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
24 static 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 */
blk_zone_cond_str(enum blk_zone_cond zone_cond)44 const 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 }
53 EXPORT_SYMBOL_GPL(blk_zone_cond_str);
54
blk_zone_start(struct request_queue * q,sector_t sector)55 static inline sector_t blk_zone_start(struct request_queue *q,
56 sector_t sector)
57 {
58 sector_t zone_mask = blk_queue_zone_sectors(q) - 1;
59
60 return sector & ~zone_mask;
61 }
62
63 /*
64 * Return true if a request is a write requests that needs zone write locking.
65 */
blk_req_needs_zone_write_lock(struct request * rq)66 bool blk_req_needs_zone_write_lock(struct request *rq)
67 {
68 if (!rq->q->seq_zones_wlock)
69 return false;
70
71 if (blk_rq_is_passthrough(rq))
72 return false;
73
74 switch (req_op(rq)) {
75 case REQ_OP_WRITE_ZEROES:
76 case REQ_OP_WRITE_SAME:
77 case REQ_OP_WRITE:
78 return blk_rq_zone_is_seq(rq);
79 default:
80 return false;
81 }
82 }
83 EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
84
blk_req_zone_write_trylock(struct request * rq)85 bool blk_req_zone_write_trylock(struct request *rq)
86 {
87 unsigned int zno = blk_rq_zone_no(rq);
88
89 if (test_and_set_bit(zno, rq->q->seq_zones_wlock))
90 return false;
91
92 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
93 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
94
95 return true;
96 }
97 EXPORT_SYMBOL_GPL(blk_req_zone_write_trylock);
98
__blk_req_zone_write_lock(struct request * rq)99 void __blk_req_zone_write_lock(struct request *rq)
100 {
101 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
102 rq->q->seq_zones_wlock)))
103 return;
104
105 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
106 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
107 }
108 EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
109
__blk_req_zone_write_unlock(struct request * rq)110 void __blk_req_zone_write_unlock(struct request *rq)
111 {
112 rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED;
113 if (rq->q->seq_zones_wlock)
114 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq),
115 rq->q->seq_zones_wlock));
116 }
117 EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
118
119 /**
120 * blkdev_nr_zones - Get number of zones
121 * @disk: Target gendisk
122 *
123 * Return the total number of zones of a zoned block device. For a block
124 * device without zone capabilities, the number of zones is always 0.
125 */
blkdev_nr_zones(struct gendisk * disk)126 unsigned int blkdev_nr_zones(struct gendisk *disk)
127 {
128 sector_t zone_sectors = blk_queue_zone_sectors(disk->queue);
129
130 if (!blk_queue_is_zoned(disk->queue))
131 return 0;
132 return (get_capacity(disk) + zone_sectors - 1) >> ilog2(zone_sectors);
133 }
134 EXPORT_SYMBOL_GPL(blkdev_nr_zones);
135
136 /**
137 * blkdev_report_zones - Get zones information
138 * @bdev: Target block device
139 * @sector: Sector from which to report zones
140 * @nr_zones: Maximum number of zones to report
141 * @cb: Callback function called for each reported zone
142 * @data: Private data for the callback
143 *
144 * Description:
145 * Get zone information starting from the zone containing @sector for at most
146 * @nr_zones, and call @cb for each zone reported by the device.
147 * To report all zones in a device starting from @sector, the BLK_ALL_ZONES
148 * constant can be passed to @nr_zones.
149 * Returns the number of zones reported by the device, or a negative errno
150 * value in case of failure.
151 *
152 * Note: The caller must use memalloc_noXX_save/restore() calls to control
153 * memory allocations done within this function.
154 */
blkdev_report_zones(struct block_device * bdev,sector_t sector,unsigned int nr_zones,report_zones_cb cb,void * data)155 int blkdev_report_zones(struct block_device *bdev, sector_t sector,
156 unsigned int nr_zones, report_zones_cb cb, void *data)
157 {
158 struct gendisk *disk = bdev->bd_disk;
159 sector_t capacity = get_capacity(disk);
160
161 if (!blk_queue_is_zoned(bdev_get_queue(bdev)) ||
162 WARN_ON_ONCE(!disk->fops->report_zones))
163 return -EOPNOTSUPP;
164
165 if (!nr_zones || sector >= capacity)
166 return 0;
167
168 return disk->fops->report_zones(disk, sector, nr_zones, cb, data);
169 }
170 EXPORT_SYMBOL_GPL(blkdev_report_zones);
171
blkdev_allow_reset_all_zones(struct block_device * bdev,sector_t sector,sector_t nr_sectors)172 static inline bool blkdev_allow_reset_all_zones(struct block_device *bdev,
173 sector_t sector,
174 sector_t nr_sectors)
175 {
176 if (!blk_queue_zone_resetall(bdev_get_queue(bdev)))
177 return false;
178
179 /*
180 * REQ_OP_ZONE_RESET_ALL can be executed only if the number of sectors
181 * of the applicable zone range is the entire disk.
