]>
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 | ||
2faf75d2 | 186 | void jbd2_journal_destroy_revoke_record_cache(void) |
8cf93332 | 187 | { |
be259fa5 TT |
188 | kmem_cache_destroy(jbd2_revoke_record_cache); |
189 | jbd2_revoke_record_cache = NULL; | |
2faf75d2 TT |
190 | } |
191 | ||
192 | void jbd2_journal_destroy_revoke_table_cache(void) | |
193 | { | |
be259fa5 TT |
194 | kmem_cache_destroy(jbd2_revoke_table_cache); |
195 | jbd2_revoke_table_cache = NULL; | |
efc6f628 | 196 | } |
8cf93332 | 197 | |
2faf75d2 | 198 | int __init jbd2_journal_init_revoke_record_cache(void) |
8cf93332 | 199 | { |
97f168b6 | 200 | J_ASSERT(!jbd2_revoke_record_cache); |
97f168b6 DW |
201 | jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s, |
202 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY); | |
97f168b6 | 203 | |
2faf75d2 TT |
204 | if (!jbd2_revoke_record_cache) { |
205 | pr_emerg("JBD2: failed to create revoke_record cache\n"); | |
206 | return -ENOMEM; | |
207 | } | |
208 | return 0; | |
209 | } | |
210 | ||
211 | int __init jbd2_journal_init_revoke_table_cache(void) | |
212 | { | |
213 | J_ASSERT(!jbd2_revoke_table_cache); | |
97f168b6 DW |
214 | jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s, |
215 | SLAB_TEMPORARY); | |
2faf75d2 TT |
216 | if (!jbd2_revoke_table_cache) { |
217 | pr_emerg("JBD2: failed to create revoke_table cache\n"); | |
97f168b6 | 218 | return -ENOMEM; |
2faf75d2 TT |
219 | } |
220 | return 0; | |
8cf93332 | 221 | } |
0e8a9560 | 222 | |
8335d3c5 | 223 | static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size) |
0e8a9560 | 224 | { |
97f168b6 DW |
225 | int shift = 0; |
226 | int tmp = hash_size; | |
227 | struct jbd2_revoke_table_s *table; | |
efc6f628 | 228 | |
97f168b6 DW |
229 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
230 | if (!table) | |
231 | goto out; | |
0e8a9560 | 232 | |
0e8a9560 TT |
233 | while((tmp >>= 1UL) != 0UL) |
234 | shift++; | |
0e8a9560 | 235 | |
97f168b6 DW |
236 | table->hash_size = hash_size; |
237 | table->hash_shift = shift; | |
238 | table->hash_table = | |
a8833696 | 239 | kmalloc_array(hash_size, sizeof(struct list_head), GFP_KERNEL); |
97f168b6 DW |
240 | if (!table->hash_table) { |
241 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
242 | table = NULL; | |
243 | goto out; | |
0e8a9560 | 244 | } |
efc6f628 | 245 | |
0e8a9560 | 246 | for (tmp = 0; tmp < hash_size; tmp++) |
97f168b6 | 247 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
efc6f628 | 248 | |
97f168b6 DW |
249 | out: |
250 | return table; | |
0e8a9560 TT |
251 | } |
252 | ||
8335d3c5 | 253 | static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) |
0e8a9560 | 254 | { |
0e8a9560 | 255 | int i; |
97f168b6 | 256 | struct list_head *hash_list; |
efc6f628 | 257 | |
97f168b6 | 258 | for (i = 0; i < table->hash_size; i++) { |
0e8a9560 | 259 | hash_list = &table->hash_table[i]; |
97f168b6 | 260 | J_ASSERT(list_empty(hash_list)); |
0e8a9560 | 261 | } |
efc6f628 | 262 | |
0e8a9560 | 263 | kfree(table->hash_table); |
97f168b6 DW |
264 | kmem_cache_free(jbd2_revoke_table_cache, table); |
265 | } | |
266 | ||
267 | /* Initialise the revoke table for a given journal to a given size. */ | |
8335d3c5 | 268 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) |
