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__ | |
125 | static void write_one_revoke_record(journal_t *, transaction_t *, | |
e5a120ae JK |
126 | struct list_head *, |
127 | struct buffer_head **, int *, | |
67c457a8 | 128 | struct jbd2_revoke_record_s *, int); |
e5a120ae | 129 | static void flush_descriptor(journal_t *, struct buffer_head *, int, 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]); | |
283 | fail0: | |
284 | return -ENOMEM; | |
285 | } | |
470decc6 | 286 | |
83c49523 | 287 | /* Destroy a journal's revoke table. The table must already be empty! */ |
f7f4bccb | 288 | void jbd2_journal_destroy_revoke(journal_t *journal) |
470decc6 | 289 | { |
470decc6 | 290 | journal->j_revoke = NULL; |
83c49523 DG |
291 | if (journal->j_revoke_table[0]) |
292 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
293 | if (journal->j_revoke_table[1]) | |
294 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
470decc6 DK |
295 | } |
296 | ||
297 | ||
298 | #ifdef __KERNEL__ | |
299 | ||
300 | /* | |
f7f4bccb | 301 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
470decc6 DK |
302 | * prevents the block from being replayed during recovery if we take a |
303 | * crash after this current transaction commits. Any subsequent | |
304 | * metadata writes of the buffer in this transaction cancel the | |
305 | * revoke. | |
306 | * | |
307 | * Note that this call may block --- it is up to the caller to make | |
308 | * sure that there are no further calls to journal_write_metadata | |
309 | * before the revoke is complete. In ext3, this implies calling the | |
310 | * revoke before clearing the block bitmap when we are deleting | |
311 | * metadata. | |
312 | * | |
f7f4bccb | 313 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
470decc6 DK |
314 | * parameter, but does _not_ forget the buffer_head if the bh was only |
315 | * found implicitly. | |
316 | * | |
317 | * bh_in may not be a journalled buffer - it may have come off | |
318 | * the hash tables without an attached journal_head. | |
319 | * | |
f7f4bccb | 320 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
470decc6 DK |
321 | * by one. |
322 | */ | |
323 | ||
18eba7aa | 324 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
470decc6 DK |
325 | struct buffer_head *bh_in) |
326 | { | |
327 | struct buffer_head *bh = NULL; | |
328 | journal_t *journal; | |
329 | struct block_device *bdev; | |
330 | int err; | |
331 | ||
332 | might_sleep(); | |
333 | if (bh_in) | |
334 | BUFFER_TRACE(bh_in, "enter"); | |
335 | ||
336 | journal = handle->h_transaction->t_journal; | |
f7f4bccb | 337 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
470decc6 DK |
338 | J_ASSERT (!"Cannot set revoke feature!"); |
339 | return -EINVAL; | |
340 | } | |
341 | ||
342 | bdev = journal->j_fs_dev; | |
343 | bh = bh_in; | |
344 | ||
345 | if (!bh) { | |
346 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
347 | if (bh) | |
348 | BUFFER_TRACE(bh, "found on hash"); | |
349 | } | |
cd02ff0b | 350 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
351 | else { |
352 | struct buffer_head *bh2; | |
353 | ||
354 | /* If there is a different buffer_head lying around in | |
355 | * memory anywhere... */ | |
356 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
357 | if (bh2) { | |
358 | /* ... and it has RevokeValid status... */ | |
359 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
360 | /* ...then it better be revoked too, | |
361 | * since it's illegal to create a revoke | |
362 | * record against a buffer_head which is | |
363 | * not marked revoked --- that would | |
364 | * risk missing a subsequent revoke | |
365 | * cancel. */ | |
366 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
367 | put_bh(bh2); | |
368 | } | |
369 | } | |
370 | #endif | |
371 | ||
372 | /* We really ought not ever to revoke twice in a row without | |
373 | first having the revoke cancelled: it's illegal to free a | |
374 | block twice without allocating it in between! */ | |
375 | if (bh) { | |
376 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
377 | "inconsistent data on disk")) { | |
378 | if (!bh_in) | |
379 | brelse(bh); | |
380 | return -EIO; | |
381 | } | |
382 | set_buffer_revoked(bh); | |
383 | set_buffer_revokevalid(bh); | |
384 | if (bh_in) { | |
f7f4bccb MC |
385 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
386 | jbd2_journal_forget(handle, bh_in); | |
470decc6 DK |
387 | } else { |
