bio.c 39 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664
  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 <scsi/sg.h> /* for struct sg_iovec */
  29. #include <trace/events/block.h>
  30. /*
  31. * Test patch to inline a certain number of bi_io_vec's inside the bio
  32. * itself, to shrink a bio data allocation from two mempool calls to one
  33. */
  34. #define BIO_INLINE_VECS 4
  35. static mempool_t *bio_split_pool __read_mostly;
  36. /*
  37. * if you change this list, also change bvec_alloc or things will
  38. * break badly! cannot be bigger than what you can fit into an
  39. * unsigned short
  40. */
  41. #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
  42. struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
  43. BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
  44. };
  45. #undef BV
  46. /*
  47. * fs_bio_set is the bio_set containing bio and iovec memory pools used by
  48. * IO code that does not need private memory pools.
  49. */
  50. struct bio_set *fs_bio_set;
  51. /*
  52. * Our slab pool management
  53. */
  54. struct bio_slab {
  55. struct kmem_cache *slab;
  56. unsigned int slab_ref;
  57. unsigned int slab_size;
  58. char name[8];
  59. };
  60. static DEFINE_MUTEX(bio_slab_lock);
  61. static struct bio_slab *bio_slabs;
  62. static unsigned int bio_slab_nr, bio_slab_max;
  63. static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
  64. {
  65. unsigned int sz = sizeof(struct bio) + extra_size;
  66. struct kmem_cache *slab = NULL;
  67. struct bio_slab *bslab;
  68. unsigned int i, entry = -1;
  69. mutex_lock(&bio_slab_lock);
  70. i = 0;
  71. while (i < bio_slab_nr) {
  72. struct bio_slab *bslab = &bio_slabs[i];
  73. if (!bslab->slab && entry == -1)
  74. entry = i;
  75. else if (bslab->slab_size == sz) {
  76. slab = bslab->slab;
  77. bslab->slab_ref++;
  78. break;
  79. }
  80. i++;
  81. }
  82. if (slab)
  83. goto out_unlock;
  84. if (bio_slab_nr == bio_slab_max && entry == -1) {
  85. bio_slab_max <<= 1;
  86. bio_slabs = krealloc(bio_slabs,
  87. bio_slab_max * sizeof(struct bio_slab),
  88. GFP_KERNEL);
  89. if (!bio_slabs)
  90. goto out_unlock;
  91. }
  92. if (entry == -1)
  93. entry = bio_slab_nr++;
  94. bslab = &bio_slabs[entry];
  95. snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
  96. slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
  97. if (!slab)
  98. goto out_unlock;
  99. printk("bio: create slab <%s> at %d\n", bslab->name, entry);
  100. bslab->slab = slab;
  101. bslab->slab_ref = 1;
  102. bslab->slab_size = sz;
  103. out_unlock:
  104. mutex_unlock(&bio_slab_lock);
  105. return slab;
  106. }
  107. static void bio_put_slab(struct bio_set *bs)
  108. {
  109. struct bio_slab *bslab = NULL;
  110. unsigned int i;
  111. mutex_lock(&bio_slab_lock);
  112. for (i = 0; i < bio_slab_nr; i++) {
  113. if (bs->bio_slab == bio_slabs[i].slab) {
  114. bslab = &bio_slabs[i];
  115. break;
  116. }
  117. }
  118. if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
  119. goto out;
  120. WARN_ON(!bslab->slab_ref);
  121. if (--bslab->slab_ref)
  122. goto out;
  123. kmem_cache_destroy(bslab->slab);
  124. bslab->slab = NULL;
  125. out:
  126. mutex_unlock(&bio_slab_lock);
  127. }
  128. unsigned int bvec_nr_vecs(unsigned short idx)
  129. {
  130. return bvec_slabs[idx].nr_vecs;
  131. }
  132. void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
  133. {
  134. BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
  135. if (idx == BIOVEC_MAX_IDX)
  136. mempool_free(bv, bs->bvec_pool);
  137. else {
  138. struct biovec_slab *bvs = bvec_slabs + idx;
  139. kmem_cache_free(bvs->slab, bv);
  140. }
  141. }
  142. struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
  143. struct bio_set *bs)
  144. {
  145. struct bio_vec *bvl;
  146. /*
  147. * see comment near bvec_array define!
  148. */
  149. switch (nr) {
  150. case 1:
  151. *idx = 0;
  152. break;
  153. case 2 ... 4:
  154. *idx = 1;
  155. break;
  156. case 5 ... 16:
  157. *idx = 2;
  158. break;
  159. case 17 ... 64:
  160. *idx = 3;
  161. break;
  162. case 65 ... 128:
  163. *idx = 4;
  164. break;
  165. case 129 ... BIO_MAX_PAGES:
  166. *idx = 5;
  167. break;
  168. default:
  169. return NULL;
  170. }
  171. /*
  172. * idx now points to the pool we want to allocate from. only the
  173. * 1-vec entry pool is mempool backed.
  174. */
  175. if (*idx == BIOVEC_MAX_IDX) {
  176. fallback:
  177. bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
  178. } else {
  179. struct biovec_slab *bvs = bvec_slabs + *idx;
  180. gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
  181. /*
  182. * Make this allocation restricted and don't dump info on
  183. * allocation failures, since we'll fallback to the mempool
  184. * in case of failure.
  185. */
  186. __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
  187. /*
  188. * Try a slab allocation. If this fails and __GFP_WAIT
  189. * is set, retry with the 1-entry mempool
  190. */
  191. bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
  192. if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
  193. *idx = BIOVEC_MAX_IDX;
  194. goto fallback;
  195. }
  196. }
  197. return bvl;
  198. }
  199. void bio_free(struct bio *bio, struct bio_set *bs)
  200. {
  201. void *p;
  202. if (bio_has_allocated_vec(bio))
  203. bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
  204. if (bio_integrity(bio))
  205. bio_integrity_free(bio, bs);
  206. /*
  207. * If we have front padding, adjust the bio pointer before freeing
  208. */
  209. p = bio;
  210. if (bs->front_pad)
  211. p -= bs->front_pad;
  212. mempool_free(p, bs->bio_pool);
  213. }
  214. void bio_init(struct bio *bio)
  215. {
  216. memset(bio, 0, sizeof(*bio));
  217. bio->bi_flags = 1 << BIO_UPTODATE;
  218. bio->bi_comp_cpu = -1;
  219. atomic_set(&bio->bi_cnt, 1);
  220. }
  221. /**
  222. * bio_alloc_bioset - allocate a bio for I/O
  223. * @gfp_mask: the GFP_ mask given to the slab allocator
  224. * @nr_iovecs: number of iovecs to pre-allocate
  225. * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
  226. *
  227. * Description:
  228. * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
  229. * If %__GFP_WAIT is set then we will block on the internal pool waiting
  230. * for a &struct bio to become free. If a %NULL @bs is passed in, we will
  231. * fall back to just using @kmalloc to allocate the required memory.