97f168b6 DW |
269 | { |
270 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
271 | J_ASSERT(is_power_of_2(hash_size)); | |
272 | ||
8335d3c5 | 273 | journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size); |
97f168b6 DW |
274 | if (!journal->j_revoke_table[0]) |
275 | goto fail0; | |
276 | ||
8335d3c5 | 277 | journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size); |
97f168b6 DW |
278 | if (!journal->j_revoke_table[1]) |
279 | goto fail1; | |
280 | ||
281 | journal->j_revoke = journal->j_revoke_table[1]; | |
282 | ||
283 | spin_lock_init(&journal->j_revoke_lock); | |
284 | ||
285 | return 0; | |
286 | ||
287 | fail1: | |
8335d3c5 | 288 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); |
be259fa5 | 289 | journal->j_revoke_table[0] = NULL; |
97f168b6 DW |
290 | fail0: |
291 | return -ENOMEM; | |
292 | } | |
293 | ||
294 | /* Destroy a journal's revoke table. The table must already be empty! */ | |
8335d3c5 | 295 | void jbd2_journal_destroy_revoke(journal_t *journal) |
97f168b6 | 296 | { |
0e8a9560 | 297 | journal->j_revoke = NULL; |
97f168b6 | 298 | if (journal->j_revoke_table[0]) |
8335d3c5 | 299 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); |
97f168b6 | 300 | if (journal->j_revoke_table[1]) |
8335d3c5 | 301 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]); |
0e8a9560 TT |
302 | } |
303 | ||
304 | ||
305 | #ifdef __KERNEL__ | |
306 | ||
efc6f628 | 307 | /* |
8335d3c5 | 308 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
0e8a9560 TT |
309 | * prevents the block from being replayed during recovery if we take a |
310 | * crash after this current transaction commits. Any subsequent | |
311 | * metadata writes of the buffer in this transaction cancel the | |
efc6f628 | 312 | * revoke. |
0e8a9560 TT |
313 | * |
314 | * Note that this call may block --- it is up to the caller to make | |
315 | * sure that there are no further calls to journal_write_metadata | |
316 | * before the revoke is complete. In ext3, this implies calling the | |
317 | * revoke before clearing the block bitmap when we are deleting | |
efc6f628 | 318 | * metadata. |
0e8a9560 | 319 | * |
8335d3c5 | 320 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
0e8a9560 | 321 | * parameter, but does _not_ forget the buffer_head if the bh was only |
efc6f628 | 322 | * found implicitly. |
0e8a9560 | 323 | * |
8cf93332 TT |
324 | * bh_in may not be a journalled buffer - it may have come off |
325 | * the hash tables without an attached journal_head. | |
326 | * | |
8335d3c5 | 327 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
8cf93332 | 328 | * by one. |
0e8a9560 TT |
329 | */ |
330 | ||
8335d3c5 | 331 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
0e8a9560 TT |
332 | struct buffer_head *bh_in) |
333 | { | |
8cf93332 | 334 | struct buffer_head *bh = NULL; |
0e8a9560 | 335 | journal_t *journal; |
97f168b6 | 336 | struct block_device *bdev; |
0e8a9560 TT |
337 | int err; |
338 | ||
97f168b6 | 339 | might_sleep(); |
8cf93332 TT |
340 | if (bh_in) |
341 | BUFFER_TRACE(bh_in, "enter"); | |
342 | ||
0e8a9560 | 343 | journal = handle->h_transaction->t_journal; |
8335d3c5 | 344 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
