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; | |
08fc99bf | 141 | jbd_debug(1, "ENOMEM in %s, retrying\n", __func__); |
1da177e4 LT |
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 | ||
1984bb76 DG |
169 | void journal_destroy_revoke_caches(void) |
170 | { | |
171 | if (revoke_record_cache) { | |
172 | kmem_cache_destroy(revoke_record_cache); | |
173 | revoke_record_cache = NULL; | |
174 | } | |
175 | if (revoke_table_cache) { | |
176 | kmem_cache_destroy(revoke_table_cache); | |
177 | revoke_table_cache = NULL; | |
178 | } | |
179 | } | |
180 | ||
1da177e4 LT |
181 | int __init journal_init_revoke_caches(void) |
182 | { | |
1984bb76 DG |
183 | J_ASSERT(!revoke_record_cache); |
184 | J_ASSERT(!revoke_table_cache); | |
185 | ||
1da177e4 LT |
186 | revoke_record_cache = kmem_cache_create("revoke_record", |
187 | sizeof(struct jbd_revoke_record_s), | |
e12ba74d MG |
188 | 0, |
189 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY, | |
190 | NULL); | |
1076d17a | 191 | if (!revoke_record_cache) |
1984bb76 | 192 | goto record_cache_failure; |
1da177e4 LT |
193 | |
194 | revoke_table_cache = kmem_cache_create("revoke_table", | |
195 | sizeof(struct jbd_revoke_table_s), | |
e12ba74d | 196 | 0, SLAB_TEMPORARY, NULL); |
1984bb76 DG |
197 | if (!revoke_table_cache) |
198 | goto table_cache_failure; | |
199 | ||
1da177e4 | 200 | return 0; |
1da177e4 | 201 | |
1984bb76 DG |
202 | table_cache_failure: |
203 | journal_destroy_revoke_caches(); | |
204 | record_cache_failure: | |
205 | return -ENOMEM; | |
1da177e4 LT |
206 | } |
207 | ||
f4d79ca2 | 208 | static struct jbd_revoke_table_s *journal_init_revoke_table(int hash_size) |
1da177e4 | 209 | { |
f4d79ca2 DG |
210 | int shift = 0; |
211 | int tmp = hash_size; | |
212 | struct jbd_revoke_table_s *table; | |
1da177e4 | 213 | |
f4d79ca2 DG |
214 | table = kmem_cache_alloc(revoke_table_cache, GFP_KERNEL); |
215 | if (!table) | |
216 | goto out; | |
1da177e4 | 217 | |
1da177e4 LT |
218 | while((tmp >>= 1UL) != 0UL) |
219 | shift++; | |
220 | ||
f4d79ca2 DG |
221 | table->hash_size = hash_size; |
222 | table->hash_shift = shift; | |
223 | table->hash_table = | |
1da177e4 | 224 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); |
f4d79ca2 DG |
225 | if (!table->hash_table) { |
226 | kmem_cache_free(revoke_table_cache, table); | |
227 | table = NULL; | |
228 | goto out; | |
1da177e4 LT |
229 | } |
230 | ||
231 | for (tmp = 0; tmp < hash_size; tmp++) | |
f4d79ca2 | 232 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
1da177e4 | 233 | |
f4d79ca2 DG |
234 | out: |
235 | return table; | |
236 | } | |
237 | ||
238 | static void journal_destroy_revoke_table(struct jbd_revoke_table_s *table) | |
239 | { | |
240 | int i; | |
241 | struct list_head *hash_list; | |
242 | ||
243 | for (i = 0; i < table->hash_size; i++) { | |
244 | hash_list = &table->hash_table[i]; | |
245 | J_ASSERT(list_empty(hash_list)); | |
1da177e4 LT |
246 | } |
247 | ||
f4d79ca2 DG |
248 | kfree(table->hash_table); |
249 | kmem_cache_free(revoke_table_cache, table); | |
250 | } | |
1da177e4 | 251 | |
f4d79ca2 DG |
252 | /* Initialise the revoke table for a given journal to a given size. */ |
253 | int journal_init_revoke(journal_t *journal, int hash_size) | |
254 | { | |
255 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
f482394c | 256 | J_ASSERT(is_power_of_2(hash_size)); |
1da177e4 | 257 | |
f4d79ca2 DG |
258 | journal->j_revoke_table[0] = journal_init_revoke_table(hash_size); |
259 | if (!journal->j_revoke_table[0]) | |
260 | goto fail0; | |
1da177e4 | 261 | |
f4d79ca2 DG |
262 | journal->j_revoke_table[1] = journal_init_revoke_table(hash_size); |
263 | if (!journal->j_revoke_table[1]) | |
264 | goto fail1; | |
1da177e4 | 265 | |
f4d79ca2 | 266 | journal->j_revoke = journal->j_revoke_table[1]; |
1da177e4 LT |
267 | |
268 | spin_lock_init(&journal->j_revoke_lock); | |
269 | ||
270 | return 0; | |
1da177e4 | 271 | |
f4d79ca2 DG |
272 | fail1: |
