quota_global.c 27 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930
  1. /*
  2. * Implementation of operations over global quota file
  3. */
  4. #include <linux/spinlock.h>
  5. #include <linux/fs.h>
  6. #include <linux/slab.h>
  7. #include <linux/quota.h>
  8. #include <linux/quotaops.h>
  9. #include <linux/dqblk_qtree.h>
  10. #include <linux/jiffies.h>
  11. #include <linux/writeback.h>
  12. #include <linux/workqueue.h>
  13. #include <cluster/masklog.h>
  14. #include "ocfs2_fs.h"
  15. #include "ocfs2.h"
  16. #include "alloc.h"
  17. #include "blockcheck.h"
  18. #include "inode.h"
  19. #include "journal.h"
  20. #include "file.h"
  21. #include "sysfile.h"
  22. #include "dlmglue.h"
  23. #include "uptodate.h"
  24. #include "super.h"
  25. #include "buffer_head_io.h"
  26. #include "quota.h"
  27. #include "ocfs2_trace.h"
  28. /*
  29. * Locking of quotas with OCFS2 is rather complex. Here are rules that
  30. * should be obeyed by all the functions:
  31. * - any write of quota structure (either to local or global file) is protected
  32. * by dqio_mutex or dquot->dq_lock.
  33. * - any modification of global quota file holds inode cluster lock, i_mutex,
  34. * and ip_alloc_sem of the global quota file (achieved by
  35. * ocfs2_lock_global_qf). It also has to hold qinfo_lock.
  36. * - an allocation of new blocks for local quota file is protected by
  37. * its ip_alloc_sem
  38. *
  39. * A rough sketch of locking dependencies (lf = local file, gf = global file):
  40. * Normal filesystem operation:
  41. * start_trans -> dqio_mutex -> write to lf
  42. * Syncing of local and global file:
  43. * ocfs2_lock_global_qf -> start_trans -> dqio_mutex -> qinfo_lock ->
  44. * write to gf
  45. * -> write to lf
  46. * Acquire dquot for the first time:
  47. * dq_lock -> ocfs2_lock_global_qf -> qinfo_lock -> read from gf
  48. * -> alloc space for gf
  49. * -> start_trans -> qinfo_lock -> write to gf
  50. * -> ip_alloc_sem of lf -> alloc space for lf
  51. * -> write to lf
  52. * Release last reference to dquot:
  53. * dq_lock -> ocfs2_lock_global_qf -> start_trans -> qinfo_lock -> write to gf
  54. * -> write to lf
  55. * Note that all the above operations also hold the inode cluster lock of lf.
  56. * Recovery:
  57. * inode cluster lock of recovered lf
  58. * -> read bitmaps -> ip_alloc_sem of lf
  59. * -> ocfs2_lock_global_qf -> start_trans -> dqio_mutex -> qinfo_lock ->
  60. * write to gf
  61. */
  62. static void qsync_work_fn(struct work_struct *work);
  63. static void ocfs2_global_disk2memdqb(struct dquot *dquot, void *dp)
  64. {
  65. struct ocfs2_global_disk_dqblk *d = dp;
  66. struct mem_dqblk *m = &dquot->dq_dqb;
  67. /* Update from disk only entries not set by the admin */
  68. if (!test_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags)) {
  69. m->dqb_ihardlimit = le64_to_cpu(d->dqb_ihardlimit);
  70. m->dqb_isoftlimit = le64_to_cpu(d->dqb_isoftlimit);
  71. }
  72. if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
  73. m->dqb_curinodes = le64_to_cpu(d->dqb_curinodes);
  74. if (!test_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags)) {
  75. m->dqb_bhardlimit = le64_to_cpu(d->dqb_bhardlimit);
  76. m->dqb_bsoftlimit = le64_to_cpu(d->dqb_bsoftlimit);
  77. }
  78. if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
  79. m->dqb_curspace = le64_to_cpu(d->dqb_curspace);
  80. if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags))
  81. m->dqb_btime = le64_to_cpu(d->dqb_btime);
  82. if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags))
  83. m->dqb_itime = le64_to_cpu(d->dqb_itime);
  84. OCFS2_DQUOT(dquot)->dq_use_count = le32_to_cpu(d->dqb_use_count);
  85. }
  86. static void ocfs2_global_mem2diskdqb(void *dp, struct dquot *dquot)
  87. {
  88. struct ocfs2_global_disk_dqblk *d = dp;
  89. struct mem_dqblk *m = &dquot->dq_dqb;
  90. d->dqb_id = cpu_to_le32(dquot->dq_id);
  91. d->dqb_use_count = cpu_to_le32(OCFS2_DQUOT(dquot)->dq_use_count);
  92. d->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
  93. d->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