1f735038 | 345 | J_ASSERT (!"Cannot set revoke feature!"); |
0e8a9560 | 346 | return -EINVAL; |
1f735038 | 347 | } |
8cf93332 | 348 | |
97f168b6 | 349 | bdev = journal->j_fs_dev; |
0e8a9560 TT |
350 | bh = bh_in; |
351 | ||
8cf93332 | 352 | if (!bh) { |
97f168b6 | 353 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); |
8cf93332 TT |
354 | if (bh) |
355 | BUFFER_TRACE(bh, "found on hash"); | |
356 | } | |
8335d3c5 | 357 | #ifdef JBD2_EXPENSIVE_CHECKING |
8cf93332 TT |
358 | else { |
359 | struct buffer_head *bh2; | |
360 | ||
361 | /* If there is a different buffer_head lying around in | |
362 | * memory anywhere... */ | |
97f168b6 | 363 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); |
8cf93332 TT |
364 | if (bh2) { |
365 | /* ... and it has RevokeValid status... */ | |
97f168b6 | 366 | if (bh2 != bh && buffer_revokevalid(bh2)) |
8cf93332 TT |
367 | /* ...then it better be revoked too, |
368 | * since it's illegal to create a revoke | |
369 | * record against a buffer_head which is | |
370 | * not marked revoked --- that would | |
371 | * risk missing a subsequent revoke | |
372 | * cancel. */ | |
97f168b6 DW |
373 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); |
374 | put_bh(bh2); | |
8cf93332 TT |
375 | } |
376 | } | |
377 | #endif | |
0e8a9560 | 378 | |
be259fa5 TT |
379 | if (WARN_ON_ONCE(handle->h_revoke_credits <= 0)) { |
380 | if (!bh_in) | |
381 | brelse(bh); | |
382 | return -EIO; | |
383 | } | |
0e8a9560 TT |
384 | /* We really ought not ever to revoke twice in a row without |
385 | first having the revoke cancelled: it's illegal to free a | |
386 | block twice without allocating it in between! */ | |
387 | if (bh) { | |
97f168b6 DW |
388 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), |
389 | "inconsistent data on disk")) { | |
390 | if (!bh_in) | |
391 | brelse(bh); | |
392 | return -EIO; | |
393 | } | |
394 | set_buffer_revoked(bh); | |
395 | set_buffer_revokevalid(bh); | |
8cf93332 | 396 | if (bh_in) { |
8335d3c5 TT |
397 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
398 | jbd2_journal_forget(handle, bh_in); | |
8cf93332 TT |
399 | } else { |
400 | BUFFER_TRACE(bh, "call brelse"); | |
401 | __brelse(bh); | |
402 | } | |
0e8a9560 | 403 | } |
be259fa5 | 404 | handle->h_revoke_credits--; |
0e8a9560 | 405 | |
97f168b6 | 406 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
8cf93332 TT |
407 | err = insert_revoke_hash(journal, blocknr, |
408 | handle->h_transaction->t_tid); | |
8cf93332 | 409 | BUFFER_TRACE(bh_in, "exit"); |
0e8a9560 TT |
410 | return err; |
411 | } | |
412 | ||
0e8a9560 TT |
413 | /* |
414 | * Cancel an outstanding revoke. For use only internally by the | |
8335d3c5 | 415 | * journaling code (called from jbd2_journal_get_write_access). |
0e8a9560 | 416 | * |
97f168b6 | 417 | * We trust buffer_revoked() on the buffer if the buffer is already |
0e8a9560 TT |
418 | * being journaled: if there is no revoke pending on the buffer, then we |
419 | * don't do anything here. | |
420 | * | |
421 | * This would break if it were possible for a buffer to be revoked and | |
422 | * discarded, and then reallocated within the same transaction. In such | |
423 | * a case we would have lost the revoked bit, but when we arrived here | |
