Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
58862699 | 2 | * linux/fs/jbd/revoke.c |
ae6ddcc5 | 3 | * |
1da177e4 LT |
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: | |
ae6ddcc5 | 18 | * |
1da177e4 LT |
19 | * + Commit: during commit we write the entire list of the current |
20 | * transaction's revoked blocks to the journal | |
ae6ddcc5 | 21 | * |
1da177e4 LT |
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: | |
ae6ddcc5 | 32 | * The desired end result is the journaling of the new block, so we |
1da177e4 LT |
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 | * | |
ae6ddcc5 | 44 | * Block is revoked and then written as data: |
1da177e4 LT |
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 | * | |
8c111b3f YY |
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 | * | |
1da177e4 LT |
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: | |
ae6ddcc5 | 61 | * buffer has been revoked. |
32433879 JK |
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 | * | |
25985edc | 78 | * Finally, also replay code uses the hash tables but at this moment no one else |
32433879 JK |
79 | * can touch them (filesystem isn't mounted yet) and hence no locking is |
80 | * needed. | |
1da177e4 LT |
81 | */ |
82 | ||
83 | #ifndef __KERNEL__ | |
84 | #include "jfs_user.h" | |
85 | #else | |
86 | #include <linux/time.h> | |
87 | #include <linux/fs.h> | |
88 | #include <linux/jbd.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/slab.h> | |
91 | #include <linux/list.h> | |
1da177e4 | 92 | #include <linux/init.h> |
38d726d1 | 93 | #include <linux/bio.h> |
1da177e4 | 94 | #endif |
f482394c | 95 | #include <linux/log2.h> |
3c9cafe0 | 96 | #include <linux/hash.h> |
1da177e4 | 97 | |
e18b890b CL |
98 | static struct kmem_cache *revoke_record_cache; |
99 | static struct kmem_cache *revoke_table_cache; | |
1da177e4 LT |
100 | |
101 | /* Each revoke record represents one single revoked block. During | |
102 | journal replay, this involves recording the transaction ID of the | |
103 | last transaction to revoke this block. */ | |
104 | ||
ae6ddcc5 | 105 | struct jbd_revoke_record_s |
1da177e4 LT |
106 | { |
107 | struct list_head hash; | |
108 | tid_t sequence; /* Used for recovery only */ | |
9c28cbcc | 109 | unsigned int blocknr; |
1da177e4 LT |
110 | }; |
111 | ||
112 | ||
113 | /* The revoke table is just a simple hash table of revoke records. */ | |
114 | struct jbd_revoke_table_s | |
115 | { | |
116 | /* It is conceivable that we might want a larger hash table | |
117 | * for recovery. Must be a power of two. */ | |
ae6ddcc5 MC |
118 | int hash_size; |
119 | int hash_shift; | |
1da177e4 LT |
120 | struct list_head *hash_table; |
121 | }; | |
122 | ||
123 | ||
124 | #ifdef __KERNEL__ | |
125 | static void write_one_revoke_record(journal_t *, transaction_t *, | |
126 | struct journal_head **, int *, | |
38d726d1 TT |
127 | struct jbd_revoke_record_s *, int); |
128 | static void flush_descriptor(journal_t *, struct journal_head *, int, int); | |
1da177e4 LT |
129 | #endif |
130 | ||
131 | /* Utility functions to maintain the revoke table */ | |
132 | ||
9c28cbcc | 133 | static inline int hash(journal_t *journal, unsigned int block) |
1da177e4 LT |
134 | { |
135 | struct jbd_revoke_table_s *table = journal->j_revoke; | |
1da177e4 | 136 | |
3c9cafe0 | 137 | return hash_32(block, table->hash_shift); |
1da177e4 LT |
138 | } |
139 | ||
9c28cbcc | 140 | static int insert_revoke_hash(journal_t *journal, unsigned int blocknr, |
022a4a7b | 141 | tid_t seq) |
1da177e4 LT |
142 | { |
143 | struct list_head *hash_list; | |
144 | struct jbd_revoke_record_s *record; | |
145 | ||
146 | repeat: | |
147 | record = kmem_cache_alloc(revoke_record_cache, GFP_NOFS); | |
