Commit | Line | Data |
---|---|---|
f5166768 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
470decc6 | 2 | /* |
58862699 | 3 | * linux/fs/jbd2/revoke.c |
470decc6 DK |
4 | * |
5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 2000 | |
6 | * | |
7 | * Copyright 2000 Red Hat corp --- All Rights Reserved | |
8 | * | |
470decc6 DK |
9 | * Journal revoke routines for the generic filesystem journaling code; |
10 | * part of the ext2fs journaling system. | |
11 | * | |
12 | * Revoke is the mechanism used to prevent old log records for deleted | |
13 | * metadata from being replayed on top of newer data using the same | |
14 | * blocks. The revoke mechanism is used in two separate places: | |
15 | * | |
16 | * + Commit: during commit we write the entire list of the current | |
17 | * transaction's revoked blocks to the journal | |
18 | * | |
19 | * + Recovery: during recovery we record the transaction ID of all | |
20 | * revoked blocks. If there are multiple revoke records in the log | |
21 | * for a single block, only the last one counts, and if there is a log | |
22 | * entry for a block beyond the last revoke, then that log entry still | |
23 | * gets replayed. | |
24 | * | |
25 | * We can get interactions between revokes and new log data within a | |
26 | * single transaction: | |
27 | * | |
28 | * Block is revoked and then journaled: | |
29 | * The desired end result is the journaling of the new block, so we | |
30 | * cancel the revoke before the transaction commits. | |
31 | * | |
32 | * Block is journaled and then revoked: | |
33 | * The revoke must take precedence over the write of the block, so we | |
34 | * need either to cancel the journal entry or to write the revoke | |
35 | * later in the log than the log block. In this case, we choose the | |
36 | * latter: journaling a block cancels any revoke record for that block | |
37 | * in the current transaction, so any revoke for that block in the | |
38 | * transaction must have happened after the block was journaled and so | |
39 | * the revoke must take precedence. | |
40 | * | |
41 | * Block is revoked and then written as data: | |
42 | * The data write is allowed to succeed, but the revoke is _not_ | |
43 | * cancelled. We still need to prevent old log records from | |
44 | * overwriting the new data. We don't even need to clear the revoke | |
45 | * bit here. | |
46 | * | |
1ba37268 YY |
47 | * We cache revoke status of a buffer in the current transaction in b_states |
48 | * bits. As the name says, revokevalid flag indicates that the cached revoke | |
49 | * status of a buffer is valid and we can rely on the cached status. | |
50 | * | |
470decc6 DK |
51 | * Revoke information on buffers is a tri-state value: |
52 | * | |
53 | * RevokeValid clear: no cached revoke status, need to look it up | |
54 | * RevokeValid set, Revoked clear: | |
55 | * buffer has not been revoked, and cancel_revoke | |
56 | * need do nothing. | |
57 | * RevokeValid set, Revoked set: | |
58 | * buffer has been revoked. | |
86db97c8 JK |
59 | * |
60 | * Locking rules: | |
61 | * We keep two hash tables of revoke records. One hashtable belongs to the | |
62 | * running transaction (is pointed to by journal->j_revoke), the other one | |
63 | * belongs to the committing transaction. Accesses to the second hash table | |
64 | * happen only from the kjournald and no other thread touches this table. Also | |
65 | * journal_switch_revoke_table() which switches which hashtable belongs to the | |
66 | * running and which to the committing transaction is called only from | |
67 | * kjournald. Therefore we need no locks when accessing the hashtable belonging | |
68 | * to the committing transaction. | |