424 | * the second time we would still have a pending revoke to cancel. So, | |
425 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
426 | * set. | |
8cf93332 | 427 | */ |
8335d3c5 | 428 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
0e8a9560 | 429 | { |
97f168b6 | 430 | struct jbd2_revoke_record_s *record; |
0e8a9560 TT |
431 | journal_t *journal = handle->h_transaction->t_journal; |
432 | int need_cancel; | |
8cf93332 TT |
433 | int did_revoke = 0; /* akpm: debug */ |
434 | struct buffer_head *bh = jh2bh(jh); | |
efc6f628 | 435 | |
71030cf8 | 436 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); |
8cf93332 | 437 | |
0e8a9560 TT |
438 | /* Is the existing Revoke bit valid? If so, we trust it, and |
439 | * only perform the full cancel if the revoke bit is set. If | |
440 | * not, we can't trust the revoke bit, and we need to do the | |
441 | * full search for a revoke record. */ | |
97f168b6 DW |
442 | if (test_set_buffer_revokevalid(bh)) { |
443 | need_cancel = test_clear_buffer_revoked(bh); | |
444 | } else { | |
0e8a9560 | 445 | need_cancel = 1; |
97f168b6 | 446 | clear_buffer_revoked(bh); |
0e8a9560 | 447 | } |
8cf93332 | 448 | |
0e8a9560 TT |
449 | if (need_cancel) { |
450 | record = find_revoke_record(journal, bh->b_blocknr); | |
451 | if (record) { | |
8cf93332 | 452 | jbd_debug(4, "cancelled existing revoke on " |
97f168b6 DW |
453 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); |
454 | spin_lock(&journal->j_revoke_lock); | |
0e8a9560 | 455 | list_del(&record->hash); |
97f168b6 DW |
456 | spin_unlock(&journal->j_revoke_lock); |
457 | kmem_cache_free(jbd2_revoke_record_cache, record); | |
8cf93332 | 458 | did_revoke = 1; |
0e8a9560 TT |
459 | } |
460 | } | |
8cf93332 | 461 | |
8335d3c5 | 462 | #ifdef JBD2_EXPENSIVE_CHECKING |
8cf93332 TT |
463 | /* There better not be one left behind by now! */ |
464 | record = find_revoke_record(journal, bh->b_blocknr); | |
465 | J_ASSERT_JH(jh, record == NULL); | |
466 | #endif | |
467 | ||
468 | /* Finally, have we just cleared revoke on an unhashed | |
469 | * buffer_head? If so, we'd better make sure we clear the | |
470 | * revoked status on any hashed alias too, otherwise the revoke | |
471 | * state machine will get very upset later on. */ | |
97f168b6 | 472 | if (need_cancel) { |
8cf93332 | 473 | struct buffer_head *bh2; |
97f168b6 | 474 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); |
8cf93332 | 475 | if (bh2) { |
97f168b6 DW |
476 | if (bh2 != bh) |
477 | clear_buffer_revoked(bh2); | |
8cf93332 TT |
478 | __brelse(bh2); |
479 | } | |
480 | } | |
8cf93332 | 481 | return did_revoke; |
0e8a9560 TT |
482 | } |
483 | ||
97f168b6 DW |
484 | /* |
485 | * journal_clear_revoked_flag clears revoked flag of buffers in | |
486 | * revoke table to reflect there is no revoked buffers in the next | |
487 | * transaction which is going to be started. | |
488 | */ | |
489 | void jbd2_clear_buffer_revoked_flags(journal_t *journal) | |
490 | { | |
491 | struct jbd2_revoke_table_s *revoke = journal->j_revoke; | |
492 | int i = 0; | |
493 | ||
494 | for (i = 0; i < revoke->hash_size; i++) { | |
495 | struct list_head *hash_list; | |
496 | struct list_head *list_entry; | |
497 | hash_list = &revoke->hash_table[i]; | |
