bio.c 39 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654
  1. /*
  2. * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
  3. *
  4. * This program is free software; you can redistribute it and/or modify
  5. * it under the terms of the GNU General Public License version 2 as
  6. * published by the Free Software Foundation.
  7. *
  8. * This program is distributed in the hope that it will be useful,
  9. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  11. * GNU General Public License for more details.
  12. *
  13. * You should have received a copy of the GNU General Public Licens
  14. * along with this program; if not, write to the Free Software
  15. * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
  16. *
  17. */
  18. #include <linux/mm.h>
  19. #include <linux/swap.h>
  20. #include <linux/bio.h>
  21. #include <linux/blkdev.h>
  22. #include <linux/slab.h>
  23. #include <linux/init.h>
  24. #include <linux/kernel.h>
  25. #include <linux/module.h>
  26. #include <linux/mempool.h>
  27. #include <linux/workqueue.h>
  28. #include <linux/blktrace_api.h>
  29. #include <trace/block.h>
  30. #include <scsi/sg.h> /* for struct sg_iovec */
  31. DEFINE_TRACE(block_split);
  32. /*
  33. * Test patch to inline a certain number of bi_io_vec's inside the bio
  34. * itself, to shrink a bio data allocation from two mempool calls to one
  35. */
  36. #define BIO_INLINE_VECS 4
  37. static mempool_t *bio_split_pool __read_mostly;
  38. /*
  39. * if you change this list, also change bvec_alloc or things will
  40. * break badly! cannot be bigger than what you can fit into an
  41. * unsigned short
  42. */
  43. #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
  44. struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
  45. BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
  46. };
  47. #undef BV
  48. /*
  49. * fs_bio_set is the bio_set containing bio and iovec memory pools used by
  50. * IO code that does not need private memory pools.
  51. */
  52. struct bio_set *fs_bio_set;
  53. /*
  54. * Our slab pool management
  55. */
  56. struct bio_slab {
  57. struct kmem_cache *slab;
  58. unsigned int slab_ref;
  59. unsigned int slab_size;
  60. char name[8];
  61. };
  62. static DEFINE_MUTEX(bio_slab_lock);
  63. static struct bio_slab *bio_slabs;
  64. static unsigned int bio_slab_nr, bio_slab_max;
  65. static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
  66. {
  67. unsigned int sz = sizeof(struct bio) + extra_size;
  68. struct kmem_cache *slab = NULL;
  69. struct bio_slab *bslab;
  70. unsigned int i, entry = -1;
  71. mutex_lock(&bio_slab_lock);
  72. i = 0;
  73. while (i < bio_slab_nr) {
  74. struct bio_slab *bslab = &bio_slabs[i];
  75. if (!bslab->slab && entry == -1)
  76. entry = i;
  77. else if (bslab->slab_size == sz) {
  78. slab = bslab->slab;
  79. bslab->slab_ref++;
  80. break;
  81. }
  82. i++;
  83. }
  84. if (slab)
  85. goto out_unlock;
  86. if (bio_slab_nr == bio_slab_max && entry == -1) {
  87. bio_slab_max <<= 1;
  88. bio_slabs = krealloc(bio_slabs,
  89. bio_slab_max * sizeof(struct bio_slab),
  90. GFP_KERNEL);
  91. if (!bio_slabs)
  92. goto out_unlock;
  93. }
  94. if (entry == -1)
  95. entry = bio_slab_nr++;
  96. bslab = &bio_slabs[entry];
  97. snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
  98. slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
  99. if (!slab)
  100. goto out_unlock;
  101. printk("bio: create slab <%s> at %d\n", bslab->name, entry);
  102. bslab->slab = slab;
  103. bslab->slab_ref = 1;
  104. bslab->slab_size = sz;
  105. out_unlock:
  106. mutex_unlock(&bio_slab_lock);
  107. return slab;
  108. }
  109. static void bio_put_slab(struct bio_set *bs)
  110. {
  111. struct bio_slab *bslab = NULL;
  112. unsigned int i;
  113. mutex_lock(&bio_slab_lock);
  114. for (i = 0; i < bio_slab_nr; i++) {
  115. if (bs->bio_slab == bio_slabs[i].slab) {
  116. bslab = &bio_slabs[i];
  117. break;
  118. }
  119. }
  120. if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
  121. goto out;
  122. WARN_ON(!bslab->slab_ref);
  123. if (--bslab->slab_ref)
  124. goto out;
  125. kmem_cache_destroy(bslab->slab);
  126. bslab->slab = NULL;
  127. out:
  128. mutex_unlock(&bio_slab_lock);
  129. }
  130. unsigned int bvec_nr_vecs(unsigned short idx)
  131. {
  132. return bvec_slabs[idx].nr_vecs;
  133. }
  134. void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
  135. {
  136. BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
  137. if (idx == BIOVEC_MAX_IDX)
  138. mempool_free(bv, bs->bvec_pool);
  139. else {
  140. struct biovec_slab *bvs = bvec_slabs + idx;
  141. kmem_cache_free(bvs->slab, bv);
  142. }
  143. }
  144. struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
  145. struct bio_set *bs)
  146. {
  147. struct bio_vec *bvl;
  148. /*
  149. * see comment near bvec_array define!
  150. */
  151. switch (nr) {
  152. case 1:
  153. *idx = 0;
  154. break;
  155. case 2 ... 4:
  156. *idx = 1;
  157. break;
  158. case 5 ... 16:
  159. *idx = 2;
  160. break;
  161. case 17 ... 64:
  162. *idx = 3;
  163. break;
  164. case 65 ... 128:
  165. *idx = 4;
  166. break;
  167. case 129 ... BIO_MAX_PAGES:
  168. *idx = 5;
  169. break;
  170. default:
  171. return NULL;
  172. }
  173. /*
  174. * idx now points to the pool we want to allocate from. only the
  175. * 1-vec entry pool is mempool backed.
  176. */
  177. if (*idx == BIOVEC_MAX_IDX) {
  178. fallback:
  179. bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
  180. } else {
  181. struct biovec_slab *bvs = bvec_slabs + *idx;
  182. gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
  183. /*
  184. * Make this allocation restricted and don't dump info on
  185. * allocation failures, since we'll fallback to the mempool
  186. * in case of failure.
