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 | * | |
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: | |
ae6ddcc5 | 57 | * buffer has been revoked. |
1da177e4 LT |
58 | */ |
59 | ||
60 | #ifndef __KERNEL__ | |
61 | #include "jfs_user.h" | |
62 | #else | |
63 | #include <linux/time.h> | |
64 | #include <linux/fs.h> | |
65 | #include <linux/jbd.h> | |
66 | #include <linux/errno.h> | |
67 | #include <linux/slab.h> | |
68 | #include <linux/list.h> | |
1da177e4 LT |
69 | #include <linux/init.h> |
70 | #endif | |
f482394c | 71 | #include <linux/log2.h> |
1da177e4 | 72 | |
e18b890b CL |
73 | static struct kmem_cache *revoke_record_cache; |
74 | static struct kmem_cache *revoke_table_cache; | |
1da177e4 LT |
75 | |
76 | /* Each revoke record represents one single revoked block. During | |
77 | journal replay, this involves recording the transaction ID of the | |
78 | last transaction to revoke this block. */ | |
79 | ||
ae6ddcc5 | 80 | struct jbd_revoke_record_s |
1da177e4 LT |
81 | { |
82 | struct list_head hash; | |
83 | tid_t sequence; /* Used for recovery only */ | |
84 | unsigned long blocknr; | |
85 | }; | |
86 | ||
87 | ||
88 | /* The revoke table is just a simple hash table of revoke records. */ | |
89 | struct jbd_revoke_table_s | |
90 | { | |
91 | /* It is conceivable that we might want a larger hash table | |
92 | * for recovery. Must be a power of two. */ | |
ae6ddcc5 MC |
93 | int hash_size; |
94 | int hash_shift; | |
1da177e4 LT |
95 | struct list_head *hash_table; |
96 | }; | |
97 | ||
98 | ||
99 | #ifdef __KERNEL__ | |
100 | static void write_one_revoke_record(journal_t *, transaction_t *, | |
101 | struct journal_head **, int *, | |
102 | struct jbd_revoke_record_s *); | |
103 | static void flush_descriptor(journal_t *, struct journal_head *, int); | |
104 | #endif | |
105 | ||
106 | /* Utility functions to maintain the revoke table */ | |
107 | ||
108 | /* Borrowed from buffer.c: this is a tried and tested block hash function */ | |
109 | static inline int hash(journal_t *journal, unsigned long block) | |
110 | { | |
111 | struct jbd_revoke_table_s *table = journal->j_revoke; | |
112 | int hash_shift = table->hash_shift; | |
113 | ||
114 | return ((block << (hash_shift - 6)) ^ | |
115 | (block >> 13) ^ | |
116 | (block << (hash_shift - 12))) & (table->hash_size - 1); | |
117 | } | |
118 | ||
022a4a7b AB |
119 | static int insert_revoke_hash(journal_t *journal, unsigned long blocknr, |
120 | tid_t seq) | |
1da177e4 LT |
121 | { |
122 | struct list_head *hash_list; | |
123 | struct jbd_revoke_record_s *record; | |
124 | ||
125 | repeat: | |
126 | record = kmem_cache_alloc(revoke_record_cache, GFP_NOFS); | |
127 | if (!record) | |
128 | goto oom; | |
129 | ||
130 | record->sequence = seq; | |
131 | record->blocknr = blocknr; | |
132 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
133 | spin_lock(&journal->j_revoke_lock); | |
134 | list_add(&record->hash, hash_list); | |
135 | spin_unlock(&journal->j_revoke_lock); | |
136 | return 0; | |
137 | ||
138 | oom: | |
139 | if (!journal_oom_retry) | |
140 | return -ENOMEM; | |
141 | jbd_debug(1, "ENOMEM in %s, retrying\n", __FUNCTION__); | |
142 | yield(); | |
143 | goto repeat; | |
144 | } | |
145 | ||
146 | /* Find a revoke record in the journal's hash table. */ | |
147 | ||
148 | static struct jbd_revoke_record_s *find_revoke_record(journal_t *journal, | |
149 | unsigned long blocknr) | |
