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> |
1da177e4 | 96 | |
e18b890b CL |
97 | static struct kmem_cache *revoke_record_cache; |
98 | static struct kmem_cache *revoke_table_cache; | |
1da177e4 LT |
99 | |
100 | /* Each revoke record represents one single revoked block. During | |
101 | journal replay, this involves recording the transaction ID of the | |
102 | last transaction to revoke this block. */ | |
103 | ||
ae6ddcc5 | 104 | struct jbd_revoke_record_s |
1da177e4 LT |
105 | { |
106 | struct list_head hash; | |
107 | tid_t sequence; /* Used for recovery only */ | |
9c28cbcc | 108 | unsigned int blocknr; |
1da177e4 LT |
109 | }; |
110 | ||
111 | ||
112 | /* The revoke table is just a simple hash table of revoke records. */ | |
113 | struct jbd_revoke_table_s | |
114 | { | |
115 | /* It is conceivable that we might want a larger hash table | |
116 | * for recovery. Must be a power of two. */ | |
ae6ddcc5 MC |
117 | int hash_size; |
118 | int hash_shift; | |
1da177e4 LT |
119 | struct list_head *hash_table; |
120 | }; | |
121 | ||
122 | ||
123 | #ifdef __KERNEL__ | |
124 | static void write_one_revoke_record(journal_t *, transaction_t *, | |
125 | struct journal_head **, int *, | |
38d726d1 TT |
126 | struct jbd_revoke_record_s *, int); |
127 | static void flush_descriptor(journal_t *, struct journal_head *, int, int); | |
1da177e4 LT |
128 | #endif |
129 | ||
130 | /* Utility functions to maintain the revoke table */ | |
131 | ||
132 | /* Borrowed from buffer.c: this is a tried and tested block hash function */ | |
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; | |
136 | int hash_shift = table->hash_shift; | |
137 | ||
138 | return ((block << (hash_shift - 6)) ^ | |
139 | (block >> 13) ^ | |
140 | (block << (hash_shift - 12))) & (table->hash_size - 1); | |
141 | } | |
142 | ||
9c28cbcc | 143 | static int insert_revoke_hash(journal_t *journal, unsigned int blocknr, |
022a4a7b | 144 | tid_t seq) |
1da177e4 LT |
145 | { |
146 | struct list_head *hash_list; | |
147 | struct jbd_revoke_record_s *record; | |
148 | ||
149 | repeat: | |
150 | record = kmem_cache_alloc(revoke_record_cache, GFP_NOFS); | |
151 | if (!record) | |
152 | goto oom; | |
153 | ||
154 | record->sequence = seq; | |
155 | record->blocknr = blocknr; | |
156 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
157 | spin_lock(&journal->j_revoke_lock); | |
158 | list_add(&record->hash, hash_list); | |
159 | spin_unlock(&journal->j_revoke_lock); | |
160 | return 0; | |
161 | ||
162 | oom: | |
163 | if (!journal_oom_retry) | |
164 | return -ENOMEM; | |
08fc99bf | 165 | jbd_debug(1, "ENOMEM in %s, retrying\n", __func__); |
1da177e4 LT |
166 | yield(); |
167 | goto repeat; | |
168 | } | |
169 | ||
170 | /* Find a revoke record in the journal's hash table. */ | |
171 | ||
172 | static struct jbd_revoke_record_s *find_revoke_record(journal_t *journal, | |
9c28cbcc | 173 | unsigned int blocknr) |
1da177e4 LT |
174 | { |
175 | struct list_head *hash_list; | |
176 | struct jbd_revoke_record_s *record; | |
177 | ||
178 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
179 | ||
180 | spin_lock(&journal->j_revoke_lock); | |
181 | record = (struct jbd_revoke_record_s *) hash_list->next; | |
182 | while (&(record->hash) != hash_list) { | |
183 | if (record->blocknr == blocknr) { | |
184 | spin_unlock(&journal->j_revoke_lock); | |
185 | return record; | |
186 | } | |
187 | record = (struct jbd_revoke_record_s *) record->hash.next; | |
