blk-lib.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441
  1. // SPDX-License-Identifier: GPL-2.0
  2. /*
  3. * Functions related to generic helpers functions
  4. */
  5. #include <linux/kernel.h>
  6. #include <linux/module.h>
  7. #include <linux/bio.h>
  8. #include <linux/blkdev.h>
  9. #include <linux/scatterlist.h>
  10. #include "blk.h"
  11. struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp)
  12. {
  13. struct bio *new = bio_alloc(gfp, nr_pages);
  14. if (bio) {
  15. bio_chain(bio, new);
  16. submit_bio(bio);
  17. }
  18. return new;
  19. }
  20. int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  21. sector_t nr_sects, gfp_t gfp_mask, int flags,
  22. struct bio **biop)
  23. {
  24. struct request_queue *q = bdev_get_queue(bdev);
  25. struct bio *bio = *biop;
  26. unsigned int op;
  27. sector_t bs_mask, part_offset = 0;
  28. if (!q)
  29. return -ENXIO;
  30. if (bdev_read_only(bdev))
  31. return -EPERM;
  32. if (flags & BLKDEV_DISCARD_SECURE) {
  33. if (!blk_queue_secure_erase(q))
  34. return -EOPNOTSUPP;
  35. op = REQ_OP_SECURE_ERASE;
  36. } else {
  37. if (!blk_queue_discard(q))
  38. return -EOPNOTSUPP;
  39. op = REQ_OP_DISCARD;
  40. }
  41. /* In case the discard granularity isn't set by buggy device driver */
  42. if (WARN_ON_ONCE(!q->limits.discard_granularity)) {
  43. char dev_name[BDEVNAME_SIZE];
  44. bdevname(bdev, dev_name);
  45. pr_err_ratelimited("%s: Error: discard_granularity is 0.\n", dev_name);
  46. return -EOPNOTSUPP;
  47. }
  48. bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  49. if ((sector | nr_sects) & bs_mask)
  50. return -EINVAL;
  51. if (!nr_sects)
  52. return -EINVAL;
  53. /* In case the discard request is in a partition */
  54. if (bdev_is_partition(bdev))
  55. part_offset = bdev->bd_part->start_sect;
  56. while (nr_sects) {
  57. sector_t granularity_aligned_lba, req_sects;
  58. sector_t sector_mapped = sector + part_offset;
  59. granularity_aligned_lba = round_up(sector_mapped,
  60. q->limits.discard_granularity >> SECTOR_SHIFT);
  61. /*
  62. * Check whether the discard bio starts at a discard_granularity
  63. * aligned LBA,
  64. * - If no: set (granularity_aligned_lba - sector_mapped) to
  65. * bi_size of the first split bio, then the second bio will
  66. * start at a discard_granularity aligned LBA on the device.
  67. * - If yes: use bio_aligned_discard_max_sectors() as the max
  68. * possible bi_size of the first split bio. Then when this bio
  69. * is split in device drive, the split ones are very probably
  70. * to be aligned to discard_granularity of the device's queue.
  71. */
  72. if (granularity_aligned_lba == sector_mapped)
  73. req_sects = min_t(sector_t, nr_sects,
  74. bio_aligned_discard_max_sectors(q));
  75. else
  76. req_sects = min_t(sector_t, nr_sects,
  77. granularity_aligned_lba - sector_mapped);
  78. WARN_ON_ONCE((req_sects << 9) > UINT_MAX);
  79. bio = blk_next_bio(bio, 0, gfp_mask);
  80. bio->bi_iter.bi_sector = sector;
  81. bio_set_dev(bio, bdev);
  82. bio_set_op_attrs(bio, op, 0);
  83. bio->bi_iter.bi_size = req_sects << 9;
  84. sector += req_sects;
  85. nr_sects -= req_sects;
  86. /*
  87. * We can loop for a long time in here, if someone does
  88. * full device discards (like mkfs). Be nice and allow
  89. * us to schedule out to avoid softlocking if preempt
  90. * is disabled.
  91. */
  92. cond_resched();
  93. }
  94. *biop = bio;
  95. return 0;
  96. }
  97. EXPORT_SYMBOL(__blkdev_issue_discard);
  98. /**
  99. * blkdev_issue_discard - queue a discard
  100. * @bdev: blockdev to issue discard for
  101. * @sector: start sector
  102. * @nr_sects: number of sectors to discard
  103. * @gfp_mask: memory allocation flags (for bio_alloc)
  104. * @flags: BLKDEV_DISCARD_* flags to control behaviour
  105. *
  106. * Description:
  107. * Issue a discard request for the sectors in question.