  232. *
  233. * Note that the caller must set ->bi_destructor on succesful return
  234. * of a bio, to do the appropriate freeing of the bio once the reference
  235. * count drops to zero.
  236. **/
  237. struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
  238. {
  239. unsigned long idx = BIO_POOL_NONE;
  240. struct bio_vec *bvl = NULL;
  241. struct bio *bio;
  242. void *p;
  243. p = mempool_alloc(bs->bio_pool, gfp_mask);
  244. if (unlikely(!p))
  245. return NULL;
  246. bio = p + bs->front_pad;
  247. bio_init(bio);
  248. if (unlikely(!nr_iovecs))
  249. goto out_set;
  250. if (nr_iovecs <= BIO_INLINE_VECS) {
  251. bvl = bio->bi_inline_vecs;
  252. nr_iovecs = BIO_INLINE_VECS;
  253. } else {
  254. bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
  255. if (unlikely(!bvl))
  256. goto err_free;
  257. nr_iovecs = bvec_nr_vecs(idx);
  258. }
  259. out_set:
  260. bio->bi_flags |= idx << BIO_POOL_OFFSET;
  261. bio->bi_max_vecs = nr_iovecs;
  262. bio->bi_io_vec = bvl;
  263. return bio;
  264. err_free:
  265. mempool_free(p, bs->bio_pool);
  266. return NULL;
  267. }
  268. static void bio_fs_destructor(struct bio *bio)
  269. {
  270. bio_free(bio, fs_bio_set);
  271. }
  272. /**
  273. * bio_alloc - allocate a new bio, memory pool backed
  274. * @gfp_mask: allocation mask to use
  275. * @nr_iovecs: number of iovecs
  276. *
  277. * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask
  278. * contains __GFP_WAIT, the allocation is guaranteed to succeed.
  279. *
  280. * RETURNS:
  281. * Pointer to new bio on success, NULL on failure.
  282. */
  283. struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
  284. {
  285. struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
  286. if (bio)
  287. bio->bi_destructor = bio_fs_destructor;
  288. return bio;
  289. }
  290. static void bio_kmalloc_destructor(struct bio *bio)
  291. {
  292. if (bio_integrity(bio))
  293. bio_integrity_free(bio, fs_bio_set);
  294. kfree(bio);
  295. }
  296. /**
  297. * bio_alloc - allocate a bio for I/O
  298. * @gfp_mask: the GFP_ mask given to the slab allocator
  299. * @nr_iovecs: number of iovecs to pre-allocate
  300. *
  301. * Description:
  302. * bio_alloc will allocate a bio and associated bio_vec array that can hold
  303. * at least @nr_iovecs entries. Allocations will be done from the
  304. * fs_bio_set. Also see @bio_alloc_bioset.
  305. *
  306. * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
  307. * a bio. This is due to the mempool guarantees. To make this work, callers
  308. * must never allocate more than 1 bio at a time from this pool. Callers
  309. * that need to allocate more than 1 bio must always submit the previously
  310. * allocated bio for IO before attempting to allocate a new one. Failure to
  311. * do so can cause livelocks under memory pressure.
  312. *
  313. **/
  314. struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
  315. {
  316. struct bio *bio;
  317. bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
  318. gfp_mask);
  319. if (unlikely(!bio))
  320. return NULL;
  321. bio_init(bio);
  322. bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
  323. bio->bi_max_vecs = nr_iovecs;
  324. bio->bi_io_vec = bio->bi_inline_vecs;
  325. bio->bi_destructor = bio_kmalloc_destructor;
  326. return bio;
  327. }
  328. void zero_fill_bio(struct bio *bio)
  329. {
  330. unsigned long flags;
  331. struct bio_vec *bv;
  332. int i;
  333. bio_for_each_segment(bv, bio, i) {
  334. char *data = bvec_kmap_irq(bv, &flags);
  335. memset(data, 0, bv->bv_len);
  336. flush_dcache_page(bv->bv_page);
  337. bvec_kunmap_irq(data, &flags);
  338. }
  339. }
  340. EXPORT_SYMBOL(zero_fill_bio);
  341. /**
  342. * bio_put - release a reference to a bio
  343. * @bio: bio to release reference to
  344. *
  345. * Description:
  346. * Put a reference to a &struct bio, either one you have gotten with
  347. * bio_alloc or bio_get. The last put of a bio will free it.
  348. **/
  349. void bio_put(struct bio *bio)
  350. {
  351. BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
  352. /*
  353. * last put frees it
  354. */
  355. if (atomic_dec_and_test(&bio->bi_cnt)) {
  356. bio->bi_next = NULL;
  357. bio->bi_destructor(bio);
  358. }
  359. }
  360. inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
  361. {
  362. if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
  363. blk_recount_segments(q, bio);
  364. return bio->bi_phys_segments;
  365. }
  366. /**
  367. * __bio_clone - clone a bio
  368. * @bio: destination bio
  369. * @bio_src: bio to clone
  370. *
  371. * Clone a &bio. Caller will own the returned bio, but not
  372. * the actual data it points to. Reference count of returned
  373. * bio will be one.
