quota.c 38 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641
  1. /*
  2. * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
  3. * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
  4. *
  5. * This copyrighted material is made available to anyone wishing to use,
  6. * modify, copy, or redistribute it subject to the terms and conditions
  7. * of the GNU General Public License version 2.
  8. */
  9. /*
  10. * Quota change tags are associated with each transaction that allocates or
  11. * deallocates space. Those changes are accumulated locally to each node (in a
  12. * per-node file) and then are periodically synced to the quota file. This
  13. * avoids the bottleneck of constantly touching the quota file, but introduces
  14. * fuzziness in the current usage value of IDs that are being used on different
  15. * nodes in the cluster simultaneously. So, it is possible for a user on
  16. * multiple nodes to overrun their quota, but that overrun is controlable.
  17. * Since quota tags are part of transactions, there is no need for a quota check
  18. * program to be run on node crashes or anything like that.
  19. *
  20. * There are couple of knobs that let the administrator manage the quota
  21. * fuzziness. "quota_quantum" sets the maximum time a quota change can be
  22. * sitting on one node before being synced to the quota file. (The default is
  23. * 60 seconds.) Another knob, "quota_scale" controls how quickly the frequency
  24. * of quota file syncs increases as the user moves closer to their limit. The
  25. * more frequent the syncs, the more accurate the quota enforcement, but that
  26. * means that there is more contention between the nodes for the quota file.
  27. * The default value is one. This sets the maximum theoretical quota overrun
  28. * (with infinite node with infinite bandwidth) to twice the user's limit. (In
  29. * practice, the maximum overrun you see should be much less.) A "quota_scale"
  30. * number greater than one makes quota syncs more frequent and reduces the
  31. * maximum overrun. Numbers less than one (but greater than zero) make quota
  32. * syncs less frequent.
  33. *
  34. * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
  35. * the quota file, so it is not being constantly read.
  36. */
  37. #include <linux/sched.h>
  38. #include <linux/slab.h>
  39. #include <linux/mm.h>
  40. #include <linux/spinlock.h>
  41. #include <linux/completion.h>
  42. #include <linux/buffer_head.h>
  43. #include <linux/sort.h>
  44. #include <linux/fs.h>
  45. #include <linux/bio.h>
  46. #include <linux/gfs2_ondisk.h>
  47. #include <linux/kthread.h>
  48. #include <linux/freezer.h>
  49. #include <linux/quota.h>
  50. #include <linux/dqblk_xfs.h>
  51. #include <linux/lockref.h>
  52. #include "gfs2.h"
  53. #include "incore.h"
  54. #include "bmap.h"
  55. #include "glock.h"
  56. #include "glops.h"
  57. #include "log.h"
  58. #include "meta_io.h"
  59. #include "quota.h"
  60. #include "rgrp.h"
  61. #include "super.h"
  62. #include "trans.h"
  63. #include "inode.h"
  64. #include "util.h"
  65. struct gfs2_quota_change_host {
  66. u64 qc_change;
  67. u32 qc_flags; /* GFS2_QCF_... */
  68. struct kqid qc_id;
  69. };
  70. static LIST_HEAD(qd_lru_list);
  71. static atomic_t qd_lru_count = ATOMIC_INIT(0);
  72. static DEFINE_SPINLOCK(qd_lock);
  73. unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
  74. struct shrink_control *sc)
  75. {
  76. struct gfs2_quota_data *qd;
  77. struct gfs2_sbd *sdp;
  78. int nr_to_scan = sc->nr_to_scan;
  79. long freed = 0;
  80. if (!(sc->gfp_mask & __GFP_FS))
  81. return SHRINK_STOP;
  82. spin_lock(&qd_lock);
  83. while (nr_to_scan && !list_empty(&qd_lru_list)) {
  84. qd = list_entry(qd_lru_list.next,
  85. struct gfs2_quota_data, qd_reclaim);
  86. sdp = qd->qd_gl->gl_sbd;
  87. /* Free from the filesystem-specific list */
  88. list_del(&qd->qd_list);
  89. gfs2_assert_warn(sdp, !qd->qd_change);
  90. gfs2_assert_warn(sdp, !qd->qd_slot_count);
  91. gfs2_assert_warn(sdp, !qd->qd_bh_count);
  92. gfs2_glock_put(qd->qd_gl);
  93. atomic_dec(&sdp->sd_quota_count);
  94. /* Delete it from the common reclaim list */
  95. list_del_init(&qd->qd_reclaim);
  96. atomic_dec(&qd_lru_count);
  97. spin_unlock(&qd_lock);
  98. kmem_cache_free(gfs2_quotad_cachep, qd);
  99. spin_lock(&qd_lock);
  100. nr_to_scan--;
  101. freed++;
  102. }
  103. spin_unlock(&qd_lock);
  104. return freed;
  105. }
  106. unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
  107. struct shrink_control *sc)
  108. {
  109. return vfs_pressure_ratio(atomic_read(&qd_lru_count));
  110. }
  111. static u64 qd2index(struct gfs2_quota_data *qd)
  112. {
  113. struct kqid qid = qd->qd_id;
  114. return (2 * (u64)from_kqid(&init_user_ns, qid)) +
  115. ((qid.type == USRQUOTA) ? 0 : 1);
  116. }
  117. static u64 qd2offset(struct gfs2_quota_data *qd)
  118. {
  119. u64 offset;
  120. offset = qd2index(qd);
  121. offset *= sizeof(struct gfs2_quota);
  122. return offset;
  123. }
  124. static int qd_alloc(struct gfs2_sbd *sdp, struct kqid qid,
  125. struct gfs2_quota_data **qdp)
  126. {
  127. struct gfs2_quota_data *qd;
  128. int error;
  129. qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
  130. if (!qd)
  131. return -ENOMEM;
  132. qd->qd_lockref.count = 1;
  133. spin_lock_init(&qd->qd_lockref.lock);
  134. qd->qd_id = qid;
  135. qd->qd_slot = -1;
  136. INIT_LIST_HEAD(&qd->qd_reclaim);
  137. error = gfs2_glock_get(sdp, qd2index(qd),
  138. &gfs2_quota_glops, CREATE, &qd->qd_gl);
  139. if (error)
  140. goto fail;
  141. *qdp = qd;
  142. return 0;
  143. fail:
  144. kmem_cache_free(gfs2_quotad_cachep, qd);
  145. return error;
  146. }
  147. static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
