * is valid, but the specified range should not contain conventional zones.
253 * The operation to execute on each zone can be a zone reset, open, close
254 * or finish request.
255 */
256int blkdev_zone_mgmt(struct block_device *bdev, enum req_op op,
257 sector_t sector, sector_t nr_sectors, gfp_t gfp_mask)
258{
259 struct request_queue *q = bdev_get_queue(bdev);
260 sector_t zone_sectors = bdev_zone_sectors(bdev);
261 sector_t capacity = bdev_nr_sectors(bdev);
262 sector_t end_sector = sector + nr_sectors;
263 struct bio *bio = NULL;
264 int ret = 0;
265
266 if (!bdev_is_zoned(bdev))
267 return -EOPNOTSUPP;
268
269 if (bdev_read_only(bdev))
270 return -EPERM;
271
272 if (!op_is_zone_mgmt(op))
273 return -EOPNOTSUPP;
274
275 if (end_sector <= sector || end_sector > capacity)
276 /* Out of range */
277 return -EINVAL;
278
279 /* Check alignment (handle eventual smaller last zone) */
280 if (sector & (zone_sectors - 1))
281 return -EINVAL;
282
283 if ((nr_sectors & (zone_sectors - 1)) && end_sector != capacity)
284 return -EINVAL;
285
286 /*
287 * In the case of a zone reset operation over all zones,
288 * REQ_OP_ZONE_RESET_ALL can be used with devices supporting this
289 * command. For other devices, we emulate this command behavior by
290 * identifying the zones needing a reset.
291 */
292 if (op == REQ_OP_ZONE_RESET && sector == 0 && nr_sectors == capacity) {
293 if (!blk_queue_zone_resetall(q))
294 return blkdev_zone_reset_all_emulated(bdev, gfp_mask);
295 return blkdev_zone_reset_all(bdev, gfp_mask);
296 }
297
298 while (sector < end_sector) {
299 bio = blk_next_bio(bio, bdev, 0, op | REQ_SYNC, gfp_mask);
300 bio->bi_iter.bi_sector = sector;
301 sector += zone_sectors;
302
303 /* This may take a while, so be nice to others */
304 cond_resched();
305 }
306
307 ret = submit_bio_wait(bio);
308 bio_put(bio);
309
310 return ret;
311}
312EXPORT_SYMBOL_GPL(blkdev_zone_mgmt);
313
314struct zone_report_args {
315 struct blk_zone __user *zones;
316};
317
318static int blkdev_copy_zone_to_user(struct blk_zone *zone, unsigned int idx,
319 void *data)
320{
321 struct zone_report_args *args = data;
322
323 if (copy_to_user(&args->zones[idx], zone, sizeof(struct blk_zone)))
324 return -EFAULT;
325 return 0;
326}
327
328/*
329 * BLKREPORTZONE ioctl processing.
330 * Called from blkdev_ioctl.
331 */
332int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
333 unsigned int cmd, unsigned long arg)
334{
335 void __user *argp = (void __user *)arg;
336 struct zone_report_args args;
337 struct request_queue *q;
338 struct blk_zone_report rep;
339 int ret;
340
341 if (!argp)
342 return -EINVAL;
343
344 q = bdev_get_queue(bdev);
345 if (!q)
346 return -ENXIO;
347
348 if (!bdev_is_zoned(bdev))
349 return -ENOTTY;
350
351 if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report)))
352 return -EFAULT;
353
354 if (!rep.nr_zones)
355 return -EINVAL;
356
357 args.zones = argp + sizeof(struct blk_zone_report);
358 ret = blkdev_report_zones(bdev, rep.sector, rep.nr_zones,
359 blkdev_copy_zone_to_user, &args);
360 if (ret < 0)
361 return ret;
362
363 rep.nr_zones = ret;
364 rep.flags = BLK_ZONE_REP_CAPACITY;
365 if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report)))
366 return -EFAULT;
367 return 0;
368}
369
370static int blkdev_truncate_zone_range(struct block_device *bdev, fmode_t mode,
371 const struct blk_zone_range *zrange)
372{
373 loff_t start, end;
374
375 if (zrange->sector + zrange->nr_sectors <= zrange->sector ||
376 zrange->sector + zrange->nr_sectors > get_capacity(bdev->bd_disk))
377 /* Out of range */
378 return -EINVAL;
379
380 start = zrange->sector << SECTOR_SHIFT;
381 end = ((zrange->sector + zrange->nr_sectors) << SECTOR_SHIFT) - 1;
382
383 return truncate_bdev_range(bdev, mode, start, end);
384}
385
386/*
387 * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
388 * Called from blkdev_ioctl.