69 | * | |
70 | * All users operating on the hash table belonging to the running transaction | |
71 | * have a handle to the transaction. Therefore they are safe from kjournald | |
72 | * switching hash tables under them. For operations on the lists of entries in | |
73 | * the hash table j_revoke_lock is used. | |
74 | * | |
25985edc | 75 | * Finally, also replay code uses the hash tables but at this moment no one else |
86db97c8 JK |
76 | * can touch them (filesystem isn't mounted yet) and hence no locking is |
77 | * needed. | |
470decc6 DK |
78 | */ |
79 | ||
80 | #ifndef __KERNEL__ | |
81 | #include "jfs_user.h" | |
82 | #else | |
83 | #include <linux/time.h> | |
84 | #include <linux/fs.h> | |
f7f4bccb | 85 | #include <linux/jbd2.h> |
470decc6 DK |
86 | #include <linux/errno.h> |
87 | #include <linux/slab.h> | |
88 | #include <linux/list.h> | |
470decc6 | 89 | #include <linux/init.h> |
67c457a8 | 90 | #include <linux/bio.h> |
f482394c | 91 | #include <linux/log2.h> |
d48458d4 | 92 | #include <linux/hash.h> |
db9ee220 | 93 | #endif |
470decc6 | 94 | |
e18b890b CL |
95 | static struct kmem_cache *jbd2_revoke_record_cache; |
96 | static struct kmem_cache *jbd2_revoke_table_cache; | |
470decc6 DK |
97 | |
98 | /* Each revoke record represents one single revoked block. During | |
99 | journal replay, this involves recording the transaction ID of the | |
100 | last transaction to revoke this block. */ | |
101 | ||
f7f4bccb | 102 | struct jbd2_revoke_record_s |
470decc6 DK |
103 | { |
104 | struct list_head hash; | |
105 | tid_t sequence; /* Used for recovery only */ | |
18eba7aa | 106 | unsigned long long blocknr; |
470decc6 DK |
107 | }; |
108 | ||
109 | ||
110 | /* The revoke table is just a simple hash table of revoke records. */ | |
f7f4bccb | 111 | struct jbd2_revoke_table_s |
470decc6 DK |
112 | { |
113 | /* It is conceivable that we might want a larger hash table | |
114 | * for recovery. Must be a power of two. */ | |
115 | int hash_size; | |
116 | int hash_shift; | |
117 | struct list_head *hash_table; | |
118 | }; | |
119 | ||
120 | ||
121 | #ifdef __KERNEL__ | |
9bcf976c | 122 | static void write_one_revoke_record(transaction_t *, |
e5a120ae JK |
123 | struct list_head *, |
124 | struct buffer_head **, int *, | |
9bcf976c JK |
125 | struct jbd2_revoke_record_s *); |
126 | static void flush_descriptor(journal_t *, struct buffer_head *, int); | |
470decc6 DK |
127 | #endif |
128 | ||
129 | /* Utility functions to maintain the revoke table */ | |
130 | ||
18eba7aa | 131 | static inline int hash(journal_t *journal, unsigned long long block) |
470decc6 | 132 | { |
d48458d4 | 133 | return hash_64(block, journal->j_revoke->hash_shift); |
470decc6 DK |
134 | } |
135 | ||
18eba7aa | 136 | static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr, |
470decc6 DK |
137 | tid_t seq) |
138 | { | |
139 | struct list_head *hash_list; | |
f7f4bccb | 140 | struct jbd2_revoke_record_s *record; |
7b506b10 | 141 | gfp_t gfp_mask = GFP_NOFS; |
470decc6 | 142 | |
7b506b10 MH |
143 | if (journal_oom_retry) |
144 | gfp_mask |= __GFP_NOFAIL; | |
145 | record = kmem_cache_alloc(jbd2_revoke_record_cache, gfp_mask); | |
470decc6 | 146 | if (!record) |
7b506b10 | 147 | return -ENOMEM; |
470decc6 DK |
148 | |
149 | record->sequence = seq; | |
150 | record->blocknr = blocknr; | |
151 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
152 | spin_lock(&journal->j_revoke_lock); | |
153 | list_add(&record->hash, hash_list); | |
154 | spin_unlock(&journal->j_revoke_lock); | |
155 | return 0; | |
470decc6 DK |
156 | } |
157 | ||
158 | /* Find a revoke record in the journal's hash table. */ | |
159 | ||
f7f4bccb | 160 | static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal, |
