revoke.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712
  1. /*
  2. * linux/fs/revoke.c
  3. *
  4. * Written by Stephen C. Tweedie <sct@redhat.com>, 2000
  5. *
  6. * Copyright 2000 Red Hat corp --- All Rights Reserved
  7. *
  8. * This file is part of the Linux kernel and is made available under
  9. * the terms of the GNU General Public License, version 2, or at your
  10. * option, any later version, incorporated herein by reference.
  11. *
  12. * Journal revoke routines for the generic filesystem journaling code;
  13. * part of the ext2fs journaling system.
  14. *
  15. * Revoke is the mechanism used to prevent old log records for deleted
  16. * metadata from being replayed on top of newer data using the same
  17. * blocks. The revoke mechanism is used in two separate places:
  18. *
  19. * + Commit: during commit we write the entire list of the current
  20. * transaction's revoked blocks to the journal
  21. *
  22. * + Recovery: during recovery we record the transaction ID of all
  23. * revoked blocks. If there are multiple revoke records in the log
  24. * for a single block, only the last one counts, and if there is a log
  25. * entry for a block beyond the last revoke, then that log entry still
  26. * gets replayed.
  27. *
  28. * We can get interactions between revokes and new log data within a
  29. * single transaction:
  30. *
  31. * Block is revoked and then journaled:
  32. * The desired end result is the journaling of the new block, so we
  33. * cancel the revoke before the transaction commits.
  34. *
  35. * Block is journaled and then revoked:
  36. * The revoke must take precedence over the write of the block, so we
  37. * need either to cancel the journal entry or to write the revoke
  38. * later in the log than the log block. In this case, we choose the
  39. * latter: journaling a block cancels any revoke record for that block
  40. * in the current transaction, so any revoke for that block in the
  41. * transaction must have happened after the block was journaled and so
  42. * the revoke must take precedence.
  43. *
  44. * Block is revoked and then written as data:
  45. * The data write is allowed to succeed, but the revoke is _not_
  46. * cancelled. We still need to prevent old log records from
  47. * overwriting the new data. We don't even need to clear the revoke
  48. * bit here.
  49. *
  50. * Revoke information on buffers is a tri-state value:
  51. *
  52. * RevokeValid clear: no cached revoke status, need to look it up
  53. * RevokeValid set, Revoked clear:
  54. * buffer has not been revoked, and cancel_revoke
  55. * need do nothing.
  56. * RevokeValid set, Revoked set:
  57. * buffer has been revoked.
  58. */
  59. #ifndef __KERNEL__
  60. #include "jfs_user.h"
  61. #else
  62. #include <linux/time.h>
  63. #include <linux/fs.h>
  64. #include <linux/jbd2.h>
  65. #include <linux/errno.h>
  66. #include <linux/slab.h>
  67. #include <linux/list.h>
  68. #include <linux/smp_lock.h>
  69. #include <linux/init.h>
  70. #endif
  71. static struct kmem_cache *jbd2_revoke_record_cache;
  72. static struct kmem_cache *jbd2_revoke_table_cache;
  73. /* Each revoke record represents one single revoked block. During
  74. journal replay, this involves recording the transaction ID of the
  75. last transaction to revoke this block. */
  76. struct jbd2_revoke_record_s
  77. {
  78. struct list_head hash;
  79. tid_t sequence; /* Used for recovery only */
  80. unsigned long long blocknr;
  81. };
  82. /* The revoke table is just a simple hash table of revoke records. */
  83. struct jbd2_revoke_table_s
  84. {
  85. /* It is conceivable that we might want a larger hash table
  86. * for recovery. Must be a power of two. */