  187. */
  188. __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
  189. /*
  190. * Try a slab allocation. If this fails and __GFP_WAIT
  191. * is set, retry with the 1-entry mempool
  192. */
  193. bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
  194. if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
  195. *idx = BIOVEC_MAX_IDX;
  196. goto fallback;
  197. }
  198. }
  199. return bvl;
  200. }
  201. void bio_free(struct bio *bio, struct bio_set *bs)
  202. {
  203. void *p;
  204. if (bio_has_allocated_vec(bio))
  205. bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
  206. if (bio_integrity(bio))
  207. bio_integrity_free(bio);
  208. /*
  209. * If we have front padding, adjust the bio pointer before freeing
  210. */
  211. p = bio;
  212. if (bs->front_pad)
  213. p -= bs->front_pad;
  214. mempool_free(p, bs->bio_pool);
  215. }
  216. void bio_init(struct bio *bio)
  217. {
  218. memset(bio, 0, sizeof(*bio));
  219. bio->bi_flags = 1 << BIO_UPTODATE;
  220. bio->bi_comp_cpu = -1;
  221. atomic_set(&bio->bi_cnt, 1);
  222. }
  223. /**
  224. * bio_alloc_bioset - allocate a bio for I/O
  225. * @gfp_mask: the GFP_ mask given to the slab allocator
  226. * @nr_iovecs: number of iovecs to pre-allocate
  227. * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
  228. *
  229. * Description:
  230. * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
  231. * If %__GFP_WAIT is set then we will block on the internal pool waiting
  232. * for a &struct bio to become free. If a %NULL @bs is passed in, we will
  233. * fall back to just using @kmalloc to allocate the required memory.
  234. *
  235. * Note that the caller must set ->bi_destructor on succesful return
  236. * of a bio, to do the appropriate freeing of the bio once the reference
  237. * count drops to zero.
  238. **/
  239. struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
  240. {
  241. unsigned long idx = BIO_POOL_NONE;
  242. struct bio_vec *bvl = NULL;
  243. struct bio *bio;
  244. void *p;
  245. p = mempool_alloc(bs->bio_pool, gfp_mask);
  246. if (unlikely(!p))
  247. return NULL;
  248. bio = p + bs->front_pad;
  249. bio_init(bio);
  250. if (unlikely(!nr_iovecs))
  251. goto out_set;
  252. if (nr_iovecs <= BIO_INLINE_VECS) {
  253. bvl = bio->bi_inline_vecs;
  254. nr_iovecs = BIO_INLINE_VECS;
  255. } else {
  256. bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
  257. if (unlikely(!bvl))
  258. goto err_free;
  259. nr_iovecs = bvec_nr_vecs(idx);
  260. }
  261. out_set:
  262. bio->bi_flags |= idx << BIO_POOL_OFFSET;
  263. bio->bi_max_vecs = nr_iovecs;
  264. bio->bi_io_vec = bvl;
  265. return bio;
  266. err_free:
  267. mempool_free(p, bs->bio_pool);
  268. return NULL;
  269. }
  270. static void bio_fs_destructor(struct bio *bio)
  271. {
  272. bio_free(bio, fs_bio_set);
  273. }
  274. /**
  275. * bio_alloc - allocate a new bio, memory pool backed
  276. * @gfp_mask: allocation mask to use
  277. * @nr_iovecs: number of iovecs
  278. *
  279. * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask
  280. * contains __GFP_WAIT, the allocation is guaranteed to succeed.
  281. *
  282. * RETURNS:
  283. * Pointer to new bio on success, NULL on failure.
  284. */
  285. struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
  286. {
  287. struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
  288. if (bio)
  289. bio->bi_destructor = bio_fs_destructor;
  290. return bio;
  291. }
  292. static void bio_kmalloc_destructor(struct bio *bio)
  293. {
  294. if (bio_integrity(bio))
  295. bio_integrity_free(bio);
  296. kfree(bio);
  297. }
  298. /**
  299. * bio_alloc - allocate a bio for I/O
  300. * @gfp_mask: the GFP_ mask given to the slab allocator
  301. * @nr_iovecs: number of iovecs to pre-allocate
  302. *
  303. * Description:
  304. * bio_alloc will allocate a bio and associated bio_vec array that can hold
  305. * at least @nr_iovecs entries. Allocations will be done from the
  306. * fs_bio_set. Also see @bio_alloc_bioset.
  307. *
  308. * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
  309. * a bio. This is due to the mempool guarantees. To make this work, callers
  310. * must never allocate more than 1 bio at the time from this pool. Callers
  311. * that need to allocate more than 1 bio must always submit the previously
  312. * allocate bio for IO before attempting to allocate a new one. Failure to
  313. * do so can cause livelocks under memory pressure.
  314. *
  315. **/
  316. struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
  317. {
  318. struct bio *bio;
  319. bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
  320. gfp_mask);
  321. if (unlikely(!bio))
  322. return NULL;
  323. bio_init(bio);
  324. bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
  325. bio->bi_max_vecs = nr_iovecs;
  326. bio->bi_io_vec = bio->bi_inline_vecs;
  327. bio->bi_destructor = bio_kmalloc_destructor;
  328. return bio;
  329. }
  330. void zero_fill_bio(struct bio *bio)
  331. {
  332. unsigned long flags;
  333. struct bio_vec *bv;
  334. int i;
  335. bio_for_each_segment(bv, bio, i) {
  336. char *data = bvec_kmap_irq(bv, &flags);
  337. memset(data, 0, bv->bv_len);
  338. flush_dcache_page(bv->bv_page);
  339. bvec_kunmap_irq(data, &flags);
  340. }
  341. }
  342. EXPORT_SYMBOL(zero_fill_bio);
  343. /**
  344. * bio_put - release a reference to a bio
  345. * @bio: bio to release reference to
  346. *
  347. * Description:
  348. * Put a reference to a &struct bio, either one you have gotten with
  349. * bio_alloc or bio_get. The last put of a bio will free it.
  350. **/
  351. void bio_put(struct bio *bio)
  352. {
  353. BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
  354. /*
  355. * last put frees it
  356. */
  357. if (atomic_dec_and_test(&bio->bi_cnt)) {
  358. bio->bi_next = NULL;
  359. bio->bi_destructor(bio);
  360. }
  361. }
  362. inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
  363. {
  364. if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
  365. blk_recount_segments(q, bio);
  366. return bio->bi_phys_segments;
  367. }
  368. /**
  369. * __bio_clone - clone a bio
  370. * @bio: destination bio
  371. * @bio_src: bio to clone
  372. *
  373. * Clone a &bio. Caller will own the returned bio, but not
  374. * the actual data it points to. Reference count of returned
  375. * bio will be one.
