]>
Commit | Line | Data |
---|---|---|
6a0cb1bc HR |
1 | /* |
2 | * Zoned block device handling | |
3 | * | |
4 | * Copyright (c) 2015, Hannes Reinecke | |
5 | * Copyright (c) 2015, SUSE Linux GmbH | |
6 | * | |
7 | * Copyright (c) 2016, Damien Le Moal | |
8 | * Copyright (c) 2016, Western Digital | |
9 | */ | |
10 | ||
11 | #include <linux/kernel.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/rbtree.h> | |
14 | #include <linux/blkdev.h> | |
15 | ||
a2d6b3a2 DLM |
16 | #include "blk.h" |
17 | ||
6a0cb1bc HR |
18 | static inline sector_t blk_zone_start(struct request_queue *q, |
19 | sector_t sector) | |
20 | { | |
f99e8648 | 21 | sector_t zone_mask = blk_queue_zone_sectors(q) - 1; |
6a0cb1bc HR |
22 | |
23 | return sector & ~zone_mask; | |
24 | } | |
25 | ||
6cc77e9c CH |
26 | /* |
27 | * Return true if a request is a write requests that needs zone write locking. | |
28 | */ | |
29 | bool blk_req_needs_zone_write_lock(struct request *rq) | |
30 | { | |
31 | if (!rq->q->seq_zones_wlock) | |
32 | return false; | |
33 | ||
34 | if (blk_rq_is_passthrough(rq)) | |
35 | return false; | |
36 | ||
37 | switch (req_op(rq)) { | |
38 | case REQ_OP_WRITE_ZEROES: | |
39 | case REQ_OP_WRITE_SAME: | |
40 | case REQ_OP_WRITE: | |
41 | return blk_rq_zone_is_seq(rq); | |
42 | default: | |
43 | return false; | |
44 | } | |
45 | } | |
46 | EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock); | |
47 | ||
48 | void __blk_req_zone_write_lock(struct request *rq) | |
49 | { | |
50 | if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq), | |
51 | rq->q->seq_zones_wlock))) | |
52 | return; | |
53 | ||
54 | WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED); | |
55 | rq->rq_flags |= RQF_ZONE_WRITE_LOCKED; | |
56 | } | |
57 | EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock); | |
58 | ||
59 | void __blk_req_zone_write_unlock(struct request *rq) | |
60 | { | |
61 | rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED; | |
62 | if (rq->q->seq_zones_wlock) | |
63 | WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq), | |
64 | rq->q->seq_zones_wlock)); | |
65 | } | |
66 | EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock); | |
67 | ||
a91e1380 DLM |
68 | static inline unsigned int __blkdev_nr_zones(struct request_queue *q, |
69 | sector_t nr_sectors) | |
70 | { | |
71 | unsigned long zone_sectors = blk_queue_zone_sectors(q); | |
72 | ||
73 | return (nr_sectors + zone_sectors - 1) >> ilog2(zone_sectors); | |
74 | } | |
75 | ||
76 | /** | |
77 | * blkdev_nr_zones - Get number of zones | |
78 | * @bdev: Target block device | |
79 | * | |
80 | * Description: | |
81 | * Return the total number of zones of a zoned block device. | |
82 | * For a regular block device, the number of zones is always 0. | |
83 | */ | |
84 | unsigned int blkdev_nr_zones(struct block_device *bdev) | |
85 | { | |
86 | struct request_queue *q = bdev_get_queue(bdev); | |
87 | ||
88 | if (!blk_queue_is_zoned(q)) | |
89 | return 0; | |
90 | ||
91 | return __blkdev_nr_zones(q, bdev->bd_part->nr_sects); | |
92 | } | |
93 | EXPORT_SYMBOL_GPL(blkdev_nr_zones); | |
94 | ||
6a0cb1bc HR |
95 | /* |
96 | * Check that a zone report belongs to the partition. | |
97 | * If yes, fix its start sector and write pointer, copy it in the | |
