revoke.c 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759
  1. /*
  2. * linux/fs/jbd2/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. * We cache revoke status of a buffer in the current transaction in b_states
  51. * bits. As the name says, revokevalid flag indicates that the cached revoke
  52. * status of a buffer is valid and we can rely on the cached status.
  53. *
  54. * Revoke information on buffers is a tri-state value:
  55. *
  56. * RevokeValid clear: no cached revoke status, need to look it up
  57. * RevokeValid set, Revoked clear:
  58. * buffer has not been revoked, and cancel_revoke
  59. * need do nothing.
  60. * RevokeValid set, Revoked set:
  61. * buffer has been revoked.
  62. *
  63. * Locking rules:
  64. * We keep two hash tables of revoke records. One hashtable belongs to the
  65. * running transaction (is pointed to by journal->j_revoke), the other one
  66. * belongs to the committing transaction. Accesses to the second hash table
  67. * happen only from the kjournald and no other thread touches this table. Also
  68. * journal_switch_revoke_table() which switches which hashtable belongs to the
  69. * running and which to the committing transaction is called only from
  70. * kjournald. Therefore we need no locks when accessing the hashtable belonging
  71. * to the committing transaction.
  72. *
  73. * All users operating on the hash table belonging to the running transaction
  74. * have a handle to the transaction. Therefore they are safe from kjournald
  75. * switching hash tables under them. For operations on the lists of entries in
  76. * the hash table j_revoke_lock is used.
  77. *
  78. * Finally, also replay code uses the hash tables but at this moment no one else
  79. * can touch them (filesystem isn't mounted yet) and hence no locking is
  80. * needed.
  81. */
  82. #ifndef __KERNEL__
  83. #include "jfs_user.h"
  84. #else
  85. #include <linux/time.h>
  86. #include <linux/fs.h>
  87. #include <linux/jbd2.h>
  88. #include <linux/errno.h>
  89. #include <linux/slab.h>
  90. #include <linux/list.h>
  91. #include <linux/init.h>
  92. #include <linux/bio.h>
  93. #include <linux/log2.h>
  94. #include <linux/hash.h>
  95. #endif
  96. static struct kmem_cache *jbd2_revoke_record_cache;
  97. static struct kmem_cache *jbd2_revoke_table_cache;
  98. /* Each revoke record represents one single revoked block. During
  99. journal replay, this involves recording the transaction ID of the
  100. last transaction to revoke this block. */
  101. struct jbd2_revoke_record_s
  102. {
  103. struct list_head hash;
  104. tid_t sequence; /* Used for recovery only */
  105. unsigned long long blocknr;
  106. };
  107. /* The revoke table is just a simple hash table of revoke records. */
  108. struct jbd2_revoke_table_s
  109. {
  110. /* It is conceivable that we might want a larger hash table
  111. * for recovery. Must be a power of two. */
  112. int hash_size;
  113. int hash_shift;
  114. struct list_head *hash_table;
  115. };
  116. #ifdef __KERNEL__
  117. static void write_one_revoke_record(journal_t *, transaction_t *,
  118. struct list_head *,
  119. struct buffer_head **, int *,
  120. struct jbd2_revoke_record_s *, int);
  121. static void flush_descriptor(journal_t *, struct buffer_head *, int, int);
  122. #endif
  123. /* Utility functions to maintain the revoke table */
  124. static inline int hash(journal_t *journal, unsigned long long block)
  125. {
  126. return hash_64(block, journal->j_revoke->hash_shift);
  127. }
  128. static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr,
  129. tid_t seq)
  130. {
  131. struct list_head *hash_list;
  132. struct jbd2_revoke_record_s *record;
  133. gfp_t gfp_mask = GFP_NOFS;
  134. if (journal_oom_retry)
  135. gfp_mask |= __GFP_NOFAIL;
  136. record = kmem_cache_alloc(jbd2_revoke_record_cache, gfp_mask);
  137. if (!record)
  138. return -ENOMEM;
  139. record->sequence = seq;
  140. record->blocknr = blocknr;
  141. hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
  142. spin_lock(&journal->j_revoke_lock);
  143. list_add(&record->hash, hash_list);
  144. spin_unlock(&journal->j_revoke_lock);
  145. return 0;
  146. }
  147. /* Find a revoke record in the journal's hash table. */
