Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
58862699 | 2 | * linux/fs/jbd2/revoke.c |
470decc6 DK |
3 | * |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 2000 | |
5 | * | |
6 | * Copyright 2000 Red Hat corp --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
12 | * Journal revoke routines for the generic filesystem journaling code; | |
13 | * part of the ext2fs journaling system. | |
14 | * | |
15 | * Revoke is the mechanism used to prevent old log records for deleted | |
16 | * metadata from being replayed on top of newer data using the same | |
17 | * blocks. The revoke mechanism is used in two separate places: | |
18 | * | |
19 | * + Commit: during commit we write the entire list of the current | |
20 | * transaction's revoked blocks to the journal | |
21 | * | |
22 | * + Recovery: during recovery we record the transaction ID of all | |
23 | * revoked blocks. If there are multiple revoke records in the log | |
24 | * for a single block, only the last one counts, and if there is a log | |
25 | * entry for a block beyond the last revoke, then that log entry still | |
26 | * gets replayed. | |
27 | * | |
28 | * We can get interactions between revokes and new log data within a | |
29 | * single transaction: | |
30 | * | |
31 | * Block is revoked and then journaled: | |
32 | * The desired end result is the journaling of the new block, so we | |
33 | * cancel the revoke before the transaction commits. | |
34 | * | |
35 | * Block is journaled and then revoked: | |
36 | * The revoke must take precedence over the write of the block, so we | |
37 | * need either to cancel the journal entry or to write the revoke | |
38 | * later in the log than the log block. In this case, we choose the | |
39 | * latter: journaling a block cancels any revoke record for that block | |
40 | * in the current transaction, so any revoke for that block in the | |
41 | * transaction must have happened after the block was journaled and so | |
42 | * the revoke must take precedence. | |
43 | * | |
44 | * Block is revoked and then written as data: | |
45 | * The data write is allowed to succeed, but the revoke is _not_ | |
46 | * cancelled. We still need to prevent old log records from | |
47 | * overwriting the new data. We don't even need to clear the revoke | |
48 | * bit here. | |
49 | * | |
50 | * 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: | |
57 | * buffer has been revoked. | |
58 | */ | |
59 | ||
60 | #ifndef __KERNEL__ | |
61 | #include "jfs_user.h" | |
62 | #else | |
63 | #include <linux/time.h> | |
64 | #include <linux/fs.h> | |
f7f4bccb | 65 | #include <linux/jbd2.h> |
470decc6 DK |
66 | #include <linux/errno.h> |
67 | #include <linux/slab.h> | |
68 | #include <linux/list.h> | |
470decc6 DK |
69 | #include <linux/init.h> |
70 | #endif | |
f482394c | 71 | #include <linux/log2.h> |
470decc6 | 72 | |
e18b890b CL |
73 | static struct kmem_cache *jbd2_revoke_record_cache; |
74 | static struct kmem_cache *jbd2_revoke_table_cache; | |
470decc6 DK |
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 | ||
f7f4bccb | 80 | struct jbd2_revoke_record_s |
470decc6 DK |
81 | { |
82 | struct list_head hash; | |
83 | tid_t sequence; /* Used for recovery only */ | |
18eba7aa | 84 | unsigned long long blocknr; |
470decc6 DK |
85 | }; |
86 | ||
87 | ||
88 | /* The revoke table is just a simple hash table of revoke records. */ | |
f7f4bccb | 89 | struct jbd2_revoke_table_s |
470decc6 DK |
90 | { |
91 | /* It is conceivable that we might want a larger hash table | |
92 | * for recovery. Must be a power of two. */ | |
93 | int hash_size; | |
94 | int hash_shift; | |
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 *, | |
f7f4bccb | 102 | struct jbd2_revoke_record_s *); |
470decc6 DK |
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 */ | |
18eba7aa | 109 | static inline int hash(journal_t *journal, unsigned long long block) |
