blk.c 8.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343
  1. // SPDX-License-Identifier: GPL-2.0-only
  2. /*
  3. * NVDIMM Block Window Driver
  4. * Copyright (c) 2014, Intel Corporation.
  5. */
  6. #include <linux/blkdev.h>
  7. #include <linux/fs.h>
  8. #include <linux/genhd.h>
  9. #include <linux/module.h>
  10. #include <linux/moduleparam.h>
  11. #include <linux/nd.h>
  12. #include <linux/sizes.h>
  13. #include "nd.h"
  14. static u32 nsblk_meta_size(struct nd_namespace_blk *nsblk)
  15. {
  16. return nsblk->lbasize - ((nsblk->lbasize >= 4096) ? 4096 : 512);
  17. }
  18. static u32 nsblk_internal_lbasize(struct nd_namespace_blk *nsblk)
  19. {
  20. return roundup(nsblk->lbasize, INT_LBASIZE_ALIGNMENT);
  21. }
  22. static u32 nsblk_sector_size(struct nd_namespace_blk *nsblk)
  23. {
  24. return nsblk->lbasize - nsblk_meta_size(nsblk);
  25. }
  26. static resource_size_t to_dev_offset(struct nd_namespace_blk *nsblk,
  27. resource_size_t ns_offset, unsigned int len)
  28. {
  29. int i;
  30. for (i = 0; i < nsblk->num_resources; i++) {
  31. if (ns_offset < resource_size(nsblk->res[i])) {
  32. if (ns_offset + len > resource_size(nsblk->res[i])) {
  33. dev_WARN_ONCE(&nsblk->common.dev, 1,
  34. "illegal request\n");
  35. return SIZE_MAX;
  36. }
  37. return nsblk->res[i]->start + ns_offset;
  38. }
  39. ns_offset -= resource_size(nsblk->res[i]);
  40. }
  41. dev_WARN_ONCE(&nsblk->common.dev, 1, "request out of range\n");
  42. return SIZE_MAX;
  43. }
  44. static struct nd_blk_region *to_ndbr(struct nd_namespace_blk *nsblk)
  45. {
  46. struct nd_region *nd_region;
  47. struct device *parent;
  48. parent = nsblk->common.dev.parent;
  49. nd_region = container_of(parent, struct nd_region, dev);
  50. return container_of(nd_region, struct nd_blk_region, nd_region);
  51. }
  52. #ifdef CONFIG_BLK_DEV_INTEGRITY
  53. static int nd_blk_rw_integrity(struct nd_namespace_blk *nsblk,
  54. struct bio_integrity_payload *bip, u64 lba, int rw)
  55. {
  56. struct nd_blk_region *ndbr = to_ndbr(nsblk);
  57. unsigned int len = nsblk_meta_size(nsblk);
  58. resource_size_t dev_offset, ns_offset;
  59. u32 internal_lbasize, sector_size;
  60. int err = 0;
  61. internal_lbasize = nsblk_internal_lbasize(nsblk);
  62. sector_size = nsblk_sector_size(nsblk);
  63. ns_offset = lba * internal_lbasize + sector_size;
  64. dev_offset = to_dev_offset(nsblk, ns_offset, len);
  65. if (dev_offset == SIZE_MAX)
  66. return -EIO;
  67. while (len) {
  68. unsigned int cur_len;
  69. struct bio_vec bv;
  70. void *iobuf;
  71. bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
  72. /*
  73. * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
  74. * .bv_offset already adjusted for iter->bi_bvec_done, and we
  75. * can use those directly
  76. */
  77. cur_len = min(len, bv.bv_len);
  78. iobuf = kmap_atomic(bv.bv_page);
  79. err = ndbr->do_io(ndbr, dev_offset, iobuf + bv.bv_offset,
  80. cur_len, rw);
  81. kunmap_atomic(iobuf);
  82. if (err)
  83. return err;
  84. len -= cur_len;
  85. dev_offset += cur_len;
  86. if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
  87. return -EIO;
  88. }
  89. return err;
  90. }
  91. #else /* CONFIG_BLK_DEV_INTEGRITY */
  92. static int nd_blk_rw_integrity(struct nd_namespace_blk *nsblk,
  93. struct bio_integrity_payload *bip, u64 lba, int rw)
  94. {
  95. return 0;
  96. }
  97. #endif
  98. static int nsblk_do_bvec(struct nd_namespace_blk *nsblk,
  99. struct bio_integrity_payload *bip, struct page *page,
  100. unsigned int len, unsigned int off, int rw, sector_t sector)
  101. {
  102. struct nd_blk_region *ndbr = to_ndbr(nsblk);
  103. resource_size_t dev_offset, ns_offset;
  104. u32 internal_lbasize, sector_size;
  105. int err = 0;
  106. void *iobuf;
  107. u64 lba;
  108. internal_lbasize = nsblk_internal_lbasize(nsblk);
  109. sector_size = nsblk_sector_size(nsblk);
  110. while (len) {
  111. unsigned int cur_len;
  112. /*
  113. * If we don't have an integrity payload, we don't have to
  114. * split the bvec into sectors, as this would cause unnecessary
  115. * Block Window setup/move steps. the do_io routine is capable
  116. * of handling len <= PAGE_SIZE.
