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; |
470decc6 DK |
144 | |
145 | repeat: | |
f7f4bccb | 146 | record = kmem_cache_alloc(jbd2_revoke_record_cache, GFP_NOFS); |
470decc6 DK |
147 | if (!record) |
148 | goto oom; | |
149 | ||
150 | record->sequence = seq; | |
151 | record->blocknr = blocknr; | |
152 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
153 | spin_lock(&journal->j_revoke_lock); | |
154 | list_add(&record->hash, hash_list); | |
155 | spin_unlock(&journal->j_revoke_lock); | |
156 | return 0; | |
157 | ||
158 | oom: | |
159 | if (!journal_oom_retry) | |
160 | return -ENOMEM; | |
329d291f | 161 | jbd_debug(1, "ENOMEM in %s, retrying\n", __func__); |
470decc6 DK |
162 | yield(); |
163 | goto repeat; | |
164 | } | |
165 | ||
166 | /* Find a revoke record in the journal's hash table. */ | |
167 | ||
f7f4bccb | 168 | static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal, |
18eba7aa | 169 | unsigned long long blocknr) |
470decc6 DK |
170 | { |
171 | struct list_head *hash_list; | |
f7f4bccb | 172 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
173 | |
174 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
175 | ||
176 | spin_lock(&journal->j_revoke_lock); | |
f7f4bccb | 177 | record = (struct jbd2_revoke_record_s *) hash_list->next; |
470decc6 DK |
178 | while (&(record->hash) != hash_list) { |
179 | if (record->blocknr == blocknr) { | |
180 | spin_unlock(&journal->j_revoke_lock); | |
181 | return record; | |
182 | } | |
f7f4bccb | 183 | record = (struct jbd2_revoke_record_s *) record->hash.next; |
470decc6 DK |
184 | } |
185 | spin_unlock(&journal->j_revoke_lock); | |
186 | return NULL; | |
187 | } | |
188 | ||
9fa27c85 DG |
189 | void jbd2_journal_destroy_revoke_caches(void) |
190 | { | |
191 | if (jbd2_revoke_record_cache) { | |
192 | kmem_cache_destroy(jbd2_revoke_record_cache); | |
193 | jbd2_revoke_record_cache = NULL; | |
194 | } | |
195 | if (jbd2_revoke_table_cache) { | |
196 | kmem_cache_destroy(jbd2_revoke_table_cache); | |
197 | jbd2_revoke_table_cache = NULL; | |
198 | } | |
199 | } | |
200 | ||
f7f4bccb | 201 | int __init jbd2_journal_init_revoke_caches(void) |
470decc6 | 202 | { |
9fa27c85 DG |
203 | J_ASSERT(!jbd2_revoke_record_cache); |
204 | J_ASSERT(!jbd2_revoke_table_cache); | |
205 | ||
9c0e00e5 YY |
206 | jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s, |
207 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY); | |
1076d17a | 208 | if (!jbd2_revoke_record_cache) |
9fa27c85 | 209 | goto record_cache_failure; |
470decc6 | 210 | |
9c0e00e5 YY |
211 | jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s, |
212 | SLAB_TEMPORARY); | |
9fa27c85 DG |
213 | if (!jbd2_revoke_table_cache) |
214 | goto table_cache_failure; | |
470decc6 | 215 | return 0; |
9fa27c85 DG |
216 | table_cache_failure: |
217 | jbd2_journal_destroy_revoke_caches(); | |
218 | record_cache_failure: | |
219 | return -ENOMEM; | |
470decc6 DK |
220 | } |
221 | ||
83c49523 | 222 | static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size) |
470decc6 | 223 | { |
83c49523 DG |
224 | int shift = 0; |
225 | int tmp = hash_size; | |
226 | struct jbd2_revoke_table_s *table; | |
470decc6 | 227 | |
83c49523 DG |
228 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
229 | if (!table) | |
230 | goto out; | |
470decc6 | 231 | |
470decc6 DK |
232 | while((tmp >>= 1UL) != 0UL) |
233 | shift++; | |
234 | ||
83c49523 DG |
235 | table->hash_size = hash_size; |
236 | table->hash_shift = shift; | |
237 | table->hash_table = | |
470decc6 | 238 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); |
83c49523 DG |
239 | if (!table->hash_table) { |
240 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
241 | table = NULL; | |
242 | goto out; | |
470decc6 DK |
243 | } |
244 | ||