18eba7aa | 161 | unsigned long long blocknr) |
470decc6 DK |
162 | { |
163 | struct list_head *hash_list; | |
f7f4bccb | 164 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
165 | |
166 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
167 | ||
168 | spin_lock(&journal->j_revoke_lock); | |
f7f4bccb | 169 | record = (struct jbd2_revoke_record_s *) hash_list->next; |
470decc6 DK |
170 | while (&(record->hash) != hash_list) { |
171 | if (record->blocknr == blocknr) { | |
172 | spin_unlock(&journal->j_revoke_lock); | |
173 | return record; | |
174 | } | |
f7f4bccb | 175 | record = (struct jbd2_revoke_record_s *) record->hash.next; |
470decc6 DK |
176 | } |
177 | spin_unlock(&journal->j_revoke_lock); | |
178 | return NULL; | |
179 | } | |
180 | ||
9fa27c85 DG |
181 | void jbd2_journal_destroy_revoke_caches(void) |
182 | { | |
183 | if (jbd2_revoke_record_cache) { | |
184 | kmem_cache_destroy(jbd2_revoke_record_cache); | |
185 | jbd2_revoke_record_cache = NULL; | |
186 | } | |
187 | if (jbd2_revoke_table_cache) { | |
188 | kmem_cache_destroy(jbd2_revoke_table_cache); | |
189 | jbd2_revoke_table_cache = NULL; | |
190 | } | |
191 | } | |
192 | ||
f7f4bccb | 193 | int __init jbd2_journal_init_revoke_caches(void) |
470decc6 | 194 | { |
9fa27c85 DG |
195 | J_ASSERT(!jbd2_revoke_record_cache); |
196 | J_ASSERT(!jbd2_revoke_table_cache); | |
197 | ||
9c0e00e5 YY |
198 | jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s, |
199 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY); | |
1076d17a | 200 | if (!jbd2_revoke_record_cache) |
9fa27c85 | 201 | goto record_cache_failure; |
470decc6 | 202 | |
9c0e00e5 YY |
203 | jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s, |
204 | SLAB_TEMPORARY); | |
9fa27c85 DG |
205 | if (!jbd2_revoke_table_cache) |
206 | goto table_cache_failure; | |
470decc6 | 207 | return 0; |
9fa27c85 DG |
208 | table_cache_failure: |
209 | jbd2_journal_destroy_revoke_caches(); | |
210 | record_cache_failure: | |
211 | return -ENOMEM; | |
470decc6 DK |
212 | } |
213 | ||
83c49523 | 214 | static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size) |
470decc6 | 215 | { |
83c49523 DG |
216 | int shift = 0; |
217 | int tmp = hash_size; | |
218 | struct jbd2_revoke_table_s *table; | |
470decc6 | 219 | |
83c49523 DG |
220 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
221 | if (!table) | |
222 | goto out; | |
470decc6 | 223 | |
470decc6 DK |
224 | while((tmp >>= 1UL) != 0UL) |
225 | shift++; | |
226 | ||
83c49523 DG |
227 | table->hash_size = hash_size; |
228 | table->hash_shift = shift; | |
229 | table->hash_table = | |
470decc6 | 230 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); |
83c49523 DG |
231 | if (!table->hash_table) { |
232 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
233 | table = NULL; | |
234 | goto out; | |
470decc6 DK |
235 | } |
236 | ||
237 | for (tmp = 0; tmp < hash_size; tmp++) | |
83c49523 | 238 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
470decc6 | 239 | |
83c49523 DG |
240 | out: |
241 | return table; | |
242 | } | |
243 | ||
244 | static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) | |
245 | { | |
246 | int i; | |
247 | struct list_head *hash_list; | |
248 | ||
249 | for (i = 0; i < table->hash_size; i++) { | |
250 | hash_list = &table->hash_table[i]; | |
251 | J_ASSERT(list_empty(hash_list)); | |
470decc6 DK |
252 | } |
253 | ||
83c49523 DG |
254 | kfree(table->hash_table); |
255 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
256 | } | |
470decc6 | 257 | |
83c49523 DG |
258 | /* Initialise the revoke table for a given journal to a given size. */ |
259 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) | |
