Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
6a0cb1bc HR |
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> | |
bf505456 | 16 | #include <linux/blk-mq.h> |
bd976e52 | 17 | #include <linux/sched/mm.h> |
6a0cb1bc | 18 | |
a2d6b3a2 DLM |
19 | #include "blk.h" |
20 | ||
6a0cb1bc HR |
21 | static inline sector_t blk_zone_start(struct request_queue *q, |
22 | sector_t sector) | |
23 | { | |
f99e8648 | 24 | sector_t zone_mask = blk_queue_zone_sectors(q) - 1; |
6a0cb1bc HR |
25 | |
26 | return sector & ~zone_mask; | |
27 | } | |
28 | ||
6cc77e9c CH |
29 | /* |
30 | * Return true if a request is a write requests that needs zone write locking. | |
31 | */ | |
32 | bool blk_req_needs_zone_write_lock(struct request *rq) | |
33 | { | |
34 | if (!rq->q->seq_zones_wlock) | |
35 | return false; | |
36 | ||
37 | if (blk_rq_is_passthrough(rq)) | |
38 | return false; | |
39 | ||
40 | switch (req_op(rq)) { | |
41 | case REQ_OP_WRITE_ZEROES: | |
42 | case REQ_OP_WRITE_SAME: | |
43 | case REQ_OP_WRITE: | |
44 | return blk_rq_zone_is_seq(rq); | |
45 | default: | |
46 | return false; | |
47 | } | |
48 | } | |
49 | EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock); | |
50 | ||
51 | void __blk_req_zone_write_lock(struct request *rq) | |
52 | { | |
53 | if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq), | |
54 | rq->q->seq_zones_wlock))) | |
55 | return; | |
56 | ||
57 | WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED); | |
58 | rq->rq_flags |= RQF_ZONE_WRITE_LOCKED; | |
59 | } | |
60 | EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock); | |
61 | ||
62 | void __blk_req_zone_write_unlock(struct request *rq) | |
63 | { | |
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)); | |
68 | } | |
69 | EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock); | |
70 | ||
a91e1380 DLM |
71 | static inline unsigned int __blkdev_nr_zones(struct request_queue *q, |
72 | sector_t nr_sectors) | |
73 | { | |
113ab72e | 74 | sector_t zone_sectors = blk_queue_zone_sectors(q); |
a91e1380 DLM |
75 | |
76 | return (nr_sectors + zone_sectors - 1) >> ilog2(zone_sectors); | |
77 | } | |
78 | ||
79 | /** | |
80 | * blkdev_nr_zones - Get number of zones | |
81 | * @bdev: Target block device | |
82 | * | |
83 | * Description: | |
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. | |
86 | */ | |
87 | unsigned int blkdev_nr_zones(struct block_device *bdev) | |
88 | { | |
89 | struct request_queue *q = bdev_get_queue(bdev); | |
90 | ||
91 | if (!blk_queue_is_zoned(q)) | |
92 | return 0; | |
93 | ||
94 | return __blkdev_nr_zones(q, bdev->bd_part->nr_sects); | |
95 | } | |
96 | EXPORT_SYMBOL_GPL(blkdev_nr_zones); | |
97 | ||
6a0cb1bc | 98 | /* |
e76239a3 CH |
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. | |
6a0cb1bc | 101 | */ |
e76239a3 | 102 | static bool blkdev_report_zone(struct block_device *bdev, struct blk_zone *rep) |
6a0cb1bc HR |
103 | { |
104 | sector_t offset = get_start_sect(bdev); | |
105 | ||
106 | if (rep->start < offset) | |
107 | return false; | |
108 | ||
109 | rep->start -= offset; | |
110 | if (rep->start + rep->len > bdev->bd_part->nr_sects) | |
111 | return false; | |
112 | ||
113 | if (rep->type == BLK_ZONE_TYPE_CONVENTIONAL) | |
114 | rep->wp = rep->start + rep->len; | |
115 | else | |
116 | rep->wp -= offset; | |
6a0cb1bc HR |
117 | return true; |
118 | } | |
119 | ||
e76239a3 | 120 | static int blk_report_zones(struct gendisk *disk, sector_t sector, |
bd976e52 | 121 | struct blk_zone *zones, unsigned int *nr_zones) |
e76239a3 CH |
122 | { |
123 | struct request_queue *q = disk->queue; | |
124 | unsigned int z = 0, n, nrz = *nr_zones; | |