150 | { | |
151 | struct list_head *hash_list; | |
152 | struct jbd_revoke_record_s *record; | |
153 | ||
154 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
155 | ||
156 | spin_lock(&journal->j_revoke_lock); | |
157 | record = (struct jbd_revoke_record_s *) hash_list->next; | |
158 | while (&(record->hash) != hash_list) { | |
159 | if (record->blocknr == blocknr) { | |
160 | spin_unlock(&journal->j_revoke_lock); | |
161 | return record; | |
162 | } | |
163 | record = (struct jbd_revoke_record_s *) record->hash.next; | |
164 | } | |
165 | spin_unlock(&journal->j_revoke_lock); | |
166 | return NULL; | |
167 | } | |
168 | ||
169 | int __init journal_init_revoke_caches(void) | |
170 | { | |
171 | revoke_record_cache = kmem_cache_create("revoke_record", | |
172 | sizeof(struct jbd_revoke_record_s), | |
20c2df83 | 173 | 0, SLAB_HWCACHE_ALIGN, NULL); |
1da177e4 LT |
174 | if (revoke_record_cache == 0) |
175 | return -ENOMEM; | |
176 | ||
177 | revoke_table_cache = kmem_cache_create("revoke_table", | |
178 | sizeof(struct jbd_revoke_table_s), | |
20c2df83 | 179 | 0, 0, NULL); |
1da177e4 LT |
180 | if (revoke_table_cache == 0) { |
181 | kmem_cache_destroy(revoke_record_cache); | |
182 | revoke_record_cache = NULL; | |
183 | return -ENOMEM; | |
184 | } | |
185 | return 0; | |
186 | } | |
187 | ||
188 | void journal_destroy_revoke_caches(void) | |
189 | { | |
190 | kmem_cache_destroy(revoke_record_cache); | |
191 | revoke_record_cache = NULL; | |
192 | kmem_cache_destroy(revoke_table_cache); | |
193 | revoke_table_cache = NULL; | |
194 | } | |
195 | ||
196 | /* Initialise the revoke table for a given journal to a given size. */ | |
197 | ||
198 | int journal_init_revoke(journal_t *journal, int hash_size) | |
199 | { | |
200 | int shift, tmp; | |
201 | ||
202 | J_ASSERT (journal->j_revoke_table[0] == NULL); | |
203 | ||
204 | shift = 0; | |
205 | tmp = hash_size; | |
206 | while((tmp >>= 1UL) != 0UL) | |
207 | shift++; | |
208 | ||
209 | journal->j_revoke_table[0] = kmem_cache_alloc(revoke_table_cache, GFP_KERNEL); | |
210 | if (!journal->j_revoke_table[0]) | |
211 | return -ENOMEM; | |
212 | journal->j_revoke = journal->j_revoke_table[0]; | |
213 | ||
214 | /* Check that the hash_size is a power of two */ | |
f482394c | 215 | J_ASSERT(is_power_of_2(hash_size)); |
1da177e4 LT |
216 | |
217 | journal->j_revoke->hash_size = hash_size; | |
218 | ||
219 | journal->j_revoke->hash_shift = shift; | |
220 | ||
221 | journal->j_revoke->hash_table = | |
222 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); | |
223 | if (!journal->j_revoke->hash_table) { | |
224 | kmem_cache_free(revoke_table_cache, journal->j_revoke_table[0]); | |
225 | journal->j_revoke = NULL; | |
226 | return -ENOMEM; | |
227 | } | |
228 | ||
229 | for (tmp = 0; tmp < hash_size; tmp++) | |
230 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]); | |
231 | ||
232 | journal->j_revoke_table[1] = kmem_cache_alloc(revoke_table_cache, GFP_KERNEL); | |
233 | if (!journal->j_revoke_table[1]) { | |
234 | kfree(journal->j_revoke_table[0]->hash_table); | |
235 | kmem_cache_free(revoke_table_cache, journal->j_revoke_table[0]); | |
236 | return -ENOMEM; | |
237 | } | |
238 | ||
239 | journal->j_revoke = journal->j_revoke_table[1]; | |
240 | ||
241 | /* Check that the hash_size is a power of two */ | |
f482394c | 242 | J_ASSERT(is_power_of_2(hash_size)); |
1da177e4 LT |
243 | |
244 | journal->j_revoke->hash_size = hash_size; | |