148 | if (!record) | |
149 | goto oom; | |
150 | ||
151 | record->sequence = seq; | |
152 | record->blocknr = blocknr; | |
153 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
154 | spin_lock(&journal->j_revoke_lock); | |
155 | list_add(&record->hash, hash_list); | |
156 | spin_unlock(&journal->j_revoke_lock); | |
157 | return 0; | |
158 | ||
159 | oom: | |
160 | if (!journal_oom_retry) | |
161 | return -ENOMEM; | |
08fc99bf | 162 | jbd_debug(1, "ENOMEM in %s, retrying\n", __func__); |
1da177e4 LT |
163 | yield(); |
164 | goto repeat; | |
165 | } | |
166 | ||
167 | /* Find a revoke record in the journal's hash table. */ | |
168 | ||
169 | static struct jbd_revoke_record_s *find_revoke_record(journal_t *journal, | |
9c28cbcc | 170 | unsigned int blocknr) |
1da177e4 LT |
171 | { |
172 | struct list_head *hash_list; | |
173 | struct jbd_revoke_record_s *record; | |
174 | ||
175 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
176 | ||
177 | spin_lock(&journal->j_revoke_lock); | |
178 | record = (struct jbd_revoke_record_s *) hash_list->next; | |
179 | while (&(record->hash) != hash_list) { | |
180 | if (record->blocknr == blocknr) { | |
181 | spin_unlock(&journal->j_revoke_lock); | |
182 | return record; | |
183 | } | |
184 | record = (struct jbd_revoke_record_s *) record->hash.next; | |
185 | } | |
186 | spin_unlock(&journal->j_revoke_lock); | |
187 | return NULL; | |
188 | } | |
189 | ||
1984bb76 DG |
190 | void journal_destroy_revoke_caches(void) |
191 | { | |
192 | if (revoke_record_cache) { | |
193 | kmem_cache_destroy(revoke_record_cache); | |
194 | revoke_record_cache = NULL; | |
195 | } | |
196 | if (revoke_table_cache) { | |
197 | kmem_cache_destroy(revoke_table_cache); | |
198 | revoke_table_cache = NULL; | |
199 | } | |
200 | } | |
201 | ||
1da177e4 LT |
202 | int __init journal_init_revoke_caches(void) |
203 | { | |
1984bb76 DG |
204 | J_ASSERT(!revoke_record_cache); |
205 | J_ASSERT(!revoke_table_cache); | |
206 | ||
1da177e4 LT |
207 | revoke_record_cache = kmem_cache_create("revoke_record", |
208 | sizeof(struct jbd_revoke_record_s), | |
e12ba74d MG |
209 | 0, |
210 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY, | |
211 | NULL); | |
1076d17a | 212 | if (!revoke_record_cache) |
1984bb76 | 213 | goto record_cache_failure; |
1da177e4 LT |
214 | |
215 | revoke_table_cache = kmem_cache_create("revoke_table", | |
216 | sizeof(struct jbd_revoke_table_s), | |
e12ba74d | 217 | 0, SLAB_TEMPORARY, NULL); |
1984bb76 DG |
218 | if (!revoke_table_cache) |
219 | goto table_cache_failure; | |
220 | ||
1da177e4 | 221 | return 0; |
1da177e4 | 222 | |
1984bb76 DG |
223 | table_cache_failure: |
224 | journal_destroy_revoke_caches(); | |
225 | record_cache_failure: | |
226 | return -ENOMEM; | |
1da177e4 LT |
227 | } |
228 | ||
f4d79ca2 | 229 | static struct jbd_revoke_table_s *journal_init_revoke_table(int hash_size) |
1da177e4 | 230 | { |
770901cc | 231 | int i; |
f4d79ca2 | 232 | struct jbd_revoke_table_s *table; |
1da177e4 | 233 | |
f4d79ca2 DG |
234 | table = kmem_cache_alloc(revoke_table_cache, GFP_KERNEL); |
235 | if (!table) | |
236 | goto out; | |
1da177e4 | 237 | |
f4d79ca2 | 238 | table->hash_size = hash_size; |
770901cc | 239 | table->hash_shift = ilog2(hash_size); |
f4d79ca2 | 240 | table->hash_table = |
1da177e4 | 241 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); |
f4d79ca2 DG |
242 | if (!table->hash_table) { |
243 | kmem_cache_free(revoke_table_cache, table); | |
244 | table = NULL; | |
245 | goto out; | |
1da177e4 LT |
246 | } |
247 | ||
770901cc FF |
248 | for (i = 0; i < hash_size; i++) |
249 | INIT_LIST_HEAD(&table->hash_table[i]); | |
1da177e4 | 250 | |