188 | } | |
189 | spin_unlock(&journal->j_revoke_lock); | |
190 | return NULL; | |
191 | } | |
192 | ||
1984bb76 DG |
193 | void journal_destroy_revoke_caches(void) |
194 | { | |
195 | if (revoke_record_cache) { | |
196 | kmem_cache_destroy(revoke_record_cache); | |
197 | revoke_record_cache = NULL; | |
198 | } | |
199 | if (revoke_table_cache) { | |
200 | kmem_cache_destroy(revoke_table_cache); | |
201 | revoke_table_cache = NULL; | |
202 | } | |
203 | } | |
204 | ||
1da177e4 LT |
205 | int __init journal_init_revoke_caches(void) |
206 | { | |
1984bb76 DG |
207 | J_ASSERT(!revoke_record_cache); |
208 | J_ASSERT(!revoke_table_cache); | |
209 | ||
1da177e4 LT |
210 | revoke_record_cache = kmem_cache_create("revoke_record", |
211 | sizeof(struct jbd_revoke_record_s), | |
e12ba74d MG |
212 | 0, |
213 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY, | |
214 | NULL); | |
1076d17a | 215 | if (!revoke_record_cache) |
1984bb76 | 216 | goto record_cache_failure; |
1da177e4 LT |
217 | |
218 | revoke_table_cache = kmem_cache_create("revoke_table", | |
219 | sizeof(struct jbd_revoke_table_s), | |
e12ba74d | 220 | 0, SLAB_TEMPORARY, NULL); |
1984bb76 DG |
221 | if (!revoke_table_cache) |
222 | goto table_cache_failure; | |
223 | ||
1da177e4 | 224 | return 0; |
1da177e4 | 225 | |
1984bb76 DG |
226 | table_cache_failure: |
227 | journal_destroy_revoke_caches(); | |
228 | record_cache_failure: | |
229 | return -ENOMEM; | |
1da177e4 LT |
230 | } |
231 | ||
f4d79ca2 | 232 | static struct jbd_revoke_table_s *journal_init_revoke_table(int hash_size) |
1da177e4 | 233 | { |
770901cc | 234 | int i; |
f4d79ca2 | 235 | struct jbd_revoke_table_s *table; |
1da177e4 | 236 | |
f4d79ca2 DG |
237 | table = kmem_cache_alloc(revoke_table_cache, GFP_KERNEL); |
238 | if (!table) | |
239 | goto out; | |
1da177e4 | 240 | |
f4d79ca2 | 241 | table->hash_size = hash_size; |
770901cc | 242 | table->hash_shift = ilog2(hash_size); |
f4d79ca2 | 243 | table->hash_table = |
1da177e4 | 244 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); |
f4d79ca2 DG |
245 | if (!table->hash_table) { |
246 | kmem_cache_free(revoke_table_cache, table); | |
247 | table = NULL; | |
248 | goto out; | |
1da177e4 LT |
249 | } |
250 | ||
770901cc FF |
251 | for (i = 0; i < hash_size; i++) |
252 | INIT_LIST_HEAD(&table->hash_table[i]); | |
1da177e4 | 253 | |
f4d79ca2 DG |
254 | out: |
255 | return table; | |
256 | } | |
257 | ||
258 | static void journal_destroy_revoke_table(struct jbd_revoke_table_s *table) | |
259 | { | |
260 | int i; | |
261 | struct list_head *hash_list; | |
262 | ||
263 | for (i = 0; i < table->hash_size; i++) { | |
264 | hash_list = &table->hash_table[i]; | |
265 | J_ASSERT(list_empty(hash_list)); | |
1da177e4 LT |
266 | } |
267 | ||
f4d79ca2 DG |
268 | kfree(table->hash_table); |
269 | kmem_cache_free(revoke_table_cache, table); | |
270 | } | |
1da177e4 | 271 | |
f4d79ca2 DG |
272 | /* Initialise the revoke table for a given journal to a given size. */ |
273 | int journal_init_revoke(journal_t *journal, int hash_size) | |
274 | { | |
275 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
f482394c | 276 | J_ASSERT(is_power_of_2(hash_size)); |
1da177e4 | 277 | |
f4d79ca2 DG |
278 | journal->j_revoke_table[0] = journal_init_revoke_table(hash_size); |
279 | if (!journal->j_revoke_table[0]) | |
280 | goto fail0; | |
1da177e4 | 281 | |
f4d79ca2 DG |
282 | journal->j_revoke_table[1] = journal_init_revoke_table(hash_size); |