  374. */
  375. void __bio_clone(struct bio *bio, struct bio *bio_src)
  376. {
  377. memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
  378. bio_src->bi_max_vecs * sizeof(struct bio_vec));
  379. /*
  380. * most users will be overriding ->bi_bdev with a new target,
  381. * so we don't set nor calculate new physical/hw segment counts here
  382. */
  383. bio->bi_sector = bio_src->bi_sector;
  384. bio->bi_bdev = bio_src->bi_bdev;
  385. bio->bi_flags |= 1 << BIO_CLONED;
  386. bio->bi_rw = bio_src->bi_rw;
  387. bio->bi_vcnt = bio_src->bi_vcnt;
  388. bio->bi_size = bio_src->bi_size;
  389. bio->bi_idx = bio_src->bi_idx;
  390. }
  391. /**
  392. * bio_clone - clone a bio
  393. * @bio: bio to clone
  394. * @gfp_mask: allocation priority
  395. *
  396. * Like __bio_clone, only also allocates the returned bio
  397. */
  398. struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
  399. {
  400. struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
  401. if (!b)
  402. return NULL;
  403. b->bi_destructor = bio_fs_destructor;
  404. __bio_clone(b, bio);
  405. if (bio_integrity(bio)) {
  406. int ret;
  407. ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
  408. if (ret < 0) {
  409. bio_put(b);
  410. return NULL;
  411. }
  412. }
  413. return b;
  414. }
  415. /**
  416. * bio_get_nr_vecs - return approx number of vecs
  417. * @bdev: I/O target
  418. *
  419. * Return the approximate number of pages we can send to this target.
  420. * There's no guarantee that you will be able to fit this number of pages
  421. * into a bio, it does not account for dynamic restrictions that vary
  422. * on offset.
  423. */
  424. int bio_get_nr_vecs(struct block_device *bdev)
  425. {
  426. struct request_queue *q = bdev_get_queue(bdev);
  427. int nr_pages;
  428. nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
  429. if (nr_pages > queue_max_phys_segments(q))
  430. nr_pages = queue_max_phys_segments(q);
  431. if (nr_pages > queue_max_hw_segments(q))
  432. nr_pages = queue_max_hw_segments(q);
  433. return nr_pages;
  434. }
  435. static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
  436. *page, unsigned int len, unsigned int offset,
  437. unsigned short max_sectors)
  438. {
  439. int retried_segments = 0;
  440. struct bio_vec *bvec;
  441. /*
  442. * cloned bio must not modify vec list
  443. */
  444. if (unlikely(bio_flagged(bio, BIO_CLONED)))
  445. return 0;
  446. if (((bio->bi_size + len) >> 9) > max_sectors)
  447. return 0;
  448. /*
  449. * For filesystems with a blocksize smaller than the pagesize
  450. * we will often be called with the same page as last time and
  451. * a consecutive offset. Optimize this special case.
  452. */
  453. if (bio->bi_vcnt > 0) {
  454. struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
  455. if (page == prev->bv_page &&
  456. offset == prev->bv_offset + prev->bv_len) {
  457. prev->bv_len += len;
  458. if (q->merge_bvec_fn) {
  459. struct bvec_merge_data bvm = {
  460. .bi_bdev = bio->bi_bdev,
  461. .bi_sector = bio->bi_sector,
  462. .bi_size = bio->bi_size,
  463. .bi_rw = bio->bi_rw,
  464. };
  465. if (q->merge_bvec_fn(q, &bvm, prev) < len) {
  466. prev->bv_len -= len;
  467. return 0;
  468. }
  469. }
  470. goto done;
  471. }
  472. }
  473. if (bio->bi_vcnt >= bio->bi_max_vecs)
  474. return 0;
  475. /*
  476. * we might lose a segment or two here, but rather that than
  477. * make this too complex.
  478. */
  479. while (bio->bi_phys_segments >= queue_max_phys_segments(q)
  480. || bio->bi_phys_segments >= queue_max_hw_segments(q)) {
  481. if (retried_segments)
  482. return 0;
  483. retried_segments = 1;
  484. blk_recount_segments(q, bio);
  485. }
  486. /*
  487. * setup the new entry, we might clear it again later if we
  488. * cannot add the page
  489. */
  490. bvec = &bio->bi_io_vec[bio->bi_vcnt];
  491. bvec->bv_page = page;
  492. bvec->bv_len = len;
  493. bvec->bv_offset = offset;
  494. /*
  495. * if queue has other restrictions (eg varying max sector size
  496. * depending on offset), it can specify a merge_bvec_fn in the
  497. * queue to get further control
  498. */
  499. if (q->merge_bvec_fn) {
  500. struct bvec_merge_data bvm = {
  501. .bi_bdev = bio->bi_bdev,
  502. .bi_sector = bio->bi_sector,
  503. .bi_size = bio->bi_size,
  504. .bi_rw = bio->bi_rw,
  505. };
  506. /*
  507. * merge_bvec_fn() returns number of bytes it can accept
  508. * at this offset
  509. */
  510. if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
  511. bvec->bv_page = NULL;
  512. bvec->bv_len = 0;
  513. bvec->bv_offset = 0;
  514. return 0;
  515. }
  516. }
  517. /* If we may be able to merge these biovecs, force a recount */
  518. if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
  519. bio->bi_flags &= ~(1 << BIO_SEG_VALID);
  520. bio->bi_vcnt++;
  521. bio->bi_phys_segments++;
  522. done:
  523. bio->bi_size += len;
  524. return len;
  525. }
  526. /**
  527. * bio_add_pc_page - attempt to add page to bio
  528. * @q: the target queue
  529. * @bio: destination bio
  530. * @page: page to add
  531. * @len: vec entry length
  532. * @offset: vec entry offset
  533. *
  534. * Attempt to add a page to the bio_vec maplist. This can fail for a
  535. * number of reasons, such as the bio being full or target block
  536. * device limitations. The target block device must allow bio's
  537. * smaller than PAGE_SIZE, so it is always possible to add a single
  538. * page to an empty bio. This should only be used by REQ_PC bios.
  539. */
  540. int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
  541. unsigned int len, unsigned int offset)
  542. {
  543. return __bio_add_page(q, bio, page, len, offset,
  544. queue_max_hw_sectors(q));
  545. }
  546. /**
  547. * bio_add_page - attempt to add page to bio
  548. * @bio: destination bio
  549. * @page: page to add
  550. * @len: vec entry length
  551. * @offset: vec entry offset
  552. *
  553. * Attempt to add a page to the bio_vec maplist. This can fail for a
  554. * number of reasons, such as the bio being full or target block
  555. * device limitations. The target block device must allow bio's
  556. * smaller than PAGE_SIZE, so it is always possible to add a single
  557. * page to an empty bio.