  94. d->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
  95. d->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
  96. d->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
  97. d->dqb_curspace = cpu_to_le64(m->dqb_curspace);
  98. d->dqb_btime = cpu_to_le64(m->dqb_btime);
  99. d->dqb_itime = cpu_to_le64(m->dqb_itime);
  100. d->dqb_pad1 = d->dqb_pad2 = 0;
  101. }
  102. static int ocfs2_global_is_id(void *dp, struct dquot *dquot)
  103. {
  104. struct ocfs2_global_disk_dqblk *d = dp;
  105. struct ocfs2_mem_dqinfo *oinfo =
  106. sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
  107. if (qtree_entry_unused(&oinfo->dqi_gi, dp))
  108. return 0;
  109. return le32_to_cpu(d->dqb_id) == dquot->dq_id;
  110. }
  111. struct qtree_fmt_operations ocfs2_global_ops = {
  112. .mem2disk_dqblk = ocfs2_global_mem2diskdqb,
  113. .disk2mem_dqblk = ocfs2_global_disk2memdqb,
  114. .is_id = ocfs2_global_is_id,
  115. };
  116. int ocfs2_validate_quota_block(struct super_block *sb, struct buffer_head *bh)
  117. {
  118. struct ocfs2_disk_dqtrailer *dqt =
  119. ocfs2_block_dqtrailer(sb->s_blocksize, bh->b_data);
  120. trace_ocfs2_validate_quota_block((unsigned long long)bh->b_blocknr);
  121. BUG_ON(!buffer_uptodate(bh));
  122. /*
  123. * If the ecc fails, we return the error but otherwise
  124. * leave the filesystem running. We know any error is
  125. * local to this block.
  126. */
  127. return ocfs2_validate_meta_ecc(sb, bh->b_data, &dqt->dq_check);
  128. }
  129. int ocfs2_read_quota_phys_block(struct inode *inode, u64 p_block,
  130. struct buffer_head **bhp)
  131. {
  132. int rc;
  133. *bhp = NULL;
  134. rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, 1, bhp, 0,
  135. ocfs2_validate_quota_block);
  136. if (rc)
  137. mlog_errno(rc);
  138. return rc;
  139. }
  140. /* Read data from global quotafile - avoid pagecache and such because we cannot
  141. * afford acquiring the locks... We use quota cluster lock to serialize
  142. * operations. Caller is responsible for acquiring it. */
  143. ssize_t ocfs2_quota_read(struct super_block *sb, int type, char *data,
  144. size_t len, loff_t off)
  145. {
  146. struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
  147. struct inode *gqinode = oinfo->dqi_gqinode;
  148. loff_t i_size = i_size_read(gqinode);
  149. int offset = off & (sb->s_blocksize - 1);
  150. sector_t blk = off >> sb->s_blocksize_bits;
  151. int err = 0;
  152. struct buffer_head *bh;
  153. size_t toread, tocopy;
  154. u64 pblock = 0, pcount = 0;
  155. if (off > i_size)
  156. return 0;
  157. if (off + len > i_size)
  158. len = i_size - off;
  159. toread = len;
  160. while (toread > 0) {
  161. tocopy = min_t(size_t, (sb->s_blocksize - offset), toread);
  162. if (!pcount) {
  163. err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock,
  164. &pcount, NULL);
  165. if (err) {
  166. mlog_errno(err);
  167. return err;
  168. }
  169. } else {
  170. pcount--;
  171. pblock++;
  172. }
  173. bh = NULL;
  174. err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
  175. if (err) {
  176. mlog_errno(err);
  177. return err;
  178. }
  179. memcpy(data, bh->b_data + offset, tocopy);
  180. brelse(bh);
  181. offset = 0;
  182. toread -= tocopy;
  183. data += tocopy;
  184. blk++;
  185. }
  186. return len;
  187. }
  188. /* Write to quotafile (we know the transaction is already started and has
  189. * enough credits) */
  190. ssize_t ocfs2_quota_write(struct super_block *sb, int type,
  191. const char *data, size_t len, loff_t off)
  192. {
  193. struct mem_dqinfo *info = sb_dqinfo(sb, type);
  194. struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
  195. struct inode *gqinode = oinfo->dqi_gqinode;
  196. int offset = off & (sb->s_blocksize - 1);
  197. sector_t blk = off >> sb->s_blocksize_bits;
  198. int err = 0, new = 0, ja_type;
  199. struct buffer_head *bh = NULL;
  200. handle_t *handle = journal_current_handle();
  201. u64 pblock, pcount;
  202. if (!handle) {
  203. mlog(ML_ERROR, "Quota write (off=%llu, len=%llu) cancelled "
  204. "because transaction was not started.\n",
  205. (unsigned long long)off, (unsigned long long)len);