  148. struct gfs2_quota_data **qdp)
  149. {
  150. struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
  151. int error, found;
  152. *qdp = NULL;
  153. for (;;) {
  154. found = 0;
  155. spin_lock(&qd_lock);
  156. list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
  157. if (qid_eq(qd->qd_id, qid)) {
  158. lockref_get(&qd->qd_lockref);
  159. if (!list_empty(&qd->qd_reclaim)) {
  160. /* Remove it from reclaim list */
  161. list_del_init(&qd->qd_reclaim);
  162. atomic_dec(&qd_lru_count);
  163. }
  164. found = 1;
  165. break;
  166. }
  167. }
  168. if (!found)
  169. qd = NULL;
  170. if (!qd && new_qd) {
  171. qd = new_qd;
  172. list_add(&qd->qd_list, &sdp->sd_quota_list);
  173. atomic_inc(&sdp->sd_quota_count);
  174. new_qd = NULL;
  175. }
  176. spin_unlock(&qd_lock);
  177. if (qd) {
  178. if (new_qd) {
  179. gfs2_glock_put(new_qd->qd_gl);
  180. kmem_cache_free(gfs2_quotad_cachep, new_qd);
  181. }
  182. *qdp = qd;
  183. return 0;
  184. }
  185. error = qd_alloc(sdp, qid, &new_qd);
  186. if (error)
  187. return error;
  188. }
  189. }
  190. static void qd_hold(struct gfs2_quota_data *qd)
  191. {
  192. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  193. gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
  194. lockref_get(&qd->qd_lockref);
  195. }
  196. static void qd_put(struct gfs2_quota_data *qd)
  197. {
  198. spin_lock(&qd_lock);
  199. if (!lockref_put_or_lock(&qd->qd_lockref)) {
  200. /* Add to the reclaim list */
  201. list_add_tail(&qd->qd_reclaim, &qd_lru_list);
  202. atomic_inc(&qd_lru_count);
  203. spin_unlock(&qd->qd_lockref.lock);
  204. }
  205. spin_unlock(&qd_lock);
  206. }
  207. static int slot_get(struct gfs2_quota_data *qd)
  208. {
  209. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  210. unsigned int c, o = 0, b;
  211. unsigned char byte = 0;
  212. spin_lock(&qd_lock);
  213. if (qd->qd_slot_count++) {
  214. spin_unlock(&qd_lock);
  215. return 0;
  216. }
  217. for (c = 0; c < sdp->sd_quota_chunks; c++)
  218. for (o = 0; o < PAGE_SIZE; o++) {
  219. byte = sdp->sd_quota_bitmap[c][o];
  220. if (byte != 0xFF)
  221. goto found;
  222. }
  223. goto fail;
  224. found:
  225. for (b = 0; b < 8; b++)
  226. if (!(byte & (1 << b)))
  227. break;
  228. qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
  229. if (qd->qd_slot >= sdp->sd_quota_slots)
  230. goto fail;
  231. sdp->sd_quota_bitmap[c][o] |= 1 << b;
  232. spin_unlock(&qd_lock);
  233. return 0;
  234. fail:
  235. qd->qd_slot_count--;
  236. spin_unlock(&qd_lock);
  237. return -ENOSPC;
  238. }
  239. static void slot_hold(struct gfs2_quota_data *qd)
  240. {
  241. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  242. spin_lock(&qd_lock);
  243. gfs2_assert(sdp, qd->qd_slot_count);
  244. qd->qd_slot_count++;
  245. spin_unlock(&qd_lock);
  246. }
  247. static void gfs2_icbit_munge(struct gfs2_sbd *sdp, unsigned char **bitmap,
  248. unsigned int bit, int new_value)
  249. {
  250. unsigned int c, o, b = bit;
  251. int old_value;
  252. c = b / (8 * PAGE_SIZE);
  253. b %= 8 * PAGE_SIZE;
  254. o = b / 8;
  255. b %= 8;
  256. old_value = (bitmap[c][o] & (1 << b));
  257. gfs2_assert_withdraw(sdp, !old_value != !new_value);
  258. if (new_value)
  259. bitmap[c][o] |= 1 << b;
  260. else
  261. bitmap[c][o] &= ~(1 << b);
  262. }
  263. static void slot_put(struct gfs2_quota_data *qd)
  264. {
  265. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  266. spin_lock(&qd_lock);
  267. gfs2_assert(sdp, qd->qd_slot_count);
  268. if (!--qd->qd_slot_count) {
  269. gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
  270. qd->qd_slot = -1;
  271. }
  272. spin_unlock(&qd_lock);
  273. }
  274. static int bh_get(struct gfs2_quota_data *qd)
  275. {
  276. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  277. struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
  278. unsigned int block, offset;
  279. struct buffer_head *bh;
  280. int error;
  281. struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
  282. mutex_lock(&sdp->sd_quota_mutex);
  283. if (qd->qd_bh_count++) {
  284. mutex_unlock(&sdp->sd_quota_mutex);
  285. return 0;
  286. }
  287. block = qd->qd_slot / sdp->sd_qc_per_block;
  288. offset = qd->qd_slot % sdp->sd_qc_per_block;
  289. bh_map.b_size = 1 << ip->i_inode.i_blkbits;
  290. error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
  291. if (error)
  292. goto fail;
  293. error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
  294. if (error)
  295. goto fail;
  296. error = -EIO;
  297. if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
  298. goto fail_brelse;
  299. qd->qd_bh = bh;
  300. qd->qd_bh_qc = (struct gfs2_quota_change *)
  301. (bh->b_data + sizeof(struct gfs2_meta_header) +
  302. offset * sizeof(struct gfs2_quota_change));
  303. mutex_unlock(&sdp->sd_quota_mutex);
  304. return 0;
  305. fail_brelse:
  306. brelse(bh);
  307. fail:
  308. qd->qd_bh_count--;
  309. mutex_unlock(&sdp->sd_quota_mutex);
  310. return error;
  311. }
  312. static void bh_put(struct gfs2_quota_data *qd)
  313. {
  314. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  315. mutex_lock(&sdp->sd_quota_mutex);
  316. gfs2_assert(sdp, qd->qd_bh_count);
  317. if (!--qd->qd_bh_count) {
  318. brelse(qd->qd_bh);
  319. qd->qd_bh = NULL;
  320. qd->qd_bh_qc = NULL;
  321. }
  322. mutex_unlock(&sdp->sd_quota_mutex);
  323. }
  324. static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
  325. u64 *sync_gen)
  326. {
  327. if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
  328. !test_bit(QDF_CHANGE, &qd->qd_flags) ||
  329. (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
  330. return 0;
  331. list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
  332. set_bit(QDF_LOCKED, &qd->qd_flags);