470decc6 | 110 | { |
f7f4bccb | 111 | struct jbd2_revoke_table_s *table = journal->j_revoke; |
470decc6 | 112 | int hash_shift = table->hash_shift; |
29971769 | 113 | int hash = (int)block ^ (int)((block >> 31) >> 1); |
470decc6 | 114 | |
29971769 MC |
115 | return ((hash << (hash_shift - 6)) ^ |
116 | (hash >> 13) ^ | |
117 | (hash << (hash_shift - 12))) & (table->hash_size - 1); | |
470decc6 DK |
118 | } |
119 | ||
18eba7aa | 120 | static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr, |
470decc6 DK |
121 | tid_t seq) |
122 | { | |
123 | struct list_head *hash_list; | |
f7f4bccb | 124 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
125 | |
126 | repeat: | |
f7f4bccb | 127 | record = kmem_cache_alloc(jbd2_revoke_record_cache, GFP_NOFS); |
470decc6 DK |
128 | if (!record) |
129 | goto oom; | |
130 | ||
131 | record->sequence = seq; | |
132 | record->blocknr = blocknr; | |
133 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
134 | spin_lock(&journal->j_revoke_lock); | |
135 | list_add(&record->hash, hash_list); | |
136 | spin_unlock(&journal->j_revoke_lock); | |
137 | return 0; | |
138 | ||
139 | oom: | |
140 | if (!journal_oom_retry) | |
141 | return -ENOMEM; | |
142 | jbd_debug(1, "ENOMEM in %s, retrying\n", __FUNCTION__); | |
143 | yield(); | |
144 | goto repeat; | |
145 | } | |
146 | ||
147 | /* Find a revoke record in the journal's hash table. */ | |
148 | ||
f7f4bccb | 149 | static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal, |
18eba7aa | 150 | unsigned long long blocknr) |
470decc6 DK |
151 | { |
152 | struct list_head *hash_list; | |
f7f4bccb | 153 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
154 | |
155 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
156 | ||
157 | spin_lock(&journal->j_revoke_lock); | |
f7f4bccb | 158 | record = (struct jbd2_revoke_record_s *) hash_list->next; |
470decc6 DK |
159 | while (&(record->hash) != hash_list) { |
160 | if (record->blocknr == blocknr) { | |
161 | spin_unlock(&journal->j_revoke_lock); | |
162 | return record; | |
163 | } | |
f7f4bccb | 164 | record = (struct jbd2_revoke_record_s *) record->hash.next; |
470decc6 DK |
165 | } |
166 | spin_unlock(&journal->j_revoke_lock); | |
167 | return NULL; | |
168 | } | |
169 | ||
9fa27c85 DG |
170 | void jbd2_journal_destroy_revoke_caches(void) |
171 | { | |
172 | if (jbd2_revoke_record_cache) { | |
173 | kmem_cache_destroy(jbd2_revoke_record_cache); | |
174 | jbd2_revoke_record_cache = NULL; | |
175 | } | |
176 | if (jbd2_revoke_table_cache) { | |
177 | kmem_cache_destroy(jbd2_revoke_table_cache); | |
178 | jbd2_revoke_table_cache = NULL; | |
179 | } | |
180 | } | |
181 | ||
f7f4bccb | 182 | int __init jbd2_journal_init_revoke_caches(void) |
470decc6 | 183 | { |
9fa27c85 DG |
184 | J_ASSERT(!jbd2_revoke_record_cache); |
185 | J_ASSERT(!jbd2_revoke_table_cache); | |
186 | ||
a920e941 | 187 | jbd2_revoke_record_cache = kmem_cache_create("jbd2_revoke_record", |
f7f4bccb | 188 | sizeof(struct jbd2_revoke_record_s), |
77160957 MC |
189 | 0, |
190 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY, | |
191 | NULL); | |
1076d17a | 192 | if (!jbd2_revoke_record_cache) |
9fa27c85 | 193 | goto record_cache_failure; |
470decc6 | 194 | |
a920e941 | 195 | jbd2_revoke_table_cache = kmem_cache_create("jbd2_revoke_table", |
f7f4bccb | 196 | sizeof(struct jbd2_revoke_table_s), |
77160957 | 197 | 0, SLAB_TEMPORARY, NULL); |
9fa27c85 DG |
198 | if (!jbd2_revoke_table_cache) |
199 | goto table_cache_failure; | |
470decc6 | 200 | return 0; |
9fa27c85 DG |
201 | table_cache_failure: |
202 | jbd2_journal_destroy_revoke_caches(); | |
203 | record_cache_failure: | |
204 | return -ENOMEM; | |
470decc6 DK |
205 | } |
206 | ||
207 | /* Initialise the revoke table for a given journal to a given size. */ | |
208 | ||
f7f4bccb | 209 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) |