f4d79ca2 DG |
251 | out: |
252 | return table; | |
253 | } | |
254 | ||
255 | static void journal_destroy_revoke_table(struct jbd_revoke_table_s *table) | |
256 | { | |
257 | int i; | |
258 | struct list_head *hash_list; | |
259 | ||
260 | for (i = 0; i < table->hash_size; i++) { | |
261 | hash_list = &table->hash_table[i]; | |
262 | J_ASSERT(list_empty(hash_list)); | |
1da177e4 LT |
263 | } |
264 | ||
f4d79ca2 DG |
265 | kfree(table->hash_table); |
266 | kmem_cache_free(revoke_table_cache, table); | |
267 | } | |
1da177e4 | 268 | |
f4d79ca2 DG |
269 | /* Initialise the revoke table for a given journal to a given size. */ |
270 | int journal_init_revoke(journal_t *journal, int hash_size) | |
271 | { | |
272 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
f482394c | 273 | J_ASSERT(is_power_of_2(hash_size)); |
1da177e4 | 274 | |
f4d79ca2 DG |
275 | journal->j_revoke_table[0] = journal_init_revoke_table(hash_size); |
276 | if (!journal->j_revoke_table[0]) | |
277 | goto fail0; | |
1da177e4 | 278 | |
f4d79ca2 DG |
279 | journal->j_revoke_table[1] = journal_init_revoke_table(hash_size); |
280 | if (!journal->j_revoke_table[1]) | |
281 | goto fail1; | |
1da177e4 | 282 | |
f4d79ca2 | 283 | journal->j_revoke = journal->j_revoke_table[1]; |
1da177e4 LT |
284 | |
285 | spin_lock_init(&journal->j_revoke_lock); | |
286 | ||
287 | return 0; | |
1da177e4 | 288 | |
f4d79ca2 DG |
289 | fail1: |
290 | journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
291 | fail0: | |
292 | return -ENOMEM; | |
293 | } | |
1da177e4 | 294 | |
f4d79ca2 | 295 | /* Destroy a journal's revoke table. The table must already be empty! */ |
1da177e4 LT |
296 | void journal_destroy_revoke(journal_t *journal) |
297 | { | |
1da177e4 | 298 | journal->j_revoke = NULL; |
f4d79ca2 DG |
299 | if (journal->j_revoke_table[0]) |
300 | journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
301 | if (journal->j_revoke_table[1]) | |
302 | journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
1da177e4 LT |
303 | } |
304 | ||
305 | ||
306 | #ifdef __KERNEL__ | |
307 | ||
ae6ddcc5 | 308 | /* |
1da177e4 LT |
309 | * journal_revoke: revoke a given buffer_head from the journal. This |
310 | * prevents the block from being replayed during recovery if we take a | |
311 | * crash after this current transaction commits. Any subsequent | |
312 | * metadata writes of the buffer in this transaction cancel the | |
ae6ddcc5 | 313 | * revoke. |
1da177e4 LT |
314 | * |
315 | * Note that this call may block --- it is up to the caller to make | |
316 | * sure that there are no further calls to journal_write_metadata | |
317 | * before the revoke is complete. In ext3, this implies calling the | |
318 | * revoke before clearing the block bitmap when we are deleting | |
ae6ddcc5 | 319 | * metadata. |
1da177e4 LT |
320 | * |
321 | * Revoke performs a journal_forget on any buffer_head passed in as a | |
322 | * parameter, but does _not_ forget the buffer_head if the bh was only | |
ae6ddcc5 | 323 | * found implicitly. |
1da177e4 LT |
324 | * |
325 | * bh_in may not be a journalled buffer - it may have come off | |
326 | * the hash tables without an attached journal_head. | |
327 | * | |
328 | * If bh_in is non-zero, journal_revoke() will decrement its b_count | |
329 | * by one. | |
330 | */ | |
331 | ||
9c28cbcc | 332 | int journal_revoke(handle_t *handle, unsigned int blocknr, |
1da177e4 LT |
333 | struct buffer_head *bh_in) |
334 | { | |
335 | struct buffer_head *bh = NULL; | |
336 | journal_t *journal; | |
337 | struct block_device *bdev; | |
338 | int err; | |
339 | ||
340 | might_sleep(); | |
341 | if (bh_in) | |
342 | BUFFER_TRACE(bh_in, "enter"); | |
343 | ||