273 | journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
274 | fail0: | |
275 | return -ENOMEM; | |
276 | } | |
1da177e4 | 277 | |
f4d79ca2 | 278 | /* Destroy a journal's revoke table. The table must already be empty! */ |
1da177e4 LT |
279 | void journal_destroy_revoke(journal_t *journal) |
280 | { | |
1da177e4 | 281 | journal->j_revoke = NULL; |
f4d79ca2 DG |
282 | if (journal->j_revoke_table[0]) |
283 | journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
284 | if (journal->j_revoke_table[1]) | |
285 | journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
1da177e4 LT |
286 | } |
287 | ||
288 | ||
289 | #ifdef __KERNEL__ | |
290 | ||
ae6ddcc5 | 291 | /* |
1da177e4 LT |
292 | * journal_revoke: revoke a given buffer_head from the journal. This |
293 | * prevents the block from being replayed during recovery if we take a | |
294 | * crash after this current transaction commits. Any subsequent | |
295 | * metadata writes of the buffer in this transaction cancel the | |
ae6ddcc5 | 296 | * revoke. |
1da177e4 LT |
297 | * |
298 | * Note that this call may block --- it is up to the caller to make | |
299 | * sure that there are no further calls to journal_write_metadata | |
300 | * before the revoke is complete. In ext3, this implies calling the | |
301 | * revoke before clearing the block bitmap when we are deleting | |
ae6ddcc5 | 302 | * metadata. |
1da177e4 LT |
303 | * |
304 | * Revoke performs a journal_forget on any buffer_head passed in as a | |
305 | * parameter, but does _not_ forget the buffer_head if the bh was only | |
ae6ddcc5 | 306 | * found implicitly. |
1da177e4 LT |
307 | * |
308 | * bh_in may not be a journalled buffer - it may have come off | |
309 | * the hash tables without an attached journal_head. | |
310 | * | |
311 | * If bh_in is non-zero, journal_revoke() will decrement its b_count | |
312 | * by one. | |
313 | */ | |
314 | ||
ae6ddcc5 | 315 | int journal_revoke(handle_t *handle, unsigned long blocknr, |
1da177e4 LT |
316 | struct buffer_head *bh_in) |
317 | { | |
318 | struct buffer_head *bh = NULL; | |
319 | journal_t *journal; | |
320 | struct block_device *bdev; | |
321 | int err; | |
322 | ||
323 | might_sleep(); | |
324 | if (bh_in) | |
325 | BUFFER_TRACE(bh_in, "enter"); | |
326 | ||
327 | journal = handle->h_transaction->t_journal; | |
328 | if (!journal_set_features(journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)){ | |
329 | J_ASSERT (!"Cannot set revoke feature!"); | |
330 | return -EINVAL; | |
331 | } | |
332 | ||
333 | bdev = journal->j_fs_dev; | |
334 | bh = bh_in; | |
335 | ||
336 | if (!bh) { | |
337 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
338 | if (bh) | |
339 | BUFFER_TRACE(bh, "found on hash"); | |
340 | } | |
341 | #ifdef JBD_EXPENSIVE_CHECKING | |
342 | else { | |
343 | struct buffer_head *bh2; | |
344 | ||
345 | /* If there is a different buffer_head lying around in | |
346 | * memory anywhere... */ | |
347 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
348 | if (bh2) { | |
349 | /* ... and it has RevokeValid status... */ | |
350 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
351 | /* ...then it better be revoked too, | |
352 | * since it's illegal to create a revoke | |
353 | * record against a buffer_head which is | |
354 | * not marked revoked --- that would | |
355 | * risk missing a subsequent revoke | |
356 | * cancel. */ | |
357 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
358 | put_bh(bh2); | |
359 | } | |
360 | } | |
361 | #endif | |
362 | ||
363 | /* We really ought not ever to revoke twice in a row without | |
364 | first having the revoke cancelled: it's illegal to free a | |
365 | block twice without allocating it in between! */ | |
366 | if (bh) { | |
367 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
368 | "inconsistent data on disk")) { | |
369 | if (!bh_in) | |
370 | brelse(bh); | |
371 | return -EIO; | |
372 | } | |
373 | set_buffer_revoked(bh); | |
374 | set_buffer_revokevalid(bh); | |