388 | BUFFER_TRACE(bh, "call brelse"); | |
389 | __brelse(bh); | |
390 | } | |
391 | } | |
392 | ||
29971769 | 393 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
470decc6 DK |
394 | err = insert_revoke_hash(journal, blocknr, |
395 | handle->h_transaction->t_tid); | |
396 | BUFFER_TRACE(bh_in, "exit"); | |
397 | return err; | |
398 | } | |
399 | ||
400 | /* | |
401 | * Cancel an outstanding revoke. For use only internally by the | |
f7f4bccb | 402 | * journaling code (called from jbd2_journal_get_write_access). |
470decc6 DK |
403 | * |
404 | * We trust buffer_revoked() on the buffer if the buffer is already | |
405 | * being journaled: if there is no revoke pending on the buffer, then we | |
406 | * don't do anything here. | |
407 | * | |
408 | * This would break if it were possible for a buffer to be revoked and | |
409 | * discarded, and then reallocated within the same transaction. In such | |
410 | * a case we would have lost the revoked bit, but when we arrived here | |
411 | * the second time we would still have a pending revoke to cancel. So, | |
412 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
413 | * set. | |
470decc6 | 414 | */ |
f7f4bccb | 415 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
470decc6 | 416 | { |
f7f4bccb | 417 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
418 | journal_t *journal = handle->h_transaction->t_journal; |
419 | int need_cancel; | |
420 | int did_revoke = 0; /* akpm: debug */ | |
421 | struct buffer_head *bh = jh2bh(jh); | |
422 | ||
423 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
424 | ||
425 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
426 | * only perform the full cancel if the revoke bit is set. If | |
427 | * not, we can't trust the revoke bit, and we need to do the | |
428 | * full search for a revoke record. */ | |
429 | if (test_set_buffer_revokevalid(bh)) { | |
430 | need_cancel = test_clear_buffer_revoked(bh); | |
431 | } else { | |
432 | need_cancel = 1; | |
433 | clear_buffer_revoked(bh); | |
434 | } | |
435 | ||
436 | if (need_cancel) { | |
437 | record = find_revoke_record(journal, bh->b_blocknr); | |
438 | if (record) { | |
439 | jbd_debug(4, "cancelled existing revoke on " | |
440 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
441 | spin_lock(&journal->j_revoke_lock); | |
442 | list_del(&record->hash); | |
443 | spin_unlock(&journal->j_revoke_lock); | |
f7f4bccb | 444 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
445 | did_revoke = 1; |
446 | } | |
447 | } | |
448 | ||
cd02ff0b | 449 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
450 | /* There better not be one left behind by now! */ |
451 | record = find_revoke_record(journal, bh->b_blocknr); | |
452 | J_ASSERT_JH(jh, record == NULL); | |
453 | #endif | |
454 | ||
455 | /* Finally, have we just cleared revoke on an unhashed | |
456 | * buffer_head? If so, we'd better make sure we clear the | |
457 | * revoked status on any hashed alias too, otherwise the revoke | |
458 | * state machine will get very upset later on. */ | |
459 | if (need_cancel) { | |
460 | struct buffer_head *bh2; | |
461 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
462 | if (bh2) { | |
463 | if (bh2 != bh) | |
464 | clear_buffer_revoked(bh2); | |
465 | __brelse(bh2); | |
466 | } | |
467 | } | |
468 | return did_revoke; | |
469 | } | |
470 | ||
1ba37268 YY |
471 | /* |
472 | * journal_clear_revoked_flag clears revoked flag of buffers in | |
473 | * revoke table to reflect there is no revoked buffers in the next | |
474 | * transaction which is going to be started. | |
475 | */ | |
476 | void jbd2_clear_buffer_revoked_flags(journal_t *journal) | |
477 | { | |
478 | struct jbd2_revoke_table_s *revoke = journal->j_revoke; | |
479 | int i = 0; | |
480 | ||
481 | for (i = 0; i < revoke->hash_size; i++) { | |
482 | struct list_head *hash_list; | |
483 | struct list_head *list_entry; | |
484 | hash_list = &revoke->hash_table[i]; | |
485 | ||
486 | list_for_each(list_entry, hash_list) { | |
487 | struct jbd2_revoke_record_s *record; | |
488 | struct buffer_head *bh; | |
489 | record = (struct jbd2_revoke_record_s *)list_entry; | |
490 | bh = __find_get_block(journal->j_fs_dev, | |
491 | record->blocknr, | |
492 | journal->j_blocksize); | |
493 | if (bh) { | |
494 | clear_buffer_revoked(bh); | |
495 | __brelse(bh); | |
496 | } | |
497 | } | |
498 | } | |
499 | } | |
500 | ||
470decc6 DK |
501 | /* journal_switch_revoke table select j_revoke for next transaction |