  148. static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal,
  149. unsigned long long blocknr)
  150. {
  151. struct list_head *hash_list;
  152. struct jbd2_revoke_record_s *record;
  153. hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
  154. spin_lock(&journal->j_revoke_lock);
  155. record = (struct jbd2_revoke_record_s *) hash_list->next;
  156. while (&(record->hash) != hash_list) {
  157. if (record->blocknr == blocknr) {
  158. spin_unlock(&journal->j_revoke_lock);
  159. return record;
  160. }
  161. record = (struct jbd2_revoke_record_s *) record->hash.next;
  162. }
  163. spin_unlock(&journal->j_revoke_lock);
  164. return NULL;
  165. }
  166. void jbd2_journal_destroy_revoke_caches(void)
  167. {
  168. if (jbd2_revoke_record_cache) {
  169. kmem_cache_destroy(jbd2_revoke_record_cache);
  170. jbd2_revoke_record_cache = NULL;
  171. }
  172. if (jbd2_revoke_table_cache) {
  173. kmem_cache_destroy(jbd2_revoke_table_cache);
  174. jbd2_revoke_table_cache = NULL;
  175. }
  176. }
  177. int __init jbd2_journal_init_revoke_caches(void)
  178. {
  179. J_ASSERT(!jbd2_revoke_record_cache);
  180. J_ASSERT(!jbd2_revoke_table_cache);
  181. jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s,
  182. SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY);
  183. if (!jbd2_revoke_record_cache)
  184. goto record_cache_failure;
  185. jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s,
  186. SLAB_TEMPORARY);
  187. if (!jbd2_revoke_table_cache)
  188. goto table_cache_failure;
  189. return 0;
  190. table_cache_failure:
  191. jbd2_journal_destroy_revoke_caches();
  192. record_cache_failure:
  193. return -ENOMEM;
  194. }
  195. static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size)
  196. {
  197. int shift = 0;
  198. int tmp = hash_size;
  199. struct jbd2_revoke_table_s *table;
  200. table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
  201. if (!table)
  202. goto out;
  203. while((tmp >>= 1UL) != 0UL)
  204. shift++;
  205. table->hash_size = hash_size;
  206. table->hash_shift = shift;
  207. table->hash_table =
  208. kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
  209. if (!table->hash_table) {
  210. kmem_cache_free(jbd2_revoke_table_cache, table);
  211. table = NULL;
  212. goto out;
  213. }
  214. for (tmp = 0; tmp < hash_size; tmp++)
  215. INIT_LIST_HEAD(&table->hash_table[tmp]);
  216. out:
  217. return table;
  218. }
  219. static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table)
  220. {
  221. int i;
  222. struct list_head *hash_list;
  223. for (i = 0; i < table->hash_size; i++) {
  224. hash_list = &table->hash_table[i];
  225. J_ASSERT(list_empty(hash_list));
  226. }
  227. kfree(table->hash_table);
  228. kmem_cache_free(jbd2_revoke_table_cache, table);
  229. }
  230. /* Initialise the revoke table for a given journal to a given size. */
  231. int jbd2_journal_init_revoke(journal_t *journal, int hash_size)
  232. {
  233. J_ASSERT(journal->j_revoke_table[0] == NULL);
  234. J_ASSERT(is_power_of_2(hash_size));
  235. journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size);
  236. if (!journal->j_revoke_table[0])
  237. goto fail0;
  238. journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size);
  239. if (!journal->j_revoke_table[1])
  240. goto fail1;
  241. journal->j_revoke = journal->j_revoke_table[1];
  242. spin_lock_init(&journal->j_revoke_lock);
  243. return 0;
  244. fail1:
  245. jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
  246. fail0:
  247. return -ENOMEM;
  248. }
  249. /* Destroy a journal's revoke table. The table must already be empty! */
  250. void jbd2_journal_destroy_revoke(journal_t *journal)
  251. {
  252. journal->j_revoke = NULL;
  253. if (journal->j_revoke_table[0])
  254. jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
  255. if (journal->j_revoke_table[1])
  256. jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]);
  257. }
  258. #ifdef __KERNEL__
  259. /*
  260. * jbd2_journal_revoke: revoke a given buffer_head from the journal. This
  261. * prevents the block from being replayed during recovery if we take a
  262. * crash after this current transaction commits. Any subsequent
  263. * metadata writes of the buffer in this transaction cancel the
  264. * revoke.