375 | if (bh_in) { | |
376 | BUFFER_TRACE(bh_in, "call journal_forget"); | |
377 | journal_forget(handle, bh_in); | |
378 | } else { | |
379 | BUFFER_TRACE(bh, "call brelse"); | |
380 | __brelse(bh); | |
381 | } | |
382 | } | |
383 | ||
384 | jbd_debug(2, "insert revoke for block %lu, bh_in=%p\n", blocknr, bh_in); | |
385 | err = insert_revoke_hash(journal, blocknr, | |
386 | handle->h_transaction->t_tid); | |
387 | BUFFER_TRACE(bh_in, "exit"); | |
388 | return err; | |
389 | } | |
390 | ||
391 | /* | |
392 | * Cancel an outstanding revoke. For use only internally by the | |
393 | * journaling code (called from journal_get_write_access). | |
394 | * | |
395 | * We trust buffer_revoked() on the buffer if the buffer is already | |
396 | * being journaled: if there is no revoke pending on the buffer, then we | |
397 | * don't do anything here. | |
398 | * | |
399 | * This would break if it were possible for a buffer to be revoked and | |
400 | * discarded, and then reallocated within the same transaction. In such | |
401 | * a case we would have lost the revoked bit, but when we arrived here | |
402 | * the second time we would still have a pending revoke to cancel. So, | |
403 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
404 | * set. | |
405 | * | |
406 | * The caller must have the journal locked. | |
407 | */ | |
408 | int journal_cancel_revoke(handle_t *handle, struct journal_head *jh) | |
409 | { | |
410 | struct jbd_revoke_record_s *record; | |
411 | journal_t *journal = handle->h_transaction->t_journal; | |
412 | int need_cancel; | |
413 | int did_revoke = 0; /* akpm: debug */ | |
414 | struct buffer_head *bh = jh2bh(jh); | |
415 | ||
416 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
417 | ||
418 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
419 | * only perform the full cancel if the revoke bit is set. If | |
420 | * not, we can't trust the revoke bit, and we need to do the | |
421 | * full search for a revoke record. */ | |
422 | if (test_set_buffer_revokevalid(bh)) { | |
423 | need_cancel = test_clear_buffer_revoked(bh); | |
424 | } else { | |
425 | need_cancel = 1; | |
426 | clear_buffer_revoked(bh); | |
427 | } | |
428 | ||
429 | if (need_cancel) { | |
430 | record = find_revoke_record(journal, bh->b_blocknr); | |
431 | if (record) { | |
432 | jbd_debug(4, "cancelled existing revoke on " | |
433 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
434 | spin_lock(&journal->j_revoke_lock); | |
435 | list_del(&record->hash); | |
436 | spin_unlock(&journal->j_revoke_lock); | |
437 | kmem_cache_free(revoke_record_cache, record); | |
438 | did_revoke = 1; | |
439 | } | |
440 | } | |
441 | ||
442 | #ifdef JBD_EXPENSIVE_CHECKING | |
443 | /* There better not be one left behind by now! */ | |
444 | record = find_revoke_record(journal, bh->b_blocknr); | |
445 | J_ASSERT_JH(jh, record == NULL); | |
446 | #endif | |
447 | ||
448 | /* Finally, have we just cleared revoke on an unhashed | |
449 | * buffer_head? If so, we'd better make sure we clear the | |
450 | * revoked status on any hashed alias too, otherwise the revoke | |
451 | * state machine will get very upset later on. */ | |
452 | if (need_cancel) { | |
453 | struct buffer_head *bh2; | |
454 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
455 | if (bh2) { | |
456 | if (bh2 != bh) | |
457 | clear_buffer_revoked(bh2); | |
458 | __brelse(bh2); | |
459 | } | |
460 | } | |
461 | return did_revoke; | |
462 | } | |
463 | ||
464 | /* journal_switch_revoke table select j_revoke for next transaction | |
465 | * we do not want to suspend any processing until all revokes are | |
466 | * written -bzzz | |
467 | */ | |
468 | void journal_switch_revoke_table(journal_t *journal) | |
469 | { | |
470 | int i; | |
471 | ||
472 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
473 | journal->j_revoke = journal->j_revoke_table[1]; | |
474 | else | |
475 | journal->j_revoke = journal->j_revoke_table[0]; | |
476 | ||
