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> |
26202928 DLM |
17 | #include <linux/mm.h> |
18 | #include <linux/vmalloc.h> | |
bd976e52 | 19 | #include <linux/sched/mm.h> |
6a0cb1bc | 20 | |
a2d6b3a2 DLM |
21 | #include "blk.h" |
22 | ||
6a0cb1bc HR |
23 | static inline sector_t blk_zone_start(struct request_queue *q, |
24 | sector_t sector) | |
25 | { | |
f99e8648 | 26 | sector_t zone_mask = blk_queue_zone_sectors(q) - 1; |
6a0cb1bc HR |
27 | |
28 | return sector & ~zone_mask; | |
29 | } | |
30 | ||
6cc77e9c CH |
31 | /* |
32 | * Return true if a request is a write requests that needs zone write locking. | |
33 | */ | |
34 | bool blk_req_needs_zone_write_lock(struct request *rq) | |
35 | { | |
36 | if (!rq->q->seq_zones_wlock) | |
37 | return false; | |
38 | ||
39 | if (blk_rq_is_passthrough(rq)) | |
40 | return false; | |
41 | ||
42 | switch (req_op(rq)) { | |
43 | case REQ_OP_WRITE_ZEROES: | |
44 | case REQ_OP_WRITE_SAME: | |
45 | case REQ_OP_WRITE: | |
46 | return blk_rq_zone_is_seq(rq); | |
47 | default: | |
48 | return false; | |
49 | } | |
50 | } | |
51 | EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock); | |
52 | ||
53 | void __blk_req_zone_write_lock(struct request *rq) | |
54 | { | |
55 | if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq), | |
56 | rq->q->seq_zones_wlock))) | |
57 | return; | |
58 | ||
59 | WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED); | |
60 | rq->rq_flags |= RQF_ZONE_WRITE_LOCKED; | |
61 | } | |
62 | EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock); | |
63 | ||
64 | void __blk_req_zone_write_unlock(struct request *rq) | |
65 | { | |
66 | rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED; | |
67 | if (rq->q->seq_zones_wlock) | |
68 | WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq), | |
69 | rq->q->seq_zones_wlock)); | |
70 | } | |
71 | EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock); | |
72 | ||
a91e1380 DLM |
73 | /** |
74 | * blkdev_nr_zones - Get number of zones | |
9b38bb4b | 75 | * @disk: Target gendisk |
a91e1380 | 76 | * |
9b38bb4b CH |
77 | * Return the total number of zones of a zoned block device. For a block |
78 | * device without zone capabilities, the number of zones is always 0. | |
a91e1380 | 79 | */ |
9b38bb4b | 80 | unsigned int blkdev_nr_zones(struct gendisk *disk) |
a91e1380 | 81 | { |
9b38bb4b | 82 | sector_t zone_sectors = blk_queue_zone_sectors(disk->queue); |
a91e1380 | 83 | |
9b38bb4b | 84 | if (!blk_queue_is_zoned(disk->queue)) |
a91e1380 | 85 | return 0; |
9b38bb4b | 86 | return (get_capacity(disk) + zone_sectors - 1) >> ilog2(zone_sectors); |
a91e1380 DLM |
87 | } |
88 | EXPORT_SYMBOL_GPL(blkdev_nr_zones); | |
89 | ||
6a0cb1bc HR |
90 | /** |
91 | * blkdev_report_zones - Get zones information | |
92 | * @bdev: Target block device | |
93 | * @sector: Sector from which to report zones | |
d4100351 CH |
94 | * @nr_zones: Maximum number of zones to report |
95 | * @cb: Callback function called for each reported zone | |
96 | * @data: Private data for the callback | |
6a0cb1bc HR |
97 | * |
98 | * Description: | |
d4100351 CH |
99 | * Get zone information starting from the zone containing @sector for at most |
100 | * @nr_zones, and call @cb for each zone reported by the device. | |
101 | * To report all zones in a device starting from @sector, the BLK_ALL_ZONES | |
102 | * constant can be passed to @nr_zones. | |
103 | * Returns the number of zones reported by the device, or a negative errno | |
104 | * value in case of failure. | |
105 | * | |
106 | * Note: The caller must use memalloc_noXX_save/restore() calls to control | |