  87. int hash_size;
  88. int hash_shift;
  89. struct list_head *hash_table;
  90. };
  91. #ifdef __KERNEL__
  92. static void write_one_revoke_record(journal_t *, transaction_t *,
  93. struct journal_head **, int *,
  94. struct jbd2_revoke_record_s *);
  95. static void flush_descriptor(journal_t *, struct journal_head *, int);
  96. #endif
  97. /* Utility functions to maintain the revoke table */
  98. /* Borrowed from buffer.c: this is a tried and tested block hash function */
  99. static inline int hash(journal_t *journal, unsigned long long block)
  100. {
  101. struct jbd2_revoke_table_s *table = journal->j_revoke;
  102. int hash_shift = table->hash_shift;
  103. int hash = (int)block ^ (int)((block >> 31) >> 1);
  104. return ((hash << (hash_shift - 6)) ^
  105. (hash >> 13) ^
  106. (hash << (hash_shift - 12))) & (table->hash_size - 1);
  107. }
  108. static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr,
  109. tid_t seq)
  110. {
  111. struct list_head *hash_list;
  112. struct jbd2_revoke_record_s *record;
  113. repeat:
  114. record = kmem_cache_alloc(jbd2_revoke_record_cache, GFP_NOFS);
  115. if (!record)
  116. goto oom;
  117. record->sequence = seq;
  118. record->blocknr = blocknr;
  119. hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
  120. spin_lock(&journal->j_revoke_lock);
  121. list_add(&record->hash, hash_list);
  122. spin_unlock(&journal->j_revoke_lock);
  123. return 0;
  124. oom:
  125. if (!journal_oom_retry)
  126. return -ENOMEM;
  127. jbd_debug(1, "ENOMEM in %s, retrying\n", __FUNCTION__);
  128. yield();
  129. goto repeat;
  130. }
  131. /* Find a revoke record in the journal's hash table. */
  132. static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal,
  133. unsigned long long blocknr)
  134. {
  135. struct list_head *hash_list;
  136. struct jbd2_revoke_record_s *record;
  137. hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
  138. spin_lock(&journal->j_revoke_lock);
  139. record = (struct jbd2_revoke_record_s *) hash_list->next;
  140. while (&(record->hash) != hash_list) {
  141. if (record->blocknr == blocknr) {
  142. spin_unlock(&journal->j_revoke_lock);
  143. return record;
  144. }
  145. record = (struct jbd2_revoke_record_s *) record->hash.next;
  146. }
  147. spin_unlock(&journal->j_revoke_lock);
  148. return NULL;
  149. }
  150. int __init jbd2_journal_init_revoke_caches(void)
  151. {
  152. jbd2_revoke_record_cache = kmem_cache_create("jbd2_revoke_record",
  153. sizeof(struct jbd2_revoke_record_s),
  154. 0, SLAB_HWCACHE_ALIGN, NULL, NULL);
  155. if (jbd2_revoke_record_cache == 0)
  156. return -ENOMEM;
  157. jbd2_revoke_table_cache = kmem_cache_create("jbd2_revoke_table",
  158. sizeof(struct jbd2_revoke_table_s),
  159. 0, 0, NULL, NULL);
  160. if (jbd2_revoke_table_cache == 0) {
  161. kmem_cache_destroy(jbd2_revoke_record_cache);
  162. jbd2_revoke_record_cache = NULL;
  163. return -ENOMEM;
  164. }
  165. return 0;
  166. }
  167. void jbd2_journal_destroy_revoke_caches(void)
  168. {
  169. kmem_cache_destroy(jbd2_revoke_record_cache);
  170. jbd2_revoke_record_cache = NULL;
  171. kmem_cache_destroy(jbd2_revoke_table_cache);
  172. jbd2_revoke_table_cache = NULL;
  173. }
  174. /* Initialise the revoke table for a given journal to a given size. */
  175. int jbd2_journal_init_revoke(journal_t *journal, int hash_size)
  176. {
  177. int shift, tmp;
  178. J_ASSERT (journal->j_revoke_table[0] == NULL);
  179. shift = 0;
  180. tmp = hash_size;
  181. while((tmp >>= 1UL) != 0UL)
  182. shift++;
  183. journal->j_revoke_table[0] = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
  184. if (!journal->j_revoke_table[0])
  185. return -ENOMEM;