245 | ||
246 | journal->j_revoke->hash_shift = shift; | |
247 | ||
248 | journal->j_revoke->hash_table = | |
249 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); | |
250 | if (!journal->j_revoke->hash_table) { | |
251 | kfree(journal->j_revoke_table[0]->hash_table); | |
252 | kmem_cache_free(revoke_table_cache, journal->j_revoke_table[0]); | |
253 | kmem_cache_free(revoke_table_cache, journal->j_revoke_table[1]); | |
254 | journal->j_revoke = NULL; | |
255 | return -ENOMEM; | |
256 | } | |
257 | ||
258 | for (tmp = 0; tmp < hash_size; tmp++) | |
259 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]); | |
260 | ||
261 | spin_lock_init(&journal->j_revoke_lock); | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
266 | /* Destoy a journal's revoke table. The table must already be empty! */ | |
267 | ||
268 | void journal_destroy_revoke(journal_t *journal) | |
269 | { | |
270 | struct jbd_revoke_table_s *table; | |
271 | struct list_head *hash_list; | |
272 | int i; | |
273 | ||
274 | table = journal->j_revoke_table[0]; | |
275 | if (!table) | |
276 | return; | |
277 | ||
278 | for (i=0; i<table->hash_size; i++) { | |
279 | hash_list = &table->hash_table[i]; | |
280 | J_ASSERT (list_empty(hash_list)); | |
281 | } | |
282 | ||
283 | kfree(table->hash_table); | |
284 | kmem_cache_free(revoke_table_cache, table); | |
285 | journal->j_revoke = NULL; | |
286 | ||
287 | table = journal->j_revoke_table[1]; | |
288 | if (!table) | |
289 | return; | |
290 | ||
291 | for (i=0; i<table->hash_size; i++) { | |
292 | hash_list = &table->hash_table[i]; | |
293 | J_ASSERT (list_empty(hash_list)); | |
294 | } | |
295 | ||
296 | kfree(table->hash_table); | |
297 | kmem_cache_free(revoke_table_cache, table); | |
298 | journal->j_revoke = NULL; | |
299 | } | |
300 | ||
301 | ||
302 | #ifdef __KERNEL__ | |
303 | ||
ae6ddcc5 | 304 | /* |
1da177e4 LT |
305 | * journal_revoke: revoke a given buffer_head from the journal. This |
306 | * prevents the block from being replayed during recovery if we take a | |
307 | * crash after this current transaction commits. Any subsequent | |
308 | * metadata writes of the buffer in this transaction cancel the | |
ae6ddcc5 | 309 | * revoke. |
1da177e4 LT |
310 | * |
311 | * Note that this call may block --- it is up to the caller to make | |
312 | * sure that there are no further calls to journal_write_metadata | |
313 | * before the revoke is complete. In ext3, this implies calling the | |
314 | * revoke before clearing the block bitmap when we are deleting | |
ae6ddcc5 | 315 | * metadata. |
1da177e4 LT |
316 | * |
317 | * Revoke performs a journal_forget on any buffer_head passed in as a | |
318 | * parameter, but does _not_ forget the buffer_head if the bh was only | |
ae6ddcc5 | 319 | * found implicitly. |
1da177e4 LT |
320 | * |
321 | * bh_in may not be a journalled buffer - it may have come off | |
322 | * the hash tables without an attached journal_head. | |
323 | * | |
324 | * If bh_in is non-zero, journal_revoke() will decrement its b_count | |
325 | * by one. | |
326 | */ | |
327 | ||
ae6ddcc5 | 328 | int journal_revoke(handle_t *handle, unsigned long blocknr, |
1da177e4 LT |
329 | struct buffer_head *bh_in) |
330 | { | |
331 | struct buffer_head *bh = NULL; | |
332 | journal_t *journal; | |
333 | struct block_device *bdev; | |
334 | int err; | |
335 | ||
336 | might_sleep(); | |
337 | if (bh_in) | |