344 | journal = handle->h_transaction->t_journal; | |
345 | if (!journal_set_features(journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)){ | |
346 | J_ASSERT (!"Cannot set revoke feature!"); | |
347 | return -EINVAL; | |
348 | } | |
349 | ||
350 | bdev = journal->j_fs_dev; | |
351 | bh = bh_in; | |
352 | ||
353 | if (!bh) { | |
354 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
355 | if (bh) | |
356 | BUFFER_TRACE(bh, "found on hash"); | |
357 | } | |
358 | #ifdef JBD_EXPENSIVE_CHECKING | |
359 | else { | |
360 | struct buffer_head *bh2; | |
361 | ||
362 | /* If there is a different buffer_head lying around in | |
363 | * memory anywhere... */ | |
364 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
365 | if (bh2) { | |
366 | /* ... and it has RevokeValid status... */ | |
367 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
368 | /* ...then it better be revoked too, | |
369 | * since it's illegal to create a revoke | |
370 | * record against a buffer_head which is | |
371 | * not marked revoked --- that would | |
372 | * risk missing a subsequent revoke | |
373 | * cancel. */ | |
374 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
375 | put_bh(bh2); | |
376 | } | |
377 | } | |
378 | #endif | |
379 | ||
380 | /* We really ought not ever to revoke twice in a row without | |
381 | first having the revoke cancelled: it's illegal to free a | |
382 | block twice without allocating it in between! */ | |
383 | if (bh) { | |
384 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
385 | "inconsistent data on disk")) { | |
386 | if (!bh_in) | |
387 | brelse(bh); | |
388 | return -EIO; | |
389 | } | |
390 | set_buffer_revoked(bh); | |
391 | set_buffer_revokevalid(bh); | |
392 | if (bh_in) { | |
393 | BUFFER_TRACE(bh_in, "call journal_forget"); | |
394 | journal_forget(handle, bh_in); | |
395 | } else { | |
396 | BUFFER_TRACE(bh, "call brelse"); | |
397 | __brelse(bh); | |
398 | } | |
399 | } | |
400 | ||
9c28cbcc | 401 | jbd_debug(2, "insert revoke for block %u, bh_in=%p\n", blocknr, bh_in); |
1da177e4 LT |
402 | err = insert_revoke_hash(journal, blocknr, |
403 | handle->h_transaction->t_tid); | |
404 | BUFFER_TRACE(bh_in, "exit"); | |
405 | return err; | |
406 | } | |
407 | ||
408 | /* | |
409 | * Cancel an outstanding revoke. For use only internally by the | |
410 | * journaling code (called from journal_get_write_access). | |
411 | * | |
412 | * We trust buffer_revoked() on the buffer if the buffer is already | |
413 | * being journaled: if there is no revoke pending on the buffer, then we | |
414 | * don't do anything here. | |
415 | * | |
416 | * This would break if it were possible for a buffer to be revoked and | |
417 | * discarded, and then reallocated within the same transaction. In such | |
418 | * a case we would have lost the revoked bit, but when we arrived here | |
419 | * the second time we would still have a pending revoke to cancel. So, | |
420 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
421 | * set. | |
1da177e4 LT |
422 | */ |
423 | int journal_cancel_revoke(handle_t *handle, struct journal_head *jh) | |
424 | { | |
425 | struct jbd_revoke_record_s *record; | |
426 | journal_t *journal = handle->h_transaction->t_journal; | |
427 | int need_cancel; | |
428 | int did_revoke = 0; /* akpm: debug */ | |
429 | struct buffer_head *bh = jh2bh(jh); | |
430 | ||
431 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
432 | ||
433 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
434 | * only perform the full cancel if the revoke bit is set. If | |
435 | * not, we can't trust the revoke bit, and we need to do the | |
436 | * full search for a revoke record. */ | |
437 | if (test_set_buffer_revokevalid(bh)) { | |
