]>
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 | * | |
1ba37268 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 | * | |
470decc6 DK |
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: | |
61 | * buffer has been revoked. | |
86db97c8 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 |
86db97c8 JK |
79 | * can touch them (filesystem isn't mounted yet) and hence no locking is |
80 | * needed. | |
470decc6 DK |
81 | */ |
82 | ||
83 | #ifndef __KERNEL__ | |
84 | #include "jfs_user.h" | |
85 | #else | |
86 | #include <linux/time.h> | |
87 | #include <linux/fs.h> | |
f7f4bccb | 88 | #include <linux/jbd2.h> |
470decc6 DK |
89 | #include <linux/errno.h> |
90 | #include <linux/slab.h> | |
91 | #include <linux/list.h> | |
470decc6 | 92 | #include <linux/init.h> |
67c457a8 | 93 | #include <linux/bio.h> |
f482394c | 94 | #include <linux/log2.h> |
d48458d4 | 95 | #include <linux/hash.h> |
db9ee220 | 96 | #endif |
470decc6 | 97 | |
e18b890b CL |
98 | static struct kmem_cache *jbd2_revoke_record_cache; |
99 | static struct kmem_cache *jbd2_revoke_table_cache; | |
470decc6 DK |
100 | |
101 | /* Each revoke record represents one single revoked block. During | |
102 | journal replay, this involves recording the transaction ID of the | |
103 | last transaction to revoke this block. */ | |
104 | ||
f7f4bccb | 105 | struct jbd2_revoke_record_s |
470decc6 DK |
106 | { |
107 | struct list_head hash; | |
108 | tid_t sequence; /* Used for recovery only */ | |
18eba7aa | 109 | unsigned long long blocknr; |
470decc6 DK |
110 | }; |
111 | ||
112 | ||
113 | /* The revoke table is just a simple hash table of revoke records. */ | |
f7f4bccb | 114 | struct jbd2_revoke_table_s |
470decc6 DK |
115 | { |
116 | /* It is conceivable that we might want a larger hash table | |
117 | * for recovery. Must be a power of two. */ | |
118 | int hash_size; | |
119 | int hash_shift; | |
120 | struct list_head *hash_table; | |
121 | }; | |
122 | ||
123 | ||
124 | #ifdef __KERNEL__ | |
9bcf976c | 125 | static void write_one_revoke_record(transaction_t *, |
e5a120ae JK |
126 | struct list_head *, |
127 | struct buffer_head **, int *, | |
9bcf976c JK |
128 | struct jbd2_revoke_record_s *); |
129 | static void flush_descriptor(journal_t *, struct buffer_head *, int); | |
470decc6 DK |
130 | #endif |
131 | ||
132 | /* Utility functions to maintain the revoke table */ | |
133 | ||
18eba7aa | 134 | static inline int hash(journal_t *journal, unsigned long long block) |
470decc6 | 135 | { |
d48458d4 | 136 | return hash_64(block, journal->j_revoke->hash_shift); |
470decc6 DK |
137 | } |
138 | ||
18eba7aa | 139 | static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr, |
470decc6 DK |
140 | tid_t seq) |
141 | { | |
142 | struct list_head *hash_list; | |
f7f4bccb | 143 | struct jbd2_revoke_record_s *record; |
7b506b10 | 144 | gfp_t gfp_mask = GFP_NOFS; |
470decc6 | 145 | |
7b506b10 MH |
146 | if (journal_oom_retry) |
147 | gfp_mask |= __GFP_NOFAIL; | |
148 | record = kmem_cache_alloc(jbd2_revoke_record_cache, gfp_mask); | |
470decc6 | 149 | if (!record) |
7b506b10 | 150 | return -ENOMEM; |
470decc6 DK |
151 | |
152 | record->sequence = seq; | |
153 | record->blocknr = blocknr; | |
154 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
155 | spin_lock(&journal->j_revoke_lock); | |
156 | list_add(&record->hash, hash_list); | |
157 | spin_unlock(&journal->j_revoke_lock); | |
158 | return 0; | |
470decc6 DK |
159 | } |
160 | ||
161 | /* Find a revoke record in the journal's hash table. */ | |
162 | ||
f7f4bccb | 163 | static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal, |
18eba7aa | 164 | unsigned long long blocknr) |
470decc6 DK |
165 | { |