260 | { | |
261 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
f482394c | 262 | J_ASSERT(is_power_of_2(hash_size)); |
470decc6 | 263 | |
83c49523 DG |
264 | journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size); |
265 | if (!journal->j_revoke_table[0]) | |
266 | goto fail0; | |
470decc6 | 267 | |
83c49523 DG |
268 | journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size); |
269 | if (!journal->j_revoke_table[1]) | |
270 | goto fail1; | |
470decc6 | 271 | |
83c49523 | 272 | journal->j_revoke = journal->j_revoke_table[1]; |
470decc6 DK |
273 | |
274 | spin_lock_init(&journal->j_revoke_lock); | |
275 | ||
276 | return 0; | |
470decc6 | 277 | |
83c49523 DG |
278 | fail1: |
279 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
cd9cb405 | 280 | journal->j_revoke_table[0] = NULL; |
83c49523 DG |
281 | fail0: |
282 | return -ENOMEM; | |
283 | } | |
470decc6 | 284 | |
83c49523 | 285 | /* Destroy a journal's revoke table. The table must already be empty! */ |
f7f4bccb | 286 | void jbd2_journal_destroy_revoke(journal_t *journal) |
470decc6 | 287 | { |
470decc6 | 288 | journal->j_revoke = NULL; |
83c49523 DG |
289 | if (journal->j_revoke_table[0]) |
290 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
291 | if (journal->j_revoke_table[1]) | |
292 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
470decc6 DK |
293 | } |
294 | ||
295 | ||
296 | #ifdef __KERNEL__ | |
297 | ||
298 | /* | |
f7f4bccb | 299 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
470decc6 DK |
300 | * prevents the block from being replayed during recovery if we take a |
301 | * crash after this current transaction commits. Any subsequent | |
302 | * metadata writes of the buffer in this transaction cancel the | |
303 | * revoke. | |
304 | * | |
305 | * Note that this call may block --- it is up to the caller to make | |
306 | * sure that there are no further calls to journal_write_metadata | |
307 | * before the revoke is complete. In ext3, this implies calling the | |
308 | * revoke before clearing the block bitmap when we are deleting | |
309 | * metadata. | |
310 | * | |
f7f4bccb | 311 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
470decc6 DK |
312 | * parameter, but does _not_ forget the buffer_head if the bh was only |
313 | * found implicitly. | |
314 | * | |
315 | * bh_in may not be a journalled buffer - it may have come off | |
316 | * the hash tables without an attached journal_head. | |
317 | * | |
f7f4bccb | 318 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
470decc6 DK |
319 | * by one. |
320 | */ | |
321 | ||
18eba7aa | 322 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
470decc6 DK |
323 | struct buffer_head *bh_in) |
324 | { | |
325 | struct buffer_head *bh = NULL; | |
326 | journal_t *journal; | |
327 | struct block_device *bdev; | |
328 | int err; | |
329 | ||
330 | might_sleep(); | |
331 | if (bh_in) | |
332 | BUFFER_TRACE(bh_in, "enter"); | |
333 | ||
334 | journal = handle->h_transaction->t_journal; | |
f7f4bccb | 335 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
470decc6 DK |
336 | J_ASSERT (!"Cannot set revoke feature!"); |
337 | return -EINVAL; | |
338 | } | |
339 | ||
340 | bdev = journal->j_fs_dev; | |
341 | bh = bh_in; | |
342 | ||
343 | if (!bh) { | |
344 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
345 | if (bh) | |
346 | BUFFER_TRACE(bh, "found on hash"); | |
347 | } | |
cd02ff0b | 348 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
349 | else { |
350 | struct buffer_head *bh2; | |
351 | ||
352 | /* If there is a different buffer_head lying around in | |
353 | * memory anywhere... */ | |
354 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
355 | if (bh2) { | |
