]>
Commit | Line | Data |
---|---|---|
71030cf8 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
0e8a9560 | 2 | /* |
97f168b6 | 3 | * linux/fs/jbd2/revoke.c |
efc6f628 | 4 | * |
0e8a9560 TT |
5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 2000 |
6 | * | |
7 | * Copyright 2000 Red Hat corp --- All Rights Reserved | |
8 | * | |
0e8a9560 TT |
9 | * Journal revoke routines for the generic filesystem journaling code; |
10 | * part of the ext2fs journaling system. | |
11 | * | |
12 | * Revoke is the mechanism used to prevent old log records for deleted | |
13 | * metadata from being replayed on top of newer data using the same | |
14 | * blocks. The revoke mechanism is used in two separate places: | |
efc6f628 | 15 | * |
0e8a9560 TT |
16 | * + Commit: during commit we write the entire list of the current |
17 | * transaction's revoked blocks to the journal | |
efc6f628 | 18 | * |
0e8a9560 TT |
19 | * + Recovery: during recovery we record the transaction ID of all |
20 | * revoked blocks. If there are multiple revoke records in the log | |
21 | * for a single block, only the last one counts, and if there is a log | |
22 | * entry for a block beyond the last revoke, then that log entry still | |
23 | * gets replayed. | |
24 | * | |
25 | * We can get interactions between revokes and new log data within a | |
26 | * single transaction: | |
27 | * | |
28 | * Block is revoked and then journaled: | |
efc6f628 | 29 | * The desired end result is the journaling of the new block, so we |
0e8a9560 TT |
30 | * cancel the revoke before the transaction commits. |
31 | * | |
32 | * Block is journaled and then revoked: | |
725c474f TT |
33 | * The revoke must take precedence over the write of the block, so we |
34 | * need either to cancel the journal entry or to write the revoke | |
0e8a9560 | 35 | * later in the log than the log block. In this case, we choose the |
725c474f TT |
36 | * latter: journaling a block cancels any revoke record for that block |
37 | * in the current transaction, so any revoke for that block in the | |
38 | * transaction must have happened after the block was journaled and so | |
39 | * the revoke must take precedence. | |
0e8a9560 | 40 | * |
efc6f628 | 41 | * Block is revoked and then written as data: |
0e8a9560 TT |
42 | * The data write is allowed to succeed, but the revoke is _not_ |
43 | * cancelled. We still need to prevent old log records from | |
44 | * overwriting the new data. We don't even need to clear the revoke | |
45 | * bit here. | |
46 | * | |
97f168b6 DW |
47 | * We cache revoke status of a buffer in the current transaction in b_states |
48 | * bits. As the name says, revokevalid flag indicates that the cached revoke | |
49 | * status of a buffer is valid and we can rely on the cached status. | |
50 | * | |
0e8a9560 TT |
51 | * Revoke information on buffers is a tri-state value: |
52 | * | |
53 | * RevokeValid clear: no cached revoke status, need to look it up | |
8cf93332 | 54 | * RevokeValid set, Revoked clear: |
0e8a9560 TT |
55 | * buffer has not been revoked, and cancel_revoke |
56 | * need do nothing. | |
8cf93332 | 57 | * RevokeValid set, Revoked set: |
efc6f628 | 58 | * buffer has been revoked. |
97f168b6 DW |
59 | * |
60 | * Locking rules: | |
61 | * We keep two hash tables of revoke records. One hashtable belongs to the | |
62 | * running transaction (is pointed to by journal->j_revoke), the other one | |
63 | * belongs to the committing transaction. Accesses to the second hash table | |
64 | * happen only from the kjournald and no other thread touches this table. Also | |