107 | * memory allocations done within this function. | |
6a0cb1bc | 108 | */ |
e76239a3 | 109 | int blkdev_report_zones(struct block_device *bdev, sector_t sector, |
d4100351 | 110 | unsigned int nr_zones, report_zones_cb cb, void *data) |
6a0cb1bc | 111 | { |
ceeb373a | 112 | struct gendisk *disk = bdev->bd_disk; |
5eac3eb3 | 113 | sector_t capacity = get_capacity(disk); |
6a0cb1bc | 114 | |
d4100351 CH |
115 | if (!blk_queue_is_zoned(bdev_get_queue(bdev)) || |
116 | WARN_ON_ONCE(!disk->fops->report_zones)) | |
e76239a3 | 117 | return -EOPNOTSUPP; |
6a0cb1bc | 118 | |
d4100351 | 119 | if (!nr_zones || sector >= capacity) |
6a0cb1bc | 120 | return 0; |
6a0cb1bc | 121 | |
d4100351 | 122 | return disk->fops->report_zones(disk, sector, nr_zones, cb, data); |
6a0cb1bc HR |
123 | } |
124 | EXPORT_SYMBOL_GPL(blkdev_report_zones); | |
125 | ||
6e33dbf2 | 126 | static inline bool blkdev_allow_reset_all_zones(struct block_device *bdev, |
c7a1d926 | 127 | sector_t sector, |
6e33dbf2 CK |
128 | sector_t nr_sectors) |
129 | { | |
130 | if (!blk_queue_zone_resetall(bdev_get_queue(bdev))) | |
131 | return false; | |
132 | ||
6e33dbf2 | 133 | /* |
5eac3eb3 DLM |
134 | * REQ_OP_ZONE_RESET_ALL can be executed only if the number of sectors |
135 | * of the applicable zone range is the entire disk. | |
6e33dbf2 | 136 | */ |
5eac3eb3 | 137 | return !sector && nr_sectors == get_capacity(bdev->bd_disk); |
6e33dbf2 CK |
138 | } |
139 | ||
6a0cb1bc | 140 | /** |
6c1b1da5 | 141 | * blkdev_zone_mgmt - Execute a zone management operation on a range of zones |
6a0cb1bc | 142 | * @bdev: Target block device |
6c1b1da5 AJ |
143 | * @op: Operation to be performed on the zones |
144 | * @sector: Start sector of the first zone to operate on | |
145 | * @nr_sectors: Number of sectors, should be at least the length of one zone and | |
146 | * must be zone size aligned. | |
6a0cb1bc HR |
147 | * @gfp_mask: Memory allocation flags (for bio_alloc) |
148 | * | |
149 | * Description: | |
6c1b1da5 | 150 | * Perform the specified operation on the range of zones specified by |
6a0cb1bc HR |
151 | * @sector..@sector+@nr_sectors. Specifying the entire disk sector range |
152 | * is valid, but the specified range should not contain conventional zones. | |
6c1b1da5 AJ |
153 | * The operation to execute on each zone can be a zone reset, open, close |
154 | * or finish request. | |
6a0cb1bc | 155 | */ |
6c1b1da5 AJ |
156 | int blkdev_zone_mgmt(struct block_device *bdev, enum req_opf op, |
157 | sector_t sector, sector_t nr_sectors, | |
158 | gfp_t gfp_mask) | |
6a0cb1bc HR |
159 | { |
160 | struct request_queue *q = bdev_get_queue(bdev); | |
6c1b1da5 | 161 | sector_t zone_sectors = blk_queue_zone_sectors(q); |
5eac3eb3 | 162 | sector_t capacity = get_capacity(bdev->bd_disk); |
6a0cb1bc | 163 | sector_t end_sector = sector + nr_sectors; |
a2d6b3a2 | 164 | struct bio *bio = NULL; |
6a0cb1bc HR |
165 | int ret; |
166 | ||
6a0cb1bc HR |
167 | if (!blk_queue_is_zoned(q)) |
168 | return -EOPNOTSUPP; | |
169 | ||
a2d6b3a2 DLM |
170 | if (bdev_read_only(bdev)) |
171 | return -EPERM; | |
172 | ||
6c1b1da5 AJ |
173 | if (!op_is_zone_mgmt(op)) |
174 | return -EOPNOTSUPP; | |
175 | ||
5eac3eb3 | 176 | if (!nr_sectors || end_sector > capacity) |
6a0cb1bc HR |
177 | /* Out of range */ |
178 | return -EINVAL; | |
179 | ||
180 | /* Check alignment (handle eventual smaller last zone) */ | |
6a0cb1bc HR |
181 | if (sector & (zone_sectors - 1)) |
182 | return -EINVAL; | |
183 | ||
5eac3eb3 | 184 | if ((nr_sectors & (zone_sectors - 1)) && end_sector != capacity) |