389 */
390int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
391 unsigned int cmd, unsigned long arg)
392{
393 void __user *argp = (void __user *)arg;
394 struct request_queue *q;
395 struct blk_zone_range zrange;
396 enum req_op op;
397 int ret;
398
399 if (!argp)
400 return -EINVAL;
401
402 q = bdev_get_queue(bdev);
403 if (!q)
404 return -ENXIO;
405
406 if (!bdev_is_zoned(bdev))
407 return -ENOTTY;
408
409 if (!(mode & FMODE_WRITE))
410 return -EBADF;
411
412 if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range)))
413 return -EFAULT;
414
415 switch (cmd) {
416 case BLKRESETZONE:
417 op = REQ_OP_ZONE_RESET;
418
419 /* Invalidate the page cache, including dirty pages. */
420 filemap_invalidate_lock(bdev->bd_inode->i_mapping);
421 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
422 if (ret)
423 goto fail;
424 break;
425 case BLKOPENZONE:
426 op = REQ_OP_ZONE_OPEN;
427 break;
428 case BLKCLOSEZONE:
429 op = REQ_OP_ZONE_CLOSE;
430 break;
431 case BLKFINISHZONE:
432 op = REQ_OP_ZONE_FINISH;
433 break;
434 default:
435 return -ENOTTY;
436 }
437
438 ret = blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors,
439 GFP_KERNEL);
440
441fail:
442 if (cmd == BLKRESETZONE)
443 filemap_invalidate_unlock(bdev->bd_inode->i_mapping);
444
445 return ret;
446}
447
448void disk_free_zone_bitmaps(struct gendisk *disk)
449{
450 kfree(disk->conv_zones_bitmap);
451 disk->conv_zones_bitmap = NULL;
452 kfree(disk->seq_zones_wlock);
453 disk->seq_zones_wlock = NULL;
454}
455
456struct blk_revalidate_zone_args {
457 struct gendisk *disk;
458 unsigned long *conv_zones_bitmap;
459 unsigned long *seq_zones_wlock;
460 unsigned int nr_zones;
461 sector_t zone_sectors;
462 sector_t sector;
463};
464
465/*
466 * Helper function to check the validity of zones of a zoned block device.
467 */
468static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx,
469 void *data)
470{
471 struct blk_revalidate_zone_args *args = data;
472 struct gendisk *disk = args->disk;
473 struct request_queue *q = disk->queue;
474 sector_t capacity = get_capacity(disk);
475
476 /*
477 * All zones must have the same size, with the exception on an eventual
478 * smaller last zone.
479 */
480 if (zone->start == 0) {
481 if (zone->len == 0 || !is_power_of_2(zone->len)) {
482 pr_warn("%s: Invalid zoned device with non power of two zone size (%llu)\n",
483 disk->disk_name, zone->len);
484 return -ENODEV;
485 }
486
487 args->zone_sectors = zone->len;
488 args->nr_zones = (capacity + zone->len - 1) >> ilog2(zone->len);
489 } else if (zone->start + args->zone_sectors < capacity) {
490 if (zone->len != args->zone_sectors) {
491 pr_warn("%s: Invalid zoned device with non constant zone size\n",
492 disk->disk_name);
493 return -ENODEV;
494 }
495 } else {
496 if (zone->len > args->zone_sectors) {
497 pr_warn("%s: Invalid zoned device with larger last zone size\n",
498 disk->disk_name);
499 return -ENODEV;
500 }
501 }
502
503 /* Check for holes in the zone report */
504 if (zone->start != args->sector) {
505 pr_warn("%s: Zone gap at sectors %llu..%llu\n",
506 disk->disk_name, args->sector, zone->start);
507 return -ENODEV;
508 }
509
510 /* Check zone type */
511 switch (zone->type) {
512 case BLK_ZONE_TYPE_CONVENTIONAL:
513 if (!args->conv_zones_bitmap) {
514 args->conv_zones_bitmap =
515 blk_alloc_zone_bitmap(q->node, args->nr_zones);
516 if (!args->conv_zones_bitmap)
517 return -ENOMEM;
518 }
519 set_bit(idx, args->conv_zones_bitmap);
520 break;
521 case BLK_ZONE_TYPE_SEQWRITE_REQ:
522 case BLK_ZONE_TYPE_SEQWRITE_PREF:
523 if (!args->seq_zones_wlock) {
524 args->seq_zones_wlock =
525 blk_alloc_zone_bitmap(q->node, args->nr_zones);
526 if (!args->seq_zones_wlock)
527 return -ENOMEM;
528 }
529 break;
530 default:
531 pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n",
532 disk->disk_name, (int)zone->type, zone->start);
533 return -ENODEV;
534 }
535
536 args->sector += zone->len;
537 return 0;
538}
539
540/**
541 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
542 * @disk: Target disk
543 * @update_driver_data: Callback to update driver data on the frozen disk
544 *
545 * Helper function for low-level device drivers to (re) allocate and initialize
546 * a disk request queue zone bitmaps. This functions should normally be called
547 * within the disk ->revalidate method for blk-mq based drivers. For BIO based
548 * drivers only q->nr_zones needs to be updated so that the sysfs exposed value
549 * is correct.
