]>
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 | ||
2e95a371 TT |
95 | static struct kmem_cache *jbd2_revoke_record_cache; |
96 | static struct kmem_cache *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; |
650ee9f8 | 141 | gfp_t gfp_mask = GFP_NOFS; |
8cf93332 | 142 | |
650ee9f8 TT |
143 | if (journal_oom_retry) |
144 | gfp_mask |= __GFP_NOFAIL; | |
145 | record = kmem_cache_alloc(jbd2_revoke_record_cache, gfp_mask); | |
0e8a9560 | 146 | if (!record) |
650ee9f8 | 147 | return -ENOMEM; |
0e8a9560 TT |
148 | |
149 | record->sequence = seq; | |
150 | record->blocknr = blocknr; | |
151 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
97f168b6 | 152 | spin_lock(&journal->j_revoke_lock); |
0e8a9560 | 153 | list_add(&record->hash, hash_list); |
97f168b6 | 154 | spin_unlock(&journal->j_revoke_lock); |
0e8a9560 TT |
155 | return 0; |
156 | } | |
157 | ||
158 | /* Find a revoke record in the journal's hash table. */ | |
159 | ||
97f168b6 DW |
160 | static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal, |
161 | unsigned long long blocknr) | |
0e8a9560 TT |
162 | { |
163 | struct list_head *hash_list; | |
97f168b6 | 164 | struct jbd2_revoke_record_s *record; |
efc6f628 | 165 | |
0e8a9560 TT |
166 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; |
167 | ||
97f168b6 DW |
168 | spin_lock(&journal->j_revoke_lock); |
169 | record = (struct jbd2_revoke_record_s *) hash_list->next; | |
0e8a9560 | 170 | while (&(record->hash) != hash_list) { |
97f168b6 DW |
171 | if (record->blocknr == blocknr) { |
172 | spin_unlock(&journal->j_revoke_lock); | |
0e8a9560 | 173 | return record; |
97f168b6 DW |
174 | } |
175 | record = (struct jbd2_revoke_record_s *) record->hash.next; | |
0e8a9560 | 176 | } |
97f168b6 | 177 | spin_unlock(&journal->j_revoke_lock); |
0e8a9560 TT |
178 | return NULL; |
179 | } | |
180 | ||
2faf75d2 | 181 | void jbd2_journal_destroy_revoke_record_cache(void) |
8cf93332 | 182 | { |
be259fa5 TT |
183 | kmem_cache_destroy(jbd2_revoke_record_cache); |
184 | jbd2_revoke_record_cache = NULL; | |
2faf75d2 TT |
185 | } |
186 | ||
187 | void jbd2_journal_destroy_revoke_table_cache(void) | |
188 | { | |
be259fa5 TT |
189 | kmem_cache_destroy(jbd2_revoke_table_cache); |
190 | jbd2_revoke_table_cache = NULL; | |
efc6f628 | 191 | } |
8cf93332 | 192 | |
2faf75d2 | 193 | int __init jbd2_journal_init_revoke_record_cache(void) |
8cf93332 | 194 | { |
97f168b6 | 195 | J_ASSERT(!jbd2_revoke_record_cache); |
97f168b6 DW |
196 | jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s, |
197 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY); | |
97f168b6 | 198 | |
2faf75d2 TT |
199 | if (!jbd2_revoke_record_cache) { |
200 | pr_emerg("JBD2: failed to create revoke_record cache\n"); | |
201 | return -ENOMEM; | |
202 | } | |
203 | return 0; | |
204 | } | |
205 | ||
206 | int __init jbd2_journal_init_revoke_table_cache(void) | |
207 | { | |
208 | J_ASSERT(!jbd2_revoke_table_cache); | |
97f168b6 DW |
209 | jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s, |
210 | SLAB_TEMPORARY); | |
2faf75d2 TT |
211 | if (!jbd2_revoke_table_cache) { |
212 | pr_emerg("JBD2: failed to create revoke_table cache\n"); | |
97f168b6 | 213 | return -ENOMEM; |
2faf75d2 TT |
214 | } |
215 | return 0; | |
8cf93332 | 216 | } |
0e8a9560 | 217 | |
8335d3c5 | 218 | static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size) |
0e8a9560 | 219 | { |
97f168b6 DW |
220 | int shift = 0; |
221 | int tmp = hash_size; | |
222 | struct jbd2_revoke_table_s *table; | |
efc6f628 | 223 | |
97f168b6 DW |