  376. */
  377. void __bio_clone(struct bio *bio, struct bio *bio_src)
  378. {
  379. memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
  380. bio_src->bi_max_vecs * sizeof(struct bio_vec));
  381. /*
  382. * most users will be overriding ->bi_bdev with a new target,
  383. * so we don't set nor calculate new physical/hw segment counts here
  384. */
  385. bio->bi_sector = bio_src->bi_sector;
  386. bio->bi_bdev = bio_src->bi_bdev;
  387. bio->bi_flags |= 1 << BIO_CLONED;
  388. bio->bi_rw = bio_src->bi_rw;
  389. bio->bi_vcnt = bio_src->bi_vcnt;
  390. bio->bi_size = bio_src->bi_size;
  391. bio->bi_idx = bio_src->bi_idx;
  392. }
  393. /**
  394. * bio_clone - clone a bio
  395. * @bio: bio to clone
  396. * @gfp_mask: allocation priority
  397. *
  398. * Like __bio_clone, only also allocates the returned bio
  399. */
  400. struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
  401. {
  402. struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
  403. if (!b)
  404. return NULL;
  405. b->bi_destructor = bio_fs_destructor;
  406. __bio_clone(b, bio);
  407. if (bio_integrity(bio)) {
  408. int ret;
  409. ret = bio_integrity_clone(b, bio, gfp_mask);
  410. if (ret < 0) {
  411. bio_put(b);
  412. return NULL;
  413. }
  414. }
  415. return b;
  416. }
  417. /**
  418. * bio_get_nr_vecs - return approx number of vecs
  419. * @bdev: I/O target
  420. *
  421. * Return the approximate number of pages we can send to this target.
  422. * There's no guarantee that you will be able to fit this number of pages
  423. * into a bio, it does not account for dynamic restrictions that vary
  424. * on offset.
  425. */
  426. int bio_get_nr_vecs(struct block_device *bdev)
  427. {
  428. struct request_queue *q = bdev_get_queue(bdev);
  429. int nr_pages;
  430. nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
  431. if (nr_pages > q->max_phys_segments)
  432. nr_pages = q->max_phys_segments;
  433. if (nr_pages > q->max_hw_segments)
  434. nr_pages = q->max_hw_segments;
  435. return nr_pages;
  436. }
  437. static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
  438. *page, unsigned int len, unsigned int offset,
  439. unsigned short max_sectors)
  440. {
  441. int retried_segments = 0;
  442. struct bio_vec *bvec;
  443. /*
  444. * cloned bio must not modify vec list
  445. */
  446. if (unlikely(bio_flagged(bio, BIO_CLONED)))
  447. return 0;
  448. if (((bio->bi_size + len) >> 9) > max_sectors)
  449. return 0;
  450. /*
  451. * For filesystems with a blocksize smaller than the pagesize
  452. * we will often be called with the same page as last time and
  453. * a consecutive offset. Optimize this special case.
  454. */
  455. if (bio->bi_vcnt > 0) {
  456. struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
  457. if (page == prev->bv_page &&
  458. offset == prev->bv_offset + prev->bv_len) {
  459. prev->bv_len += len;
  460. if (q->merge_bvec_fn) {
  461. struct bvec_merge_data bvm = {
  462. .bi_bdev = bio->bi_bdev,
  463. .bi_sector = bio->bi_sector,
  464. .bi_size = bio->bi_size,
  465. .bi_rw = bio->bi_rw,
  466. };
  467. if (q->merge_bvec_fn(q, &bvm, prev) < len) {
  468. prev->bv_len -= len;
  469. return 0;
  470. }
  471. }
  472. goto done;
  473. }
  474. }
  475. if (bio->bi_vcnt >= bio->bi_max_vecs)
  476. return 0;
  477. /*
  478. * we might lose a segment or two here, but rather that than
  479. * make this too complex.
  480. */
  481. while (bio->bi_phys_segments >= q->max_phys_segments
  482. || bio->bi_phys_segments >= q->max_hw_segments) {
  483. if (retried_segments)
  484. return 0;
  485. retried_segments = 1;
  486. blk_recount_segments(q, bio);
  487. }
  488. /*
  489. * setup the new entry, we might clear it again later if we
  490. * cannot add the page
  491. */
  492. bvec = &bio->bi_io_vec[bio->bi_vcnt];
  493. bvec->bv_page = page;
  494. bvec->bv_len = len;
  495. bvec->bv_offset = offset;
  496. /*
  497. * if queue has other restrictions (eg varying max sector size
  498. * depending on offset), it can specify a merge_bvec_fn in the
  499. * queue to get further control
  500. */
  501. if (q->merge_bvec_fn) {
  502. struct bvec_merge_data bvm = {
  503. .bi_bdev = bio->bi_bdev,
  504. .bi_sector = bio->bi_sector,
  505. .bi_size = bio->bi_size,
  506. .bi_rw = bio->bi_rw,
  507. };
  508. /*
  509. * merge_bvec_fn() returns number of bytes it can accept
  510. * at this offset
  511. */
  512. if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
  513. bvec->bv_page = NULL;
  514. bvec->bv_len = 0;
  515. bvec->bv_offset = 0;
  516. return 0;
  517. }
  518. }
  519. /* If we may be able to merge these biovecs, force a recount */
  520. if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
  521. bio->bi_flags &= ~(1 << BIO_SEG_VALID);
  522. bio->bi_vcnt++;
  523. bio->bi_phys_segments++;
  524. done:
  525. bio->bi_size += len;
  526. return len;
  527. }
  528. /**
  529. * bio_add_pc_page - attempt to add page to bio
  530. * @q: the target queue
  531. * @bio: destination bio
  532. * @page: page to add
  533. * @len: vec entry length
  534. * @offset: vec entry offset
  535. *
  536. * Attempt to add a page to the bio_vec maplist. This can fail for a
  537. * number of reasons, such as the bio being full or target block
  538. * device limitations. The target block device must allow bio's
  539. * smaller than PAGE_SIZE, so it is always possible to add a single
  540. * page to an empty bio. This should only be used by REQ_PC bios.
