]>
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 | ||
0d52154b | 181 | void jbd2_journal_destroy_revoke_record_cache(void) |
9fa27c85 | 182 | { |
8bdd5b60 WL |
183 | kmem_cache_destroy(jbd2_revoke_record_cache); |
184 | jbd2_revoke_record_cache = NULL; | |
0d52154b CX |
185 | } |
186 | ||
187 | void jbd2_journal_destroy_revoke_table_cache(void) | |
188 | { | |
8bdd5b60 WL |
189 | kmem_cache_destroy(jbd2_revoke_table_cache); |
190 | jbd2_revoke_table_cache = NULL; | |
9fa27c85 DG |
191 | } |
192 | ||
0d52154b | 193 | int __init jbd2_journal_init_revoke_record_cache(void) |
470decc6 | 194 | { |
9fa27c85 | 195 | J_ASSERT(!jbd2_revoke_record_cache); |
9c0e00e5 YY |
196 | jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s, |
197 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY); | |
470decc6 | 198 | |
0d52154b CX |
199 | if (!jbd2_revoke_record_cache) { |
200 | pr_emerg("JBD2: failed to create revoke_record cache\n"); | |
201 | return -ENOMEM; | |
202 | } | |
203 | return 0; | |
204 | } | |
205 | ||
206 | int __init jbd2_journal_init_revoke_table_cache(void) | |
207 | { | |
208 | J_ASSERT(!jbd2_revoke_table_cache); | |
9c0e00e5 YY |
209 | jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s, |
210 | SLAB_TEMPORARY); | |
0d52154b CX |
211 | if (!jbd2_revoke_table_cache) { |
212 | pr_emerg("JBD2: failed to create revoke_table cache\n"); | |
9fa27c85 | 213 | return -ENOMEM; |
0d52154b CX |
214 | } |
215 | return 0; | |
470decc6 DK |
216 | } |
217 | ||
83c49523 | 218 | static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size) |
470decc6 | 219 | { |
83c49523 DG |
220 | int shift = 0; |
221 | int tmp = hash_size; | |
222 | struct jbd2_revoke_table_s *table; | |
470decc6 | 223 | |
83c49523 DG |
224 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
225 | if (!table) | |
226 | goto out; | |
470decc6 | 227 | |
470decc6 DK |
228 | while((tmp >>= 1UL) != 0UL) |
229 | shift++; | |
230 | ||
83c49523 DG |
231 | table->hash_size = hash_size; |
232 | table->hash_shift = shift; | |
233 | table->hash_table = | |
6da2ec56 | 234 | kmalloc_array(hash_size, sizeof(struct list_head), GFP_KERNEL); |
83c49523 DG |
235 | if (!table->hash_table) { |
236 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
237 | table = NULL; | |
238 | goto out; | |
470decc6 DK |
239 | } |
240 | ||
241 | for (tmp = 0; tmp < hash_size; tmp++) | |
83c49523 | 242 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
470decc6 | 243 | |
83c49523 DG |
244 | out: |
245 | return table; | |
246 | } | |
247 | ||
248 | static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) | |
249 | { | |
250 | int i; | |
251 | struct list_head *hash_list; | |
252 | ||
253 | for (i = 0; i < table->hash_size; i++) { | |
254 | hash_list = &table->hash_table[i]; | |
255 | J_ASSERT(list_empty(hash_list)); | |
470decc6 DK |
256 | } |
257 | ||
83c49523 DG |
258 | kfree(table->hash_table); |
259 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
260 | } | |
470decc6 | 261 | |
83c49523 DG |
262 | /* Initialise the revoke table for a given journal to a given size. */ |
263 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) | |
264 | { | |
265 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
f482394c | 266 | J_ASSERT(is_power_of_2(hash_size)); |
470decc6 | 267 | |
83c49523 DG |
268 | journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size); |
269 | if (!journal->j_revoke_table[0]) | |
270 | goto fail0; | |
470decc6 | 271 | |
83c49523 DG |
272 | journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size); |
273 | if (!journal->j_revoke_table[1]) | |
274 | goto fail1; | |