338 | BUFFER_TRACE(bh_in, "enter"); | |
339 | ||
340 | journal = handle->h_transaction->t_journal; | |
341 | if (!journal_set_features(journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)){ | |
342 | J_ASSERT (!"Cannot set revoke feature!"); | |
343 | return -EINVAL; | |
344 | } | |
345 | ||
346 | bdev = journal->j_fs_dev; | |
347 | bh = bh_in; | |
348 | ||
349 | if (!bh) { | |
350 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
351 | if (bh) | |
352 | BUFFER_TRACE(bh, "found on hash"); | |
353 | } | |
354 | #ifdef JBD_EXPENSIVE_CHECKING | |
355 | else { | |
356 | struct buffer_head *bh2; | |
357 | ||
358 | /* If there is a different buffer_head lying around in | |
359 | * memory anywhere... */ | |
360 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
361 | if (bh2) { | |
362 | /* ... and it has RevokeValid status... */ | |
363 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
364 | /* ...then it better be revoked too, | |
365 | * since it's illegal to create a revoke | |
366 | * record against a buffer_head which is | |
367 | * not marked revoked --- that would | |
368 | * risk missing a subsequent revoke | |
369 | * cancel. */ | |
370 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
371 | put_bh(bh2); | |
372 | } | |
373 | } | |
374 | #endif | |
375 | ||
376 | /* We really ought not ever to revoke twice in a row without | |
377 | first having the revoke cancelled: it's illegal to free a | |
378 | block twice without allocating it in between! */ | |
379 | if (bh) { | |
380 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
381 | "inconsistent data on disk")) { | |
382 | if (!bh_in) | |
383 | brelse(bh); | |
384 | return -EIO; | |
385 | } | |
386 | set_buffer_revoked(bh); | |
387 | set_buffer_revokevalid(bh); | |
388 | if (bh_in) { | |
389 | BUFFER_TRACE(bh_in, "call journal_forget"); | |
390 | journal_forget(handle, bh_in); | |
391 | } else { | |
392 | BUFFER_TRACE(bh, "call brelse"); | |
393 | __brelse(bh); | |
394 | } | |
395 | } | |
396 | ||
397 | jbd_debug(2, "insert revoke for block %lu, bh_in=%p\n", blocknr, bh_in); | |
398 | err = insert_revoke_hash(journal, blocknr, | |
399 | handle->h_transaction->t_tid); | |
400 | BUFFER_TRACE(bh_in, "exit"); | |
401 | return err; | |
402 | } | |
403 | ||
404 | /* | |
405 | * Cancel an outstanding revoke. For use only internally by the | |
406 | * journaling code (called from journal_get_write_access). | |
407 | * | |
408 | * We trust buffer_revoked() on the buffer if the buffer is already | |
409 | * being journaled: if there is no revoke pending on the buffer, then we | |
410 | * don't do anything here. | |
411 | * | |
412 | * This would break if it were possible for a buffer to be revoked and | |
413 | * discarded, and then reallocated within the same transaction. In such | |
414 | * a case we would have lost the revoked bit, but when we arrived here | |
415 | * the second time we would still have a pending revoke to cancel. So, | |
416 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
417 | * set. | |
418 | * | |
419 | * The caller must have the journal locked. | |
420 | */ | |
421 | int journal_cancel_revoke(handle_t *handle, struct journal_head *jh) | |
422 | { | |
423 | struct jbd_revoke_record_s *record; | |
424 | journal_t *journal = handle->h_transaction->t_journal; | |
425 | int need_cancel; | |
426 | int did_revoke = 0; /* akpm: debug */ | |
427 | struct buffer_head *bh = jh2bh(jh); | |