166 | struct list_head *hash_list; | |
f7f4bccb | 167 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
168 | |
169 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
170 | ||
171 | spin_lock(&journal->j_revoke_lock); | |
f7f4bccb | 172 | record = (struct jbd2_revoke_record_s *) hash_list->next; |
470decc6 DK |
173 | while (&(record->hash) != hash_list) { |
174 | if (record->blocknr == blocknr) { | |
175 | spin_unlock(&journal->j_revoke_lock); | |
176 | return record; | |
177 | } | |
f7f4bccb | 178 | record = (struct jbd2_revoke_record_s *) record->hash.next; |
470decc6 DK |
179 | } |
180 | spin_unlock(&journal->j_revoke_lock); | |
181 | return NULL; | |
182 | } | |
183 | ||
9fa27c85 DG |
184 | void jbd2_journal_destroy_revoke_caches(void) |
185 | { | |
186 | if (jbd2_revoke_record_cache) { | |
187 | kmem_cache_destroy(jbd2_revoke_record_cache); | |
188 | jbd2_revoke_record_cache = NULL; | |
189 | } | |
190 | if (jbd2_revoke_table_cache) { | |
191 | kmem_cache_destroy(jbd2_revoke_table_cache); | |
192 | jbd2_revoke_table_cache = NULL; | |
193 | } | |
194 | } | |
195 | ||
f7f4bccb | 196 | int __init jbd2_journal_init_revoke_caches(void) |
470decc6 | 197 | { |
9fa27c85 DG |
198 | J_ASSERT(!jbd2_revoke_record_cache); |
199 | J_ASSERT(!jbd2_revoke_table_cache); | |
200 | ||
9c0e00e5 YY |
201 | jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s, |
202 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY); | |
1076d17a | 203 | if (!jbd2_revoke_record_cache) |
9fa27c85 | 204 | goto record_cache_failure; |
470decc6 | 205 | |
9c0e00e5 YY |
206 | jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s, |
207 | SLAB_TEMPORARY); | |
9fa27c85 DG |
208 | if (!jbd2_revoke_table_cache) |
209 | goto table_cache_failure; | |
470decc6 | 210 | return 0; |
9fa27c85 DG |
211 | table_cache_failure: |
212 | jbd2_journal_destroy_revoke_caches(); | |
213 | record_cache_failure: | |
214 | return -ENOMEM; | |
470decc6 DK |
215 | } |
216 | ||
83c49523 | 217 | static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size) |
470decc6 | 218 | { |
83c49523 DG |
219 | int shift = 0; |
220 | int tmp = hash_size; | |
221 | struct jbd2_revoke_table_s *table; | |
470decc6 | 222 | |
83c49523 DG |
223 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
224 | if (!table) | |
225 | goto out; | |
470decc6 | 226 | |
470decc6 DK |
227 | while((tmp >>= 1UL) != 0UL) |
228 | shift++; | |
229 | ||
83c49523 DG |
230 | table->hash_size = hash_size; |
231 | table->hash_shift = shift; | |
232 | table->hash_table = | |
470decc6 | 233 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); |
83c49523 DG |
234 | if (!table->hash_table) { |
235 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
236 | table = NULL; | |
237 | goto out; | |
470decc6 DK |
238 | } |
239 | ||
240 | for (tmp = 0; tmp < hash_size; tmp++) | |
83c49523 | 241 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
470decc6 | 242 | |
83c49523 DG |
243 | out: |
244 | return table; | |
245 | } | |
246 | ||
247 | static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) | |
248 | { | |
249 | int i; | |
250 | struct list_head *hash_list; | |
251 | ||
252 | for (i = 0; i < table->hash_size; i++) { | |
253 | hash_list = &table->hash_table[i]; | |
254 | J_ASSERT(list_empty(hash_list)); | |
470decc6 DK |
255 | } |
256 | ||
83c49523 DG |
257 | kfree(table->hash_table); |
258 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
259 | } | |
470decc6 | 260 | |
83c49523 DG |
261 | /* Initialise the revoke table for a given journal to a given size. */ |
262 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) | |
263 | { | |
264 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