  186. journal->j_revoke = journal->j_revoke_table[0];
  187. /* Check that the hash_size is a power of two */
  188. J_ASSERT ((hash_size & (hash_size-1)) == 0);
  189. journal->j_revoke->hash_size = hash_size;
  190. journal->j_revoke->hash_shift = shift;
  191. journal->j_revoke->hash_table =
  192. kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
  193. if (!journal->j_revoke->hash_table) {
  194. kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[0]);
  195. journal->j_revoke = NULL;
  196. return -ENOMEM;
  197. }
  198. for (tmp = 0; tmp < hash_size; tmp++)
  199. INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]);
  200. journal->j_revoke_table[1] = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
  201. if (!journal->j_revoke_table[1]) {
  202. kfree(journal->j_revoke_table[0]->hash_table);
  203. kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[0]);
  204. return -ENOMEM;
  205. }
  206. journal->j_revoke = journal->j_revoke_table[1];
  207. /* Check that the hash_size is a power of two */
  208. J_ASSERT ((hash_size & (hash_size-1)) == 0);
  209. journal->j_revoke->hash_size = hash_size;
  210. journal->j_revoke->hash_shift = shift;
  211. journal->j_revoke->hash_table =
  212. kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
  213. if (!journal->j_revoke->hash_table) {
  214. kfree(journal->j_revoke_table[0]->hash_table);
  215. kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[0]);
  216. kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[1]);
  217. journal->j_revoke = NULL;
  218. return -ENOMEM;
  219. }
  220. for (tmp = 0; tmp < hash_size; tmp++)
  221. INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]);
  222. spin_lock_init(&journal->j_revoke_lock);
  223. return 0;
  224. }
  225. /* Destoy a journal's revoke table. The table must already be empty! */
  226. void jbd2_journal_destroy_revoke(journal_t *journal)
  227. {
  228. struct jbd2_revoke_table_s *table;
  229. struct list_head *hash_list;
  230. int i;
  231. table = journal->j_revoke_table[0];
  232. if (!table)
  233. return;
  234. for (i=0; i<table->hash_size; i++) {
  235. hash_list = &table->hash_table[i];
  236. J_ASSERT (list_empty(hash_list));
  237. }
  238. kfree(table->hash_table);
  239. kmem_cache_free(jbd2_revoke_table_cache, table);
  240. journal->j_revoke = NULL;
  241. table = journal->j_revoke_table[1];
  242. if (!table)
  243. return;
  244. for (i=0; i<table->hash_size; i++) {
  245. hash_list = &table->hash_table[i];
  246. J_ASSERT (list_empty(hash_list));
  247. }
  248. kfree(table->hash_table);
  249. kmem_cache_free(jbd2_revoke_table_cache, table);
  250. journal->j_revoke = NULL;
  251. }
  252. #ifdef __KERNEL__
  253. /*
  254. * jbd2_journal_revoke: revoke a given buffer_head from the journal. This
  255. * prevents the block from being replayed during recovery if we take a
  256. * crash after this current transaction commits. Any subsequent
  257. * metadata writes of the buffer in this transaction cancel the
  258. * revoke.
  259. *
  260. * Note that this call may block --- it is up to the caller to make
  261. * sure that there are no further calls to journal_write_metadata
  262. * before the revoke is complete. In ext3, this implies calling the
  263. * revoke before clearing the block bitmap when we are deleting
  264. * metadata.
  265. *
  266. * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a
  267. * parameter, but does _not_ forget the buffer_head if the bh was only
  268. * found implicitly.
  269. *
  270. * bh_in may not be a journalled buffer - it may have come off
  271. * the hash tables without an attached journal_head.
  272. *
  273. * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count
  274. * by one.