470decc6 | 275 | |
83c49523 | 276 | journal->j_revoke = journal->j_revoke_table[1]; |
470decc6 DK |
277 | |
278 | spin_lock_init(&journal->j_revoke_lock); | |
279 | ||
280 | return 0; | |
470decc6 | 281 | |
83c49523 DG |
282 | fail1: |
283 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
cd9cb405 | 284 | journal->j_revoke_table[0] = NULL; |
83c49523 DG |
285 | fail0: |
286 | return -ENOMEM; | |
287 | } | |
470decc6 | 288 | |
83c49523 | 289 | /* Destroy a journal's revoke table. The table must already be empty! */ |
f7f4bccb | 290 | void jbd2_journal_destroy_revoke(journal_t *journal) |
470decc6 | 291 | { |
470decc6 | 292 | journal->j_revoke = NULL; |
83c49523 DG |
293 | if (journal->j_revoke_table[0]) |
294 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
295 | if (journal->j_revoke_table[1]) | |
296 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
470decc6 DK |
297 | } |
298 | ||
299 | ||
300 | #ifdef __KERNEL__ | |
301 | ||
302 | /* | |
f7f4bccb | 303 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
470decc6 DK |
304 | * prevents the block from being replayed during recovery if we take a |
305 | * crash after this current transaction commits. Any subsequent | |
306 | * metadata writes of the buffer in this transaction cancel the | |
307 | * revoke. | |
308 | * | |
309 | * Note that this call may block --- it is up to the caller to make | |
310 | * sure that there are no further calls to journal_write_metadata | |
311 | * before the revoke is complete. In ext3, this implies calling the | |
312 | * revoke before clearing the block bitmap when we are deleting | |
313 | * metadata. | |
314 | * | |
f7f4bccb | 315 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
470decc6 DK |
316 | * parameter, but does _not_ forget the buffer_head if the bh was only |
317 | * found implicitly. | |
318 | * | |
319 | * bh_in may not be a journalled buffer - it may have come off | |
320 | * the hash tables without an attached journal_head. | |
321 | * | |
f7f4bccb | 322 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
470decc6 DK |
323 | * by one. |
324 | */ | |
325 | ||
18eba7aa | 326 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
470decc6 DK |
327 | struct buffer_head *bh_in) |
328 | { | |
329 | struct buffer_head *bh = NULL; | |
330 | journal_t *journal; | |
331 | struct block_device *bdev; | |
332 | int err; | |
333 | ||
334 | might_sleep(); | |
335 | if (bh_in) | |
336 | BUFFER_TRACE(bh_in, "enter"); | |
337 | ||
338 | journal = handle->h_transaction->t_journal; | |
f7f4bccb | 339 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
470decc6 DK |
340 | J_ASSERT (!"Cannot set revoke feature!"); |
341 | return -EINVAL; | |
342 | } | |
343 | ||
344 | bdev = journal->j_fs_dev; | |
345 | bh = bh_in; | |
346 | ||
347 | if (!bh) { | |
348 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
349 | if (bh) | |
350 | BUFFER_TRACE(bh, "found on hash"); | |
351 | } | |
cd02ff0b | 352 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
353 | else { |
354 | struct buffer_head *bh2; | |
355 | ||
356 | /* If there is a different buffer_head lying around in | |
357 | * memory anywhere... */ | |
358 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
359 | if (bh2) { | |
360 | /* ... and it has RevokeValid status... */ | |
361 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
362 | /* ...then it better be revoked too, | |
363 | * since it's illegal to create a revoke | |
364 | * record against a buffer_head which is | |
365 | * not marked revoked --- that would | |
366 | * risk missing a subsequent revoke | |
367 | * cancel. */ | |