f482394c | 265 | J_ASSERT(is_power_of_2(hash_size)); |
470decc6 | 266 | |
83c49523 DG |
267 | journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size); |
268 | if (!journal->j_revoke_table[0]) | |
269 | goto fail0; | |
470decc6 | 270 | |
83c49523 DG |
271 | journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size); |
272 | if (!journal->j_revoke_table[1]) | |
273 | goto fail1; | |
470decc6 | 274 | |
83c49523 | 275 | journal->j_revoke = journal->j_revoke_table[1]; |
470decc6 DK |
276 | |
277 | spin_lock_init(&journal->j_revoke_lock); | |
278 | ||
279 | return 0; | |
470decc6 | 280 | |
83c49523 DG |
281 | fail1: |
282 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
cd9cb405 | 283 | journal->j_revoke_table[0] = NULL; |
83c49523 DG |
284 | fail0: |
285 | return -ENOMEM; | |
286 | } | |
470decc6 | 287 | |
83c49523 | 288 | /* Destroy a journal's revoke table. The table must already be empty! */ |
f7f4bccb | 289 | void jbd2_journal_destroy_revoke(journal_t *journal) |
470decc6 | 290 | { |
470decc6 | 291 | journal->j_revoke = NULL; |
83c49523 DG |
292 | if (journal->j_revoke_table[0]) |
293 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
294 | if (journal->j_revoke_table[1]) | |
295 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
470decc6 DK |
296 | } |
297 | ||
298 | ||
299 | #ifdef __KERNEL__ | |
300 | ||
301 | /* | |
f7f4bccb | 302 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
470decc6 DK |
303 | * prevents the block from being replayed during recovery if we take a |
304 | * crash after this current transaction commits. Any subsequent | |
305 | * metadata writes of the buffer in this transaction cancel the | |
306 | * revoke. | |
307 | * | |
308 | * Note that this call may block --- it is up to the caller to make | |
309 | * sure that there are no further calls to journal_write_metadata | |
310 | * before the revoke is complete. In ext3, this implies calling the | |
311 | * revoke before clearing the block bitmap when we are deleting | |
312 | * metadata. | |
313 | * | |
f7f4bccb | 314 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
470decc6 DK |
315 | * parameter, but does _not_ forget the buffer_head if the bh was only |
316 | * found implicitly. | |
317 | * | |
318 | * bh_in may not be a journalled buffer - it may have come off | |
319 | * the hash tables without an attached journal_head. | |
320 | * | |
f7f4bccb | 321 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
470decc6 DK |
322 | * by one. |
323 | */ | |
324 | ||
18eba7aa | 325 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
470decc6 DK |
326 | struct buffer_head *bh_in) |
327 | { | |
328 | struct buffer_head *bh = NULL; | |
329 | journal_t *journal; | |
330 | struct block_device *bdev; | |
331 | int err; | |
332 | ||
333 | might_sleep(); | |
334 | if (bh_in) | |
335 | BUFFER_TRACE(bh_in, "enter"); | |
336 | ||
337 | journal = handle->h_transaction->t_journal; | |
f7f4bccb | 338 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
470decc6 DK |
339 | J_ASSERT (!"Cannot set revoke feature!"); |
340 | return -EINVAL; | |
341 | } | |
342 | ||
343 | bdev = journal->j_fs_dev; | |
344 | bh = bh_in; | |
345 | ||
346 | if (!bh) { | |
347 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
348 | if (bh) | |
349 | BUFFER_TRACE(bh, "found on hash"); | |
350 | } | |
cd02ff0b | 351 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
352 | else { |
353 | struct buffer_head *bh2; | |
354 | ||
355 | /* If there is a different buffer_head lying around in | |
356 | * memory anywhere... */ | |
357 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
358 | if (bh2) { | |
359 | /* ... and it has RevokeValid status... */ | |