65 | * journal_switch_revoke_table() which switches which hashtable belongs to the | |
66 | * running and which to the committing transaction is called only from | |
67 | * kjournald. Therefore we need no locks when accessing the hashtable belonging | |
68 | * to the committing transaction. | |
69 | * | |
70 | * All users operating on the hash table belonging to the running transaction | |
71 | * have a handle to the transaction. Therefore they are safe from kjournald | |
72 | * switching hash tables under them. For operations on the lists of entries in | |
73 | * the hash table j_revoke_lock is used. | |
74 | * | |
75 | * Finally, also replay code uses the hash tables but at this moment no one else | |
76 | * can touch them (filesystem isn't mounted yet) and hence no locking is | |
77 | * needed. | |
0e8a9560 TT |
78 | */ |
79 | ||
80 | #ifndef __KERNEL__ | |
81 | #include "jfs_user.h" | |
82 | #else | |
97f168b6 | 83 | #include <linux/time.h> |
0e8a9560 | 84 | #include <linux/fs.h> |
97f168b6 | 85 | #include <linux/jbd2.h> |
0e8a9560 TT |
86 | #include <linux/errno.h> |
87 | #include <linux/slab.h> | |
0e8a9560 | 88 | #include <linux/list.h> |
8cf93332 | 89 | #include <linux/init.h> |
97f168b6 DW |
90 | #include <linux/bio.h> |
91 | #include <linux/log2.h> | |
71030cf8 | 92 | #include <linux/hash.h> |
0e8a9560 TT |
93 | #endif |
94 | ||
97f168b6 DW |
95 | static lkmem_cache_t *jbd2_revoke_record_cache; |
96 | static lkmem_cache_t *jbd2_revoke_table_cache; | |
0e8a9560 TT |
97 | |
98 | /* Each revoke record represents one single revoked block. During | |
99 | journal replay, this involves recording the transaction ID of the | |
100 | last transaction to revoke this block. */ | |
101 | ||
97f168b6 | 102 | struct jbd2_revoke_record_s |
0e8a9560 TT |
103 | { |
104 | struct list_head hash; | |
105 | tid_t sequence; /* Used for recovery only */ | |
97f168b6 | 106 | unsigned long long blocknr; |
0e8a9560 TT |
107 | }; |
108 | ||
109 | ||
110 | /* The revoke table is just a simple hash table of revoke records. */ | |
97f168b6 | 111 | struct jbd2_revoke_table_s |
0e8a9560 TT |
112 | { |
113 | /* It is conceivable that we might want a larger hash table | |
114 | * for recovery. Must be a power of two. */ | |
efc6f628 TT |
115 | int hash_size; |
116 | int hash_shift; | |
0e8a9560 TT |
117 | struct list_head *hash_table; |
118 | }; | |
119 | ||
120 | ||
121 | #ifdef __KERNEL__ | |
be259fa5 | 122 | static void write_one_revoke_record(transaction_t *, |
97f168b6 DW |
123 | struct list_head *, |
124 | struct buffer_head **, int *, | |
be259fa5 TT |
125 | struct jbd2_revoke_record_s *); |
126 | static void flush_descriptor(journal_t *, struct buffer_head *, int); | |
0e8a9560 TT |
127 | #endif |
128 | ||
129 | /* Utility functions to maintain the revoke table */ | |
130 | ||
97f168b6 | 131 | static inline int hash(journal_t *journal, unsigned long long block) |
0e8a9560 | 132 | { |
71030cf8 | 133 | return hash_64(block, journal->j_revoke->hash_shift); |
0e8a9560 TT |
134 | } |
135 | ||
97f168b6 | 136 | static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr, |
3e699064 | 137 | tid_t seq) |
0e8a9560 TT |
138 | { |
139 | struct list_head *hash_list; | |
97f168b6 | 140 | struct jbd2_revoke_record_s *record; |
8cf93332 TT |
141 | |
142 | repeat: | |
97f168b6 | 143 | record = kmem_cache_alloc(jbd2_revoke_record_cache, GFP_NOFS); |
0e8a9560 | 144 | if (!record) |
8cf93332 | 145 | goto oom; |
0e8a9560 TT |
146 | |
147 | record->sequence = seq; | |
148 | record->blocknr = blocknr; | |
149 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
97f168b6 | 150 | spin_lock(&journal->j_revoke_lock); |