98 | * zone information array and return true. Return false otherwise. | |
99 | */ | |
100 | static bool blkdev_report_zone(struct block_device *bdev, | |
101 | struct blk_zone *rep, | |
102 | struct blk_zone *zone) | |
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; | |
117 | memcpy(zone, rep, sizeof(struct blk_zone)); | |
118 | ||
119 | return true; | |
120 | } | |
121 | ||
122 | /** | |
123 | * blkdev_report_zones - Get zones information | |
124 | * @bdev: Target block device | |
125 | * @sector: Sector from which to report zones | |
126 | * @zones: Array of zone structures where to return the zones information | |
127 | * @nr_zones: Number of zone structures in the zone array | |
128 | * @gfp_mask: Memory allocation flags (for bio_alloc) | |
129 | * | |
130 | * Description: | |
131 | * Get zone information starting from the zone containing @sector. | |
132 | * The number of zone information reported may be less than the number | |
133 | * requested by @nr_zones. The number of zones actually reported is | |
134 | * returned in @nr_zones. | |
135 | */ | |
136 | int blkdev_report_zones(struct block_device *bdev, | |
137 | sector_t sector, | |
138 | struct blk_zone *zones, | |
139 | unsigned int *nr_zones, | |
140 | gfp_t gfp_mask) | |
141 | { | |
142 | struct request_queue *q = bdev_get_queue(bdev); | |
143 | struct blk_zone_report_hdr *hdr; | |
144 | unsigned int nrz = *nr_zones; | |
145 | struct page *page; | |
146 | unsigned int nr_rep; | |
147 | size_t rep_bytes; | |
148 | unsigned int nr_pages; | |
149 | struct bio *bio; | |
150 | struct bio_vec *bv; | |
151 | unsigned int i, n, nz; | |
152 | unsigned int ofst; | |
153 | void *addr; | |
3c4da758 | 154 | int ret; |
6a0cb1bc HR |
155 | |
156 | if (!q) | |
157 | return -ENXIO; | |
158 | ||
159 | if (!blk_queue_is_zoned(q)) | |
160 | return -EOPNOTSUPP; | |
161 | ||
162 | if (!nrz) | |
163 | return 0; | |
164 | ||
165 | if (sector > bdev->bd_part->nr_sects) { | |
166 | *nr_zones = 0; | |
167 | return 0; | |
168 | } | |
169 | ||
170 | /* | |
171 | * The zone report has a header. So make room for it in the | |
172 | * payload. Also make sure that the report fits in a single BIO | |
173 | * that will not be split down the stack. | |
174 | */ | |
175 | rep_bytes = sizeof(struct blk_zone_report_hdr) + | |
176 | sizeof(struct blk_zone) * nrz; | |
177 | rep_bytes = (rep_bytes + PAGE_SIZE - 1) & PAGE_MASK; | |
178 | if (rep_bytes > (queue_max_sectors(q) << 9)) | |
179 | rep_bytes = queue_max_sectors(q) << 9; | |
180 | ||
181 | nr_pages = min_t(unsigned int, BIO_MAX_PAGES, | |
182 | rep_bytes >> PAGE_SHIFT); | |
183 | nr_pages = min_t(unsigned int, nr_pages, | |
184 | queue_max_segments(q)); | |
185 | ||
186 | bio = bio_alloc(gfp_mask, nr_pages); | |
187 | if (!bio) | |
188 | return -ENOMEM; | |
189 | ||
74d46992 | 190 | bio_set_dev(bio, bdev); |
6a0cb1bc HR |
191 | bio->bi_iter.bi_sector = blk_zone_start(q, sector); |
192 | bio_set_op_attrs(bio, REQ_OP_ZONE_REPORT, 0); | |
193 | ||
194 | for (i = 0; i < nr_pages; i++) { | |
195 | page = alloc_page(gfp_mask); | |
196 | if (!page) { | |
197 | ret = -ENOMEM; | |
198 | goto out; | |
199 | } | |
200 | if (!bio_add_page(bio, page, PAGE_SIZE, 0)) { | |
201 | __free_page(page); | |
202 | break; | |
203 | } | |
204 | } | |
205 | ||
206 | if (i == 0) | |