470decc6 DK |
210 | { |
211 | int shift, tmp; | |
212 | ||
213 | J_ASSERT (journal->j_revoke_table[0] == NULL); | |
214 | ||
215 | shift = 0; | |
216 | tmp = hash_size; | |
217 | while((tmp >>= 1UL) != 0UL) | |
218 | shift++; | |
219 | ||
f7f4bccb | 220 | journal->j_revoke_table[0] = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
470decc6 DK |
221 | if (!journal->j_revoke_table[0]) |
222 | return -ENOMEM; | |
223 | journal->j_revoke = journal->j_revoke_table[0]; | |
224 | ||
225 | /* Check that the hash_size is a power of two */ | |
f482394c | 226 | J_ASSERT(is_power_of_2(hash_size)); |
470decc6 DK |
227 | |
228 | journal->j_revoke->hash_size = hash_size; | |
229 | ||
230 | journal->j_revoke->hash_shift = shift; | |
231 | ||
232 | journal->j_revoke->hash_table = | |
233 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); | |
234 | if (!journal->j_revoke->hash_table) { | |
f7f4bccb | 235 | kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[0]); |
470decc6 DK |
236 | journal->j_revoke = NULL; |
237 | return -ENOMEM; | |
238 | } | |
239 | ||
240 | for (tmp = 0; tmp < hash_size; tmp++) | |
241 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]); | |
242 | ||
f7f4bccb | 243 | journal->j_revoke_table[1] = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
470decc6 DK |
244 | if (!journal->j_revoke_table[1]) { |
245 | kfree(journal->j_revoke_table[0]->hash_table); | |
f7f4bccb | 246 | kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[0]); |
470decc6 DK |
247 | return -ENOMEM; |
248 | } | |
249 | ||
250 | journal->j_revoke = journal->j_revoke_table[1]; | |
251 | ||
252 | /* Check that the hash_size is a power of two */ | |
f482394c | 253 | J_ASSERT(is_power_of_2(hash_size)); |
470decc6 DK |
254 | |
255 | journal->j_revoke->hash_size = hash_size; | |
256 | ||
257 | journal->j_revoke->hash_shift = shift; | |
258 | ||
259 | journal->j_revoke->hash_table = | |
260 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); | |
261 | if (!journal->j_revoke->hash_table) { | |
262 | kfree(journal->j_revoke_table[0]->hash_table); | |
f7f4bccb MC |
263 | kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[0]); |
264 | kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[1]); | |
470decc6 DK |
265 | journal->j_revoke = NULL; |
266 | return -ENOMEM; | |
267 | } | |
268 | ||
269 | for (tmp = 0; tmp < hash_size; tmp++) | |
270 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]); | |
271 | ||
272 | spin_lock_init(&journal->j_revoke_lock); | |
273 | ||
274 | return 0; | |
275 | } | |
276 | ||
277 | /* Destoy a journal's revoke table. The table must already be empty! */ | |
278 | ||
f7f4bccb | 279 | void jbd2_journal_destroy_revoke(journal_t *journal) |
470decc6 | 280 | { |
f7f4bccb | 281 | struct jbd2_revoke_table_s *table; |
470decc6 DK |
282 | struct list_head *hash_list; |
283 | int i; | |
284 | ||
285 | table = journal->j_revoke_table[0]; | |
286 | if (!table) | |
287 | return; | |
288 | ||
289 | for (i=0; i<table->hash_size; i++) { | |
290 | hash_list = &table->hash_table[i]; | |
291 | J_ASSERT (list_empty(hash_list)); | |
292 | } | |
293 | ||
294 | kfree(table->hash_table); | |
f7f4bccb | 295 | kmem_cache_free(jbd2_revoke_table_cache, table); |
470decc6 DK |
296 | journal->j_revoke = NULL; |
297 | ||
298 | table = journal->j_revoke_table[1]; | |
299 | if (!table) | |
300 | return; | |
301 | ||
302 | for (i=0; i<table->hash_size; i++) { | |
303 | hash_list = &table->hash_table[i]; | |
304 | J_ASSERT (list_empty(hash_list)); | |
305 | } | |
306 | ||
307 | kfree(table->hash_table); | |
f7f4bccb | 308 | kmem_cache_free(jbd2_revoke_table_cache, table); |
470decc6 DK |
309 | journal->j_revoke = NULL; |
310 | } | |
311 | ||
312 | ||
313 | #ifdef __KERNEL__ | |
314 | ||
315 | /* | |
f7f4bccb | 316 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
470decc6 DK |