368 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
369 | put_bh(bh2); | |
370 | } | |
371 | } | |
372 | #endif | |
373 | ||
374 | /* We really ought not ever to revoke twice in a row without | |
375 | first having the revoke cancelled: it's illegal to free a | |
376 | block twice without allocating it in between! */ | |
377 | if (bh) { | |
378 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
379 | "inconsistent data on disk")) { | |
380 | if (!bh_in) | |
381 | brelse(bh); | |
382 | return -EIO; | |
383 | } | |
384 | set_buffer_revoked(bh); | |
385 | set_buffer_revokevalid(bh); | |
386 | if (bh_in) { | |
f7f4bccb MC |
387 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
388 | jbd2_journal_forget(handle, bh_in); | |
470decc6 DK |
389 | } else { |
390 | BUFFER_TRACE(bh, "call brelse"); | |
391 | __brelse(bh); | |
392 | } | |
393 | } | |
394 | ||
29971769 | 395 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
470decc6 DK |
396 | err = insert_revoke_hash(journal, blocknr, |
397 | handle->h_transaction->t_tid); | |
398 | BUFFER_TRACE(bh_in, "exit"); | |
399 | return err; | |
400 | } | |
401 | ||
402 | /* | |
403 | * Cancel an outstanding revoke. For use only internally by the | |
f7f4bccb | 404 | * journaling code (called from jbd2_journal_get_write_access). |
470decc6 DK |
405 | * |
406 | * We trust buffer_revoked() on the buffer if the buffer is already | |
407 | * being journaled: if there is no revoke pending on the buffer, then we | |
408 | * don't do anything here. | |
409 | * | |
410 | * This would break if it were possible for a buffer to be revoked and | |
411 | * discarded, and then reallocated within the same transaction. In such | |
412 | * a case we would have lost the revoked bit, but when we arrived here | |
413 | * the second time we would still have a pending revoke to cancel. So, | |
414 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
415 | * set. | |
470decc6 | 416 | */ |
f7f4bccb | 417 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
470decc6 | 418 | { |
f7f4bccb | 419 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
420 | journal_t *journal = handle->h_transaction->t_journal; |
421 | int need_cancel; | |
422 | int did_revoke = 0; /* akpm: debug */ | |
423 | struct buffer_head *bh = jh2bh(jh); | |
424 | ||
425 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
426 | ||
427 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
428 | * only perform the full cancel if the revoke bit is set. If | |
429 | * not, we can't trust the revoke bit, and we need to do the | |
430 | * full search for a revoke record. */ | |
431 | if (test_set_buffer_revokevalid(bh)) { | |
432 | need_cancel = test_clear_buffer_revoked(bh); | |
433 | } else { | |
434 | need_cancel = 1; | |
435 | clear_buffer_revoked(bh); | |
436 | } | |
437 | ||
438 | if (need_cancel) { | |
439 | record = find_revoke_record(journal, bh->b_blocknr); | |
440 | if (record) { | |
441 | jbd_debug(4, "cancelled existing revoke on " | |
442 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
443 | spin_lock(&journal->j_revoke_lock); | |
444 | list_del(&record->hash); | |
445 | spin_unlock(&journal->j_revoke_lock); | |
f7f4bccb | 446 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
447 | did_revoke = 1; |
448 | } | |
449 | } | |
450 | ||
cd02ff0b | 451 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
452 | /* There better not be one left behind by now! */ |
453 | record = find_revoke_record(journal, bh->b_blocknr); | |
454 | J_ASSERT_JH(jh, record == NULL); | |
455 | #endif | |
456 | ||
457 | /* Finally, have we just cleared revoke on an unhashed | |