438 | need_cancel = test_clear_buffer_revoked(bh); | |
439 | } else { | |
440 | need_cancel = 1; | |
441 | clear_buffer_revoked(bh); | |
442 | } | |
443 | ||
444 | if (need_cancel) { | |
445 | record = find_revoke_record(journal, bh->b_blocknr); | |
446 | if (record) { | |
447 | jbd_debug(4, "cancelled existing revoke on " | |
448 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
449 | spin_lock(&journal->j_revoke_lock); | |
450 | list_del(&record->hash); | |
451 | spin_unlock(&journal->j_revoke_lock); | |
452 | kmem_cache_free(revoke_record_cache, record); | |
453 | did_revoke = 1; | |
454 | } | |
455 | } | |
456 | ||
457 | #ifdef JBD_EXPENSIVE_CHECKING | |
458 | /* There better not be one left behind by now! */ | |
459 | record = find_revoke_record(journal, bh->b_blocknr); | |
460 | J_ASSERT_JH(jh, record == NULL); | |
461 | #endif | |
462 | ||
463 | /* Finally, have we just cleared revoke on an unhashed | |
464 | * buffer_head? If so, we'd better make sure we clear the | |
465 | * revoked status on any hashed alias too, otherwise the revoke | |
466 | * state machine will get very upset later on. */ | |
467 | if (need_cancel) { | |
468 | struct buffer_head *bh2; | |
469 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
470 | if (bh2) { | |
471 | if (bh2 != bh) | |
472 | clear_buffer_revoked(bh2); | |
473 | __brelse(bh2); | |
474 | } | |
475 | } | |
476 | return did_revoke; | |
477 | } | |
478 | ||
8c111b3f YY |
479 | /* |
480 | * journal_clear_revoked_flags clears revoked flag of buffers in | |
481 | * revoke table to reflect there is no revoked buffer in the next | |
482 | * transaction which is going to be started. | |
483 | */ | |
484 | void journal_clear_buffer_revoked_flags(journal_t *journal) | |
485 | { | |
486 | struct jbd_revoke_table_s *revoke = journal->j_revoke; | |
487 | int i = 0; | |
488 | ||
489 | for (i = 0; i < revoke->hash_size; i++) { | |
490 | struct list_head *hash_list; | |
491 | struct list_head *list_entry; | |
492 | hash_list = &revoke->hash_table[i]; | |
493 | ||
494 | list_for_each(list_entry, hash_list) { | |
495 | struct jbd_revoke_record_s *record; | |
496 | struct buffer_head *bh; | |
497 | record = (struct jbd_revoke_record_s *)list_entry; | |
498 | bh = __find_get_block(journal->j_fs_dev, | |
499 | record->blocknr, | |
500 | journal->j_blocksize); | |
501 | if (bh) { | |
502 | clear_buffer_revoked(bh); | |
503 | __brelse(bh); | |
504 | } | |
505 | } | |
506 | } | |
507 | } | |
508 | ||
1da177e4 LT |
509 | /* journal_switch_revoke table select j_revoke for next transaction |
510 | * we do not want to suspend any processing until all revokes are | |
511 | * written -bzzz | |
512 | */ | |
513 | void journal_switch_revoke_table(journal_t *journal) | |
514 | { | |
515 | int i; | |
516 | ||
517 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
518 | journal->j_revoke = journal->j_revoke_table[1]; | |
519 | else | |
520 | journal->j_revoke = journal->j_revoke_table[0]; | |
521 | ||
ae6ddcc5 | 522 | for (i = 0; i < journal->j_revoke->hash_size; i++) |
1da177e4 LT |
523 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); |
524 | } | |
525 | ||
526 | /* | |
527 | * Write revoke records to the journal for all entries in the current | |
528 | * revoke hash, deleting the entries as we go. | |
1da177e4 | 529 | */ |
ae6ddcc5 | 530 | void journal_write_revoke_records(journal_t *journal, |
38d726d1 | 531 | transaction_t *transaction, int write_op) |
1da177e4 LT |
532 | { |
533 | struct journal_head *descriptor; | |
534 | struct jbd_revoke_record_s *record; | |
535 | struct jbd_revoke_table_s *revoke; | |
536 | struct list_head *hash_list; | |
537 | int i, offset, count; | |
538 | ||