  333. gfs2_assert_warn(sdp, !__lockref_is_dead(&qd->qd_lockref));
  334. lockref_get(&qd->qd_lockref);
  335. qd->qd_change_sync = qd->qd_change;
  336. gfs2_assert_warn(sdp, qd->qd_slot_count);
  337. qd->qd_slot_count++;
  338. return 1;
  339. }
  340. static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
  341. {
  342. struct gfs2_quota_data *qd = NULL;
  343. int error;
  344. int found = 0;
  345. *qdp = NULL;
  346. if (sdp->sd_vfs->s_flags & MS_RDONLY)
  347. return 0;
  348. spin_lock(&qd_lock);
  349. list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
  350. found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
  351. if (found)
  352. break;
  353. }
  354. if (!found)
  355. qd = NULL;
  356. spin_unlock(&qd_lock);
  357. if (qd) {
  358. gfs2_assert_warn(sdp, qd->qd_change_sync);
  359. error = bh_get(qd);
  360. if (error) {
  361. clear_bit(QDF_LOCKED, &qd->qd_flags);
  362. slot_put(qd);
  363. qd_put(qd);
  364. return error;
  365. }
  366. }
  367. *qdp = qd;
  368. return 0;
  369. }
  370. static void qd_unlock(struct gfs2_quota_data *qd)
  371. {
  372. gfs2_assert_warn(qd->qd_gl->gl_sbd,
  373. test_bit(QDF_LOCKED, &qd->qd_flags));
  374. clear_bit(QDF_LOCKED, &qd->qd_flags);
  375. bh_put(qd);
  376. slot_put(qd);
  377. qd_put(qd);
  378. }
  379. static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
  380. struct gfs2_quota_data **qdp)
  381. {
  382. int error;
  383. error = qd_get(sdp, qid, qdp);
  384. if (error)
  385. return error;
  386. error = slot_get(*qdp);
  387. if (error)
  388. goto fail;
  389. error = bh_get(*qdp);
  390. if (error)
  391. goto fail_slot;
  392. return 0;
  393. fail_slot:
  394. slot_put(*qdp);
  395. fail:
  396. qd_put(*qdp);
  397. return error;
  398. }
  399. static void qdsb_put(struct gfs2_quota_data *qd)
  400. {
  401. bh_put(qd);
  402. slot_put(qd);
  403. qd_put(qd);
  404. }
  405. int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
  406. {
  407. struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
  408. struct gfs2_quota_data **qd;
  409. int error;
  410. if (ip->i_res == NULL) {
  411. error = gfs2_rs_alloc(ip);
  412. if (error)
  413. return error;
  414. }
  415. qd = ip->i_res->rs_qa_qd;
  416. if (gfs2_assert_warn(sdp, !ip->i_res->rs_qa_qd_num) ||
  417. gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
  418. return -EIO;
  419. if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
  420. return 0;
  421. error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
  422. if (error)
  423. goto out;
  424. ip->i_res->rs_qa_qd_num++;
  425. qd++;
  426. error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
  427. if (error)
  428. goto out;
  429. ip->i_res->rs_qa_qd_num++;
  430. qd++;
  431. if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
  432. !uid_eq(uid, ip->i_inode.i_uid)) {
  433. error = qdsb_get(sdp, make_kqid_uid(uid), qd);
  434. if (error)
  435. goto out;
  436. ip->i_res->rs_qa_qd_num++;
  437. qd++;
  438. }
  439. if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
  440. !gid_eq(gid, ip->i_inode.i_gid)) {
  441. error = qdsb_get(sdp, make_kqid_gid(gid), qd);
  442. if (error)
  443. goto out;
  444. ip->i_res->rs_qa_qd_num++;
  445. qd++;
  446. }
  447. out:
  448. if (error)
  449. gfs2_quota_unhold(ip);
  450. return error;
  451. }
  452. void gfs2_quota_unhold(struct gfs2_inode *ip)
  453. {
  454. struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
  455. unsigned int x;
  456. if (ip->i_res == NULL)
  457. return;
  458. gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
  459. for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
  460. qdsb_put(ip->i_res->rs_qa_qd[x]);
  461. ip->i_res->rs_qa_qd[x] = NULL;
  462. }
  463. ip->i_res->rs_qa_qd_num = 0;
  464. }
  465. static int sort_qd(const void *a, const void *b)
  466. {
  467. const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
  468. const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
  469. if (qid_lt(qd_a->qd_id, qd_b->qd_id))
  470. return -1;
  471. if (qid_lt(qd_b->qd_id, qd_a->qd_id))
  472. return 1;
  473. return 0;
  474. }
  475. static void do_qc(struct gfs2_quota_data *qd, s64 change)
  476. {
  477. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  478. struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
  479. struct gfs2_quota_change *qc = qd->qd_bh_qc;
  480. s64 x;
  481. mutex_lock(&sdp->sd_quota_mutex);
  482. gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
  483. if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
  484. qc->qc_change = 0;
  485. qc->qc_flags = 0;
  486. if (qd->qd_id.type == USRQUOTA)
  487. qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
  488. qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
  489. }
  490. x = be64_to_cpu(qc->qc_change) + change;
  491. qc->qc_change = cpu_to_be64(x);
  492. spin_lock(&qd_lock);
  493. qd->qd_change = x;
  494. spin_unlock(&qd_lock);
  495. if (!x) {
  496. gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
  497. clear_bit(QDF_CHANGE, &qd->qd_flags);
  498. qc->qc_flags = 0;
  499. qc->qc_id = 0;
  500. slot_put(qd);
  501. qd_put(qd);
  502. } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
  503. qd_hold(qd);
  504. slot_hold(qd);
  505. }
  506. mutex_unlock(&sdp->sd_quota_mutex);
  507. }
  508. /**
  509. * gfs2_adjust_quota - adjust record of current block usage
  510. * @ip: The quota inode
  511. * @loc: Offset of the entry in the quota file
  512. * @change: The amount of usage change to record
  513. * @qd: The quota data
  514. * @fdq: The updated limits to record
  515. *
  516. * This function was mostly borrowed from gfs2_block_truncate_page which was
  517. * in turn mostly borrowed from ext3
  518. *
  519. * Returns: 0 or -ve on error
  520. */