458 | * buffer_head? If so, we'd better make sure we clear the | |
459 | * revoked status on any hashed alias too, otherwise the revoke | |
460 | * state machine will get very upset later on. */ | |
461 | if (need_cancel) { | |
462 | struct buffer_head *bh2; | |
463 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
464 | if (bh2) { | |
465 | if (bh2 != bh) | |
466 | clear_buffer_revoked(bh2); | |
467 | __brelse(bh2); | |
468 | } | |
469 | } | |
470 | return did_revoke; | |
471 | } | |
472 | ||
1ba37268 YY |
473 | /* |
474 | * journal_clear_revoked_flag clears revoked flag of buffers in | |
475 | * revoke table to reflect there is no revoked buffers in the next | |
476 | * transaction which is going to be started. | |
477 | */ | |
478 | void jbd2_clear_buffer_revoked_flags(journal_t *journal) | |
479 | { | |
480 | struct jbd2_revoke_table_s *revoke = journal->j_revoke; | |
481 | int i = 0; | |
482 | ||
483 | for (i = 0; i < revoke->hash_size; i++) { | |
484 | struct list_head *hash_list; | |
485 | struct list_head *list_entry; | |
486 | hash_list = &revoke->hash_table[i]; | |
487 | ||
488 | list_for_each(list_entry, hash_list) { | |
489 | struct jbd2_revoke_record_s *record; | |
490 | struct buffer_head *bh; | |
491 | record = (struct jbd2_revoke_record_s *)list_entry; | |
492 | bh = __find_get_block(journal->j_fs_dev, | |
493 | record->blocknr, | |
494 | journal->j_blocksize); | |
495 | if (bh) { | |
496 | clear_buffer_revoked(bh); | |
497 | __brelse(bh); | |
498 | } | |
499 | } | |
500 | } | |
501 | } | |
502 | ||
470decc6 DK |
503 | /* journal_switch_revoke table select j_revoke for next transaction |
504 | * we do not want to suspend any processing until all revokes are | |
505 | * written -bzzz | |
506 | */ | |
f7f4bccb | 507 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
470decc6 DK |
508 | { |
509 | int i; | |
510 | ||
511 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
512 | journal->j_revoke = journal->j_revoke_table[1]; | |
513 | else | |
514 | journal->j_revoke = journal->j_revoke_table[0]; | |
515 | ||
516 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
517 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
518 | } | |
519 | ||
520 | /* | |
521 | * Write revoke records to the journal for all entries in the current | |
522 | * revoke hash, deleting the entries as we go. | |
470decc6 | 523 | */ |
9bcf976c JK |
524 | void jbd2_journal_write_revoke_records(transaction_t *transaction, |
525 | struct list_head *log_bufs) | |
470decc6 | 526 | { |
9bcf976c | 527 | journal_t *journal = transaction->t_journal; |
e5a120ae | 528 | struct buffer_head *descriptor; |
f7f4bccb MC |
529 | struct jbd2_revoke_record_s *record; |
530 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
531 | struct list_head *hash_list; |
532 | int i, offset, count; | |
533 | ||
534 | descriptor = NULL; | |
535 | offset = 0; | |
536 | count = 0; | |
537 | ||
538 | /* select revoke table for committing transaction */ | |
539 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
540 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
541 | ||
542 | for (i = 0; i < revoke->hash_size; i++) { | |
543 | hash_list = &revoke->hash_table[i]; | |
544 | ||
545 | while (!list_empty(hash_list)) { | |
f7f4bccb | 546 | record = (struct jbd2_revoke_record_s *) |
470decc6 | 547 | hash_list->next; |
9bcf976c JK |
548 | write_one_revoke_record(transaction, log_bufs, |
549 | &descriptor, &offset, record); | |
470decc6 DK |
550 | count++; |
551 | list_del(&record->hash); | |
f7f4bccb | 552 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
553 | } |