  265. *
  266. * Note that this call may block --- it is up to the caller to make
  267. * sure that there are no further calls to journal_write_metadata
  268. * before the revoke is complete. In ext3, this implies calling the
  269. * revoke before clearing the block bitmap when we are deleting
  270. * metadata.
  271. *
  272. * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a
  273. * parameter, but does _not_ forget the buffer_head if the bh was only
  274. * found implicitly.
  275. *
  276. * bh_in may not be a journalled buffer - it may have come off
  277. * the hash tables without an attached journal_head.
  278. *
  279. * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count
  280. * by one.
  281. */
  282. int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr,
  283. struct buffer_head *bh_in)
  284. {
  285. struct buffer_head *bh = NULL;
  286. journal_t *journal;
  287. struct block_device *bdev;
  288. int err;
  289. might_sleep();
  290. if (bh_in)
  291. BUFFER_TRACE(bh_in, "enter");
  292. journal = handle->h_transaction->t_journal;
  293. if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){
  294. J_ASSERT (!"Cannot set revoke feature!");
  295. return -EINVAL;
  296. }
  297. bdev = journal->j_fs_dev;
  298. bh = bh_in;
  299. if (!bh) {
  300. bh = __find_get_block(bdev, blocknr, journal->j_blocksize);
  301. if (bh)
  302. BUFFER_TRACE(bh, "found on hash");
  303. }
  304. #ifdef JBD2_EXPENSIVE_CHECKING
  305. else {
  306. struct buffer_head *bh2;
  307. /* If there is a different buffer_head lying around in
  308. * memory anywhere... */
  309. bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize);
  310. if (bh2) {
  311. /* ... and it has RevokeValid status... */
  312. if (bh2 != bh && buffer_revokevalid(bh2))
  313. /* ...then it better be revoked too,
  314. * since it's illegal to create a revoke
  315. * record against a buffer_head which is
  316. * not marked revoked --- that would
  317. * risk missing a subsequent revoke
  318. * cancel. */
  319. J_ASSERT_BH(bh2, buffer_revoked(bh2));
  320. put_bh(bh2);
  321. }
  322. }
  323. #endif
  324. /* We really ought not ever to revoke twice in a row without
  325. first having the revoke cancelled: it's illegal to free a
  326. block twice without allocating it in between! */
  327. if (bh) {
  328. if (!J_EXPECT_BH(bh, !buffer_revoked(bh),
  329. "inconsistent data on disk")) {
  330. if (!bh_in)
  331. brelse(bh);
  332. return -EIO;
  333. }
  334. set_buffer_revoked(bh);
  335. set_buffer_revokevalid(bh);
  336. if (bh_in) {
  337. BUFFER_TRACE(bh_in, "call jbd2_journal_forget");
  338. jbd2_journal_forget(handle, bh_in);
  339. } else {
  340. BUFFER_TRACE(bh, "call brelse");
  341. __brelse(bh);
  342. }
  343. }
  344. jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in);
  345. err = insert_revoke_hash(journal, blocknr,
  346. handle->h_transaction->t_tid);
  347. BUFFER_TRACE(bh_in, "exit");
  348. return err;
  349. }
  350. /*
  351. * Cancel an outstanding revoke. For use only internally by the
  352. * journaling code (called from jbd2_journal_get_write_access).
  353. *
  354. * We trust buffer_revoked() on the buffer if the buffer is already
  355. * being journaled: if there is no revoke pending on the buffer, then we
  356. * don't do anything here.
  357. *
  358. * This would break if it were possible for a buffer to be revoked and
  359. * discarded, and then reallocated within the same transaction. In such
  360. * a case we would have lost the revoked bit, but when we arrived here
  361. * the second time we would still have a pending revoke to cancel. So,
  362. * do not trust the Revoked bit on buffers unless RevokeValid is also
  363. * set.