125 | sector_t capacity = get_capacity(disk); | |
126 | int ret; | |
127 | ||
128 | while (z < nrz && sector < capacity) { | |
129 | n = nrz - z; | |
bd976e52 | 130 | ret = disk->fops->report_zones(disk, sector, &zones[z], &n); |
e76239a3 CH |
131 | if (ret) |
132 | return ret; | |
133 | if (!n) | |
134 | break; | |
135 | sector += blk_queue_zone_sectors(q) * n; | |
136 | z += n; | |
137 | } | |
138 | ||
139 | WARN_ON(z > *nr_zones); | |
140 | *nr_zones = z; | |
141 | ||
142 | return 0; | |
143 | } | |
144 | ||
6a0cb1bc HR |
145 | /** |
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 | |
6a0cb1bc HR |
151 | * |
152 | * Description: | |
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. | |
bd976e52 DLM |
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). | |
6a0cb1bc | 160 | */ |
e76239a3 | 161 | int blkdev_report_zones(struct block_device *bdev, sector_t sector, |
bd976e52 | 162 | struct blk_zone *zones, unsigned int *nr_zones) |
6a0cb1bc HR |
163 | { |
164 | struct request_queue *q = bdev_get_queue(bdev); | |
e76239a3 | 165 | unsigned int i, nrz; |
3c4da758 | 166 | int ret; |
6a0cb1bc | 167 | |
6a0cb1bc HR |
168 | if (!blk_queue_is_zoned(q)) |
169 | return -EOPNOTSUPP; | |
170 | ||
e76239a3 CH |
171 | /* |
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. | |
175 | */ | |
176 | if (WARN_ON_ONCE(!bdev->bd_disk->fops->report_zones)) | |
177 | return -EOPNOTSUPP; | |
6a0cb1bc | 178 | |
e76239a3 | 179 | if (!*nr_zones || sector >= bdev->bd_part->nr_sects) { |
6a0cb1bc HR |
180 | *nr_zones = 0; |
181 | return 0; | |
182 | } | |
183 | ||
e76239a3 CH |
184 | nrz = min(*nr_zones, |
185 | __blkdev_nr_zones(q, bdev->bd_part->nr_sects - sector)); | |
186 | ret = blk_report_zones(bdev->bd_disk, get_start_sect(bdev) + sector, | |
bd976e52 | 187 | zones, &nrz); |
6a0cb1bc | 188 | if (ret) |
e76239a3 | 189 | return ret; |
6a0cb1bc | 190 | |
e76239a3 CH |
191 | for (i = 0; i < nrz; i++) { |
192 | if (!blkdev_report_zone(bdev, zones)) | |
6a0cb1bc | 193 | break; |
e76239a3 | 194 | zones++; |
6a0cb1bc HR |
195 | } |
196 | ||
e76239a3 | 197 | *nr_zones = i; |
6a0cb1bc | 198 | |
e76239a3 | 199 | return 0; |
6a0cb1bc HR |
200 | } |
201 | EXPORT_SYMBOL_GPL(blkdev_report_zones); | |
202 | ||
203 | /** | |
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) | |
209 | * | |
210 | * Description: | |
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. | |
214 | */ | |
215 | int blkdev_reset_zones(struct block_device *bdev, | |
216 | sector_t sector, sector_t nr_sectors, | |
217 | gfp_t gfp_mask) | |
218 | { | |
219 | struct request_queue *q = bdev_get_queue(bdev); | |
220 | sector_t zone_sectors; | |
221 | sector_t end_sector = sector + nr_sectors; | |
a2d6b3a2 DLM |
222 | struct bio *bio = NULL; |
223 | struct blk_plug plug; | |
6a0cb1bc HR |
224 | int ret; |
225 | ||
6a0cb1bc HR |
226 | if (!blk_queue_is_zoned(q)) |
227 | return -EOPNOTSUPP; | |
228 | ||
a2d6b3a2 DLM |
229 | if (bdev_read_only(bdev)) |
230 | return -EPERM; | |
231 | ||
232 | if (!nr_sectors || end_sector > bdev->bd_part->nr_sects) | |
6a0cb1bc HR |
233 | /* Out of range */ |
234 | return -EINVAL; | |
235 | ||
236 | /* Check alignment (handle eventual smaller last zone) */ | |
f99e8648 | 237 | zone_sectors = blk_queue_zone_sectors(q); |
6a0cb1bc HR |
238 | if (sector & (zone_sectors - 1)) |
239 | return -EINVAL; | |
240 | ||
241 | if ((nr_sectors & (zone_sectors - 1)) && | |
242 | end_sector != bdev->bd_part->nr_sects) | |
243 | return -EINVAL; | |
244 | ||
a2d6b3a2 | 245 | blk_start_plug(&plug); |
6a0cb1bc HR |
246 | while (sector < end_sector) { |
247 | ||
a2d6b3a2 | 248 | bio = blk_next_bio(bio, 0, gfp_mask); |