245 | for (tmp = 0; tmp < hash_size; tmp++) | |
83c49523 | 246 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
470decc6 | 247 | |
83c49523 DG |
248 | out: |
249 | return table; | |
250 | } | |
251 | ||
252 | static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) | |
253 | { | |
254 | int i; | |
255 | struct list_head *hash_list; | |
256 | ||
257 | for (i = 0; i < table->hash_size; i++) { | |
258 | hash_list = &table->hash_table[i]; | |
259 | J_ASSERT(list_empty(hash_list)); | |
470decc6 DK |
260 | } |
261 | ||
83c49523 DG |
262 | kfree(table->hash_table); |
263 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
264 | } | |
470decc6 | 265 | |
83c49523 DG |
266 | /* Initialise the revoke table for a given journal to a given size. */ |
267 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) | |
268 | { | |
269 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
f482394c | 270 | J_ASSERT(is_power_of_2(hash_size)); |
470decc6 | 271 | |
83c49523 DG |
272 | journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size); |
273 | if (!journal->j_revoke_table[0]) | |
274 | goto fail0; | |
470decc6 | 275 | |
83c49523 DG |
276 | journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size); |
277 | if (!journal->j_revoke_table[1]) | |
278 | goto fail1; | |
470decc6 | 279 | |
83c49523 | 280 | journal->j_revoke = journal->j_revoke_table[1]; |
470decc6 DK |
281 | |
282 | spin_lock_init(&journal->j_revoke_lock); | |
283 | ||
284 | return 0; | |
470decc6 | 285 | |
83c49523 DG |
286 | fail1: |
287 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
288 | fail0: | |
289 | return -ENOMEM; | |
290 | } | |
470decc6 | 291 | |
83c49523 | 292 | /* Destroy a journal's revoke table. The table must already be empty! */ |
f7f4bccb | 293 | void jbd2_journal_destroy_revoke(journal_t *journal) |
470decc6 | 294 | { |
470decc6 | 295 | journal->j_revoke = NULL; |
83c49523 DG |
296 | if (journal->j_revoke_table[0]) |
297 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
298 | if (journal->j_revoke_table[1]) | |
299 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
470decc6 DK |
300 | } |
301 | ||
302 | ||
303 | #ifdef __KERNEL__ | |
304 | ||
305 | /* | |
f7f4bccb | 306 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
470decc6 DK |
307 | * prevents the block from being replayed during recovery if we take a |
308 | * crash after this current transaction commits. Any subsequent | |
309 | * metadata writes of the buffer in this transaction cancel the | |
310 | * revoke. | |
311 | * | |
312 | * Note that this call may block --- it is up to the caller to make | |
313 | * sure that there are no further calls to journal_write_metadata | |
314 | * before the revoke is complete. In ext3, this implies calling the | |
315 | * revoke before clearing the block bitmap when we are deleting | |
316 | * metadata. | |
317 | * | |
f7f4bccb | 318 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
470decc6 DK |
319 | * parameter, but does _not_ forget the buffer_head if the bh was only |
320 | * found implicitly. | |
321 | * | |
322 | * bh_in may not be a journalled buffer - it may have come off | |
323 | * the hash tables without an attached journal_head. | |
324 | * | |
f7f4bccb | 325 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
470decc6 DK |
326 | * by one. |
327 | */ | |
328 | ||
18eba7aa | 329 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
470decc6 DK |
330 | struct buffer_head *bh_in) |
331 | { | |
332 | struct buffer_head *bh = NULL; | |
333 | journal_t *journal; | |
334 | struct block_device *bdev; | |
335 | int err; | |
336 | ||
337 | might_sleep(); | |
338 | if (bh_in) | |
339 | BUFFER_TRACE(bh_in, "enter"); | |
340 | ||
341 | journal = handle->h_transaction->t_journal; | |
f7f4bccb | 342 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