356 | /* ... and it has RevokeValid status... */ | |
357 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
358 | /* ...then it better be revoked too, | |
359 | * since it's illegal to create a revoke | |
360 | * record against a buffer_head which is | |
361 | * not marked revoked --- that would | |
362 | * risk missing a subsequent revoke | |
363 | * cancel. */ | |
364 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
365 | put_bh(bh2); | |
366 | } | |
367 | } | |
368 | #endif | |
369 | ||
370 | /* We really ought not ever to revoke twice in a row without | |
371 | first having the revoke cancelled: it's illegal to free a | |
372 | block twice without allocating it in between! */ | |
373 | if (bh) { | |
374 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
375 | "inconsistent data on disk")) { | |
376 | if (!bh_in) | |
377 | brelse(bh); | |
378 | return -EIO; | |
379 | } | |
380 | set_buffer_revoked(bh); | |
381 | set_buffer_revokevalid(bh); | |
382 | if (bh_in) { | |
f7f4bccb MC |
383 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
384 | jbd2_journal_forget(handle, bh_in); | |
470decc6 DK |
385 | } else { |
386 | BUFFER_TRACE(bh, "call brelse"); | |
387 | __brelse(bh); | |
388 | } | |
389 | } | |
390 | ||
29971769 | 391 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
470decc6 DK |
392 | err = insert_revoke_hash(journal, blocknr, |
393 | handle->h_transaction->t_tid); | |
394 | BUFFER_TRACE(bh_in, "exit"); | |
395 | return err; | |
396 | } | |
397 | ||
398 | /* | |
399 | * Cancel an outstanding revoke. For use only internally by the | |
f7f4bccb | 400 | * journaling code (called from jbd2_journal_get_write_access). |
470decc6 DK |
401 | * |
402 | * We trust buffer_revoked() on the buffer if the buffer is already | |
403 | * being journaled: if there is no revoke pending on the buffer, then we | |
404 | * don't do anything here. | |
405 | * | |
406 | * This would break if it were possible for a buffer to be revoked and | |
407 | * discarded, and then reallocated within the same transaction. In such | |
408 | * a case we would have lost the revoked bit, but when we arrived here | |
409 | * the second time we would still have a pending revoke to cancel. So, | |
410 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
411 | * set. | |
470decc6 | 412 | */ |
f7f4bccb | 413 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
470decc6 | 414 | { |
f7f4bccb | 415 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
416 | journal_t *journal = handle->h_transaction->t_journal; |
417 | int need_cancel; | |
418 | int did_revoke = 0; /* akpm: debug */ | |
419 | struct buffer_head *bh = jh2bh(jh); | |
420 | ||
421 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
422 | ||
423 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
424 | * only perform the full cancel if the revoke bit is set. If | |
425 | * not, we can't trust the revoke bit, and we need to do the | |
426 | * full search for a revoke record. */ | |
427 | if (test_set_buffer_revokevalid(bh)) { | |
428 | need_cancel = test_clear_buffer_revoked(bh); | |
429 | } else { | |
430 | need_cancel = 1; | |
431 | clear_buffer_revoked(bh); | |
432 | } | |
433 | ||
434 | if (need_cancel) { | |
435 | record = find_revoke_record(journal, bh->b_blocknr); | |
436 | if (record) { | |
437 | jbd_debug(4, "cancelled existing revoke on " | |
438 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
439 | spin_lock(&journal->j_revoke_lock); | |
440 | list_del(&record->hash); | |
441 | spin_unlock(&journal->j_revoke_lock); | |
f7f4bccb | 442 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
443 | did_revoke = 1; |
444 | } | |
445 | } | |
446 | ||
cd02ff0b | 447 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