  558. */
  559. int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
  560. unsigned int offset)
  561. {
  562. struct request_queue *q = bdev_get_queue(bio->bi_bdev);
  563. return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
  564. }
  565. struct bio_map_data {
  566. struct bio_vec *iovecs;
  567. struct sg_iovec *sgvecs;
  568. int nr_sgvecs;
  569. int is_our_pages;
  570. };
  571. static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
  572. struct sg_iovec *iov, int iov_count,
  573. int is_our_pages)
  574. {
  575. memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
  576. memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
  577. bmd->nr_sgvecs = iov_count;
  578. bmd->is_our_pages = is_our_pages;
  579. bio->bi_private = bmd;
  580. }
  581. static void bio_free_map_data(struct bio_map_data *bmd)
  582. {
  583. kfree(bmd->iovecs);
  584. kfree(bmd->sgvecs);
  585. kfree(bmd);
  586. }
  587. static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
  588. gfp_t gfp_mask)
  589. {
  590. struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
  591. if (!bmd)
  592. return NULL;
  593. bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
  594. if (!bmd->iovecs) {
  595. kfree(bmd);
  596. return NULL;
  597. }
  598. bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
  599. if (bmd->sgvecs)
  600. return bmd;
  601. kfree(bmd->iovecs);
  602. kfree(bmd);
  603. return NULL;
  604. }
  605. static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
  606. struct sg_iovec *iov, int iov_count,
  607. int to_user, int from_user, int do_free_page)
  608. {
  609. int ret = 0, i;
  610. struct bio_vec *bvec;
  611. int iov_idx = 0;
  612. unsigned int iov_off = 0;
  613. __bio_for_each_segment(bvec, bio, i, 0) {
  614. char *bv_addr = page_address(bvec->bv_page);
  615. unsigned int bv_len = iovecs[i].bv_len;
  616. while (bv_len && iov_idx < iov_count) {
  617. unsigned int bytes;
  618. char __user *iov_addr;
  619. bytes = min_t(unsigned int,
  620. iov[iov_idx].iov_len - iov_off, bv_len);
  621. iov_addr = iov[iov_idx].iov_base + iov_off;
  622. if (!ret) {
  623. if (to_user)
  624. ret = copy_to_user(iov_addr, bv_addr,
  625. bytes);
  626. if (from_user)
  627. ret = copy_from_user(bv_addr, iov_addr,
  628. bytes);
  629. if (ret)
  630. ret = -EFAULT;
  631. }
  632. bv_len -= bytes;
  633. bv_addr += bytes;
  634. iov_addr += bytes;
  635. iov_off += bytes;
  636. if (iov[iov_idx].iov_len == iov_off) {
  637. iov_idx++;
  638. iov_off = 0;
  639. }
  640. }
  641. if (do_free_page)
  642. __free_page(bvec->bv_page);
  643. }
  644. return ret;
  645. }
  646. /**
  647. * bio_uncopy_user - finish previously mapped bio
  648. * @bio: bio being terminated
  649. *
  650. * Free pages allocated from bio_copy_user() and write back data
  651. * to user space in case of a read.
  652. */
  653. int bio_uncopy_user(struct bio *bio)
  654. {
  655. struct bio_map_data *bmd = bio->bi_private;
  656. int ret = 0;
  657. if (!bio_flagged(bio, BIO_NULL_MAPPED))
  658. ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
  659. bmd->nr_sgvecs, bio_data_dir(bio) == READ,
  660. 0, bmd->is_our_pages);
  661. bio_free_map_data(bmd);
  662. bio_put(bio);
  663. return ret;
  664. }
  665. /**
  666. * bio_copy_user_iov - copy user data to bio
  667. * @q: destination block queue
  668. * @map_data: pointer to the rq_map_data holding pages (if necessary)
  669. * @iov: the iovec.
  670. * @iov_count: number of elements in the iovec
  671. * @write_to_vm: bool indicating writing to pages or not
  672. * @gfp_mask: memory allocation flags
  673. *
  674. * Prepares and returns a bio for indirect user io, bouncing data
  675. * to/from kernel pages as necessary. Must be paired with
  676. * call bio_uncopy_user() on io completion.
  677. */
  678. struct bio *bio_copy_user_iov(struct request_queue *q,
  679. struct rq_map_data *map_data,
  680. struct sg_iovec *iov, int iov_count,
  681. int write_to_vm, gfp_t gfp_mask)
  682. {
  683. struct bio_map_data *bmd;
  684. struct bio_vec *bvec;
  685. struct page *page;
  686. struct bio *bio;
  687. int i, ret;
  688. int nr_pages = 0;
  689. unsigned int len = 0;
  690. unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
  691. for (i = 0; i < iov_count; i++) {
  692. unsigned long uaddr;
  693. unsigned long end;
  694. unsigned long start;
  695. uaddr = (unsigned long)iov[i].iov_base;
  696. end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
  697. start = uaddr >> PAGE_SHIFT;
  698. nr_pages += end - start;
  699. len += iov[i].iov_len;
  700. }
  701. if (offset)
  702. nr_pages++;
  703. bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
  704. if (!bmd)
  705. return ERR_PTR(-ENOMEM);
  706. ret = -ENOMEM;
  707. bio = bio_kmalloc(gfp_mask, nr_pages);
  708. if (!bio)
  709. goto out_bmd;
  710. bio->bi_rw |= (!write_to_vm << BIO_RW);
  711. ret = 0;
  712. if (map_data) {
  713. nr_pages = 1 << map_data->page_order;
  714. i = map_data->offset / PAGE_SIZE;
  715. }
  716. while (len) {
  717. unsigned int bytes = PAGE_SIZE;
  718. bytes -= offset;
  719. if (bytes > len)
  720. bytes = len;
  721. if (map_data) {
  722. if (i == map_data->nr_entries * nr_pages) {
  723. ret = -ENOMEM;
  724. break;
  725. }
  726. page = map_data->pages[i / nr_pages];
  727. page += (i % nr_pages);
  728. i++;
  729. } else {
  730. page = alloc_page(q->bounce_gfp | gfp_mask);
  731. if (!page) {
  732. ret = -ENOMEM;
  733. break;
  734. }
  735. }
  736. if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
  737. break;
  738. len -= bytes;
  739. offset = 0;
  740. }
  741. if (ret)
  742. goto cleanup;
  743. /*
  744. * success
  745. */
  746. if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
  747. (map_data && map_data->from_user)) {
  748. ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
  749. if (ret)
  750. goto cleanup;
  751. }
  752. bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
  753. return bio;
  754. cleanup:
  755. if (!map_data)
  756. bio_for_each_segment(bvec, bio, i)
  757. __free_page(bvec->bv_page);
  758. bio_put(bio);
  759. out_bmd:
  760. bio_free_map_data(bmd);
  761. return ERR_PTR(ret);
  762. }
  763. /**
  764. * bio_copy_user - copy user data to bio
  765. * @q: destination block queue
  766. * @map_data: pointer to the rq_map_data holding pages (if necessary)
  767. * @uaddr: start of user address
  768. * @len: length in bytes
  769. * @write_to_vm: bool indicating writing to pages or not
  770. * @gfp_mask: memory allocation flags
  771. *
  772. * Prepares and returns a bio for indirect user io, bouncing data
  773. * to/from kernel pages as necessary. Must be paired with
  774. * call bio_uncopy_user() on io completion.