470decc6 DK |
343 | J_ASSERT (!"Cannot set revoke feature!"); |
344 | return -EINVAL; | |
345 | } | |
346 | ||
347 | bdev = journal->j_fs_dev; | |
348 | bh = bh_in; | |
349 | ||
350 | if (!bh) { | |
351 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
352 | if (bh) | |
353 | BUFFER_TRACE(bh, "found on hash"); | |
354 | } | |
cd02ff0b | 355 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
356 | else { |
357 | struct buffer_head *bh2; | |
358 | ||
359 | /* If there is a different buffer_head lying around in | |
360 | * memory anywhere... */ | |
361 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
362 | if (bh2) { | |
363 | /* ... and it has RevokeValid status... */ | |
364 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
365 | /* ...then it better be revoked too, | |
366 | * since it's illegal to create a revoke | |
367 | * record against a buffer_head which is | |
368 | * not marked revoked --- that would | |
369 | * risk missing a subsequent revoke | |
370 | * cancel. */ | |
371 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
372 | put_bh(bh2); | |
373 | } | |
374 | } | |
375 | #endif | |
376 | ||
377 | /* We really ought not ever to revoke twice in a row without | |
378 | first having the revoke cancelled: it's illegal to free a | |
379 | block twice without allocating it in between! */ | |
380 | if (bh) { | |
381 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
382 | "inconsistent data on disk")) { | |
383 | if (!bh_in) | |
384 | brelse(bh); | |
385 | return -EIO; | |
386 | } | |
387 | set_buffer_revoked(bh); | |
388 | set_buffer_revokevalid(bh); | |
389 | if (bh_in) { | |
f7f4bccb MC |
390 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
391 | jbd2_journal_forget(handle, bh_in); | |
470decc6 DK |
392 | } else { |
393 | BUFFER_TRACE(bh, "call brelse"); | |
394 | __brelse(bh); | |
395 | } | |
396 | } | |
397 | ||
29971769 | 398 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
470decc6 DK |
399 | err = insert_revoke_hash(journal, blocknr, |
400 | handle->h_transaction->t_tid); | |
401 | BUFFER_TRACE(bh_in, "exit"); | |
402 | return err; | |
403 | } | |
404 | ||
405 | /* | |
406 | * Cancel an outstanding revoke. For use only internally by the | |
f7f4bccb | 407 | * journaling code (called from jbd2_journal_get_write_access). |
470decc6 DK |
408 | * |
409 | * We trust buffer_revoked() on the buffer if the buffer is already | |
410 | * being journaled: if there is no revoke pending on the buffer, then we | |
411 | * don't do anything here. | |
412 | * | |
413 | * This would break if it were possible for a buffer to be revoked and | |
414 | * discarded, and then reallocated within the same transaction. In such | |
415 | * a case we would have lost the revoked bit, but when we arrived here | |
416 | * the second time we would still have a pending revoke to cancel. So, | |
417 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
418 | * set. | |
470decc6 | 419 | */ |
f7f4bccb | 420 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
470decc6 | 421 | { |
f7f4bccb | 422 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
423 | journal_t *journal = handle->h_transaction->t_journal; |
424 | int need_cancel; | |
425 | int did_revoke = 0; /* akpm: debug */ | |
426 | struct buffer_head *bh = jh2bh(jh); | |
427 | ||
428 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
429 | ||
430 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
431 | * only perform the full cancel if the revoke bit is set. If | |
432 | * not, we can't trust the revoke bit, and we need to do the | |
433 | * full search for a revoke record. */ | |
434 | if (test_set_buffer_revokevalid(bh)) { | |
435 | need_cancel = test_clear_buffer_revoked(bh); | |
436 | } else { | |
437 | need_cancel = 1; | |