0e8a9560 | 151 | list_add(&record->hash, hash_list); |
97f168b6 | 152 | spin_unlock(&journal->j_revoke_lock); |
0e8a9560 | 153 | return 0; |
8cf93332 TT |
154 | |
155 | oom: | |
8cf93332 TT |
156 | if (!journal_oom_retry) |
157 | return -ENOMEM; | |
97f168b6 DW |
158 | jbd_debug(1, "ENOMEM in %s, retrying\n", __func__); |
159 | yield(); | |
8cf93332 | 160 | goto repeat; |
0e8a9560 TT |
161 | } |
162 | ||
163 | /* Find a revoke record in the journal's hash table. */ | |
164 | ||
97f168b6 DW |
165 | static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal, |
166 | unsigned long long blocknr) | |
0e8a9560 TT |
167 | { |
168 | struct list_head *hash_list; | |
97f168b6 | 169 | struct jbd2_revoke_record_s *record; |
efc6f628 | 170 | |
0e8a9560 TT |
171 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; |
172 | ||
97f168b6 DW |
173 | spin_lock(&journal->j_revoke_lock); |
174 | record = (struct jbd2_revoke_record_s *) hash_list->next; | |
0e8a9560 | 175 | while (&(record->hash) != hash_list) { |
97f168b6 DW |
176 | if (record->blocknr == blocknr) { |
177 | spin_unlock(&journal->j_revoke_lock); | |
0e8a9560 | 178 | return record; |
97f168b6 DW |
179 | } |
180 | record = (struct jbd2_revoke_record_s *) record->hash.next; | |
0e8a9560 | 181 | } |
97f168b6 | 182 | spin_unlock(&journal->j_revoke_lock); |
0e8a9560 TT |
183 | return NULL; |
184 | } | |
185 | ||
8335d3c5 | 186 | void jbd2_journal_destroy_revoke_caches(void) |
8cf93332 | 187 | { |
be259fa5 TT |
188 | kmem_cache_destroy(jbd2_revoke_record_cache); |
189 | jbd2_revoke_record_cache = NULL; | |
190 | kmem_cache_destroy(jbd2_revoke_table_cache); | |
191 | jbd2_revoke_table_cache = NULL; | |
efc6f628 | 192 | } |
8cf93332 | 193 | |
8335d3c5 | 194 | int __init jbd2_journal_init_revoke_caches(void) |
8cf93332 | 195 | { |
97f168b6 DW |
196 | J_ASSERT(!jbd2_revoke_record_cache); |
197 | J_ASSERT(!jbd2_revoke_table_cache); | |
198 | ||
199 | jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s, | |
200 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY); | |
201 | if (!jbd2_revoke_record_cache) | |
202 | goto record_cache_failure; | |
203 | ||
204 | jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s, | |
205 | SLAB_TEMPORARY); | |
206 | if (!jbd2_revoke_table_cache) | |
207 | goto table_cache_failure; | |
208 | return 0; | |
209 | table_cache_failure: | |
8335d3c5 | 210 | jbd2_journal_destroy_revoke_caches(); |
97f168b6 DW |
211 | record_cache_failure: |
212 | return -ENOMEM; | |
8cf93332 | 213 | } |
0e8a9560 | 214 | |
8335d3c5 | 215 | static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size) |
0e8a9560 | 216 | { |
97f168b6 DW |
217 | int shift = 0; |
218 | int tmp = hash_size; | |
219 | struct jbd2_revoke_table_s *table; | |
efc6f628 | 220 | |
97f168b6 DW |
221 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
222 | if (!table) | |
223 | goto out; | |
0e8a9560 | 224 | |
0e8a9560 TT |
225 | while((tmp >>= 1UL) != 0UL) |
226 | shift++; | |
0e8a9560 | 227 | |
97f168b6 DW |
228 | table->hash_size = hash_size; |
229 | table->hash_shift = shift; | |
230 | table->hash_table = | |
0e8a9560 | 231 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); |
97f168b6 DW |
232 | if (!table->hash_table) { |
233 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
234 | table = NULL; | |
235 | goto out; | |
0e8a9560 | 236 | } |
efc6f628 | 237 | |
0e8a9560 | 238 | for (tmp = 0; tmp < hash_size; tmp++) |
97f168b6 | 239 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
efc6f628 | 240 | |
97f168b6 DW |
241 | out: |
242 | return table; | |
0e8a9560 TT |
243 | } |
244 | ||