  275. */
  276. int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr,
  277. struct buffer_head *bh_in)
  278. {
  279. struct buffer_head *bh = NULL;
  280. journal_t *journal;
  281. struct block_device *bdev;
  282. int err;
  283. might_sleep();
  284. if (bh_in)
  285. BUFFER_TRACE(bh_in, "enter");
  286. journal = handle->h_transaction->t_journal;
  287. if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){
  288. J_ASSERT (!"Cannot set revoke feature!");
  289. return -EINVAL;
  290. }
  291. bdev = journal->j_fs_dev;
  292. bh = bh_in;
  293. if (!bh) {
  294. bh = __find_get_block(bdev, blocknr, journal->j_blocksize);
  295. if (bh)
  296. BUFFER_TRACE(bh, "found on hash");
  297. }
  298. #ifdef JBD_EXPENSIVE_CHECKING
  299. else {
  300. struct buffer_head *bh2;
  301. /* If there is a different buffer_head lying around in
  302. * memory anywhere... */
  303. bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize);
  304. if (bh2) {
  305. /* ... and it has RevokeValid status... */
  306. if (bh2 != bh && buffer_revokevalid(bh2))
  307. /* ...then it better be revoked too,
  308. * since it's illegal to create a revoke
  309. * record against a buffer_head which is
  310. * not marked revoked --- that would
  311. * risk missing a subsequent revoke
  312. * cancel. */
  313. J_ASSERT_BH(bh2, buffer_revoked(bh2));
  314. put_bh(bh2);
  315. }
  316. }
  317. #endif
  318. /* We really ought not ever to revoke twice in a row without
  319. first having the revoke cancelled: it's illegal to free a
  320. block twice without allocating it in between! */
  321. if (bh) {
  322. if (!J_EXPECT_BH(bh, !buffer_revoked(bh),
  323. "inconsistent data on disk")) {
  324. if (!bh_in)
  325. brelse(bh);
  326. return -EIO;
  327. }
  328. set_buffer_revoked(bh);
  329. set_buffer_revokevalid(bh);
  330. if (bh_in) {
  331. BUFFER_TRACE(bh_in, "call jbd2_journal_forget");
  332. jbd2_journal_forget(handle, bh_in);
  333. } else {
  334. BUFFER_TRACE(bh, "call brelse");
  335. __brelse(bh);
  336. }
  337. }
  338. jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in);
  339. err = insert_revoke_hash(journal, blocknr,
  340. handle->h_transaction->t_tid);
  341. BUFFER_TRACE(bh_in, "exit");
  342. return err;
  343. }
  344. /*
  345. * Cancel an outstanding revoke. For use only internally by the
  346. * journaling code (called from jbd2_journal_get_write_access).
  347. *
  348. * We trust buffer_revoked() on the buffer if the buffer is already
  349. * being journaled: if there is no revoke pending on the buffer, then we
  350. * don't do anything here.
  351. *
  352. * This would break if it were possible for a buffer to be revoked and
  353. * discarded, and then reallocated within the same transaction. In such
  354. * a case we would have lost the revoked bit, but when we arrived here
  355. * the second time we would still have a pending revoke to cancel. So,
  356. * do not trust the Revoked bit on buffers unless RevokeValid is also
  357. * set.
  358. *
  359. * The caller must have the journal locked.
  360. */
  361. int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
  362. {
  363. struct jbd2_revoke_record_s *record;
  364. journal_t *journal = handle->h_transaction->t_journal;
  365. int need_cancel;
  366. int did_revoke = 0; /* akpm: debug */
  367. struct buffer_head *bh = jh2bh(jh);
  368. jbd_debug(4, "journal_head %p, cancelling revoke\n", jh);
  369. /* Is the existing Revoke bit valid? If so, we trust it, and
  370. * only perform the full cancel if the revoke bit is set. If
  371. * not, we can't trust the revoke bit, and we need to do the
  372. * full search for a revoke record. */
  373. if (test_set_buffer_revokevalid(bh)) {
  374. need_cancel = test_clear_buffer_revoked(bh);
  375. } else {
  376. need_cancel = 1;
  377. clear_buffer_revoked(bh);
  378. }
  379. if (need_cancel) {
  380. record = find_revoke_record(journal, bh->b_blocknr);
  381. if (record) {
  382. jbd_debug(4, "cancelled existing revoke on "
  383. "blocknr %llu\n", (unsigned long long)bh->b_blocknr);
  384. spin_lock(&journal->j_revoke_lock);
  385. list_del(&record->hash);
  386. spin_unlock(&journal->j_revoke_lock);
  387. kmem_cache_free(jbd2_revoke_record_cache, record);
  388. did_revoke = 1;
  389. }
  390. }
  391. #ifdef JBD_EXPENSIVE_CHECKING
  392. /* There better not be one left behind by now! */
  393. record = find_revoke_record(journal, bh->b_blocknr);
  394. J_ASSERT_JH(jh, record == NULL);
  395. #endif
  396. /* Finally, have we just cleared revoke on an unhashed
  397. * buffer_head? If so, we'd better make sure we clear the
  398. * revoked status on any hashed alias too, otherwise the revoke
  399. * state machine will get very upset later on. */
  400. if (need_cancel) {
  401. struct buffer_head *bh2;
  402. bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size);
  403. if (bh2) {
  404. if (bh2 != bh)
  405. clear_buffer_revoked(bh2);
  406. __brelse(bh2);
  407. }
  408. }
  409. return did_revoke;
  410. }
  411. /* journal_switch_revoke table select j_revoke for next transaction
  412. * we do not want to suspend any processing until all revokes are
  413. * written -bzzz
  414. */
  415. void jbd2_journal_switch_revoke_table(journal_t *journal)
  416. {
  417. int i;
  418. if (journal->j_revoke == journal->j_revoke_table[0])
  419. journal->j_revoke = journal->j_revoke_table[1];
  420. else
  421. journal->j_revoke = journal->j_revoke_table[0];
  422. for (i = 0; i < journal->j_revoke->hash_size; i++)
  423. INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]);
  424. }
  425. /*
  426. * Write revoke records to the journal for all entries in the current
  427. * revoke hash, deleting the entries as we go.