224 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
225 | if (!table) | |
226 | goto out; | |
0e8a9560 | 227 | |
0e8a9560 TT |
228 | while((tmp >>= 1UL) != 0UL) |
229 | shift++; | |
0e8a9560 | 230 | |
97f168b6 DW |
231 | table->hash_size = hash_size; |
232 | table->hash_shift = shift; | |
233 | table->hash_table = | |
a8833696 | 234 | kmalloc_array(hash_size, sizeof(struct list_head), GFP_KERNEL); |
97f168b6 DW |
235 | if (!table->hash_table) { |
236 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
237 | table = NULL; | |
238 | goto out; | |
0e8a9560 | 239 | } |
efc6f628 | 240 | |
0e8a9560 | 241 | for (tmp = 0; tmp < hash_size; tmp++) |
97f168b6 | 242 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
efc6f628 | 243 | |
97f168b6 DW |
244 | out: |
245 | return table; | |
0e8a9560 TT |
246 | } |
247 | ||
8335d3c5 | 248 | static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) |
0e8a9560 | 249 | { |
0e8a9560 | 250 | int i; |
97f168b6 | 251 | struct list_head *hash_list; |
efc6f628 | 252 | |
97f168b6 | 253 | for (i = 0; i < table->hash_size; i++) { |
0e8a9560 | 254 | hash_list = &table->hash_table[i]; |
97f168b6 | 255 | J_ASSERT(list_empty(hash_list)); |
0e8a9560 | 256 | } |
efc6f628 | 257 | |
0e8a9560 | 258 | kfree(table->hash_table); |
97f168b6 DW |
259 | kmem_cache_free(jbd2_revoke_table_cache, table); |
260 | } | |
261 | ||
262 | /* Initialise the revoke table for a given journal to a given size. */ | |
8335d3c5 | 263 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) |
97f168b6 DW |
264 | { |
265 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
266 | J_ASSERT(is_power_of_2(hash_size)); | |
267 | ||
8335d3c5 | 268 | journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size); |
97f168b6 DW |
269 | if (!journal->j_revoke_table[0]) |
270 | goto fail0; | |
271 | ||
8335d3c5 | 272 | journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size); |
97f168b6 DW |
273 | if (!journal->j_revoke_table[1]) |
274 | goto fail1; | |
275 | ||
276 | journal->j_revoke = journal->j_revoke_table[1]; | |
277 | ||
278 | spin_lock_init(&journal->j_revoke_lock); | |
279 | ||
280 | return 0; | |
281 | ||
282 | fail1: | |
8335d3c5 | 283 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); |
be259fa5 | 284 | journal->j_revoke_table[0] = NULL; |
97f168b6 DW |
285 | fail0: |
286 | return -ENOMEM; | |
287 | } | |
288 | ||
289 | /* Destroy a journal's revoke table. The table must already be empty! */ | |
8335d3c5 | 290 | void jbd2_journal_destroy_revoke(journal_t *journal) |
97f168b6 | 291 | { |
0e8a9560 | 292 | journal->j_revoke = NULL; |
97f168b6 | 293 | if (journal->j_revoke_table[0]) |
8335d3c5 | 294 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); |
97f168b6 | 295 | if (journal->j_revoke_table[1]) |
8335d3c5 | 296 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]); |
0e8a9560 TT |
297 | } |
298 | ||
299 | ||
300 | #ifdef __KERNEL__ | |
301 | ||
efc6f628 | 302 | /* |
8335d3c5 | 303 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
0e8a9560 TT |
304 | * prevents the block from being replayed during recovery if we take a |
305 | * crash after this current transaction commits. Any subsequent | |
306 | * metadata writes of the buffer in this transaction cancel the | |
efc6f628 | 307 | * revoke. |
0e8a9560 TT |
308 | * |
309 | * Note that this call may block --- it is up to the caller to make | |
310 | * sure that there are no further calls to journal_write_metadata | |
311 | * before the revoke is complete. In ext3, this implies calling the | |