360 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
361 | /* ...then it better be revoked too, | |
362 | * since it's illegal to create a revoke | |
363 | * record against a buffer_head which is | |
364 | * not marked revoked --- that would | |
365 | * risk missing a subsequent revoke | |
366 | * cancel. */ | |
367 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
368 | put_bh(bh2); | |
369 | } | |
370 | } | |
371 | #endif | |
372 | ||
373 | /* We really ought not ever to revoke twice in a row without | |
374 | first having the revoke cancelled: it's illegal to free a | |
375 | block twice without allocating it in between! */ | |
376 | if (bh) { | |
377 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
378 | "inconsistent data on disk")) { | |
379 | if (!bh_in) | |
380 | brelse(bh); | |
381 | return -EIO; | |
382 | } | |
383 | set_buffer_revoked(bh); | |
384 | set_buffer_revokevalid(bh); | |
385 | if (bh_in) { | |
f7f4bccb MC |
386 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
387 | jbd2_journal_forget(handle, bh_in); | |
470decc6 DK |
388 | } else { |
389 | BUFFER_TRACE(bh, "call brelse"); | |
390 | __brelse(bh); | |
391 | } | |
392 | } | |
393 | ||
29971769 | 394 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
470decc6 DK |
395 | err = insert_revoke_hash(journal, blocknr, |
396 | handle->h_transaction->t_tid); | |
397 | BUFFER_TRACE(bh_in, "exit"); | |
398 | return err; | |
399 | } | |
400 | ||
401 | /* | |
402 | * Cancel an outstanding revoke. For use only internally by the | |
f7f4bccb | 403 | * journaling code (called from jbd2_journal_get_write_access). |
470decc6 DK |
404 | * |
405 | * We trust buffer_revoked() on the buffer if the buffer is already | |
406 | * being journaled: if there is no revoke pending on the buffer, then we | |
407 | * don't do anything here. | |
408 | * | |
409 | * This would break if it were possible for a buffer to be revoked and | |
410 | * discarded, and then reallocated within the same transaction. In such | |
411 | * a case we would have lost the revoked bit, but when we arrived here | |
412 | * the second time we would still have a pending revoke to cancel. So, | |
413 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
414 | * set. | |
470decc6 | 415 | */ |
f7f4bccb | 416 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
470decc6 | 417 | { |
f7f4bccb | 418 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
419 | journal_t *journal = handle->h_transaction->t_journal; |
420 | int need_cancel; | |
421 | int did_revoke = 0; /* akpm: debug */ | |
422 | struct buffer_head *bh = jh2bh(jh); | |
423 | ||
424 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
425 | ||
426 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
427 | * only perform the full cancel if the revoke bit is set. If | |
428 | * not, we can't trust the revoke bit, and we need to do the | |
429 | * full search for a revoke record. */ | |
430 | if (test_set_buffer_revokevalid(bh)) { | |
431 | need_cancel = test_clear_buffer_revoked(bh); | |
432 | } else { | |
433 | need_cancel = 1; | |
434 | clear_buffer_revoked(bh); | |
435 | } | |
436 | ||
437 | if (need_cancel) { | |
438 | record = find_revoke_record(journal, bh->b_blocknr); | |
439 | if (record) { | |
440 | jbd_debug(4, "cancelled existing revoke on " | |
441 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
442 | spin_lock(&journal->j_revoke_lock); | |
443 | list_del(&record->hash); | |
444 | spin_unlock(&journal->j_revoke_lock); | |
f7f4bccb | 445 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
446 | did_revoke = 1; |
447 | } | |
448 | } | |
449 | ||
cd02ff0b | 450 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