317 | * prevents the block from being replayed during recovery if we take a |
318 | * crash after this current transaction commits. Any subsequent | |
319 | * metadata writes of the buffer in this transaction cancel the | |
320 | * revoke. | |
321 | * | |
322 | * Note that this call may block --- it is up to the caller to make | |
323 | * sure that there are no further calls to journal_write_metadata | |
324 | * before the revoke is complete. In ext3, this implies calling the | |
325 | * revoke before clearing the block bitmap when we are deleting | |
326 | * metadata. | |
327 | * | |
f7f4bccb | 328 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
470decc6 DK |
329 | * parameter, but does _not_ forget the buffer_head if the bh was only |
330 | * found implicitly. | |
331 | * | |
332 | * bh_in may not be a journalled buffer - it may have come off | |
333 | * the hash tables without an attached journal_head. | |
334 | * | |
f7f4bccb | 335 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
470decc6 DK |
336 | * by one. |
337 | */ | |
338 | ||
18eba7aa | 339 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
470decc6 DK |
340 | struct buffer_head *bh_in) |
341 | { | |
342 | struct buffer_head *bh = NULL; | |
343 | journal_t *journal; | |
344 | struct block_device *bdev; | |
345 | int err; | |
346 | ||
347 | might_sleep(); | |
348 | if (bh_in) | |
349 | BUFFER_TRACE(bh_in, "enter"); | |
350 | ||
351 | journal = handle->h_transaction->t_journal; | |
f7f4bccb | 352 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
470decc6 DK |
353 | J_ASSERT (!"Cannot set revoke feature!"); |
354 | return -EINVAL; | |
355 | } | |
356 | ||
357 | bdev = journal->j_fs_dev; | |
358 | bh = bh_in; | |
359 | ||
360 | if (!bh) { | |
361 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
362 | if (bh) | |
363 | BUFFER_TRACE(bh, "found on hash"); | |
364 | } | |
cd02ff0b | 365 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
366 | else { |
367 | struct buffer_head *bh2; | |
368 | ||
369 | /* If there is a different buffer_head lying around in | |
370 | * memory anywhere... */ | |
371 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
372 | if (bh2) { | |
373 | /* ... and it has RevokeValid status... */ | |
374 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
375 | /* ...then it better be revoked too, | |
376 | * since it's illegal to create a revoke | |
377 | * record against a buffer_head which is | |
378 | * not marked revoked --- that would | |
379 | * risk missing a subsequent revoke | |
380 | * cancel. */ | |
381 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
382 | put_bh(bh2); | |
383 | } | |
384 | } | |
385 | #endif | |
386 | ||
387 | /* We really ought not ever to revoke twice in a row without | |
388 | first having the revoke cancelled: it's illegal to free a | |
389 | block twice without allocating it in between! */ | |
390 | if (bh) { | |
391 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
392 | "inconsistent data on disk")) { | |
393 | if (!bh_in) | |
394 | brelse(bh); | |
395 | return -EIO; | |
396 | } | |
397 | set_buffer_revoked(bh); | |
398 | set_buffer_revokevalid(bh); | |
399 | if (bh_in) { | |
f7f4bccb MC |
400 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
401 | jbd2_journal_forget(handle, bh_in); | |
470decc6 DK |
402 | } else { |
403 | BUFFER_TRACE(bh, "call brelse"); | |
404 | __brelse(bh); | |
405 | } | |
406 | } | |
407 | ||
29971769 | 408 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
470decc6 DK |
409 | err = insert_revoke_hash(journal, blocknr, |
410 | handle->h_transaction->t_tid); | |
411 | BUFFER_TRACE(bh_in, "exit"); | |
412 | return err; | |
413 | } | |
414 | ||
415 | /* | |
416 | * Cancel an outstanding revoke. For use only internally by the | |
f7f4bccb | 417 | * journaling code (called from jbd2_journal_get_write_access). |
470decc6 DK |
418 | * |
