]>
Commit | Line | Data |
---|---|---|
0e8a9560 | 1 | /* |
97f168b6 | 2 | * linux/fs/jbd2/revoke.c |
efc6f628 | 3 | * |
0e8a9560 TT |
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: | |
efc6f628 | 18 | * |
0e8a9560 TT |
19 | * + Commit: during commit we write the entire list of the current |
20 | * transaction's revoked blocks to the journal | |
efc6f628 | 21 | * |
0e8a9560 TT |
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: | |
efc6f628 | 32 | * The desired end result is the journaling of the new block, so we |
0e8a9560 TT |
33 | * cancel the revoke before the transaction commits. |
34 | * | |
35 | * Block is journaled and then revoked: | |
725c474f TT |
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 | |
0e8a9560 | 38 | * later in the log than the log block. In this case, we choose the |
725c474f TT |
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. | |
0e8a9560 | 43 | * |
efc6f628 | 44 | * Block is revoked and then written as data: |
0e8a9560 TT |
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 | * | |
97f168b6 DW |
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 | * | |
0e8a9560 TT |
54 | * Revoke information on buffers is a tri-state value: |
55 | * | |
56 | * RevokeValid clear: no cached revoke status, need to look it up | |
8cf93332 | 57 | * RevokeValid set, Revoked clear: |
0e8a9560 TT |
58 | * buffer has not been revoked, and cancel_revoke |
59 | * need do nothing. | |
8cf93332 | 60 | * RevokeValid set, Revoked set: |
efc6f628 | 61 | * buffer has been revoked. |
97f168b6 DW |
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 | * | |
78 | * Finally, also replay code uses the hash tables but at this moment no one else | |
79 | * can touch them (filesystem isn't mounted yet) and hence no locking is | |
80 | * needed. | |
0e8a9560 TT |
81 | */ |
82 | ||
83 | #ifndef __KERNEL__ | |
84 | #include "jfs_user.h" | |
85 | #else | |
97f168b6 | 86 | #include <linux/time.h> |
0e8a9560 | 87 | #include <linux/fs.h> |
97f168b6 | 88 | #include <linux/jbd2.h> |
0e8a9560 TT |
89 | #include <linux/errno.h> |
90 | #include <linux/slab.h> | |
0e8a9560 | 91 | #include <linux/list.h> |
8cf93332 | 92 | #include <linux/init.h> |
97f168b6 DW |
93 | #include <linux/bio.h> |
94 | #include <linux/log2.h> | |
0e8a9560 TT |
95 | #endif |
96 | ||
97f168b6 DW |
97 | static lkmem_cache_t *jbd2_revoke_record_cache; |
98 | static lkmem_cache_t *jbd2_revoke_table_cache; | |
0e8a9560 TT |
99 | |
100 | /* Each revoke record represents one single revoked block. During | |
101 | journal replay, this involves recording the transaction ID of the | |
102 | last transaction to revoke this block. */ | |
103 | ||
97f168b6 | 104 | struct jbd2_revoke_record_s |
0e8a9560 TT |
105 | { |
106 | struct list_head hash; | |
107 | tid_t sequence; /* Used for recovery only */ | |
97f168b6 | 108 | unsigned long long blocknr; |
0e8a9560 TT |
109 | }; |
110 | ||
111 | ||
112 | /* The revoke table is just a simple hash table of revoke records. */ | |
97f168b6 | 113 | struct jbd2_revoke_table_s |
0e8a9560 TT |
114 | { |
115 | /* It is conceivable that we might want a larger hash table | |
116 | * for recovery. Must be a power of two. */ | |
efc6f628 TT |
117 | int hash_size; |
118 | int hash_shift; | |
0e8a9560 TT |
119 | struct list_head *hash_table; |
120 | }; | |
121 | ||
122 | ||
123 | #ifdef __KERNEL__ | |
124 | static void write_one_revoke_record(journal_t *, transaction_t *, | |
97f168b6 DW |
125 | struct list_head *, |
126 | struct buffer_head **, int *, | |
127 | struct jbd2_revoke_record_s *, int); | |
128 | static void flush_descriptor(journal_t *, struct buffer_head *, int, int); | |
0e8a9560 TT |
129 | #endif |
130 | ||