451 | /* There better not be one left behind by now! */ |
452 | record = find_revoke_record(journal, bh->b_blocknr); | |
453 | J_ASSERT_JH(jh, record == NULL); | |
454 | #endif | |
455 | ||
456 | /* Finally, have we just cleared revoke on an unhashed | |
457 | * buffer_head? If so, we'd better make sure we clear the | |
458 | * revoked status on any hashed alias too, otherwise the revoke | |
459 | * state machine will get very upset later on. */ | |
460 | if (need_cancel) { | |
461 | struct buffer_head *bh2; | |
462 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
463 | if (bh2) { | |
464 | if (bh2 != bh) | |
465 | clear_buffer_revoked(bh2); | |
466 | __brelse(bh2); | |
467 | } | |
468 | } | |
469 | return did_revoke; | |
470 | } | |
471 | ||
1ba37268 YY |
472 | /* |
473 | * journal_clear_revoked_flag clears revoked flag of buffers in | |
474 | * revoke table to reflect there is no revoked buffers in the next | |
475 | * transaction which is going to be started. | |
476 | */ | |
477 | void jbd2_clear_buffer_revoked_flags(journal_t *journal) | |
478 | { | |
479 | struct jbd2_revoke_table_s *revoke = journal->j_revoke; | |
480 | int i = 0; | |
481 | ||
482 | for (i = 0; i < revoke->hash_size; i++) { | |
483 | struct list_head *hash_list; | |
484 | struct list_head *list_entry; | |
485 | hash_list = &revoke->hash_table[i]; | |
486 | ||
487 | list_for_each(list_entry, hash_list) { | |
488 | struct jbd2_revoke_record_s *record; | |
489 | struct buffer_head *bh; | |
490 | record = (struct jbd2_revoke_record_s *)list_entry; | |
491 | bh = __find_get_block(journal->j_fs_dev, | |
492 | record->blocknr, | |
493 | journal->j_blocksize); | |
494 | if (bh) { | |
495 | clear_buffer_revoked(bh); | |
496 | __brelse(bh); | |
497 | } | |
498 | } | |
499 | } | |
500 | } | |
501 | ||
470decc6 DK |
502 | /* journal_switch_revoke table select j_revoke for next transaction |
503 | * we do not want to suspend any processing until all revokes are | |
504 | * written -bzzz | |
505 | */ | |
f7f4bccb | 506 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
470decc6 DK |
507 | { |
508 | int i; | |
509 | ||
510 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
511 | journal->j_revoke = journal->j_revoke_table[1]; | |
512 | else | |
513 | journal->j_revoke = journal->j_revoke_table[0]; | |
514 | ||
515 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
516 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
517 | } | |
518 | ||
519 | /* | |
520 | * Write revoke records to the journal for all entries in the current | |
521 | * revoke hash, deleting the entries as we go. | |
470decc6 | 522 | */ |
9bcf976c JK |
523 | void jbd2_journal_write_revoke_records(transaction_t *transaction, |
524 | struct list_head *log_bufs) | |
470decc6 | 525 | { |
9bcf976c | 526 | journal_t *journal = transaction->t_journal; |
e5a120ae | 527 | struct buffer_head *descriptor; |
f7f4bccb MC |
528 | struct jbd2_revoke_record_s *record; |
529 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
530 | struct list_head *hash_list; |
531 | int i, offset, count; | |
532 | ||
533 | descriptor = NULL; | |
534 | offset = 0; | |
535 | count = 0; | |
536 | ||
537 | /* select revoke table for committing transaction */ | |
538 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
539 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
540 | ||
541 | for (i = 0; i < revoke->hash_size; i++) { | |
542 | hash_list = &revoke->hash_table[i]; | |
543 | ||
544 | while (!list_empty(hash_list)) { | |
f7f4bccb | 545 | record = (struct jbd2_revoke_record_s *) |
470decc6 | 546 | hash_list->next; |
9bcf976c JK |
547 | write_one_revoke_record(transaction, log_bufs, |
548 | &descriptor, &offset, record); | |