207 | ret = -ENOMEM; | |
208 | else | |
209 | ret = submit_bio_wait(bio); | |
210 | if (ret) | |
211 | goto out; | |
212 | ||
213 | /* | |
214 | * Process the report result: skip the header and go through the | |
215 | * reported zones to fixup and fixup the zone information for | |
216 | * partitions. At the same time, return the zone information into | |
217 | * the zone array. | |
218 | */ | |
219 | n = 0; | |
220 | nz = 0; | |
221 | nr_rep = 0; | |
222 | bio_for_each_segment_all(bv, bio, i) { | |
223 | ||
224 | if (!bv->bv_page) | |
225 | break; | |
226 | ||
227 | addr = kmap_atomic(bv->bv_page); | |
228 | ||
229 | /* Get header in the first page */ | |
230 | ofst = 0; | |
231 | if (!nr_rep) { | |
f441108f | 232 | hdr = addr; |
6a0cb1bc HR |
233 | nr_rep = hdr->nr_zones; |
234 | ofst = sizeof(struct blk_zone_report_hdr); | |
235 | } | |
236 | ||
237 | /* Fixup and report zones */ | |
238 | while (ofst < bv->bv_len && | |
239 | n < nr_rep && nz < nrz) { | |
240 | if (blkdev_report_zone(bdev, addr + ofst, &zones[nz])) | |
241 | nz++; | |
242 | ofst += sizeof(struct blk_zone); | |
243 | n++; | |
244 | } | |
245 | ||
246 | kunmap_atomic(addr); | |
247 | ||
248 | if (n >= nr_rep || nz >= nrz) | |
249 | break; | |
250 | ||
251 | } | |
252 | ||
3c4da758 | 253 | *nr_zones = nz; |
6a0cb1bc HR |
254 | out: |
255 | bio_for_each_segment_all(bv, bio, i) | |
256 | __free_page(bv->bv_page); | |
257 | bio_put(bio); | |
258 | ||
6a0cb1bc HR |
259 | return ret; |
260 | } | |
261 | EXPORT_SYMBOL_GPL(blkdev_report_zones); | |
262 | ||
263 | /** | |
264 | * blkdev_reset_zones - Reset zones write pointer | |
265 | * @bdev: Target block device | |
266 | * @sector: Start sector of the first zone to reset | |
267 | * @nr_sectors: Number of sectors, at least the length of one zone | |
268 | * @gfp_mask: Memory allocation flags (for bio_alloc) | |
269 | * | |
270 | * Description: | |
271 | * Reset the write pointer of the zones contained in the range | |
272 | * @sector..@sector+@nr_sectors. Specifying the entire disk sector range | |
273 | * is valid, but the specified range should not contain conventional zones. | |
274 | */ | |
275 | int blkdev_reset_zones(struct block_device *bdev, | |
276 | sector_t sector, sector_t nr_sectors, | |
277 | gfp_t gfp_mask) | |
278 | { | |
279 | struct request_queue *q = bdev_get_queue(bdev); | |
280 | sector_t zone_sectors; | |
281 | sector_t end_sector = sector + nr_sectors; | |
a2d6b3a2 DLM |
282 | struct bio *bio = NULL; |
283 | struct blk_plug plug; | |
6a0cb1bc HR |
284 | int ret; |
285 | ||
6a0cb1bc HR |
286 | if (!blk_queue_is_zoned(q)) |
287 | return -EOPNOTSUPP; | |
288 | ||
a2d6b3a2 DLM |
289 | if (bdev_read_only(bdev)) |
290 | return -EPERM; | |
291 | ||
292 | if (!nr_sectors || end_sector > bdev->bd_part->nr_sects) | |
6a0cb1bc HR |
293 | /* Out of range */ |
294 | return -EINVAL; | |
295 | ||
296 | /* Check alignment (handle eventual smaller last zone) */ | |
f99e8648 | 297 | zone_sectors = blk_queue_zone_sectors(q); |
6a0cb1bc HR |
298 | if (sector & (zone_sectors - 1)) |
299 | return -EINVAL; | |
300 | ||
301 | if ((nr_sectors & (zone_sectors - 1)) && | |
302 | end_sector != bdev->bd_part->nr_sects) | |
303 | return -EINVAL; | |
304 | ||
a2d6b3a2 | 305 | blk_start_plug(&plug); |
6a0cb1bc HR |
306 | while (sector < end_sector) { |