182 */
183 return !sector && nr_sectors == get_capacity(bdev->bd_disk);
184 }
185
186 /**
187 * blkdev_zone_mgmt - Execute a zone management operation on a range of zones
188 * @bdev: Target block device
189 * @op: Operation to be performed on the zones
190 * @sector: Start sector of the first zone to operate on
191 * @nr_sectors: Number of sectors, should be at least the length of one zone and
192 * must be zone size aligned.
193 * @gfp_mask: Memory allocation flags (for bio_alloc)
194 *
195 * Description:
196 * Perform the specified operation on the range of zones specified by
197 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
198 * is valid, but the specified range should not contain conventional zones.
199 * The operation to execute on each zone can be a zone reset, open, close
200 * or finish request.
201 */
blkdev_zone_mgmt(struct block_device * bdev,enum req_opf op,sector_t sector,sector_t nr_sectors,gfp_t gfp_mask)202 int blkdev_zone_mgmt(struct block_device *bdev, enum req_opf op,
203 sector_t sector, sector_t nr_sectors,
204 gfp_t gfp_mask)
205 {
206 struct request_queue *q = bdev_get_queue(bdev);
207 sector_t zone_sectors = blk_queue_zone_sectors(q);
208 sector_t capacity = get_capacity(bdev->bd_disk);
209 sector_t end_sector = sector + nr_sectors;
210 struct bio *bio = NULL;
211 int ret;
212
213 if (!blk_queue_is_zoned(q))
214 return -EOPNOTSUPP;
215
216 if (bdev_read_only(bdev))
217 return -EPERM;
218
219 if (!op_is_zone_mgmt(op))
220 return -EOPNOTSUPP;
221
222 if (end_sector <= sector || end_sector > capacity)
223 /* Out of range */
224 return -EINVAL;
225
226 /* Check alignment (handle eventual smaller last zone) */
227 if (sector & (zone_sectors - 1))
228 return -EINVAL;
229
230 if ((nr_sectors & (zone_sectors - 1)) && end_sector != capacity)
231 return -EINVAL;
232
233 while (sector < end_sector) {
234 bio = blk_next_bio(bio, 0, gfp_mask);
235 bio_set_dev(bio, bdev);
236
237 /*
238 * Special case for the zone reset operation that reset all
239 * zones, this is useful for applications like mkfs.
240 */
241 if (op == REQ_OP_ZONE_RESET &&
242 blkdev_allow_reset_all_zones(bdev, sector, nr_sectors)) {
243 bio->bi_opf = REQ_OP_ZONE_RESET_ALL | REQ_SYNC;
244 break;
245 }
246
247 bio->bi_opf = op | REQ_SYNC;
248 bio->bi_iter.bi_sector = sector;
249 sector += zone_sectors;
250
251 /* This may take a while, so be nice to others */
252 cond_resched();
253 }
254
255 ret = submit_bio_wait(bio);
256 bio_put(bio);
257
258 return ret;
259 }
260 EXPORT_SYMBOL_GPL(blkdev_zone_mgmt);
261
262 struct zone_report_args {
263 struct blk_zone __user *zones;
264 };
265
blkdev_copy_zone_to_user(struct blk_zone * zone,unsigned int idx,void * data)266 static int blkdev_copy_zone_to_user(struct blk_zone *zone, unsigned int idx,
267 void *data)
268 {
269 struct zone_report_args *args = data;
270
271 if (copy_to_user(&args->zones[idx], zone, sizeof(struct blk_zone)))
272 return -EFAULT;
273 return 0;
274 }
275
276 /*
277 * BLKREPORTZONE ioctl processing.
278 * Called from blkdev_ioctl.