  117. */
  118. cur_len = bip ? min(len, sector_size) : len;
  119. lba = div_u64(sector << SECTOR_SHIFT, sector_size);
  120. ns_offset = lba * internal_lbasize;
  121. dev_offset = to_dev_offset(nsblk, ns_offset, cur_len);
  122. if (dev_offset == SIZE_MAX)
  123. return -EIO;
  124. iobuf = kmap_atomic(page);
  125. err = ndbr->do_io(ndbr, dev_offset, iobuf + off, cur_len, rw);
  126. kunmap_atomic(iobuf);
  127. if (err)
  128. return err;
  129. if (bip) {
  130. err = nd_blk_rw_integrity(nsblk, bip, lba, rw);
  131. if (err)
  132. return err;
  133. }
  134. len -= cur_len;
  135. off += cur_len;
  136. sector += sector_size >> SECTOR_SHIFT;
  137. }
  138. return err;
  139. }
  140. static blk_qc_t nd_blk_submit_bio(struct bio *bio)
  141. {
  142. struct bio_integrity_payload *bip;
  143. struct nd_namespace_blk *nsblk = bio->bi_disk->private_data;
  144. struct bvec_iter iter;
  145. unsigned long start;
  146. struct bio_vec bvec;
  147. int err = 0, rw;
  148. bool do_acct;
  149. if (!bio_integrity_prep(bio))
  150. return BLK_QC_T_NONE;
  151. bip = bio_integrity(bio);
  152. rw = bio_data_dir(bio);
  153. do_acct = blk_queue_io_stat(bio->bi_disk->queue);
  154. if (do_acct)
  155. start = bio_start_io_acct(bio);
  156. bio_for_each_segment(bvec, bio, iter) {
  157. unsigned int len = bvec.bv_len;
  158. BUG_ON(len > PAGE_SIZE);
  159. err = nsblk_do_bvec(nsblk, bip, bvec.bv_page, len,
  160. bvec.bv_offset, rw, iter.bi_sector);
  161. if (err) {
  162. dev_dbg(&nsblk->common.dev,
  163. "io error in %s sector %lld, len %d,\n",
  164. (rw == READ) ? "READ" : "WRITE",
  165. (unsigned long long) iter.bi_sector, len);
  166. bio->bi_status = errno_to_blk_status(err);
  167. break;
  168. }
  169. }
  170. if (do_acct)
  171. bio_end_io_acct(bio, start);
  172. bio_endio(bio);
  173. return BLK_QC_T_NONE;
  174. }
  175. static int nsblk_rw_bytes(struct nd_namespace_common *ndns,
  176. resource_size_t offset, void *iobuf, size_t n, int rw,
  177. unsigned long flags)
  178. {
  179. struct nd_namespace_blk *nsblk = to_nd_namespace_blk(&ndns->dev);
  180. struct nd_blk_region *ndbr = to_ndbr(nsblk);
  181. resource_size_t dev_offset;
  182. dev_offset = to_dev_offset(nsblk, offset, n);
  183. if (unlikely(offset + n > nsblk->size)) {
  184. dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
  185. return -EFAULT;
  186. }
  187. if (dev_offset == SIZE_MAX)
  188. return -EIO;
  189. return ndbr->do_io(ndbr, dev_offset, iobuf, n, rw);
  190. }
  191. static const struct block_device_operations nd_blk_fops = {
  192. .owner = THIS_MODULE,
  193. .submit_bio = nd_blk_submit_bio,
  194. };
  195. static void nd_blk_release_queue(void *q)
  196. {
  197. blk_cleanup_queue(q);
  198. }
  199. static void nd_blk_release_disk(void *disk)
  200. {
  201. del_gendisk(disk);
  202. put_disk(disk);
  203. }
  204. static int nsblk_attach_disk(struct nd_namespace_blk *nsblk)
