extents_status.c 60 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194
  1. // SPDX-License-Identifier: GPL-2.0
  2. /*
  3. * fs/ext4/extents_status.c
  4. *
  5. * Written by Yongqiang Yang <xiaoqiangnk@gmail.com>
  6. * Modified by
  7. * Allison Henderson <achender@linux.vnet.ibm.com>
  8. * Hugh Dickins <hughd@google.com>
  9. * Zheng Liu <wenqing.lz@taobao.com>
  10. *
  11. * Ext4 extents status tree core functions.
  12. */
  13. #include <linux/list_sort.h>
  14. #include <linux/proc_fs.h>
  15. #include <linux/seq_file.h>
  16. #include "ext4.h"
  17. #include <trace/events/ext4.h>
  18. /*
  19. * According to previous discussion in Ext4 Developer Workshop, we
  20. * will introduce a new structure called io tree to track all extent
  21. * status in order to solve some problems that we have met
  22. * (e.g. Reservation space warning), and provide extent-level locking.
  23. * Delay extent tree is the first step to achieve this goal. It is
  24. * original built by Yongqiang Yang. At that time it is called delay
  25. * extent tree, whose goal is only track delayed extents in memory to
  26. * simplify the implementation of fiemap and bigalloc, and introduce
  27. * lseek SEEK_DATA/SEEK_HOLE support. That is why it is still called
  28. * delay extent tree at the first commit. But for better understand
  29. * what it does, it has been rename to extent status tree.
  30. *
  31. * Step1:
  32. * Currently the first step has been done. All delayed extents are
  33. * tracked in the tree. It maintains the delayed extent when a delayed
  34. * allocation is issued, and the delayed extent is written out or
  35. * invalidated. Therefore the implementation of fiemap and bigalloc
  36. * are simplified, and SEEK_DATA/SEEK_HOLE are introduced.
  37. *
  38. * The following comment describes the implemenmtation of extent
  39. * status tree and future works.
  40. *
  41. * Step2:
  42. * In this step all extent status are tracked by extent status tree.
  43. * Thus, we can first try to lookup a block mapping in this tree before
  44. * finding it in extent tree. Hence, single extent cache can be removed
  45. * because extent status tree can do a better job. Extents in status
  46. * tree are loaded on-demand. Therefore, the extent status tree may not
  47. * contain all of the extents in a file. Meanwhile we define a shrinker
  48. * to reclaim memory from extent status tree because fragmented extent
  49. * tree will make status tree cost too much memory. written/unwritten/-
  50. * hole extents in the tree will be reclaimed by this shrinker when we
  51. * are under high memory pressure. Delayed extents will not be
  52. * reclimed because fiemap, bigalloc, and seek_data/hole need it.
  53. */
  54. /*
  55. * Extent status tree implementation for ext4.
  56. *
  57. *
  58. * ==========================================================================
  59. * Extent status tree tracks all extent status.
  60. *
  61. * 1. Why we need to implement extent status tree?
  62. *
  63. * Without extent status tree, ext4 identifies a delayed extent by looking
  64. * up page cache, this has several deficiencies - complicated, buggy,
  65. * and inefficient code.
  66. *
  67. * FIEMAP, SEEK_HOLE/DATA, bigalloc, and writeout all need to know if a
  68. * block or a range of blocks are belonged to a delayed extent.
  69. *
  70. * Let us have a look at how they do without extent status tree.
  71. * -- FIEMAP
  72. * FIEMAP looks up page cache to identify delayed allocations from holes.
  73. *
  74. * -- SEEK_HOLE/DATA
  75. * SEEK_HOLE/DATA has the same problem as FIEMAP.
  76. *
  77. * -- bigalloc
  78. * bigalloc looks up page cache to figure out if a block is
  79. * already under delayed allocation or not to determine whether
  80. * quota reserving is needed for the cluster.
  81. *
  82. * -- writeout
  83. * Writeout looks up whole page cache to see if a buffer is
  84. * mapped, If there are not very many delayed buffers, then it is
  85. * time consuming.
  86. *
  87. * With extent status tree implementation, FIEMAP, SEEK_HOLE/DATA,
  88. * bigalloc and writeout can figure out if a block or a range of
  89. * blocks is under delayed allocation(belonged to a delayed extent) or
  90. * not by searching the extent tree.
  91. *
  92. *
  93. * ==========================================================================
  94. * 2. Ext4 extent status tree impelmentation
  95. *
  96. * -- extent
  97. * A extent is a range of blocks which are contiguous logically and
  98. * physically. Unlike extent in extent tree, this extent in ext4 is
  99. * a in-memory struct, there is no corresponding on-disk data. There
  100. * is no limit on length of extent, so an extent can contain as many
  101. * blocks as they are contiguous logically and physically.
  102. *
  103. * -- extent status tree
  104. * Every inode has an extent status tree and all allocation blocks
  105. * are added to the tree with different status. The extent in the
  106. * tree are ordered by logical block no.
  107. *
  108. * -- operations on a extent status tree
  109. * There are three important operations on a delayed extent tree: find
  110. * next extent, adding a extent(a range of blocks) and removing a extent.
  111. *
  112. * -- race on a extent status tree
  113. * Extent status tree is protected by inode->i_es_lock.
  114. *
  115. * -- memory consumption
  116. * Fragmented extent tree will make extent status tree cost too much
  117. * memory. Hence, we will reclaim written/unwritten/hole extents from
  118. * the tree under a heavy memory pressure.
  119. *
  120. *
  121. * ==========================================================================
  122. * 3. Performance analysis
  123. *
  124. * -- overhead
  125. * 1. There is a cache extent for write access, so if writes are
  126. * not very random, adding space operaions are in O(1) time.
  127. *
  128. * -- gain
  129. * 2. Code is much simpler, more readable, more maintainable and
  130. * more efficient.
  131. *
  132. *
  133. * ==========================================================================
  134. * 4. TODO list
  135. *
  136. * -- Refactor delayed space reservation
  137. *
  138. * -- Extent-level locking
  139. */
  140. static struct kmem_cache *ext4_es_cachep;
  141. static struct kmem_cache *ext4_pending_cachep;
  142. static int __es_insert_extent(struct inode *inode, struct extent_status *newes);
  143. static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
  144. ext4_lblk_t end, int *reserved);
  145. static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan);
  146. static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
  147. struct ext4_inode_info *locked_ei);
  148. static void __revise_pending(struct inode *inode, ext4_lblk_t lblk,
  149. ext4_lblk_t len);
  150. int __init ext4_init_es(void)
  151. {
  152. ext4_es_cachep = kmem_cache_create("ext4_extent_status",
  153. sizeof(struct extent_status),
  154. 0, (SLAB_RECLAIM_ACCOUNT), NULL);
  155. if (ext4_es_cachep == NULL)
  156. return -ENOMEM;
  157. return 0;
  158. }
  159. void ext4_exit_es(void)
  160. {
  161. kmem_cache_destroy(ext4_es_cachep);
  162. }
  163. void ext4_es_init_tree(struct ext4_es_tree *tree)
  164. {
  165. tree->root = RB_ROOT;
  166. tree->cache_es = NULL;
  167. }
  168. #ifdef ES_DEBUG__
  169. static void ext4_es_print_tree(struct inode *inode)
  170. {
  171. struct ext4_es_tree *tree;
  172. struct rb_node *node;
  173. printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino);
  174. tree = &EXT4_I(inode)->i_es_tree;
  175. node = rb_first(&tree->root);
  176. while (node) {
  177. struct extent_status *es;
  178. es = rb_entry(node, struct extent_status, rb_node);
  179. printk(KERN_DEBUG " [%u/%u) %llu %x",
  180. es->es_lblk, es->es_len,
  181. ext4_es_pblock(es), ext4_es_status(es));
  182. node = rb_next(node);
  183. }
  184. printk(KERN_DEBUG "\n");
  185. }
  186. #else
  187. #define ext4_es_print_tree(inode)
  188. #endif
  189. static inline ext4_lblk_t ext4_es_end(struct extent_status *es)
  190. {
  191. BUG_ON(es->es_lblk + es->es_len < es->es_lblk);
  192. return es->es_lblk + es->es_len - 1;
  193. }
  194. /*
  195. * search through the tree for an delayed extent with a given offset. If
  196. * it can't be found, try to find next extent.
  197. */
  198. static struct extent_status *__es_tree_search(struct rb_root *root,
  199. ext4_lblk_t lblk)
  200. {
  201. struct rb_node *node = root->rb_node;
  202. struct extent_status *es = NULL;
  203. while (node) {
  204. es = rb_entry(node, struct extent_status, rb_node);
  205. if (lblk < es->es_lblk)
  206. node = node->rb_left;
  207. else if (lblk > ext4_es_end(es))
  208. node = node->rb_right;
  209. else
  210. return es;
  211. }
  212. if (es && lblk < es->es_lblk)
  213. return es;
  214. if (es && lblk > ext4_es_end(es)) {
  215. node = rb_next(&es->rb_node);
  216. return node ? rb_entry(node, struct extent_status, rb_node) :
  217. NULL;
  218. }
  219. return NULL;
  220. }
  221. /*
  222. * ext4_es_find_extent_range - find extent with specified status within block
  223. * range or next extent following block range in
  224. * extents status tree
  225. *
  226. * @inode - file containing the range
  227. * @matching_fn - pointer to function that matches extents with desired status
  228. * @lblk - logical block defining start of range
  229. * @end - logical block defining end of range
  230. * @es - extent found, if any
  231. *
  232. * Find the first extent within the block range specified by @lblk and @end
  233. * in the extents status tree that satisfies @matching_fn. If a match
  234. * is found, it's returned in @es. If not, and a matching extent is found
  235. * beyond the block range, it's returned in @es. If no match is found, an
  236. * extent is returned in @es whose es_lblk, es_len, and es_pblk components
  237. * are 0.