  521. static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
  522. s64 change, struct gfs2_quota_data *qd,
  523. struct fs_disk_quota *fdq)
  524. {
  525. struct inode *inode = &ip->i_inode;
  526. struct gfs2_sbd *sdp = GFS2_SB(inode);
  527. struct address_space *mapping = inode->i_mapping;
  528. unsigned long index = loc >> PAGE_CACHE_SHIFT;
  529. unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
  530. unsigned blocksize, iblock, pos;
  531. struct buffer_head *bh;
  532. struct page *page;
  533. void *kaddr, *ptr;
  534. struct gfs2_quota q, *qp;
  535. int err, nbytes;
  536. u64 size;
  537. if (gfs2_is_stuffed(ip)) {
  538. err = gfs2_unstuff_dinode(ip, NULL);
  539. if (err)
  540. return err;
  541. }
  542. memset(&q, 0, sizeof(struct gfs2_quota));
  543. err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
  544. if (err < 0)
  545. return err;
  546. err = -EIO;
  547. qp = &q;
  548. qp->qu_value = be64_to_cpu(qp->qu_value);
  549. qp->qu_value += change;
  550. qp->qu_value = cpu_to_be64(qp->qu_value);
  551. qd->qd_qb.qb_value = qp->qu_value;
  552. if (fdq) {
  553. if (fdq->d_fieldmask & FS_DQ_BSOFT) {
  554. qp->qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
  555. qd->qd_qb.qb_warn = qp->qu_warn;
  556. }
  557. if (fdq->d_fieldmask & FS_DQ_BHARD) {
  558. qp->qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
  559. qd->qd_qb.qb_limit = qp->qu_limit;
  560. }
  561. if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
  562. qp->qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
  563. qd->qd_qb.qb_value = qp->qu_value;
  564. }
  565. }
  566. /* Write the quota into the quota file on disk */
  567. ptr = qp;
  568. nbytes = sizeof(struct gfs2_quota);
  569. get_a_page:
  570. page = find_or_create_page(mapping, index, GFP_NOFS);
  571. if (!page)
  572. return -ENOMEM;
  573. blocksize = inode->i_sb->s_blocksize;
  574. iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
  575. if (!page_has_buffers(page))
  576. create_empty_buffers(page, blocksize, 0);
  577. bh = page_buffers(page);
  578. pos = blocksize;
  579. while (offset >= pos) {
  580. bh = bh->b_this_page;
  581. iblock++;
  582. pos += blocksize;
  583. }
  584. if (!buffer_mapped(bh)) {
  585. gfs2_block_map(inode, iblock, bh, 1);
  586. if (!buffer_mapped(bh))
  587. goto unlock_out;
  588. /* If it's a newly allocated disk block for quota, zero it */
  589. if (buffer_new(bh))
  590. zero_user(page, pos - blocksize, bh->b_size);
  591. }
  592. if (PageUptodate(page))
  593. set_buffer_uptodate(bh);
  594. if (!buffer_uptodate(bh)) {
  595. ll_rw_block(READ | REQ_META, 1, &bh);
  596. wait_on_buffer(bh);
  597. if (!buffer_uptodate(bh))
  598. goto unlock_out;
  599. }
  600. gfs2_trans_add_data(ip->i_gl, bh);
  601. kaddr = kmap_atomic(page);
  602. if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
  603. nbytes = PAGE_CACHE_SIZE - offset;
  604. memcpy(kaddr + offset, ptr, nbytes);
  605. flush_dcache_page(page);
  606. kunmap_atomic(kaddr);
  607. unlock_page(page);
  608. page_cache_release(page);
  609. /* If quota straddles page boundary, we need to update the rest of the
  610. * quota at the beginning of the next page */
  611. if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
  612. ptr = ptr + nbytes;
  613. nbytes = sizeof(struct gfs2_quota) - nbytes;
  614. offset = 0;
  615. index++;
  616. goto get_a_page;
  617. }
  618. size = loc + sizeof(struct gfs2_quota);
  619. if (size > inode->i_size)
  620. i_size_write(inode, size);
  621. inode->i_mtime = inode->i_atime = CURRENT_TIME;
  622. mark_inode_dirty(inode);
  623. return 0;
  624. unlock_out:
  625. unlock_page(page);
  626. page_cache_release(page);
  627. return err;
  628. }
  629. static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
  630. {
  631. struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
  632. struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
  633. struct gfs2_alloc_parms ap = { .aflags = 0, };
  634. unsigned int data_blocks, ind_blocks;
  635. struct gfs2_holder *ghs, i_gh;
  636. unsigned int qx, x;
  637. struct gfs2_quota_data *qd;
  638. unsigned reserved;
  639. loff_t offset;
  640. unsigned int nalloc = 0, blocks;
  641. int error;
  642. error = gfs2_rs_alloc(ip);
  643. if (error)
  644. return error;
  645. gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
  646. &data_blocks, &ind_blocks);
  647. ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
  648. if (!ghs)
  649. return -ENOMEM;
  650. sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
  651. mutex_lock(&ip->i_inode.i_mutex);
  652. for (qx = 0; qx < num_qd; qx++) {
  653. error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
  654. GL_NOCACHE, &ghs[qx]);
  655. if (error)
  656. goto out;
  657. }
  658. error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
  659. if (error)
  660. goto out;
  661. for (x = 0; x < num_qd; x++) {
  662. offset = qd2offset(qda[x]);
  663. if (gfs2_write_alloc_required(ip, offset,
  664. sizeof(struct gfs2_quota)))
  665. nalloc++;
  666. }
  667. /*
  668. * 1 blk for unstuffing inode if stuffed. We add this extra
  669. * block to the reservation unconditionally. If the inode
  670. * doesn't need unstuffing, the block will be released to the
  671. * rgrp since it won't be allocated during the transaction
  672. */
  673. /* +3 in the end for unstuffing block, inode size update block
  674. * and another block in case quota straddles page boundary and
  675. * two blocks need to be updated instead of 1 */
  676. blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
  677. reserved = 1 + (nalloc * (data_blocks + ind_blocks));
  678. ap.target = reserved;
  679. error = gfs2_inplace_reserve(ip, &ap);
  680. if (error)
  681. goto out_alloc;
  682. if (nalloc)