419 | * We trust buffer_revoked() on the buffer if the buffer is already | |
420 | * being journaled: if there is no revoke pending on the buffer, then we | |
421 | * don't do anything here. | |
422 | * | |
423 | * This would break if it were possible for a buffer to be revoked and | |
424 | * discarded, and then reallocated within the same transaction. In such | |
425 | * a case we would have lost the revoked bit, but when we arrived here | |
426 | * the second time we would still have a pending revoke to cancel. So, | |
427 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
428 | * set. | |
429 | * | |
430 | * The caller must have the journal locked. | |
431 | */ | |
f7f4bccb | 432 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
470decc6 | 433 | { |
f7f4bccb | 434 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
435 | journal_t *journal = handle->h_transaction->t_journal; |
436 | int need_cancel; | |
437 | int did_revoke = 0; /* akpm: debug */ | |
438 | struct buffer_head *bh = jh2bh(jh); | |
439 | ||
440 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
441 | ||
442 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
443 | * only perform the full cancel if the revoke bit is set. If | |
444 | * not, we can't trust the revoke bit, and we need to do the | |
445 | * full search for a revoke record. */ | |
446 | if (test_set_buffer_revokevalid(bh)) { | |
447 | need_cancel = test_clear_buffer_revoked(bh); | |
448 | } else { | |
449 | need_cancel = 1; | |
450 | clear_buffer_revoked(bh); | |
451 | } | |
452 | ||
453 | if (need_cancel) { | |
454 | record = find_revoke_record(journal, bh->b_blocknr); | |
455 | if (record) { | |
456 | jbd_debug(4, "cancelled existing revoke on " | |
457 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
458 | spin_lock(&journal->j_revoke_lock); | |
459 | list_del(&record->hash); | |
460 | spin_unlock(&journal->j_revoke_lock); | |
f7f4bccb | 461 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
462 | did_revoke = 1; |
463 | } | |
464 | } | |
465 | ||
cd02ff0b | 466 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
467 | /* There better not be one left behind by now! */ |
468 | record = find_revoke_record(journal, bh->b_blocknr); | |
469 | J_ASSERT_JH(jh, record == NULL); | |
470 | #endif | |
471 | ||
472 | /* Finally, have we just cleared revoke on an unhashed | |
473 | * buffer_head? If so, we'd better make sure we clear the | |
474 | * revoked status on any hashed alias too, otherwise the revoke | |
475 | * state machine will get very upset later on. */ | |
476 | if (need_cancel) { | |
477 | struct buffer_head *bh2; | |
478 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
479 | if (bh2) { | |
480 | if (bh2 != bh) | |
481 | clear_buffer_revoked(bh2); | |
482 | __brelse(bh2); | |
483 | } | |
484 | } | |
485 | return did_revoke; | |
486 | } | |
487 | ||
488 | /* journal_switch_revoke table select j_revoke for next transaction | |
489 | * we do not want to suspend any processing until all revokes are | |
490 | * written -bzzz | |
491 | */ | |
f7f4bccb | 492 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
470decc6 DK |
493 | { |
494 | int i; | |
495 | ||
496 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
497 | journal->j_revoke = journal->j_revoke_table[1]; | |
498 | else | |
499 | journal->j_revoke = journal->j_revoke_table[0]; | |
500 | ||
501 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
502 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
503 | } | |
504 | ||
505 | /* | |
506 | * Write revoke records to the journal for all entries in the current | |
507 | * revoke hash, deleting the entries as we go. | |
508 | * | |
509 | * Called with the journal lock held. | |
510 | */ | |
511 | ||
f7f4bccb | 512 | void jbd2_journal_write_revoke_records(journal_t *journal, |
470decc6 DK |
513 | transaction_t *transaction) |
514 | { | |
515 | struct journal_head *descriptor; | |