  238. */
  239. static void __es_find_extent_range(struct inode *inode,
  240. int (*matching_fn)(struct extent_status *es),
  241. ext4_lblk_t lblk, ext4_lblk_t end,
  242. struct extent_status *es)
  243. {
  244. struct ext4_es_tree *tree = NULL;
  245. struct extent_status *es1 = NULL;
  246. struct rb_node *node;
  247. WARN_ON(es == NULL);
  248. WARN_ON(end < lblk);
  249. tree = &EXT4_I(inode)->i_es_tree;
  250. /* see if the extent has been cached */
  251. es->es_lblk = es->es_len = es->es_pblk = 0;
  252. if (tree->cache_es) {
  253. es1 = tree->cache_es;
  254. if (in_range(lblk, es1->es_lblk, es1->es_len)) {
  255. es_debug("%u cached by [%u/%u) %llu %x\n",
  256. lblk, es1->es_lblk, es1->es_len,
  257. ext4_es_pblock(es1), ext4_es_status(es1));
  258. goto out;
  259. }
  260. }
  261. es1 = __es_tree_search(&tree->root, lblk);
  262. out:
  263. if (es1 && !matching_fn(es1)) {
  264. while ((node = rb_next(&es1->rb_node)) != NULL) {
  265. es1 = rb_entry(node, struct extent_status, rb_node);
  266. if (es1->es_lblk > end) {
  267. es1 = NULL;
  268. break;
  269. }
  270. if (matching_fn(es1))
  271. break;
  272. }
  273. }
  274. if (es1 && matching_fn(es1)) {
  275. tree->cache_es = es1;
  276. es->es_lblk = es1->es_lblk;
  277. es->es_len = es1->es_len;
  278. es->es_pblk = es1->es_pblk;
  279. }
  280. }
  281. /*
  282. * Locking for __es_find_extent_range() for external use
  283. */
  284. void ext4_es_find_extent_range(struct inode *inode,
  285. int (*matching_fn)(struct extent_status *es),
  286. ext4_lblk_t lblk, ext4_lblk_t end,
  287. struct extent_status *es)
  288. {
  289. if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
  290. return;
  291. trace_ext4_es_find_extent_range_enter(inode, lblk);
  292. read_lock(&EXT4_I(inode)->i_es_lock);
  293. __es_find_extent_range(inode, matching_fn, lblk, end, es);
  294. read_unlock(&EXT4_I(inode)->i_es_lock);
  295. trace_ext4_es_find_extent_range_exit(inode, es);
  296. }
  297. /*
  298. * __es_scan_range - search block range for block with specified status
  299. * in extents status tree
  300. *
  301. * @inode - file containing the range
  302. * @matching_fn - pointer to function that matches extents with desired status
  303. * @lblk - logical block defining start of range
  304. * @end - logical block defining end of range
  305. *
  306. * Returns true if at least one block in the specified block range satisfies
  307. * the criterion specified by @matching_fn, and false if not. If at least
  308. * one extent has the specified status, then there is at least one block
  309. * in the cluster with that status. Should only be called by code that has
  310. * taken i_es_lock.
  311. */
  312. static bool __es_scan_range(struct inode *inode,
  313. int (*matching_fn)(struct extent_status *es),
  314. ext4_lblk_t start, ext4_lblk_t end)
  315. {
  316. struct extent_status es;
  317. __es_find_extent_range(inode, matching_fn, start, end, &es);
  318. if (es.es_len == 0)
  319. return false; /* no matching extent in the tree */
  320. else if (es.es_lblk <= start &&
  321. start < es.es_lblk + es.es_len)
  322. return true;
  323. else if (start <= es.es_lblk && es.es_lblk <= end)
  324. return true;
  325. else
  326. return false;
  327. }
  328. /*
  329. * Locking for __es_scan_range() for external use
  330. */
  331. bool ext4_es_scan_range(struct inode *inode,
  332. int (*matching_fn)(struct extent_status *es),
  333. ext4_lblk_t lblk, ext4_lblk_t end)
  334. {
  335. bool ret;
  336. if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
  337. return false;
  338. read_lock(&EXT4_I(inode)->i_es_lock);
  339. ret = __es_scan_range(inode, matching_fn, lblk, end);
  340. read_unlock(&EXT4_I(inode)->i_es_lock);
  341. return ret;
  342. }
  343. /*
  344. * __es_scan_clu - search cluster for block with specified status in
  345. * extents status tree
  346. *
  347. * @inode - file containing the cluster
  348. * @matching_fn - pointer to function that matches extents with desired status
  349. * @lblk - logical block in cluster to be searched
  350. *
  351. * Returns true if at least one extent in the cluster containing @lblk
  352. * satisfies the criterion specified by @matching_fn, and false if not. If at
  353. * least one extent has the specified status, then there is at least one block
  354. * in the cluster with that status. Should only be called by code that has
  355. * taken i_es_lock.
  356. */
  357. static bool __es_scan_clu(struct inode *inode,
  358. int (*matching_fn)(struct extent_status *es),
  359. ext4_lblk_t lblk)
  360. {
  361. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  362. ext4_lblk_t lblk_start, lblk_end;
  363. lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
  364. lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
  365. return __es_scan_range(inode, matching_fn, lblk_start, lblk_end);
  366. }
  367. /*
  368. * Locking for __es_scan_clu() for external use
  369. */
  370. bool ext4_es_scan_clu(struct inode *inode,
  371. int (*matching_fn)(struct extent_status *es),
  372. ext4_lblk_t lblk)
  373. {
  374. bool ret;
  375. if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
  376. return false;
  377. read_lock(&EXT4_I(inode)->i_es_lock);
  378. ret = __es_scan_clu(inode, matching_fn, lblk);
  379. read_unlock(&EXT4_I(inode)->i_es_lock);
  380. return ret;
  381. }
  382. static void ext4_es_list_add(struct inode *inode)
  383. {
  384. struct ext4_inode_info *ei = EXT4_I(inode);
  385. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  386. if (!list_empty(&ei->i_es_list))
  387. return;
  388. spin_lock(&sbi->s_es_lock);
  389. if (list_empty(&ei->i_es_list)) {
  390. list_add_tail(&ei->i_es_list, &sbi->s_es_list);
  391. sbi->s_es_nr_inode++;
  392. }
  393. spin_unlock(&sbi->s_es_lock);
  394. }
  395. static void ext4_es_list_del(struct inode *inode)
  396. {
  397. struct ext4_inode_info *ei = EXT4_I(inode);
  398. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  399. spin_lock(&sbi->s_es_lock);
  400. if (!list_empty(&ei->i_es_list)) {
  401. list_del_init(&ei->i_es_list);
  402. sbi->s_es_nr_inode--;
  403. WARN_ON_ONCE(sbi->s_es_nr_inode < 0);
  404. }
  405. spin_unlock(&sbi->s_es_lock);
  406. }
  407. static struct extent_status *
  408. ext4_es_alloc_extent(struct inode *inode, ext4_lblk_t lblk, ext4_lblk_t len,
  409. ext4_fsblk_t pblk)
  410. {
  411. struct extent_status *es;
  412. es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC);
  413. if (es == NULL)
  414. return NULL;
  415. es->es_lblk = lblk;
  416. es->es_len = len;
  417. es->es_pblk = pblk;
  418. /*
  419. * We don't count delayed extent because we never try to reclaim them
  420. */
  421. if (!ext4_es_is_delayed(es)) {
  422. if (!EXT4_I(inode)->i_es_shk_nr++)
  423. ext4_es_list_add(inode);
  424. percpu_counter_inc(&EXT4_SB(inode->i_sb)->
  425. s_es_stats.es_stats_shk_cnt);
  426. }
  427. EXT4_I(inode)->i_es_all_nr++;
  428. percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
  429. return es;
  430. }
  431. static void ext4_es_free_extent(struct inode *inode, struct extent_status *es)
  432. {
  433. EXT4_I(inode)->i_es_all_nr--;
  434. percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
  435. /* Decrease the shrink counter when this es is not delayed */
  436. if (!ext4_es_is_delayed(es)) {
  437. BUG_ON(EXT4_I(inode)->i_es_shk_nr == 0);
  438. if (!--EXT4_I(inode)->i_es_shk_nr)
  439. ext4_es_list_del(inode);
  440. percpu_counter_dec(&EXT4_SB(inode->i_sb)->
  441. s_es_stats.es_stats_shk_cnt);
  442. }
  443. kmem_cache_free(ext4_es_cachep, es);
  444. }
  445. /*
  446. * Check whether or not two extents can be merged
  447. * Condition:
  448. * - logical block number is contiguous
  449. * - physical block number is contiguous
  450. * - status is equal
  451. */
  452. static int ext4_es_can_be_merged(struct extent_status *es1,
  453. struct extent_status *es2)
  454. {
  455. if (ext4_es_type(es1) != ext4_es_type(es2))
  456. return 0;
  457. if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) {
  458. pr_warn("ES assertion failed when merging extents. "
  459. "The sum of lengths of es1 (%d) and es2 (%d) "
  460. "is bigger than allowed file size (%d)\n",
  461. es1->es_len, es2->es_len, EXT_MAX_BLOCKS);
  462. WARN_ON(1);
  463. return 0;
  464. }
  465. if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk)
  466. return 0;
  467. if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) &&
  468. (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2)))
  469. return 1;
  470. if (ext4_es_is_hole(es1))
  471. return 1;
  472. /* we need to check delayed extent is without unwritten status */
  473. if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1))
  474. return 1;
  475. return 0;
  476. }
  477. static struct extent_status *
  478. ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es)
  479. {
  480. struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
  481. struct extent_status *es1;
  482. struct rb_node *node;
  483. node = rb_prev(&es->rb_node);
  484. if (!node)
  485. return es;
  486. es1 = rb_entry(node, struct extent_status, rb_node);
  487. if (ext4_es_can_be_merged(es1, es)) {
  488. es1->es_len += es->es_len;
  489. if (ext4_es_is_referenced(es))
  490. ext4_es_set_referenced(es1);
  491. rb_erase(&es->rb_node, &tree->root);
  492. ext4_es_free_extent(inode, es);
  493. es = es1;
  494. }
  495. return es;
  496. }
  497. static struct extent_status *
  498. ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es)
  499. {
  500. struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
  501. struct extent_status *es1;
  502. struct rb_node *node;
  503. node = rb_next(&es->rb_node);
  504. if (!node)
  505. return es;
  506. es1 = rb_entry(node, struct extent_status, rb_node);
  507. if (ext4_es_can_be_merged(es, es1)) {
  508. es->es_len += es1->es_len;
  509. if (ext4_es_is_referenced(es1))
  510. ext4_es_set_referenced(es);
  511. rb_erase(node, &tree->root);
  512. ext4_es_free_extent(inode, es1);
  513. }
  514. return es;
  515. }
  516. #ifdef ES_AGGRESSIVE_TEST
  517. #include "ext4_extents.h" /* Needed when ES_AGGRESSIVE_TEST is defined */
  518. static void ext4_es_insert_extent_ext_check(struct inode *inode,
  519. struct extent_status *es)
  520. {
  521. struct ext4_ext_path *path = NULL;
  522. struct ext4_extent *ex;
  523. ext4_lblk_t ee_block;
  524. ext4_fsblk_t ee_start;
  525. unsigned short ee_len;
  526. int depth, ee_status, es_status;
  527. path = ext4_find_extent(inode, es->es_lblk, NULL, EXT4_EX_NOCACHE);
  528. if (IS_ERR(path))
  529. return;
  530. depth = ext_depth(inode);
  531. ex = path[depth].p_ext;
  532. if (ex) {
  533. ee_block = le32_to_cpu(ex->ee_block);
  534. ee_start = ext4_ext_pblock(ex);
  535. ee_len = ext4_ext_get_actual_len(ex);
  536. ee_status = ext4_ext_is_unwritten(ex) ? 1 : 0;
  537. es_status = ext4_es_is_unwritten(es) ? 1 : 0;
  538. /*
  539. * Make sure ex and es are not overlap when we try to insert
  540. * a delayed/hole extent.