502 | * we do not want to suspend any processing until all revokes are | |
503 | * written -bzzz | |
504 | */ | |
f7f4bccb | 505 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
470decc6 DK |
506 | { |
507 | int i; | |
508 | ||
509 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
510 | journal->j_revoke = journal->j_revoke_table[1]; | |
511 | else | |
512 | journal->j_revoke = journal->j_revoke_table[0]; | |
513 | ||
514 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
515 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
516 | } | |
517 | ||
518 | /* | |
519 | * Write revoke records to the journal for all entries in the current | |
520 | * revoke hash, deleting the entries as we go. | |
470decc6 | 521 | */ |
f7f4bccb | 522 | void jbd2_journal_write_revoke_records(journal_t *journal, |
67c457a8 | 523 | transaction_t *transaction, |
e5a120ae | 524 | struct list_head *log_bufs, |
67c457a8 | 525 | int write_op) |
470decc6 | 526 | { |
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; |
e5a120ae | 547 | write_one_revoke_record(journal, transaction, log_bufs, |
470decc6 | 548 | &descriptor, &offset, |
67c457a8 | 549 | record, write_op); |
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) | |
67c457a8 | 556 | flush_descriptor(journal, descriptor, offset, write_op); |
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 | ||
565 | static void write_one_revoke_record(journal_t *journal, | |
566 | transaction_t *transaction, | |
e5a120ae JK |
567 | struct list_head *log_bufs, |
568 | struct buffer_head **descriptorp, | |
470decc6 | 569 | int *offsetp, |
67c457a8 TT |
570 | struct jbd2_revoke_record_s *record, |
571 | int write_op) | |
470decc6 | 572 | { |
42a7106d | 573 | int csum_size = 0; |
e5a120ae | 574 | struct buffer_head *descriptor; |
e531d0bc | 575 | int sz, offset; |
470decc6 DK |
576 | journal_header_t *header; |
577 | ||
578 | /* If we are already aborting, this all becomes a noop. We | |
579 | still need to go round the loop in | |
f7f4bccb | 580 | jbd2_journal_write_revoke_records in order to free all of the |
470decc6 DK |
581 | revoke records: only the IO to the journal is omitted. */ |
582 | if (is_journal_aborted(journal)) | |
583 | return; | |
584 | ||
585 | descriptor = *descriptorp; | |
586 | offset = *offsetp; | |
587 | ||
42a7106d | 588 | /* Do we need to leave space at the end for a checksum? */ |
db9ee220 | 589 | if (jbd2_journal_has_csum_v2or3(journal)) |
42a7106d DW |
590 | csum_size = sizeof(struct jbd2_journal_revoke_tail); |
591 | ||
56316a0d | 592 | if (jbd2_has_feature_64bit(journal)) |
e531d0bc DW |
593 | sz = 8; |
594 | else | |
595 | sz = 4; | |
596 | ||
470decc6 DK |
597 | /* Make sure we have a descriptor with space left for the record */ |
598 | if (descriptor) { | |
e531d0bc | 599 | if (offset + sz > journal->j_blocksize - csum_size) { |
67c457a8 | 600 | flush_descriptor(journal, descriptor, offset, write_op); |
470decc6 DK |
601 | descriptor = NULL; |
602 | } | |
603 | } | |
604 | ||
605 | if (!descriptor) { | |
f7f4bccb | 606 | descriptor = jbd2_journal_get_descriptor_buffer(journal); |
470decc6 DK |
607 | if (!descriptor) |
608 | return; | |
e5a120ae | 609 | header = (journal_header_t *)descriptor->b_data; |
f7f4bccb MC |
610 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); |
611 | header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK); | |
470decc6 DK |
612 | header->h_sequence = cpu_to_be32(transaction->t_tid); |
613 | ||
614 | /* Record it so that we can wait for IO completion later */ | |
e5a120ae JK |
615 | BUFFER_TRACE(descriptor, "file in log_bufs"); |
616 | jbd2_file_log_bh(log_bufs, descriptor); | |
470decc6 | 617 | |
f7f4bccb | 618 | offset = sizeof(jbd2_journal_revoke_header_t); |
470decc6 DK |
619 | *descriptorp = descriptor; |
620 | } | |
621 | ||
56316a0d | 622 | if (jbd2_has_feature_64bit(journal)) |
e5a120ae | 623 | * ((__be64 *)(&descriptor->b_data[offset])) = |
b517bea1 | 624 | cpu_to_be64(record->blocknr); |
e531d0bc | 625 | else |
e5a120ae | 626 | * ((__be32 *)(&descriptor->b_data[offset])) = |
b517bea1 | 627 | cpu_to_be32(record->blocknr); |
e531d0bc | 628 | offset += sz; |
b517bea1 | 629 | |
470decc6 DK |
630 | *offsetp = offset; |
631 | } | |
632 | ||
e5a120ae | 633 | static void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh) |
42a7106d DW |
634 | { |
635 | struct jbd2_journal_revoke_tail *tail; | |
636 | __u32 csum; | |
637 | ||