312 | * revoke before clearing the block bitmap when we are deleting | |
efc6f628 | 313 | * metadata. |
0e8a9560 | 314 | * |
8335d3c5 | 315 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
0e8a9560 | 316 | * parameter, but does _not_ forget the buffer_head if the bh was only |
efc6f628 | 317 | * found implicitly. |
0e8a9560 | 318 | * |
8cf93332 TT |
319 | * bh_in may not be a journalled buffer - it may have come off |
320 | * the hash tables without an attached journal_head. | |
321 | * | |
8335d3c5 | 322 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
8cf93332 | 323 | * by one. |
0e8a9560 TT |
324 | */ |
325 | ||
8335d3c5 | 326 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
0e8a9560 TT |
327 | struct buffer_head *bh_in) |
328 | { | |
8cf93332 | 329 | struct buffer_head *bh = NULL; |
0e8a9560 | 330 | journal_t *journal; |
97f168b6 | 331 | struct block_device *bdev; |
0e8a9560 TT |
332 | int err; |
333 | ||
97f168b6 | 334 | might_sleep(); |
8cf93332 TT |
335 | if (bh_in) |
336 | BUFFER_TRACE(bh_in, "enter"); | |
337 | ||
0e8a9560 | 338 | journal = handle->h_transaction->t_journal; |
8335d3c5 | 339 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
1f735038 | 340 | J_ASSERT (!"Cannot set revoke feature!"); |
0e8a9560 | 341 | return -EINVAL; |
1f735038 | 342 | } |
8cf93332 | 343 | |
97f168b6 | 344 | bdev = journal->j_fs_dev; |
0e8a9560 TT |
345 | bh = bh_in; |
346 | ||
8cf93332 | 347 | if (!bh) { |
97f168b6 | 348 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); |
8cf93332 TT |
349 | if (bh) |
350 | BUFFER_TRACE(bh, "found on hash"); | |
351 | } | |
8335d3c5 | 352 | #ifdef JBD2_EXPENSIVE_CHECKING |
8cf93332 TT |
353 | else { |
354 | struct buffer_head *bh2; | |
355 | ||
356 | /* If there is a different buffer_head lying around in | |
357 | * memory anywhere... */ | |
97f168b6 | 358 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); |
8cf93332 TT |
359 | if (bh2) { |
360 | /* ... and it has RevokeValid status... */ | |
97f168b6 | 361 | if (bh2 != bh && buffer_revokevalid(bh2)) |
8cf93332 TT |
362 | /* ...then it better be revoked too, |
363 | * since it's illegal to create a revoke | |
364 | * record against a buffer_head which is | |
365 | * not marked revoked --- that would | |
366 | * risk missing a subsequent revoke | |
367 | * cancel. */ | |
97f168b6 DW |
368 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); |
369 | put_bh(bh2); | |
8cf93332 TT |
370 | } |
371 | } | |
372 | #endif | |
0e8a9560 | 373 | |
be259fa5 TT |
374 | if (WARN_ON_ONCE(handle->h_revoke_credits <= 0)) { |
375 | if (!bh_in) | |
376 | brelse(bh); | |
377 | return -EIO; | |
378 | } | |
0e8a9560 TT |
379 | /* We really ought not ever to revoke twice in a row without |
380 | first having the revoke cancelled: it's illegal to free a | |
381 | block twice without allocating it in between! */ | |
382 | if (bh) { | |
97f168b6 DW |
383 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), |
384 | "inconsistent data on disk")) { | |
385 | if (!bh_in) | |
386 | brelse(bh); | |
387 | return -EIO; | |
388 | } | |
389 | set_buffer_revoked(bh); | |
390 | set_buffer_revokevalid(bh); | |
8cf93332 | 391 | if (bh_in) { |
8335d3c5 TT |
392 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
393 | jbd2_journal_forget(handle, bh_in); | |
8cf93332 TT |
394 | } else { |
395 | BUFFER_TRACE(bh, "call brelse"); | |
396 | __brelse(bh); | |
397 | } | |
0e8a9560 | 398 | } |
be259fa5 | 399 | handle->h_revoke_credits--; |
0e8a9560 | 400 | |
97f168b6 | 401 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
8cf93332 TT |