  205. {
  206. struct device *dev = &nsblk->common.dev;
  207. resource_size_t available_disk_size;
  208. struct request_queue *q;
  209. struct gendisk *disk;
  210. u64 internal_nlba;
  211. internal_nlba = div_u64(nsblk->size, nsblk_internal_lbasize(nsblk));
  212. available_disk_size = internal_nlba * nsblk_sector_size(nsblk);
  213. q = blk_alloc_queue(NUMA_NO_NODE);
  214. if (!q)
  215. return -ENOMEM;
  216. if (devm_add_action_or_reset(dev, nd_blk_release_queue, q))
  217. return -ENOMEM;
  218. blk_queue_max_hw_sectors(q, UINT_MAX);
  219. blk_queue_logical_block_size(q, nsblk_sector_size(nsblk));
  220. blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
  221. disk = alloc_disk(0);
  222. if (!disk)
  223. return -ENOMEM;
  224. disk->first_minor = 0;
  225. disk->fops = &nd_blk_fops;
  226. disk->queue = q;
  227. disk->flags = GENHD_FL_EXT_DEVT;
  228. disk->private_data = nsblk;
  229. nvdimm_namespace_disk_name(&nsblk->common, disk->disk_name);
  230. if (devm_add_action_or_reset(dev, nd_blk_release_disk, disk))
  231. return -ENOMEM;
  232. if (nsblk_meta_size(nsblk)) {
  233. int rc = nd_integrity_init(disk, nsblk_meta_size(nsblk));
  234. if (rc)
  235. return rc;
  236. }
  237. set_capacity(disk, available_disk_size >> SECTOR_SHIFT);
  238. device_add_disk(dev, disk, NULL);
  239. nvdimm_check_and_set_ro(disk);
  240. return 0;
  241. }
  242. static int nd_blk_probe(struct device *dev)
  243. {
  244. struct nd_namespace_common *ndns;
  245. struct nd_namespace_blk *nsblk;
  246. ndns = nvdimm_namespace_common_probe(dev);
  247. if (IS_ERR(ndns))
  248. return PTR_ERR(ndns);
  249. nsblk = to_nd_namespace_blk(&ndns->dev);
  250. nsblk->size = nvdimm_namespace_capacity(ndns);
  251. dev_set_drvdata(dev, nsblk);
  252. ndns->rw_bytes = nsblk_rw_bytes;
  253. if (is_nd_btt(dev))
  254. return nvdimm_namespace_attach_btt(ndns);
  255. else if (nd_btt_probe(dev, ndns) == 0) {
  256. /* we'll come back as btt-blk */
  257. return -ENXIO;
  258. } else
  259. return nsblk_attach_disk(nsblk);
  260. }
  261. static int nd_blk_remove(struct device *dev)
  262. {
  263. if (is_nd_btt(dev))
  264. nvdimm_namespace_detach_btt(to_nd_btt(dev));
  265. return 0;
  266. }
  267. static struct nd_device_driver nd_blk_driver = {
  268. .probe = nd_blk_probe,
  269. .remove = nd_blk_remove,
  270. .drv = {
  271. .name = "nd_blk",
  272. },
  273. .type = ND_DRIVER_NAMESPACE_BLK,
  274. };
  275. static int __init nd_blk_init(void)
  276. {
  277. return nd_driver_register(&nd_blk_driver);
  278. }
  279. static void __exit nd_blk_exit(void)
  280. {
  281. driver_unregister(&nd_blk_driver.drv);
  282. }
  283. MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
  284. MODULE_LICENSE("GPL v2");
  285. MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_BLK);
  286. module_init(nd_blk_init);
  287. module_exit(nd_blk_exit);