  364. */
  365. int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
  366. {
  367. struct jbd2_revoke_record_s *record;
  368. journal_t *journal = handle->h_transaction->t_journal;
  369. int need_cancel;
  370. int did_revoke = 0; /* akpm: debug */
  371. struct buffer_head *bh = jh2bh(jh);
  372. jbd_debug(4, "journal_head %p, cancelling revoke\n", jh);
  373. /* Is the existing Revoke bit valid? If so, we trust it, and
  374. * only perform the full cancel if the revoke bit is set. If
  375. * not, we can't trust the revoke bit, and we need to do the
  376. * full search for a revoke record. */
  377. if (test_set_buffer_revokevalid(bh)) {
  378. need_cancel = test_clear_buffer_revoked(bh);
  379. } else {
  380. need_cancel = 1;
  381. clear_buffer_revoked(bh);
  382. }
  383. if (need_cancel) {
  384. record = find_revoke_record(journal, bh->b_blocknr);
  385. if (record) {
  386. jbd_debug(4, "cancelled existing revoke on "
  387. "blocknr %llu\n", (unsigned long long)bh->b_blocknr);
  388. spin_lock(&journal->j_revoke_lock);
  389. list_del(&record->hash);
  390. spin_unlock(&journal->j_revoke_lock);
  391. kmem_cache_free(jbd2_revoke_record_cache, record);
  392. did_revoke = 1;
  393. }
  394. }
  395. #ifdef JBD2_EXPENSIVE_CHECKING
  396. /* There better not be one left behind by now! */
  397. record = find_revoke_record(journal, bh->b_blocknr);
  398. J_ASSERT_JH(jh, record == NULL);
  399. #endif
  400. /* Finally, have we just cleared revoke on an unhashed
  401. * buffer_head? If so, we'd better make sure we clear the
  402. * revoked status on any hashed alias too, otherwise the revoke
  403. * state machine will get very upset later on. */
  404. if (need_cancel) {
  405. struct buffer_head *bh2;
  406. bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size);
  407. if (bh2) {
  408. if (bh2 != bh)
  409. clear_buffer_revoked(bh2);
  410. __brelse(bh2);
  411. }
  412. }
  413. return did_revoke;
  414. }
  415. /*
  416. * journal_clear_revoked_flag clears revoked flag of buffers in
  417. * revoke table to reflect there is no revoked buffers in the next
  418. * transaction which is going to be started.
  419. */
  420. void jbd2_clear_buffer_revoked_flags(journal_t *journal)
  421. {
  422. struct jbd2_revoke_table_s *revoke = journal->j_revoke;
  423. int i = 0;
  424. for (i = 0; i < revoke->hash_size; i++) {
  425. struct list_head *hash_list;
  426. struct list_head *list_entry;
  427. hash_list = &revoke->hash_table[i];
  428. list_for_each(list_entry, hash_list) {
  429. struct jbd2_revoke_record_s *record;
  430. struct buffer_head *bh;
  431. record = (struct jbd2_revoke_record_s *)list_entry;
  432. bh = __find_get_block(journal->j_fs_dev,
  433. record->blocknr,
  434. journal->j_blocksize);
  435. if (bh) {
  436. clear_buffer_revoked(bh);
  437. __brelse(bh);
  438. }
  439. }
  440. }
  441. }
  442. /* journal_switch_revoke table select j_revoke for next transaction
  443. * we do not want to suspend any processing until all revokes are
  444. * written -bzzz
  445. */
  446. void jbd2_journal_switch_revoke_table(journal_t *journal)
  447. {
  448. int i;
  449. if (journal->j_revoke == journal->j_revoke_table[0])
  450. journal->j_revoke = journal->j_revoke_table[1];
  451. else
  452. journal->j_revoke = journal->j_revoke_table[0];
  453. for (i = 0; i < journal->j_revoke->hash_size; i++)
  454. INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]);
  455. }
  456. /*
  457. * Write revoke records to the journal for all entries in the current
  458. * revoke hash, deleting the entries as we go.