402 | err = insert_revoke_hash(journal, blocknr, |
403 | handle->h_transaction->t_tid); | |
8cf93332 | 404 | BUFFER_TRACE(bh_in, "exit"); |
0e8a9560 TT |
405 | return err; |
406 | } | |
407 | ||
0e8a9560 TT |
408 | /* |
409 | * Cancel an outstanding revoke. For use only internally by the | |
8335d3c5 | 410 | * journaling code (called from jbd2_journal_get_write_access). |
0e8a9560 | 411 | * |
97f168b6 | 412 | * We trust buffer_revoked() on the buffer if the buffer is already |
0e8a9560 TT |
413 | * being journaled: if there is no revoke pending on the buffer, then we |
414 | * don't do anything here. | |
415 | * | |
416 | * This would break if it were possible for a buffer to be revoked and | |
417 | * discarded, and then reallocated within the same transaction. In such | |
418 | * a case we would have lost the revoked bit, but when we arrived here | |
419 | * the second time we would still have a pending revoke to cancel. So, | |
420 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
421 | * set. | |
8cf93332 | 422 | */ |
8335d3c5 | 423 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
0e8a9560 | 424 | { |
97f168b6 | 425 | struct jbd2_revoke_record_s *record; |
0e8a9560 TT |
426 | journal_t *journal = handle->h_transaction->t_journal; |
427 | int need_cancel; | |
8cf93332 TT |
428 | int did_revoke = 0; /* akpm: debug */ |
429 | struct buffer_head *bh = jh2bh(jh); | |
efc6f628 | 430 | |
71030cf8 | 431 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); |
8cf93332 | 432 | |
0e8a9560 TT |
433 | /* Is the existing Revoke bit valid? If so, we trust it, and |
434 | * only perform the full cancel if the revoke bit is set. If | |
435 | * not, we can't trust the revoke bit, and we need to do the | |
436 | * full search for a revoke record. */ | |
97f168b6 DW |
437 | if (test_set_buffer_revokevalid(bh)) { |
438 | need_cancel = test_clear_buffer_revoked(bh); | |
439 | } else { | |
0e8a9560 | 440 | need_cancel = 1; |
97f168b6 | 441 | clear_buffer_revoked(bh); |
0e8a9560 | 442 | } |
8cf93332 | 443 | |
0e8a9560 TT |
444 | if (need_cancel) { |
445 | record = find_revoke_record(journal, bh->b_blocknr); | |
446 | if (record) { | |
8cf93332 | 447 | jbd_debug(4, "cancelled existing revoke on " |
97f168b6 DW |
448 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); |
449 | spin_lock(&journal->j_revoke_lock); | |
0e8a9560 | 450 | list_del(&record->hash); |
97f168b6 DW |
451 | spin_unlock(&journal->j_revoke_lock); |
452 | kmem_cache_free(jbd2_revoke_record_cache, record); | |
8cf93332 | 453 | did_revoke = 1; |
0e8a9560 TT |
454 | } |
455 | } | |
8cf93332 | 456 | |
8335d3c5 | 457 | #ifdef JBD2_EXPENSIVE_CHECKING |
8cf93332 TT |
458 | /* There better not be one left behind by now! */ |
459 | record = find_revoke_record(journal, bh->b_blocknr); | |
460 | J_ASSERT_JH(jh, record == NULL); | |
461 | #endif | |
462 | ||
463 | /* Finally, have we just cleared revoke on an unhashed | |
464 | * buffer_head? If so, we'd better make sure we clear the | |
465 | * revoked status on any hashed alias too, otherwise the revoke | |
466 | * state machine will get very upset later on. */ | |
97f168b6 | 467 | if (need_cancel) { |
8cf93332 | 468 | struct buffer_head *bh2; |
97f168b6 | 469 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); |
8cf93332 | 470 | if (bh2) { |
97f168b6 DW |
471 | if (bh2 != bh) |
472 | clear_buffer_revoked(bh2); | |
8cf93332 TT |
473 | __brelse(bh2); |
474 | } | |
475 | } | |
8cf93332 | 476 | return did_revoke; |
0e8a9560 TT |
477 | } |
478 | ||
97f168b6 DW |
479 | /* |