ae6ddcc5 | 539 | descriptor = NULL; |
1da177e4 LT |
540 | offset = 0; |
541 | count = 0; | |
542 | ||
543 | /* select revoke table for committing transaction */ | |
544 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
545 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
546 | ||
547 | for (i = 0; i < revoke->hash_size; i++) { | |
548 | hash_list = &revoke->hash_table[i]; | |
549 | ||
550 | while (!list_empty(hash_list)) { | |
ae6ddcc5 | 551 | record = (struct jbd_revoke_record_s *) |
1da177e4 LT |
552 | hash_list->next; |
553 | write_one_revoke_record(journal, transaction, | |
ae6ddcc5 | 554 | &descriptor, &offset, |
38d726d1 | 555 | record, write_op); |
1da177e4 LT |
556 | count++; |
557 | list_del(&record->hash); | |
558 | kmem_cache_free(revoke_record_cache, record); | |
559 | } | |
560 | } | |
561 | if (descriptor) | |
38d726d1 | 562 | flush_descriptor(journal, descriptor, offset, write_op); |
1da177e4 LT |
563 | jbd_debug(1, "Wrote %d revoke records\n", count); |
564 | } | |
565 | ||
ae6ddcc5 | 566 | /* |
1da177e4 | 567 | * Write out one revoke record. We need to create a new descriptor |
ae6ddcc5 | 568 | * block if the old one is full or if we have not already created one. |
1da177e4 LT |
569 | */ |
570 | ||
ae6ddcc5 | 571 | static void write_one_revoke_record(journal_t *journal, |
1da177e4 | 572 | transaction_t *transaction, |
ae6ddcc5 | 573 | struct journal_head **descriptorp, |
1da177e4 | 574 | int *offsetp, |
38d726d1 TT |
575 | struct jbd_revoke_record_s *record, |
576 | int write_op) | |
1da177e4 LT |
577 | { |
578 | struct journal_head *descriptor; | |
579 | int offset; | |
580 | journal_header_t *header; | |
581 | ||
582 | /* If we are already aborting, this all becomes a noop. We | |
583 | still need to go round the loop in | |
584 | journal_write_revoke_records in order to free all of the | |
585 | revoke records: only the IO to the journal is omitted. */ | |
586 | if (is_journal_aborted(journal)) | |
587 | return; | |
588 | ||
589 | descriptor = *descriptorp; | |
590 | offset = *offsetp; | |
591 | ||
592 | /* Make sure we have a descriptor with space left for the record */ | |
593 | if (descriptor) { | |
594 | if (offset == journal->j_blocksize) { | |
38d726d1 | 595 | flush_descriptor(journal, descriptor, offset, write_op); |
1da177e4 LT |
596 | descriptor = NULL; |
597 | } | |
598 | } | |
599 | ||
600 | if (!descriptor) { | |
601 | descriptor = journal_get_descriptor_buffer(journal); | |
602 | if (!descriptor) | |
603 | return; | |
604 | header = (journal_header_t *) &jh2bh(descriptor)->b_data[0]; | |
605 | header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER); | |
606 | header->h_blocktype = cpu_to_be32(JFS_REVOKE_BLOCK); | |
607 | header->h_sequence = cpu_to_be32(transaction->t_tid); | |
608 | ||
609 | /* Record it so that we can wait for IO completion later */ | |
610 | JBUFFER_TRACE(descriptor, "file as BJ_LogCtl"); | |
611 | journal_file_buffer(descriptor, transaction, BJ_LogCtl); | |
612 | ||
613 | offset = sizeof(journal_revoke_header_t); | |
614 | *descriptorp = descriptor; | |
615 | } | |
616 | ||
ae6ddcc5 | 617 | * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) = |
1da177e4 LT |
618 | cpu_to_be32(record->blocknr); |
619 | offset += 4; | |
620 | *offsetp = offset; | |
621 | } | |
622 | ||
ae6ddcc5 | 623 | /* |
1da177e4 LT |
624 | * Flush a revoke descriptor out to the journal. If we are aborting, |
625 | * this is a noop; otherwise we are generating a buffer which needs to | |
626 | * be waited for during commit, so it has to go onto the appropriate | |
627 | * journal buffer list. | |
628 | */ | |
629 | ||
ae6ddcc5 MC |
630 | static void flush_descriptor(journal_t *journal, |