279 */
blkdev_report_zones_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)280 int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
281 unsigned int cmd, unsigned long arg)
282 {
283 void __user *argp = (void __user *)arg;
284 struct zone_report_args args;
285 struct request_queue *q;
286 struct blk_zone_report rep;
287 int ret;
288
289 if (!argp)
290 return -EINVAL;
291
292 q = bdev_get_queue(bdev);
293 if (!q)
294 return -ENXIO;
295
296 if (!blk_queue_is_zoned(q))
297 return -ENOTTY;
298
299 if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report)))
300 return -EFAULT;
301
302 if (!rep.nr_zones)
303 return -EINVAL;
304
305 args.zones = argp + sizeof(struct blk_zone_report);
306 ret = blkdev_report_zones(bdev, rep.sector, rep.nr_zones,
307 blkdev_copy_zone_to_user, &args);
308 if (ret < 0)
309 return ret;
310
311 rep.nr_zones = ret;
312 rep.flags = BLK_ZONE_REP_CAPACITY;
313 if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report)))
314 return -EFAULT;
315 return 0;
316 }
317
blkdev_truncate_zone_range(struct block_device * bdev,fmode_t mode,const struct blk_zone_range * zrange)318 static int blkdev_truncate_zone_range(struct block_device *bdev, fmode_t mode,
319 const struct blk_zone_range *zrange)
320 {
321 loff_t start, end;
322
323 if (zrange->sector + zrange->nr_sectors <= zrange->sector ||
324 zrange->sector + zrange->nr_sectors > get_capacity(bdev->bd_disk))
325 /* Out of range */
326 return -EINVAL;
327
328 start = zrange->sector << SECTOR_SHIFT;
329 end = ((zrange->sector + zrange->nr_sectors) << SECTOR_SHIFT) - 1;
330
331 return truncate_bdev_range(bdev, mode, start, end);
332 }
333
334 /*
335 * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
336 * Called from blkdev_ioctl.
337 */
blkdev_zone_mgmt_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)338 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
339 unsigned int cmd, unsigned long arg)
340 {
341 void __user *argp = (void __user *)arg;
342 struct request_queue *q;
343 struct blk_zone_range zrange;
344 enum req_opf op;
345 int ret;
346
347 if (!argp)
348 return -EINVAL;
349
350 q = bdev_get_queue(bdev);
351 if (!q)
352 return -ENXIO;
353
354 if (!blk_queue_is_zoned(q))
355 return -ENOTTY;
356
357 if (!(mode & FMODE_WRITE))
358 return -EBADF;
359
360 if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range)))
361 return -EFAULT;
362
363 switch (cmd) {
364 case BLKRESETZONE:
365 op = REQ_OP_ZONE_RESET;
366
367 /* Invalidate the page cache, including dirty pages. */
368 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
369 if (ret)
370 return ret;
371 break;
372 case BLKOPENZONE:
373 op = REQ_OP_ZONE_OPEN;
374 break;
375 case BLKCLOSEZONE:
376 op = REQ_OP_ZONE_CLOSE;
377 break;
378 case BLKFINISHZONE:
379 op = REQ_OP_ZONE_FINISH;
380 break;
381 default:
382 return -ENOTTY;
383 }
384
385 ret = blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors,
386 GFP_KERNEL);
387
388 /*
389 * Invalidate the page cache again for zone reset: writes can only be
390 * direct for zoned devices so concurrent writes would not add any page
391 * to the page cache after/during reset. The page cache may be filled
392 * again due to concurrent reads though and dropping the pages for
393 * these is fine.
394 */
395 if (!ret && cmd == BLKRESETZONE)
396 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
397
398 return ret;
399 }
400
blk_alloc_zone_bitmap(int node,unsigned int nr_zones)401 static inline unsigned long *blk_alloc_zone_bitmap(int node,
402 unsigned int nr_zones)
403 {
404 return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long),
405 GFP_NOIO, node);
406 }
407
blk_queue_free_zone_bitmaps(struct request_queue * q)408 void blk_queue_free_zone_bitmaps(struct request_queue *q)
409 {
410 kfree(q->conv_zones_bitmap);
411 q->conv_zones_bitmap = NULL;
412 kfree(q->seq_zones_wlock);
413 q->seq_zones_wlock = NULL;
414 }
415
416 struct blk_revalidate_zone_args {
417 struct gendisk *disk;
418 unsigned long *conv_zones_bitmap;
419 unsigned long *seq_zones_wlock;
420 unsigned int nr_zones;
421 sector_t zone_sectors;
422 sector_t sector;
423 };
424
425 /*
426 * Helper function to check the validity of zones of a zoned block device.
427 */
blk_revalidate_zone_cb(struct blk_zone * zone,unsigned int idx,void * data)428 static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx,
429 void *data)
430 {
431 struct blk_revalidate_zone_args *args = data;
432 struct gendisk *disk = args->disk;
433 struct request_queue *q = disk->queue;
434 sector_t capacity = get_capacity(disk);
435
436 /*
437 * All zones must have the same size, with the exception on an eventual
438 * smaller last zone.