448 | /* There better not be one left behind by now! */ |
449 | record = find_revoke_record(journal, bh->b_blocknr); | |
450 | J_ASSERT_JH(jh, record == NULL); | |
451 | #endif | |
452 | ||
453 | /* Finally, have we just cleared revoke on an unhashed | |
454 | * buffer_head? If so, we'd better make sure we clear the | |
455 | * revoked status on any hashed alias too, otherwise the revoke | |
456 | * state machine will get very upset later on. */ | |
457 | if (need_cancel) { | |
458 | struct buffer_head *bh2; | |
459 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
460 | if (bh2) { | |
461 | if (bh2 != bh) | |
462 | clear_buffer_revoked(bh2); | |
463 | __brelse(bh2); | |
464 | } | |
465 | } | |
466 | return did_revoke; | |
467 | } | |
468 | ||
1ba37268 YY |
469 | /* |
470 | * journal_clear_revoked_flag clears revoked flag of buffers in | |
471 | * revoke table to reflect there is no revoked buffers in the next | |
472 | * transaction which is going to be started. | |
473 | */ | |
474 | void jbd2_clear_buffer_revoked_flags(journal_t *journal) | |
475 | { | |
476 | struct jbd2_revoke_table_s *revoke = journal->j_revoke; | |
477 | int i = 0; | |
478 | ||
479 | for (i = 0; i < revoke->hash_size; i++) { | |
480 | struct list_head *hash_list; | |
481 | struct list_head *list_entry; | |
482 | hash_list = &revoke->hash_table[i]; | |
483 | ||
484 | list_for_each(list_entry, hash_list) { | |
485 | struct jbd2_revoke_record_s *record; | |
486 | struct buffer_head *bh; | |
487 | record = (struct jbd2_revoke_record_s *)list_entry; | |
488 | bh = __find_get_block(journal->j_fs_dev, | |
489 | record->blocknr, | |
490 | journal->j_blocksize); | |
491 | if (bh) { | |
492 | clear_buffer_revoked(bh); | |
493 | __brelse(bh); | |
494 | } | |
495 | } | |
496 | } | |
497 | } | |
498 | ||
470decc6 DK |
499 | /* journal_switch_revoke table select j_revoke for next transaction |
500 | * we do not want to suspend any processing until all revokes are | |
501 | * written -bzzz | |
502 | */ | |
f7f4bccb | 503 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
470decc6 DK |
504 | { |
505 | int i; | |
506 | ||
507 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
508 | journal->j_revoke = journal->j_revoke_table[1]; | |
509 | else | |
510 | journal->j_revoke = journal->j_revoke_table[0]; | |
511 | ||
512 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
513 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
514 | } | |
515 | ||
516 | /* | |
517 | * Write revoke records to the journal for all entries in the current | |
518 | * revoke hash, deleting the entries as we go. | |
470decc6 | 519 | */ |
9bcf976c JK |
520 | void jbd2_journal_write_revoke_records(transaction_t *transaction, |
521 | struct list_head *log_bufs) | |
470decc6 | 522 | { |
9bcf976c | 523 | journal_t *journal = transaction->t_journal; |
e5a120ae | 524 | struct buffer_head *descriptor; |
f7f4bccb MC |
525 | struct jbd2_revoke_record_s *record; |
526 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
527 | struct list_head *hash_list; |
528 | int i, offset, count; | |
529 | ||
530 | descriptor = NULL; | |
531 | offset = 0; | |
532 | count = 0; | |
533 | ||
534 | /* select revoke table for committing transaction */ | |
535 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
536 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
537 | ||
538 | for (i = 0; i < revoke->hash_size; i++) { | |
539 | hash_list = &revoke->hash_table[i]; | |
540 | ||
541 | while (!list_empty(hash_list)) { | |
f7f4bccb | 542 | record = (struct jbd2_revoke_record_s *) |
470decc6 | 543 | hash_list->next; |
9bcf976c JK |
544 | write_one_revoke_record(transaction, log_bufs, |