  541. */
  542. if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) {
  543. if (in_range(es->es_lblk, ee_block, ee_len)) {
  544. pr_warn("ES insert assertion failed for "
  545. "inode: %lu we can find an extent "
  546. "at block [%d/%d/%llu/%c], but we "
  547. "want to add a delayed/hole extent "
  548. "[%d/%d/%llu/%x]\n",
  549. inode->i_ino, ee_block, ee_len,
  550. ee_start, ee_status ? 'u' : 'w',
  551. es->es_lblk, es->es_len,
  552. ext4_es_pblock(es), ext4_es_status(es));
  553. }
  554. goto out;
  555. }
  556. /*
  557. * We don't check ee_block == es->es_lblk, etc. because es
  558. * might be a part of whole extent, vice versa.
  559. */
  560. if (es->es_lblk < ee_block ||
  561. ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) {
  562. pr_warn("ES insert assertion failed for inode: %lu "
  563. "ex_status [%d/%d/%llu/%c] != "
  564. "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
  565. ee_block, ee_len, ee_start,
  566. ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
  567. ext4_es_pblock(es), es_status ? 'u' : 'w');
  568. goto out;
  569. }
  570. if (ee_status ^ es_status) {
  571. pr_warn("ES insert assertion failed for inode: %lu "
  572. "ex_status [%d/%d/%llu/%c] != "
  573. "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
  574. ee_block, ee_len, ee_start,
  575. ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
  576. ext4_es_pblock(es), es_status ? 'u' : 'w');
  577. }
  578. } else {
  579. /*
  580. * We can't find an extent on disk. So we need to make sure
  581. * that we don't want to add an written/unwritten extent.
  582. */
  583. if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) {
  584. pr_warn("ES insert assertion failed for inode: %lu "
  585. "can't find an extent at block %d but we want "
  586. "to add a written/unwritten extent "
  587. "[%d/%d/%llu/%x]\n", inode->i_ino,
  588. es->es_lblk, es->es_lblk, es->es_len,
  589. ext4_es_pblock(es), ext4_es_status(es));
  590. }
  591. }
  592. out:
  593. ext4_ext_drop_refs(path);
  594. kfree(path);
  595. }
  596. static void ext4_es_insert_extent_ind_check(struct inode *inode,
  597. struct extent_status *es)
  598. {
  599. struct ext4_map_blocks map;
  600. int retval;
  601. /*
  602. * Here we call ext4_ind_map_blocks to lookup a block mapping because
  603. * 'Indirect' structure is defined in indirect.c. So we couldn't
  604. * access direct/indirect tree from outside. It is too dirty to define
  605. * this function in indirect.c file.
  606. */
  607. map.m_lblk = es->es_lblk;
  608. map.m_len = es->es_len;
  609. retval = ext4_ind_map_blocks(NULL, inode, &map, 0);
  610. if (retval > 0) {
  611. if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) {
  612. /*
  613. * We want to add a delayed/hole extent but this
  614. * block has been allocated.
  615. */
  616. pr_warn("ES insert assertion failed for inode: %lu "
  617. "We can find blocks but we want to add a "
  618. "delayed/hole extent [%d/%d/%llu/%x]\n",
  619. inode->i_ino, es->es_lblk, es->es_len,
  620. ext4_es_pblock(es), ext4_es_status(es));
  621. return;
  622. } else if (ext4_es_is_written(es)) {
  623. if (retval != es->es_len) {
  624. pr_warn("ES insert assertion failed for "
  625. "inode: %lu retval %d != es_len %d\n",
  626. inode->i_ino, retval, es->es_len);
  627. return;
  628. }
  629. if (map.m_pblk != ext4_es_pblock(es)) {
  630. pr_warn("ES insert assertion failed for "
  631. "inode: %lu m_pblk %llu != "
  632. "es_pblk %llu\n",
  633. inode->i_ino, map.m_pblk,
  634. ext4_es_pblock(es));
  635. return;
  636. }
  637. } else {
  638. /*
  639. * We don't need to check unwritten extent because
  640. * indirect-based file doesn't have it.
  641. */
  642. BUG();
  643. }
  644. } else if (retval == 0) {
  645. if (ext4_es_is_written(es)) {
  646. pr_warn("ES insert assertion failed for inode: %lu "
  647. "We can't find the block but we want to add "
  648. "a written extent [%d/%d/%llu/%x]\n",
  649. inode->i_ino, es->es_lblk, es->es_len,
  650. ext4_es_pblock(es), ext4_es_status(es));
  651. return;
  652. }
  653. }
  654. }
  655. static inline void ext4_es_insert_extent_check(struct inode *inode,
  656. struct extent_status *es)
  657. {
  658. /*
  659. * We don't need to worry about the race condition because
  660. * caller takes i_data_sem locking.
  661. */
  662. BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
  663. if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
  664. ext4_es_insert_extent_ext_check(inode, es);
  665. else
  666. ext4_es_insert_extent_ind_check(inode, es);
  667. }
  668. #else
  669. static inline void ext4_es_insert_extent_check(struct inode *inode,
  670. struct extent_status *es)
  671. {
  672. }
  673. #endif
  674. static int __es_insert_extent(struct inode *inode, struct extent_status *newes)
  675. {
  676. struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
  677. struct rb_node **p = &tree->root.rb_node;
  678. struct rb_node *parent = NULL;
  679. struct extent_status *es;
  680. while (*p) {
  681. parent = *p;
  682. es = rb_entry(parent, struct extent_status, rb_node);
  683. if (newes->es_lblk < es->es_lblk) {
  684. if (ext4_es_can_be_merged(newes, es)) {
  685. /*
  686. * Here we can modify es_lblk directly
  687. * because it isn't overlapped.
  688. */
  689. es->es_lblk = newes->es_lblk;
  690. es->es_len += newes->es_len;
  691. if (ext4_es_is_written(es) ||
  692. ext4_es_is_unwritten(es))
  693. ext4_es_store_pblock(es,
  694. newes->es_pblk);
  695. es = ext4_es_try_to_merge_left(inode, es);
  696. goto out;
  697. }
  698. p = &(*p)->rb_left;
  699. } else if (newes->es_lblk > ext4_es_end(es)) {
  700. if (ext4_es_can_be_merged(es, newes)) {
  701. es->es_len += newes->es_len;
  702. es = ext4_es_try_to_merge_right(inode, es);
  703. goto out;
  704. }
  705. p = &(*p)->rb_right;
  706. } else {
  707. BUG();
  708. return -EINVAL;
  709. }
  710. }
  711. es = ext4_es_alloc_extent(inode, newes->es_lblk, newes->es_len,
  712. newes->es_pblk);
  713. if (!es)
  714. return -ENOMEM;
  715. rb_link_node(&es->rb_node, parent, p);
  716. rb_insert_color(&es->rb_node, &tree->root);
  717. out:
  718. tree->cache_es = es;
  719. return 0;
  720. }
  721. /*
  722. * ext4_es_insert_extent() adds information to an inode's extent
  723. * status tree.
  724. *
  725. * Return 0 on success, error code on failure.