131 | /* Utility functions to maintain the revoke table */ | |
132 | ||
133 | /* Borrowed from buffer.c: this is a tried and tested block hash function */ | |
97f168b6 | 134 | static inline int hash(journal_t *journal, unsigned long long block) |
0e8a9560 | 135 | { |
97f168b6 | 136 | struct jbd2_revoke_table_s *table = journal->j_revoke; |
0e8a9560 | 137 | int hash_shift = table->hash_shift; |
97f168b6 | 138 | int hash = (int)block ^ (int)((block >> 31) >> 1); |
efc6f628 | 139 | |
97f168b6 DW |
140 | return ((hash << (hash_shift - 6)) ^ |
141 | (hash >> 13) ^ | |
142 | (hash << (hash_shift - 12))) & (table->hash_size - 1); | |
0e8a9560 TT |
143 | } |
144 | ||
97f168b6 | 145 | static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr, |
3e699064 | 146 | tid_t seq) |
0e8a9560 TT |
147 | { |
148 | struct list_head *hash_list; | |
97f168b6 | 149 | struct jbd2_revoke_record_s *record; |
8cf93332 TT |
150 | |
151 | repeat: | |
97f168b6 | 152 | record = kmem_cache_alloc(jbd2_revoke_record_cache, GFP_NOFS); |
0e8a9560 | 153 | if (!record) |
8cf93332 | 154 | goto oom; |
0e8a9560 TT |
155 | |
156 | record->sequence = seq; | |
157 | record->blocknr = blocknr; | |
158 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
97f168b6 | 159 | spin_lock(&journal->j_revoke_lock); |
0e8a9560 | 160 | list_add(&record->hash, hash_list); |
97f168b6 | 161 | spin_unlock(&journal->j_revoke_lock); |
0e8a9560 | 162 | return 0; |
8cf93332 TT |
163 | |
164 | oom: | |
8cf93332 TT |
165 | if (!journal_oom_retry) |
166 | return -ENOMEM; | |
97f168b6 DW |
167 | jbd_debug(1, "ENOMEM in %s, retrying\n", __func__); |
168 | yield(); | |
8cf93332 | 169 | goto repeat; |
0e8a9560 TT |
170 | } |
171 | ||
172 | /* Find a revoke record in the journal's hash table. */ | |
173 | ||
97f168b6 DW |
174 | static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal, |
175 | unsigned long long blocknr) | |
0e8a9560 TT |
176 | { |
177 | struct list_head *hash_list; | |
97f168b6 | 178 | struct jbd2_revoke_record_s *record; |
efc6f628 | 179 | |
0e8a9560 TT |
180 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; |
181 | ||
97f168b6 DW |
182 | spin_lock(&journal->j_revoke_lock); |
183 | record = (struct jbd2_revoke_record_s *) hash_list->next; | |
0e8a9560 | 184 | while (&(record->hash) != hash_list) { |
97f168b6 DW |
185 | if (record->blocknr == blocknr) { |
186 | spin_unlock(&journal->j_revoke_lock); | |
0e8a9560 | 187 | return record; |
97f168b6 DW |
188 | } |
189 | record = (struct jbd2_revoke_record_s *) record->hash.next; | |
0e8a9560 | 190 | } |
97f168b6 | 191 | spin_unlock(&journal->j_revoke_lock); |
0e8a9560 TT |
192 | return NULL; |
193 | } | |
194 | ||
97f168b6 | 195 | void journal_destroy_revoke_caches(void) |
8cf93332 | 196 | { |
97f168b6 DW |
197 | if (jbd2_revoke_record_cache) { |
198 | kmem_cache_destroy(jbd2_revoke_record_cache); | |
199 | jbd2_revoke_record_cache = NULL; | |
200 | } | |
201 | if (jbd2_revoke_table_cache) { | |
202 | kmem_cache_destroy(jbd2_revoke_table_cache); | |
203 | jbd2_revoke_table_cache = NULL; | |
8cf93332 | 204 | } |
efc6f628 | 205 | } |
8cf93332 | 206 | |
97f168b6 | 207 | int __init journal_init_revoke_caches(void) |
8cf93332 | 208 | { |
97f168b6 DW |
209 | J_ASSERT(!jbd2_revoke_record_cache); |
210 | J_ASSERT(!jbd2_revoke_table_cache); | |
211 | ||
212 | jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s, | |
213 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY); | |
214 | if (!jbd2_revoke_record_cache) | |
215 | goto record_cache_failure; | |
216 | ||
217 | jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s, | |
218 | SLAB_TEMPORARY); | |
219 | if (!jbd2_revoke_table_cache) | |
220 | goto table_cache_failure; | |
221 | return 0; | |