  683. blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
  684. error = gfs2_trans_begin(sdp, blocks, 0);
  685. if (error)
  686. goto out_ipres;
  687. for (x = 0; x < num_qd; x++) {
  688. qd = qda[x];
  689. offset = qd2offset(qd);
  690. error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
  691. if (error)
  692. goto out_end_trans;
  693. do_qc(qd, -qd->qd_change_sync);
  694. set_bit(QDF_REFRESH, &qd->qd_flags);
  695. }
  696. error = 0;
  697. out_end_trans:
  698. gfs2_trans_end(sdp);
  699. out_ipres:
  700. gfs2_inplace_release(ip);
  701. out_alloc:
  702. gfs2_glock_dq_uninit(&i_gh);
  703. out:
  704. while (qx--)
  705. gfs2_glock_dq_uninit(&ghs[qx]);
  706. mutex_unlock(&ip->i_inode.i_mutex);
  707. kfree(ghs);
  708. gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
  709. return error;
  710. }
  711. static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
  712. {
  713. struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
  714. struct gfs2_quota q;
  715. struct gfs2_quota_lvb *qlvb;
  716. loff_t pos;
  717. int error;
  718. memset(&q, 0, sizeof(struct gfs2_quota));
  719. pos = qd2offset(qd);
  720. error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
  721. if (error < 0)
  722. return error;
  723. qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
  724. qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
  725. qlvb->__pad = 0;
  726. qlvb->qb_limit = q.qu_limit;
  727. qlvb->qb_warn = q.qu_warn;
  728. qlvb->qb_value = q.qu_value;
  729. qd->qd_qb = *qlvb;
  730. return 0;
  731. }
  732. static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
  733. struct gfs2_holder *q_gh)
  734. {
  735. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  736. struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
  737. struct gfs2_holder i_gh;
  738. int error;
  739. restart:
  740. error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
  741. if (error)
  742. return error;
  743. qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
  744. if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
  745. gfs2_glock_dq_uninit(q_gh);
  746. error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
  747. GL_NOCACHE, q_gh);
  748. if (error)
  749. return error;
  750. error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
  751. if (error)
  752. goto fail;
  753. error = update_qd(sdp, qd);
  754. if (error)
  755. goto fail_gunlock;
  756. gfs2_glock_dq_uninit(&i_gh);
  757. gfs2_glock_dq_uninit(q_gh);
  758. force_refresh = 0;
  759. goto restart;
  760. }
  761. return 0;
  762. fail_gunlock:
  763. gfs2_glock_dq_uninit(&i_gh);
  764. fail:
  765. gfs2_glock_dq_uninit(q_gh);
  766. return error;
  767. }
  768. int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
  769. {
  770. struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
  771. struct gfs2_quota_data *qd;
  772. unsigned int x;
  773. int error = 0;
  774. error = gfs2_quota_hold(ip, uid, gid);
  775. if (error)
  776. return error;
  777. if (capable(CAP_SYS_RESOURCE) ||
  778. sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
  779. return 0;
  780. sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
  781. sizeof(struct gfs2_quota_data *), sort_qd, NULL);
  782. for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
  783. int force = NO_FORCE;
  784. qd = ip->i_res->rs_qa_qd[x];
  785. if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
  786. force = FORCE;
  787. error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
  788. if (error)
  789. break;
  790. }
  791. if (!error)
  792. set_bit(GIF_QD_LOCKED, &ip->i_flags);
  793. else {
  794. while (x--)
  795. gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
  796. gfs2_quota_unhold(ip);
  797. }
  798. return error;
  799. }
  800. static int need_sync(struct gfs2_quota_data *qd)
  801. {
  802. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  803. struct gfs2_tune *gt = &sdp->sd_tune;
  804. s64 value;
  805. unsigned int num, den;
  806. int do_sync = 1;
  807. if (!qd->qd_qb.qb_limit)
  808. return 0;
  809. spin_lock(&qd_lock);
  810. value = qd->qd_change;
  811. spin_unlock(&qd_lock);
  812. spin_lock(&gt->gt_spin);
  813. num = gt->gt_quota_scale_num;
  814. den = gt->gt_quota_scale_den;
  815. spin_unlock(&gt->gt_spin);
  816. if (value < 0)
  817. do_sync = 0;
  818. else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
  819. (s64)be64_to_cpu(qd->qd_qb.qb_limit))
  820. do_sync = 0;
  821. else {
  822. value *= gfs2_jindex_size(sdp) * num;
  823. value = div_s64(value, den);
  824. value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
  825. if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
  826. do_sync = 0;
  827. }
  828. return do_sync;
  829. }
  830. void gfs2_quota_unlock(struct gfs2_inode *ip)
  831. {
  832. struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
  833. struct gfs2_quota_data *qda[4];
  834. unsigned int count = 0;
  835. unsigned int x;
  836. int found;
  837. if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
  838. goto out;
  839. for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
  840. struct gfs2_quota_data *qd;
  841. int sync;
  842. qd = ip->i_res->rs_qa_qd[x];
  843. sync = need_sync(qd);
  844. gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
  845. if (!sync)
  846. continue;
  847. spin_lock(&qd_lock);
  848. found = qd_check_sync(sdp, qd, NULL);
  849. spin_unlock(&qd_lock);
  850. if (!found)
  851. continue;
  852. gfs2_assert_warn(sdp, qd->qd_change_sync);
  853. if (bh_get(qd)) {
  854. clear_bit(QDF_LOCKED, &qd->qd_flags);
  855. slot_put(qd);
  856. qd_put(qd);
  857. continue;
  858. }
  859. qda[count++] = qd;
  860. }
  861. if (count) {
  862. do_sync(count, qda);
  863. for (x = 0; x < count; x++)