438 | clear_buffer_revoked(bh); | |
439 | } | |
440 | ||
441 | if (need_cancel) { | |
442 | record = find_revoke_record(journal, bh->b_blocknr); | |
443 | if (record) { | |
444 | jbd_debug(4, "cancelled existing revoke on " | |
445 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
446 | spin_lock(&journal->j_revoke_lock); | |
447 | list_del(&record->hash); | |
448 | spin_unlock(&journal->j_revoke_lock); | |
f7f4bccb | 449 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
450 | did_revoke = 1; |
451 | } | |
452 | } | |
453 | ||
cd02ff0b | 454 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
455 | /* There better not be one left behind by now! */ |
456 | record = find_revoke_record(journal, bh->b_blocknr); | |
457 | J_ASSERT_JH(jh, record == NULL); | |
458 | #endif | |
459 | ||
460 | /* Finally, have we just cleared revoke on an unhashed | |
461 | * buffer_head? If so, we'd better make sure we clear the | |
462 | * revoked status on any hashed alias too, otherwise the revoke | |
463 | * state machine will get very upset later on. */ | |
464 | if (need_cancel) { | |
465 | struct buffer_head *bh2; | |
466 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
467 | if (bh2) { | |
468 | if (bh2 != bh) | |
469 | clear_buffer_revoked(bh2); | |
470 | __brelse(bh2); | |
471 | } | |
472 | } | |
473 | return did_revoke; | |
474 | } | |
475 | ||
1ba37268 YY |
476 | /* |
477 | * journal_clear_revoked_flag clears revoked flag of buffers in | |
478 | * revoke table to reflect there is no revoked buffers in the next | |
479 | * transaction which is going to be started. | |
480 | */ | |
481 | void jbd2_clear_buffer_revoked_flags(journal_t *journal) | |
482 | { | |
483 | struct jbd2_revoke_table_s *revoke = journal->j_revoke; | |
484 | int i = 0; | |
485 | ||
486 | for (i = 0; i < revoke->hash_size; i++) { | |
487 | struct list_head *hash_list; | |
488 | struct list_head *list_entry; | |
489 | hash_list = &revoke->hash_table[i]; | |
490 | ||
491 | list_for_each(list_entry, hash_list) { | |
492 | struct jbd2_revoke_record_s *record; | |
493 | struct buffer_head *bh; | |
494 | record = (struct jbd2_revoke_record_s *)list_entry; | |
495 | bh = __find_get_block(journal->j_fs_dev, | |
496 | record->blocknr, | |
497 | journal->j_blocksize); | |
498 | if (bh) { | |
499 | clear_buffer_revoked(bh); | |
500 | __brelse(bh); | |
501 | } | |
502 | } | |
503 | } | |
504 | } | |
505 | ||
470decc6 DK |
506 | /* journal_switch_revoke table select j_revoke for next transaction |
507 | * we do not want to suspend any processing until all revokes are | |
508 | * written -bzzz | |
509 | */ | |
f7f4bccb | 510 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
470decc6 DK |
511 | { |
512 | int i; | |
513 | ||
514 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
515 | journal->j_revoke = journal->j_revoke_table[1]; | |
516 | else | |
517 | journal->j_revoke = journal->j_revoke_table[0]; | |
518 | ||
519 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
520 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
521 | } | |
522 | ||
523 | /* | |
524 | * Write revoke records to the journal for all entries in the current | |
525 | * revoke hash, deleting the entries as we go. | |
470decc6 | 526 | */ |
f7f4bccb | 527 | void jbd2_journal_write_revoke_records(journal_t *journal, |
67c457a8 | 528 | transaction_t *transaction, |
e5a120ae | 529 | struct list_head *log_bufs, |
67c457a8 | 530 | int write_op) |
470decc6 | 531 | { |
e5a120ae | 532 | struct buffer_head *descriptor; |
f7f4bccb MC |
533 | struct jbd2_revoke_record_s *record; |
534 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
535 | struct list_head *hash_list; |
536 | int i, offset, count; | |
537 | ||
538 | descriptor = NULL; | |
539 | offset = 0; | |
540 | count = 0; | |
541 | ||