480 | * journal_clear_revoked_flag clears revoked flag of buffers in | |
481 | * revoke table to reflect there is no revoked buffers in the next | |
482 | * transaction which is going to be started. | |
483 | */ | |
484 | void jbd2_clear_buffer_revoked_flags(journal_t *journal) | |
485 | { | |
486 | struct jbd2_revoke_table_s *revoke = journal->j_revoke; | |
487 | int i = 0; | |
488 | ||
489 | for (i = 0; i < revoke->hash_size; i++) { | |
490 | struct list_head *hash_list; | |
491 | struct list_head *list_entry; | |
492 | hash_list = &revoke->hash_table[i]; | |
493 | ||
494 | list_for_each(list_entry, hash_list) { | |
495 | struct jbd2_revoke_record_s *record; | |
496 | struct buffer_head *bh; | |
497 | record = (struct jbd2_revoke_record_s *)list_entry; | |
498 | bh = __find_get_block(journal->j_fs_dev, | |
499 | record->blocknr, | |
500 | journal->j_blocksize); | |
501 | if (bh) { | |
502 | clear_buffer_revoked(bh); | |
503 | __brelse(bh); | |
504 | } | |
505 | } | |
506 | } | |
507 | } | |
508 | ||
509 | /* journal_switch_revoke table select j_revoke for next transaction | |
510 | * we do not want to suspend any processing until all revokes are | |
511 | * written -bzzz | |
512 | */ | |
8335d3c5 | 513 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
97f168b6 DW |
514 | { |
515 | int i; | |
516 | ||
517 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
518 | journal->j_revoke = journal->j_revoke_table[1]; | |
519 | else | |
520 | journal->j_revoke = journal->j_revoke_table[0]; | |
521 | ||
522 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
523 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
524 | } | |
0e8a9560 TT |
525 | |
526 | /* | |
527 | * Write revoke records to the journal for all entries in the current | |
528 | * revoke hash, deleting the entries as we go. | |
0e8a9560 | 529 | */ |
be259fa5 TT |
530 | void jbd2_journal_write_revoke_records(transaction_t *transaction, |
531 | struct list_head *log_bufs) | |
0e8a9560 | 532 | { |
be259fa5 | 533 | journal_t *journal = transaction->t_journal; |
97f168b6 DW |
534 | struct buffer_head *descriptor; |
535 | struct jbd2_revoke_record_s *record; | |
536 | struct jbd2_revoke_table_s *revoke; | |
0e8a9560 | 537 | struct list_head *hash_list; |
1f735038 | 538 | int i, offset, count; |
8cf93332 | 539 | |
efc6f628 | 540 | descriptor = NULL; |
0e8a9560 | 541 | offset = 0; |
1f735038 | 542 | count = 0; |
97f168b6 DW |
543 | |
544 | /* select revoke table for committing transaction */ | |
545 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
546 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
efc6f628 | 547 | |
0e8a9560 TT |
548 | for (i = 0; i < revoke->hash_size; i++) { |
549 | hash_list = &revoke->hash_table[i]; | |
550 | ||
551 | while (!list_empty(hash_list)) { | |
97f168b6 | 552 | record = (struct jbd2_revoke_record_s *) |
0e8a9560 | 553 | hash_list->next; |
be259fa5 TT |
554 | write_one_revoke_record(transaction, log_bufs, |
555 | &descriptor, &offset, record); | |
1f735038 | 556 | count++; |
0e8a9560 | 557 | list_del(&record->hash); |
97f168b6 | 558 | kmem_cache_free(jbd2_revoke_record_cache, record); |
0e8a9560 TT |
559 | } |
560 | } | |
efc6f628 | 561 | if (descriptor) |
be259fa5 | 562 | flush_descriptor(journal, descriptor, offset); |
8cf93332 | 563 | jbd_debug(1, "Wrote %d revoke records\n", count); |
0e8a9560 TT |
564 | } |
565 | ||
efc6f628 | 566 | /* |
0e8a9560 | 567 | * Write out one revoke record. We need to create a new descriptor |
efc6f628 | 568 | * block if the old one is full or if we have not already created one. |
0e8a9560 TT |