  541. */
  542. int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
  543. unsigned int len, unsigned int offset)
  544. {
  545. return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
  546. }
  547. /**
  548. * bio_add_page - attempt to add page to bio
  549. * @bio: destination bio
  550. * @page: page to add
  551. * @len: vec entry length
  552. * @offset: vec entry offset
  553. *
  554. * Attempt to add a page to the bio_vec maplist. This can fail for a
  555. * number of reasons, such as the bio being full or target block
  556. * device limitations. The target block device must allow bio's
  557. * smaller than PAGE_SIZE, so it is always possible to add a single
  558. * page to an empty bio.
  559. */
  560. int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
  561. unsigned int offset)
  562. {
  563. struct request_queue *q = bdev_get_queue(bio->bi_bdev);
  564. return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
  565. }
  566. struct bio_map_data {
  567. struct bio_vec *iovecs;
  568. struct sg_iovec *sgvecs;
  569. int nr_sgvecs;
  570. int is_our_pages;
  571. };
  572. static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
  573. struct sg_iovec *iov, int iov_count,
  574. int is_our_pages)
  575. {
  576. memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
  577. memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
  578. bmd->nr_sgvecs = iov_count;
  579. bmd->is_our_pages = is_our_pages;
  580. bio->bi_private = bmd;
  581. }
  582. static void bio_free_map_data(struct bio_map_data *bmd)
  583. {
  584. kfree(bmd->iovecs);
  585. kfree(bmd->sgvecs);
  586. kfree(bmd);
  587. }
  588. static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
  589. gfp_t gfp_mask)
  590. {
  591. struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
  592. if (!bmd)
  593. return NULL;
  594. bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
  595. if (!bmd->iovecs) {
  596. kfree(bmd);
  597. return NULL;
  598. }
  599. bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
  600. if (bmd->sgvecs)
  601. return bmd;
  602. kfree(bmd->iovecs);
  603. kfree(bmd);
  604. return NULL;
  605. }
  606. static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
  607. struct sg_iovec *iov, int iov_count, int uncopy,
  608. int do_free_page)
  609. {
  610. int ret = 0, i;
  611. struct bio_vec *bvec;
  612. int iov_idx = 0;
  613. unsigned int iov_off = 0;
  614. int read = bio_data_dir(bio) == READ;
  615. __bio_for_each_segment(bvec, bio, i, 0) {
  616. char *bv_addr = page_address(bvec->bv_page);
  617. unsigned int bv_len = iovecs[i].bv_len;
  618. while (bv_len && iov_idx < iov_count) {
  619. unsigned int bytes;
  620. char *iov_addr;
  621. bytes = min_t(unsigned int,
  622. iov[iov_idx].iov_len - iov_off, bv_len);
  623. iov_addr = iov[iov_idx].iov_base + iov_off;
  624. if (!ret) {
  625. if (!read && !uncopy)
  626. ret = copy_from_user(bv_addr, iov_addr,
  627. bytes);
  628. if (read && uncopy)
  629. ret = copy_to_user(iov_addr, bv_addr,
  630. bytes);
  631. if (ret)
  632. ret = -EFAULT;
  633. }
  634. bv_len -= bytes;
  635. bv_addr += bytes;
  636. iov_addr += bytes;
  637. iov_off += bytes;
  638. if (iov[iov_idx].iov_len == iov_off) {
  639. iov_idx++;
  640. iov_off = 0;
  641. }
  642. }
  643. if (do_free_page)
  644. __free_page(bvec->bv_page);
  645. }
  646. return ret;
  647. }
  648. /**
  649. * bio_uncopy_user - finish previously mapped bio
  650. * @bio: bio being terminated
  651. *
  652. * Free pages allocated from bio_copy_user() and write back data
  653. * to user space in case of a read.
  654. */
  655. int bio_uncopy_user(struct bio *bio)
  656. {
  657. struct bio_map_data *bmd = bio->bi_private;
  658. int ret = 0;
  659. if (!bio_flagged(bio, BIO_NULL_MAPPED))
  660. ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
  661. bmd->nr_sgvecs, 1, bmd->is_our_pages);
  662. bio_free_map_data(bmd);
  663. bio_put(bio);
  664. return ret;
  665. }
  666. /**
  667. * bio_copy_user_iov - copy user data to bio
  668. * @q: destination block queue
  669. * @map_data: pointer to the rq_map_data holding pages (if necessary)
  670. * @iov: the iovec.
  671. * @iov_count: number of elements in the iovec
  672. * @write_to_vm: bool indicating writing to pages or not
  673. * @gfp_mask: memory allocation flags
  674. *
  675. * Prepares and returns a bio for indirect user io, bouncing data
  676. * to/from kernel pages as necessary. Must be paired with
  677. * call bio_uncopy_user() on io completion.
  678. */
  679. struct bio *bio_copy_user_iov(struct request_queue *q,
  680. struct rq_map_data *map_data,
  681. struct sg_iovec *iov, int iov_count,
  682. int write_to_vm, gfp_t gfp_mask)
  683. {
  684. struct bio_map_data *bmd;
  685. struct bio_vec *bvec;
  686. struct page *page;
  687. struct bio *bio;
  688. int i, ret;
  689. int nr_pages = 0;
  690. unsigned int len = 0;
  691. unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
  692. for (i = 0; i < iov_count; i++) {
  693. unsigned long uaddr;
  694. unsigned long end;
  695. unsigned long start;
  696. uaddr = (unsigned long)iov[i].iov_base;
  697. end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
  698. start = uaddr >> PAGE_SHIFT;
  699. nr_pages += end - start;
  700. len += iov[i].iov_len;
  701. }
  702. bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
  703. if (!bmd)
  704. return ERR_PTR(-ENOMEM);
  705. ret = -ENOMEM;
  706. bio = bio_alloc(gfp_mask, nr_pages);
  707. if (!bio)
  708. goto out_bmd;
  709. bio->bi_rw |= (!write_to_vm << BIO_RW);
  710. ret = 0;
  711. if (map_data) {
  712. nr_pages = 1 << map_data->page_order;
  713. i = map_data->offset / PAGE_SIZE;
  714. }
  715. while (len) {
  716. unsigned int bytes = PAGE_SIZE;
  717. bytes -= offset;
  718. if (bytes > len)
  719. bytes = len;
  720. if (map_data) {
  721. if (i == map_data->nr_entries * nr_pages) {
  722. ret = -ENOMEM;
  723. break;
  724. }
  725. page = map_data->pages[i / nr_pages];
  726. page += (i % nr_pages);
  727. i++;
  728. } else {
  729. page = alloc_page(q->bounce_gfp | gfp_mask);
  730. if (!page) {
  731. ret = -ENOMEM;
  732. break;
  733. }
  734. }
  735. if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
  736. break;
  737. len -= bytes;
  738. offset = 0;
  739. }
  740. if (ret)
  741. goto cleanup;
  742. /*
  743. * success
  744. */
  745. if (!write_to_vm && (!map_data || !map_data->null_mapped)) {
  746. ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
  747. if (ret)
  748. goto cleanup;
  749. }
  750. bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
  751. return bio;
  752. cleanup:
  753. if (!map_data)
  754. bio_for_each_segment(bvec, bio, i)
  755. __free_page(bvec->bv_page);
  756. bio_put(bio);
  757. out_bmd:
  758. bio_free_map_data(bmd);
  759. return ERR_PTR(ret);
  760. }
  761. /**
  762. * bio_copy_user - copy user data to bio
  763. * @q: destination block queue
  764. * @map_data: pointer to the rq_map_data holding pages (if necessary)
  765. * @uaddr: start of user address
  766. * @len: length in bytes
  767. * @write_to_vm: bool indicating writing to pages or not
  768. * @gfp_mask: memory allocation flags
  769. *
  770. * Prepares and returns a bio for indirect user io, bouncing data
  771. * to/from kernel pages as necessary. Must be paired with
  772. * call bio_uncopy_user() on io completion.