  864. qd_unlock(qda[x]);
  865. }
  866. out:
  867. gfs2_quota_unhold(ip);
  868. }
  869. #define MAX_LINE 256
  870. static int print_message(struct gfs2_quota_data *qd, char *type)
  871. {
  872. struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
  873. printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
  874. sdp->sd_fsname, type,
  875. (qd->qd_id.type == USRQUOTA) ? "user" : "group",
  876. from_kqid(&init_user_ns, qd->qd_id));
  877. return 0;
  878. }
  879. int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
  880. {
  881. struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
  882. struct gfs2_quota_data *qd;
  883. s64 value;
  884. unsigned int x;
  885. int error = 0;
  886. if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
  887. return 0;
  888. if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
  889. return 0;
  890. for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
  891. qd = ip->i_res->rs_qa_qd[x];
  892. if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
  893. qid_eq(qd->qd_id, make_kqid_gid(gid))))
  894. continue;
  895. value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
  896. spin_lock(&qd_lock);
  897. value += qd->qd_change;
  898. spin_unlock(&qd_lock);
  899. if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
  900. print_message(qd, "exceeded");
  901. quota_send_warning(qd->qd_id,
  902. sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
  903. error = -EDQUOT;
  904. break;
  905. } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
  906. (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
  907. time_after_eq(jiffies, qd->qd_last_warn +
  908. gfs2_tune_get(sdp,
  909. gt_quota_warn_period) * HZ)) {
  910. quota_send_warning(qd->qd_id,
  911. sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
  912. error = print_message(qd, "warning");
  913. qd->qd_last_warn = jiffies;
  914. }
  915. }
  916. return error;
  917. }
  918. void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
  919. kuid_t uid, kgid_t gid)
  920. {
  921. struct gfs2_quota_data *qd;
  922. unsigned int x;
  923. if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
  924. return;
  925. if (ip->i_diskflags & GFS2_DIF_SYSTEM)
  926. return;
  927. for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
  928. qd = ip->i_res->rs_qa_qd[x];
  929. if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
  930. qid_eq(qd->qd_id, make_kqid_gid(gid))) {
  931. do_qc(qd, change);
  932. }
  933. }
  934. }
  935. int gfs2_quota_sync(struct super_block *sb, int type)
  936. {
  937. struct gfs2_sbd *sdp = sb->s_fs_info;
  938. struct gfs2_quota_data **qda;
  939. unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
  940. unsigned int num_qd;
  941. unsigned int x;
  942. int error = 0;
  943. qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
  944. if (!qda)
  945. return -ENOMEM;
  946. mutex_lock(&sdp->sd_quota_sync_mutex);
  947. sdp->sd_quota_sync_gen++;
  948. do {
  949. num_qd = 0;
  950. for (;;) {
  951. error = qd_fish(sdp, qda + num_qd);
  952. if (error || !qda[num_qd])
  953. break;
  954. if (++num_qd == max_qd)
  955. break;
  956. }
  957. if (num_qd) {
  958. if (!error)
  959. error = do_sync(num_qd, qda);
  960. if (!error)
  961. for (x = 0; x < num_qd; x++)
  962. qda[x]->qd_sync_gen =
  963. sdp->sd_quota_sync_gen;
  964. for (x = 0; x < num_qd; x++)
  965. qd_unlock(qda[x]);
  966. }
  967. } while (!error && num_qd == max_qd);
  968. mutex_unlock(&sdp->sd_quota_sync_mutex);
  969. kfree(qda);
  970. return error;
  971. }
  972. int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
  973. {
  974. struct gfs2_quota_data *qd;
  975. struct gfs2_holder q_gh;
  976. int error;
  977. error = qd_get(sdp, qid, &qd);
  978. if (error)
  979. return error;
  980. error = do_glock(qd, FORCE, &q_gh);
  981. if (!error)
  982. gfs2_glock_dq_uninit(&q_gh);
  983. qd_put(qd);
  984. return error;
  985. }
  986. static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
  987. {
  988. const struct gfs2_quota_change *str = buf;
  989. qc->qc_change = be64_to_cpu(str->qc_change);
  990. qc->qc_flags = be32_to_cpu(str->qc_flags);
  991. qc->qc_id = make_kqid(&init_user_ns,
  992. (qc->qc_flags & GFS2_QCF_USER)?USRQUOTA:GRPQUOTA,
  993. be32_to_cpu(str->qc_id));
  994. }
  995. int gfs2_quota_init(struct gfs2_sbd *sdp)
  996. {
  997. struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
  998. u64 size = i_size_read(sdp->sd_qc_inode);
  999. unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
  1000. unsigned int x, slot = 0;
  1001. unsigned int found = 0;
  1002. u64 dblock;
  1003. u32 extlen = 0;
  1004. int error;
  1005. if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
  1006. return -EIO;
  1007. sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
  1008. sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
  1009. error = -ENOMEM;
  1010. sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
  1011. sizeof(unsigned char *), GFP_NOFS);
  1012. if (!sdp->sd_quota_bitmap)
  1013. return error;
  1014. for (x = 0; x < sdp->sd_quota_chunks; x++) {
  1015. sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
  1016. if (!sdp->sd_quota_bitmap[x])
  1017. goto fail;
  1018. }
  1019. for (x = 0; x < blocks; x++) {
  1020. struct buffer_head *bh;
  1021. unsigned int y;
  1022. if (!extlen) {
  1023. int new = 0;
  1024. error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
  1025. if (error)
  1026. goto fail;
  1027. }
  1028. error = -EIO;
  1029. bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
  1030. if (!bh)
  1031. goto fail;
  1032. if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
  1033. brelse(bh);
  1034. goto fail;
  1035. }
  1036. for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