283 | if (!journal->j_revoke_table[1]) | |
284 | goto fail1; | |
1da177e4 | 285 | |
f4d79ca2 | 286 | journal->j_revoke = journal->j_revoke_table[1]; |
1da177e4 LT |
287 | |
288 | spin_lock_init(&journal->j_revoke_lock); | |
289 | ||
290 | return 0; | |
1da177e4 | 291 | |
f4d79ca2 DG |
292 | fail1: |
293 | journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
294 | fail0: | |
295 | return -ENOMEM; | |
296 | } | |
1da177e4 | 297 | |
f4d79ca2 | 298 | /* Destroy a journal's revoke table. The table must already be empty! */ |
1da177e4 LT |
299 | void journal_destroy_revoke(journal_t *journal) |
300 | { | |
1da177e4 | 301 | journal->j_revoke = NULL; |
f4d79ca2 DG |
302 | if (journal->j_revoke_table[0]) |
303 | journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
304 | if (journal->j_revoke_table[1]) | |
305 | journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
1da177e4 LT |
306 | } |
307 | ||
308 | ||
309 | #ifdef __KERNEL__ | |
310 | ||
ae6ddcc5 | 311 | /* |
1da177e4 LT |
312 | * journal_revoke: revoke a given buffer_head from the journal. This |
313 | * prevents the block from being replayed during recovery if we take a | |
314 | * crash after this current transaction commits. Any subsequent | |
315 | * metadata writes of the buffer in this transaction cancel the | |
ae6ddcc5 | 316 | * revoke. |
1da177e4 LT |
317 | * |
318 | * Note that this call may block --- it is up to the caller to make | |
319 | * sure that there are no further calls to journal_write_metadata | |
320 | * before the revoke is complete. In ext3, this implies calling the | |
321 | * revoke before clearing the block bitmap when we are deleting | |
ae6ddcc5 | 322 | * metadata. |
1da177e4 LT |
323 | * |
324 | * Revoke performs a journal_forget on any buffer_head passed in as a | |
325 | * parameter, but does _not_ forget the buffer_head if the bh was only | |
ae6ddcc5 | 326 | * found implicitly. |
1da177e4 LT |
327 | * |
328 | * bh_in may not be a journalled buffer - it may have come off | |
329 | * the hash tables without an attached journal_head. | |
330 | * | |
331 | * If bh_in is non-zero, journal_revoke() will decrement its b_count | |
332 | * by one. | |
333 | */ | |
334 | ||
9c28cbcc | 335 | int journal_revoke(handle_t *handle, unsigned int blocknr, |
1da177e4 LT |
336 | struct buffer_head *bh_in) |
337 | { | |
338 | struct buffer_head *bh = NULL; | |
339 | journal_t *journal; | |
340 | struct block_device *bdev; | |
341 | int err; | |
342 | ||
343 | might_sleep(); | |
344 | if (bh_in) | |
345 | BUFFER_TRACE(bh_in, "enter"); | |
346 | ||
347 | journal = handle->h_transaction->t_journal; | |
348 | if (!journal_set_features(journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)){ | |
349 | J_ASSERT (!"Cannot set revoke feature!"); | |
350 | return -EINVAL; | |
351 | } | |
352 | ||
353 | bdev = journal->j_fs_dev; | |
354 | bh = bh_in; | |
355 | ||
356 | if (!bh) { | |
357 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
358 | if (bh) | |
359 | BUFFER_TRACE(bh, "found on hash"); | |
360 | } | |
361 | #ifdef JBD_EXPENSIVE_CHECKING | |
362 | else { | |
363 | struct buffer_head *bh2; | |
364 | ||
365 | /* If there is a different buffer_head lying around in | |
366 | * memory anywhere... */ | |
367 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
368 | if (bh2) { | |
369 | /* ... and it has RevokeValid status... */ | |
370 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
371 | /* ...then it better be revoked too, | |