6a0cb1bc HR |
185 | return -EINVAL; |
186 | ||
187 | while (sector < end_sector) { | |
a2d6b3a2 | 188 | bio = blk_next_bio(bio, 0, gfp_mask); |
74d46992 | 189 | bio_set_dev(bio, bdev); |
6a0cb1bc | 190 | |
c7a1d926 DLM |
191 | /* |
192 | * Special case for the zone reset operation that reset all | |
193 | * zones, this is useful for applications like mkfs. | |
194 | */ | |
6c1b1da5 AJ |
195 | if (op == REQ_OP_ZONE_RESET && |
196 | blkdev_allow_reset_all_zones(bdev, sector, nr_sectors)) { | |
c7a1d926 DLM |
197 | bio->bi_opf = REQ_OP_ZONE_RESET_ALL; |
198 | break; | |
199 | } | |
200 | ||
8e42d239 | 201 | bio->bi_opf = op | REQ_SYNC; |
c7a1d926 | 202 | bio->bi_iter.bi_sector = sector; |
6a0cb1bc HR |
203 | sector += zone_sectors; |
204 | ||
205 | /* This may take a while, so be nice to others */ | |
206 | cond_resched(); | |
6a0cb1bc HR |
207 | } |
208 | ||
a2d6b3a2 DLM |
209 | ret = submit_bio_wait(bio); |
210 | bio_put(bio); | |
211 | ||
a2d6b3a2 | 212 | return ret; |
6a0cb1bc | 213 | } |
6c1b1da5 | 214 | EXPORT_SYMBOL_GPL(blkdev_zone_mgmt); |
3ed05a98 | 215 | |
d4100351 CH |
216 | struct zone_report_args { |
217 | struct blk_zone __user *zones; | |
218 | }; | |
219 | ||
220 | static int blkdev_copy_zone_to_user(struct blk_zone *zone, unsigned int idx, | |
221 | void *data) | |
222 | { | |
223 | struct zone_report_args *args = data; | |
224 | ||
225 | if (copy_to_user(&args->zones[idx], zone, sizeof(struct blk_zone))) | |
226 | return -EFAULT; | |
227 | return 0; | |
228 | } | |
229 | ||
56c4bddb | 230 | /* |
3ed05a98 ST |
231 | * BLKREPORTZONE ioctl processing. |
232 | * Called from blkdev_ioctl. | |
233 | */ | |
234 | int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode, | |
235 | unsigned int cmd, unsigned long arg) | |
236 | { | |
237 | void __user *argp = (void __user *)arg; | |
d4100351 | 238 | struct zone_report_args args; |
3ed05a98 ST |
239 | struct request_queue *q; |
240 | struct blk_zone_report rep; | |
3ed05a98 ST |
241 | int ret; |
242 | ||
243 | if (!argp) | |
244 | return -EINVAL; | |
245 | ||
246 | q = bdev_get_queue(bdev); | |
247 | if (!q) | |
248 | return -ENXIO; | |
249 | ||
250 | if (!blk_queue_is_zoned(q)) | |
251 | return -ENOTTY; | |
252 | ||
253 | if (!capable(CAP_SYS_ADMIN)) | |
254 | return -EACCES; | |
255 | ||
256 | if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report))) | |
257 | return -EFAULT; | |
258 | ||
259 | if (!rep.nr_zones) | |
260 | return -EINVAL; | |
261 | ||
d4100351 CH |
262 | args.zones = argp + sizeof(struct blk_zone_report); |
263 | ret = blkdev_report_zones(bdev, rep.sector, rep.nr_zones, | |
264 | blkdev_copy_zone_to_user, &args); | |
265 | if (ret < 0) | |
266 | return ret; | |
3ed05a98 | 267 | |
d4100351 CH |
268 | rep.nr_zones = ret; |
269 | if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report))) | |
270 | return -EFAULT; | |
271 | return 0; | |
3ed05a98 ST |
272 | } |
273 | ||
56c4bddb | 274 | /* |
e876df1f | 275 | * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing. |
3ed05a98 ST |
276 | * Called from blkdev_ioctl. |
277 | */ | |
e876df1f AJ |
278 | int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode, |
279 | unsigned int cmd, unsigned long arg) | |
3ed05a98 ST |
280 | { |
281 | void __user *argp = (void __user *)arg; | |
282 | struct request_queue *q; | |
283 | struct blk_zone_range zrange; | |
e876df1f | 284 | enum req_opf op; |
3ed05a98 ST |
285 | |
286 | if (!argp) | |
287 | return -EINVAL; | |
288 | ||
289 | q = bdev_get_queue(bdev); | |
290 | if (!q) | |
291 | return -ENXIO; | |
292 | ||