6a0cb1bc | 249 | bio->bi_iter.bi_sector = sector; |
74d46992 | 250 | bio_set_dev(bio, bdev); |
6a0cb1bc HR |
251 | bio_set_op_attrs(bio, REQ_OP_ZONE_RESET, 0); |
252 | ||
6a0cb1bc HR |
253 | sector += zone_sectors; |
254 | ||
255 | /* This may take a while, so be nice to others */ | |
256 | cond_resched(); | |
257 | ||
258 | } | |
259 | ||
a2d6b3a2 DLM |
260 | ret = submit_bio_wait(bio); |
261 | bio_put(bio); | |
262 | ||
263 | blk_finish_plug(&plug); | |
264 | ||
265 | return ret; | |
6a0cb1bc HR |
266 | } |
267 | EXPORT_SYMBOL_GPL(blkdev_reset_zones); | |
3ed05a98 | 268 | |
56c4bddb | 269 | /* |
3ed05a98 ST |
270 | * BLKREPORTZONE ioctl processing. |
271 | * Called from blkdev_ioctl. | |
272 | */ | |
273 | int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode, | |
274 | unsigned int cmd, unsigned long arg) | |
275 | { | |
276 | void __user *argp = (void __user *)arg; | |
277 | struct request_queue *q; | |
278 | struct blk_zone_report rep; | |
279 | struct blk_zone *zones; | |
280 | int ret; | |
281 | ||
282 | if (!argp) | |
283 | return -EINVAL; | |
284 | ||
285 | q = bdev_get_queue(bdev); | |
286 | if (!q) | |
287 | return -ENXIO; | |
288 | ||
289 | if (!blk_queue_is_zoned(q)) | |
290 | return -ENOTTY; | |
291 | ||
292 | if (!capable(CAP_SYS_ADMIN)) | |
293 | return -EACCES; | |
294 | ||
295 | if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report))) | |
296 | return -EFAULT; | |
297 | ||
298 | if (!rep.nr_zones) | |
299 | return -EINVAL; | |
300 | ||
2e85fbaf | 301 | rep.nr_zones = min(blkdev_nr_zones(bdev), rep.nr_zones); |
327ea4ad | 302 | |
344476e1 KC |
303 | zones = kvmalloc_array(rep.nr_zones, sizeof(struct blk_zone), |
304 | GFP_KERNEL | __GFP_ZERO); | |
3ed05a98 ST |
305 | if (!zones) |
306 | return -ENOMEM; | |
307 | ||
bd976e52 | 308 | ret = blkdev_report_zones(bdev, rep.sector, zones, &rep.nr_zones); |
3ed05a98 ST |
309 | if (ret) |
310 | goto out; | |
311 | ||
312 | if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report))) { | |
313 | ret = -EFAULT; | |
314 | goto out; | |
315 | } | |
316 | ||
317 | if (rep.nr_zones) { | |
318 | if (copy_to_user(argp + sizeof(struct blk_zone_report), zones, | |
319 | sizeof(struct blk_zone) * rep.nr_zones)) | |
320 | ret = -EFAULT; | |
321 | } | |
322 | ||
323 | out: | |
327ea4ad | 324 | kvfree(zones); |
3ed05a98 ST |
325 | |
326 | return ret; | |
327 | } | |
328 | ||
56c4bddb | 329 | /* |
3ed05a98 ST |
330 | * BLKRESETZONE ioctl processing. |
331 | * Called from blkdev_ioctl. | |
332 | */ | |
333 | int blkdev_reset_zones_ioctl(struct block_device *bdev, fmode_t mode, | |
334 | unsigned int cmd, unsigned long arg) | |
335 | { | |
336 | void __user *argp = (void __user *)arg; | |
337 | struct request_queue *q; | |
338 | struct blk_zone_range zrange; | |
339 | ||
340 | if (!argp) | |
341 | return -EINVAL; | |
342 | ||
343 | q = bdev_get_queue(bdev); | |
344 | if (!q) | |
345 | return -ENXIO; | |
346 | ||
347 | if (!blk_queue_is_zoned(q)) | |
348 | return -ENOTTY; | |
349 | ||
350 | if (!capable(CAP_SYS_ADMIN)) | |
351 | return -EACCES; | |
352 | ||
353 | if (!(mode & FMODE_WRITE)) | |
354 | return -EBADF; | |
355 | ||
356 | if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range))) | |
357 | return -EFAULT; | |
358 | ||
359 | return blkdev_reset_zones(bdev, zrange.sector, zrange.nr_sectors, | |
360 | GFP_KERNEL); | |
361 | } | |
bf505456 DLM |
362 | |
363 | static inline unsigned long *blk_alloc_zone_bitmap(int node, | |
364 | unsigned int nr_zones) | |
365 | { | |
366 | return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long), | |
367 | GFP_NOIO, node); | |
368 | } | |
369 | ||