  726. */
  727. int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk,
  728. ext4_lblk_t len, ext4_fsblk_t pblk,
  729. unsigned int status)
  730. {
  731. struct extent_status newes;
  732. ext4_lblk_t end = lblk + len - 1;
  733. int err = 0;
  734. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  735. if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
  736. return 0;
  737. es_debug("add [%u/%u) %llu %x to extent status tree of inode %lu\n",
  738. lblk, len, pblk, status, inode->i_ino);
  739. if (!len)
  740. return 0;
  741. BUG_ON(end < lblk);
  742. if ((status & EXTENT_STATUS_DELAYED) &&
  743. (status & EXTENT_STATUS_WRITTEN)) {
  744. ext4_warning(inode->i_sb, "Inserting extent [%u/%u] as "
  745. " delayed and written which can potentially "
  746. " cause data loss.", lblk, len);
  747. WARN_ON(1);
  748. }
  749. newes.es_lblk = lblk;
  750. newes.es_len = len;
  751. ext4_es_store_pblock_status(&newes, pblk, status);
  752. trace_ext4_es_insert_extent(inode, &newes);
  753. ext4_es_insert_extent_check(inode, &newes);
  754. write_lock(&EXT4_I(inode)->i_es_lock);
  755. err = __es_remove_extent(inode, lblk, end, NULL);
  756. if (err != 0)
  757. goto error;
  758. retry:
  759. err = __es_insert_extent(inode, &newes);
  760. if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb),
  761. 128, EXT4_I(inode)))
  762. goto retry;
  763. if (err == -ENOMEM && !ext4_es_is_delayed(&newes))
  764. err = 0;
  765. if (sbi->s_cluster_ratio > 1 && test_opt(inode->i_sb, DELALLOC) &&
  766. (status & EXTENT_STATUS_WRITTEN ||
  767. status & EXTENT_STATUS_UNWRITTEN))
  768. __revise_pending(inode, lblk, len);
  769. error:
  770. write_unlock(&EXT4_I(inode)->i_es_lock);
  771. ext4_es_print_tree(inode);
  772. return err;
  773. }
  774. /*
  775. * ext4_es_cache_extent() inserts information into the extent status
  776. * tree if and only if there isn't information about the range in
  777. * question already.
  778. */
  779. void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
  780. ext4_lblk_t len, ext4_fsblk_t pblk,
  781. unsigned int status)
  782. {
  783. struct extent_status *es;
  784. struct extent_status newes;
  785. ext4_lblk_t end = lblk + len - 1;
  786. if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
  787. return;
  788. newes.es_lblk = lblk;
  789. newes.es_len = len;
  790. ext4_es_store_pblock_status(&newes, pblk, status);
  791. trace_ext4_es_cache_extent(inode, &newes);
  792. if (!len)
  793. return;
  794. BUG_ON(end < lblk);
  795. write_lock(&EXT4_I(inode)->i_es_lock);
  796. es = __es_tree_search(&EXT4_I(inode)->i_es_tree.root, lblk);
  797. if (!es || es->es_lblk > end)
  798. __es_insert_extent(inode, &newes);
  799. write_unlock(&EXT4_I(inode)->i_es_lock);
  800. }
  801. /*
  802. * ext4_es_lookup_extent() looks up an extent in extent status tree.
  803. *
  804. * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks.
  805. *
  806. * Return: 1 on found, 0 on not
  807. */
  808. int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
  809. ext4_lblk_t *next_lblk,
  810. struct extent_status *es)
  811. {
  812. struct ext4_es_tree *tree;
  813. struct ext4_es_stats *stats;
  814. struct extent_status *es1 = NULL;
  815. struct rb_node *node;
  816. int found = 0;
  817. if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
  818. return 0;
  819. trace_ext4_es_lookup_extent_enter(inode, lblk);
  820. es_debug("lookup extent in block %u\n", lblk);
  821. tree = &EXT4_I(inode)->i_es_tree;
  822. read_lock(&EXT4_I(inode)->i_es_lock);
  823. /* find extent in cache firstly */
  824. es->es_lblk = es->es_len = es->es_pblk = 0;
  825. if (tree->cache_es) {
  826. es1 = tree->cache_es;
  827. if (in_range(lblk, es1->es_lblk, es1->es_len)) {
  828. es_debug("%u cached by [%u/%u)\n",
  829. lblk, es1->es_lblk, es1->es_len);
  830. found = 1;
  831. goto out;
  832. }
  833. }
  834. node = tree->root.rb_node;
  835. while (node) {
  836. es1 = rb_entry(node, struct extent_status, rb_node);
  837. if (lblk < es1->es_lblk)
  838. node = node->rb_left;
  839. else if (lblk > ext4_es_end(es1))
  840. node = node->rb_right;
  841. else {
  842. found = 1;
  843. break;
  844. }
  845. }
  846. out:
  847. stats = &EXT4_SB(inode->i_sb)->s_es_stats;
  848. if (found) {
  849. BUG_ON(!es1);
  850. es->es_lblk = es1->es_lblk;
  851. es->es_len = es1->es_len;
  852. es->es_pblk = es1->es_pblk;
  853. if (!ext4_es_is_referenced(es1))
  854. ext4_es_set_referenced(es1);
  855. percpu_counter_inc(&stats->es_stats_cache_hits);
  856. if (next_lblk) {
  857. node = rb_next(&es1->rb_node);
  858. if (node) {
  859. es1 = rb_entry(node, struct extent_status,
  860. rb_node);
  861. *next_lblk = es1->es_lblk;
  862. } else
  863. *next_lblk = 0;
  864. }
  865. } else {
  866. percpu_counter_inc(&stats->es_stats_cache_misses);
  867. }
  868. read_unlock(&EXT4_I(inode)->i_es_lock);
  869. trace_ext4_es_lookup_extent_exit(inode, es, found);
  870. return found;
  871. }
  872. struct rsvd_count {
  873. int ndelonly;
  874. bool first_do_lblk_found;
  875. ext4_lblk_t first_do_lblk;
  876. ext4_lblk_t last_do_lblk;
  877. struct extent_status *left_es;
  878. bool partial;
  879. ext4_lblk_t lclu;
  880. };
  881. /*
  882. * init_rsvd - initialize reserved count data before removing block range
  883. * in file from extent status tree
  884. *
  885. * @inode - file containing range
  886. * @lblk - first block in range
  887. * @es - pointer to first extent in range
  888. * @rc - pointer to reserved count data
  889. *
  890. * Assumes es is not NULL
  891. */
  892. static void init_rsvd(struct inode *inode, ext4_lblk_t lblk,
  893. struct extent_status *es, struct rsvd_count *rc)
  894. {
  895. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  896. struct rb_node *node;
  897. rc->ndelonly = 0;
  898. /*
  899. * for bigalloc, note the first delonly block in the range has not
  900. * been found, record the extent containing the block to the left of
  901. * the region to be removed, if any, and note that there's no partial
  902. * cluster to track
  903. */
  904. if (sbi->s_cluster_ratio > 1) {
  905. rc->first_do_lblk_found = false;
  906. if (lblk > es->es_lblk) {
  907. rc->left_es = es;
  908. } else {
  909. node = rb_prev(&es->rb_node);
  910. rc->left_es = node ? rb_entry(node,
  911. struct extent_status,
  912. rb_node) : NULL;
  913. }
  914. rc->partial = false;
  915. }
  916. }
  917. /*
  918. * count_rsvd - count the clusters containing delayed and not unwritten
  919. * (delonly) blocks in a range within an extent and add to
  920. * the running tally in rsvd_count
  921. *
  922. * @inode - file containing extent
  923. * @lblk - first block in range
  924. * @len - length of range in blocks
  925. * @es - pointer to extent containing clusters to be counted
  926. * @rc - pointer to reserved count data
  927. *
  928. * Tracks partial clusters found at the beginning and end of extents so
  929. * they aren't overcounted when they span adjacent extents
  930. */
  931. static void count_rsvd(struct inode *inode, ext4_lblk_t lblk, long len,
  932. struct extent_status *es, struct rsvd_count *rc)
  933. {
  934. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  935. ext4_lblk_t i, end, nclu;
  936. if (!ext4_es_is_delonly(es))
  937. return;
  938. WARN_ON(len <= 0);
  939. if (sbi->s_cluster_ratio == 1) {
  940. rc->ndelonly += (int) len;
  941. return;
  942. }
  943. /* bigalloc */
  944. i = (lblk < es->es_lblk) ? es->es_lblk : lblk;
  945. end = lblk + (ext4_lblk_t) len - 1;
  946. end = (end > ext4_es_end(es)) ? ext4_es_end(es) : end;
  947. /* record the first block of the first delonly extent seen */
  948. if (!rc->first_do_lblk_found) {
  949. rc->first_do_lblk = i;
  950. rc->first_do_lblk_found = true;
  951. }
  952. /* update the last lblk in the region seen so far */
  953. rc->last_do_lblk = end;
  954. /*
  955. * if we're tracking a partial cluster and the current extent
  956. * doesn't start with it, count it and stop tracking
  957. */
  958. if (rc->partial && (rc->lclu != EXT4_B2C(sbi, i))) {
  959. rc->ndelonly++;
  960. rc->partial = false;
  961. }
  962. /*
  963. * if the first cluster doesn't start on a cluster boundary but
  964. * ends on one, count it
  965. */
  966. if (EXT4_LBLK_COFF(sbi, i) != 0) {
  967. if (end >= EXT4_LBLK_CFILL(sbi, i)) {
  968. rc->ndelonly++;
  969. rc->partial = false;
  970. i = EXT4_LBLK_CFILL(sbi, i) + 1;
  971. }
  972. }
  973. /*
  974. * if the current cluster starts on a cluster boundary, count the
  975. * number of whole delonly clusters in the extent
  976. */
  977. if ((i + sbi->s_cluster_ratio - 1) <= end) {
  978. nclu = (end - i + 1) >> sbi->s_cluster_bits;
  979. rc->ndelonly += nclu;
  980. i += nclu << sbi->s_cluster_bits;
  981. }
  982. /*
  983. * start tracking a partial cluster if there's a partial at the end
  984. * of the current extent and we're not already tracking one
  985. */
  986. if (!rc->partial && i <= end) {
  987. rc->partial = true;
  988. rc->lclu = EXT4_B2C(sbi, i);
  989. }
  990. }
  991. /*
  992. * __pr_tree_search - search for a pending cluster reservation
  993. *
  994. * @root - root of pending reservation tree
  995. * @lclu - logical cluster to search for
  996. *
  997. * Returns the pending reservation for the cluster identified by @lclu
  998. * if found. If not, returns a reservation for the next cluster if any,
  999. * and if not, returns NULL.