db9ee220 | 638 | if (!jbd2_journal_has_csum_v2or3(j)) |
42a7106d DW |
639 | return; |
640 | ||
e5a120ae | 641 | tail = (struct jbd2_journal_revoke_tail *)(bh->b_data + j->j_blocksize - |
42a7106d DW |
642 | sizeof(struct jbd2_journal_revoke_tail)); |
643 | tail->r_checksum = 0; | |
e5a120ae | 644 | csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); |
42a7106d DW |
645 | tail->r_checksum = cpu_to_be32(csum); |
646 | } | |
647 | ||
470decc6 DK |
648 | /* |
649 | * Flush a revoke descriptor out to the journal. If we are aborting, | |
650 | * this is a noop; otherwise we are generating a buffer which needs to | |
651 | * be waited for during commit, so it has to go onto the appropriate | |
652 | * journal buffer list. | |
653 | */ | |
654 | ||
655 | static void flush_descriptor(journal_t *journal, | |
e5a120ae | 656 | struct buffer_head *descriptor, |
67c457a8 | 657 | int offset, int write_op) |
470decc6 | 658 | { |
f7f4bccb | 659 | jbd2_journal_revoke_header_t *header; |
470decc6 DK |
660 | |
661 | if (is_journal_aborted(journal)) { | |
e5a120ae | 662 | put_bh(descriptor); |
470decc6 DK |
663 | return; |
664 | } | |
665 | ||
e5a120ae | 666 | header = (jbd2_journal_revoke_header_t *)descriptor->b_data; |
470decc6 | 667 | header->r_count = cpu_to_be32(offset); |
42a7106d DW |
668 | jbd2_revoke_csum_set(journal, descriptor); |
669 | ||
e5a120ae JK |
670 | set_buffer_jwrite(descriptor); |
671 | BUFFER_TRACE(descriptor, "write"); | |
672 | set_buffer_dirty(descriptor); | |
673 | write_dirty_buffer(descriptor, write_op); | |
470decc6 DK |
674 | } |
675 | #endif | |
676 | ||
677 | /* | |
678 | * Revoke support for recovery. | |
679 | * | |
680 | * Recovery needs to be able to: | |
681 | * | |
682 | * record all revoke records, including the tid of the latest instance | |
683 | * of each revoke in the journal | |
684 | * | |
685 | * check whether a given block in a given transaction should be replayed | |
686 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
687 | * transaction) | |
688 | * | |
689 | * empty the revoke table after recovery. | |
690 | */ | |
691 | ||
692 | /* | |
693 | * First, setting revoke records. We create a new revoke record for | |
694 | * every block ever revoked in the log as we scan it for recovery, and | |
695 | * we update the existing records if we find multiple revokes for a | |
696 | * single block. | |
697 | */ | |
698 | ||
f7f4bccb | 699 | int jbd2_journal_set_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 | /* If we have multiple occurrences, only record the | |
708 | * latest sequence number in the hashed record */ | |
709 | if (tid_gt(sequence, record->sequence)) | |
710 | record->sequence = sequence; | |
711 | return 0; | |
712 | } | |
713 | return insert_revoke_hash(journal, blocknr, sequence); | |
714 | } | |
715 | ||
716 | /* | |
717 | * Test revoke records. For a given block referenced in the log, has | |
718 | * that block been revoked? A revoke record with a given transaction | |
719 | * sequence number revokes all blocks in that transaction and earlier | |
720 | * ones, but later transactions still need replayed. | |
721 | */ | |
722 | ||
f7f4bccb | 723 | int jbd2_journal_test_revoke(journal_t *journal, |
18eba7aa | 724 | unsigned long long blocknr, |
470decc6 DK |
725 | tid_t sequence) |
726 | { | |
f7f4bccb | 727 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
728 | |
729 | record = find_revoke_record(journal, blocknr); | |
730 | if (!record) | |
731 | return 0; | |
732 | if (tid_gt(sequence, record->sequence)) | |
733 | return 0; | |
734 | return 1; | |
735 | } | |
736 | ||
737 | /* | |
738 | * Finally, once recovery is over, we need to clear the revoke table so | |
739 | * that it can be reused by the running filesystem. | |
740 | */ | |
741 | ||
f7f4bccb | 742 | void jbd2_journal_clear_revoke(journal_t *journal) |
470decc6 DK |
743 | { |
744 | int i; | |
745 | struct list_head *hash_list; | |
f7f4bccb MC |
746 | struct jbd2_revoke_record_s *record; |
747 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
748 | |
749 | revoke = journal->j_revoke; | |
750 | ||
751 | for (i = 0; i < revoke->hash_size; i++) { | |
752 | hash_list = &revoke->hash_table[i]; | |
753 | while (!list_empty(hash_list)) { | |
f7f4bccb | 754 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
470decc6 | 755 | list_del(&record->hash); |
f7f4bccb | 756 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
757 | } |
758 | } | |
759 | } |