307 | ||
a2d6b3a2 | 308 | bio = blk_next_bio(bio, 0, gfp_mask); |
6a0cb1bc | 309 | bio->bi_iter.bi_sector = sector; |
74d46992 | 310 | bio_set_dev(bio, bdev); |
6a0cb1bc HR |
311 | bio_set_op_attrs(bio, REQ_OP_ZONE_RESET, 0); |
312 | ||
6a0cb1bc HR |
313 | sector += zone_sectors; |
314 | ||
315 | /* This may take a while, so be nice to others */ | |
316 | cond_resched(); | |
317 | ||
318 | } | |
319 | ||
a2d6b3a2 DLM |
320 | ret = submit_bio_wait(bio); |
321 | bio_put(bio); | |
322 | ||
323 | blk_finish_plug(&plug); | |
324 | ||
325 | return ret; | |
6a0cb1bc HR |
326 | } |
327 | EXPORT_SYMBOL_GPL(blkdev_reset_zones); | |
3ed05a98 | 328 | |
56c4bddb | 329 | /* |
3ed05a98 ST |
330 | * BLKREPORTZONE ioctl processing. |
331 | * Called from blkdev_ioctl. | |
332 | */ | |
333 | int blkdev_report_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_report rep; | |
339 | struct blk_zone *zones; | |
340 | int ret; | |
341 | ||
342 | if (!argp) | |
343 | return -EINVAL; | |
344 | ||
345 | q = bdev_get_queue(bdev); | |
346 | if (!q) | |
347 | return -ENXIO; | |
348 | ||
349 | if (!blk_queue_is_zoned(q)) | |
350 | return -ENOTTY; | |
351 | ||
352 | if (!capable(CAP_SYS_ADMIN)) | |
353 | return -EACCES; | |
354 | ||
355 | if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report))) | |
356 | return -EFAULT; | |
357 | ||
358 | if (!rep.nr_zones) | |
359 | return -EINVAL; | |
360 | ||
2e85fbaf | 361 | rep.nr_zones = min(blkdev_nr_zones(bdev), rep.nr_zones); |
327ea4ad | 362 | |
344476e1 KC |
363 | zones = kvmalloc_array(rep.nr_zones, sizeof(struct blk_zone), |
364 | GFP_KERNEL | __GFP_ZERO); | |
3ed05a98 ST |
365 | if (!zones) |
366 | return -ENOMEM; | |
367 | ||
368 | ret = blkdev_report_zones(bdev, rep.sector, | |
369 | zones, &rep.nr_zones, | |
370 | GFP_KERNEL); | |
371 | if (ret) | |
372 | goto out; | |
373 | ||
374 | if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report))) { | |
375 | ret = -EFAULT; | |
376 | goto out; | |
377 | } | |
378 | ||
379 | if (rep.nr_zones) { | |
380 | if (copy_to_user(argp + sizeof(struct blk_zone_report), zones, | |
381 | sizeof(struct blk_zone) * rep.nr_zones)) | |
382 | ret = -EFAULT; | |
383 | } | |
384 | ||
385 | out: | |
327ea4ad | 386 | kvfree(zones); |
3ed05a98 ST |
387 | |
388 | return ret; | |
389 | } | |
390 | ||
56c4bddb | 391 | /* |
3ed05a98 ST |
392 | * BLKRESETZONE ioctl processing. |
393 | * Called from blkdev_ioctl. | |
394 | */ | |
395 | int blkdev_reset_zones_ioctl(struct block_device *bdev, fmode_t mode, | |
396 | unsigned int cmd, unsigned long arg) | |
397 | { | |
398 | void __user *argp = (void __user *)arg; | |
399 | struct request_queue *q; | |
400 | struct blk_zone_range zrange; | |
401 | ||
402 | if (!argp) | |
403 | return -EINVAL; | |
404 | ||
405 | q = bdev_get_queue(bdev); | |
406 | if (!q) | |
407 | return -ENXIO; | |
408 | ||
409 | if (!blk_queue_is_zoned(q)) | |
410 | return -ENOTTY; | |
411 | ||
412 | if (!capable(CAP_SYS_ADMIN)) | |
413 | return -EACCES; | |
414 | ||
415 | if (!(mode & FMODE_WRITE)) | |
416 | return -EBADF; | |
417 | ||
418 | if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range))) | |
419 | return -EFAULT; | |
420 | ||
421 | return blkdev_reset_zones(bdev, zrange.sector, zrange.nr_sectors, | |
422 | GFP_KERNEL); | |
423 | } |