  775. */
  776. struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
  777. unsigned long uaddr, unsigned int len,
  778. int write_to_vm, gfp_t gfp_mask)
  779. {
  780. struct sg_iovec iov;
  781. iov.iov_base = (void __user *)uaddr;
  782. iov.iov_len = len;
  783. return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
  784. }
  785. static struct bio *__bio_map_user_iov(struct request_queue *q,
  786. struct block_device *bdev,
  787. struct sg_iovec *iov, int iov_count,
  788. int write_to_vm, gfp_t gfp_mask)
  789. {
  790. int i, j;
  791. int nr_pages = 0;
  792. struct page **pages;
  793. struct bio *bio;
  794. int cur_page = 0;
  795. int ret, offset;
  796. for (i = 0; i < iov_count; i++) {
  797. unsigned long uaddr = (unsigned long)iov[i].iov_base;
  798. unsigned long len = iov[i].iov_len;
  799. unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
  800. unsigned long start = uaddr >> PAGE_SHIFT;
  801. nr_pages += end - start;
  802. /*
  803. * buffer must be aligned to at least hardsector size for now
  804. */
  805. if (uaddr & queue_dma_alignment(q))
  806. return ERR_PTR(-EINVAL);
  807. }
  808. if (!nr_pages)
  809. return ERR_PTR(-EINVAL);
  810. bio = bio_kmalloc(gfp_mask, nr_pages);
  811. if (!bio)
  812. return ERR_PTR(-ENOMEM);
  813. ret = -ENOMEM;
  814. pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
  815. if (!pages)
  816. goto out;
  817. for (i = 0; i < iov_count; i++) {
  818. unsigned long uaddr = (unsigned long)iov[i].iov_base;
  819. unsigned long len = iov[i].iov_len;
  820. unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
  821. unsigned long start = uaddr >> PAGE_SHIFT;
  822. const int local_nr_pages = end - start;
  823. const int page_limit = cur_page + local_nr_pages;
  824. ret = get_user_pages_fast(uaddr, local_nr_pages,
  825. write_to_vm, &pages[cur_page]);
  826. if (ret < local_nr_pages) {
  827. ret = -EFAULT;
  828. goto out_unmap;
  829. }
  830. offset = uaddr & ~PAGE_MASK;
  831. for (j = cur_page; j < page_limit; j++) {
  832. unsigned int bytes = PAGE_SIZE - offset;
  833. if (len <= 0)
  834. break;
  835. if (bytes > len)
  836. bytes = len;
  837. /*
  838. * sorry...
  839. */
  840. if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
  841. bytes)
  842. break;
  843. len -= bytes;
  844. offset = 0;
  845. }
  846. cur_page = j;
  847. /*
  848. * release the pages we didn't map into the bio, if any
  849. */
  850. while (j < page_limit)
  851. page_cache_release(pages[j++]);
  852. }
  853. kfree(pages);
  854. /*
  855. * set data direction, and check if mapped pages need bouncing
  856. */
  857. if (!write_to_vm)
  858. bio->bi_rw |= (1 << BIO_RW);
  859. bio->bi_bdev = bdev;
  860. bio->bi_flags |= (1 << BIO_USER_MAPPED);
  861. return bio;
  862. out_unmap:
  863. for (i = 0; i < nr_pages; i++) {
  864. if(!pages[i])
  865. break;
  866. page_cache_release(pages[i]);
  867. }
  868. out:
  869. kfree(pages);
  870. bio_put(bio);
  871. return ERR_PTR(ret);
  872. }
  873. /**
  874. * bio_map_user - map user address into bio
  875. * @q: the struct request_queue for the bio
  876. * @bdev: destination block device
  877. * @uaddr: start of user address
  878. * @len: length in bytes
  879. * @write_to_vm: bool indicating writing to pages or not
  880. * @gfp_mask: memory allocation flags
  881. *
  882. * Map the user space address into a bio suitable for io to a block
  883. * device. Returns an error pointer in case of error.
  884. */
  885. struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
  886. unsigned long uaddr, unsigned int len, int write_to_vm,
  887. gfp_t gfp_mask)
  888. {
  889. struct sg_iovec iov;
  890. iov.iov_base = (void __user *)uaddr;
  891. iov.iov_len = len;
  892. return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
  893. }
  894. /**
  895. * bio_map_user_iov - map user sg_iovec table into bio
  896. * @q: the struct request_queue for the bio
  897. * @bdev: destination block device
  898. * @iov: the iovec.
  899. * @iov_count: number of elements in the iovec
  900. * @write_to_vm: bool indicating writing to pages or not
  901. * @gfp_mask: memory allocation flags
  902. *
  903. * Map the user space address into a bio suitable for io to a block
  904. * device. Returns an error pointer in case of error.
  905. */
  906. struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
  907. struct sg_iovec *iov, int iov_count,
  908. int write_to_vm, gfp_t gfp_mask)
  909. {
  910. struct bio *bio;
  911. bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
  912. gfp_mask);
  913. if (IS_ERR(bio))
  914. return bio;
  915. /*
  916. * subtle -- if __bio_map_user() ended up bouncing a bio,
  917. * it would normally disappear when its bi_end_io is run.