ae6ddcc5 | 477 | for (i = 0; i < journal->j_revoke->hash_size; i++) |
1da177e4 LT |
478 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); |
479 | } | |
480 | ||
481 | /* | |
482 | * Write revoke records to the journal for all entries in the current | |
483 | * revoke hash, deleting the entries as we go. | |
484 | * | |
485 | * Called with the journal lock held. | |
486 | */ | |
487 | ||
ae6ddcc5 | 488 | void journal_write_revoke_records(journal_t *journal, |
1da177e4 LT |
489 | transaction_t *transaction) |
490 | { | |
491 | struct journal_head *descriptor; | |
492 | struct jbd_revoke_record_s *record; | |
493 | struct jbd_revoke_table_s *revoke; | |
494 | struct list_head *hash_list; | |
495 | int i, offset, count; | |
496 | ||
ae6ddcc5 | 497 | descriptor = NULL; |
1da177e4 LT |
498 | offset = 0; |
499 | count = 0; | |
500 | ||
501 | /* select revoke table for committing transaction */ | |
502 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
503 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
504 | ||
505 | for (i = 0; i < revoke->hash_size; i++) { | |
506 | hash_list = &revoke->hash_table[i]; | |
507 | ||
508 | while (!list_empty(hash_list)) { | |
ae6ddcc5 | 509 | record = (struct jbd_revoke_record_s *) |
1da177e4 LT |
510 | hash_list->next; |
511 | write_one_revoke_record(journal, transaction, | |
ae6ddcc5 | 512 | &descriptor, &offset, |
1da177e4 LT |
513 | record); |
514 | count++; | |
515 | list_del(&record->hash); | |
516 | kmem_cache_free(revoke_record_cache, record); | |
517 | } | |
518 | } | |
519 | if (descriptor) | |
520 | flush_descriptor(journal, descriptor, offset); | |
521 | jbd_debug(1, "Wrote %d revoke records\n", count); | |
522 | } | |
523 | ||
ae6ddcc5 | 524 | /* |
1da177e4 | 525 | * Write out one revoke record. We need to create a new descriptor |
ae6ddcc5 | 526 | * block if the old one is full or if we have not already created one. |
1da177e4 LT |
527 | */ |
528 | ||
ae6ddcc5 | 529 | static void write_one_revoke_record(journal_t *journal, |
1da177e4 | 530 | transaction_t *transaction, |
ae6ddcc5 | 531 | struct journal_head **descriptorp, |
1da177e4 LT |
532 | int *offsetp, |
533 | struct jbd_revoke_record_s *record) | |
534 | { | |
535 | struct journal_head *descriptor; | |
536 | int offset; | |
537 | journal_header_t *header; | |
538 | ||
539 | /* If we are already aborting, this all becomes a noop. We | |
540 | still need to go round the loop in | |
541 | journal_write_revoke_records in order to free all of the | |
542 | revoke records: only the IO to the journal is omitted. */ | |
543 | if (is_journal_aborted(journal)) | |
544 | return; | |
545 | ||
546 | descriptor = *descriptorp; | |
547 | offset = *offsetp; | |
548 | ||
549 | /* Make sure we have a descriptor with space left for the record */ | |
550 | if (descriptor) { | |
551 | if (offset == journal->j_blocksize) { | |
552 | flush_descriptor(journal, descriptor, offset); | |
553 | descriptor = NULL; | |
554 | } | |
555 | } | |
556 | ||
557 | if (!descriptor) { | |
558 | descriptor = journal_get_descriptor_buffer(journal); | |
559 | if (!descriptor) | |
560 | return; | |
561 | header = (journal_header_t *) &jh2bh(descriptor)->b_data[0]; | |
562 | header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER); | |
563 | header->h_blocktype = cpu_to_be32(JFS_REVOKE_BLOCK); | |
564 | header->h_sequence = cpu_to_be32(transaction->t_tid); | |
565 | ||
566 | /* Record it so that we can wait for IO completion later */ | |
567 | JBUFFER_TRACE(descriptor, "file as BJ_LogCtl"); | |
568 | journal_file_buffer(descriptor, transaction, BJ_LogCtl); | |
569 | ||
570 | offset = sizeof(journal_revoke_header_t); | |
571 | *descriptorp = descriptor; | |
572 | } | |
573 | ||
ae6ddcc5 | 574 | * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) = |
1da177e4 LT |
575 | cpu_to_be32(record->blocknr); |
576 | offset += 4; | |
577 | *offsetp = offset; | |
578 | } | |
579 | ||
ae6ddcc5 | 580 | /* |
1da177e4 LT |
581 | * Flush a revoke descriptor out to the journal. If we are aborting, |