8335d3c5 | 245 | static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) |
0e8a9560 | 246 | { |
0e8a9560 | 247 | int i; |
97f168b6 | 248 | struct list_head *hash_list; |
efc6f628 | 249 | |
97f168b6 | 250 | for (i = 0; i < table->hash_size; i++) { |
0e8a9560 | 251 | hash_list = &table->hash_table[i]; |
97f168b6 | 252 | J_ASSERT(list_empty(hash_list)); |
0e8a9560 | 253 | } |
efc6f628 | 254 | |
0e8a9560 | 255 | kfree(table->hash_table); |
97f168b6 DW |
256 | kmem_cache_free(jbd2_revoke_table_cache, table); |
257 | } | |
258 | ||
259 | /* Initialise the revoke table for a given journal to a given size. */ | |
8335d3c5 | 260 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) |
97f168b6 DW |
261 | { |
262 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
263 | J_ASSERT(is_power_of_2(hash_size)); | |
264 | ||
8335d3c5 | 265 | journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size); |
97f168b6 DW |
266 | if (!journal->j_revoke_table[0]) |
267 | goto fail0; | |
268 | ||
8335d3c5 | 269 | journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size); |
97f168b6 DW |
270 | if (!journal->j_revoke_table[1]) |
271 | goto fail1; | |
272 | ||
273 | journal->j_revoke = journal->j_revoke_table[1]; | |
274 | ||
275 | spin_lock_init(&journal->j_revoke_lock); | |
276 | ||
277 | return 0; | |
278 | ||
279 | fail1: | |
8335d3c5 | 280 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); |
be259fa5 | 281 | journal->j_revoke_table[0] = NULL; |
97f168b6 DW |
282 | fail0: |
283 | return -ENOMEM; | |
284 | } | |
285 | ||
286 | /* Destroy a journal's revoke table. The table must already be empty! */ | |
8335d3c5 | 287 | void jbd2_journal_destroy_revoke(journal_t *journal) |
97f168b6 | 288 | { |
0e8a9560 | 289 | journal->j_revoke = NULL; |
97f168b6 | 290 | if (journal->j_revoke_table[0]) |
8335d3c5 | 291 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); |
97f168b6 | 292 | if (journal->j_revoke_table[1]) |
8335d3c5 | 293 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]); |
0e8a9560 TT |
294 | } |
295 | ||
296 | ||
297 | #ifdef __KERNEL__ | |
298 | ||
efc6f628 | 299 | /* |
8335d3c5 | 300 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
0e8a9560 TT |
301 | * prevents the block from being replayed during recovery if we take a |
302 | * crash after this current transaction commits. Any subsequent | |
303 | * metadata writes of the buffer in this transaction cancel the | |
efc6f628 | 304 | * revoke. |
0e8a9560 TT |
305 | * |
306 | * Note that this call may block --- it is up to the caller to make | |
307 | * sure that there are no further calls to journal_write_metadata | |
308 | * before the revoke is complete. In ext3, this implies calling the | |
309 | * revoke before clearing the block bitmap when we are deleting | |
efc6f628 | 310 | * metadata. |
0e8a9560 | 311 | * |
8335d3c5 | 312 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
0e8a9560 | 313 | * parameter, but does _not_ forget the buffer_head if the bh was only |
efc6f628 | 314 | * found implicitly. |
0e8a9560 | 315 | * |
8cf93332 TT |
316 | * bh_in may not be a journalled buffer - it may have come off |
317 | * the hash tables without an attached journal_head. | |
318 | * | |
8335d3c5 | 319 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
8cf93332 | 320 | * by one. |
0e8a9560 TT |
321 | */ |
322 | ||
8335d3c5 | 323 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
0e8a9560 TT |
324 | struct buffer_head *bh_in) |
325 | { | |
8cf93332 | 326 | struct buffer_head *bh = NULL; |
0e8a9560 | 327 | journal_t *journal; |
97f168b6 | 328 | struct block_device *bdev; |
0e8a9560 TT |