498 | ||
499 | list_for_each(list_entry, hash_list) { | |
500 | struct jbd2_revoke_record_s *record; | |
501 | struct buffer_head *bh; | |
502 | record = (struct jbd2_revoke_record_s *)list_entry; | |
503 | bh = __find_get_block(journal->j_fs_dev, | |
504 | record->blocknr, | |
505 | journal->j_blocksize); | |
506 | if (bh) { | |
507 | clear_buffer_revoked(bh); | |
508 | __brelse(bh); | |
509 | } | |
510 | } | |
511 | } | |
512 | } | |
513 | ||
514 | /* journal_switch_revoke table select j_revoke for next transaction | |
515 | * we do not want to suspend any processing until all revokes are | |
516 | * written -bzzz | |
517 | */ | |
8335d3c5 | 518 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
97f168b6 DW |
519 | { |
520 | int i; | |
521 | ||
522 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
523 | journal->j_revoke = journal->j_revoke_table[1]; | |
524 | else | |
525 | journal->j_revoke = journal->j_revoke_table[0]; | |
526 | ||
527 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
528 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
529 | } | |
0e8a9560 TT |
530 | |
531 | /* | |
532 | * Write revoke records to the journal for all entries in the current | |
533 | * revoke hash, deleting the entries as we go. | |
0e8a9560 | 534 | */ |
be259fa5 TT |
535 | void jbd2_journal_write_revoke_records(transaction_t *transaction, |
536 | struct list_head *log_bufs) | |
0e8a9560 | 537 | { |
be259fa5 | 538 | journal_t *journal = transaction->t_journal; |
97f168b6 DW |
539 | struct buffer_head *descriptor; |
540 | struct jbd2_revoke_record_s *record; | |
541 | struct jbd2_revoke_table_s *revoke; | |
0e8a9560 | 542 | struct list_head *hash_list; |
1f735038 | 543 | int i, offset, count; |
8cf93332 | 544 | |
efc6f628 | 545 | descriptor = NULL; |
0e8a9560 | 546 | offset = 0; |
1f735038 | 547 | count = 0; |
97f168b6 DW |
548 | |
549 | /* select revoke table for committing transaction */ | |
550 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
551 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
efc6f628 | 552 | |
0e8a9560 TT |
553 | for (i = 0; i < revoke->hash_size; i++) { |
554 | hash_list = &revoke->hash_table[i]; | |
555 | ||
556 | while (!list_empty(hash_list)) { | |
97f168b6 | 557 | record = (struct jbd2_revoke_record_s *) |
0e8a9560 | 558 | hash_list->next; |
be259fa5 TT |
559 | write_one_revoke_record(transaction, log_bufs, |
560 | &descriptor, &offset, record); | |
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) |
be259fa5 | 567 | flush_descriptor(journal, descriptor, offset); |
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 | ||
be259fa5 | 576 | static void write_one_revoke_record(transaction_t *transaction, |
97f168b6 DW |
577 | struct list_head *log_bufs, |
578 | struct buffer_head **descriptorp, | |
0e8a9560 | 579 | int *offsetp, |
be259fa5 | 580 | struct jbd2_revoke_record_s *record) |
0e8a9560 | 581 | { |
be259fa5 | 582 | journal_t *journal = transaction->t_journal; |
97f168b6 DW |
583 | int csum_size = 0; |
584 | struct buffer_head *descriptor; | |
04c66cb2 | 585 | int sz, offset; |
8cf93332 | 586 | |
0e8a9560 TT |
587 | /* If we are already aborting, this all becomes a noop. We |
588 | still need to go round the loop in | |