222 | table_cache_failure: | |
223 | journal_destroy_revoke_caches(); | |
224 | record_cache_failure: | |
225 | return -ENOMEM; | |
8cf93332 | 226 | } |
0e8a9560 | 227 | |
97f168b6 | 228 | static struct jbd2_revoke_table_s *journal_init_revoke_table(int hash_size) |
0e8a9560 | 229 | { |
97f168b6 DW |
230 | int shift = 0; |
231 | int tmp = hash_size; | |
232 | struct jbd2_revoke_table_s *table; | |
efc6f628 | 233 | |
97f168b6 DW |
234 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
235 | if (!table) | |
236 | goto out; | |
0e8a9560 | 237 | |
0e8a9560 TT |
238 | while((tmp >>= 1UL) != 0UL) |
239 | shift++; | |
0e8a9560 | 240 | |
97f168b6 DW |
241 | table->hash_size = hash_size; |
242 | table->hash_shift = shift; | |
243 | table->hash_table = | |
0e8a9560 | 244 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); |
97f168b6 DW |
245 | if (!table->hash_table) { |
246 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
247 | table = NULL; | |
248 | goto out; | |
0e8a9560 | 249 | } |
efc6f628 | 250 | |
0e8a9560 | 251 | for (tmp = 0; tmp < hash_size; tmp++) |
97f168b6 | 252 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
efc6f628 | 253 | |
97f168b6 DW |
254 | out: |
255 | return table; | |
0e8a9560 TT |
256 | } |
257 | ||
97f168b6 | 258 | static void journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) |
0e8a9560 | 259 | { |
0e8a9560 | 260 | int i; |
97f168b6 | 261 | struct list_head *hash_list; |
efc6f628 | 262 | |
97f168b6 | 263 | for (i = 0; i < table->hash_size; i++) { |
0e8a9560 | 264 | hash_list = &table->hash_table[i]; |
97f168b6 | 265 | J_ASSERT(list_empty(hash_list)); |
0e8a9560 | 266 | } |
efc6f628 | 267 | |
0e8a9560 | 268 | kfree(table->hash_table); |
97f168b6 DW |
269 | kmem_cache_free(jbd2_revoke_table_cache, table); |
270 | } | |
271 | ||
272 | /* Initialise the revoke table for a given journal to a given size. */ | |
273 | int journal_init_revoke(journal_t *journal, int hash_size) | |
274 | { | |
275 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
276 | J_ASSERT(is_power_of_2(hash_size)); | |
277 | ||
278 | journal->j_revoke_table[0] = journal_init_revoke_table(hash_size); | |
279 | if (!journal->j_revoke_table[0]) | |
280 | goto fail0; | |
281 | ||
282 | journal->j_revoke_table[1] = journal_init_revoke_table(hash_size); | |
283 | if (!journal->j_revoke_table[1]) | |
284 | goto fail1; | |
285 | ||
286 | journal->j_revoke = journal->j_revoke_table[1]; | |
287 | ||
288 | spin_lock_init(&journal->j_revoke_lock); | |
289 | ||
290 | return 0; | |
291 | ||
292 | fail1: | |
293 | journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
294 | fail0: | |
295 | return -ENOMEM; | |
296 | } | |
297 | ||
298 | /* Destroy a journal's revoke table. The table must already be empty! */ | |
299 | void journal_destroy_revoke(journal_t *journal) | |
300 | { | |
0e8a9560 | 301 | journal->j_revoke = NULL; |
97f168b6 DW |
302 | if (journal->j_revoke_table[0]) |
303 | journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
304 | if (journal->j_revoke_table[1]) | |
305 | journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
0e8a9560 TT |
306 | } |
307 | ||
308 | ||
309 | #ifdef __KERNEL__ | |
310 | ||
efc6f628 | 311 | /* |
0e8a9560 TT |
312 | * journal_revoke: revoke a given buffer_head from the journal. This |
313 | * prevents the block from being replayed during recovery if we take a | |
314 | * crash after this current transaction commits. Any subsequent | |
315 | * metadata writes of the buffer in this transaction cancel the | |
efc6f628 | 316 | * revoke. |
0e8a9560 TT |
317 | * |
318 | * Note that this call may block --- it is up to the caller to make | |
319 | * sure that there are no further calls to journal_write_metadata | |