582 | * this is a noop; otherwise we are generating a buffer which needs to | |
583 | * be waited for during commit, so it has to go onto the appropriate | |
584 | * journal buffer list. | |
585 | */ | |
586 | ||
ae6ddcc5 MC |
587 | static void flush_descriptor(journal_t *journal, |
588 | struct journal_head *descriptor, | |
1da177e4 LT |
589 | int offset) |
590 | { | |
591 | journal_revoke_header_t *header; | |
592 | struct buffer_head *bh = jh2bh(descriptor); | |
593 | ||
594 | if (is_journal_aborted(journal)) { | |
595 | put_bh(bh); | |
596 | return; | |
597 | } | |
598 | ||
599 | header = (journal_revoke_header_t *) jh2bh(descriptor)->b_data; | |
600 | header->r_count = cpu_to_be32(offset); | |
601 | set_buffer_jwrite(bh); | |
602 | BUFFER_TRACE(bh, "write"); | |
603 | set_buffer_dirty(bh); | |
26707699 | 604 | ll_rw_block(SWRITE, 1, &bh); |
1da177e4 LT |
605 | } |
606 | #endif | |
607 | ||
ae6ddcc5 | 608 | /* |
1da177e4 LT |
609 | * Revoke support for recovery. |
610 | * | |
611 | * Recovery needs to be able to: | |
612 | * | |
613 | * record all revoke records, including the tid of the latest instance | |
614 | * of each revoke in the journal | |
615 | * | |
616 | * check whether a given block in a given transaction should be replayed | |
617 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
618 | * transaction) | |
ae6ddcc5 | 619 | * |
1da177e4 LT |
620 | * empty the revoke table after recovery. |
621 | */ | |
622 | ||
623 | /* | |
624 | * First, setting revoke records. We create a new revoke record for | |
625 | * every block ever revoked in the log as we scan it for recovery, and | |
626 | * we update the existing records if we find multiple revokes for a | |
ae6ddcc5 | 627 | * single block. |
1da177e4 LT |
628 | */ |
629 | ||
ae6ddcc5 MC |
630 | int journal_set_revoke(journal_t *journal, |
631 | unsigned long blocknr, | |
1da177e4 LT |
632 | tid_t sequence) |
633 | { | |
634 | struct jbd_revoke_record_s *record; | |
635 | ||
636 | record = find_revoke_record(journal, blocknr); | |
637 | if (record) { | |
638 | /* If we have multiple occurrences, only record the | |
639 | * latest sequence number in the hashed record */ | |
640 | if (tid_gt(sequence, record->sequence)) | |
641 | record->sequence = sequence; | |
642 | return 0; | |
ae6ddcc5 | 643 | } |
1da177e4 LT |
644 | return insert_revoke_hash(journal, blocknr, sequence); |
645 | } | |
646 | ||
ae6ddcc5 | 647 | /* |
1da177e4 LT |
648 | * Test revoke records. For a given block referenced in the log, has |
649 | * that block been revoked? A revoke record with a given transaction | |
650 | * sequence number revokes all blocks in that transaction and earlier | |
651 | * ones, but later transactions still need replayed. | |
652 | */ | |
653 | ||
ae6ddcc5 | 654 | int journal_test_revoke(journal_t *journal, |
1da177e4 LT |
655 | unsigned long blocknr, |
656 | tid_t sequence) | |
657 | { | |
658 | struct jbd_revoke_record_s *record; | |
659 | ||
660 | record = find_revoke_record(journal, blocknr); | |
661 | if (!record) | |
662 | return 0; | |
663 | if (tid_gt(sequence, record->sequence)) | |
664 | return 0; | |
665 | return 1; | |
666 | } | |
667 | ||
668 | /* | |
669 | * Finally, once recovery is over, we need to clear the revoke table so | |
670 | * that it can be reused by the running filesystem. | |
671 | */ | |
672 | ||
673 | void journal_clear_revoke(journal_t *journal) | |
674 | { | |
675 | int i; | |
676 | struct list_head *hash_list; | |
677 | struct jbd_revoke_record_s *record; | |
678 | struct jbd_revoke_table_s *revoke; | |
679 | ||
680 | revoke = journal->j_revoke; | |
681 | ||
682 | for (i = 0; i < revoke->hash_size; i++) { | |
683 | hash_list = &revoke->hash_table[i]; | |
684 | while (!list_empty(hash_list)) { | |
685 | record = (struct jbd_revoke_record_s*) hash_list->next; | |
686 | list_del(&record->hash); | |
687 | kmem_cache_free(revoke_record_cache, record); | |
688 | } | |
689 | } | |
690 | } |