  206. return -EIO;
  207. }
  208. if (len > sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset) {
  209. WARN_ON(1);
  210. len = sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset;
  211. }
  212. if (gqinode->i_size < off + len) {
  213. loff_t rounded_end =
  214. ocfs2_align_bytes_to_blocks(sb, off + len);
  215. /* Space is already allocated in ocfs2_acquire_dquot() */
  216. err = ocfs2_simple_size_update(gqinode,
  217. oinfo->dqi_gqi_bh,
  218. rounded_end);
  219. if (err < 0)
  220. goto out;
  221. new = 1;
  222. }
  223. err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock, &pcount, NULL);
  224. if (err) {
  225. mlog_errno(err);
  226. goto out;
  227. }
  228. /* Not rewriting whole block? */
  229. if ((offset || len < sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE) &&
  230. !new) {
  231. err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
  232. ja_type = OCFS2_JOURNAL_ACCESS_WRITE;
  233. } else {
  234. bh = sb_getblk(sb, pblock);
  235. if (!bh)
  236. err = -ENOMEM;
  237. ja_type = OCFS2_JOURNAL_ACCESS_CREATE;
  238. }
  239. if (err) {
  240. mlog_errno(err);
  241. goto out;
  242. }
  243. lock_buffer(bh);
  244. if (new)
  245. memset(bh->b_data, 0, sb->s_blocksize);
  246. memcpy(bh->b_data + offset, data, len);
  247. flush_dcache_page(bh->b_page);
  248. set_buffer_uptodate(bh);
  249. unlock_buffer(bh);
  250. ocfs2_set_buffer_uptodate(INODE_CACHE(gqinode), bh);
  251. err = ocfs2_journal_access_dq(handle, INODE_CACHE(gqinode), bh,
  252. ja_type);
  253. if (err < 0) {
  254. brelse(bh);
  255. goto out;
  256. }
  257. ocfs2_journal_dirty(handle, bh);
  258. brelse(bh);
  259. out:
  260. if (err) {
  261. mlog_errno(err);
  262. return err;
  263. }
  264. gqinode->i_version++;
  265. ocfs2_mark_inode_dirty(handle, gqinode, oinfo->dqi_gqi_bh);
  266. return len;
  267. }
  268. int ocfs2_lock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
  269. {
  270. int status;
  271. struct buffer_head *bh = NULL;
  272. status = ocfs2_inode_lock(oinfo->dqi_gqinode, &bh, ex);
  273. if (status < 0)
  274. return status;
  275. spin_lock(&dq_data_lock);
  276. if (!oinfo->dqi_gqi_count++)
  277. oinfo->dqi_gqi_bh = bh;
  278. else
  279. WARN_ON(bh != oinfo->dqi_gqi_bh);
  280. spin_unlock(&dq_data_lock);
  281. if (ex) {
  282. mutex_lock(&oinfo->dqi_gqinode->i_mutex);
  283. down_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
  284. } else {
  285. down_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
  286. }
  287. return 0;
  288. }
  289. void ocfs2_unlock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
  290. {
  291. if (ex) {
  292. up_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
  293. mutex_unlock(&oinfo->dqi_gqinode->i_mutex);
  294. } else {
  295. up_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
  296. }
  297. ocfs2_inode_unlock(oinfo->dqi_gqinode, ex);
  298. brelse(oinfo->dqi_gqi_bh);
  299. spin_lock(&dq_data_lock);
  300. if (!--oinfo->dqi_gqi_count)
  301. oinfo->dqi_gqi_bh = NULL;
  302. spin_unlock(&dq_data_lock);
  303. }
  304. /* Read information header from global quota file */
  305. int ocfs2_global_read_info(struct super_block *sb, int type)
  306. {
  307. struct inode *gqinode = NULL;
  308. unsigned int ino[MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
  309. GROUP_QUOTA_SYSTEM_INODE };
  310. struct ocfs2_global_disk_dqinfo dinfo;
  311. struct mem_dqinfo *info = sb_dqinfo(sb, type);
  312. struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
  313. u64 pcount;
  314. int status;
  315. /* Read global header */
  316. gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
  317. OCFS2_INVALID_SLOT);
  318. if (!gqinode) {
  319. mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
  320. type);
  321. status = -EINVAL;
  322. goto out_err;
  323. }
  324. oinfo->dqi_gi.dqi_sb = sb;
  325. oinfo->dqi_gi.dqi_type = type;
  326. ocfs2_qinfo_lock_res_init(&oinfo->dqi_gqlock, oinfo);
  327. oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
  328. oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
  329. oinfo->dqi_gqi_bh = NULL;