542 | /* select revoke table for committing transaction */ | |
543 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
544 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
545 | ||
546 | for (i = 0; i < revoke->hash_size; i++) { | |
547 | hash_list = &revoke->hash_table[i]; | |
548 | ||
549 | while (!list_empty(hash_list)) { | |
f7f4bccb | 550 | record = (struct jbd2_revoke_record_s *) |
470decc6 | 551 | hash_list->next; |
e5a120ae | 552 | write_one_revoke_record(journal, transaction, log_bufs, |
470decc6 | 553 | &descriptor, &offset, |
67c457a8 | 554 | record, write_op); |
470decc6 DK |
555 | count++; |
556 | list_del(&record->hash); | |
f7f4bccb | 557 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
558 | } |
559 | } | |
560 | if (descriptor) | |
67c457a8 | 561 | flush_descriptor(journal, descriptor, offset, write_op); |
470decc6 DK |
562 | jbd_debug(1, "Wrote %d revoke records\n", count); |
563 | } | |
564 | ||
565 | /* | |
566 | * Write out one revoke record. We need to create a new descriptor | |
567 | * block if the old one is full or if we have not already created one. | |
568 | */ | |
569 | ||
570 | static void write_one_revoke_record(journal_t *journal, | |
571 | transaction_t *transaction, | |
e5a120ae JK |
572 | struct list_head *log_bufs, |
573 | struct buffer_head **descriptorp, | |
470decc6 | 574 | int *offsetp, |
67c457a8 TT |
575 | struct jbd2_revoke_record_s *record, |
576 | int write_op) | |
470decc6 | 577 | { |
42a7106d | 578 | int csum_size = 0; |
e5a120ae | 579 | struct buffer_head *descriptor; |
470decc6 DK |
580 | int offset; |
581 | journal_header_t *header; | |
582 | ||
583 | /* If we are already aborting, this all becomes a noop. We | |
584 | still need to go round the loop in | |
f7f4bccb | 585 | jbd2_journal_write_revoke_records in order to free all of the |
470decc6 DK |
586 | revoke records: only the IO to the journal is omitted. */ |
587 | if (is_journal_aborted(journal)) | |
588 | return; | |
589 | ||
590 | descriptor = *descriptorp; | |
591 | offset = *offsetp; | |
592 | ||
42a7106d | 593 | /* Do we need to leave space at the end for a checksum? */ |
db9ee220 | 594 | if (jbd2_journal_has_csum_v2or3(journal)) |
42a7106d DW |
595 | csum_size = sizeof(struct jbd2_journal_revoke_tail); |
596 | ||
470decc6 DK |
597 | /* Make sure we have a descriptor with space left for the record */ |
598 | if (descriptor) { | |
42a7106d | 599 | if (offset >= 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 | ||
b517bea1 | 622 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) { |
e5a120ae | 623 | * ((__be64 *)(&descriptor->b_data[offset])) = |
b517bea1 ZB |
624 | cpu_to_be64(record->blocknr); |
625 | offset += 8; | |
626 | ||
627 | } else { | |
e5a120ae | 628 | * ((__be32 *)(&descriptor->b_data[offset])) = |
b517bea1 ZB |
629 | cpu_to_be32(record->blocknr); |
630 | offset += 4; | |
631 | } | |
632 | ||
470decc6 DK |
633 | *offsetp = offset; |
634 | } | |
635 | ||
e5a120ae | 636 | static void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh) |
42a7106d DW |
637 | { |
638 | struct jbd2_journal_revoke_tail *tail; | |
639 | __u32 csum; | |
640 | ||
db9ee220 | 641 | if (!jbd2_journal_has_csum_v2or3(j)) |
42a7106d DW |
642 | return; |
643 | ||
e5a120ae | 644 | tail = (struct jbd2_journal_revoke_tail *)(bh->b_data + j->j_blocksize - |
42a7106d DW |
645 | sizeof(struct jbd2_journal_revoke_tail)); |
646 | tail->r_checksum = 0; | |
e5a120ae | 647 | csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); |
42a7106d DW |
648 | tail->r_checksum = cpu_to_be32(csum); |
649 | } | |
650 | ||
470decc6 DK |
651 | /* |
652 | * Flush a revoke descriptor out to the journal. If we are aborting, | |
653 | * this is a noop; otherwise we are generating a buffer which needs to | |