  918. * however, we need it for the unmap, so grab an extra
  919. * reference to it
  920. */
  921. bio_get(bio);
  922. return bio;
  923. }
  924. static void __bio_unmap_user(struct bio *bio)
  925. {
  926. struct bio_vec *bvec;
  927. int i;
  928. /*
  929. * make sure we dirty pages we wrote to
  930. */
  931. __bio_for_each_segment(bvec, bio, i, 0) {
  932. if (bio_data_dir(bio) == READ)
  933. set_page_dirty_lock(bvec->bv_page);
  934. page_cache_release(bvec->bv_page);
  935. }
  936. bio_put(bio);
  937. }
  938. /**
  939. * bio_unmap_user - unmap a bio
  940. * @bio: the bio being unmapped
  941. *
  942. * Unmap a bio previously mapped by bio_map_user(). Must be called with
  943. * a process context.
  944. *
  945. * bio_unmap_user() may sleep.
  946. */
  947. void bio_unmap_user(struct bio *bio)
  948. {
  949. __bio_unmap_user(bio);
  950. bio_put(bio);
  951. }
  952. static void bio_map_kern_endio(struct bio *bio, int err)
  953. {
  954. bio_put(bio);
  955. }
  956. static struct bio *__bio_map_kern(struct request_queue *q, void *data,
  957. unsigned int len, gfp_t gfp_mask)
  958. {
  959. unsigned long kaddr = (unsigned long)data;
  960. unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
  961. unsigned long start = kaddr >> PAGE_SHIFT;
  962. const int nr_pages = end - start;
  963. int offset, i;
  964. struct bio *bio;
  965. bio = bio_kmalloc(gfp_mask, nr_pages);
  966. if (!bio)
  967. return ERR_PTR(-ENOMEM);
  968. offset = offset_in_page(kaddr);
  969. for (i = 0; i < nr_pages; i++) {
  970. unsigned int bytes = PAGE_SIZE - offset;
  971. if (len <= 0)
  972. break;
  973. if (bytes > len)
  974. bytes = len;
  975. if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
  976. offset) < bytes)
  977. break;
  978. data += bytes;
  979. len -= bytes;
  980. offset = 0;
  981. }
  982. bio->bi_end_io = bio_map_kern_endio;
  983. return bio;
  984. }
  985. /**
  986. * bio_map_kern - map kernel address into bio
  987. * @q: the struct request_queue for the bio
  988. * @data: pointer to buffer to map
  989. * @len: length in bytes
  990. * @gfp_mask: allocation flags for bio allocation
  991. *
  992. * Map the kernel address into a bio suitable for io to a block
  993. * device. Returns an error pointer in case of error.
  994. */
  995. struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
  996. gfp_t gfp_mask)
  997. {
  998. struct bio *bio;
  999. bio = __bio_map_kern(q, data, len, gfp_mask);
  1000. if (IS_ERR(bio))
  1001. return bio;
  1002. if (bio->bi_size == len)
  1003. return bio;
  1004. /*
  1005. * Don't support partial mappings.
  1006. */
  1007. bio_put(bio);
  1008. return ERR_PTR(-EINVAL);
  1009. }
  1010. static void bio_copy_kern_endio(struct bio *bio, int err)
  1011. {
  1012. struct bio_vec *bvec;
  1013. const int read = bio_data_dir(bio) == READ;
  1014. struct bio_map_data *bmd = bio->bi_private;
  1015. int i;
  1016. char *p = bmd->sgvecs[0].iov_base;
  1017. __bio_for_each_segment(bvec, bio, i, 0) {
  1018. char *addr = page_address(bvec->bv_page);
  1019. int len = bmd->iovecs[i].bv_len;
  1020. if (read)
  1021. memcpy(p, addr, len);
  1022. __free_page(bvec->bv_page);
  1023. p += len;
  1024. }
  1025. bio_free_map_data(bmd);
  1026. bio_put(bio);
  1027. }
  1028. /**
  1029. * bio_copy_kern - copy kernel address into bio
  1030. * @q: the struct request_queue for the bio
  1031. * @data: pointer to buffer to copy
  1032. * @len: length in bytes
  1033. * @gfp_mask: allocation flags for bio and page allocation
  1034. * @reading: data direction is READ
  1035. *
  1036. * copy the kernel address into a bio suitable for io to a block
  1037. * device. Returns an error pointer in case of error.
  1038. */
  1039. struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
  1040. gfp_t gfp_mask, int reading)
  1041. {
  1042. struct bio *bio;
  1043. struct bio_vec *bvec;
  1044. int i;
  1045. bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
  1046. if (IS_ERR(bio))
  1047. return bio;
  1048. if (!reading) {
  1049. void *p = data;
  1050. bio_for_each_segment(bvec, bio, i) {
  1051. char *addr = page_address(bvec->bv_page);
  1052. memcpy(addr, p, bvec->bv_len);
  1053. p += bvec->bv_len;
  1054. }
  1055. }
  1056. bio->bi_end_io = bio_copy_kern_endio;
  1057. return bio;
  1058. }
  1059. /*
  1060. * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
  1061. * for performing direct-IO in BIOs.
  1062. *
  1063. * The problem is that we cannot run set_page_dirty() from interrupt context
  1064. * because the required locks are not interrupt-safe. So what we can do is to
  1065. * mark the pages dirty _before_ performing IO. And in interrupt context,
  1066. * check that the pages are still dirty. If so, fine. If not, redirty them
  1067. * in process context.
  1068. *
  1069. * We special-case compound pages here: normally this means reads into hugetlb
  1070. * pages. The logic in here doesn't really work right for compound pages
  1071. * because the VM does not uniformly chase down the head page in all cases.
  1072. * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
  1073. * handle them at all. So we skip compound pages here at an early stage.
  1074. *
  1075. * Note that this code is very hard to test under normal circumstances because
  1076. * direct-io pins the pages with get_user_pages(). This makes
  1077. * is_page_cache_freeable return false, and the VM will not clean the pages.
  1078. * But other code (eg, pdflush) could clean the pages if they are mapped
  1079. * pagecache.
  1080. *
  1081. * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
  1082. * deferred bio dirtying paths.
  1083. */
  1084. /*
  1085. * bio_set_pages_dirty() will mark all the bio's pages as dirty.