  330. oinfo->dqi_gqi_count = 0;
  331. oinfo->dqi_gqinode = gqinode;
  332. status = ocfs2_lock_global_qf(oinfo, 0);
  333. if (status < 0) {
  334. mlog_errno(status);
  335. goto out_err;
  336. }
  337. status = ocfs2_extent_map_get_blocks(gqinode, 0, &oinfo->dqi_giblk,
  338. &pcount, NULL);
  339. if (status < 0)
  340. goto out_unlock;
  341. status = ocfs2_qinfo_lock(oinfo, 0);
  342. if (status < 0)
  343. goto out_unlock;
  344. status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
  345. sizeof(struct ocfs2_global_disk_dqinfo),
  346. OCFS2_GLOBAL_INFO_OFF);
  347. ocfs2_qinfo_unlock(oinfo, 0);
  348. ocfs2_unlock_global_qf(oinfo, 0);
  349. if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
  350. mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
  351. status);
  352. if (status >= 0)
  353. status = -EIO;
  354. mlog_errno(status);
  355. goto out_err;
  356. }
  357. info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
  358. info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
  359. oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
  360. oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
  361. oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
  362. oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
  363. oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
  364. oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
  365. OCFS2_QBLK_RESERVED_SPACE;
  366. oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
  367. INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
  368. schedule_delayed_work(&oinfo->dqi_sync_work,
  369. msecs_to_jiffies(oinfo->dqi_syncms));
  370. out_err:
  371. if (status)
  372. mlog_errno(status);
  373. return status;
  374. out_unlock:
  375. ocfs2_unlock_global_qf(oinfo, 0);
  376. mlog_errno(status);
  377. goto out_err;
  378. }
  379. /* Write information to global quota file. Expects exlusive lock on quota
  380. * file inode and quota info */
  381. static int __ocfs2_global_write_info(struct super_block *sb, int type)
  382. {
  383. struct mem_dqinfo *info = sb_dqinfo(sb, type);
  384. struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
  385. struct ocfs2_global_disk_dqinfo dinfo;
  386. ssize_t size;
  387. spin_lock(&dq_data_lock);
  388. info->dqi_flags &= ~DQF_INFO_DIRTY;
  389. dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
  390. dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
  391. spin_unlock(&dq_data_lock);
  392. dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
  393. dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
  394. dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
  395. dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
  396. size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
  397. sizeof(struct ocfs2_global_disk_dqinfo),
  398. OCFS2_GLOBAL_INFO_OFF);
  399. if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
  400. mlog(ML_ERROR, "Cannot write global quota info structure\n");
  401. if (size >= 0)
  402. size = -EIO;
  403. return size;
  404. }
  405. return 0;
  406. }
  407. int ocfs2_global_write_info(struct super_block *sb, int type)
  408. {
  409. int err;
  410. struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
  411. err = ocfs2_qinfo_lock(info, 1);
  412. if (err < 0)
  413. return err;
  414. err = __ocfs2_global_write_info(sb, type);
  415. ocfs2_qinfo_unlock(info, 1);
  416. return err;
  417. }
  418. static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
  419. {
  420. struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
  421. /*
  422. * We may need to allocate tree blocks and a leaf block but not the
  423. * root block
  424. */
  425. return oinfo->dqi_gi.dqi_qtree_depth;
  426. }
  427. static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
  428. {
  429. /* We modify all the allocated blocks, tree root, info block and
  430. * the inode */
  431. return (ocfs2_global_qinit_alloc(sb, type) + 2) *
  432. OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
  433. }
  434. /* Sync local information about quota modifications with global quota file.