329 | int err; |
330 | ||
97f168b6 | 331 | might_sleep(); |
8cf93332 TT |
332 | if (bh_in) |
333 | BUFFER_TRACE(bh_in, "enter"); | |
334 | ||
0e8a9560 | 335 | journal = handle->h_transaction->t_journal; |
8335d3c5 | 336 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
1f735038 | 337 | J_ASSERT (!"Cannot set revoke feature!"); |
0e8a9560 | 338 | return -EINVAL; |
1f735038 | 339 | } |
8cf93332 | 340 | |
97f168b6 | 341 | bdev = journal->j_fs_dev; |
0e8a9560 TT |
342 | bh = bh_in; |
343 | ||
8cf93332 | 344 | if (!bh) { |
97f168b6 | 345 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); |
8cf93332 TT |
346 | if (bh) |
347 | BUFFER_TRACE(bh, "found on hash"); | |
348 | } | |
8335d3c5 | 349 | #ifdef JBD2_EXPENSIVE_CHECKING |
8cf93332 TT |
350 | else { |
351 | struct buffer_head *bh2; | |
352 | ||
353 | /* If there is a different buffer_head lying around in | |
354 | * memory anywhere... */ | |
97f168b6 | 355 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); |
8cf93332 TT |
356 | if (bh2) { |
357 | /* ... and it has RevokeValid status... */ | |
97f168b6 | 358 | if (bh2 != bh && buffer_revokevalid(bh2)) |
8cf93332 TT |
359 | /* ...then it better be revoked too, |
360 | * since it's illegal to create a revoke | |
361 | * record against a buffer_head which is | |
362 | * not marked revoked --- that would | |
363 | * risk missing a subsequent revoke | |
364 | * cancel. */ | |
97f168b6 DW |
365 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); |
366 | put_bh(bh2); | |
8cf93332 TT |
367 | } |
368 | } | |
369 | #endif | |
0e8a9560 | 370 | |
be259fa5 TT |
371 | if (WARN_ON_ONCE(handle->h_revoke_credits <= 0)) { |
372 | if (!bh_in) | |
373 | brelse(bh); | |
374 | return -EIO; | |
375 | } | |
0e8a9560 TT |
376 | /* We really ought not ever to revoke twice in a row without |
377 | first having the revoke cancelled: it's illegal to free a | |
378 | block twice without allocating it in between! */ | |
379 | if (bh) { | |
97f168b6 DW |
380 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), |
381 | "inconsistent data on disk")) { | |
382 | if (!bh_in) | |
383 | brelse(bh); | |
384 | return -EIO; | |
385 | } | |
386 | set_buffer_revoked(bh); | |
387 | set_buffer_revokevalid(bh); | |
8cf93332 | 388 | if (bh_in) { |
8335d3c5 TT |
389 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
390 | jbd2_journal_forget(handle, bh_in); | |
8cf93332 TT |
391 | } else { |
392 | BUFFER_TRACE(bh, "call brelse"); | |
393 | __brelse(bh); | |
394 | } | |
0e8a9560 | 395 | } |
be259fa5 | 396 | handle->h_revoke_credits--; |
0e8a9560 | 397 | |
97f168b6 | 398 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
8cf93332 TT |
399 | err = insert_revoke_hash(journal, blocknr, |
400 | handle->h_transaction->t_tid); | |
8cf93332 | 401 | BUFFER_TRACE(bh_in, "exit"); |
0e8a9560 TT |
402 | return err; |
403 | } | |
404 | ||
0e8a9560 TT |
405 | /* |
406 | * Cancel an outstanding revoke. For use only internally by the | |
8335d3c5 | 407 | * journaling code (called from jbd2_journal_get_write_access). |
0e8a9560 | 408 | * |
97f168b6 | 409 | * We trust buffer_revoked() on the buffer if the buffer is already |
0e8a9560 TT |
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. | |
8cf93332 | 419 | */ |
8335d3c5 | 420 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
0e8a9560 | 421 | { |
97f168b6 | 422 | struct jbd2_revoke_record_s *record; |
0e8a9560 TT |
423 | journal_t *journal = handle->h_transaction->t_journal; |
424 | int need_cancel; | |
8cf93332 TT |
425 | int did_revoke = 0; /* akpm: debug */ |
426 | struct buffer_head *bh = jh2bh(jh); | |
efc6f628 | 427 | |