428 | ||
429 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
430 | ||
431 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
432 | * only perform the full cancel if the revoke bit is set. If | |
433 | * not, we can't trust the revoke bit, and we need to do the | |
434 | * full search for a revoke record. */ | |
435 | if (test_set_buffer_revokevalid(bh)) { | |
436 | need_cancel = test_clear_buffer_revoked(bh); | |
437 | } else { | |
438 | need_cancel = 1; | |
439 | clear_buffer_revoked(bh); | |
440 | } | |
441 | ||
442 | if (need_cancel) { | |
443 | record = find_revoke_record(journal, bh->b_blocknr); | |
444 | if (record) { | |
445 | jbd_debug(4, "cancelled existing revoke on " | |
446 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
447 | spin_lock(&journal->j_revoke_lock); | |
448 | list_del(&record->hash); | |
449 | spin_unlock(&journal->j_revoke_lock); | |
450 | kmem_cache_free(revoke_record_cache, record); | |
451 | did_revoke = 1; | |
452 | } | |
453 | } | |
454 | ||
455 | #ifdef JBD_EXPENSIVE_CHECKING | |
456 | /* There better not be one left behind by now! */ | |
457 | record = find_revoke_record(journal, bh->b_blocknr); | |
458 | J_ASSERT_JH(jh, record == NULL); | |
459 | #endif | |
460 | ||
461 | /* Finally, have we just cleared revoke on an unhashed | |
462 | * buffer_head? If so, we'd better make sure we clear the | |
463 | * revoked status on any hashed alias too, otherwise the revoke | |
464 | * state machine will get very upset later on. */ | |
465 | if (need_cancel) { | |
466 | struct buffer_head *bh2; | |
467 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
468 | if (bh2) { | |
469 | if (bh2 != bh) | |
470 | clear_buffer_revoked(bh2); | |
471 | __brelse(bh2); | |
472 | } | |
473 | } | |
474 | return did_revoke; | |
475 | } | |
476 | ||
477 | /* journal_switch_revoke table select j_revoke for next transaction | |
478 | * we do not want to suspend any processing until all revokes are | |
479 | * written -bzzz | |
480 | */ | |
481 | void journal_switch_revoke_table(journal_t *journal) | |
482 | { | |
483 | int i; | |
484 | ||
485 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
486 | journal->j_revoke = journal->j_revoke_table[1]; | |
487 | else | |
488 | journal->j_revoke = journal->j_revoke_table[0]; | |
489 | ||
ae6ddcc5 | 490 | for (i = 0; i < journal->j_revoke->hash_size; i++) |
1da177e4 LT |
491 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); |
492 | } | |
493 | ||
494 | /* | |
495 | * Write revoke records to the journal for all entries in the current | |
496 | * revoke hash, deleting the entries as we go. | |
497 | * | |
498 | * Called with the journal lock held. | |
499 | */ | |
500 | ||
ae6ddcc5 | 501 | void journal_write_revoke_records(journal_t *journal, |
1da177e4 LT |
502 | transaction_t *transaction) |
503 | { | |
504 | struct journal_head *descriptor; | |
505 | struct jbd_revoke_record_s *record; | |
506 | struct jbd_revoke_table_s *revoke; | |
507 | struct list_head *hash_list; | |
508 | int i, offset, count; | |
509 | ||
ae6ddcc5 | 510 | descriptor = NULL; |
1da177e4 LT |
511 | offset = 0; |
512 | count = 0; | |
513 | ||
514 | /* select revoke table for committing transaction */ | |
515 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
516 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
517 | ||
518 | for (i = 0; i < revoke->hash_size; i++) { | |
519 | hash_list = &revoke->hash_table[i]; | |