370 | /* | |
371 | * Allocate an array of struct blk_zone to get nr_zones zone information. | |
372 | * The allocated array may be smaller than nr_zones. | |
373 | */ | |
374 | static struct blk_zone *blk_alloc_zones(int node, unsigned int *nr_zones) | |
375 | { | |
376 | size_t size = *nr_zones * sizeof(struct blk_zone); | |
377 | struct page *page; | |
378 | int order; | |
379 | ||
927b6b2d | 380 | for (order = get_order(size); order >= 0; order--) { |
bf505456 DLM |
381 | page = alloc_pages_node(node, GFP_NOIO | __GFP_ZERO, order); |
382 | if (page) { | |
383 | *nr_zones = min_t(unsigned int, *nr_zones, | |
384 | (PAGE_SIZE << order) / sizeof(struct blk_zone)); | |
385 | return page_address(page); | |
386 | } | |
387 | } | |
388 | ||
389 | return NULL; | |
390 | } | |
391 | ||
392 | void blk_queue_free_zone_bitmaps(struct request_queue *q) | |
393 | { | |
394 | kfree(q->seq_zones_bitmap); | |
395 | q->seq_zones_bitmap = NULL; | |
396 | kfree(q->seq_zones_wlock); | |
397 | q->seq_zones_wlock = NULL; | |
398 | } | |
399 | ||
400 | /** | |
401 | * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps | |
402 | * @disk: Target disk | |
403 | * | |
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 | |
407 | * is allocated. | |
408 | */ | |
409 | int blk_revalidate_disk_zones(struct gendisk *disk) | |
410 | { | |
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; | |
bd976e52 | 416 | unsigned int noio_flag; |
bf505456 DLM |
417 | sector_t sector = 0; |
418 | int ret = 0; | |
419 | ||
420 | /* | |
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. | |
423 | */ | |
344e9ffc | 424 | if (!queue_is_mq(q)) { |
bf505456 DLM |
425 | q->nr_zones = nr_zones; |
426 | return 0; | |
427 | } | |
428 | ||
bd976e52 DLM |
429 | /* |
430 | * Ensure that all memory allocations in this context are done as | |
431 | * if GFP_NOIO was specified. | |
432 | */ | |
433 | noio_flag = memalloc_noio_save(); | |
434 | ||
bf505456 DLM |
435 | if (!blk_queue_is_zoned(q) || !nr_zones) { |
436 | nr_zones = 0; | |
437 | goto update; | |
438 | } | |
439 | ||
440 | /* Allocate bitmaps */ | |
441 | ret = -ENOMEM; | |
442 | seq_zones_wlock = blk_alloc_zone_bitmap(q->node, nr_zones); | |
443 | if (!seq_zones_wlock) | |
444 | goto out; | |
445 | seq_zones_bitmap = blk_alloc_zone_bitmap(q->node, nr_zones); | |
446 | if (!seq_zones_bitmap) | |
447 | goto out; | |
448 | ||
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); | |
452 | if (!zones) | |
453 | goto out; | |
454 | ||
455 | while (z < nr_zones) { | |
456 | nrz = min(nr_zones - z, rep_nr_zones); | |
bd976e52 | 457 | ret = blk_report_zones(disk, sector, zones, &nrz); |
bf505456 DLM |
458 | if (ret) |
459 | goto out; | |
460 | if (!nrz) | |
461 | break; | |
462 | for (i = 0; i < nrz; i++) { | |
463 | if (zones[i].type != BLK_ZONE_TYPE_CONVENTIONAL) | |
464 | set_bit(z, seq_zones_bitmap); | |
465 | z++; | |
466 | } | |
467 | sector += nrz * blk_queue_zone_sectors(q); | |
468 | } | |
469 | ||
470 | if (WARN_ON(z != nr_zones)) { | |
471 | ret = -EIO; | |
472 | goto out; | |
473 | } | |
474 | ||
475 | update: | |
476 | /* | |
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 | |
479 | * bitmaps). | |
480 | */ | |
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); | |
486 | ||
487 | out: | |
bd976e52 DLM |
488 | memalloc_noio_restore(noio_flag); |
489 | ||
bf505456 DLM |
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); | |
494 | ||
495 | if (ret) { | |
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); | |
500 | } | |
501 | ||
502 | return ret; | |
503 | } | |
504 | EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones); | |
505 |