293 | if (!blk_queue_is_zoned(q)) | |
294 | return -ENOTTY; | |
295 | ||
296 | if (!capable(CAP_SYS_ADMIN)) | |
297 | return -EACCES; | |
298 | ||
299 | if (!(mode & FMODE_WRITE)) | |
300 | return -EBADF; | |
301 | ||
302 | if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range))) | |
303 | return -EFAULT; | |
304 | ||
e876df1f AJ |
305 | switch (cmd) { |
306 | case BLKRESETZONE: | |
307 | op = REQ_OP_ZONE_RESET; | |
308 | break; | |
309 | case BLKOPENZONE: | |
310 | op = REQ_OP_ZONE_OPEN; | |
311 | break; | |
312 | case BLKCLOSEZONE: | |
313 | op = REQ_OP_ZONE_CLOSE; | |
314 | break; | |
315 | case BLKFINISHZONE: | |
316 | op = REQ_OP_ZONE_FINISH; | |
317 | break; | |
318 | default: | |
319 | return -ENOTTY; | |
320 | } | |
321 | ||
322 | return blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors, | |
323 | GFP_KERNEL); | |
3ed05a98 | 324 | } |
bf505456 DLM |
325 | |
326 | static inline unsigned long *blk_alloc_zone_bitmap(int node, | |
327 | unsigned int nr_zones) | |
328 | { | |
329 | return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long), | |
330 | GFP_NOIO, node); | |
331 | } | |
332 | ||
bf505456 DLM |
333 | void blk_queue_free_zone_bitmaps(struct request_queue *q) |
334 | { | |
f216fdd7 CH |
335 | kfree(q->conv_zones_bitmap); |
336 | q->conv_zones_bitmap = NULL; | |
bf505456 DLM |
337 | kfree(q->seq_zones_wlock); |
338 | q->seq_zones_wlock = NULL; | |
339 | } | |
340 | ||
d4100351 CH |
341 | struct blk_revalidate_zone_args { |
342 | struct gendisk *disk; | |
f216fdd7 | 343 | unsigned long *conv_zones_bitmap; |
d4100351 | 344 | unsigned long *seq_zones_wlock; |
e94f5819 | 345 | unsigned int nr_zones; |
6c6b3549 | 346 | sector_t zone_sectors; |
d4100351 CH |
347 | sector_t sector; |
348 | }; | |
349 | ||
d9dd7308 DLM |
350 | /* |
351 | * Helper function to check the validity of zones of a zoned block device. | |
352 | */ | |
d4100351 CH |
353 | static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx, |
354 | void *data) | |
d9dd7308 | 355 | { |
d4100351 CH |
356 | struct blk_revalidate_zone_args *args = data; |
357 | struct gendisk *disk = args->disk; | |
d9dd7308 | 358 | struct request_queue *q = disk->queue; |
d9dd7308 DLM |
359 | sector_t capacity = get_capacity(disk); |
360 | ||
361 | /* | |
362 | * All zones must have the same size, with the exception on an eventual | |
363 | * smaller last zone. | |
364 | */ | |
6c6b3549 CH |
365 | if (zone->start == 0) { |
366 | if (zone->len == 0 || !is_power_of_2(zone->len)) { | |
367 | pr_warn("%s: Invalid zoned device with non power of two zone size (%llu)\n", | |
368 | disk->disk_name, zone->len); | |
369 | return -ENODEV; | |
370 | } | |
d9dd7308 | 371 | |
6c6b3549 CH |
372 | args->zone_sectors = zone->len; |
373 | args->nr_zones = (capacity + zone->len - 1) >> ilog2(zone->len); | |
374 | } else if (zone->start + args->zone_sectors < capacity) { | |
375 | if (zone->len != args->zone_sectors) { | |
376 | pr_warn("%s: Invalid zoned device with non constant zone size\n", | |
377 | disk->disk_name); | |
378 | return -ENODEV; | |
379 | } | |
380 | } else { | |
381 | if (zone->len > args->zone_sectors) { | |
382 | pr_warn("%s: Invalid zoned device with larger last zone size\n", | |
383 | disk->disk_name); | |
384 | return -ENODEV; | |
385 | } | |
d9dd7308 DLM |
386 | } |
387 | ||
388 | /* Check for holes in the zone report */ | |
d4100351 | 389 | if (zone->start != args->sector) { |
d9dd7308 | 390 | pr_warn("%s: Zone gap at sectors %llu..%llu\n", |
d4100351 CH |
391 | disk->disk_name, args->sector, zone->start); |
392 | return -ENODEV; | |
d9dd7308 DLM |