554 | } | |
555 | if (descriptor) | |
9bcf976c | 556 | flush_descriptor(journal, descriptor, offset); |
470decc6 DK |
557 | jbd_debug(1, "Wrote %d revoke records\n", count); |
558 | } | |
559 | ||
560 | /* | |
561 | * Write out one revoke record. We need to create a new descriptor | |
562 | * block if the old one is full or if we have not already created one. | |
563 | */ | |
564 | ||
9bcf976c | 565 | static void write_one_revoke_record(transaction_t *transaction, |
e5a120ae JK |
566 | struct list_head *log_bufs, |
567 | struct buffer_head **descriptorp, | |
470decc6 | 568 | int *offsetp, |
9bcf976c | 569 | struct jbd2_revoke_record_s *record) |
470decc6 | 570 | { |
9bcf976c | 571 | journal_t *journal = transaction->t_journal; |
42a7106d | 572 | int csum_size = 0; |
e5a120ae | 573 | struct buffer_head *descriptor; |
e531d0bc | 574 | int sz, offset; |
470decc6 DK |
575 | |
576 | /* If we are already aborting, this all becomes a noop. We | |
577 | still need to go round the loop in | |
f7f4bccb | 578 | jbd2_journal_write_revoke_records in order to free all of the |
470decc6 DK |
579 | revoke records: only the IO to the journal is omitted. */ |
580 | if (is_journal_aborted(journal)) | |
581 | return; | |
582 | ||
583 | descriptor = *descriptorp; | |
584 | offset = *offsetp; | |
585 | ||
42a7106d | 586 | /* Do we need to leave space at the end for a checksum? */ |
db9ee220 | 587 | if (jbd2_journal_has_csum_v2or3(journal)) |
1101cd4d | 588 | csum_size = sizeof(struct jbd2_journal_block_tail); |
42a7106d | 589 | |
56316a0d | 590 | if (jbd2_has_feature_64bit(journal)) |
e531d0bc DW |
591 | sz = 8; |
592 | else | |
593 | sz = 4; | |
594 | ||
470decc6 DK |
595 | /* Make sure we have a descriptor with space left for the record */ |
596 | if (descriptor) { | |
e531d0bc | 597 | if (offset + sz > journal->j_blocksize - csum_size) { |
9bcf976c | 598 | flush_descriptor(journal, descriptor, offset); |
470decc6 DK |
599 | descriptor = NULL; |
600 | } | |
601 | } | |
602 | ||
603 | if (!descriptor) { | |
32ab6715 JK |
604 | descriptor = jbd2_journal_get_descriptor_buffer(transaction, |
605 | JBD2_REVOKE_BLOCK); | |
470decc6 DK |
606 | if (!descriptor) |
607 | return; | |
470decc6 DK |
608 | |
609 | /* Record it so that we can wait for IO completion later */ | |
e5a120ae JK |
610 | BUFFER_TRACE(descriptor, "file in log_bufs"); |
611 | jbd2_file_log_bh(log_bufs, descriptor); | |
470decc6 | 612 | |
f7f4bccb | 613 | offset = sizeof(jbd2_journal_revoke_header_t); |
470decc6 DK |
614 | *descriptorp = descriptor; |
615 | } | |
616 | ||
56316a0d | 617 | if (jbd2_has_feature_64bit(journal)) |
e5a120ae | 618 | * ((__be64 *)(&descriptor->b_data[offset])) = |
b517bea1 | 619 | cpu_to_be64(record->blocknr); |
e531d0bc | 620 | else |
e5a120ae | 621 | * ((__be32 *)(&descriptor->b_data[offset])) = |
b517bea1 | 622 | cpu_to_be32(record->blocknr); |
e531d0bc | 623 | offset += sz; |
b517bea1 | 624 | |
470decc6 DK |
625 | *offsetp = offset; |
626 | } | |
627 | ||
628 | /* | |
629 | * Flush a revoke descriptor out to the journal. If we are aborting, | |
630 | * this is a noop; otherwise we are generating a buffer which needs to | |
631 | * be waited for during commit, so it has to go onto the appropriate | |
632 | * journal buffer list. | |
633 | */ | |
634 | ||
635 | static void flush_descriptor(journal_t *journal, | |
e5a120ae | 636 | struct buffer_head *descriptor, |
9bcf976c | 637 | int offset) |
470decc6 | 638 | { |
f7f4bccb | 639 | jbd2_journal_revoke_header_t *header; |
470decc6 DK |
640 | |
641 | if (is_journal_aborted(journal)) { | |
e5a120ae | 642 | put_bh(descriptor); |