  1000. */
  1001. static struct pending_reservation *__pr_tree_search(struct rb_root *root,
  1002. ext4_lblk_t lclu)
  1003. {
  1004. struct rb_node *node = root->rb_node;
  1005. struct pending_reservation *pr = NULL;
  1006. while (node) {
  1007. pr = rb_entry(node, struct pending_reservation, rb_node);
  1008. if (lclu < pr->lclu)
  1009. node = node->rb_left;
  1010. else if (lclu > pr->lclu)
  1011. node = node->rb_right;
  1012. else
  1013. return pr;
  1014. }
  1015. if (pr && lclu < pr->lclu)
  1016. return pr;
  1017. if (pr && lclu > pr->lclu) {
  1018. node = rb_next(&pr->rb_node);
  1019. return node ? rb_entry(node, struct pending_reservation,
  1020. rb_node) : NULL;
  1021. }
  1022. return NULL;
  1023. }
  1024. /*
  1025. * get_rsvd - calculates and returns the number of cluster reservations to be
  1026. * released when removing a block range from the extent status tree
  1027. * and releases any pending reservations within the range
  1028. *
  1029. * @inode - file containing block range
  1030. * @end - last block in range
  1031. * @right_es - pointer to extent containing next block beyond end or NULL
  1032. * @rc - pointer to reserved count data
  1033. *
  1034. * The number of reservations to be released is equal to the number of
  1035. * clusters containing delayed and not unwritten (delonly) blocks within
  1036. * the range, minus the number of clusters still containing delonly blocks
  1037. * at the ends of the range, and minus the number of pending reservations
  1038. * within the range.
  1039. */
  1040. static unsigned int get_rsvd(struct inode *inode, ext4_lblk_t end,
  1041. struct extent_status *right_es,
  1042. struct rsvd_count *rc)
  1043. {
  1044. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  1045. struct pending_reservation *pr;
  1046. struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
  1047. struct rb_node *node;
  1048. ext4_lblk_t first_lclu, last_lclu;
  1049. bool left_delonly, right_delonly, count_pending;
  1050. struct extent_status *es;
  1051. if (sbi->s_cluster_ratio > 1) {
  1052. /* count any remaining partial cluster */
  1053. if (rc->partial)
  1054. rc->ndelonly++;
  1055. if (rc->ndelonly == 0)
  1056. return 0;
  1057. first_lclu = EXT4_B2C(sbi, rc->first_do_lblk);
  1058. last_lclu = EXT4_B2C(sbi, rc->last_do_lblk);
  1059. /*
  1060. * decrease the delonly count by the number of clusters at the
  1061. * ends of the range that still contain delonly blocks -
  1062. * these clusters still need to be reserved
  1063. */
  1064. left_delonly = right_delonly = false;
  1065. es = rc->left_es;
  1066. while (es && ext4_es_end(es) >=
  1067. EXT4_LBLK_CMASK(sbi, rc->first_do_lblk)) {
  1068. if (ext4_es_is_delonly(es)) {
  1069. rc->ndelonly--;
  1070. left_delonly = true;
  1071. break;
  1072. }
  1073. node = rb_prev(&es->rb_node);
  1074. if (!node)
  1075. break;
  1076. es = rb_entry(node, struct extent_status, rb_node);
  1077. }
  1078. if (right_es && (!left_delonly || first_lclu != last_lclu)) {
  1079. if (end < ext4_es_end(right_es)) {
  1080. es = right_es;
  1081. } else {
  1082. node = rb_next(&right_es->rb_node);
  1083. es = node ? rb_entry(node, struct extent_status,
  1084. rb_node) : NULL;
  1085. }
  1086. while (es && es->es_lblk <=
  1087. EXT4_LBLK_CFILL(sbi, rc->last_do_lblk)) {
  1088. if (ext4_es_is_delonly(es)) {
  1089. rc->ndelonly--;
  1090. right_delonly = true;
  1091. break;
  1092. }
  1093. node = rb_next(&es->rb_node);
  1094. if (!node)
  1095. break;
  1096. es = rb_entry(node, struct extent_status,
  1097. rb_node);
  1098. }
  1099. }
  1100. /*
  1101. * Determine the block range that should be searched for
  1102. * pending reservations, if any. Clusters on the ends of the
  1103. * original removed range containing delonly blocks are
  1104. * excluded. They've already been accounted for and it's not
  1105. * possible to determine if an associated pending reservation
  1106. * should be released with the information available in the
  1107. * extents status tree.
  1108. */
  1109. if (first_lclu == last_lclu) {
  1110. if (left_delonly | right_delonly)
  1111. count_pending = false;
  1112. else
  1113. count_pending = true;
  1114. } else {
  1115. if (left_delonly)
  1116. first_lclu++;
  1117. if (right_delonly)
  1118. last_lclu--;
  1119. if (first_lclu <= last_lclu)
  1120. count_pending = true;
  1121. else
  1122. count_pending = false;
  1123. }
  1124. /*
  1125. * a pending reservation found between first_lclu and last_lclu
  1126. * represents an allocated cluster that contained at least one
  1127. * delonly block, so the delonly total must be reduced by one
  1128. * for each pending reservation found and released
  1129. */
  1130. if (count_pending) {
  1131. pr = __pr_tree_search(&tree->root, first_lclu);
  1132. while (pr && pr->lclu <= last_lclu) {
  1133. rc->ndelonly--;
  1134. node = rb_next(&pr->rb_node);
  1135. rb_erase(&pr->rb_node, &tree->root);
  1136. kmem_cache_free(ext4_pending_cachep, pr);
  1137. if (!node)
  1138. break;
  1139. pr = rb_entry(node, struct pending_reservation,
  1140. rb_node);
  1141. }
  1142. }
  1143. }
  1144. return rc->ndelonly;
  1145. }
  1146. /*
  1147. * __es_remove_extent - removes block range from extent status tree
  1148. *
  1149. * @inode - file containing range
  1150. * @lblk - first block in range
  1151. * @end - last block in range
  1152. * @reserved - number of cluster reservations released
  1153. *
  1154. * If @reserved is not NULL and delayed allocation is enabled, counts
  1155. * block/cluster reservations freed by removing range and if bigalloc
  1156. * enabled cancels pending reservations as needed. Returns 0 on success,
  1157. * error code on failure.
  1158. */
  1159. static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
  1160. ext4_lblk_t end, int *reserved)
  1161. {
  1162. struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
  1163. struct rb_node *node;
  1164. struct extent_status *es;
  1165. struct extent_status orig_es;
  1166. ext4_lblk_t len1, len2;
  1167. ext4_fsblk_t block;
  1168. int err;
  1169. bool count_reserved = true;
  1170. struct rsvd_count rc;
  1171. if (reserved == NULL || !test_opt(inode->i_sb, DELALLOC))
  1172. count_reserved = false;
  1173. retry:
  1174. err = 0;
  1175. es = __es_tree_search(&tree->root, lblk);
  1176. if (!es)
  1177. goto out;
  1178. if (es->es_lblk > end)
  1179. goto out;
  1180. /* Simply invalidate cache_es. */
  1181. tree->cache_es = NULL;
  1182. if (count_reserved)
  1183. init_rsvd(inode, lblk, es, &rc);
  1184. orig_es.es_lblk = es->es_lblk;
  1185. orig_es.es_len = es->es_len;
  1186. orig_es.es_pblk = es->es_pblk;
  1187. len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0;
  1188. len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0;
  1189. if (len1 > 0)
  1190. es->es_len = len1;
  1191. if (len2 > 0) {
  1192. if (len1 > 0) {
  1193. struct extent_status newes;
  1194. newes.es_lblk = end + 1;
  1195. newes.es_len = len2;
  1196. block = 0x7FDEADBEEFULL;
  1197. if (ext4_es_is_written(&orig_es) ||
  1198. ext4_es_is_unwritten(&orig_es))
  1199. block = ext4_es_pblock(&orig_es) +
  1200. orig_es.es_len - len2;
  1201. ext4_es_store_pblock_status(&newes, block,
  1202. ext4_es_status(&orig_es));
  1203. err = __es_insert_extent(inode, &newes);
  1204. if (err) {
  1205. es->es_lblk = orig_es.es_lblk;
  1206. es->es_len = orig_es.es_len;
  1207. if ((err == -ENOMEM) &&
  1208. __es_shrink(EXT4_SB(inode->i_sb),
  1209. 128, EXT4_I(inode)))
  1210. goto retry;
  1211. goto out;
  1212. }
  1213. } else {
  1214. es->es_lblk = end + 1;
  1215. es->es_len = len2;
  1216. if (ext4_es_is_written(es) ||
  1217. ext4_es_is_unwritten(es)) {
  1218. block = orig_es.es_pblk + orig_es.es_len - len2;
  1219. ext4_es_store_pblock(es, block);
  1220. }
  1221. }
  1222. if (count_reserved)
  1223. count_rsvd(inode, lblk, orig_es.es_len - len1 - len2,
  1224. &orig_es, &rc);
  1225. goto out;
  1226. }
  1227. if (len1 > 0) {
  1228. if (count_reserved)
  1229. count_rsvd(inode, lblk, orig_es.es_len - len1,
  1230. &orig_es, &rc);
  1231. node = rb_next(&es->rb_node);
  1232. if (node)
  1233. es = rb_entry(node, struct extent_status, rb_node);
  1234. else
  1235. es = NULL;
  1236. }
  1237. while (es && ext4_es_end(es) <= end) {
  1238. if (count_reserved)
  1239. count_rsvd(inode, es->es_lblk, es->es_len, es, &rc);
  1240. node = rb_next(&es->rb_node);
  1241. rb_erase(&es->rb_node, &tree->root);
  1242. ext4_es_free_extent(inode, es);
  1243. if (!node) {
  1244. es = NULL;
  1245. break;
  1246. }
  1247. es = rb_entry(node, struct extent_status, rb_node);
  1248. }
  1249. if (es && es->es_lblk < end + 1) {
  1250. ext4_lblk_t orig_len = es->es_len;
  1251. len1 = ext4_es_end(es) - end;
  1252. if (count_reserved)
  1253. count_rsvd(inode, es->es_lblk, orig_len - len1,
  1254. es, &rc);
  1255. es->es_lblk = end + 1;
  1256. es->es_len = len1;
  1257. if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) {
  1258. block = es->es_pblk + orig_len - len1;
  1259. ext4_es_store_pblock(es, block);
  1260. }
  1261. }
  1262. if (count_reserved)
  1263. *reserved = get_rsvd(inode, end, es, &rc);
  1264. out:
  1265. return err;
  1266. }
  1267. /*
  1268. * ext4_es_remove_extent - removes block range from extent status tree
  1269. *
  1270. * @inode - file containing range
  1271. * @lblk - first block in range
  1272. * @len - number of blocks to remove
  1273. *
  1274. * Reduces block/cluster reservation count and for bigalloc cancels pending
  1275. * reservations as needed. Returns 0 on success, error code on failure.