372 | * since it's illegal to create a revoke | |
373 | * record against a buffer_head which is | |
374 | * not marked revoked --- that would | |
375 | * risk missing a subsequent revoke | |
376 | * cancel. */ | |
377 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
378 | put_bh(bh2); | |
379 | } | |
380 | } | |
381 | #endif | |
382 | ||
383 | /* We really ought not ever to revoke twice in a row without | |
384 | first having the revoke cancelled: it's illegal to free a | |
385 | block twice without allocating it in between! */ | |
386 | if (bh) { | |
387 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
388 | "inconsistent data on disk")) { | |
389 | if (!bh_in) | |
390 | brelse(bh); | |
391 | return -EIO; | |
392 | } | |
393 | set_buffer_revoked(bh); | |
394 | set_buffer_revokevalid(bh); | |
395 | if (bh_in) { | |
396 | BUFFER_TRACE(bh_in, "call journal_forget"); | |
397 | journal_forget(handle, bh_in); | |
398 | } else { | |
399 | BUFFER_TRACE(bh, "call brelse"); | |
400 | __brelse(bh); | |
401 | } | |
402 | } | |
403 | ||
9c28cbcc | 404 | jbd_debug(2, "insert revoke for block %u, bh_in=%p\n", blocknr, bh_in); |
1da177e4 LT |
405 | err = insert_revoke_hash(journal, blocknr, |
406 | handle->h_transaction->t_tid); | |
407 | BUFFER_TRACE(bh_in, "exit"); | |
408 | return err; | |
409 | } | |
410 | ||
411 | /* | |
412 | * Cancel an outstanding revoke. For use only internally by the | |
413 | * journaling code (called from journal_get_write_access). | |
414 | * | |
415 | * We trust buffer_revoked() on the buffer if the buffer is already | |
416 | * being journaled: if there is no revoke pending on the buffer, then we | |
417 | * don't do anything here. | |
418 | * | |
419 | * This would break if it were possible for a buffer to be revoked and | |
420 | * discarded, and then reallocated within the same transaction. In such | |
421 | * a case we would have lost the revoked bit, but when we arrived here | |
422 | * the second time we would still have a pending revoke to cancel. So, | |
423 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
424 | * set. | |
1da177e4 LT |
425 | */ |
426 | int journal_cancel_revoke(handle_t *handle, struct journal_head *jh) | |
427 | { | |
428 | struct jbd_revoke_record_s *record; | |
429 | journal_t *journal = handle->h_transaction->t_journal; | |
430 | int need_cancel; | |
431 | int did_revoke = 0; /* akpm: debug */ | |
432 | struct buffer_head *bh = jh2bh(jh); | |
433 | ||
434 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
435 | ||
436 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
437 | * only perform the full cancel if the revoke bit is set. If | |
438 | * not, we can't trust the revoke bit, and we need to do the | |
439 | * full search for a revoke record. */ | |
440 | if (test_set_buffer_revokevalid(bh)) { | |
441 | need_cancel = test_clear_buffer_revoked(bh); | |
442 | } else { | |
443 | need_cancel = 1; | |
444 | clear_buffer_revoked(bh); | |
445 | } | |
446 | ||
447 | if (need_cancel) { | |
448 | record = find_revoke_record(journal, bh->b_blocknr); | |
449 | if (record) { | |
450 | jbd_debug(4, "cancelled existing revoke on " | |
451 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
452 | spin_lock(&journal->j_revoke_lock); | |
453 | list_del(&record->hash); | |
454 | spin_unlock(&journal->j_revoke_lock); | |
455 | kmem_cache_free(revoke_record_cache, record); | |
456 | did_revoke = 1; | |
457 | } | |
458 | } | |
459 | ||
460 | #ifdef JBD_EXPENSIVE_CHECKING | |