545 | &descriptor, &offset, record); | |
470decc6 DK |
546 | count++; |
547 | list_del(&record->hash); | |
f7f4bccb | 548 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
549 | } |
550 | } | |
551 | if (descriptor) | |
9bcf976c | 552 | flush_descriptor(journal, descriptor, offset); |
470decc6 DK |
553 | jbd_debug(1, "Wrote %d revoke records\n", count); |
554 | } | |
555 | ||
556 | /* | |
557 | * Write out one revoke record. We need to create a new descriptor | |
558 | * block if the old one is full or if we have not already created one. | |
559 | */ | |
560 | ||
9bcf976c | 561 | static void write_one_revoke_record(transaction_t *transaction, |
e5a120ae JK |
562 | struct list_head *log_bufs, |
563 | struct buffer_head **descriptorp, | |
470decc6 | 564 | int *offsetp, |
9bcf976c | 565 | struct jbd2_revoke_record_s *record) |
470decc6 | 566 | { |
9bcf976c | 567 | journal_t *journal = transaction->t_journal; |
42a7106d | 568 | int csum_size = 0; |
e5a120ae | 569 | struct buffer_head *descriptor; |
e531d0bc | 570 | int sz, offset; |
470decc6 DK |
571 | |
572 | /* If we are already aborting, this all becomes a noop. We | |
573 | still need to go round the loop in | |
f7f4bccb | 574 | jbd2_journal_write_revoke_records in order to free all of the |
470decc6 DK |
575 | revoke records: only the IO to the journal is omitted. */ |
576 | if (is_journal_aborted(journal)) | |
577 | return; | |
578 | ||
579 | descriptor = *descriptorp; | |
580 | offset = *offsetp; | |
581 | ||
42a7106d | 582 | /* Do we need to leave space at the end for a checksum? */ |
db9ee220 | 583 | if (jbd2_journal_has_csum_v2or3(journal)) |
1101cd4d | 584 | csum_size = sizeof(struct jbd2_journal_block_tail); |
42a7106d | 585 | |
56316a0d | 586 | if (jbd2_has_feature_64bit(journal)) |
e531d0bc DW |
587 | sz = 8; |
588 | else | |
589 | sz = 4; | |
590 | ||
470decc6 DK |
591 | /* Make sure we have a descriptor with space left for the record */ |
592 | if (descriptor) { | |
e531d0bc | 593 | if (offset + sz > journal->j_blocksize - csum_size) { |
9bcf976c | 594 | flush_descriptor(journal, descriptor, offset); |
470decc6 DK |
595 | descriptor = NULL; |
596 | } | |
597 | } | |
598 | ||
599 | if (!descriptor) { | |
32ab6715 JK |
600 | descriptor = jbd2_journal_get_descriptor_buffer(transaction, |
601 | JBD2_REVOKE_BLOCK); | |
470decc6 DK |
602 | if (!descriptor) |
603 | return; | |
470decc6 DK |
604 | |
605 | /* Record it so that we can wait for IO completion later */ | |
e5a120ae JK |
606 | BUFFER_TRACE(descriptor, "file in log_bufs"); |
607 | jbd2_file_log_bh(log_bufs, descriptor); | |
470decc6 | 608 | |
f7f4bccb | 609 | offset = sizeof(jbd2_journal_revoke_header_t); |
470decc6 DK |
610 | *descriptorp = descriptor; |
611 | } | |
612 | ||
56316a0d | 613 | if (jbd2_has_feature_64bit(journal)) |
e5a120ae | 614 | * ((__be64 *)(&descriptor->b_data[offset])) = |
b517bea1 | 615 | cpu_to_be64(record->blocknr); |
e531d0bc | 616 | else |
e5a120ae | 617 | * ((__be32 *)(&descriptor->b_data[offset])) = |
b517bea1 | 618 | cpu_to_be32(record->blocknr); |
e531d0bc | 619 | offset += sz; |
b517bea1 | 620 | |
470decc6 DK |
621 | *offsetp = offset; |
622 | } | |
623 | ||
624 | /* | |
625 | * Flush a revoke descriptor out to the journal. If we are aborting, | |
626 | * this is a noop; otherwise we are generating a buffer which needs to | |
627 | * be waited for during commit, so it has to go onto the appropriate | |
628 | * journal buffer list. | |
629 | */ | |
630 | ||
631 | static void flush_descriptor(journal_t *journal, | |
e5a120ae | 632 | struct buffer_head *descriptor, |
9bcf976c | 633 | int offset) |
470decc6 | 634 | { |
f7f4bccb | 635 | jbd2_journal_revoke_header_t *header; |
470decc6 DK |