  108. */
  109. int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  110. sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
  111. {
  112. struct bio *bio = NULL;
  113. struct blk_plug plug;
  114. int ret;
  115. blk_start_plug(&plug);
  116. ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
  117. &bio);
  118. if (!ret && bio) {
  119. ret = submit_bio_wait(bio);
  120. if (ret == -EOPNOTSUPP)
  121. ret = 0;
  122. bio_put(bio);
  123. }
  124. blk_finish_plug(&plug);
  125. return ret;
  126. }
  127. EXPORT_SYMBOL(blkdev_issue_discard);
  128. /**
  129. * __blkdev_issue_write_same - generate number of bios with same page
  130. * @bdev: target blockdev
  131. * @sector: start sector
  132. * @nr_sects: number of sectors to write
  133. * @gfp_mask: memory allocation flags (for bio_alloc)
  134. * @page: page containing data to write
  135. * @biop: pointer to anchor bio
  136. *
  137. * Description:
  138. * Generate and issue number of bios(REQ_OP_WRITE_SAME) with same page.
  139. */
  140. static int __blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
  141. sector_t nr_sects, gfp_t gfp_mask, struct page *page,
  142. struct bio **biop)
  143. {
  144. struct request_queue *q = bdev_get_queue(bdev);
  145. unsigned int max_write_same_sectors;
  146. struct bio *bio = *biop;
  147. sector_t bs_mask;
  148. if (!q)
  149. return -ENXIO;
  150. if (bdev_read_only(bdev))
  151. return -EPERM;
  152. bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  153. if ((sector | nr_sects) & bs_mask)
  154. return -EINVAL;
  155. if (!bdev_write_same(bdev))
  156. return -EOPNOTSUPP;
  157. /* Ensure that max_write_same_sectors doesn't overflow bi_size */
  158. max_write_same_sectors = bio_allowed_max_sectors(q);
  159. while (nr_sects) {
  160. bio = blk_next_bio(bio, 1, gfp_mask);
  161. bio->bi_iter.bi_sector = sector;
  162. bio_set_dev(bio, bdev);
  163. bio->bi_vcnt = 1;
  164. bio->bi_io_vec->bv_page = page;
  165. bio->bi_io_vec->bv_offset = 0;
  166. bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
  167. bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
  168. if (nr_sects > max_write_same_sectors) {
  169. bio->bi_iter.bi_size = max_write_same_sectors << 9;
  170. nr_sects -= max_write_same_sectors;
  171. sector += max_write_same_sectors;
  172. } else {
  173. bio->bi_iter.bi_size = nr_sects << 9;
  174. nr_sects = 0;
  175. }
  176. cond_resched();
  177. }
  178. *biop = bio;
  179. return 0;
  180. }
  181. /**
  182. * blkdev_issue_write_same - queue a write same operation
  183. * @bdev: target blockdev
  184. * @sector: start sector
  185. * @nr_sects: number of sectors to write
  186. * @gfp_mask: memory allocation flags (for bio_alloc)
  187. * @page: page containing data
  188. *
  189. * Description:
  190. * Issue a write same request for the sectors in question.
  191. */
  192. int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
  193. sector_t nr_sects, gfp_t gfp_mask,
  194. struct page *page)
  195. {
  196. struct bio *bio = NULL;
  197. struct blk_plug plug;
  198. int ret;
  199. blk_start_plug(&plug);
  200. ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
  201. &bio);
  202. if (ret == 0 && bio) {
  203. ret = submit_bio_wait(bio);
  204. bio_put(bio);
  205. }
  206. blk_finish_plug(&plug);
  207. return ret;
  208. }
  209. EXPORT_SYMBOL(blkdev_issue_write_same);
  210. static int __blkdev_issue_write_zeroes(struct block_device *bdev,
  211. sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
  212. struct bio **biop, unsigned flags)
  213. {
  214. struct bio *bio = *biop;
  215. unsigned int max_write_zeroes_sectors;
  216. struct request_queue *q = bdev_get_queue(bdev);
  217. if (!q)
  218. return -ENXIO;
  219. if (bdev_read_only(bdev))
  220. return -EPERM;
  221. /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
  222. max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
  223. if (max_write_zeroes_sectors == 0)
  224. return -EOPNOTSUPP;
  225. while (nr_sects) {
  226. bio = blk_next_bio(bio, 0, gfp_mask);
  227. bio->bi_iter.bi_sector = sector;
  228. bio_set_dev(bio, bdev);
  229. bio->bi_opf = REQ_OP_WRITE_ZEROES;
  230. if (flags & BLKDEV_ZERO_NOUNMAP)
  231. bio->bi_opf |= REQ_NOUNMAP;
  232. if (nr_sects > max_write_zeroes_sectors) {
  233. bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
  234. nr_sects -= max_write_zeroes_sectors;
  235. sector += max_write_zeroes_sectors;
  236. } else {
  237. bio->bi_iter.bi_size = nr_sects << 9;
  238. nr_sects = 0;
  239. }
  240. cond_resched();
  241. }
  242. *biop = bio;
  243. return 0;
  244. }
  245. /*
  246. * Convert a number of 512B sectors to a number of pages.