631 | struct journal_head *descriptor, | |
38d726d1 | 632 | int offset, int write_op) |
1da177e4 LT |
633 | { |
634 | journal_revoke_header_t *header; | |
635 | struct buffer_head *bh = jh2bh(descriptor); | |
636 | ||
637 | if (is_journal_aborted(journal)) { | |
638 | put_bh(bh); | |
639 | return; | |
640 | } | |
641 | ||
642 | header = (journal_revoke_header_t *) jh2bh(descriptor)->b_data; | |
643 | header->r_count = cpu_to_be32(offset); | |
644 | set_buffer_jwrite(bh); | |
645 | BUFFER_TRACE(bh, "write"); | |
646 | set_buffer_dirty(bh); | |
9cb569d6 | 647 | write_dirty_buffer(bh, write_op); |
1da177e4 LT |
648 | } |
649 | #endif | |
650 | ||
ae6ddcc5 | 651 | /* |
1da177e4 LT |
652 | * Revoke support for recovery. |
653 | * | |
654 | * Recovery needs to be able to: | |
655 | * | |
656 | * record all revoke records, including the tid of the latest instance | |
657 | * of each revoke in the journal | |
658 | * | |
659 | * check whether a given block in a given transaction should be replayed | |
660 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
661 | * transaction) | |
ae6ddcc5 | 662 | * |
1da177e4 LT |
663 | * empty the revoke table after recovery. |
664 | */ | |
665 | ||
666 | /* | |
667 | * First, setting revoke records. We create a new revoke record for | |
668 | * every block ever revoked in the log as we scan it for recovery, and | |
669 | * we update the existing records if we find multiple revokes for a | |
ae6ddcc5 | 670 | * single block. |
1da177e4 LT |
671 | */ |
672 | ||
ae6ddcc5 | 673 | int journal_set_revoke(journal_t *journal, |
9c28cbcc | 674 | unsigned int blocknr, |
1da177e4 LT |
675 | tid_t sequence) |
676 | { | |
677 | struct jbd_revoke_record_s *record; | |
678 | ||
679 | record = find_revoke_record(journal, blocknr); | |
680 | if (record) { | |
681 | /* If we have multiple occurrences, only record the | |
682 | * latest sequence number in the hashed record */ | |
683 | if (tid_gt(sequence, record->sequence)) | |
684 | record->sequence = sequence; | |
685 | return 0; | |
ae6ddcc5 | 686 | } |
1da177e4 LT |
687 | return insert_revoke_hash(journal, blocknr, sequence); |
688 | } | |
689 | ||
ae6ddcc5 | 690 | /* |
1da177e4 LT |
691 | * Test revoke records. For a given block referenced in the log, has |
692 | * that block been revoked? A revoke record with a given transaction | |
693 | * sequence number revokes all blocks in that transaction and earlier | |
694 | * ones, but later transactions still need replayed. | |
695 | */ | |
696 | ||
ae6ddcc5 | 697 | int journal_test_revoke(journal_t *journal, |
9c28cbcc | 698 | unsigned int blocknr, |
1da177e4 LT |
699 | tid_t sequence) |
700 | { | |
701 | struct jbd_revoke_record_s *record; | |
702 | ||
703 | record = find_revoke_record(journal, blocknr); | |
704 | if (!record) | |
705 | return 0; | |
706 | if (tid_gt(sequence, record->sequence)) | |
707 | return 0; | |
708 | return 1; | |
709 | } | |
710 | ||
711 | /* | |
712 | * Finally, once recovery is over, we need to clear the revoke table so | |
713 | * that it can be reused by the running filesystem. | |
714 | */ | |
715 | ||
716 | void journal_clear_revoke(journal_t *journal) | |
717 | { | |
718 | int i; | |
719 | struct list_head *hash_list; | |
720 | struct jbd_revoke_record_s *record; | |
721 | struct jbd_revoke_table_s *revoke; | |
722 | ||
723 | revoke = journal->j_revoke; | |
724 | ||
725 | for (i = 0; i < revoke->hash_size; i++) { | |
726 | hash_list = &revoke->hash_table[i]; | |
727 | while (!list_empty(hash_list)) { | |
728 | record = (struct jbd_revoke_record_s*) hash_list->next; | |
729 | list_del(&record->hash); | |
730 | kmem_cache_free(revoke_record_cache, record); | |
731 | } | |
732 | } | |
733 | } |