  773. */
  774. struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
  775. unsigned long uaddr, unsigned int len,
  776. int write_to_vm, gfp_t gfp_mask)
  777. {
  778. struct sg_iovec iov;
  779. iov.iov_base = (void __user *)uaddr;
  780. iov.iov_len = len;
  781. return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
  782. }
  783. static struct bio *__bio_map_user_iov(struct request_queue *q,
  784. struct block_device *bdev,
  785. struct sg_iovec *iov, int iov_count,
  786. int write_to_vm, gfp_t gfp_mask)
  787. {
  788. int i, j;
  789. int nr_pages = 0;
  790. struct page **pages;
  791. struct bio *bio;
  792. int cur_page = 0;
  793. int ret, offset;
  794. for (i = 0; i < iov_count; i++) {
  795. unsigned long uaddr = (unsigned long)iov[i].iov_base;
  796. unsigned long len = iov[i].iov_len;
  797. unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
  798. unsigned long start = uaddr >> PAGE_SHIFT;
  799. nr_pages += end - start;
  800. /*
  801. * buffer must be aligned to at least hardsector size for now
  802. */
  803. if (uaddr & queue_dma_alignment(q))
  804. return ERR_PTR(-EINVAL);
  805. }
  806. if (!nr_pages)
  807. return ERR_PTR(-EINVAL);
  808. bio = bio_alloc(gfp_mask, nr_pages);
  809. if (!bio)
  810. return ERR_PTR(-ENOMEM);
  811. ret = -ENOMEM;
  812. pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
  813. if (!pages)
  814. goto out;
  815. for (i = 0; i < iov_count; i++) {
  816. unsigned long uaddr = (unsigned long)iov[i].iov_base;
  817. unsigned long len = iov[i].iov_len;
  818. unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
  819. unsigned long start = uaddr >> PAGE_SHIFT;
  820. const int local_nr_pages = end - start;
  821. const int page_limit = cur_page + local_nr_pages;
  822. ret = get_user_pages_fast(uaddr, local_nr_pages,
  823. write_to_vm, &pages[cur_page]);
  824. if (ret < local_nr_pages) {
  825. ret = -EFAULT;
  826. goto out_unmap;
  827. }
  828. offset = uaddr & ~PAGE_MASK;
  829. for (j = cur_page; j < page_limit; j++) {
  830. unsigned int bytes = PAGE_SIZE - offset;
  831. if (len <= 0)
  832. break;
  833. if (bytes > len)
  834. bytes = len;
  835. /*
  836. * sorry...
  837. */
  838. if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
  839. bytes)
  840. break;
  841. len -= bytes;
  842. offset = 0;
  843. }
  844. cur_page = j;
  845. /*
  846. * release the pages we didn't map into the bio, if any
  847. */
  848. while (j < page_limit)
  849. page_cache_release(pages[j++]);
  850. }
  851. kfree(pages);
  852. /*
  853. * set data direction, and check if mapped pages need bouncing
  854. */
  855. if (!write_to_vm)
  856. bio->bi_rw |= (1 << BIO_RW);
  857. bio->bi_bdev = bdev;
  858. bio->bi_flags |= (1 << BIO_USER_MAPPED);
  859. return bio;
  860. out_unmap:
  861. for (i = 0; i < nr_pages; i++) {
  862. if(!pages[i])
  863. break;
  864. page_cache_release(pages[i]);
  865. }
  866. out:
  867. kfree(pages);
  868. bio_put(bio);
  869. return ERR_PTR(ret);
  870. }
  871. /**
  872. * bio_map_user - map user address into bio
  873. * @q: the struct request_queue for the bio
  874. * @bdev: destination block device
  875. * @uaddr: start of user address
  876. * @len: length in bytes
  877. * @write_to_vm: bool indicating writing to pages or not
  878. * @gfp_mask: memory allocation flags
  879. *
  880. * Map the user space address into a bio suitable for io to a block
  881. * device. Returns an error pointer in case of error.
  882. */
  883. struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
  884. unsigned long uaddr, unsigned int len, int write_to_vm,
  885. gfp_t gfp_mask)
  886. {
  887. struct sg_iovec iov;
  888. iov.iov_base = (void __user *)uaddr;
  889. iov.iov_len = len;
  890. return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
  891. }
  892. /**
  893. * bio_map_user_iov - map user sg_iovec table into bio
  894. * @q: the struct request_queue for the bio
  895. * @bdev: destination block device
  896. * @iov: the iovec.
  897. * @iov_count: number of elements in the iovec
  898. * @write_to_vm: bool indicating writing to pages or not
  899. * @gfp_mask: memory allocation flags
  900. *
  901. * Map the user space address into a bio suitable for io to a block
  902. * device. Returns an error pointer in case of error.
  903. */
  904. struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
  905. struct sg_iovec *iov, int iov_count,
  906. int write_to_vm, gfp_t gfp_mask)
  907. {
  908. struct bio *bio;
  909. bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
  910. gfp_mask);
  911. if (IS_ERR(bio))
  912. return bio;
  913. /*
  914. * subtle -- if __bio_map_user() ended up bouncing a bio,
  915. * it would normally disappear when its bi_end_io is run.