461 | /* There better not be one left behind by now! */ | |
462 | record = find_revoke_record(journal, bh->b_blocknr); | |
463 | J_ASSERT_JH(jh, record == NULL); | |
464 | #endif | |
465 | ||
466 | /* Finally, have we just cleared revoke on an unhashed | |
467 | * buffer_head? If so, we'd better make sure we clear the | |
468 | * revoked status on any hashed alias too, otherwise the revoke | |
469 | * state machine will get very upset later on. */ | |
470 | if (need_cancel) { | |
471 | struct buffer_head *bh2; | |
472 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
473 | if (bh2) { | |
474 | if (bh2 != bh) | |
475 | clear_buffer_revoked(bh2); | |
476 | __brelse(bh2); | |
477 | } | |
478 | } | |
479 | return did_revoke; | |
480 | } | |
481 | ||
8c111b3f YY |
482 | /* |
483 | * journal_clear_revoked_flags clears revoked flag of buffers in | |
484 | * revoke table to reflect there is no revoked buffer in the next | |
485 | * transaction which is going to be started. | |
486 | */ | |
487 | void journal_clear_buffer_revoked_flags(journal_t *journal) | |
488 | { | |
489 | struct jbd_revoke_table_s *revoke = journal->j_revoke; | |
490 | int i = 0; | |
491 | ||
492 | for (i = 0; i < revoke->hash_size; i++) { | |
493 | struct list_head *hash_list; | |
494 | struct list_head *list_entry; | |
495 | hash_list = &revoke->hash_table[i]; | |
496 | ||
497 | list_for_each(list_entry, hash_list) { | |
498 | struct jbd_revoke_record_s *record; | |
499 | struct buffer_head *bh; | |
500 | record = (struct jbd_revoke_record_s *)list_entry; | |
501 | bh = __find_get_block(journal->j_fs_dev, | |
502 | record->blocknr, | |
503 | journal->j_blocksize); | |
504 | if (bh) { | |
505 | clear_buffer_revoked(bh); | |
506 | __brelse(bh); | |
507 | } | |
508 | } | |
509 | } | |
510 | } | |
511 | ||
1da177e4 LT |
512 | /* journal_switch_revoke table select j_revoke for next transaction |
513 | * we do not want to suspend any processing until all revokes are | |
514 | * written -bzzz | |
515 | */ | |
516 | void journal_switch_revoke_table(journal_t *journal) | |
517 | { | |
518 | int i; | |
519 | ||
520 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
521 | journal->j_revoke = journal->j_revoke_table[1]; | |
522 | else | |
523 | journal->j_revoke = journal->j_revoke_table[0]; | |
524 | ||
ae6ddcc5 | 525 | for (i = 0; i < journal->j_revoke->hash_size; i++) |
1da177e4 LT |
526 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); |
527 | } | |
528 | ||
529 | /* | |
530 | * Write revoke records to the journal for all entries in the current | |
531 | * revoke hash, deleting the entries as we go. | |
1da177e4 | 532 | */ |
ae6ddcc5 | 533 | void journal_write_revoke_records(journal_t *journal, |
38d726d1 | 534 | transaction_t *transaction, int write_op) |
1da177e4 LT |
535 | { |
536 | struct journal_head *descriptor; | |
537 | struct jbd_revoke_record_s *record; | |
538 | struct jbd_revoke_table_s *revoke; | |
539 | struct list_head *hash_list; | |
540 | int i, offset, count; | |
541 | ||
ae6ddcc5 | 542 | descriptor = NULL; |
1da177e4 LT |
543 | offset = 0; |
544 | count = 0; | |
545 | ||
546 | /* select revoke table for committing transaction */ | |
547 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
548 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
549 | ||
550 | for (i = 0; i < revoke->hash_size; i++) { | |
551 | hash_list = &revoke->hash_table[i]; | |
552 | ||
553 | while (!list_empty(hash_list)) { | |