f7f4bccb MC |
516 | struct jbd2_revoke_record_s *record; |
517 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
518 | struct list_head *hash_list; |
519 | int i, offset, count; | |
520 | ||
521 | descriptor = NULL; | |
522 | offset = 0; | |
523 | count = 0; | |
524 | ||
525 | /* select revoke table for committing transaction */ | |
526 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
527 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
528 | ||
529 | for (i = 0; i < revoke->hash_size; i++) { | |
530 | hash_list = &revoke->hash_table[i]; | |
531 | ||
532 | while (!list_empty(hash_list)) { | |
f7f4bccb | 533 | record = (struct jbd2_revoke_record_s *) |
470decc6 DK |
534 | hash_list->next; |
535 | write_one_revoke_record(journal, transaction, | |
536 | &descriptor, &offset, | |
537 | record); | |
538 | count++; | |
539 | list_del(&record->hash); | |
f7f4bccb | 540 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
541 | } |
542 | } | |
543 | if (descriptor) | |
544 | flush_descriptor(journal, descriptor, offset); | |
545 | jbd_debug(1, "Wrote %d revoke records\n", count); | |
546 | } | |
547 | ||
548 | /* | |
549 | * Write out one revoke record. We need to create a new descriptor | |
550 | * block if the old one is full or if we have not already created one. | |
551 | */ | |
552 | ||
553 | static void write_one_revoke_record(journal_t *journal, | |
554 | transaction_t *transaction, | |
555 | struct journal_head **descriptorp, | |
556 | int *offsetp, | |
f7f4bccb | 557 | struct jbd2_revoke_record_s *record) |
470decc6 DK |
558 | { |
559 | struct journal_head *descriptor; | |
560 | int offset; | |
561 | journal_header_t *header; | |
562 | ||
563 | /* If we are already aborting, this all becomes a noop. We | |
564 | still need to go round the loop in | |
f7f4bccb | 565 | jbd2_journal_write_revoke_records in order to free all of the |
470decc6 DK |
566 | revoke records: only the IO to the journal is omitted. */ |
567 | if (is_journal_aborted(journal)) | |
568 | return; | |
569 | ||
570 | descriptor = *descriptorp; | |
571 | offset = *offsetp; | |
572 | ||
573 | /* Make sure we have a descriptor with space left for the record */ | |
574 | if (descriptor) { | |
575 | if (offset == journal->j_blocksize) { | |
576 | flush_descriptor(journal, descriptor, offset); | |
577 | descriptor = NULL; | |
578 | } | |
579 | } | |
580 | ||
581 | if (!descriptor) { | |
f7f4bccb | 582 | descriptor = jbd2_journal_get_descriptor_buffer(journal); |
470decc6 DK |
583 | if (!descriptor) |
584 | return; | |
585 | header = (journal_header_t *) &jh2bh(descriptor)->b_data[0]; | |
f7f4bccb MC |
586 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); |
587 | header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK); | |
470decc6 DK |
588 | header->h_sequence = cpu_to_be32(transaction->t_tid); |
589 | ||
590 | /* Record it so that we can wait for IO completion later */ | |
591 | JBUFFER_TRACE(descriptor, "file as BJ_LogCtl"); | |
f7f4bccb | 592 | jbd2_journal_file_buffer(descriptor, transaction, BJ_LogCtl); |
470decc6 | 593 | |
f7f4bccb | 594 | offset = sizeof(jbd2_journal_revoke_header_t); |
470decc6 DK |
595 | *descriptorp = descriptor; |
596 | } | |
597 | ||
b517bea1 ZB |
598 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) { |
599 | * ((__be64 *)(&jh2bh(descriptor)->b_data[offset])) = | |
600 | cpu_to_be64(record->blocknr); | |
601 | offset += 8; | |
602 | ||
603 | } else { | |
604 | * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) = | |
605 | cpu_to_be32(record->blocknr); | |
606 | offset += 4; | |
607 | } | |
608 | ||
470decc6 DK |
609 | *offsetp = offset; |
610 | } | |
611 | ||
612 | /* | |
613 | * Flush a revoke descriptor out to the journal. If we are aborting, | |
614 | * this is a noop; otherwise we are generating a buffer which needs to | |
615 | * be waited for during commit, so it has to go onto the appropriate | |