  1086. */
  1087. void bio_set_pages_dirty(struct bio *bio)
  1088. {
  1089. struct bio_vec *bvec = bio->bi_io_vec;
  1090. int i;
  1091. for (i = 0; i < bio->bi_vcnt; i++) {
  1092. struct page *page = bvec[i].bv_page;
  1093. if (page && !PageCompound(page))
  1094. set_page_dirty_lock(page);
  1095. }
  1096. }
  1097. static void bio_release_pages(struct bio *bio)
  1098. {
  1099. struct bio_vec *bvec = bio->bi_io_vec;
  1100. int i;
  1101. for (i = 0; i < bio->bi_vcnt; i++) {
  1102. struct page *page = bvec[i].bv_page;
  1103. if (page)
  1104. put_page(page);
  1105. }
  1106. }
  1107. /*
  1108. * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
  1109. * If they are, then fine. If, however, some pages are clean then they must
  1110. * have been written out during the direct-IO read. So we take another ref on
  1111. * the BIO and the offending pages and re-dirty the pages in process context.
  1112. *
  1113. * It is expected that bio_check_pages_dirty() will wholly own the BIO from
  1114. * here on. It will run one page_cache_release() against each page and will
  1115. * run one bio_put() against the BIO.
  1116. */
  1117. static void bio_dirty_fn(struct work_struct *work);
  1118. static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
  1119. static DEFINE_SPINLOCK(bio_dirty_lock);
  1120. static struct bio *bio_dirty_list;
  1121. /*
  1122. * This runs in process context
  1123. */
  1124. static void bio_dirty_fn(struct work_struct *work)
  1125. {
  1126. unsigned long flags;
  1127. struct bio *bio;
  1128. spin_lock_irqsave(&bio_dirty_lock, flags);
  1129. bio = bio_dirty_list;
  1130. bio_dirty_list = NULL;
  1131. spin_unlock_irqrestore(&bio_dirty_lock, flags);
  1132. while (bio) {
  1133. struct bio *next = bio->bi_private;
  1134. bio_set_pages_dirty(bio);
  1135. bio_release_pages(bio);
  1136. bio_put(bio);
  1137. bio = next;
  1138. }
  1139. }
  1140. void bio_check_pages_dirty(struct bio *bio)
  1141. {
  1142. struct bio_vec *bvec = bio->bi_io_vec;
  1143. int nr_clean_pages = 0;
  1144. int i;
  1145. for (i = 0; i < bio->bi_vcnt; i++) {
  1146. struct page *page = bvec[i].bv_page;
  1147. if (PageDirty(page) || PageCompound(page)) {
  1148. page_cache_release(page);
  1149. bvec[i].bv_page = NULL;
  1150. } else {
  1151. nr_clean_pages++;
  1152. }
  1153. }
  1154. if (nr_clean_pages) {
  1155. unsigned long flags;
  1156. spin_lock_irqsave(&bio_dirty_lock, flags);
  1157. bio->bi_private = bio_dirty_list;
  1158. bio_dirty_list = bio;
  1159. spin_unlock_irqrestore(&bio_dirty_lock, flags);
  1160. schedule_work(&bio_dirty_work);
  1161. } else {
  1162. bio_put(bio);
  1163. }
  1164. }
  1165. /**
  1166. * bio_endio - end I/O on a bio
  1167. * @bio: bio
  1168. * @error: error, if any
  1169. *
  1170. * Description:
  1171. * bio_endio() will end I/O on the whole bio. bio_endio() is the
  1172. * preferred way to end I/O on a bio, it takes care of clearing
  1173. * BIO_UPTODATE on error. @error is 0 on success, and and one of the
  1174. * established -Exxxx (-EIO, for instance) error values in case
  1175. * something went wrong. Noone should call bi_end_io() directly on a
  1176. * bio unless they own it and thus know that it has an end_io
  1177. * function.
  1178. **/
  1179. void bio_endio(struct bio *bio, int error)
  1180. {
  1181. if (error)
  1182. clear_bit(BIO_UPTODATE, &bio->bi_flags);
  1183. else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
  1184. error = -EIO;
  1185. if (bio->bi_end_io)
  1186. bio->bi_end_io(bio, error);
  1187. }
  1188. void bio_pair_release(struct bio_pair *bp)
  1189. {
  1190. if (atomic_dec_and_test(&bp->cnt)) {
  1191. struct bio *master = bp->bio1.bi_private;
  1192. bio_endio(master, bp->error);
  1193. mempool_free(bp, bp->bio2.bi_private);
  1194. }
  1195. }
  1196. static void bio_pair_end_1(struct bio *bi, int err)
  1197. {
  1198. struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
  1199. if (err)
  1200. bp->error = err;
  1201. bio_pair_release(bp);
  1202. }
  1203. static void bio_pair_end_2(struct bio *bi, int err)
  1204. {
  1205. struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
  1206. if (err)
  1207. bp->error = err;
  1208. bio_pair_release(bp);
  1209. }
  1210. /*
  1211. * split a bio - only worry about a bio with a single page in its iovec
  1212. */
  1213. struct bio_pair *bio_split(struct bio *bi, int first_sectors)
  1214. {
  1215. struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
  1216. if (!bp)
  1217. return bp;
  1218. trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
  1219. bi->bi_sector + first_sectors);
  1220. BUG_ON(bi->bi_vcnt != 1);
  1221. BUG_ON(bi->bi_idx != 0);
  1222. atomic_set(&bp->cnt, 3);
  1223. bp->error = 0;
  1224. bp->bio1 = *bi;
  1225. bp->bio2 = *bi;
  1226. bp->bio2.bi_sector += first_sectors;
  1227. bp->bio2.bi_size -= first_sectors << 9;
  1228. bp->bio1.bi_size = first_sectors << 9;
  1229. bp->bv1 = bi->bi_io_vec[0];
  1230. bp->bv2 = bi->bi_io_vec[0];
  1231. bp->bv2.bv_offset += first_sectors << 9;
  1232. bp->bv2.bv_len -= first_sectors << 9;
  1233. bp->bv1.bv_len = first_sectors << 9;
  1234. bp->bio1.bi_io_vec = &bp->bv1;
  1235. bp->bio2.bi_io_vec = &bp->bv2;
  1236. bp->bio1.bi_max_vecs = 1;
  1237. bp->bio2.bi_max_vecs = 1;
  1238. bp->bio1.bi_end_io = bio_pair_end_1;
  1239. bp->bio2.bi_end_io = bio_pair_end_2;
  1240. bp->bio1.bi_private = bi;
  1241. bp->bio2.bi_private = bio_split_pool;
  1242. if (bio_integrity(bi))
  1243. bio_integrity_split(bi, bp, first_sectors);
  1244. return bp;
  1245. }
  1246. /**
  1247. * bio_sector_offset - Find hardware sector offset in bio
  1248. * @bio: bio to inspect
  1249. * @index: bio_vec index
  1250. * @offset: offset in bv_page
  1251. *
  1252. * Return the number of hardware sectors between beginning of bio
  1253. * and an end point indicated by a bio_vec index and an offset
  1254. * within that vector's page.