  1276. */
  1277. int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
  1278. ext4_lblk_t len)
  1279. {
  1280. ext4_lblk_t end;
  1281. int err = 0;
  1282. int reserved = 0;
  1283. if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
  1284. return 0;
  1285. trace_ext4_es_remove_extent(inode, lblk, len);
  1286. es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
  1287. lblk, len, inode->i_ino);
  1288. if (!len)
  1289. return err;
  1290. end = lblk + len - 1;
  1291. BUG_ON(end < lblk);
  1292. /*
  1293. * ext4_clear_inode() depends on us taking i_es_lock unconditionally
  1294. * so that we are sure __es_shrink() is done with the inode before it
  1295. * is reclaimed.
  1296. */
  1297. write_lock(&EXT4_I(inode)->i_es_lock);
  1298. err = __es_remove_extent(inode, lblk, end, &reserved);
  1299. write_unlock(&EXT4_I(inode)->i_es_lock);
  1300. ext4_es_print_tree(inode);
  1301. ext4_da_release_space(inode, reserved);
  1302. return err;
  1303. }
  1304. static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
  1305. struct ext4_inode_info *locked_ei)
  1306. {
  1307. struct ext4_inode_info *ei;
  1308. struct ext4_es_stats *es_stats;
  1309. ktime_t start_time;
  1310. u64 scan_time;
  1311. int nr_to_walk;
  1312. int nr_shrunk = 0;
  1313. int retried = 0, nr_skipped = 0;
  1314. es_stats = &sbi->s_es_stats;
  1315. start_time = ktime_get();
  1316. retry:
  1317. spin_lock(&sbi->s_es_lock);
  1318. nr_to_walk = sbi->s_es_nr_inode;
  1319. while (nr_to_walk-- > 0) {
  1320. if (list_empty(&sbi->s_es_list)) {
  1321. spin_unlock(&sbi->s_es_lock);
  1322. goto out;
  1323. }
  1324. ei = list_first_entry(&sbi->s_es_list, struct ext4_inode_info,
  1325. i_es_list);
  1326. /* Move the inode to the tail */
  1327. list_move_tail(&ei->i_es_list, &sbi->s_es_list);
  1328. /*
  1329. * Normally we try hard to avoid shrinking precached inodes,
  1330. * but we will as a last resort.
  1331. */
  1332. if (!retried && ext4_test_inode_state(&ei->vfs_inode,
  1333. EXT4_STATE_EXT_PRECACHED)) {
  1334. nr_skipped++;
  1335. continue;
  1336. }
  1337. if (ei == locked_ei || !write_trylock(&ei->i_es_lock)) {
  1338. nr_skipped++;
  1339. continue;
  1340. }
  1341. /*
  1342. * Now we hold i_es_lock which protects us from inode reclaim
  1343. * freeing inode under us
  1344. */
  1345. spin_unlock(&sbi->s_es_lock);
  1346. nr_shrunk += es_reclaim_extents(ei, &nr_to_scan);
  1347. write_unlock(&ei->i_es_lock);
  1348. if (nr_to_scan <= 0)
  1349. goto out;
  1350. spin_lock(&sbi->s_es_lock);
  1351. }
  1352. spin_unlock(&sbi->s_es_lock);
  1353. /*
  1354. * If we skipped any inodes, and we weren't able to make any
  1355. * forward progress, try again to scan precached inodes.
  1356. */
  1357. if ((nr_shrunk == 0) && nr_skipped && !retried) {
  1358. retried++;
  1359. goto retry;
  1360. }
  1361. if (locked_ei && nr_shrunk == 0)
  1362. nr_shrunk = es_reclaim_extents(locked_ei, &nr_to_scan);
  1363. out:
  1364. scan_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
  1365. if (likely(es_stats->es_stats_scan_time))
  1366. es_stats->es_stats_scan_time = (scan_time +
  1367. es_stats->es_stats_scan_time*3) / 4;
  1368. else
  1369. es_stats->es_stats_scan_time = scan_time;
  1370. if (scan_time > es_stats->es_stats_max_scan_time)
  1371. es_stats->es_stats_max_scan_time = scan_time;
  1372. if (likely(es_stats->es_stats_shrunk))
  1373. es_stats->es_stats_shrunk = (nr_shrunk +
  1374. es_stats->es_stats_shrunk*3) / 4;
  1375. else
  1376. es_stats->es_stats_shrunk = nr_shrunk;
  1377. trace_ext4_es_shrink(sbi->s_sb, nr_shrunk, scan_time,
  1378. nr_skipped, retried);
  1379. return nr_shrunk;
  1380. }
  1381. static unsigned long ext4_es_count(struct shrinker *shrink,
  1382. struct shrink_control *sc)
  1383. {
  1384. unsigned long nr;
  1385. struct ext4_sb_info *sbi;
  1386. sbi = container_of(shrink, struct ext4_sb_info, s_es_shrinker);
  1387. nr = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
  1388. trace_ext4_es_shrink_count(sbi->s_sb, sc->nr_to_scan, nr);
  1389. return nr;
  1390. }
  1391. static unsigned long ext4_es_scan(struct shrinker *shrink,
  1392. struct shrink_control *sc)
  1393. {
  1394. struct ext4_sb_info *sbi = container_of(shrink,
  1395. struct ext4_sb_info, s_es_shrinker);
  1396. int nr_to_scan = sc->nr_to_scan;
  1397. int ret, nr_shrunk;
  1398. ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
  1399. trace_ext4_es_shrink_scan_enter(sbi->s_sb, nr_to_scan, ret);
  1400. nr_shrunk = __es_shrink(sbi, nr_to_scan, NULL);
  1401. ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
  1402. trace_ext4_es_shrink_scan_exit(sbi->s_sb, nr_shrunk, ret);
  1403. return nr_shrunk;
  1404. }
  1405. int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v)
  1406. {
  1407. struct ext4_sb_info *sbi = EXT4_SB((struct super_block *) seq->private);
  1408. struct ext4_es_stats *es_stats = &sbi->s_es_stats;
  1409. struct ext4_inode_info *ei, *max = NULL;
  1410. unsigned int inode_cnt = 0;
  1411. if (v != SEQ_START_TOKEN)
  1412. return 0;
  1413. /* here we just find an inode that has the max nr. of objects */
  1414. spin_lock(&sbi->s_es_lock);
  1415. list_for_each_entry(ei, &sbi->s_es_list, i_es_list) {
  1416. inode_cnt++;
  1417. if (max && max->i_es_all_nr < ei->i_es_all_nr)
  1418. max = ei;
  1419. else if (!max)
  1420. max = ei;
  1421. }
  1422. spin_unlock(&sbi->s_es_lock);
  1423. seq_printf(seq, "stats:\n %lld objects\n %lld reclaimable objects\n",
  1424. percpu_counter_sum_positive(&es_stats->es_stats_all_cnt),
  1425. percpu_counter_sum_positive(&es_stats->es_stats_shk_cnt));
  1426. seq_printf(seq, " %lld/%lld cache hits/misses\n",
  1427. percpu_counter_sum_positive(&es_stats->es_stats_cache_hits),
  1428. percpu_counter_sum_positive(&es_stats->es_stats_cache_misses));
  1429. if (inode_cnt)
  1430. seq_printf(seq, " %d inodes on list\n", inode_cnt);
  1431. seq_printf(seq, "average:\n %llu us scan time\n",
  1432. div_u64(es_stats->es_stats_scan_time, 1000));
  1433. seq_printf(seq, " %lu shrunk objects\n", es_stats->es_stats_shrunk);
  1434. if (inode_cnt)
  1435. seq_printf(seq,
  1436. "maximum:\n %lu inode (%u objects, %u reclaimable)\n"
  1437. " %llu us max scan time\n",
  1438. max->vfs_inode.i_ino, max->i_es_all_nr, max->i_es_shk_nr,
  1439. div_u64(es_stats->es_stats_max_scan_time, 1000));
  1440. return 0;
  1441. }
  1442. int ext4_es_register_shrinker(struct ext4_sb_info *sbi)
  1443. {
  1444. int err;
  1445. /* Make sure we have enough bits for physical block number */
  1446. BUILD_BUG_ON(ES_SHIFT < 48);
  1447. INIT_LIST_HEAD(&sbi->s_es_list);
  1448. sbi->s_es_nr_inode = 0;
  1449. spin_lock_init(&sbi->s_es_lock);
  1450. sbi->s_es_stats.es_stats_shrunk = 0;
  1451. err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_hits, 0,
  1452. GFP_KERNEL);
  1453. if (err)
  1454. return err;
  1455. err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_misses, 0,
  1456. GFP_KERNEL);
  1457. if (err)
  1458. goto err1;
  1459. sbi->s_es_stats.es_stats_scan_time = 0;
  1460. sbi->s_es_stats.es_stats_max_scan_time = 0;
  1461. err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL);
  1462. if (err)
  1463. goto err2;
  1464. err = percpu_counter_init(&sbi->s_es_stats.es_stats_shk_cnt, 0, GFP_KERNEL);
  1465. if (err)
  1466. goto err3;
  1467. sbi->s_es_shrinker.scan_objects = ext4_es_scan;
  1468. sbi->s_es_shrinker.count_objects = ext4_es_count;
  1469. sbi->s_es_shrinker.seeks = DEFAULT_SEEKS;
  1470. err = register_shrinker(&sbi->s_es_shrinker);
  1471. if (err)
  1472. goto err4;
  1473. return 0;
  1474. err4:
  1475. percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
  1476. err3:
  1477. percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
  1478. err2:
  1479. percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
  1480. err1:
  1481. percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
  1482. return err;
  1483. }
  1484. void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi)
  1485. {
  1486. percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
  1487. percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
  1488. percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
  1489. percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
  1490. unregister_shrinker(&sbi->s_es_shrinker);
  1491. }
  1492. /*
  1493. * Shrink extents in given inode from ei->i_es_shrink_lblk till end. Scan at
  1494. * most *nr_to_scan extents, update *nr_to_scan accordingly.