320 | * before the revoke is complete. In ext3, this implies calling the | |
321 | * revoke before clearing the block bitmap when we are deleting | |
efc6f628 | 322 | * metadata. |
0e8a9560 TT |
323 | * |
324 | * Revoke performs a journal_forget on any buffer_head passed in as a | |
325 | * parameter, but does _not_ forget the buffer_head if the bh was only | |
efc6f628 | 326 | * found implicitly. |
0e8a9560 | 327 | * |
8cf93332 TT |
328 | * bh_in may not be a journalled buffer - it may have come off |
329 | * the hash tables without an attached journal_head. | |
330 | * | |
331 | * If bh_in is non-zero, journal_revoke() will decrement its b_count | |
332 | * by one. | |
0e8a9560 TT |
333 | */ |
334 | ||
97f168b6 | 335 | int journal_revoke(handle_t *handle, unsigned long long blocknr, |
0e8a9560 TT |
336 | struct buffer_head *bh_in) |
337 | { | |
8cf93332 | 338 | struct buffer_head *bh = NULL; |
0e8a9560 | 339 | journal_t *journal; |
97f168b6 | 340 | struct block_device *bdev; |
0e8a9560 TT |
341 | int err; |
342 | ||
97f168b6 | 343 | might_sleep(); |
8cf93332 TT |
344 | if (bh_in) |
345 | BUFFER_TRACE(bh_in, "enter"); | |
346 | ||
0e8a9560 | 347 | journal = handle->h_transaction->t_journal; |
1f735038 TT |
348 | if (!journal_set_features(journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)){ |
349 | J_ASSERT (!"Cannot set revoke feature!"); | |
0e8a9560 | 350 | return -EINVAL; |
1f735038 | 351 | } |
8cf93332 | 352 | |
97f168b6 | 353 | bdev = journal->j_fs_dev; |
0e8a9560 TT |
354 | bh = bh_in; |
355 | ||
8cf93332 | 356 | if (!bh) { |
97f168b6 | 357 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); |
8cf93332 TT |
358 | if (bh) |
359 | BUFFER_TRACE(bh, "found on hash"); | |
360 | } | |
97f168b6 | 361 | #ifdef JFS_EXPENSIVE_CHECKING |
8cf93332 TT |
362 | else { |
363 | struct buffer_head *bh2; | |
364 | ||
365 | /* If there is a different buffer_head lying around in | |
366 | * memory anywhere... */ | |
97f168b6 | 367 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); |
8cf93332 TT |
368 | if (bh2) { |
369 | /* ... and it has RevokeValid status... */ | |
97f168b6 | 370 | if (bh2 != bh && buffer_revokevalid(bh2)) |
8cf93332 TT |
371 | /* ...then it better be revoked too, |
372 | * since it's illegal to create a revoke | |
373 | * record against a buffer_head which is | |
374 | * not marked revoked --- that would | |
375 | * risk missing a subsequent revoke | |
376 | * cancel. */ | |
97f168b6 DW |
377 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); |
378 | put_bh(bh2); | |
8cf93332 TT |
379 | } |
380 | } | |
381 | #endif | |
0e8a9560 TT |
382 | |
383 | /* We really ought not ever to revoke twice in a row without | |
384 | first having the revoke cancelled: it's illegal to free a | |
385 | block twice without allocating it in between! */ | |
386 | if (bh) { | |
97f168b6 DW |
387 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), |
388 | "inconsistent data on disk")) { | |
389 | if (!bh_in) | |
390 | brelse(bh); | |
391 | return -EIO; | |
392 | } | |
393 | set_buffer_revoked(bh); | |
394 | set_buffer_revokevalid(bh); | |
8cf93332 TT |
395 | if (bh_in) { |
396 | BUFFER_TRACE(bh_in, "call journal_forget"); | |
0e8a9560 | 397 | journal_forget(handle, bh_in); |
8cf93332 TT |
398 | } else { |
399 | BUFFER_TRACE(bh, "call brelse"); | |
400 | __brelse(bh); | |
401 | } | |
0e8a9560 TT |
402 | } |
403 | ||
97f168b6 | 404 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
8cf93332 TT |
405 | err = insert_revoke_hash(journal, blocknr, |
406 | handle->h_transaction->t_tid); | |
8cf93332 | 407 | BUFFER_TRACE(bh_in, "exit"); |
0e8a9560 TT |
408 | return err; |
409 | } | |
410 | ||
0e8a9560 TT |
411 | /* |