  916. * however, we need it for the unmap, so grab an extra
  917. * reference to it
  918. */
  919. bio_get(bio);
  920. return bio;
  921. }
  922. static void __bio_unmap_user(struct bio *bio)
  923. {
  924. struct bio_vec *bvec;
  925. int i;
  926. /*
  927. * make sure we dirty pages we wrote to
  928. */
  929. __bio_for_each_segment(bvec, bio, i, 0) {
  930. if (bio_data_dir(bio) == READ)
  931. set_page_dirty_lock(bvec->bv_page);
  932. page_cache_release(bvec->bv_page);
  933. }
  934. bio_put(bio);
  935. }
  936. /**
  937. * bio_unmap_user - unmap a bio
  938. * @bio: the bio being unmapped
  939. *
  940. * Unmap a bio previously mapped by bio_map_user(). Must be called with
  941. * a process context.
  942. *
  943. * bio_unmap_user() may sleep.
  944. */
  945. void bio_unmap_user(struct bio *bio)
  946. {
  947. __bio_unmap_user(bio);
  948. bio_put(bio);
  949. }
  950. static void bio_map_kern_endio(struct bio *bio, int err)
  951. {
  952. bio_put(bio);
  953. }
  954. static struct bio *__bio_map_kern(struct request_queue *q, void *data,
  955. unsigned int len, gfp_t gfp_mask)
  956. {
  957. unsigned long kaddr = (unsigned long)data;
  958. unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
  959. unsigned long start = kaddr >> PAGE_SHIFT;
  960. const int nr_pages = end - start;
  961. int offset, i;
  962. struct bio *bio;
  963. bio = bio_alloc(gfp_mask, nr_pages);
  964. if (!bio)
  965. return ERR_PTR(-ENOMEM);
  966. offset = offset_in_page(kaddr);
  967. for (i = 0; i < nr_pages; i++) {
  968. unsigned int bytes = PAGE_SIZE - offset;
  969. if (len <= 0)
  970. break;
  971. if (bytes > len)
  972. bytes = len;
  973. if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
  974. offset) < bytes)
  975. break;
  976. data += bytes;
  977. len -= bytes;
  978. offset = 0;
  979. }
  980. bio->bi_end_io = bio_map_kern_endio;
  981. return bio;
  982. }
  983. /**
  984. * bio_map_kern - map kernel address into bio
  985. * @q: the struct request_queue for the bio
  986. * @data: pointer to buffer to map
  987. * @len: length in bytes
  988. * @gfp_mask: allocation flags for bio allocation
  989. *
  990. * Map the kernel address into a bio suitable for io to a block
  991. * device. Returns an error pointer in case of error.
  992. */
  993. struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
  994. gfp_t gfp_mask)
  995. {
  996. struct bio *bio;
  997. bio = __bio_map_kern(q, data, len, gfp_mask);
  998. if (IS_ERR(bio))
  999. return bio;
  1000. if (bio->bi_size == len)
  1001. return bio;
  1002. /*
  1003. * Don't support partial mappings.
  1004. */
  1005. bio_put(bio);
  1006. return ERR_PTR(-EINVAL);
  1007. }
  1008. static void bio_copy_kern_endio(struct bio *bio, int err)
  1009. {
  1010. struct bio_vec *bvec;
  1011. const int read = bio_data_dir(bio) == READ;
  1012. struct bio_map_data *bmd = bio->bi_private;
  1013. int i;
  1014. char *p = bmd->sgvecs[0].iov_base;
  1015. __bio_for_each_segment(bvec, bio, i, 0) {
  1016. char *addr = page_address(bvec->bv_page);
  1017. int len = bmd->iovecs[i].bv_len;
  1018. if (read && !err)
  1019. memcpy(p, addr, len);
  1020. __free_page(bvec->bv_page);
  1021. p += len;
  1022. }
  1023. bio_free_map_data(bmd);
  1024. bio_put(bio);
  1025. }
  1026. /**
  1027. * bio_copy_kern - copy kernel address into bio
  1028. * @q: the struct request_queue for the bio
  1029. * @data: pointer to buffer to copy
  1030. * @len: length in bytes
  1031. * @gfp_mask: allocation flags for bio and page allocation
  1032. * @reading: data direction is READ
  1033. *
  1034. * copy the kernel address into a bio suitable for io to a block
  1035. * device. Returns an error pointer in case of error.
  1036. */
  1037. struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
  1038. gfp_t gfp_mask, int reading)
  1039. {
  1040. struct bio *bio;
  1041. struct bio_vec *bvec;
  1042. int i;
  1043. bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
  1044. if (IS_ERR(bio))
  1045. return bio;
  1046. if (!reading) {
  1047. void *p = data;
  1048. bio_for_each_segment(bvec, bio, i) {
  1049. char *addr = page_address(bvec->bv_page);
  1050. memcpy(addr, p, bvec->bv_len);
  1051. p += bvec->bv_len;
  1052. }
  1053. }
  1054. bio->bi_end_io = bio_copy_kern_endio;
  1055. return bio;
  1056. }
  1057. /*
  1058. * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
  1059. * for performing direct-IO in BIOs.
  1060. *
  1061. * The problem is that we cannot run set_page_dirty() from interrupt context
  1062. * because the required locks are not interrupt-safe. So what we can do is to
  1063. * mark the pages dirty _before_ performing IO. And in interrupt context,
  1064. * check that the pages are still dirty. If so, fine. If not, redirty them
  1065. * in process context.
  1066. *
  1067. * We special-case compound pages here: normally this means reads into hugetlb
  1068. * pages. The logic in here doesn't really work right for compound pages
  1069. * because the VM does not uniformly chase down the head page in all cases.
  1070. * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
  1071. * handle them at all. So we skip compound pages here at an early stage.
  1072. *
  1073. * Note that this code is very hard to test under normal circumstances because
  1074. * direct-io pins the pages with get_user_pages(). This makes
  1075. * is_page_cache_freeable return false, and the VM will not clean the pages.
  1076. * But other code (eg, pdflush) could clean the pages if they are mapped
  1077. * pagecache.
  1078. *
  1079. * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
  1080. * deferred bio dirtying paths.
  1081. */
  1082. /*
  1083. * bio_set_pages_dirty() will mark all the bio's pages as dirty.