  435. * Caller must have started the transaction and obtained exclusive lock for
  436. * global quota file inode */
  437. int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
  438. {
  439. int err, err2;
  440. struct super_block *sb = dquot->dq_sb;
  441. int type = dquot->dq_type;
  442. struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
  443. struct ocfs2_global_disk_dqblk dqblk;
  444. s64 spacechange, inodechange;
  445. time_t olditime, oldbtime;
  446. err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
  447. sizeof(struct ocfs2_global_disk_dqblk),
  448. dquot->dq_off);
  449. if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
  450. if (err >= 0) {
  451. mlog(ML_ERROR, "Short read from global quota file "
  452. "(%u read)\n", err);
  453. err = -EIO;
  454. }
  455. goto out;
  456. }
  457. /* Update space and inode usage. Get also other information from
  458. * global quota file so that we don't overwrite any changes there.
  459. * We are */
  460. spin_lock(&dq_data_lock);
  461. spacechange = dquot->dq_dqb.dqb_curspace -
  462. OCFS2_DQUOT(dquot)->dq_origspace;
  463. inodechange = dquot->dq_dqb.dqb_curinodes -
  464. OCFS2_DQUOT(dquot)->dq_originodes;
  465. olditime = dquot->dq_dqb.dqb_itime;
  466. oldbtime = dquot->dq_dqb.dqb_btime;
  467. ocfs2_global_disk2memdqb(dquot, &dqblk);
  468. trace_ocfs2_sync_dquot(dquot->dq_id, dquot->dq_dqb.dqb_curspace,
  469. (long long)spacechange,
  470. dquot->dq_dqb.dqb_curinodes,
  471. (long long)inodechange);
  472. if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
  473. dquot->dq_dqb.dqb_curspace += spacechange;
  474. if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
  475. dquot->dq_dqb.dqb_curinodes += inodechange;
  476. /* Set properly space grace time... */
  477. if (dquot->dq_dqb.dqb_bsoftlimit &&
  478. dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
  479. if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
  480. oldbtime > 0) {
  481. if (dquot->dq_dqb.dqb_btime > 0)
  482. dquot->dq_dqb.dqb_btime =
  483. min(dquot->dq_dqb.dqb_btime, oldbtime);
  484. else
  485. dquot->dq_dqb.dqb_btime = oldbtime;
  486. }
  487. } else {
  488. dquot->dq_dqb.dqb_btime = 0;
  489. clear_bit(DQ_BLKS_B, &dquot->dq_flags);
  490. }
  491. /* Set properly inode grace time... */
  492. if (dquot->dq_dqb.dqb_isoftlimit &&
  493. dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
  494. if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
  495. olditime > 0) {
  496. if (dquot->dq_dqb.dqb_itime > 0)
  497. dquot->dq_dqb.dqb_itime =
  498. min(dquot->dq_dqb.dqb_itime, olditime);
  499. else
  500. dquot->dq_dqb.dqb_itime = olditime;
  501. }
  502. } else {
  503. dquot->dq_dqb.dqb_itime = 0;
  504. clear_bit(DQ_INODES_B, &dquot->dq_flags);
  505. }
  506. /* All information is properly updated, clear the flags */
  507. __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
  508. __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
  509. __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
  510. __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
  511. __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
  512. __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
  513. OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
  514. OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
  515. spin_unlock(&dq_data_lock);
  516. err = ocfs2_qinfo_lock(info, freeing);
  517. if (err < 0) {
  518. mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
  519. " (type=%d, id=%u)\n", dquot->dq_type,
  520. (unsigned)dquot->dq_id);
  521. goto out;
  522. }
  523. if (freeing)
  524. OCFS2_DQUOT(dquot)->dq_use_count--;
  525. err = qtree_write_dquot(&info->dqi_gi, dquot);
  526. if (err < 0)
  527. goto out_qlock;
  528. if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
  529. err = qtree_release_dquot(&info->dqi_gi, dquot);
  530. if (info_dirty(sb_dqinfo(sb, type))) {
  531. err2 = __ocfs2_global_write_info(sb, type);
  532. if (!err)
  533. err = err2;
  534. }
  535. }
  536. out_qlock:
  537. ocfs2_qinfo_unlock(info, freeing);
  538. out:
  539. if (err < 0)
  540. mlog_errno(err);