  1495. *
  1496. * Return 0 if we hit end of tree / interval, 1 if we exhausted nr_to_scan.
  1497. * Increment *nr_shrunk by the number of reclaimed extents. Also update
  1498. * ei->i_es_shrink_lblk to where we should continue scanning.
  1499. */
  1500. static int es_do_reclaim_extents(struct ext4_inode_info *ei, ext4_lblk_t end,
  1501. int *nr_to_scan, int *nr_shrunk)
  1502. {
  1503. struct inode *inode = &ei->vfs_inode;
  1504. struct ext4_es_tree *tree = &ei->i_es_tree;
  1505. struct extent_status *es;
  1506. struct rb_node *node;
  1507. es = __es_tree_search(&tree->root, ei->i_es_shrink_lblk);
  1508. if (!es)
  1509. goto out_wrap;
  1510. while (*nr_to_scan > 0) {
  1511. if (es->es_lblk > end) {
  1512. ei->i_es_shrink_lblk = end + 1;
  1513. return 0;
  1514. }
  1515. (*nr_to_scan)--;
  1516. node = rb_next(&es->rb_node);
  1517. /*
  1518. * We can't reclaim delayed extent from status tree because
  1519. * fiemap, bigallic, and seek_data/hole need to use it.
  1520. */
  1521. if (ext4_es_is_delayed(es))
  1522. goto next;
  1523. if (ext4_es_is_referenced(es)) {
  1524. ext4_es_clear_referenced(es);
  1525. goto next;
  1526. }
  1527. rb_erase(&es->rb_node, &tree->root);
  1528. ext4_es_free_extent(inode, es);
  1529. (*nr_shrunk)++;
  1530. next:
  1531. if (!node)
  1532. goto out_wrap;
  1533. es = rb_entry(node, struct extent_status, rb_node);
  1534. }
  1535. ei->i_es_shrink_lblk = es->es_lblk;
  1536. return 1;
  1537. out_wrap:
  1538. ei->i_es_shrink_lblk = 0;
  1539. return 0;
  1540. }
  1541. static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan)
  1542. {
  1543. struct inode *inode = &ei->vfs_inode;
  1544. int nr_shrunk = 0;
  1545. ext4_lblk_t start = ei->i_es_shrink_lblk;
  1546. static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
  1547. DEFAULT_RATELIMIT_BURST);
  1548. if (ei->i_es_shk_nr == 0)
  1549. return 0;
  1550. if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) &&
  1551. __ratelimit(&_rs))
  1552. ext4_warning(inode->i_sb, "forced shrink of precached extents");
  1553. if (!es_do_reclaim_extents(ei, EXT_MAX_BLOCKS, nr_to_scan, &nr_shrunk) &&
  1554. start != 0)
  1555. es_do_reclaim_extents(ei, start - 1, nr_to_scan, &nr_shrunk);
  1556. ei->i_es_tree.cache_es = NULL;
  1557. return nr_shrunk;
  1558. }
  1559. /*
  1560. * Called to support EXT4_IOC_CLEAR_ES_CACHE. We can only remove
  1561. * discretionary entries from the extent status cache. (Some entries
  1562. * must be present for proper operations.)
  1563. */
  1564. void ext4_clear_inode_es(struct inode *inode)
  1565. {
  1566. struct ext4_inode_info *ei = EXT4_I(inode);
  1567. struct extent_status *es;
  1568. struct ext4_es_tree *tree;
  1569. struct rb_node *node;
  1570. write_lock(&ei->i_es_lock);
  1571. tree = &EXT4_I(inode)->i_es_tree;
  1572. tree->cache_es = NULL;
  1573. node = rb_first(&tree->root);
  1574. while (node) {
  1575. es = rb_entry(node, struct extent_status, rb_node);
  1576. node = rb_next(node);
  1577. if (!ext4_es_is_delayed(es)) {
  1578. rb_erase(&es->rb_node, &tree->root);
  1579. ext4_es_free_extent(inode, es);
  1580. }
  1581. }
  1582. ext4_clear_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
  1583. write_unlock(&ei->i_es_lock);
  1584. }
  1585. #ifdef ES_DEBUG__
  1586. static void ext4_print_pending_tree(struct inode *inode)
  1587. {
  1588. struct ext4_pending_tree *tree;
  1589. struct rb_node *node;
  1590. struct pending_reservation *pr;
  1591. printk(KERN_DEBUG "pending reservations for inode %lu:", inode->i_ino);
  1592. tree = &EXT4_I(inode)->i_pending_tree;
  1593. node = rb_first(&tree->root);
  1594. while (node) {
  1595. pr = rb_entry(node, struct pending_reservation, rb_node);
  1596. printk(KERN_DEBUG " %u", pr->lclu);
  1597. node = rb_next(node);
  1598. }
  1599. printk(KERN_DEBUG "\n");
  1600. }
  1601. #else
  1602. #define ext4_print_pending_tree(inode)
  1603. #endif
  1604. int __init ext4_init_pending(void)
  1605. {
  1606. ext4_pending_cachep = kmem_cache_create("ext4_pending_reservation",
  1607. sizeof(struct pending_reservation),
  1608. 0, (SLAB_RECLAIM_ACCOUNT), NULL);
  1609. if (ext4_pending_cachep == NULL)
  1610. return -ENOMEM;
  1611. return 0;
  1612. }
  1613. void ext4_exit_pending(void)
  1614. {
  1615. kmem_cache_destroy(ext4_pending_cachep);
  1616. }
  1617. void ext4_init_pending_tree(struct ext4_pending_tree *tree)
  1618. {
  1619. tree->root = RB_ROOT;
  1620. }
  1621. /*
  1622. * __get_pending - retrieve a pointer to a pending reservation
  1623. *
  1624. * @inode - file containing the pending cluster reservation
  1625. * @lclu - logical cluster of interest
  1626. *
  1627. * Returns a pointer to a pending reservation if it's a member of
  1628. * the set, and NULL if not. Must be called holding i_es_lock.
  1629. */
  1630. static struct pending_reservation *__get_pending(struct inode *inode,
  1631. ext4_lblk_t lclu)
  1632. {
  1633. struct ext4_pending_tree *tree;
  1634. struct rb_node *node;
  1635. struct pending_reservation *pr = NULL;
  1636. tree = &EXT4_I(inode)->i_pending_tree;
  1637. node = (&tree->root)->rb_node;
  1638. while (node) {
  1639. pr = rb_entry(node, struct pending_reservation, rb_node);
  1640. if (lclu < pr->lclu)
  1641. node = node->rb_left;
  1642. else if (lclu > pr->lclu)
  1643. node = node->rb_right;
  1644. else if (lclu == pr->lclu)
  1645. return pr;
  1646. }
  1647. return NULL;
  1648. }
  1649. /*
  1650. * __insert_pending - adds a pending cluster reservation to the set of
  1651. * pending reservations
  1652. *
  1653. * @inode - file containing the cluster
  1654. * @lblk - logical block in the cluster to be added
  1655. *
  1656. * Returns 0 on successful insertion and -ENOMEM on failure. If the
  1657. * pending reservation is already in the set, returns successfully.
  1658. */
  1659. static int __insert_pending(struct inode *inode, ext4_lblk_t lblk)
  1660. {
  1661. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  1662. struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
  1663. struct rb_node **p = &tree->root.rb_node;
  1664. struct rb_node *parent = NULL;
  1665. struct pending_reservation *pr;
  1666. ext4_lblk_t lclu;
  1667. int ret = 0;
  1668. lclu = EXT4_B2C(sbi, lblk);
  1669. /* search to find parent for insertion */
  1670. while (*p) {
  1671. parent = *p;
  1672. pr = rb_entry(parent, struct pending_reservation, rb_node);
  1673. if (lclu < pr->lclu) {
  1674. p = &(*p)->rb_left;
  1675. } else if (lclu > pr->lclu) {
  1676. p = &(*p)->rb_right;
  1677. } else {
  1678. /* pending reservation already inserted */
  1679. goto out;
  1680. }
  1681. }
  1682. pr = kmem_cache_alloc(ext4_pending_cachep, GFP_ATOMIC);
  1683. if (pr == NULL) {
  1684. ret = -ENOMEM;
  1685. goto out;
  1686. }
  1687. pr->lclu = lclu;
  1688. rb_link_node(&pr->rb_node, parent, p);
  1689. rb_insert_color(&pr->rb_node, &tree->root);
  1690. out:
  1691. return ret;
  1692. }
  1693. /*
  1694. * __remove_pending - removes a pending cluster reservation from the set
  1695. * of pending reservations
  1696. *
  1697. * @inode - file containing the cluster
  1698. * @lblk - logical block in the pending cluster reservation to be removed
  1699. *
  1700. * Returns successfully if pending reservation is not a member of the set.