  1037. y++, slot++) {
  1038. struct gfs2_quota_change_host qc;
  1039. struct gfs2_quota_data *qd;
  1040. gfs2_quota_change_in(&qc, bh->b_data +
  1041. sizeof(struct gfs2_meta_header) +
  1042. y * sizeof(struct gfs2_quota_change));
  1043. if (!qc.qc_change)
  1044. continue;
  1045. error = qd_alloc(sdp, qc.qc_id, &qd);
  1046. if (error) {
  1047. brelse(bh);
  1048. goto fail;
  1049. }
  1050. set_bit(QDF_CHANGE, &qd->qd_flags);
  1051. qd->qd_change = qc.qc_change;
  1052. qd->qd_slot = slot;
  1053. qd->qd_slot_count = 1;
  1054. spin_lock(&qd_lock);
  1055. gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
  1056. list_add(&qd->qd_list, &sdp->sd_quota_list);
  1057. atomic_inc(&sdp->sd_quota_count);
  1058. spin_unlock(&qd_lock);
  1059. found++;
  1060. }
  1061. brelse(bh);
  1062. dblock++;
  1063. extlen--;
  1064. }
  1065. if (found)
  1066. fs_info(sdp, "found %u quota changes\n", found);
  1067. return 0;
  1068. fail:
  1069. gfs2_quota_cleanup(sdp);
  1070. return error;
  1071. }
  1072. void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
  1073. {
  1074. struct list_head *head = &sdp->sd_quota_list;
  1075. struct gfs2_quota_data *qd;
  1076. unsigned int x;
  1077. spin_lock(&qd_lock);
  1078. while (!list_empty(head)) {
  1079. qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
  1080. /*
  1081. * To be removed in due course... we should be able to
  1082. * ensure that all refs to the qd have done by this point
  1083. * so that this rather odd test is not required
  1084. */
  1085. spin_lock(&qd->qd_lockref.lock);
  1086. if (qd->qd_lockref.count > 1 ||
  1087. (qd->qd_lockref.count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
  1088. spin_unlock(&qd->qd_lockref.lock);
  1089. list_move(&qd->qd_list, head);
  1090. spin_unlock(&qd_lock);
  1091. schedule();
  1092. spin_lock(&qd_lock);
  1093. continue;
  1094. }
  1095. spin_unlock(&qd->qd_lockref.lock);
  1096. list_del(&qd->qd_list);
  1097. /* Also remove if this qd exists in the reclaim list */
  1098. if (!list_empty(&qd->qd_reclaim)) {
  1099. list_del_init(&qd->qd_reclaim);
  1100. atomic_dec(&qd_lru_count);
  1101. }
  1102. atomic_dec(&sdp->sd_quota_count);
  1103. spin_unlock(&qd_lock);
  1104. if (!qd->qd_lockref.count) {
  1105. gfs2_assert_warn(sdp, !qd->qd_change);
  1106. gfs2_assert_warn(sdp, !qd->qd_slot_count);
  1107. } else
  1108. gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
  1109. gfs2_assert_warn(sdp, !qd->qd_bh_count);
  1110. gfs2_glock_put(qd->qd_gl);
  1111. kmem_cache_free(gfs2_quotad_cachep, qd);
  1112. spin_lock(&qd_lock);
  1113. }
  1114. spin_unlock(&qd_lock);
  1115. gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
  1116. if (sdp->sd_quota_bitmap) {
  1117. for (x = 0; x < sdp->sd_quota_chunks; x++)
  1118. kfree(sdp->sd_quota_bitmap[x]);
  1119. kfree(sdp->sd_quota_bitmap);
  1120. }
  1121. }
  1122. static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
  1123. {
  1124. if (error == 0 || error == -EROFS)
  1125. return;
  1126. if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
  1127. fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
  1128. }
  1129. static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
  1130. int (*fxn)(struct super_block *sb, int type),
  1131. unsigned long t, unsigned long *timeo,
  1132. unsigned int *new_timeo)
  1133. {
  1134. if (t >= *timeo) {
  1135. int error = fxn(sdp->sd_vfs, 0);
  1136. quotad_error(sdp, msg, error);
  1137. *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
  1138. } else {
  1139. *timeo -= t;
  1140. }
  1141. }
  1142. static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
  1143. {
  1144. struct gfs2_inode *ip;
  1145. while(1) {
  1146. ip = NULL;
  1147. spin_lock(&sdp->sd_trunc_lock);
  1148. if (!list_empty(&sdp->sd_trunc_list)) {
  1149. ip = list_entry(sdp->sd_trunc_list.next,
  1150. struct gfs2_inode, i_trunc_list);
  1151. list_del_init(&ip->i_trunc_list);
  1152. }
  1153. spin_unlock(&sdp->sd_trunc_lock);
  1154. if (ip == NULL)
  1155. return;
  1156. gfs2_glock_finish_truncate(ip);
  1157. }
  1158. }
  1159. void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
  1160. if (!sdp->sd_statfs_force_sync) {
  1161. sdp->sd_statfs_force_sync = 1;
  1162. wake_up(&sdp->sd_quota_wait);
  1163. }
  1164. }
  1165. /**
  1166. * gfs2_quotad - Write cached quota changes into the quota file
  1167. * @sdp: Pointer to GFS2 superblock
  1168. *
  1169. */
  1170. int gfs2_quotad(void *data)
  1171. {
  1172. struct gfs2_sbd *sdp = data;
  1173. struct gfs2_tune *tune = &sdp->sd_tune;
  1174. unsigned long statfs_timeo = 0;
  1175. unsigned long quotad_timeo = 0;
  1176. unsigned long t = 0;
  1177. DEFINE_WAIT(wait);
  1178. int empty;
  1179. while (!kthread_should_stop()) {
  1180. /* Update the master statfs file */
  1181. if (sdp->sd_statfs_force_sync) {
  1182. int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
  1183. quotad_error(sdp, "statfs", error);
  1184. statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
  1185. }
  1186. else
  1187. quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
  1188. &statfs_timeo,
  1189. &tune->gt_statfs_quantum);
  1190. /* Update quota file */
  1191. quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
  1192. &quotad_timeo, &tune->gt_quota_quantum);
  1193. /* Check for & recover partially truncated inodes */
  1194. quotad_check_trunc_list(sdp);
  1195. try_to_freeze();
  1196. t = min(quotad_timeo, statfs_timeo);
  1197. prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
  1198. spin_lock(&sdp->sd_trunc_lock);
  1199. empty = list_empty(&sdp->sd_trunc_list);
  1200. spin_unlock(&sdp->sd_trunc_lock);
  1201. if (empty && !sdp->sd_statfs_force_sync)
  1202. t -= schedule_timeout(t);