71030cf8 | 428 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); |
8cf93332 | 429 | |
0e8a9560 TT |
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. */ | |
97f168b6 DW |
434 | if (test_set_buffer_revokevalid(bh)) { |
435 | need_cancel = test_clear_buffer_revoked(bh); | |
436 | } else { | |
0e8a9560 | 437 | need_cancel = 1; |
97f168b6 | 438 | clear_buffer_revoked(bh); |
0e8a9560 | 439 | } |
8cf93332 | 440 | |
0e8a9560 TT |
441 | if (need_cancel) { |
442 | record = find_revoke_record(journal, bh->b_blocknr); | |
443 | if (record) { | |
8cf93332 | 444 | jbd_debug(4, "cancelled existing revoke on " |
97f168b6 DW |
445 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); |
446 | spin_lock(&journal->j_revoke_lock); | |
0e8a9560 | 447 | list_del(&record->hash); |
97f168b6 DW |
448 | spin_unlock(&journal->j_revoke_lock); |
449 | kmem_cache_free(jbd2_revoke_record_cache, record); | |
8cf93332 | 450 | did_revoke = 1; |
0e8a9560 TT |
451 | } |
452 | } | |
8cf93332 | 453 | |
8335d3c5 | 454 | #ifdef JBD2_EXPENSIVE_CHECKING |
8cf93332 TT |
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. */ | |
97f168b6 | 464 | if (need_cancel) { |
8cf93332 | 465 | struct buffer_head *bh2; |
97f168b6 | 466 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); |
8cf93332 | 467 | if (bh2) { |
97f168b6 DW |
468 | if (bh2 != bh) |
469 | clear_buffer_revoked(bh2); | |
8cf93332 TT |
470 | __brelse(bh2); |
471 | } | |
472 | } | |
8cf93332 | 473 | return did_revoke; |
0e8a9560 TT |
474 | } |
475 | ||
97f168b6 DW |
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 | ||
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 | */ | |
8335d3c5 | 510 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
97f168b6 DW |
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 | } | |
0e8a9560 TT |
522 | |
523 | /* | |
524 | * Write revoke records to the journal for all entries in the current | |
525 | * revoke hash, deleting the entries as we go. | |
0e8a9560 | 526 | */ |
be259fa5 TT |
527 | void jbd2_journal_write_revoke_records(transaction_t *transaction, |
528 | struct list_head *log_bufs) | |
0e8a9560 | 529 | { |
be259fa5 | 530 | journal_t *journal = transaction->t_journal; |
97f168b6 DW |
531 | struct buffer_head *descriptor; |
532 | struct jbd2_revoke_record_s *record; | |
533 | struct jbd2_revoke_table_s *revoke; | |
0e8a9560 | 534 | struct list_head *hash_list; |
1f735038 | 535 | int i, offset, count; |
8cf93332 | 536 | |
efc6f628 | 537 | descriptor = NULL; |
0e8a9560 | 538 | offset = 0; |
1f735038 | 539 | count = 0; |
97f168b6 DW |
540 | |
541 | /* select revoke table for committing transaction */ | |
542 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
543 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
efc6f628 | 544 | |
0e8a9560 TT |
545 | for (i = 0; i < revoke->hash_size; i++) { |
546 | hash_list = &revoke->hash_table[i]; | |
547 | ||
548 | while (!list_empty(hash_list)) { | |
97f168b6 | 549 | record = (struct jbd2_revoke_record_s *) |
0e8a9560 | 550 | hash_list->next; |
be259fa5 TT |
551 | write_one_revoke_record(transaction, log_bufs, |
552 | &descriptor, &offset, record); | |
1f735038 | 553 | count++; |
0e8a9560 | 554 | list_del(&record->hash); |
97f168b6 | 555 | kmem_cache_free(jbd2_revoke_record_cache, record); |
0e8a9560 TT |
556 | } |
557 | } | |
efc6f628 | 558 | if (descriptor) |
be259fa5 | 559 | flush_descriptor(journal, descriptor, offset); |
8cf93332 | 560 | jbd_debug(1, "Wrote %d revoke records\n", count); |
0e8a9560 TT |
561 | } |
562 | ||
efc6f628 | 563 | /* |
0e8a9560 | 564 | * Write out one revoke record. We need to create a new descriptor |