470decc6 DK |
549 | count++; |
550 | list_del(&record->hash); | |
f7f4bccb | 551 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
552 | } |
553 | } | |
554 | if (descriptor) | |
9bcf976c | 555 | flush_descriptor(journal, descriptor, offset); |
470decc6 DK |
556 | jbd_debug(1, "Wrote %d revoke records\n", count); |
557 | } | |
558 | ||
559 | /* | |
560 | * Write out one revoke record. We need to create a new descriptor | |
561 | * block if the old one is full or if we have not already created one. | |
562 | */ | |
563 | ||
9bcf976c | 564 | static void write_one_revoke_record(transaction_t *transaction, |
e5a120ae JK |
565 | struct list_head *log_bufs, |
566 | struct buffer_head **descriptorp, | |
470decc6 | 567 | int *offsetp, |
9bcf976c | 568 | struct jbd2_revoke_record_s *record) |
470decc6 | 569 | { |
9bcf976c | 570 | journal_t *journal = transaction->t_journal; |
42a7106d | 571 | int csum_size = 0; |
e5a120ae | 572 | struct buffer_head *descriptor; |
e531d0bc | 573 | int sz, offset; |
470decc6 DK |
574 | |
575 | /* If we are already aborting, this all becomes a noop. We | |
576 | still need to go round the loop in | |
f7f4bccb | 577 | jbd2_journal_write_revoke_records in order to free all of the |
470decc6 DK |
578 | revoke records: only the IO to the journal is omitted. */ |
579 | if (is_journal_aborted(journal)) | |
580 | return; | |
581 | ||
582 | descriptor = *descriptorp; | |
583 | offset = *offsetp; | |
584 | ||
42a7106d | 585 | /* Do we need to leave space at the end for a checksum? */ |
db9ee220 | 586 | if (jbd2_journal_has_csum_v2or3(journal)) |
1101cd4d | 587 | csum_size = sizeof(struct jbd2_journal_block_tail); |
42a7106d | 588 | |
56316a0d | 589 | if (jbd2_has_feature_64bit(journal)) |
e531d0bc DW |
590 | sz = 8; |
591 | else | |
592 | sz = 4; | |
593 | ||
470decc6 DK |
594 | /* Make sure we have a descriptor with space left for the record */ |
595 | if (descriptor) { | |
e531d0bc | 596 | if (offset + sz > journal->j_blocksize - csum_size) { |
9bcf976c | 597 | flush_descriptor(journal, descriptor, offset); |
470decc6 DK |
598 | descriptor = NULL; |
599 | } | |
600 | } | |
601 | ||
602 | if (!descriptor) { | |
32ab6715 JK |
603 | descriptor = jbd2_journal_get_descriptor_buffer(transaction, |
604 | JBD2_REVOKE_BLOCK); | |
470decc6 DK |
605 | if (!descriptor) |
606 | return; | |
470decc6 DK |
607 | |
608 | /* Record it so that we can wait for IO completion later */ | |
e5a120ae JK |
609 | BUFFER_TRACE(descriptor, "file in log_bufs"); |
610 | jbd2_file_log_bh(log_bufs, descriptor); | |
470decc6 | 611 | |
f7f4bccb | 612 | offset = sizeof(jbd2_journal_revoke_header_t); |
470decc6 DK |
613 | *descriptorp = descriptor; |
614 | } | |
615 | ||
56316a0d | 616 | if (jbd2_has_feature_64bit(journal)) |
e5a120ae | 617 | * ((__be64 *)(&descriptor->b_data[offset])) = |
b517bea1 | 618 | cpu_to_be64(record->blocknr); |
e531d0bc | 619 | else |
e5a120ae | 620 | * ((__be32 *)(&descriptor->b_data[offset])) = |
b517bea1 | 621 | cpu_to_be32(record->blocknr); |
e531d0bc | 622 | offset += sz; |
b517bea1 | 623 | |
470decc6 DK |
624 | *offsetp = offset; |
625 | } | |
626 | ||
627 | /* | |
628 | * Flush a revoke descriptor out to the journal. If we are aborting, | |
629 | * this is a noop; otherwise we are generating a buffer which needs to | |
630 | * be waited for during commit, so it has to go onto the appropriate | |
631 | * journal buffer list. | |
632 | */ | |
633 | ||
634 | static void flush_descriptor(journal_t *journal, | |
e5a120ae | 635 | struct buffer_head *descriptor, |
9bcf976c | 636 | int offset) |
470decc6 | 637 | { |
f7f4bccb | 638 | jbd2_journal_revoke_header_t *header; |
470decc6 DK |
639 | |