  541. return err;
  542. }
  543. /*
  544. * Functions for periodic syncing of dquots with global file
  545. */
  546. static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
  547. {
  548. handle_t *handle;
  549. struct super_block *sb = dquot->dq_sb;
  550. struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
  551. struct ocfs2_super *osb = OCFS2_SB(sb);
  552. int status = 0;
  553. trace_ocfs2_sync_dquot_helper(dquot->dq_id, dquot->dq_type,
  554. type, sb->s_id);
  555. if (type != dquot->dq_type)
  556. goto out;
  557. status = ocfs2_lock_global_qf(oinfo, 1);
  558. if (status < 0)
  559. goto out;
  560. handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
  561. if (IS_ERR(handle)) {
  562. status = PTR_ERR(handle);
  563. mlog_errno(status);
  564. goto out_ilock;
  565. }
  566. mutex_lock(&sb_dqopt(sb)->dqio_mutex);
  567. status = ocfs2_sync_dquot(dquot);
  568. if (status < 0)
  569. mlog_errno(status);
  570. /* We have to write local structure as well... */
  571. status = ocfs2_local_write_dquot(dquot);
  572. if (status < 0)
  573. mlog_errno(status);
  574. mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
  575. ocfs2_commit_trans(osb, handle);
  576. out_ilock:
  577. ocfs2_unlock_global_qf(oinfo, 1);
  578. out:
  579. return status;
  580. }
  581. static void qsync_work_fn(struct work_struct *work)
  582. {
  583. struct ocfs2_mem_dqinfo *oinfo = container_of(work,
  584. struct ocfs2_mem_dqinfo,
  585. dqi_sync_work.work);
  586. struct super_block *sb = oinfo->dqi_gqinode->i_sb;
  587. dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
  588. schedule_delayed_work(&oinfo->dqi_sync_work,
  589. msecs_to_jiffies(oinfo->dqi_syncms));
  590. }
  591. /*
  592. * Wrappers for generic quota functions
  593. */
  594. static int ocfs2_write_dquot(struct dquot *dquot)
  595. {
  596. handle_t *handle;
  597. struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
  598. int status = 0;
  599. trace_ocfs2_write_dquot(dquot->dq_id, dquot->dq_type);
  600. handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
  601. if (IS_ERR(handle)) {
  602. status = PTR_ERR(handle);
  603. mlog_errno(status);
  604. goto out;
  605. }
  606. mutex_lock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
  607. status = ocfs2_local_write_dquot(dquot);
  608. mutex_unlock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
  609. ocfs2_commit_trans(osb, handle);
  610. out:
  611. return status;
  612. }
  613. static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
  614. {
  615. struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
  616. /*
  617. * We modify tree, leaf block, global info, local chunk header,
  618. * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
  619. * accounts for inode update
  620. */
  621. return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
  622. OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
  623. OCFS2_QINFO_WRITE_CREDITS +
  624. OCFS2_INODE_UPDATE_CREDITS;
  625. }
  626. static int ocfs2_release_dquot(struct dquot *dquot)
  627. {
  628. handle_t *handle;
  629. struct ocfs2_mem_dqinfo *oinfo =
  630. sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
  631. struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
  632. int status = 0;
  633. trace_ocfs2_release_dquot(dquot->dq_id, dquot->dq_type);
  634. mutex_lock(&dquot->dq_lock);
  635. /* Check whether we are not racing with some other dqget() */
  636. if (atomic_read(&dquot->dq_count) > 1)
  637. goto out;
  638. status = ocfs2_lock_global_qf(oinfo, 1);
  639. if (status < 0)
  640. goto out;
  641. handle = ocfs2_start_trans(osb,
  642. ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_type));
  643. if (IS_ERR(handle)) {
  644. status = PTR_ERR(handle);
  645. mlog_errno(status);
  646. goto out_ilock;
  647. }
  648. status = ocfs2_global_release_dquot(dquot);
  649. if (status < 0) {
  650. mlog_errno(status);
  651. goto out_trans;
  652. }
  653. status = ocfs2_local_release_dquot(handle, dquot);
  654. /*
  655. * If we fail here, we cannot do much as global structure is
  656. * already released. So just complain...