393 | } |
394 | ||
395 | /* Check zone type */ | |
396 | switch (zone->type) { | |
397 | case BLK_ZONE_TYPE_CONVENTIONAL: | |
e94f5819 CH |
398 | if (!args->conv_zones_bitmap) { |
399 | args->conv_zones_bitmap = | |
400 | blk_alloc_zone_bitmap(q->node, args->nr_zones); | |
401 | if (!args->conv_zones_bitmap) | |
402 | return -ENOMEM; | |
403 | } | |
404 | set_bit(idx, args->conv_zones_bitmap); | |
405 | break; | |
d9dd7308 DLM |
406 | case BLK_ZONE_TYPE_SEQWRITE_REQ: |
407 | case BLK_ZONE_TYPE_SEQWRITE_PREF: | |
e94f5819 CH |
408 | if (!args->seq_zones_wlock) { |
409 | args->seq_zones_wlock = | |
410 | blk_alloc_zone_bitmap(q->node, args->nr_zones); | |
411 | if (!args->seq_zones_wlock) | |
412 | return -ENOMEM; | |
413 | } | |
d9dd7308 DLM |
414 | break; |
415 | default: | |
416 | pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n", | |
417 | disk->disk_name, (int)zone->type, zone->start); | |
d4100351 | 418 | return -ENODEV; |
d9dd7308 DLM |
419 | } |
420 | ||
d4100351 CH |
421 | args->sector += zone->len; |
422 | return 0; | |
423 | } | |
424 | ||
bf505456 DLM |
425 | /** |
426 | * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps | |
427 | * @disk: Target disk | |
428 | * | |
429 | * Helper function for low-level device drivers to (re) allocate and initialize | |
430 | * a disk request queue zone bitmaps. This functions should normally be called | |
ae58954d CH |
431 | * within the disk ->revalidate method for blk-mq based drivers. For BIO based |
432 | * drivers only q->nr_zones needs to be updated so that the sysfs exposed value | |
433 | * is correct. | |
bf505456 DLM |
434 | */ |
435 | int blk_revalidate_disk_zones(struct gendisk *disk) | |
436 | { | |
437 | struct request_queue *q = disk->queue; | |
e94f5819 CH |
438 | struct blk_revalidate_zone_args args = { |
439 | .disk = disk, | |
e94f5819 | 440 | }; |
6c6b3549 CH |
441 | unsigned int noio_flag; |
442 | int ret; | |
bf505456 | 443 | |
c98c3d09 CH |
444 | if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) |
445 | return -EIO; | |
ae58954d CH |
446 | if (WARN_ON_ONCE(!queue_is_mq(q))) |
447 | return -EIO; | |
bf505456 | 448 | |
e94f5819 | 449 | /* |
6c6b3549 CH |
450 | * Ensure that all memory allocations in this context are done as if |
451 | * GFP_NOIO was specified. | |
e94f5819 | 452 | */ |
6c6b3549 CH |
453 | noio_flag = memalloc_noio_save(); |
454 | ret = disk->fops->report_zones(disk, 0, UINT_MAX, | |
455 | blk_revalidate_zone_cb, &args); | |
456 | memalloc_noio_restore(noio_flag); | |
bf505456 | 457 | |
bf505456 | 458 | /* |
6c6b3549 CH |
459 | * Install the new bitmaps and update nr_zones only once the queue is |
460 | * stopped and all I/Os are completed (i.e. a scheduler is not | |
461 | * referencing the bitmaps). | |
bf505456 DLM |
462 | */ |
463 | blk_mq_freeze_queue(q); | |
d4100351 | 464 | if (ret >= 0) { |
6c6b3549 | 465 | blk_queue_chunk_sectors(q, args.zone_sectors); |
e94f5819 | 466 | q->nr_zones = args.nr_zones; |
d4100351 | 467 | swap(q->seq_zones_wlock, args.seq_zones_wlock); |
f216fdd7 | 468 | swap(q->conv_zones_bitmap, args.conv_zones_bitmap); |
d4100351 CH |
469 | ret = 0; |
470 | } else { | |
bf505456 | 471 | pr_warn("%s: failed to revalidate zones\n", disk->disk_name); |
bf505456 | 472 | blk_queue_free_zone_bitmaps(q); |
bf505456 | 473 | } |
d4100351 | 474 | blk_mq_unfreeze_queue(q); |
bf505456 | 475 | |
d4100351 | 476 | kfree(args.seq_zones_wlock); |
f216fdd7 | 477 | kfree(args.conv_zones_bitmap); |
bf505456 DLM |
478 | return ret; |
479 | } | |
480 | EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones); |