  459. */
  460. void jbd2_journal_write_revoke_records(journal_t *journal,
  461. transaction_t *transaction,
  462. struct list_head *log_bufs,
  463. int write_op)
  464. {
  465. struct buffer_head *descriptor;
  466. struct jbd2_revoke_record_s *record;
  467. struct jbd2_revoke_table_s *revoke;
  468. struct list_head *hash_list;
  469. int i, offset, count;
  470. descriptor = NULL;
  471. offset = 0;
  472. count = 0;
  473. /* select revoke table for committing transaction */
  474. revoke = journal->j_revoke == journal->j_revoke_table[0] ?
  475. journal->j_revoke_table[1] : journal->j_revoke_table[0];
  476. for (i = 0; i < revoke->hash_size; i++) {
  477. hash_list = &revoke->hash_table[i];
  478. while (!list_empty(hash_list)) {
  479. record = (struct jbd2_revoke_record_s *)
  480. hash_list->next;
  481. write_one_revoke_record(journal, transaction, log_bufs,
  482. &descriptor, &offset,
  483. record, write_op);
  484. count++;
  485. list_del(&record->hash);
  486. kmem_cache_free(jbd2_revoke_record_cache, record);
  487. }
  488. }
  489. if (descriptor)
  490. flush_descriptor(journal, descriptor, offset, write_op);
  491. jbd_debug(1, "Wrote %d revoke records\n", count);
  492. }
  493. /*
  494. * Write out one revoke record. We need to create a new descriptor
  495. * block if the old one is full or if we have not already created one.
  496. */
  497. static void write_one_revoke_record(journal_t *journal,
  498. transaction_t *transaction,
  499. struct list_head *log_bufs,
  500. struct buffer_head **descriptorp,
  501. int *offsetp,
  502. struct jbd2_revoke_record_s *record,
  503. int write_op)
  504. {
  505. int csum_size = 0;
  506. struct buffer_head *descriptor;
  507. int sz, offset;
  508. journal_header_t *header;
  509. /* If we are already aborting, this all becomes a noop. We
  510. still need to go round the loop in
  511. jbd2_journal_write_revoke_records in order to free all of the
  512. revoke records: only the IO to the journal is omitted. */
  513. if (is_journal_aborted(journal))
  514. return;
  515. descriptor = *descriptorp;
  516. offset = *offsetp;
  517. /* Do we need to leave space at the end for a checksum? */
  518. if (jbd2_journal_has_csum_v2or3(journal))
  519. csum_size = sizeof(struct jbd2_journal_revoke_tail);
  520. if (jbd2_has_feature_64bit(journal))
  521. sz = 8;
  522. else
  523. sz = 4;
  524. /* Make sure we have a descriptor with space left for the record */
  525. if (descriptor) {
  526. if (offset + sz > journal->j_blocksize - csum_size) {
  527. flush_descriptor(journal, descriptor, offset, write_op);
  528. descriptor = NULL;
  529. }
  530. }
  531. if (!descriptor) {
  532. descriptor = jbd2_journal_get_descriptor_buffer(journal);
  533. if (!descriptor)
  534. return;
  535. header = (journal_header_t *)descriptor->b_data;
  536. header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
  537. header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK);
  538. header->h_sequence = cpu_to_be32(transaction->t_tid);
  539. /* Record it so that we can wait for IO completion later */
  540. BUFFER_TRACE(descriptor, "file in log_bufs");
  541. jbd2_file_log_bh(log_bufs, descriptor);
  542. offset = sizeof(jbd2_journal_revoke_header_t);
  543. *descriptorp = descriptor;
  544. }
  545. if (jbd2_has_feature_64bit(journal))
  546. * ((__be64 *)(&descriptor->b_data[offset])) =
  547. cpu_to_be64(record->blocknr);
  548. else
  549. * ((__be32 *)(&descriptor->b_data[offset])) =
  550. cpu_to_be32(record->blocknr);
  551. offset += sz;
  552. *offsetp = offset;
  553. }
  554. static void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh)
  555. {
  556. struct jbd2_journal_revoke_tail *tail;
  557. __u32 csum;
  558. if (!jbd2_journal_has_csum_v2or3(j))
  559. return;
  560. tail = (struct jbd2_journal_revoke_tail *)(bh->b_data + j->j_blocksize -
  561. sizeof(struct jbd2_journal_revoke_tail));
  562. tail->r_checksum = 0;
  563. csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
  564. tail->r_checksum = cpu_to_be32(csum);
  565. }
  566. /*
  567. * Flush a revoke descriptor out to the journal. If we are aborting,
  568. * this is a noop; otherwise we are generating a buffer which needs to
  569. * be waited for during commit, so it has to go onto the appropriate
  570. * journal buffer list.