  1084. */
  1085. void bio_set_pages_dirty(struct bio *bio)
  1086. {
  1087. struct bio_vec *bvec = bio->bi_io_vec;
  1088. int i;
  1089. for (i = 0; i < bio->bi_vcnt; i++) {
  1090. struct page *page = bvec[i].bv_page;
  1091. if (page && !PageCompound(page))
  1092. set_page_dirty_lock(page);
  1093. }
  1094. }
  1095. static void bio_release_pages(struct bio *bio)
  1096. {
  1097. struct bio_vec *bvec = bio->bi_io_vec;
  1098. int i;
  1099. for (i = 0; i < bio->bi_vcnt; i++) {
  1100. struct page *page = bvec[i].bv_page;
  1101. if (page)
  1102. put_page(page);
  1103. }
  1104. }
  1105. /*
  1106. * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
  1107. * If they are, then fine. If, however, some pages are clean then they must
  1108. * have been written out during the direct-IO read. So we take another ref on
  1109. * the BIO and the offending pages and re-dirty the pages in process context.
  1110. *
  1111. * It is expected that bio_check_pages_dirty() will wholly own the BIO from
  1112. * here on. It will run one page_cache_release() against each page and will
  1113. * run one bio_put() against the BIO.
  1114. */
  1115. static void bio_dirty_fn(struct work_struct *work);
  1116. static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
  1117. static DEFINE_SPINLOCK(bio_dirty_lock);
  1118. static struct bio *bio_dirty_list;
  1119. /*
  1120. * This runs in process context
  1121. */
  1122. static void bio_dirty_fn(struct work_struct *work)
  1123. {
  1124. unsigned long flags;
  1125. struct bio *bio;
  1126. spin_lock_irqsave(&bio_dirty_lock, flags);
  1127. bio = bio_dirty_list;
  1128. bio_dirty_list = NULL;
  1129. spin_unlock_irqrestore(&bio_dirty_lock, flags);
  1130. while (bio) {
  1131. struct bio *next = bio->bi_private;
  1132. bio_set_pages_dirty(bio);
  1133. bio_release_pages(bio);
  1134. bio_put(bio);
  1135. bio = next;
  1136. }
  1137. }
  1138. void bio_check_pages_dirty(struct bio *bio)
  1139. {
  1140. struct bio_vec *bvec = bio->bi_io_vec;
  1141. int nr_clean_pages = 0;
  1142. int i;
  1143. for (i = 0; i < bio->bi_vcnt; i++) {
  1144. struct page *page = bvec[i].bv_page;
  1145. if (PageDirty(page) || PageCompound(page)) {
  1146. page_cache_release(page);
  1147. bvec[i].bv_page = NULL;
  1148. } else {
  1149. nr_clean_pages++;
  1150. }
  1151. }
  1152. if (nr_clean_pages) {
  1153. unsigned long flags;
  1154. spin_lock_irqsave(&bio_dirty_lock, flags);
  1155. bio->bi_private = bio_dirty_list;
  1156. bio_dirty_list = bio;
  1157. spin_unlock_irqrestore(&bio_dirty_lock, flags);
  1158. schedule_work(&bio_dirty_work);
  1159. } else {
  1160. bio_put(bio);
  1161. }
  1162. }
  1163. /**
  1164. * bio_endio - end I/O on a bio
  1165. * @bio: bio
  1166. * @error: error, if any
  1167. *
  1168. * Description:
  1169. * bio_endio() will end I/O on the whole bio. bio_endio() is the
  1170. * preferred way to end I/O on a bio, it takes care of clearing
  1171. * BIO_UPTODATE on error. @error is 0 on success, and and one of the
  1172. * established -Exxxx (-EIO, for instance) error values in case
  1173. * something went wrong. Noone should call bi_end_io() directly on a
  1174. * bio unless they own it and thus know that it has an end_io
  1175. * function.
  1176. **/
  1177. void bio_endio(struct bio *bio, int error)
  1178. {
  1179. if (error)
  1180. clear_bit(BIO_UPTODATE, &bio->bi_flags);
  1181. else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
  1182. error = -EIO;
  1183. if (bio->bi_end_io)
  1184. bio->bi_end_io(bio, error);
  1185. }
  1186. void bio_pair_release(struct bio_pair *bp)
  1187. {
  1188. if (atomic_dec_and_test(&bp->cnt)) {
  1189. struct bio *master = bp->bio1.bi_private;
  1190. bio_endio(master, bp->error);
  1191. mempool_free(bp, bp->bio2.bi_private);
  1192. }
  1193. }
  1194. static void bio_pair_end_1(struct bio *bi, int err)
  1195. {
  1196. struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
  1197. if (err)
  1198. bp->error = err;
  1199. bio_pair_release(bp);
  1200. }
  1201. static void bio_pair_end_2(struct bio *bi, int err)
  1202. {
  1203. struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
  1204. if (err)
  1205. bp->error = err;
  1206. bio_pair_release(bp);
  1207. }
  1208. /*
  1209. * split a bio - only worry about a bio with a single page in its iovec
  1210. */
  1211. struct bio_pair *bio_split(struct bio *bi, int first_sectors)
  1212. {
  1213. struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
  1214. if (!bp)
  1215. return bp;
  1216. trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
  1217. bi->bi_sector + first_sectors);
  1218. BUG_ON(bi->bi_vcnt != 1);
  1219. BUG_ON(bi->bi_idx != 0);
  1220. atomic_set(&bp->cnt, 3);
  1221. bp->error = 0;
  1222. bp->bio1 = *bi;
  1223. bp->bio2 = *bi;
  1224. bp->bio2.bi_sector += first_sectors;
  1225. bp->bio2.bi_size -= first_sectors << 9;
  1226. bp->bio1.bi_size = first_sectors << 9;
  1227. bp->bv1 = bi->bi_io_vec[0];
  1228. bp->bv2 = bi->bi_io_vec[0];
  1229. bp->bv2.bv_offset += first_sectors << 9;
  1230. bp->bv2.bv_len -= first_sectors << 9;
  1231. bp->bv1.bv_len = first_sectors << 9;
  1232. bp->bio1.bi_io_vec = &bp->bv1;
  1233. bp->bio2.bi_io_vec = &bp->bv2;
  1234. bp->bio1.bi_max_vecs = 1;
  1235. bp->bio2.bi_max_vecs = 1;
  1236. bp->bio1.bi_end_io = bio_pair_end_1;
  1237. bp->bio2.bi_end_io = bio_pair_end_2;
  1238. bp->bio1.bi_private = bi;
  1239. bp->bio2.bi_private = bio_split_pool;
  1240. if (bio_integrity(bi))
  1241. bio_integrity_split(bi, bp, first_sectors);
  1242. return bp;
  1243. }
  1244. /**
  1245. * bio_sector_offset - Find hardware sector offset in bio
  1246. * @bio: bio to inspect
  1247. * @index: bio_vec index
  1248. * @offset: offset in bv_page
  1249. *
  1250. * Return the number of hardware sectors between beginning of bio
  1251. * and an end point indicated by a bio_vec index and an offset
  1252. * within that vector's page.