8335d3c5 | 589 | jbd2_journal_write_revoke_records in order to free all of the |
0e8a9560 | 590 | revoke records: only the IO to the journal is omitted. */ |
8cf93332 | 591 | if (is_journal_aborted(journal)) |
0e8a9560 TT |
592 | return; |
593 | ||
594 | descriptor = *descriptorp; | |
595 | offset = *offsetp; | |
8cf93332 | 596 | |
97f168b6 | 597 | /* Do we need to leave space at the end for a checksum? */ |
8335d3c5 | 598 | if (jbd2_journal_has_csum_v2or3(journal)) |
1942e33c | 599 | csum_size = sizeof(struct jbd2_journal_block_tail); |
97f168b6 | 600 | |
8335d3c5 | 601 | if (jbd2_has_feature_64bit(journal)) |
04c66cb2 DW |
602 | sz = 8; |
603 | else | |
604 | sz = 4; | |
605 | ||
0e8a9560 TT |
606 | /* Make sure we have a descriptor with space left for the record */ |
607 | if (descriptor) { | |
04c66cb2 | 608 | if (offset + sz > journal->j_blocksize - csum_size) { |
be259fa5 | 609 | flush_descriptor(journal, descriptor, offset); |
0e8a9560 TT |
610 | descriptor = NULL; |
611 | } | |
612 | } | |
efc6f628 | 613 | |
0e8a9560 | 614 | if (!descriptor) { |
be259fa5 TT |
615 | descriptor = jbd2_journal_get_descriptor_buffer(transaction, |
616 | JBD2_REVOKE_BLOCK); | |
8cf93332 TT |
617 | if (!descriptor) |
618 | return; | |
0e8a9560 TT |
619 | |
620 | /* Record it so that we can wait for IO completion later */ | |
97f168b6 DW |
621 | BUFFER_TRACE(descriptor, "file in log_bufs"); |
622 | jbd2_file_log_bh(log_bufs, descriptor); | |
8cf93332 | 623 | |
8335d3c5 | 624 | offset = sizeof(jbd2_journal_revoke_header_t); |
0e8a9560 TT |
625 | *descriptorp = descriptor; |
626 | } | |
efc6f628 | 627 | |
8335d3c5 | 628 | if (jbd2_has_feature_64bit(journal)) |
04c66cb2 DW |
629 | * ((__be64 *)(&descriptor->b_data[offset])) = |
630 | cpu_to_be64(record->blocknr); | |
631 | else | |
632 | * ((__be32 *)(&descriptor->b_data[offset])) = | |
633 | cpu_to_be32(record->blocknr); | |
634 | offset += sz; | |
97f168b6 | 635 | |
0e8a9560 TT |
636 | *offsetp = offset; |
637 | } | |
638 | ||
efc6f628 | 639 | /* |
0e8a9560 TT |
640 | * Flush a revoke descriptor out to the journal. If we are aborting, |
641 | * this is a noop; otherwise we are generating a buffer which needs to | |
642 | * be waited for during commit, so it has to go onto the appropriate | |
643 | * journal buffer list. | |
644 | */ | |
645 | ||
efc6f628 | 646 | static void flush_descriptor(journal_t *journal, |
97f168b6 | 647 | struct buffer_head *descriptor, |
be259fa5 | 648 | int offset) |
0e8a9560 | 649 | { |
8335d3c5 | 650 | jbd2_journal_revoke_header_t *header; |
8cf93332 | 651 | |
be259fa5 | 652 | if (is_journal_aborted(journal)) |
0e8a9560 | 653 | return; |
efc6f628 | 654 | |
8335d3c5 | 655 | header = (jbd2_journal_revoke_header_t *)descriptor->b_data; |
97f168b6 | 656 | header->r_count = ext2fs_cpu_to_be32(offset); |
1942e33c | 657 | jbd2_descriptor_block_csum_set(journal, descriptor); |
0e8a9560 | 658 | |
97f168b6 DW |
659 | set_buffer_jwrite(descriptor); |
660 | BUFFER_TRACE(descriptor, "write"); | |
661 | set_buffer_dirty(descriptor); | |
be259fa5 | 662 | write_dirty_buffer(descriptor, REQ_SYNC); |
97f168b6 | 663 | } |
0e8a9560 TT |
664 | #endif |
665 | ||
efc6f628 | 666 | /* |
0e8a9560 TT |
667 | * Revoke support for recovery. |