efc6f628 | 565 | * block if the old one is full or if we have not already created one. |
0e8a9560 TT |
566 | */ |
567 | ||
be259fa5 | 568 | static void write_one_revoke_record(transaction_t *transaction, |
97f168b6 DW |
569 | struct list_head *log_bufs, |
570 | struct buffer_head **descriptorp, | |
0e8a9560 | 571 | int *offsetp, |
be259fa5 | 572 | struct jbd2_revoke_record_s *record) |
0e8a9560 | 573 | { |
be259fa5 | 574 | journal_t *journal = transaction->t_journal; |
97f168b6 DW |
575 | int csum_size = 0; |
576 | struct buffer_head *descriptor; | |
04c66cb2 | 577 | int sz, offset; |
8cf93332 | 578 | |
0e8a9560 TT |
579 | /* If we are already aborting, this all becomes a noop. We |
580 | still need to go round the loop in | |
8335d3c5 | 581 | jbd2_journal_write_revoke_records in order to free all of the |
0e8a9560 | 582 | revoke records: only the IO to the journal is omitted. */ |
8cf93332 | 583 | if (is_journal_aborted(journal)) |
0e8a9560 TT |
584 | return; |
585 | ||
586 | descriptor = *descriptorp; | |
587 | offset = *offsetp; | |
8cf93332 | 588 | |
97f168b6 | 589 | /* Do we need to leave space at the end for a checksum? */ |
8335d3c5 | 590 | if (jbd2_journal_has_csum_v2or3(journal)) |
1942e33c | 591 | csum_size = sizeof(struct jbd2_journal_block_tail); |
97f168b6 | 592 | |
8335d3c5 | 593 | if (jbd2_has_feature_64bit(journal)) |
04c66cb2 DW |
594 | sz = 8; |
595 | else | |
596 | sz = 4; | |
597 | ||
0e8a9560 TT |
598 | /* Make sure we have a descriptor with space left for the record */ |
599 | if (descriptor) { | |
04c66cb2 | 600 | if (offset + sz > journal->j_blocksize - csum_size) { |
be259fa5 | 601 | flush_descriptor(journal, descriptor, offset); |
0e8a9560 TT |
602 | descriptor = NULL; |
603 | } | |
604 | } | |
efc6f628 | 605 | |
0e8a9560 | 606 | if (!descriptor) { |
be259fa5 TT |
607 | descriptor = jbd2_journal_get_descriptor_buffer(transaction, |
608 | JBD2_REVOKE_BLOCK); | |
8cf93332 TT |
609 | if (!descriptor) |
610 | return; | |
0e8a9560 TT |
611 | |
612 | /* Record it so that we can wait for IO completion later */ | |
97f168b6 DW |
613 | BUFFER_TRACE(descriptor, "file in log_bufs"); |
614 | jbd2_file_log_bh(log_bufs, descriptor); | |
8cf93332 | 615 | |
8335d3c5 | 616 | offset = sizeof(jbd2_journal_revoke_header_t); |
0e8a9560 TT |
617 | *descriptorp = descriptor; |
618 | } | |
efc6f628 | 619 | |
8335d3c5 | 620 | if (jbd2_has_feature_64bit(journal)) |
04c66cb2 DW |
621 | * ((__be64 *)(&descriptor->b_data[offset])) = |
622 | cpu_to_be64(record->blocknr); | |
623 | else | |
624 | * ((__be32 *)(&descriptor->b_data[offset])) = | |
625 | cpu_to_be32(record->blocknr); | |
626 | offset += sz; | |
97f168b6 | 627 | |
0e8a9560 TT |
628 | *offsetp = offset; |
629 | } | |
630 | ||
efc6f628 | 631 | /* |
0e8a9560 TT |
632 | * Flush a revoke descriptor out to the journal. If we are aborting, |
633 | * this is a noop; otherwise we are generating a buffer which needs to | |
634 | * be waited for during commit, so it has to go onto the appropriate | |
635 | * journal buffer list. | |
636 | */ | |
637 | ||
efc6f628 | 638 | static void flush_descriptor(journal_t *journal, |
97f168b6 | 639 | struct buffer_head *descriptor, |
be259fa5 | 640 | int offset) |
0e8a9560 | 641 | { |
8335d3c5 | 642 | jbd2_journal_revoke_header_t *header; |
8cf93332 | 643 | |
be259fa5 | 644 | if (is_journal_aborted(journal)) |
0e8a9560 | 645 | return; |
efc6f628 | 646 | |
8335d3c5 | 647 | header = (jbd2_journal_revoke_header_t *)descriptor->b_data; |
97f168b6 | 648 | header->r_count = ext2fs_cpu_to_be32(offset); |
1942e33c | 649 | jbd2_descriptor_block_csum_set(journal, descriptor); |