616 | * journal buffer list. | |
617 | */ | |
618 | ||
619 | static void flush_descriptor(journal_t *journal, | |
620 | struct journal_head *descriptor, | |
621 | int offset) | |
622 | { | |
f7f4bccb | 623 | jbd2_journal_revoke_header_t *header; |
470decc6 DK |
624 | struct buffer_head *bh = jh2bh(descriptor); |
625 | ||
626 | if (is_journal_aborted(journal)) { | |
627 | put_bh(bh); | |
628 | return; | |
629 | } | |
630 | ||
f7f4bccb | 631 | header = (jbd2_journal_revoke_header_t *) jh2bh(descriptor)->b_data; |
470decc6 DK |
632 | header->r_count = cpu_to_be32(offset); |
633 | set_buffer_jwrite(bh); | |
634 | BUFFER_TRACE(bh, "write"); | |
635 | set_buffer_dirty(bh); | |
636 | ll_rw_block(SWRITE, 1, &bh); | |
637 | } | |
638 | #endif | |
639 | ||
640 | /* | |
641 | * Revoke support for recovery. | |
642 | * | |
643 | * Recovery needs to be able to: | |
644 | * | |
645 | * record all revoke records, including the tid of the latest instance | |
646 | * of each revoke in the journal | |
647 | * | |
648 | * check whether a given block in a given transaction should be replayed | |
649 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
650 | * transaction) | |
651 | * | |
652 | * empty the revoke table after recovery. | |
653 | */ | |
654 | ||
655 | /* | |
656 | * First, setting revoke records. We create a new revoke record for | |
657 | * every block ever revoked in the log as we scan it for recovery, and | |
658 | * we update the existing records if we find multiple revokes for a | |
659 | * single block. | |
660 | */ | |
661 | ||
f7f4bccb | 662 | int jbd2_journal_set_revoke(journal_t *journal, |
18eba7aa | 663 | unsigned long long blocknr, |
470decc6 DK |
664 | tid_t sequence) |
665 | { | |
f7f4bccb | 666 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
667 | |
668 | record = find_revoke_record(journal, blocknr); | |
669 | if (record) { | |
670 | /* If we have multiple occurrences, only record the | |
671 | * latest sequence number in the hashed record */ | |
672 | if (tid_gt(sequence, record->sequence)) | |
673 | record->sequence = sequence; | |
674 | return 0; | |
675 | } | |
676 | return insert_revoke_hash(journal, blocknr, sequence); | |
677 | } | |
678 | ||
679 | /* | |
680 | * Test revoke records. For a given block referenced in the log, has | |
681 | * that block been revoked? A revoke record with a given transaction | |
682 | * sequence number revokes all blocks in that transaction and earlier | |
683 | * ones, but later transactions still need replayed. | |
684 | */ | |
685 | ||
f7f4bccb | 686 | int jbd2_journal_test_revoke(journal_t *journal, |
18eba7aa | 687 | unsigned long long blocknr, |
470decc6 DK |
688 | tid_t sequence) |
689 | { | |
f7f4bccb | 690 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
691 | |
692 | record = find_revoke_record(journal, blocknr); | |
693 | if (!record) | |
694 | return 0; | |
695 | if (tid_gt(sequence, record->sequence)) | |
696 | return 0; | |
697 | return 1; | |
698 | } | |
699 | ||
700 | /* | |
701 | * Finally, once recovery is over, we need to clear the revoke table so | |
702 | * that it can be reused by the running filesystem. | |
703 | */ | |
704 | ||
f7f4bccb | 705 | void jbd2_journal_clear_revoke(journal_t *journal) |
470decc6 DK |
706 | { |
707 | int i; | |
708 | struct list_head *hash_list; | |
f7f4bccb MC |
709 | struct jbd2_revoke_record_s *record; |
710 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
711 | |
712 | revoke = journal->j_revoke; | |
713 | ||
714 | for (i = 0; i < revoke->hash_size; i++) { | |
715 | hash_list = &revoke->hash_table[i]; | |
716 | while (!list_empty(hash_list)) { | |
f7f4bccb | 717 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
470decc6 | 718 | list_del(&record->hash); |
f7f4bccb | 719 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
720 | } |
721 | } | |
722 | } |