668 | * | |
669 | * Recovery needs to be able to: | |
670 | * | |
671 | * record all revoke records, including the tid of the latest instance | |
672 | * of each revoke in the journal | |
673 | * | |
674 | * check whether a given block in a given transaction should be replayed | |
675 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
676 | * transaction) | |
efc6f628 | 677 | * |
0e8a9560 TT |
678 | * empty the revoke table after recovery. |
679 | */ | |
680 | ||
681 | /* | |
682 | * First, setting revoke records. We create a new revoke record for | |
683 | * every block ever revoked in the log as we scan it for recovery, and | |
684 | * we update the existing records if we find multiple revokes for a | |
efc6f628 | 685 | * single block. |
0e8a9560 TT |
686 | */ |
687 | ||
8335d3c5 | 688 | int jbd2_journal_set_revoke(journal_t *journal, |
97f168b6 | 689 | unsigned long long blocknr, |
0e8a9560 TT |
690 | tid_t sequence) |
691 | { | |
97f168b6 | 692 | struct jbd2_revoke_record_s *record; |
efc6f628 | 693 | |
0e8a9560 TT |
694 | record = find_revoke_record(journal, blocknr); |
695 | if (record) { | |
97f168b6 | 696 | /* If we have multiple occurrences, only record the |
0e8a9560 | 697 | * latest sequence number in the hashed record */ |
725c474f | 698 | if (tid_gt(sequence, record->sequence)) |
0e8a9560 TT |
699 | record->sequence = sequence; |
700 | return 0; | |
efc6f628 | 701 | } |
0e8a9560 TT |
702 | return insert_revoke_hash(journal, blocknr, sequence); |
703 | } | |
704 | ||
efc6f628 | 705 | /* |
0e8a9560 TT |
706 | * Test revoke records. For a given block referenced in the log, has |
707 | * that block been revoked? A revoke record with a given transaction | |
708 | * sequence number revokes all blocks in that transaction and earlier | |
709 | * ones, but later transactions still need replayed. | |
710 | */ | |
711 | ||
8335d3c5 | 712 | int jbd2_journal_test_revoke(journal_t *journal, |
97f168b6 | 713 | unsigned long long blocknr, |
0e8a9560 TT |
714 | tid_t sequence) |
715 | { | |
97f168b6 | 716 | struct jbd2_revoke_record_s *record; |
efc6f628 | 717 | |
0e8a9560 TT |
718 | record = find_revoke_record(journal, blocknr); |
719 | if (!record) | |
720 | return 0; | |
725c474f | 721 | if (tid_gt(sequence, record->sequence)) |
0e8a9560 TT |
722 | return 0; |
723 | return 1; | |
724 | } | |
725 | ||
726 | /* | |
727 | * Finally, once recovery is over, we need to clear the revoke table so | |
728 | * that it can be reused by the running filesystem. | |
729 | */ | |
730 | ||
8335d3c5 | 731 | void jbd2_journal_clear_revoke(journal_t *journal) |
0e8a9560 TT |
732 | { |
733 | int i; | |
734 | struct list_head *hash_list; | |
97f168b6 DW |
735 | struct jbd2_revoke_record_s *record; |
736 | struct jbd2_revoke_table_s *revoke; | |
efc6f628 | 737 | |
0e8a9560 | 738 | revoke = journal->j_revoke; |
efc6f628 | 739 | |
0e8a9560 TT |
740 | for (i = 0; i < revoke->hash_size; i++) { |
741 | hash_list = &revoke->hash_table[i]; | |
742 | while (!list_empty(hash_list)) { | |
97f168b6 | 743 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
0e8a9560 | 744 | list_del(&record->hash); |
97f168b6 | 745 | kmem_cache_free(jbd2_revoke_record_cache, record); |
0e8a9560 TT |
746 | } |
747 | } | |
748 | } |