550 * If the @update_driver_data callback function is not NULL, the callback is
551 * executed with the device request queue frozen after all zones have been
552 * checked.
553 */
554int blk_revalidate_disk_zones(struct gendisk *disk,
555 void (*update_driver_data)(struct gendisk *disk))
556{
557 struct request_queue *q = disk->queue;
558 struct blk_revalidate_zone_args args = {
559 .disk = disk,
560 };
561 unsigned int noio_flag;
562 int ret;
563
564 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
565 return -EIO;
566 if (WARN_ON_ONCE(!queue_is_mq(q)))
567 return -EIO;
568
569 if (!get_capacity(disk))
570 return -EIO;
571
572 /*
573 * Ensure that all memory allocations in this context are done as if
574 * GFP_NOIO was specified.
575 */
576 noio_flag = memalloc_noio_save();
577 ret = disk->fops->report_zones(disk, 0, UINT_MAX,
578 blk_revalidate_zone_cb, &args);
579 if (!ret) {
580 pr_warn("%s: No zones reported\n", disk->disk_name);
581 ret = -ENODEV;
582 }
583 memalloc_noio_restore(noio_flag);
584
585 /*
586 * If zones where reported, make sure that the entire disk capacity
587 * has been checked.
588 */
589 if (ret > 0 && args.sector != get_capacity(disk)) {
590 pr_warn("%s: Missing zones from sector %llu\n",
591 disk->disk_name, args.sector);
592 ret = -ENODEV;
593 }
594
595 /*
596 * Install the new bitmaps and update nr_zones only once the queue is
597 * stopped and all I/Os are completed (i.e. a scheduler is not
598 * referencing the bitmaps).
599 */
600 blk_mq_freeze_queue(q);
601 if (ret > 0) {
602 blk_queue_chunk_sectors(q, args.zone_sectors);
603 disk->nr_zones = args.nr_zones;
604 swap(disk->seq_zones_wlock, args.seq_zones_wlock);
605 swap(disk->conv_zones_bitmap, args.conv_zones_bitmap);
606 if (update_driver_data)
607 update_driver_data(disk);
608 ret = 0;
609 } else {
610 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
611 disk_free_zone_bitmaps(disk);
612 }
613 blk_mq_unfreeze_queue(q);
614
615 kfree(args.seq_zones_wlock);
616 kfree(args.conv_zones_bitmap);
617 return ret;
618}
619EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);
620
621void disk_clear_zone_settings(struct gendisk *disk)
622{
623 struct request_queue *q = disk->queue;
624
625 blk_mq_freeze_queue(q);
626
627 disk_free_zone_bitmaps(disk);
628 blk_queue_flag_clear(QUEUE_FLAG_ZONE_RESETALL, q);
629 q->required_elevator_features &= ~ELEVATOR_F_ZBD_SEQ_WRITE;
630 disk->nr_zones = 0;
631 disk->max_open_zones = 0;
632 disk->max_active_zones = 0;
633 q->limits.chunk_sectors = 0;
634 q->limits.zone_write_granularity = 0;
635 q->limits.max_zone_append_sectors = 0;
636
637 blk_mq_unfreeze_queue(q);
638}