  1253. */
  1254. sector_t bio_sector_offset(struct bio *bio, unsigned short index,
  1255. unsigned int offset)
  1256. {
  1257. unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue);
  1258. struct bio_vec *bv;
  1259. sector_t sectors;
  1260. int i;
  1261. sectors = 0;
  1262. if (index >= bio->bi_idx)
  1263. index = bio->bi_vcnt - 1;
  1264. __bio_for_each_segment(bv, bio, i, 0) {
  1265. if (i == index) {
  1266. if (offset > bv->bv_offset)
  1267. sectors += (offset - bv->bv_offset) / sector_sz;
  1268. break;
  1269. }
  1270. sectors += bv->bv_len / sector_sz;
  1271. }
  1272. return sectors;
  1273. }
  1274. EXPORT_SYMBOL(bio_sector_offset);
  1275. /*
  1276. * create memory pools for biovec's in a bio_set.
  1277. * use the global biovec slabs created for general use.
  1278. */
  1279. static int biovec_create_pools(struct bio_set *bs, int pool_entries)
  1280. {
  1281. struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
  1282. bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
  1283. if (!bs->bvec_pool)
  1284. return -ENOMEM;
  1285. return 0;
  1286. }
  1287. static void biovec_free_pools(struct bio_set *bs)
  1288. {
  1289. mempool_destroy(bs->bvec_pool);
  1290. }
  1291. void bioset_free(struct bio_set *bs)
  1292. {
  1293. if (bs->bio_pool)
  1294. mempool_destroy(bs->bio_pool);
  1295. biovec_free_pools(bs);
  1296. bio_put_slab(bs);
  1297. kfree(bs);
  1298. }
  1299. /**
  1300. * bioset_create - Create a bio_set
  1301. * @pool_size: Number of bio and bio_vecs to cache in the mempool
  1302. * @front_pad: Number of bytes to allocate in front of the returned bio
  1303. *
  1304. * Description:
  1305. * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
  1306. * to ask for a number of bytes to be allocated in front of the bio.
  1307. * Front pad allocation is useful for embedding the bio inside
  1308. * another structure, to avoid allocating extra data to go with the bio.
  1309. * Note that the bio must be embedded at the END of that structure always,
  1310. * or things will break badly.
  1311. */
  1312. struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
  1313. {
  1314. unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
  1315. struct bio_set *bs;
  1316. bs = kzalloc(sizeof(*bs), GFP_KERNEL);
  1317. if (!bs)
  1318. return NULL;
  1319. bs->front_pad = front_pad;
  1320. bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
  1321. if (!bs->bio_slab) {
  1322. kfree(bs);
  1323. return NULL;
  1324. }
  1325. bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
  1326. if (!bs->bio_pool)
  1327. goto bad;
  1328. if (!biovec_create_pools(bs, pool_size))
  1329. return bs;
  1330. bad:
  1331. bioset_free(bs);
  1332. return NULL;
  1333. }
  1334. static void __init biovec_init_slabs(void)
  1335. {
  1336. int i;
  1337. for (i = 0; i < BIOVEC_NR_POOLS; i++) {
  1338. int size;
  1339. struct biovec_slab *bvs = bvec_slabs + i;
  1340. #ifndef CONFIG_BLK_DEV_INTEGRITY
  1341. if (bvs->nr_vecs <= BIO_INLINE_VECS) {
  1342. bvs->slab = NULL;
  1343. continue;
  1344. }
  1345. #endif
  1346. size = bvs->nr_vecs * sizeof(struct bio_vec);
  1347. bvs->slab = kmem_cache_create(bvs->name, size, 0,
  1348. SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
  1349. }
  1350. }
  1351. static int __init init_bio(void)
  1352. {
  1353. bio_slab_max = 2;
  1354. bio_slab_nr = 0;
  1355. bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
  1356. if (!bio_slabs)
  1357. panic("bio: can't allocate bios\n");
  1358. biovec_init_slabs();
  1359. fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
  1360. if (!fs_bio_set)
  1361. panic("bio: can't allocate bios\n");
  1362. bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
  1363. sizeof(struct bio_pair));
  1364. if (!bio_split_pool)
  1365. panic("bio: can't create split pool\n");
  1366. return 0;
  1367. }
  1368. subsys_initcall(init_bio);
  1369. EXPORT_SYMBOL(bio_alloc);
  1370. EXPORT_SYMBOL(bio_kmalloc);
  1371. EXPORT_SYMBOL(bio_put);
  1372. EXPORT_SYMBOL(bio_free);
  1373. EXPORT_SYMBOL(bio_endio);
  1374. EXPORT_SYMBOL(bio_init);
  1375. EXPORT_SYMBOL(__bio_clone);
  1376. EXPORT_SYMBOL(bio_clone);
  1377. EXPORT_SYMBOL(bio_phys_segments);
  1378. EXPORT_SYMBOL(bio_add_page);
  1379. EXPORT_SYMBOL(bio_add_pc_page);
  1380. EXPORT_SYMBOL(bio_get_nr_vecs);
  1381. EXPORT_SYMBOL(bio_map_user);
  1382. EXPORT_SYMBOL(bio_unmap_user);
  1383. EXPORT_SYMBOL(bio_map_kern);
  1384. EXPORT_SYMBOL(bio_copy_kern);
  1385. EXPORT_SYMBOL(bio_pair_release);
  1386. EXPORT_SYMBOL(bio_split);
  1387. EXPORT_SYMBOL(bio_copy_user);
  1388. EXPORT_SYMBOL(bio_uncopy_user);
  1389. EXPORT_SYMBOL(bioset_create);
  1390. EXPORT_SYMBOL(bioset_free);
  1391. EXPORT_SYMBOL(bio_alloc_bioset);