  1701. */
  1702. static void __remove_pending(struct inode *inode, ext4_lblk_t lblk)
  1703. {
  1704. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  1705. struct pending_reservation *pr;
  1706. struct ext4_pending_tree *tree;
  1707. pr = __get_pending(inode, EXT4_B2C(sbi, lblk));
  1708. if (pr != NULL) {
  1709. tree = &EXT4_I(inode)->i_pending_tree;
  1710. rb_erase(&pr->rb_node, &tree->root);
  1711. kmem_cache_free(ext4_pending_cachep, pr);
  1712. }
  1713. }
  1714. /*
  1715. * ext4_remove_pending - removes a pending cluster reservation from the set
  1716. * of pending reservations
  1717. *
  1718. * @inode - file containing the cluster
  1719. * @lblk - logical block in the pending cluster reservation to be removed
  1720. *
  1721. * Locking for external use of __remove_pending.
  1722. */
  1723. void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk)
  1724. {
  1725. struct ext4_inode_info *ei = EXT4_I(inode);
  1726. write_lock(&ei->i_es_lock);
  1727. __remove_pending(inode, lblk);
  1728. write_unlock(&ei->i_es_lock);
  1729. }
  1730. /*
  1731. * ext4_is_pending - determine whether a cluster has a pending reservation
  1732. * on it
  1733. *
  1734. * @inode - file containing the cluster
  1735. * @lblk - logical block in the cluster
  1736. *
  1737. * Returns true if there's a pending reservation for the cluster in the
  1738. * set of pending reservations, and false if not.
  1739. */
  1740. bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk)
  1741. {
  1742. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  1743. struct ext4_inode_info *ei = EXT4_I(inode);
  1744. bool ret;
  1745. read_lock(&ei->i_es_lock);
  1746. ret = (bool)(__get_pending(inode, EXT4_B2C(sbi, lblk)) != NULL);
  1747. read_unlock(&ei->i_es_lock);
  1748. return ret;
  1749. }
  1750. /*
  1751. * ext4_es_insert_delayed_block - adds a delayed block to the extents status
  1752. * tree, adding a pending reservation where
  1753. * needed
  1754. *
  1755. * @inode - file containing the newly added block
  1756. * @lblk - logical block to be added
  1757. * @allocated - indicates whether a physical cluster has been allocated for
  1758. * the logical cluster that contains the block
  1759. *
  1760. * Returns 0 on success, negative error code on failure.
  1761. */
  1762. int ext4_es_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk,
  1763. bool allocated)
  1764. {
  1765. struct extent_status newes;
  1766. int err = 0;
  1767. if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
  1768. return 0;
  1769. es_debug("add [%u/1) delayed to extent status tree of inode %lu\n",
  1770. lblk, inode->i_ino);
  1771. newes.es_lblk = lblk;
  1772. newes.es_len = 1;
  1773. ext4_es_store_pblock_status(&newes, ~0, EXTENT_STATUS_DELAYED);
  1774. trace_ext4_es_insert_delayed_block(inode, &newes, allocated);
  1775. ext4_es_insert_extent_check(inode, &newes);
  1776. write_lock(&EXT4_I(inode)->i_es_lock);
  1777. err = __es_remove_extent(inode, lblk, lblk, NULL);
  1778. if (err != 0)
  1779. goto error;
  1780. retry:
  1781. err = __es_insert_extent(inode, &newes);
  1782. if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb),
  1783. 128, EXT4_I(inode)))
  1784. goto retry;
  1785. if (err != 0)
  1786. goto error;
  1787. if (allocated)
  1788. __insert_pending(inode, lblk);
  1789. error:
  1790. write_unlock(&EXT4_I(inode)->i_es_lock);
  1791. ext4_es_print_tree(inode);
  1792. ext4_print_pending_tree(inode);
  1793. return err;
  1794. }
  1795. /*
  1796. * __es_delayed_clu - count number of clusters containing blocks that
  1797. * are delayed only
  1798. *
  1799. * @inode - file containing block range
  1800. * @start - logical block defining start of range
  1801. * @end - logical block defining end of range
  1802. *
  1803. * Returns the number of clusters containing only delayed (not delayed
  1804. * and unwritten) blocks in the range specified by @start and @end. Any
  1805. * cluster or part of a cluster within the range and containing a delayed
  1806. * and not unwritten block within the range is counted as a whole cluster.
  1807. */
  1808. static unsigned int __es_delayed_clu(struct inode *inode, ext4_lblk_t start,
  1809. ext4_lblk_t end)
  1810. {
  1811. struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
  1812. struct extent_status *es;
  1813. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  1814. struct rb_node *node;
  1815. ext4_lblk_t first_lclu, last_lclu;
  1816. unsigned long long last_counted_lclu;
  1817. unsigned int n = 0;
  1818. /* guaranteed to be unequal to any ext4_lblk_t value */
  1819. last_counted_lclu = ~0ULL;
  1820. es = __es_tree_search(&tree->root, start);
  1821. while (es && (es->es_lblk <= end)) {
  1822. if (ext4_es_is_delonly(es)) {
  1823. if (es->es_lblk <= start)
  1824. first_lclu = EXT4_B2C(sbi, start);
  1825. else
  1826. first_lclu = EXT4_B2C(sbi, es->es_lblk);
  1827. if (ext4_es_end(es) >= end)
  1828. last_lclu = EXT4_B2C(sbi, end);
  1829. else
  1830. last_lclu = EXT4_B2C(sbi, ext4_es_end(es));
  1831. if (first_lclu == last_counted_lclu)
  1832. n += last_lclu - first_lclu;
  1833. else
  1834. n += last_lclu - first_lclu + 1;
  1835. last_counted_lclu = last_lclu;
  1836. }
  1837. node = rb_next(&es->rb_node);
  1838. if (!node)
  1839. break;
  1840. es = rb_entry(node, struct extent_status, rb_node);
  1841. }
  1842. return n;
  1843. }
  1844. /*
  1845. * ext4_es_delayed_clu - count number of clusters containing blocks that
  1846. * are both delayed and unwritten
  1847. *
  1848. * @inode - file containing block range
  1849. * @lblk - logical block defining start of range
  1850. * @len - number of blocks in range
  1851. *
  1852. * Locking for external use of __es_delayed_clu().
  1853. */
  1854. unsigned int ext4_es_delayed_clu(struct inode *inode, ext4_lblk_t lblk,
  1855. ext4_lblk_t len)
  1856. {
  1857. struct ext4_inode_info *ei = EXT4_I(inode);
  1858. ext4_lblk_t end;
  1859. unsigned int n;
  1860. if (len == 0)
  1861. return 0;
  1862. end = lblk + len - 1;
  1863. WARN_ON(end < lblk);
  1864. read_lock(&ei->i_es_lock);
  1865. n = __es_delayed_clu(inode, lblk, end);
  1866. read_unlock(&ei->i_es_lock);
  1867. return n;
  1868. }
  1869. /*
  1870. * __revise_pending - makes, cancels, or leaves unchanged pending cluster
  1871. * reservations for a specified block range depending
  1872. * upon the presence or absence of delayed blocks
  1873. * outside the range within clusters at the ends of the
  1874. * range
  1875. *
  1876. * @inode - file containing the range
  1877. * @lblk - logical block defining the start of range
  1878. * @len - length of range in blocks
  1879. *
  1880. * Used after a newly allocated extent is added to the extents status tree.
  1881. * Requires that the extents in the range have either written or unwritten
  1882. * status. Must be called while holding i_es_lock.
  1883. */
  1884. static void __revise_pending(struct inode *inode, ext4_lblk_t lblk,
  1885. ext4_lblk_t len)
  1886. {
  1887. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  1888. ext4_lblk_t end = lblk + len - 1;
  1889. ext4_lblk_t first, last;
  1890. bool f_del = false, l_del = false;
  1891. if (len == 0)
  1892. return;
  1893. /*
  1894. * Two cases - block range within single cluster and block range
  1895. * spanning two or more clusters. Note that a cluster belonging
  1896. * to a range starting and/or ending on a cluster boundary is treated
  1897. * as if it does not contain a delayed extent. The new range may
  1898. * have allocated space for previously delayed blocks out to the
  1899. * cluster boundary, requiring that any pre-existing pending
  1900. * reservation be canceled. Because this code only looks at blocks
  1901. * outside the range, it should revise pending reservations
  1902. * correctly even if the extent represented by the range can't be
  1903. * inserted in the extents status tree due to ENOSPC.
  1904. */
  1905. if (EXT4_B2C(sbi, lblk) == EXT4_B2C(sbi, end)) {
  1906. first = EXT4_LBLK_CMASK(sbi, lblk);
  1907. if (first != lblk)
  1908. f_del = __es_scan_range(inode, &ext4_es_is_delonly,
  1909. first, lblk - 1);
  1910. if (f_del) {
  1911. __insert_pending(inode, first);
  1912. } else {
  1913. last = EXT4_LBLK_CMASK(sbi, end) +
  1914. sbi->s_cluster_ratio - 1;
  1915. if (last != end)
  1916. l_del = __es_scan_range(inode,
  1917. &ext4_es_is_delonly,
  1918. end + 1, last);
  1919. if (l_del)
  1920. __insert_pending(inode, last);
  1921. else
  1922. __remove_pending(inode, last);
  1923. }
  1924. } else {
  1925. first = EXT4_LBLK_CMASK(sbi, lblk);
  1926. if (first != lblk)
  1927. f_del = __es_scan_range(inode, &ext4_es_is_delonly,
  1928. first, lblk - 1);
  1929. if (f_del)
  1930. __insert_pending(inode, first);
  1931. else
  1932. __remove_pending(inode, first);
  1933. last = EXT4_LBLK_CMASK(sbi, end) + sbi->s_cluster_ratio - 1;
  1934. if (last != end)
  1935. l_del = __es_scan_range(inode, &ext4_es_is_delonly,
  1936. end + 1, last);
  1937. if (l_del)
  1938. __insert_pending(inode, last);
  1939. else
  1940. __remove_pending(inode, last);
  1941. }
  1942. }