470decc6 DK |
643 | return; |
644 | } | |
645 | ||
e5a120ae | 646 | header = (jbd2_journal_revoke_header_t *)descriptor->b_data; |
470decc6 | 647 | header->r_count = cpu_to_be32(offset); |
1101cd4d | 648 | jbd2_descriptor_block_csum_set(journal, descriptor); |
42a7106d | 649 | |
e5a120ae JK |
650 | set_buffer_jwrite(descriptor); |
651 | BUFFER_TRACE(descriptor, "write"); | |
652 | set_buffer_dirty(descriptor); | |
70fd7614 | 653 | write_dirty_buffer(descriptor, REQ_SYNC); |
470decc6 DK |
654 | } |
655 | #endif | |
656 | ||
657 | /* | |
658 | * Revoke support for recovery. | |
659 | * | |
660 | * Recovery needs to be able to: | |
661 | * | |
662 | * record all revoke records, including the tid of the latest instance | |
663 | * of each revoke in the journal | |
664 | * | |
665 | * check whether a given block in a given transaction should be replayed | |
666 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
667 | * transaction) | |
668 | * | |
669 | * empty the revoke table after recovery. | |
670 | */ | |
671 | ||
672 | /* | |
673 | * First, setting revoke records. We create a new revoke record for | |
674 | * every block ever revoked in the log as we scan it for recovery, and | |
675 | * we update the existing records if we find multiple revokes for a | |
676 | * single block. | |
677 | */ | |
678 | ||
f7f4bccb | 679 | int jbd2_journal_set_revoke(journal_t *journal, |
18eba7aa | 680 | unsigned long long blocknr, |
470decc6 DK |
681 | tid_t sequence) |
682 | { | |
f7f4bccb | 683 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
684 | |
685 | record = find_revoke_record(journal, blocknr); | |
686 | if (record) { | |
687 | /* If we have multiple occurrences, only record the | |
688 | * latest sequence number in the hashed record */ | |
689 | if (tid_gt(sequence, record->sequence)) | |
690 | record->sequence = sequence; | |
691 | return 0; | |
692 | } | |
693 | return insert_revoke_hash(journal, blocknr, sequence); | |
694 | } | |
695 | ||
696 | /* | |
697 | * Test revoke records. For a given block referenced in the log, has | |
698 | * that block been revoked? A revoke record with a given transaction | |
699 | * sequence number revokes all blocks in that transaction and earlier | |
700 | * ones, but later transactions still need replayed. | |
701 | */ | |
702 | ||
f7f4bccb | 703 | int jbd2_journal_test_revoke(journal_t *journal, |
18eba7aa | 704 | unsigned long long blocknr, |
470decc6 DK |
705 | tid_t sequence) |
706 | { | |
f7f4bccb | 707 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
708 | |
709 | record = find_revoke_record(journal, blocknr); | |
710 | if (!record) | |
711 | return 0; | |
712 | if (tid_gt(sequence, record->sequence)) | |
713 | return 0; | |
714 | return 1; | |
715 | } | |
716 | ||
717 | /* | |
718 | * Finally, once recovery is over, we need to clear the revoke table so | |
719 | * that it can be reused by the running filesystem. | |
720 | */ | |
721 | ||
f7f4bccb | 722 | void jbd2_journal_clear_revoke(journal_t *journal) |
470decc6 DK |
723 | { |
724 | int i; | |
725 | struct list_head *hash_list; | |
f7f4bccb MC |
726 | struct jbd2_revoke_record_s *record; |
727 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
728 | |
729 | revoke = journal->j_revoke; | |
730 | ||
731 | for (i = 0; i < revoke->hash_size; i++) { | |
732 | hash_list = &revoke->hash_table[i]; | |
733 | while (!list_empty(hash_list)) { | |
f7f4bccb | 734 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
470decc6 | 735 | list_del(&record->hash); |
f7f4bccb | 736 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
737 | } |
738 | } | |
739 | } |