ae6ddcc5 | 554 | record = (struct jbd_revoke_record_s *) |
1da177e4 LT |
555 | hash_list->next; |
556 | write_one_revoke_record(journal, transaction, | |
ae6ddcc5 | 557 | &descriptor, &offset, |
38d726d1 | 558 | record, write_op); |
1da177e4 LT |
559 | count++; |
560 | list_del(&record->hash); | |
561 | kmem_cache_free(revoke_record_cache, record); | |
562 | } | |
563 | } | |
564 | if (descriptor) | |
38d726d1 | 565 | flush_descriptor(journal, descriptor, offset, write_op); |
1da177e4 LT |
566 | jbd_debug(1, "Wrote %d revoke records\n", count); |
567 | } | |
568 | ||
ae6ddcc5 | 569 | /* |
1da177e4 | 570 | * Write out one revoke record. We need to create a new descriptor |
ae6ddcc5 | 571 | * block if the old one is full or if we have not already created one. |
1da177e4 LT |
572 | */ |
573 | ||
ae6ddcc5 | 574 | static void write_one_revoke_record(journal_t *journal, |
1da177e4 | 575 | transaction_t *transaction, |
ae6ddcc5 | 576 | struct journal_head **descriptorp, |
1da177e4 | 577 | int *offsetp, |
38d726d1 TT |
578 | struct jbd_revoke_record_s *record, |
579 | int write_op) | |
1da177e4 LT |
580 | { |
581 | struct journal_head *descriptor; | |
582 | int offset; | |
583 | journal_header_t *header; | |
584 | ||
585 | /* If we are already aborting, this all becomes a noop. We | |
586 | still need to go round the loop in | |
587 | journal_write_revoke_records in order to free all of the | |
588 | revoke records: only the IO to the journal is omitted. */ | |
589 | if (is_journal_aborted(journal)) | |
590 | return; | |
591 | ||
592 | descriptor = *descriptorp; | |
593 | offset = *offsetp; | |
594 | ||
595 | /* Make sure we have a descriptor with space left for the record */ | |
596 | if (descriptor) { | |
597 | if (offset == journal->j_blocksize) { | |
38d726d1 | 598 | flush_descriptor(journal, descriptor, offset, write_op); |
1da177e4 LT |
599 | descriptor = NULL; |
600 | } | |
601 | } | |
602 | ||
603 | if (!descriptor) { | |
604 | descriptor = journal_get_descriptor_buffer(journal); | |
605 | if (!descriptor) | |
606 | return; | |
607 | header = (journal_header_t *) &jh2bh(descriptor)->b_data[0]; | |
608 | header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER); | |
609 | header->h_blocktype = cpu_to_be32(JFS_REVOKE_BLOCK); | |
610 | header->h_sequence = cpu_to_be32(transaction->t_tid); | |
611 | ||
612 | /* Record it so that we can wait for IO completion later */ | |
613 | JBUFFER_TRACE(descriptor, "file as BJ_LogCtl"); | |
614 | journal_file_buffer(descriptor, transaction, BJ_LogCtl); | |
615 | ||
616 | offset = sizeof(journal_revoke_header_t); | |
617 | *descriptorp = descriptor; | |
618 | } | |
619 | ||
ae6ddcc5 | 620 | * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) = |
1da177e4 LT |
621 | cpu_to_be32(record->blocknr); |
622 | offset += 4; | |
623 | *offsetp = offset; | |
624 | } | |
625 | ||
ae6ddcc5 | 626 | /* |
1da177e4 LT |
627 | * Flush a revoke descriptor out to the journal. If we are aborting, |
628 | * this is a noop; otherwise we are generating a buffer which needs to | |
629 | * be waited for during commit, so it has to go onto the appropriate | |
630 | * journal buffer list. | |
631 | */ | |
632 | ||
ae6ddcc5 MC |
633 | static void flush_descriptor(journal_t *journal, |
634 | struct journal_head *descriptor, | |
38d726d1 | 635 | int offset, int write_op) |
1da177e4 LT |
636 | { |
637 | journal_revoke_header_t *header; | |
638 | struct buffer_head *bh = jh2bh(descriptor); | |
639 | ||