439 */
440 if (zone->start == 0) {
441 if (zone->len == 0 || !is_power_of_2(zone->len)) {
442 pr_warn("%s: Invalid zoned device with non power of two zone size (%llu)\n",
443 disk->disk_name, zone->len);
444 return -ENODEV;
445 }
446
447 args->zone_sectors = zone->len;
448 args->nr_zones = (capacity + zone->len - 1) >> ilog2(zone->len);
449 } else if (zone->start + args->zone_sectors < capacity) {
450 if (zone->len != args->zone_sectors) {
451 pr_warn("%s: Invalid zoned device with non constant zone size\n",
452 disk->disk_name);
453 return -ENODEV;
454 }
455 } else {
456 if (zone->len > args->zone_sectors) {
457 pr_warn("%s: Invalid zoned device with larger last zone size\n",
458 disk->disk_name);
459 return -ENODEV;
460 }
461 }
462
463 /* Check for holes in the zone report */
464 if (zone->start != args->sector) {
465 pr_warn("%s: Zone gap at sectors %llu..%llu\n",
466 disk->disk_name, args->sector, zone->start);
467 return -ENODEV;
468 }
469
470 /* Check zone type */
471 switch (zone->type) {
472 case BLK_ZONE_TYPE_CONVENTIONAL:
473 if (!args->conv_zones_bitmap) {
474 args->conv_zones_bitmap =
475 blk_alloc_zone_bitmap(q->node, args->nr_zones);
476 if (!args->conv_zones_bitmap)
477 return -ENOMEM;
478 }
479 set_bit(idx, args->conv_zones_bitmap);
480 break;
481 case BLK_ZONE_TYPE_SEQWRITE_REQ:
482 case BLK_ZONE_TYPE_SEQWRITE_PREF:
483 if (!args->seq_zones_wlock) {
484 args->seq_zones_wlock =
485 blk_alloc_zone_bitmap(q->node, args->nr_zones);
486 if (!args->seq_zones_wlock)
487 return -ENOMEM;
488 }
489 break;
490 default:
491 pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n",
492 disk->disk_name, (int)zone->type, zone->start);
493 return -ENODEV;
494 }
495
496 args->sector += zone->len;
497 return 0;
498 }
499
500 /**
501 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
502 * @disk: Target disk
503 * @update_driver_data: Callback to update driver data on the frozen disk
504 *
505 * Helper function for low-level device drivers to (re) allocate and initialize
506 * a disk request queue zone bitmaps. This functions should normally be called
507 * within the disk ->revalidate method for blk-mq based drivers. For BIO based
508 * drivers only q->nr_zones needs to be updated so that the sysfs exposed value
509 * is correct.
510 * If the @update_driver_data callback function is not NULL, the callback is
511 * executed with the device request queue frozen after all zones have been
512 * checked.
513 */
blk_revalidate_disk_zones(struct gendisk * disk,void (* update_driver_data)(struct gendisk * disk))514 int blk_revalidate_disk_zones(struct gendisk *disk,
515 void (*update_driver_data)(struct gendisk *disk))
516 {
517 struct request_queue *q = disk->queue;
518 struct blk_revalidate_zone_args args = {
519 .disk = disk,
520 };
521 unsigned int noio_flag;
522 int ret;
523
524 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
525 return -EIO;
526 if (WARN_ON_ONCE(!queue_is_mq(q)))
527 return -EIO;
528
529 if (!get_capacity(disk))
530 return -EIO;
531
532 /*
533 * Ensure that all memory allocations in this context are done as if
534 * GFP_NOIO was specified.
535 */
536 noio_flag = memalloc_noio_save();
537 ret = disk->fops->report_zones(disk, 0, UINT_MAX,
538 blk_revalidate_zone_cb, &args);
539 memalloc_noio_restore(noio_flag);
540
541 /*
542 * Install the new bitmaps and update nr_zones only once the queue is
543 * stopped and all I/Os are completed (i.e. a scheduler is not
544 * referencing the bitmaps).
545 */
546 blk_mq_freeze_queue(q);
547 if (ret >= 0) {
548 blk_queue_chunk_sectors(q, args.zone_sectors);
549 q->nr_zones = args.nr_zones;
550 swap(q->seq_zones_wlock, args.seq_zones_wlock);
551 swap(q->conv_zones_bitmap, args.conv_zones_bitmap);
552 if (update_driver_data)
553 update_driver_data(disk);
554 ret = 0;
555 } else {
556 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
557 blk_queue_free_zone_bitmaps(q);
558 }
559 blk_mq_unfreeze_queue(q);
560
561 kfree(args.seq_zones_wlock);
562 kfree(args.conv_zones_bitmap);
563 return ret;
564 }
565 EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);
566