520 | ||
521 | while (!list_empty(hash_list)) { | |
ae6ddcc5 | 522 | record = (struct jbd_revoke_record_s *) |
1da177e4 LT |
523 | hash_list->next; |
524 | write_one_revoke_record(journal, transaction, | |
ae6ddcc5 | 525 | &descriptor, &offset, |
1da177e4 LT |
526 | record); |
527 | count++; | |
528 | list_del(&record->hash); | |
529 | kmem_cache_free(revoke_record_cache, record); | |
530 | } | |
531 | } | |
532 | if (descriptor) | |
533 | flush_descriptor(journal, descriptor, offset); | |
534 | jbd_debug(1, "Wrote %d revoke records\n", count); | |
535 | } | |
536 | ||
ae6ddcc5 | 537 | /* |
1da177e4 | 538 | * Write out one revoke record. We need to create a new descriptor |
ae6ddcc5 | 539 | * block if the old one is full or if we have not already created one. |
1da177e4 LT |
540 | */ |
541 | ||
ae6ddcc5 | 542 | static void write_one_revoke_record(journal_t *journal, |
1da177e4 | 543 | transaction_t *transaction, |
ae6ddcc5 | 544 | struct journal_head **descriptorp, |
1da177e4 LT |
545 | int *offsetp, |
546 | struct jbd_revoke_record_s *record) | |
547 | { | |
548 | struct journal_head *descriptor; | |
549 | int offset; | |
550 | journal_header_t *header; | |
551 | ||
552 | /* If we are already aborting, this all becomes a noop. We | |
553 | still need to go round the loop in | |
554 | journal_write_revoke_records in order to free all of the | |
555 | revoke records: only the IO to the journal is omitted. */ | |
556 | if (is_journal_aborted(journal)) | |
557 | return; | |
558 | ||
559 | descriptor = *descriptorp; | |
560 | offset = *offsetp; | |
561 | ||
562 | /* Make sure we have a descriptor with space left for the record */ | |
563 | if (descriptor) { | |
564 | if (offset == journal->j_blocksize) { | |
565 | flush_descriptor(journal, descriptor, offset); | |
566 | descriptor = NULL; | |
567 | } | |
568 | } | |
569 | ||
570 | if (!descriptor) { | |
571 | descriptor = journal_get_descriptor_buffer(journal); | |
572 | if (!descriptor) | |
573 | return; | |
574 | header = (journal_header_t *) &jh2bh(descriptor)->b_data[0]; | |
575 | header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER); | |
576 | header->h_blocktype = cpu_to_be32(JFS_REVOKE_BLOCK); | |
577 | header->h_sequence = cpu_to_be32(transaction->t_tid); | |
578 | ||
579 | /* Record it so that we can wait for IO completion later */ | |
580 | JBUFFER_TRACE(descriptor, "file as BJ_LogCtl"); | |
581 | journal_file_buffer(descriptor, transaction, BJ_LogCtl); | |
582 | ||
583 | offset = sizeof(journal_revoke_header_t); | |
584 | *descriptorp = descriptor; | |
585 | } | |
586 | ||
ae6ddcc5 | 587 | * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) = |
1da177e4 LT |
588 | cpu_to_be32(record->blocknr); |
589 | offset += 4; | |
590 | *offsetp = offset; | |
591 | } | |
592 | ||
ae6ddcc5 | 593 | /* |
1da177e4 LT |
594 | * Flush a revoke descriptor out to the journal. If we are aborting, |
595 | * this is a noop; otherwise we are generating a buffer which needs to | |
596 | * be waited for during commit, so it has to go onto the appropriate | |
597 | * journal buffer list. | |
598 | */ | |
599 | ||
ae6ddcc5 MC |
600 | static void flush_descriptor(journal_t *journal, |
601 | struct journal_head *descriptor, | |
1da177e4 LT |
602 | int offset) |
603 | { | |
604 | journal_revoke_header_t *header; | |
605 | struct buffer_head *bh = jh2bh(descriptor); | |
606 | ||
607 | if (is_journal_aborted(journal)) { | |