412 | * Cancel an outstanding revoke. For use only internally by the | |
413 | * journaling code (called from journal_get_write_access). | |
414 | * | |
97f168b6 | 415 | * We trust buffer_revoked() on the buffer if the buffer is already |
0e8a9560 TT |
416 | * being journaled: if there is no revoke pending on the buffer, then we |
417 | * don't do anything here. | |
418 | * | |
419 | * This would break if it were possible for a buffer to be revoked and | |
420 | * discarded, and then reallocated within the same transaction. In such | |
421 | * a case we would have lost the revoked bit, but when we arrived here | |
422 | * the second time we would still have a pending revoke to cancel. So, | |
423 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
424 | * set. | |
8cf93332 TT |
425 | */ |
426 | int journal_cancel_revoke(handle_t *handle, struct journal_head *jh) | |
0e8a9560 | 427 | { |
97f168b6 | 428 | struct jbd2_revoke_record_s *record; |
0e8a9560 TT |
429 | journal_t *journal = handle->h_transaction->t_journal; |
430 | int need_cancel; | |
8cf93332 TT |
431 | int did_revoke = 0; /* akpm: debug */ |
432 | struct buffer_head *bh = jh2bh(jh); | |
efc6f628 | 433 | |
8cf93332 TT |
434 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); |
435 | ||
0e8a9560 TT |
436 | /* Is the existing Revoke bit valid? If so, we trust it, and |
437 | * only perform the full cancel if the revoke bit is set. If | |
438 | * not, we can't trust the revoke bit, and we need to do the | |
439 | * full search for a revoke record. */ | |
97f168b6 DW |
440 | if (test_set_buffer_revokevalid(bh)) { |
441 | need_cancel = test_clear_buffer_revoked(bh); | |
442 | } else { | |
0e8a9560 | 443 | need_cancel = 1; |
97f168b6 | 444 | clear_buffer_revoked(bh); |
0e8a9560 | 445 | } |
8cf93332 | 446 | |
0e8a9560 TT |
447 | if (need_cancel) { |
448 | record = find_revoke_record(journal, bh->b_blocknr); | |
449 | if (record) { | |
8cf93332 | 450 | jbd_debug(4, "cancelled existing revoke on " |
97f168b6 DW |
451 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); |
452 | spin_lock(&journal->j_revoke_lock); | |
0e8a9560 | 453 | list_del(&record->hash); |
97f168b6 DW |
454 | spin_unlock(&journal->j_revoke_lock); |
455 | kmem_cache_free(jbd2_revoke_record_cache, record); | |
8cf93332 | 456 | did_revoke = 1; |
0e8a9560 TT |
457 | } |
458 | } | |
8cf93332 | 459 | |
97f168b6 | 460 | #ifdef JFS_EXPENSIVE_CHECKING |
8cf93332 TT |
461 | /* There better not be one left behind by now! */ |
462 | record = find_revoke_record(journal, bh->b_blocknr); | |
463 | J_ASSERT_JH(jh, record == NULL); | |
464 | #endif | |
465 | ||
466 | /* Finally, have we just cleared revoke on an unhashed | |
467 | * buffer_head? If so, we'd better make sure we clear the | |
468 | * revoked status on any hashed alias too, otherwise the revoke | |
469 | * state machine will get very upset later on. */ | |
97f168b6 | 470 | if (need_cancel) { |
8cf93332 | 471 | struct buffer_head *bh2; |
97f168b6 | 472 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); |
8cf93332 | 473 | if (bh2) { |
97f168b6 DW |
474 | if (bh2 != bh) |
475 | clear_buffer_revoked(bh2); | |
8cf93332 TT |
476 | __brelse(bh2); |
477 | } | |
478 | } | |
8cf93332 | 479 | return did_revoke; |
0e8a9560 TT |
480 | } |
481 | ||
97f168b6 DW |
482 | /* |
483 | * journal_clear_revoked_flag clears revoked flag of buffers in | |
484 | * revoke table to reflect there is no revoked buffers in the next | |
485 | * transaction which is going to be started. | |
486 | */ | |
487 | void jbd2_clear_buffer_revoked_flags(journal_t *journal) | |
488 | { | |
489 | struct jbd2_revoke_table_s *revoke = journal->j_revoke; | |
490 | int i = 0; | |
491 | ||