  247. * The result is limited to a number of pages that can fit into a BIO.
  248. * Also make sure that the result is always at least 1 (page) for the cases
  249. * where nr_sects is lower than the number of sectors in a page.
  250. */
  251. static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
  252. {
  253. sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
  254. return min(pages, (sector_t)BIO_MAX_PAGES);
  255. }
  256. static int __blkdev_issue_zero_pages(struct block_device *bdev,
  257. sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
  258. struct bio **biop)
  259. {
  260. struct request_queue *q = bdev_get_queue(bdev);
  261. struct bio *bio = *biop;
  262. int bi_size = 0;
  263. unsigned int sz;
  264. if (!q)
  265. return -ENXIO;
  266. if (bdev_read_only(bdev))
  267. return -EPERM;
  268. while (nr_sects != 0) {
  269. bio = blk_next_bio(bio, __blkdev_sectors_to_bio_pages(nr_sects),
  270. gfp_mask);
  271. bio->bi_iter.bi_sector = sector;
  272. bio_set_dev(bio, bdev);
  273. bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
  274. while (nr_sects != 0) {
  275. sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
  276. bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
  277. nr_sects -= bi_size >> 9;
  278. sector += bi_size >> 9;
  279. if (bi_size < sz)
  280. break;
  281. }
  282. cond_resched();
  283. }
  284. *biop = bio;
  285. return 0;
  286. }
  287. /**
  288. * __blkdev_issue_zeroout - generate number of zero filed write bios
  289. * @bdev: blockdev to issue
  290. * @sector: start sector
  291. * @nr_sects: number of sectors to write
  292. * @gfp_mask: memory allocation flags (for bio_alloc)
  293. * @biop: pointer to anchor bio
  294. * @flags: controls detailed behavior
  295. *
  296. * Description:
  297. * Zero-fill a block range, either using hardware offload or by explicitly
  298. * writing zeroes to the device.
  299. *
  300. * If a device is using logical block provisioning, the underlying space will
  301. * not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
  302. *
  303. * If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
  304. * -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
  305. */
  306. int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
  307. sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
  308. unsigned flags)
  309. {
  310. int ret;
  311. sector_t bs_mask;
  312. bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  313. if ((sector | nr_sects) & bs_mask)
  314. return -EINVAL;
  315. ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
  316. biop, flags);
  317. if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
  318. return ret;
  319. return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
  320. biop);
  321. }
  322. EXPORT_SYMBOL(__blkdev_issue_zeroout);
  323. /**
  324. * blkdev_issue_zeroout - zero-fill a block range
  325. * @bdev: blockdev to write
  326. * @sector: start sector
  327. * @nr_sects: number of sectors to write
  328. * @gfp_mask: memory allocation flags (for bio_alloc)
  329. * @flags: controls detailed behavior
  330. *
  331. * Description:
  332. * Zero-fill a block range, either using hardware offload or by explicitly
  333. * writing zeroes to the device. See __blkdev_issue_zeroout() for the
  334. * valid values for %flags.
  335. */
  336. int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
  337. sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
  338. {
  339. int ret = 0;
  340. sector_t bs_mask;
  341. struct bio *bio;
  342. struct blk_plug plug;
  343. bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
  344. bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  345. if ((sector | nr_sects) & bs_mask)
  346. return -EINVAL;
  347. retry:
  348. bio = NULL;
  349. blk_start_plug(&plug);
  350. if (try_write_zeroes) {
  351. ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
  352. gfp_mask, &bio, flags);
  353. } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
  354. ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
  355. gfp_mask, &bio);
  356. } else {
  357. /* No zeroing offload support */
  358. ret = -EOPNOTSUPP;
  359. }
  360. if (ret == 0 && bio) {
  361. ret = submit_bio_wait(bio);
  362. bio_put(bio);
  363. }
  364. blk_finish_plug(&plug);
  365. if (ret && try_write_zeroes) {
  366. if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
  367. try_write_zeroes = false;
  368. goto retry;
  369. }
  370. if (!bdev_write_zeroes_sectors(bdev)) {
  371. /*
  372. * Zeroing offload support was indicated, but the
  373. * device reported ILLEGAL REQUEST (for some devices
  374. * there is no non-destructive way to verify whether
  375. * WRITE ZEROES is actually supported).
  376. */
  377. ret = -EOPNOTSUPP;
  378. }
  379. }
  380. return ret;
  381. }
  382. EXPORT_SYMBOL(blkdev_issue_zeroout);