  1203. else
  1204. t = 0;
  1205. finish_wait(&sdp->sd_quota_wait, &wait);
  1206. }
  1207. return 0;
  1208. }
  1209. static int gfs2_quota_get_xstate(struct super_block *sb,
  1210. struct fs_quota_stat *fqs)
  1211. {
  1212. struct gfs2_sbd *sdp = sb->s_fs_info;
  1213. memset(fqs, 0, sizeof(struct fs_quota_stat));
  1214. fqs->qs_version = FS_QSTAT_VERSION;
  1215. switch (sdp->sd_args.ar_quota) {
  1216. case GFS2_QUOTA_ON:
  1217. fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
  1218. /*FALLTHRU*/
  1219. case GFS2_QUOTA_ACCOUNT:
  1220. fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
  1221. break;
  1222. case GFS2_QUOTA_OFF:
  1223. break;
  1224. }
  1225. if (sdp->sd_quota_inode) {
  1226. fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
  1227. fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
  1228. }
  1229. fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
  1230. fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
  1231. fqs->qs_incoredqs = atomic_read(&qd_lru_count);
  1232. return 0;
  1233. }
  1234. static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
  1235. struct fs_disk_quota *fdq)
  1236. {
  1237. struct gfs2_sbd *sdp = sb->s_fs_info;
  1238. struct gfs2_quota_lvb *qlvb;
  1239. struct gfs2_quota_data *qd;
  1240. struct gfs2_holder q_gh;
  1241. int error;
  1242. memset(fdq, 0, sizeof(struct fs_disk_quota));
  1243. if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
  1244. return -ESRCH; /* Crazy XFS error code */
  1245. if ((qid.type != USRQUOTA) &&
  1246. (qid.type != GRPQUOTA))
  1247. return -EINVAL;
  1248. error = qd_get(sdp, qid, &qd);
  1249. if (error)
  1250. return error;
  1251. error = do_glock(qd, FORCE, &q_gh);
  1252. if (error)
  1253. goto out;
  1254. qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
  1255. fdq->d_version = FS_DQUOT_VERSION;
  1256. fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
  1257. fdq->d_id = from_kqid_munged(current_user_ns(), qid);
  1258. fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
  1259. fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
  1260. fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
  1261. gfs2_glock_dq_uninit(&q_gh);
  1262. out:
  1263. qd_put(qd);
  1264. return error;
  1265. }
  1266. /* GFS2 only supports a subset of the XFS fields */
  1267. #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
  1268. static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
  1269. struct fs_disk_quota *fdq)
  1270. {
  1271. struct gfs2_sbd *sdp = sb->s_fs_info;
  1272. struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
  1273. struct gfs2_quota_data *qd;
  1274. struct gfs2_holder q_gh, i_gh;
  1275. unsigned int data_blocks, ind_blocks;
  1276. unsigned int blocks = 0;
  1277. int alloc_required;
  1278. loff_t offset;
  1279. int error;
  1280. if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
  1281. return -ESRCH; /* Crazy XFS error code */
  1282. if ((qid.type != USRQUOTA) &&
  1283. (qid.type != GRPQUOTA))
  1284. return -EINVAL;
  1285. if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
  1286. return -EINVAL;
  1287. error = qd_get(sdp, qid, &qd);
  1288. if (error)
  1289. return error;
  1290. error = gfs2_rs_alloc(ip);
  1291. if (error)
  1292. goto out_put;
  1293. mutex_lock(&ip->i_inode.i_mutex);
  1294. error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
  1295. if (error)
  1296. goto out_unlockput;
  1297. error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
  1298. if (error)
  1299. goto out_q;
  1300. /* Check for existing entry, if none then alloc new blocks */
  1301. error = update_qd(sdp, qd);
  1302. if (error)
  1303. goto out_i;
  1304. /* If nothing has changed, this is a no-op */
  1305. if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
  1306. ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
  1307. fdq->d_fieldmask ^= FS_DQ_BSOFT;
  1308. if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
  1309. ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
  1310. fdq->d_fieldmask ^= FS_DQ_BHARD;
  1311. if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
  1312. ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
  1313. fdq->d_fieldmask ^= FS_DQ_BCOUNT;
  1314. if (fdq->d_fieldmask == 0)
  1315. goto out_i;
  1316. offset = qd2offset(qd);
  1317. alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
  1318. if (gfs2_is_stuffed(ip))
  1319. alloc_required = 1;
  1320. if (alloc_required) {
  1321. struct gfs2_alloc_parms ap = { .aflags = 0, };
  1322. gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
  1323. &data_blocks, &ind_blocks);
  1324. blocks = 1 + data_blocks + ind_blocks;
  1325. ap.target = blocks;
  1326. error = gfs2_inplace_reserve(ip, &ap);
  1327. if (error)
  1328. goto out_i;
  1329. blocks += gfs2_rg_blocks(ip, blocks);
  1330. }
  1331. /* Some quotas span block boundaries and can update two blocks,
  1332. adding an extra block to the transaction to handle such quotas */
  1333. error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
  1334. if (error)
  1335. goto out_release;
  1336. /* Apply changes */
  1337. error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
  1338. gfs2_trans_end(sdp);
  1339. out_release:
  1340. if (alloc_required)
  1341. gfs2_inplace_release(ip);
  1342. out_i:
  1343. gfs2_glock_dq_uninit(&i_gh);
  1344. out_q:
  1345. gfs2_glock_dq_uninit(&q_gh);
  1346. out_unlockput:
  1347. mutex_unlock(&ip->i_inode.i_mutex);
  1348. out_put:
  1349. qd_put(qd);
  1350. return error;
  1351. }
  1352. const struct quotactl_ops gfs2_quotactl_ops = {
  1353. .quota_sync = gfs2_quota_sync,
  1354. .get_xstate = gfs2_quota_get_xstate,
  1355. .get_dqblk = gfs2_get_dqblk,
  1356. .set_dqblk = gfs2_set_dqblk,
  1357. };