569 | */ |
570 | ||
be259fa5 | 571 | static void write_one_revoke_record(transaction_t *transaction, |
97f168b6 DW |
572 | struct list_head *log_bufs, |
573 | struct buffer_head **descriptorp, | |
0e8a9560 | 574 | int *offsetp, |
be259fa5 | 575 | struct jbd2_revoke_record_s *record) |
0e8a9560 | 576 | { |
be259fa5 | 577 | journal_t *journal = transaction->t_journal; |
97f168b6 DW |
578 | int csum_size = 0; |
579 | struct buffer_head *descriptor; | |
04c66cb2 | 580 | int sz, offset; |
8cf93332 | 581 | |
0e8a9560 TT |
582 | /* If we are already aborting, this all becomes a noop. We |
583 | still need to go round the loop in | |
8335d3c5 | 584 | jbd2_journal_write_revoke_records in order to free all of the |
0e8a9560 | 585 | revoke records: only the IO to the journal is omitted. */ |
8cf93332 | 586 | if (is_journal_aborted(journal)) |
0e8a9560 TT |
587 | return; |
588 | ||
589 | descriptor = *descriptorp; | |
590 | offset = *offsetp; | |
8cf93332 | 591 | |
97f168b6 | 592 | /* Do we need to leave space at the end for a checksum? */ |
8335d3c5 | 593 | if (jbd2_journal_has_csum_v2or3(journal)) |
1942e33c | 594 | csum_size = sizeof(struct jbd2_journal_block_tail); |
97f168b6 | 595 | |
8335d3c5 | 596 | if (jbd2_has_feature_64bit(journal)) |
04c66cb2 DW |
597 | sz = 8; |
598 | else | |
599 | sz = 4; | |
600 | ||
0e8a9560 TT |
601 | /* Make sure we have a descriptor with space left for the record */ |
602 | if (descriptor) { | |
04c66cb2 | 603 | if (offset + sz > journal->j_blocksize - csum_size) { |
be259fa5 | 604 | flush_descriptor(journal, descriptor, offset); |
0e8a9560 TT |
605 | descriptor = NULL; |
606 | } | |
607 | } | |
efc6f628 | 608 | |
0e8a9560 | 609 | if (!descriptor) { |
be259fa5 TT |
610 | descriptor = jbd2_journal_get_descriptor_buffer(transaction, |
611 | JBD2_REVOKE_BLOCK); | |
8cf93332 TT |
612 | if (!descriptor) |
613 | return; | |
0e8a9560 TT |
614 | |
615 | /* Record it so that we can wait for IO completion later */ | |
97f168b6 DW |
616 | BUFFER_TRACE(descriptor, "file in log_bufs"); |
617 | jbd2_file_log_bh(log_bufs, descriptor); | |
8cf93332 | 618 | |
8335d3c5 | 619 | offset = sizeof(jbd2_journal_revoke_header_t); |
0e8a9560 TT |
620 | *descriptorp = descriptor; |
621 | } | |
efc6f628 | 622 | |
8335d3c5 | 623 | if (jbd2_has_feature_64bit(journal)) |
04c66cb2 DW |
624 | * ((__be64 *)(&descriptor->b_data[offset])) = |
625 | cpu_to_be64(record->blocknr); | |
626 | else | |
627 | * ((__be32 *)(&descriptor->b_data[offset])) = | |
628 | cpu_to_be32(record->blocknr); | |
629 | offset += sz; | |
97f168b6 | 630 | |
0e8a9560 TT |
631 | *offsetp = offset; |
632 | } | |
633 | ||
efc6f628 | 634 | /* |
0e8a9560 TT |
635 | * Flush a revoke descriptor out to the journal. If we are aborting, |
636 | * this is a noop; otherwise we are generating a buffer which needs to | |
637 | * be waited for during commit, so it has to go onto the appropriate | |
638 | * journal buffer list. | |
639 | */ | |
640 | ||
efc6f628 | 641 | static void flush_descriptor(journal_t *journal, |
97f168b6 | 642 | struct buffer_head *descriptor, |
be259fa5 | 643 | int offset) |
0e8a9560 | 644 | { |
8335d3c5 | 645 | jbd2_journal_revoke_header_t *header; |
8cf93332 | 646 | |
be259fa5 | 647 | if (is_journal_aborted(journal)) |
0e8a9560 | 648 | return; |
efc6f628 | 649 | |
8335d3c5 | 650 | header = (jbd2_journal_revoke_header_t *)descriptor->b_data; |
46e1286a | 651 | header->r_count = cpu_to_be32(offset); |
1942e33c | 652 | jbd2_descriptor_block_csum_set(journal, descriptor); |
0e8a9560 | 653 | |
97f168b6 DW |