654 | * be waited for during commit, so it has to go onto the appropriate | |
655 | * journal buffer list. | |
656 | */ | |
657 | ||
658 | static void flush_descriptor(journal_t *journal, | |
e5a120ae | 659 | struct buffer_head *descriptor, |
67c457a8 | 660 | int offset, int write_op) |
470decc6 | 661 | { |
f7f4bccb | 662 | jbd2_journal_revoke_header_t *header; |
470decc6 DK |
663 | |
664 | if (is_journal_aborted(journal)) { | |
e5a120ae | 665 | put_bh(descriptor); |
470decc6 DK |
666 | return; |
667 | } | |
668 | ||
e5a120ae | 669 | header = (jbd2_journal_revoke_header_t *)descriptor->b_data; |
470decc6 | 670 | header->r_count = cpu_to_be32(offset); |
42a7106d DW |
671 | jbd2_revoke_csum_set(journal, descriptor); |
672 | ||
e5a120ae JK |
673 | set_buffer_jwrite(descriptor); |
674 | BUFFER_TRACE(descriptor, "write"); | |
675 | set_buffer_dirty(descriptor); | |
676 | write_dirty_buffer(descriptor, write_op); | |
470decc6 DK |
677 | } |
678 | #endif | |
679 | ||
680 | /* | |
681 | * Revoke support for recovery. | |
682 | * | |
683 | * Recovery needs to be able to: | |
684 | * | |
685 | * record all revoke records, including the tid of the latest instance | |
686 | * of each revoke in the journal | |
687 | * | |
688 | * check whether a given block in a given transaction should be replayed | |
689 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
690 | * transaction) | |
691 | * | |
692 | * empty the revoke table after recovery. | |
693 | */ | |
694 | ||
695 | /* | |
696 | * First, setting revoke records. We create a new revoke record for | |
697 | * every block ever revoked in the log as we scan it for recovery, and | |
698 | * we update the existing records if we find multiple revokes for a | |
699 | * single block. | |
700 | */ | |
701 | ||
f7f4bccb | 702 | int jbd2_journal_set_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 | /* If we have multiple occurrences, only record the | |
711 | * latest sequence number in the hashed record */ | |
712 | if (tid_gt(sequence, record->sequence)) | |
713 | record->sequence = sequence; | |
714 | return 0; | |
715 | } | |
716 | return insert_revoke_hash(journal, blocknr, sequence); | |
717 | } | |
718 | ||
719 | /* | |
720 | * Test revoke records. For a given block referenced in the log, has | |
721 | * that block been revoked? A revoke record with a given transaction | |
722 | * sequence number revokes all blocks in that transaction and earlier | |
723 | * ones, but later transactions still need replayed. | |
724 | */ | |
725 | ||
f7f4bccb | 726 | int jbd2_journal_test_revoke(journal_t *journal, |
18eba7aa | 727 | unsigned long long blocknr, |
470decc6 DK |
728 | tid_t sequence) |
729 | { | |
f7f4bccb | 730 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
731 | |
732 | record = find_revoke_record(journal, blocknr); | |
733 | if (!record) | |
734 | return 0; | |
735 | if (tid_gt(sequence, record->sequence)) | |
736 | return 0; | |
737 | return 1; | |
738 | } | |
739 | ||
740 | /* | |
741 | * Finally, once recovery is over, we need to clear the revoke table so | |
742 | * that it can be reused by the running filesystem. | |
743 | */ | |
744 | ||
f7f4bccb | 745 | void jbd2_journal_clear_revoke(journal_t *journal) |
470decc6 DK |
746 | { |
747 | int i; | |
748 | struct list_head *hash_list; | |
f7f4bccb MC |
749 | struct jbd2_revoke_record_s *record; |
750 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
751 | |
752 | revoke = journal->j_revoke; | |
753 | ||
754 | for (i = 0; i < revoke->hash_size; i++) { | |
755 | hash_list = &revoke->hash_table[i]; | |
756 | while (!list_empty(hash_list)) { | |
f7f4bccb | 757 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
470decc6 | 758 | list_del(&record->hash); |
f7f4bccb | 759 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
760 | } |
761 | } | |
762 | } |