  657. */
  658. if (status < 0)
  659. mlog_errno(status);
  660. /*
  661. * Clear dq_off so that we search for the structure in quota file next
  662. * time we acquire it. The structure might be deleted and reallocated
  663. * elsewhere by another node while our dquot structure is on freelist.
  664. */
  665. dquot->dq_off = 0;
  666. clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
  667. out_trans:
  668. ocfs2_commit_trans(osb, handle);
  669. out_ilock:
  670. ocfs2_unlock_global_qf(oinfo, 1);
  671. out:
  672. mutex_unlock(&dquot->dq_lock);
  673. if (status)
  674. mlog_errno(status);
  675. return status;
  676. }
  677. /*
  678. * Read global dquot structure from disk or create it if it does
  679. * not exist. Also update use count of the global structure and
  680. * create structure in node-local quota file.
  681. */
  682. static int ocfs2_acquire_dquot(struct dquot *dquot)
  683. {
  684. int status = 0, err;
  685. int ex = 0;
  686. struct super_block *sb = dquot->dq_sb;
  687. struct ocfs2_super *osb = OCFS2_SB(sb);
  688. int type = dquot->dq_type;
  689. struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
  690. struct inode *gqinode = info->dqi_gqinode;
  691. int need_alloc = ocfs2_global_qinit_alloc(sb, type);
  692. handle_t *handle;
  693. trace_ocfs2_acquire_dquot(dquot->dq_id, type);
  694. mutex_lock(&dquot->dq_lock);
  695. /*
  696. * We need an exclusive lock, because we're going to update use count
  697. * and instantiate possibly new dquot structure
  698. */
  699. status = ocfs2_lock_global_qf(info, 1);
  700. if (status < 0)
  701. goto out;
  702. status = ocfs2_qinfo_lock(info, 0);
  703. if (status < 0)
  704. goto out_dq;
  705. /*
  706. * We always want to read dquot structure from disk because we don't
  707. * know what happened with it while it was on freelist.
  708. */
  709. status = qtree_read_dquot(&info->dqi_gi, dquot);
  710. ocfs2_qinfo_unlock(info, 0);
  711. if (status < 0)
  712. goto out_dq;
  713. OCFS2_DQUOT(dquot)->dq_use_count++;
  714. OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
  715. OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
  716. if (!dquot->dq_off) { /* No real quota entry? */
  717. ex = 1;
  718. /*
  719. * Add blocks to quota file before we start a transaction since
  720. * locking allocators ranks above a transaction start
  721. */
  722. WARN_ON(journal_current_handle());
  723. status = ocfs2_extend_no_holes(gqinode, NULL,
  724. gqinode->i_size + (need_alloc << sb->s_blocksize_bits),
  725. gqinode->i_size);
  726. if (status < 0)
  727. goto out_dq;
  728. }
  729. handle = ocfs2_start_trans(osb,
  730. ocfs2_calc_global_qinit_credits(sb, type));
  731. if (IS_ERR(handle)) {
  732. status = PTR_ERR(handle);
  733. goto out_dq;
  734. }
  735. status = ocfs2_qinfo_lock(info, ex);
  736. if (status < 0)
  737. goto out_trans;
  738. status = qtree_write_dquot(&info->dqi_gi, dquot);
  739. if (ex && info_dirty(sb_dqinfo(sb, type))) {
  740. err = __ocfs2_global_write_info(sb, type);
  741. if (!status)
  742. status = err;
  743. }
  744. ocfs2_qinfo_unlock(info, ex);
  745. out_trans:
  746. ocfs2_commit_trans(osb, handle);
  747. out_dq:
  748. ocfs2_unlock_global_qf(info, 1);
  749. if (status < 0)
  750. goto out;
  751. status = ocfs2_create_local_dquot(dquot);
  752. if (status < 0)
  753. goto out;
  754. set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
  755. out:
  756. mutex_unlock(&dquot->dq_lock);
  757. if (status)
  758. mlog_errno(status);
  759. return status;
  760. }