608 | put_bh(bh); | |
609 | return; | |
610 | } | |
611 | ||
612 | header = (journal_revoke_header_t *) jh2bh(descriptor)->b_data; | |
613 | header->r_count = cpu_to_be32(offset); | |
614 | set_buffer_jwrite(bh); | |
615 | BUFFER_TRACE(bh, "write"); | |
616 | set_buffer_dirty(bh); | |
26707699 | 617 | ll_rw_block(SWRITE, 1, &bh); |
1da177e4 LT |
618 | } |
619 | #endif | |
620 | ||
ae6ddcc5 | 621 | /* |
1da177e4 LT |
622 | * Revoke support for recovery. |
623 | * | |
624 | * Recovery needs to be able to: | |
625 | * | |
626 | * record all revoke records, including the tid of the latest instance | |
627 | * of each revoke in the journal | |
628 | * | |
629 | * check whether a given block in a given transaction should be replayed | |
630 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
631 | * transaction) | |
ae6ddcc5 | 632 | * |
1da177e4 LT |
633 | * empty the revoke table after recovery. |
634 | */ | |
635 | ||
636 | /* | |
637 | * First, setting revoke records. We create a new revoke record for | |
638 | * every block ever revoked in the log as we scan it for recovery, and | |
639 | * we update the existing records if we find multiple revokes for a | |
ae6ddcc5 | 640 | * single block. |
1da177e4 LT |
641 | */ |
642 | ||
ae6ddcc5 MC |
643 | int journal_set_revoke(journal_t *journal, |
644 | unsigned long blocknr, | |
1da177e4 LT |
645 | tid_t sequence) |
646 | { | |
647 | struct jbd_revoke_record_s *record; | |
648 | ||
649 | record = find_revoke_record(journal, blocknr); | |
650 | if (record) { | |
651 | /* If we have multiple occurrences, only record the | |
652 | * latest sequence number in the hashed record */ | |
653 | if (tid_gt(sequence, record->sequence)) | |
654 | record->sequence = sequence; | |
655 | return 0; | |
ae6ddcc5 | 656 | } |
1da177e4 LT |
657 | return insert_revoke_hash(journal, blocknr, sequence); |
658 | } | |
659 | ||
ae6ddcc5 | 660 | /* |
1da177e4 LT |
661 | * Test revoke records. For a given block referenced in the log, has |
662 | * that block been revoked? A revoke record with a given transaction | |
663 | * sequence number revokes all blocks in that transaction and earlier | |
664 | * ones, but later transactions still need replayed. | |
665 | */ | |
666 | ||
ae6ddcc5 | 667 | int journal_test_revoke(journal_t *journal, |
1da177e4 LT |
668 | unsigned long blocknr, |
669 | tid_t sequence) | |
670 | { | |
671 | struct jbd_revoke_record_s *record; | |
672 | ||
673 | record = find_revoke_record(journal, blocknr); | |
674 | if (!record) | |
675 | return 0; | |
676 | if (tid_gt(sequence, record->sequence)) | |
677 | return 0; | |
678 | return 1; | |
679 | } | |
680 | ||
681 | /* | |
682 | * Finally, once recovery is over, we need to clear the revoke table so | |
683 | * that it can be reused by the running filesystem. | |
684 | */ | |
685 | ||
686 | void journal_clear_revoke(journal_t *journal) | |
687 | { | |
688 | int i; | |
689 | struct list_head *hash_list; | |
690 | struct jbd_revoke_record_s *record; | |
691 | struct jbd_revoke_table_s *revoke; | |
692 | ||
693 | revoke = journal->j_revoke; | |
694 | ||
695 | for (i = 0; i < revoke->hash_size; i++) { | |
696 | hash_list = &revoke->hash_table[i]; | |
697 | while (!list_empty(hash_list)) { | |
698 | record = (struct jbd_revoke_record_s*) hash_list->next; | |
699 | list_del(&record->hash); | |
700 | kmem_cache_free(revoke_record_cache, record); | |
701 | } | |
702 | } | |
703 | } |