492 | for (i = 0; i < revoke->hash_size; i++) { | |
493 | struct list_head *hash_list; | |
494 | struct list_head *list_entry; | |
495 | hash_list = &revoke->hash_table[i]; | |
496 | ||
497 | list_for_each(list_entry, hash_list) { | |
498 | struct jbd2_revoke_record_s *record; | |
499 | struct buffer_head *bh; | |
500 | record = (struct jbd2_revoke_record_s *)list_entry; | |
501 | bh = __find_get_block(journal->j_fs_dev, | |
502 | record->blocknr, | |
503 | journal->j_blocksize); | |
504 | if (bh) { | |
505 | clear_buffer_revoked(bh); | |
506 | __brelse(bh); | |
507 | } | |
508 | } | |
509 | } | |
510 | } | |
511 | ||
512 | /* journal_switch_revoke table select j_revoke for next transaction | |
513 | * we do not want to suspend any processing until all revokes are | |
514 | * written -bzzz | |
515 | */ | |
516 | void journal_switch_revoke_table(journal_t *journal) | |
517 | { | |
518 | int i; | |
519 | ||
520 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
521 | journal->j_revoke = journal->j_revoke_table[1]; | |
522 | else | |
523 | journal->j_revoke = journal->j_revoke_table[0]; | |
524 | ||
525 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
526 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
527 | } | |
0e8a9560 TT |
528 | |
529 | /* | |
530 | * Write revoke records to the journal for all entries in the current | |
531 | * revoke hash, deleting the entries as we go. | |
0e8a9560 | 532 | */ |
efc6f628 | 533 | void journal_write_revoke_records(journal_t *journal, |
97f168b6 DW |
534 | transaction_t *transaction, |
535 | struct list_head *log_bufs, | |
536 | int write_op) | |
0e8a9560 | 537 | { |
97f168b6 DW |
538 | struct buffer_head *descriptor; |
539 | struct jbd2_revoke_record_s *record; | |
540 | struct jbd2_revoke_table_s *revoke; | |
0e8a9560 | 541 | struct list_head *hash_list; |
1f735038 | 542 | int i, offset, count; |
8cf93332 | 543 | |
efc6f628 | 544 | descriptor = NULL; |
0e8a9560 | 545 | offset = 0; |
1f735038 | 546 | count = 0; |
97f168b6 DW |
547 | |
548 | /* select revoke table for committing transaction */ | |
549 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
550 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
efc6f628 | 551 | |
0e8a9560 TT |
552 | for (i = 0; i < revoke->hash_size; i++) { |
553 | hash_list = &revoke->hash_table[i]; | |
554 | ||
555 | while (!list_empty(hash_list)) { | |
97f168b6 | 556 | record = (struct jbd2_revoke_record_s *) |
0e8a9560 | 557 | hash_list->next; |
97f168b6 | 558 | write_one_revoke_record(journal, transaction, log_bufs, |
efc6f628 | 559 | &descriptor, &offset, |
97f168b6 | 560 | record, write_op); |
1f735038 | 561 | count++; |
0e8a9560 | 562 | list_del(&record->hash); |
97f168b6 | 563 | kmem_cache_free(jbd2_revoke_record_cache, record); |
0e8a9560 TT |
564 | } |
565 | } | |
efc6f628 | 566 | if (descriptor) |
97f168b6 | 567 | flush_descriptor(journal, descriptor, offset, write_op); |
8cf93332 | 568 | jbd_debug(1, "Wrote %d revoke records\n", count); |
0e8a9560 TT |
569 | } |
570 | ||
efc6f628 | 571 | /* |
0e8a9560 | 572 | * Write out one revoke record. We need to create a new descriptor |
efc6f628 | 573 | * block if the old one is full or if we have not already created one. |
0e8a9560 TT |
574 | */ |
575 | ||
efc6f628 | 576 | static void write_one_revoke_record(journal_t *journal, |
0e8a9560 | 577 | transaction_t *transaction, |
97f168b6 DW |
578 | struct list_head *log_bufs, |
579 | struct buffer_head **descriptorp, | |
0e8a9560 | 580 | int *offsetp, |
97f168b6 DW |
581 | struct jbd2_revoke_record_s *record, |
582 | int write_op) | |
0e8a9560 | 583 | { |
97f168b6 DW |
584 | int csum_size = 0; |
585 | struct buffer_head *descriptor; | |
04c66cb2 | 586 | int sz, offset; |