0e8a9560 | 650 | |
97f168b6 DW |
651 | set_buffer_jwrite(descriptor); |
652 | BUFFER_TRACE(descriptor, "write"); | |
653 | set_buffer_dirty(descriptor); | |
be259fa5 | 654 | write_dirty_buffer(descriptor, REQ_SYNC); |
97f168b6 | 655 | } |
0e8a9560 TT |
656 | #endif |
657 | ||
efc6f628 | 658 | /* |
0e8a9560 TT |
659 | * Revoke support for recovery. |
660 | * | |
661 | * Recovery needs to be able to: | |
662 | * | |
663 | * record all revoke records, including the tid of the latest instance | |
664 | * of each revoke in the journal | |
665 | * | |
666 | * check whether a given block in a given transaction should be replayed | |
667 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
668 | * transaction) | |
efc6f628 | 669 | * |
0e8a9560 TT |
670 | * empty the revoke table after recovery. |
671 | */ | |
672 | ||
673 | /* | |
674 | * First, setting revoke records. We create a new revoke record for | |
675 | * every block ever revoked in the log as we scan it for recovery, and | |
676 | * we update the existing records if we find multiple revokes for a | |
efc6f628 | 677 | * single block. |
0e8a9560 TT |
678 | */ |
679 | ||
8335d3c5 | 680 | int jbd2_journal_set_revoke(journal_t *journal, |
97f168b6 | 681 | unsigned long long blocknr, |
0e8a9560 TT |
682 | tid_t sequence) |
683 | { | |
97f168b6 | 684 | struct jbd2_revoke_record_s *record; |
efc6f628 | 685 | |
0e8a9560 TT |
686 | record = find_revoke_record(journal, blocknr); |
687 | if (record) { | |
97f168b6 | 688 | /* If we have multiple occurrences, only record the |
0e8a9560 | 689 | * latest sequence number in the hashed record */ |
725c474f | 690 | if (tid_gt(sequence, record->sequence)) |
0e8a9560 TT |
691 | record->sequence = sequence; |
692 | return 0; | |
efc6f628 | 693 | } |
0e8a9560 TT |
694 | return insert_revoke_hash(journal, blocknr, sequence); |
695 | } | |
696 | ||
efc6f628 | 697 | /* |
0e8a9560 TT |
698 | * Test revoke records. For a given block referenced in the log, has |
699 | * that block been revoked? A revoke record with a given transaction | |
700 | * sequence number revokes all blocks in that transaction and earlier | |
701 | * ones, but later transactions still need replayed. | |
702 | */ | |
703 | ||
8335d3c5 | 704 | int jbd2_journal_test_revoke(journal_t *journal, |
97f168b6 | 705 | unsigned long long blocknr, |
0e8a9560 TT |
706 | tid_t sequence) |
707 | { | |
97f168b6 | 708 | struct jbd2_revoke_record_s *record; |
efc6f628 | 709 | |
0e8a9560 TT |
710 | record = find_revoke_record(journal, blocknr); |
711 | if (!record) | |
712 | return 0; | |
725c474f | 713 | if (tid_gt(sequence, record->sequence)) |
0e8a9560 TT |
714 | return 0; |
715 | return 1; | |
716 | } | |
717 | ||
718 | /* | |
719 | * Finally, once recovery is over, we need to clear the revoke table so | |
720 | * that it can be reused by the running filesystem. | |
721 | */ | |
722 | ||
8335d3c5 | 723 | void jbd2_journal_clear_revoke(journal_t *journal) |
0e8a9560 TT |
724 | { |
725 | int i; | |
726 | struct list_head *hash_list; | |
97f168b6 DW |
727 | struct jbd2_revoke_record_s *record; |
728 | struct jbd2_revoke_table_s *revoke; | |
efc6f628 | 729 | |
0e8a9560 | 730 | revoke = journal->j_revoke; |
efc6f628 | 731 | |
0e8a9560 TT |
732 | for (i = 0; i < revoke->hash_size; i++) { |
733 | hash_list = &revoke->hash_table[i]; | |
734 | while (!list_empty(hash_list)) { | |
97f168b6 | 735 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
0e8a9560 | 736 | list_del(&record->hash); |
97f168b6 | 737 | kmem_cache_free(jbd2_revoke_record_cache, record); |
0e8a9560 TT |
738 | } |
739 | } | |
740 | } |