  428. *
  429. * Called with the journal lock held.
  430. */
  431. void jbd2_journal_write_revoke_records(journal_t *journal,
  432. transaction_t *transaction)
  433. {
  434. struct journal_head *descriptor;
  435. struct jbd2_revoke_record_s *record;
  436. struct jbd2_revoke_table_s *revoke;
  437. struct list_head *hash_list;
  438. int i, offset, count;
  439. descriptor = NULL;
  440. offset = 0;
  441. count = 0;
  442. /* select revoke table for committing transaction */
  443. revoke = journal->j_revoke == journal->j_revoke_table[0] ?
  444. journal->j_revoke_table[1] : journal->j_revoke_table[0];
  445. for (i = 0; i < revoke->hash_size; i++) {
  446. hash_list = &revoke->hash_table[i];
  447. while (!list_empty(hash_list)) {
  448. record = (struct jbd2_revoke_record_s *)
  449. hash_list->next;
  450. write_one_revoke_record(journal, transaction,
  451. &descriptor, &offset,
  452. record);
  453. count++;
  454. list_del(&record->hash);
  455. kmem_cache_free(jbd2_revoke_record_cache, record);
  456. }
  457. }
  458. if (descriptor)
  459. flush_descriptor(journal, descriptor, offset);
  460. jbd_debug(1, "Wrote %d revoke records\n", count);
  461. }
  462. /*
  463. * Write out one revoke record. We need to create a new descriptor
  464. * block if the old one is full or if we have not already created one.
  465. */
  466. static void write_one_revoke_record(journal_t *journal,
  467. transaction_t *transaction,
  468. struct journal_head **descriptorp,
  469. int *offsetp,
  470. struct jbd2_revoke_record_s *record)
  471. {
  472. struct journal_head *descriptor;
  473. int offset;
  474. journal_header_t *header;
  475. /* If we are already aborting, this all becomes a noop. We
  476. still need to go round the loop in
  477. jbd2_journal_write_revoke_records in order to free all of the
  478. revoke records: only the IO to the journal is omitted. */
  479. if (is_journal_aborted(journal))
  480. return;
  481. descriptor = *descriptorp;
  482. offset = *offsetp;
  483. /* Make sure we have a descriptor with space left for the record */
  484. if (descriptor) {
  485. if (offset == journal->j_blocksize) {
  486. flush_descriptor(journal, descriptor, offset);
  487. descriptor = NULL;
  488. }
  489. }
  490. if (!descriptor) {
  491. descriptor = jbd2_journal_get_descriptor_buffer(journal);
  492. if (!descriptor)
  493. return;
  494. header = (journal_header_t *) &jh2bh(descriptor)->b_data[0];
  495. header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
  496. header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK);
  497. header->h_sequence = cpu_to_be32(transaction->t_tid);
  498. /* Record it so that we can wait for IO completion later */
  499. JBUFFER_TRACE(descriptor, "file as BJ_LogCtl");
  500. jbd2_journal_file_buffer(descriptor, transaction, BJ_LogCtl);
  501. offset = sizeof(jbd2_journal_revoke_header_t);
  502. *descriptorp = descriptor;
  503. }
  504. if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) {
  505. * ((__be64 *)(&jh2bh(descriptor)->b_data[offset])) =
  506. cpu_to_be64(record->blocknr);
  507. offset += 8;
  508. } else {
  509. * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) =
  510. cpu_to_be32(record->blocknr);
  511. offset += 4;
  512. }
  513. *offsetp = offset;
  514. }
  515. /*
  516. * Flush a revoke descriptor out to the journal. If we are aborting,
  517. * this is a noop; otherwise we are generating a buffer which needs to
  518. * be waited for during commit, so it has to go onto the appropriate
  519. * journal buffer list.