0e8a9560 | 587 | journal_header_t *header; |
8cf93332 | 588 | |
0e8a9560 TT |
589 | /* If we are already aborting, this all becomes a noop. We |
590 | still need to go round the loop in | |
591 | journal_write_revoke_records in order to free all of the | |
592 | revoke records: only the IO to the journal is omitted. */ | |
8cf93332 | 593 | if (is_journal_aborted(journal)) |
0e8a9560 TT |
594 | return; |
595 | ||
596 | descriptor = *descriptorp; | |
597 | offset = *offsetp; | |
8cf93332 | 598 | |
97f168b6 | 599 | /* Do we need to leave space at the end for a checksum? */ |
38d5adf3 | 600 | if (journal_has_csum_v2or3(journal)) |
97f168b6 DW |
601 | csum_size = sizeof(struct journal_revoke_tail); |
602 | ||
04c66cb2 DW |
603 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) |
604 | sz = 8; | |
605 | else | |
606 | sz = 4; | |
607 | ||
0e8a9560 TT |
608 | /* Make sure we have a descriptor with space left for the record */ |
609 | if (descriptor) { | |
04c66cb2 | 610 | if (offset + sz > journal->j_blocksize - csum_size) { |
97f168b6 | 611 | flush_descriptor(journal, descriptor, offset, write_op); |
0e8a9560 TT |
612 | descriptor = NULL; |
613 | } | |
614 | } | |
efc6f628 | 615 | |
0e8a9560 TT |
616 | if (!descriptor) { |
617 | descriptor = journal_get_descriptor_buffer(journal); | |
8cf93332 TT |
618 | if (!descriptor) |
619 | return; | |
97f168b6 DW |
620 | header = (journal_header_t *)descriptor->b_data; |
621 | header->h_magic = ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER); | |
622 | header->h_blocktype = ext2fs_cpu_to_be32(JFS_REVOKE_BLOCK); | |
623 | header->h_sequence = ext2fs_cpu_to_be32(transaction->t_tid); | |
0e8a9560 TT |
624 | |
625 | /* Record it so that we can wait for IO completion later */ | |
97f168b6 DW |
626 | BUFFER_TRACE(descriptor, "file in log_bufs"); |
627 | jbd2_file_log_bh(log_bufs, descriptor); | |
8cf93332 | 628 | |
0e8a9560 TT |
629 | offset = sizeof(journal_revoke_header_t); |
630 | *descriptorp = descriptor; | |
631 | } | |
efc6f628 | 632 | |
04c66cb2 DW |
633 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) { |
634 | * ((__be64 *)(&descriptor->b_data[offset])) = | |
635 | cpu_to_be64(record->blocknr); | |
636 | else | |
637 | * ((__be32 *)(&descriptor->b_data[offset])) = | |
638 | cpu_to_be32(record->blocknr); | |
639 | offset += sz; | |
97f168b6 | 640 | |
0e8a9560 TT |
641 | *offsetp = offset; |
642 | } | |
643 | ||
97f168b6 DW |
644 | static void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh) |
645 | { | |
646 | struct journal_revoke_tail *tail; | |
647 | __u32 csum; | |
648 | ||
38d5adf3 | 649 | if (!journal_has_csum_v2or3(j)) |
97f168b6 DW |
650 | return; |
651 | ||
652 | tail = (struct journal_revoke_tail *)(bh->b_data + j->j_blocksize - | |
653 | sizeof(struct journal_revoke_tail)); | |
654 | tail->r_checksum = 0; | |
655 | csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); | |
656 | tail->r_checksum = ext2fs_cpu_to_be32(csum); | |
657 | } | |
658 | ||
efc6f628 | 659 | /* |
0e8a9560 TT |
660 | * Flush a revoke descriptor out to the journal. If we are aborting, |
661 | * this is a noop; otherwise we are generating a buffer which needs to | |
662 | * be waited for during commit, so it has to go onto the appropriate | |
663 | * journal buffer list. | |
664 | */ | |
665 | ||
efc6f628 | 666 | static void flush_descriptor(journal_t *journal, |
97f168b6 DW |
667 | struct buffer_head *descriptor, |
668 | int offset, int write_op) | |
0e8a9560 TT |
669 | { |
670 | journal_revoke_header_t *header; | |
8cf93332 TT |
671 | |
672 | if (is_journal_aborted(journal)) { | |
97f168b6 | 673 | put_bh(descriptor); |
0e8a9560 TT |
674 | return; |
675 | } | |
efc6f628 | 676 | |
97f168b6 DW |