  571. */
  572. static void flush_descriptor(journal_t *journal,
  573. struct buffer_head *descriptor,
  574. int offset, int write_op)
  575. {
  576. jbd2_journal_revoke_header_t *header;
  577. if (is_journal_aborted(journal)) {
  578. put_bh(descriptor);
  579. return;
  580. }
  581. header = (jbd2_journal_revoke_header_t *)descriptor->b_data;
  582. header->r_count = cpu_to_be32(offset);
  583. jbd2_revoke_csum_set(journal, descriptor);
  584. set_buffer_jwrite(descriptor);
  585. BUFFER_TRACE(descriptor, "write");
  586. set_buffer_dirty(descriptor);
  587. write_dirty_buffer(descriptor, write_op);
  588. }
  589. #endif
  590. /*
  591. * Revoke support for recovery.
  592. *
  593. * Recovery needs to be able to:
  594. *
  595. * record all revoke records, including the tid of the latest instance
  596. * of each revoke in the journal
  597. *
  598. * check whether a given block in a given transaction should be replayed
  599. * (ie. has not been revoked by a revoke record in that or a subsequent
  600. * transaction)
  601. *
  602. * empty the revoke table after recovery.
  603. */
  604. /*
  605. * First, setting revoke records. We create a new revoke record for
  606. * every block ever revoked in the log as we scan it for recovery, and
  607. * we update the existing records if we find multiple revokes for a
  608. * single block.
  609. */
  610. int jbd2_journal_set_revoke(journal_t *journal,
  611. unsigned long long blocknr,
  612. tid_t sequence)
  613. {
  614. struct jbd2_revoke_record_s *record;
  615. record = find_revoke_record(journal, blocknr);
  616. if (record) {
  617. /* If we have multiple occurrences, only record the
  618. * latest sequence number in the hashed record */
  619. if (tid_gt(sequence, record->sequence))
  620. record->sequence = sequence;
  621. return 0;
  622. }
  623. return insert_revoke_hash(journal, blocknr, sequence);
  624. }
  625. /*
  626. * Test revoke records. For a given block referenced in the log, has
  627. * that block been revoked? A revoke record with a given transaction
  628. * sequence number revokes all blocks in that transaction and earlier
  629. * ones, but later transactions still need replayed.
  630. */
  631. int jbd2_journal_test_revoke(journal_t *journal,
  632. unsigned long long blocknr,
  633. tid_t sequence)
  634. {
  635. struct jbd2_revoke_record_s *record;
  636. record = find_revoke_record(journal, blocknr);
  637. if (!record)
  638. return 0;
  639. if (tid_gt(sequence, record->sequence))
  640. return 0;
  641. return 1;
  642. }
  643. /*
  644. * Finally, once recovery is over, we need to clear the revoke table so
  645. * that it can be reused by the running filesystem.
  646. */
  647. void jbd2_journal_clear_revoke(journal_t *journal)
  648. {
  649. int i;
  650. struct list_head *hash_list;
  651. struct jbd2_revoke_record_s *record;
  652. struct jbd2_revoke_table_s *revoke;
  653. revoke = journal->j_revoke;
  654. for (i = 0; i < revoke->hash_size; i++) {
  655. hash_list = &revoke->hash_table[i];
  656. while (!list_empty(hash_list)) {
  657. record = (struct jbd2_revoke_record_s*) hash_list->next;
  658. list_del(&record->hash);
  659. kmem_cache_free(jbd2_revoke_record_cache, record);
  660. }
  661. }
  662. }