636 | |
637 | if (is_journal_aborted(journal)) { | |
e5a120ae | 638 | put_bh(descriptor); |
470decc6 DK |
639 | return; |
640 | } | |
641 | ||
e5a120ae | 642 | header = (jbd2_journal_revoke_header_t *)descriptor->b_data; |
470decc6 | 643 | header->r_count = cpu_to_be32(offset); |
1101cd4d | 644 | jbd2_descriptor_block_csum_set(journal, descriptor); |
42a7106d | 645 | |
e5a120ae JK |
646 | set_buffer_jwrite(descriptor); |
647 | BUFFER_TRACE(descriptor, "write"); | |
648 | set_buffer_dirty(descriptor); | |
70fd7614 | 649 | write_dirty_buffer(descriptor, REQ_SYNC); |
470decc6 DK |
650 | } |
651 | #endif | |
652 | ||
653 | /* | |
654 | * Revoke support for recovery. | |
655 | * | |
656 | * Recovery needs to be able to: | |
657 | * | |
658 | * record all revoke records, including the tid of the latest instance | |
659 | * of each revoke in the journal | |
660 | * | |
661 | * check whether a given block in a given transaction should be replayed | |
662 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
663 | * transaction) | |
664 | * | |
665 | * empty the revoke table after recovery. | |
666 | */ | |
667 | ||
668 | /* | |
669 | * First, setting revoke records. We create a new revoke record for | |
670 | * every block ever revoked in the log as we scan it for recovery, and | |
671 | * we update the existing records if we find multiple revokes for a | |
672 | * single block. | |
673 | */ | |
674 | ||
f7f4bccb | 675 | int jbd2_journal_set_revoke(journal_t *journal, |
18eba7aa | 676 | unsigned long long blocknr, |
470decc6 DK |
677 | tid_t sequence) |
678 | { | |
f7f4bccb | 679 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
680 | |
681 | record = find_revoke_record(journal, blocknr); | |
682 | if (record) { | |
683 | /* If we have multiple occurrences, only record the | |
684 | * latest sequence number in the hashed record */ | |
685 | if (tid_gt(sequence, record->sequence)) | |
686 | record->sequence = sequence; | |
687 | return 0; | |
688 | } | |
689 | return insert_revoke_hash(journal, blocknr, sequence); | |
690 | } | |
691 | ||
692 | /* | |
693 | * Test revoke records. For a given block referenced in the log, has | |
694 | * that block been revoked? A revoke record with a given transaction | |
695 | * sequence number revokes all blocks in that transaction and earlier | |
696 | * ones, but later transactions still need replayed. | |
697 | */ | |
698 | ||
f7f4bccb | 699 | int jbd2_journal_test_revoke(journal_t *journal, |
18eba7aa | 700 | unsigned long long blocknr, |
470decc6 DK |
701 | tid_t sequence) |
702 | { | |
f7f4bccb | 703 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
704 | |
705 | record = find_revoke_record(journal, blocknr); | |
706 | if (!record) | |
707 | return 0; | |
708 | if (tid_gt(sequence, record->sequence)) | |
709 | return 0; | |
710 | return 1; | |
711 | } | |
712 | ||
713 | /* | |
714 | * Finally, once recovery is over, we need to clear the revoke table so | |
715 | * that it can be reused by the running filesystem. | |
716 | */ | |
717 | ||
f7f4bccb | 718 | void jbd2_journal_clear_revoke(journal_t *journal) |
470decc6 DK |
719 | { |
720 | int i; | |
721 | struct list_head *hash_list; | |
f7f4bccb MC |
722 | struct jbd2_revoke_record_s *record; |
723 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
724 | |
725 | revoke = journal->j_revoke; | |
726 | ||
727 | for (i = 0; i < revoke->hash_size; i++) { | |
728 | hash_list = &revoke->hash_table[i]; | |
729 | while (!list_empty(hash_list)) { | |
f7f4bccb | 730 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
470decc6 | 731 | list_del(&record->hash); |
f7f4bccb | 732 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
733 | } |
734 | } | |
735 | } |