677 | header = (journal_revoke_header_t *)descriptor->b_data; |
678 | header->r_count = ext2fs_cpu_to_be32(offset); | |
679 | jbd2_revoke_csum_set(journal, descriptor); | |
0e8a9560 | 680 | |
97f168b6 DW |
681 | set_buffer_jwrite(descriptor); |
682 | BUFFER_TRACE(descriptor, "write"); | |
683 | set_buffer_dirty(descriptor); | |
684 | write_dirty_buffer(descriptor, write_op); | |
685 | } | |
0e8a9560 TT |
686 | #endif |
687 | ||
efc6f628 | 688 | /* |
0e8a9560 TT |
689 | * Revoke support for recovery. |
690 | * | |
691 | * Recovery needs to be able to: | |
692 | * | |
693 | * record all revoke records, including the tid of the latest instance | |
694 | * of each revoke in the journal | |
695 | * | |
696 | * check whether a given block in a given transaction should be replayed | |
697 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
698 | * transaction) | |
efc6f628 | 699 | * |
0e8a9560 TT |
700 | * empty the revoke table after recovery. |
701 | */ | |
702 | ||
703 | /* | |
704 | * First, setting revoke records. We create a new revoke record for | |
705 | * every block ever revoked in the log as we scan it for recovery, and | |
706 | * we update the existing records if we find multiple revokes for a | |
efc6f628 | 707 | * single block. |
0e8a9560 TT |
708 | */ |
709 | ||
efc6f628 | 710 | int journal_set_revoke(journal_t *journal, |
97f168b6 | 711 | unsigned long long blocknr, |
0e8a9560 TT |
712 | tid_t sequence) |
713 | { | |
97f168b6 | 714 | struct jbd2_revoke_record_s *record; |
efc6f628 | 715 | |
0e8a9560 TT |
716 | record = find_revoke_record(journal, blocknr); |
717 | if (record) { | |
97f168b6 | 718 | /* If we have multiple occurrences, only record the |
0e8a9560 | 719 | * latest sequence number in the hashed record */ |
725c474f | 720 | if (tid_gt(sequence, record->sequence)) |
0e8a9560 TT |
721 | record->sequence = sequence; |
722 | return 0; | |
efc6f628 | 723 | } |
0e8a9560 TT |
724 | return insert_revoke_hash(journal, blocknr, sequence); |
725 | } | |
726 | ||
efc6f628 | 727 | /* |
0e8a9560 TT |
728 | * Test revoke records. For a given block referenced in the log, has |
729 | * that block been revoked? A revoke record with a given transaction | |
730 | * sequence number revokes all blocks in that transaction and earlier | |
731 | * ones, but later transactions still need replayed. | |
732 | */ | |
733 | ||
efc6f628 | 734 | int journal_test_revoke(journal_t *journal, |
97f168b6 | 735 | unsigned long long blocknr, |
0e8a9560 TT |
736 | tid_t sequence) |
737 | { | |
97f168b6 | 738 | struct jbd2_revoke_record_s *record; |
efc6f628 | 739 | |
0e8a9560 TT |
740 | record = find_revoke_record(journal, blocknr); |
741 | if (!record) | |
742 | return 0; | |
725c474f | 743 | if (tid_gt(sequence, record->sequence)) |
0e8a9560 TT |
744 | return 0; |
745 | return 1; | |
746 | } | |
747 | ||
748 | /* | |
749 | * Finally, once recovery is over, we need to clear the revoke table so | |
750 | * that it can be reused by the running filesystem. | |
751 | */ | |
752 | ||
753 | void journal_clear_revoke(journal_t *journal) | |
754 | { | |
755 | int i; | |
756 | struct list_head *hash_list; | |
97f168b6 DW |
757 | struct jbd2_revoke_record_s *record; |
758 | struct jbd2_revoke_table_s *revoke; | |
efc6f628 | 759 | |
0e8a9560 | 760 | revoke = journal->j_revoke; |
efc6f628 | 761 | |
0e8a9560 TT |
762 | for (i = 0; i < revoke->hash_size; i++) { |
763 | hash_list = &revoke->hash_table[i]; | |
764 | while (!list_empty(hash_list)) { | |
97f168b6 | 765 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
0e8a9560 | 766 | list_del(&record->hash); |
97f168b6 | 767 | kmem_cache_free(jbd2_revoke_record_cache, record); |
0e8a9560 TT |
768 | } |
769 | } | |
770 | } |