  761. static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
  762. {
  763. unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
  764. (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
  765. (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
  766. (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
  767. (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
  768. (1 << (DQ_LASTSET_B + QIF_ITIME_B));
  769. int sync = 0;
  770. int status;
  771. struct super_block *sb = dquot->dq_sb;
  772. int type = dquot->dq_type;
  773. struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
  774. handle_t *handle;
  775. struct ocfs2_super *osb = OCFS2_SB(sb);
  776. trace_ocfs2_mark_dquot_dirty(dquot->dq_id, type);
  777. /* In case user set some limits, sync dquot immediately to global
  778. * quota file so that information propagates quicker */
  779. spin_lock(&dq_data_lock);
  780. if (dquot->dq_flags & mask)
  781. sync = 1;
  782. spin_unlock(&dq_data_lock);
  783. /* This is a slight hack but we can't afford getting global quota
  784. * lock if we already have a transaction started. */
  785. if (!sync || journal_current_handle()) {
  786. status = ocfs2_write_dquot(dquot);
  787. goto out;
  788. }
  789. status = ocfs2_lock_global_qf(oinfo, 1);
  790. if (status < 0)
  791. goto out;
  792. handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
  793. if (IS_ERR(handle)) {
  794. status = PTR_ERR(handle);
  795. mlog_errno(status);
  796. goto out_ilock;
  797. }
  798. mutex_lock(&sb_dqopt(sb)->dqio_mutex);
  799. status = ocfs2_sync_dquot(dquot);
  800. if (status < 0) {
  801. mlog_errno(status);
  802. goto out_dlock;
  803. }
  804. /* Now write updated local dquot structure */
  805. status = ocfs2_local_write_dquot(dquot);
  806. out_dlock:
  807. mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
  808. ocfs2_commit_trans(osb, handle);
  809. out_ilock:
  810. ocfs2_unlock_global_qf(oinfo, 1);
  811. out:
  812. if (status)
  813. mlog_errno(status);
  814. return status;
  815. }
  816. /* This should happen only after set_dqinfo(). */
  817. static int ocfs2_write_info(struct super_block *sb, int type)
  818. {
  819. handle_t *handle;
  820. int status = 0;
  821. struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
  822. status = ocfs2_lock_global_qf(oinfo, 1);
  823. if (status < 0)
  824. goto out;
  825. handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
  826. if (IS_ERR(handle)) {
  827. status = PTR_ERR(handle);
  828. mlog_errno(status);
  829. goto out_ilock;
  830. }
  831. status = dquot_commit_info(sb, type);
  832. ocfs2_commit_trans(OCFS2_SB(sb), handle);
  833. out_ilock:
  834. ocfs2_unlock_global_qf(oinfo, 1);
  835. out:
  836. if (status)
  837. mlog_errno(status);
  838. return status;
  839. }
  840. static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
  841. {
  842. struct ocfs2_dquot *dquot =
  843. kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
  844. if (!dquot)
  845. return NULL;
  846. return &dquot->dq_dquot;
  847. }
  848. static void ocfs2_destroy_dquot(struct dquot *dquot)
  849. {
  850. kmem_cache_free(ocfs2_dquot_cachep, dquot);
  851. }
  852. const struct dquot_operations ocfs2_quota_operations = {
  853. /* We never make dquot dirty so .write_dquot is never called */
  854. .acquire_dquot = ocfs2_acquire_dquot,
  855. .release_dquot = ocfs2_release_dquot,
  856. .mark_dirty = ocfs2_mark_dquot_dirty,
  857. .write_info = ocfs2_write_info,
  858. .alloc_dquot = ocfs2_alloc_dquot,
  859. .destroy_dquot = ocfs2_destroy_dquot,
  860. };