640 | if (is_journal_aborted(journal)) { | |
641 | put_bh(bh); | |
642 | return; | |
643 | } | |
644 | ||
645 | header = (journal_revoke_header_t *) jh2bh(descriptor)->b_data; | |
646 | header->r_count = cpu_to_be32(offset); | |
647 | set_buffer_jwrite(bh); | |
648 | BUFFER_TRACE(bh, "write"); | |
649 | set_buffer_dirty(bh); | |
9cb569d6 | 650 | write_dirty_buffer(bh, write_op); |
1da177e4 LT |
651 | } |
652 | #endif | |
653 | ||
ae6ddcc5 | 654 | /* |
1da177e4 LT |
655 | * Revoke support for recovery. |
656 | * | |
657 | * Recovery needs to be able to: | |
658 | * | |
659 | * record all revoke records, including the tid of the latest instance | |
660 | * of each revoke in the journal | |
661 | * | |
662 | * check whether a given block in a given transaction should be replayed | |
663 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
664 | * transaction) | |
ae6ddcc5 | 665 | * |
1da177e4 LT |
666 | * empty the revoke table after recovery. |
667 | */ | |
668 | ||
669 | /* | |
670 | * First, setting revoke records. We create a new revoke record for | |
671 | * every block ever revoked in the log as we scan it for recovery, and | |
672 | * we update the existing records if we find multiple revokes for a | |
ae6ddcc5 | 673 | * single block. |
1da177e4 LT |
674 | */ |
675 | ||
ae6ddcc5 | 676 | int journal_set_revoke(journal_t *journal, |
9c28cbcc | 677 | unsigned int blocknr, |
1da177e4 LT |
678 | tid_t sequence) |
679 | { | |
680 | struct jbd_revoke_record_s *record; | |
681 | ||
682 | record = find_revoke_record(journal, blocknr); | |
683 | if (record) { | |
684 | /* If we have multiple occurrences, only record the | |
685 | * latest sequence number in the hashed record */ | |
686 | if (tid_gt(sequence, record->sequence)) | |
687 | record->sequence = sequence; | |
688 | return 0; | |
ae6ddcc5 | 689 | } |
1da177e4 LT |
690 | return insert_revoke_hash(journal, blocknr, sequence); |
691 | } | |
692 | ||
ae6ddcc5 | 693 | /* |
1da177e4 LT |
694 | * Test revoke records. For a given block referenced in the log, has |
695 | * that block been revoked? A revoke record with a given transaction | |
696 | * sequence number revokes all blocks in that transaction and earlier | |
697 | * ones, but later transactions still need replayed. | |
698 | */ | |
699 | ||
ae6ddcc5 | 700 | int journal_test_revoke(journal_t *journal, |
9c28cbcc | 701 | unsigned int blocknr, |
1da177e4 LT |
702 | tid_t sequence) |
703 | { | |
704 | struct jbd_revoke_record_s *record; | |
705 | ||
706 | record = find_revoke_record(journal, blocknr); | |
707 | if (!record) | |
708 | return 0; | |
709 | if (tid_gt(sequence, record->sequence)) | |
710 | return 0; | |
711 | return 1; | |
712 | } | |
713 | ||
714 | /* | |
715 | * Finally, once recovery is over, we need to clear the revoke table so | |
716 | * that it can be reused by the running filesystem. | |
717 | */ | |
718 | ||
719 | void journal_clear_revoke(journal_t *journal) | |
720 | { | |
721 | int i; | |
722 | struct list_head *hash_list; | |
723 | struct jbd_revoke_record_s *record; | |
724 | struct jbd_revoke_table_s *revoke; | |
725 | ||
726 | revoke = journal->j_revoke; | |
727 | ||
728 | for (i = 0; i < revoke->hash_size; i++) { | |
729 | hash_list = &revoke->hash_table[i]; | |
730 | while (!list_empty(hash_list)) { | |
731 | record = (struct jbd_revoke_record_s*) hash_list->next; | |
732 | list_del(&record->hash); | |
733 | kmem_cache_free(revoke_record_cache, record); | |
734 | } | |
735 | } | |
736 | } |