654 | set_buffer_jwrite(descriptor); |
655 | BUFFER_TRACE(descriptor, "write"); | |
656 | set_buffer_dirty(descriptor); | |
be259fa5 | 657 | write_dirty_buffer(descriptor, REQ_SYNC); |
97f168b6 | 658 | } |
0e8a9560 TT |
659 | #endif |
660 | ||
efc6f628 | 661 | /* |
0e8a9560 TT |
662 | * Revoke support for recovery. |
663 | * | |
664 | * Recovery needs to be able to: | |
665 | * | |
666 | * record all revoke records, including the tid of the latest instance | |
667 | * of each revoke in the journal | |
668 | * | |
669 | * check whether a given block in a given transaction should be replayed | |
670 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
671 | * transaction) | |
efc6f628 | 672 | * |
0e8a9560 TT |
673 | * empty the revoke table after recovery. |
674 | */ | |
675 | ||
676 | /* | |
677 | * First, setting revoke records. We create a new revoke record for | |
678 | * every block ever revoked in the log as we scan it for recovery, and | |
679 | * we update the existing records if we find multiple revokes for a | |
efc6f628 | 680 | * single block. |
0e8a9560 TT |
681 | */ |
682 | ||
8335d3c5 | 683 | int jbd2_journal_set_revoke(journal_t *journal, |
97f168b6 | 684 | unsigned long long blocknr, |
0e8a9560 TT |
685 | tid_t sequence) |
686 | { | |
97f168b6 | 687 | struct jbd2_revoke_record_s *record; |
efc6f628 | 688 | |
0e8a9560 TT |
689 | record = find_revoke_record(journal, blocknr); |
690 | if (record) { | |
97f168b6 | 691 | /* If we have multiple occurrences, only record the |
0e8a9560 | 692 | * latest sequence number in the hashed record */ |
725c474f | 693 | if (tid_gt(sequence, record->sequence)) |
0e8a9560 TT |
694 | record->sequence = sequence; |
695 | return 0; | |
efc6f628 | 696 | } |
0e8a9560 TT |
697 | return insert_revoke_hash(journal, blocknr, sequence); |
698 | } | |
699 | ||
efc6f628 | 700 | /* |
0e8a9560 TT |
701 | * Test revoke records. For a given block referenced in the log, has |
702 | * that block been revoked? A revoke record with a given transaction | |
703 | * sequence number revokes all blocks in that transaction and earlier | |
704 | * ones, but later transactions still need replayed. | |
705 | */ | |
706 | ||
8335d3c5 | 707 | int jbd2_journal_test_revoke(journal_t *journal, |
97f168b6 | 708 | unsigned long long blocknr, |
0e8a9560 TT |
709 | tid_t sequence) |
710 | { | |
97f168b6 | 711 | struct jbd2_revoke_record_s *record; |
efc6f628 | 712 | |
0e8a9560 TT |
713 | record = find_revoke_record(journal, blocknr); |
714 | if (!record) | |
715 | return 0; | |
725c474f | 716 | if (tid_gt(sequence, record->sequence)) |
0e8a9560 TT |
717 | return 0; |
718 | return 1; | |
719 | } | |
720 | ||
721 | /* | |
722 | * Finally, once recovery is over, we need to clear the revoke table so | |
723 | * that it can be reused by the running filesystem. | |
724 | */ | |
725 | ||
8335d3c5 | 726 | void jbd2_journal_clear_revoke(journal_t *journal) |
0e8a9560 TT |
727 | { |
728 | int i; | |
729 | struct list_head *hash_list; | |
97f168b6 DW |
730 | struct jbd2_revoke_record_s *record; |
731 | struct jbd2_revoke_table_s *revoke; | |
efc6f628 | 732 | |
0e8a9560 | 733 | revoke = journal->j_revoke; |
efc6f628 | 734 | |
0e8a9560 TT |
735 | for (i = 0; i < revoke->hash_size; i++) { |
736 | hash_list = &revoke->hash_table[i]; | |
737 | while (!list_empty(hash_list)) { | |
97f168b6 | 738 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
0e8a9560 | 739 | list_del(&record->hash); |
97f168b6 | 740 | kmem_cache_free(jbd2_revoke_record_cache, record); |
0e8a9560 TT |
741 | } |
742 | } | |
743 | } |