  520. */
  521. static void flush_descriptor(journal_t *journal,
  522. struct journal_head *descriptor,
  523. int offset)
  524. {
  525. jbd2_journal_revoke_header_t *header;
  526. struct buffer_head *bh = jh2bh(descriptor);
  527. if (is_journal_aborted(journal)) {
  528. put_bh(bh);
  529. return;
  530. }
  531. header = (jbd2_journal_revoke_header_t *) jh2bh(descriptor)->b_data;
  532. header->r_count = cpu_to_be32(offset);
  533. set_buffer_jwrite(bh);
  534. BUFFER_TRACE(bh, "write");
  535. set_buffer_dirty(bh);
  536. ll_rw_block(SWRITE, 1, &bh);
  537. }
  538. #endif
  539. /*
  540. * Revoke support for recovery.
  541. *
  542. * Recovery needs to be able to:
  543. *
  544. * record all revoke records, including the tid of the latest instance
  545. * of each revoke in the journal
  546. *
  547. * check whether a given block in a given transaction should be replayed
  548. * (ie. has not been revoked by a revoke record in that or a subsequent
  549. * transaction)
  550. *
  551. * empty the revoke table after recovery.
  552. */
  553. /*
  554. * First, setting revoke records. We create a new revoke record for
  555. * every block ever revoked in the log as we scan it for recovery, and
  556. * we update the existing records if we find multiple revokes for a
  557. * single block.
  558. */
  559. int jbd2_journal_set_revoke(journal_t *journal,
  560. unsigned long long blocknr,
  561. tid_t sequence)
  562. {
  563. struct jbd2_revoke_record_s *record;
  564. record = find_revoke_record(journal, blocknr);
  565. if (record) {
  566. /* If we have multiple occurrences, only record the
  567. * latest sequence number in the hashed record */
  568. if (tid_gt(sequence, record->sequence))
  569. record->sequence = sequence;
  570. return 0;
  571. }
  572. return insert_revoke_hash(journal, blocknr, sequence);
  573. }
  574. /*
  575. * Test revoke records. For a given block referenced in the log, has
  576. * that block been revoked? A revoke record with a given transaction
  577. * sequence number revokes all blocks in that transaction and earlier
  578. * ones, but later transactions still need replayed.
  579. */
  580. int jbd2_journal_test_revoke(journal_t *journal,
  581. unsigned long long blocknr,
  582. tid_t sequence)
  583. {
  584. struct jbd2_revoke_record_s *record;
  585. record = find_revoke_record(journal, blocknr);
  586. if (!record)
  587. return 0;
  588. if (tid_gt(sequence, record->sequence))
  589. return 0;
  590. return 1;
  591. }
  592. /*
  593. * Finally, once recovery is over, we need to clear the revoke table so
  594. * that it can be reused by the running filesystem.
  595. */
  596. void jbd2_journal_clear_revoke(journal_t *journal)
  597. {
  598. int i;
  599. struct list_head *hash_list;
  600. struct jbd2_revoke_record_s *record;
  601. struct jbd2_revoke_table_s *revoke;
  602. revoke = journal->j_revoke;
  603. for (i = 0; i < revoke->hash_size; i++) {
  604. hash_list = &revoke->hash_table[i];
  605. while (!list_empty(hash_list)) {
  606. record = (struct jbd2_revoke_record_s*) hash_list->next;
  607. list_del(&record->hash);
  608. kmem_cache_free(jbd2_revoke_record_cache, record);
  609. }
  610. }
  611. }