  1255. */
  1256. sector_t bio_sector_offset(struct bio *bio, unsigned short index,
  1257. unsigned int offset)
  1258. {
  1259. unsigned int sector_sz;
  1260. struct bio_vec *bv;
  1261. sector_t sectors;
  1262. int i;
  1263. sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
  1264. sectors = 0;
  1265. if (index >= bio->bi_idx)
  1266. index = bio->bi_vcnt - 1;
  1267. __bio_for_each_segment(bv, bio, i, 0) {
  1268. if (i == index) {
  1269. if (offset > bv->bv_offset)
  1270. sectors += (offset - bv->bv_offset) / sector_sz;
  1271. break;
  1272. }
  1273. sectors += bv->bv_len / sector_sz;
  1274. }
  1275. return sectors;
  1276. }
  1277. EXPORT_SYMBOL(bio_sector_offset);
  1278. /*
  1279. * create memory pools for biovec's in a bio_set.
  1280. * use the global biovec slabs created for general use.
  1281. */
  1282. static int biovec_create_pools(struct bio_set *bs, int pool_entries)
  1283. {
  1284. struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
  1285. bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
  1286. if (!bs->bvec_pool)
  1287. return -ENOMEM;
  1288. return 0;
  1289. }
  1290. static void biovec_free_pools(struct bio_set *bs)
  1291. {
  1292. mempool_destroy(bs->bvec_pool);
  1293. }
  1294. void bioset_free(struct bio_set *bs)
  1295. {
  1296. if (bs->bio_pool)
  1297. mempool_destroy(bs->bio_pool);
  1298. bioset_integrity_free(bs);
  1299. biovec_free_pools(bs);
  1300. bio_put_slab(bs);
  1301. kfree(bs);
  1302. }
  1303. /**
  1304. * bioset_create - Create a bio_set
  1305. * @pool_size: Number of bio and bio_vecs to cache in the mempool
  1306. * @front_pad: Number of bytes to allocate in front of the returned bio
  1307. *
  1308. * Description:
  1309. * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
  1310. * to ask for a number of bytes to be allocated in front of the bio.
  1311. * Front pad allocation is useful for embedding the bio inside
  1312. * another structure, to avoid allocating extra data to go with the bio.
  1313. * Note that the bio must be embedded at the END of that structure always,
  1314. * or things will break badly.
  1315. */
  1316. struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
  1317. {
  1318. unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
  1319. struct bio_set *bs;
  1320. bs = kzalloc(sizeof(*bs), GFP_KERNEL);
  1321. if (!bs)
  1322. return NULL;
  1323. bs->front_pad = front_pad;
  1324. bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
  1325. if (!bs->bio_slab) {
  1326. kfree(bs);
  1327. return NULL;
  1328. }
  1329. bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
  1330. if (!bs->bio_pool)
  1331. goto bad;
  1332. if (bioset_integrity_create(bs, pool_size))
  1333. goto bad;
  1334. if (!biovec_create_pools(bs, pool_size))
  1335. return bs;
  1336. bad:
  1337. bioset_free(bs);
  1338. return NULL;
  1339. }
  1340. static void __init biovec_init_slabs(void)
  1341. {
  1342. int i;
  1343. for (i = 0; i < BIOVEC_NR_POOLS; i++) {
  1344. int size;
  1345. struct biovec_slab *bvs = bvec_slabs + i;
  1346. #ifndef CONFIG_BLK_DEV_INTEGRITY
  1347. if (bvs->nr_vecs <= BIO_INLINE_VECS) {
  1348. bvs->slab = NULL;
  1349. continue;
  1350. }
  1351. #endif
  1352. size = bvs->nr_vecs * sizeof(struct bio_vec);
  1353. bvs->slab = kmem_cache_create(bvs->name, size, 0,
  1354. SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
  1355. }
  1356. }
  1357. static int __init init_bio(void)
  1358. {
  1359. bio_slab_max = 2;
  1360. bio_slab_nr = 0;
  1361. bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
  1362. if (!bio_slabs)
  1363. panic("bio: can't allocate bios\n");
  1364. bio_integrity_init();
  1365. biovec_init_slabs();
  1366. fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
  1367. if (!fs_bio_set)
  1368. panic("bio: can't allocate bios\n");
  1369. bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
  1370. sizeof(struct bio_pair));
  1371. if (!bio_split_pool)
  1372. panic("bio: can't create split pool\n");
  1373. return 0;
  1374. }
  1375. subsys_initcall(init_bio);
  1376. EXPORT_SYMBOL(bio_alloc);
  1377. EXPORT_SYMBOL(bio_kmalloc);
  1378. EXPORT_SYMBOL(bio_put);
  1379. EXPORT_SYMBOL(bio_free);
  1380. EXPORT_SYMBOL(bio_endio);
  1381. EXPORT_SYMBOL(bio_init);
  1382. EXPORT_SYMBOL(__bio_clone);
  1383. EXPORT_SYMBOL(bio_clone);
  1384. EXPORT_SYMBOL(bio_phys_segments);
  1385. EXPORT_SYMBOL(bio_add_page);
  1386. EXPORT_SYMBOL(bio_add_pc_page);
  1387. EXPORT_SYMBOL(bio_get_nr_vecs);
  1388. EXPORT_SYMBOL(bio_map_user);
  1389. EXPORT_SYMBOL(bio_unmap_user);
  1390. EXPORT_SYMBOL(bio_map_kern);
  1391. EXPORT_SYMBOL(bio_copy_kern);
  1392. EXPORT_SYMBOL(bio_pair_release);
  1393. EXPORT_SYMBOL(bio_split);
  1394. EXPORT_SYMBOL(bio_copy_user);
  1395. EXPORT_SYMBOL(bio_uncopy_user);
  1396. EXPORT_SYMBOL(bioset_create);
  1397. EXPORT_SYMBOL(bioset_free);
  1398. EXPORT_SYMBOL(bio_alloc_bioset);