640 | if (is_journal_aborted(journal)) { | |
e5a120ae | 641 | put_bh(descriptor); |
470decc6 DK |
642 | return; |
643 | } | |
644 | ||
e5a120ae | 645 | header = (jbd2_journal_revoke_header_t *)descriptor->b_data; |
470decc6 | 646 | header->r_count = cpu_to_be32(offset); |
1101cd4d | 647 | jbd2_descriptor_block_csum_set(journal, descriptor); |
42a7106d | 648 | |
e5a120ae JK |
649 | set_buffer_jwrite(descriptor); |
650 | BUFFER_TRACE(descriptor, "write"); | |
651 | set_buffer_dirty(descriptor); | |
70fd7614 | 652 | write_dirty_buffer(descriptor, REQ_SYNC); |
470decc6 DK |
653 | } |
654 | #endif | |
655 | ||
656 | /* | |
657 | * Revoke support for recovery. | |
658 | * | |
659 | * Recovery needs to be able to: | |
660 | * | |
661 | * record all revoke records, including the tid of the latest instance | |
662 | * of each revoke in the journal | |
663 | * | |
664 | * check whether a given block in a given transaction should be replayed | |
665 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
666 | * transaction) | |
667 | * | |
668 | * empty the revoke table after recovery. | |
669 | */ | |
670 | ||
671 | /* | |
672 | * First, setting revoke records. We create a new revoke record for | |
673 | * every block ever revoked in the log as we scan it for recovery, and | |
674 | * we update the existing records if we find multiple revokes for a | |
675 | * single block. | |
676 | */ | |
677 | ||
f7f4bccb | 678 | int jbd2_journal_set_revoke(journal_t *journal, |
18eba7aa | 679 | unsigned long long blocknr, |
470decc6 DK |
680 | tid_t sequence) |
681 | { | |
f7f4bccb | 682 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
683 | |
684 | record = find_revoke_record(journal, blocknr); | |
685 | if (record) { | |
686 | /* If we have multiple occurrences, only record the | |
687 | * latest sequence number in the hashed record */ | |
688 | if (tid_gt(sequence, record->sequence)) | |
689 | record->sequence = sequence; | |
690 | return 0; | |
691 | } | |
692 | return insert_revoke_hash(journal, blocknr, sequence); | |
693 | } | |
694 | ||
695 | /* | |
696 | * Test revoke records. For a given block referenced in the log, has | |
697 | * that block been revoked? A revoke record with a given transaction | |
698 | * sequence number revokes all blocks in that transaction and earlier | |
699 | * ones, but later transactions still need replayed. | |
700 | */ | |
701 | ||
f7f4bccb | 702 | int jbd2_journal_test_revoke(journal_t *journal, |
18eba7aa | 703 | unsigned long long blocknr, |
470decc6 DK |
704 | tid_t sequence) |
705 | { | |
f7f4bccb | 706 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
707 | |
708 | record = find_revoke_record(journal, blocknr); | |
709 | if (!record) | |
710 | return 0; | |
711 | if (tid_gt(sequence, record->sequence)) | |
712 | return 0; | |
713 | return 1; | |
714 | } | |
715 | ||
716 | /* | |
717 | * Finally, once recovery is over, we need to clear the revoke table so | |
718 | * that it can be reused by the running filesystem. | |
719 | */ | |
720 | ||
f7f4bccb | 721 | void jbd2_journal_clear_revoke(journal_t *journal) |
470decc6 DK |
722 | { |
723 | int i; | |
724 | struct list_head *hash_list; | |
f7f4bccb MC |
725 | struct jbd2_revoke_record_s *record; |
726 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
727 | |
728 | revoke = journal->j_revoke; | |
729 | ||
730 | for (i = 0; i < revoke->hash_size; i++) { | |
731 | hash_list = &revoke->hash_table[i]; | |
732 | while (!list_empty(hash_list)) { | |
f7f4bccb | 733 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
470decc6 | 734 | list_del(&record->hash); |
f7f4bccb | 735 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
736 | } |
737 | } | |
738 | } |