blob: 3310a1d7ace9d79cfc135c24703b64142c55be1b [file] [log] [blame]
Dave Kleikamp470decc2006-10-11 01:20:57 -07001/*
2 * linux/fs/revoke.c
3 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 2000
5 *
6 * Copyright 2000 Red Hat corp --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Journal revoke routines for the generic filesystem journaling code;
13 * part of the ext2fs journaling system.
14 *
15 * Revoke is the mechanism used to prevent old log records for deleted
16 * metadata from being replayed on top of newer data using the same
17 * blocks. The revoke mechanism is used in two separate places:
18 *
19 * + Commit: during commit we write the entire list of the current
20 * transaction's revoked blocks to the journal
21 *
22 * + Recovery: during recovery we record the transaction ID of all
23 * revoked blocks. If there are multiple revoke records in the log
24 * for a single block, only the last one counts, and if there is a log
25 * entry for a block beyond the last revoke, then that log entry still
26 * gets replayed.
27 *
28 * We can get interactions between revokes and new log data within a
29 * single transaction:
30 *
31 * Block is revoked and then journaled:
32 * The desired end result is the journaling of the new block, so we
33 * cancel the revoke before the transaction commits.
34 *
35 * Block is journaled and then revoked:
36 * The revoke must take precedence over the write of the block, so we
37 * need either to cancel the journal entry or to write the revoke
38 * later in the log than the log block. In this case, we choose the
39 * latter: journaling a block cancels any revoke record for that block
40 * in the current transaction, so any revoke for that block in the
41 * transaction must have happened after the block was journaled and so
42 * the revoke must take precedence.
43 *
44 * Block is revoked and then written as data:
45 * The data write is allowed to succeed, but the revoke is _not_
46 * cancelled. We still need to prevent old log records from
47 * overwriting the new data. We don't even need to clear the revoke
48 * bit here.
49 *
50 * Revoke information on buffers is a tri-state value:
51 *
52 * RevokeValid clear: no cached revoke status, need to look it up
53 * RevokeValid set, Revoked clear:
54 * buffer has not been revoked, and cancel_revoke
55 * need do nothing.
56 * RevokeValid set, Revoked set:
57 * buffer has been revoked.
58 */
59
60#ifndef __KERNEL__
61#include "jfs_user.h"
62#else
63#include <linux/time.h>
64#include <linux/fs.h>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070065#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070066#include <linux/errno.h>
67#include <linux/slab.h>
68#include <linux/list.h>
69#include <linux/smp_lock.h>
70#include <linux/init.h>
71#endif
72
Mingming Caof7f4bcc2006-10-11 01:20:59 -070073static kmem_cache_t *jbd2_revoke_record_cache;
74static kmem_cache_t *jbd2_revoke_table_cache;
Dave Kleikamp470decc2006-10-11 01:20:57 -070075
76/* Each revoke record represents one single revoked block. During
77 journal replay, this involves recording the transaction ID of the
78 last transaction to revoke this block. */
79
Mingming Caof7f4bcc2006-10-11 01:20:59 -070080struct jbd2_revoke_record_s
Dave Kleikamp470decc2006-10-11 01:20:57 -070081{
82 struct list_head hash;
83 tid_t sequence; /* Used for recovery only */
Mingming Cao29971762006-10-11 01:21:09 -070084 sector_t blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -070085};
86
87
88/* The revoke table is just a simple hash table of revoke records. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -070089struct jbd2_revoke_table_s
Dave Kleikamp470decc2006-10-11 01:20:57 -070090{
91 /* It is conceivable that we might want a larger hash table
92 * for recovery. Must be a power of two. */
93 int hash_size;
94 int hash_shift;
95 struct list_head *hash_table;
96};
97
98
99#ifdef __KERNEL__
100static void write_one_revoke_record(journal_t *, transaction_t *,
101 struct journal_head **, int *,
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700102 struct jbd2_revoke_record_s *);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700103static void flush_descriptor(journal_t *, struct journal_head *, int);
104#endif
105
106/* Utility functions to maintain the revoke table */
107
108/* Borrowed from buffer.c: this is a tried and tested block hash function */
Mingming Cao29971762006-10-11 01:21:09 -0700109static inline int hash(journal_t *journal, sector_t block)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700110{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700111 struct jbd2_revoke_table_s *table = journal->j_revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700112 int hash_shift = table->hash_shift;
Mingming Cao29971762006-10-11 01:21:09 -0700113 int hash = (int)block ^ (int)((block >> 31) >> 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700114
Mingming Cao29971762006-10-11 01:21:09 -0700115 return ((hash << (hash_shift - 6)) ^
116 (hash >> 13) ^
117 (hash << (hash_shift - 12))) & (table->hash_size - 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700118}
119
Mingming Cao29971762006-10-11 01:21:09 -0700120static int insert_revoke_hash(journal_t *journal, sector_t blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700121 tid_t seq)
122{
123 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700124 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700125
126repeat:
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700127 record = kmem_cache_alloc(jbd2_revoke_record_cache, GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700128 if (!record)
129 goto oom;
130
131 record->sequence = seq;
132 record->blocknr = blocknr;
133 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
134 spin_lock(&journal->j_revoke_lock);
135 list_add(&record->hash, hash_list);
136 spin_unlock(&journal->j_revoke_lock);
137 return 0;
138
139oom:
140 if (!journal_oom_retry)
141 return -ENOMEM;
142 jbd_debug(1, "ENOMEM in %s, retrying\n", __FUNCTION__);
143 yield();
144 goto repeat;
145}
146
147/* Find a revoke record in the journal's hash table. */
148
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700149static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal,
Mingming Cao29971762006-10-11 01:21:09 -0700150 sector_t blocknr)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700151{
152 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700153 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700154
155 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
156
157 spin_lock(&journal->j_revoke_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700158 record = (struct jbd2_revoke_record_s *) hash_list->next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700159 while (&(record->hash) != hash_list) {
160 if (record->blocknr == blocknr) {
161 spin_unlock(&journal->j_revoke_lock);
162 return record;
163 }
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700164 record = (struct jbd2_revoke_record_s *) record->hash.next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700165 }
166 spin_unlock(&journal->j_revoke_lock);
167 return NULL;
168}
169
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700170int __init jbd2_journal_init_revoke_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700171{
Johann Lombardia920e942006-10-11 01:21:00 -0700172 jbd2_revoke_record_cache = kmem_cache_create("jbd2_revoke_record",
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700173 sizeof(struct jbd2_revoke_record_s),
Dave Kleikamp470decc2006-10-11 01:20:57 -0700174 0, SLAB_HWCACHE_ALIGN, NULL, NULL);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700175 if (jbd2_revoke_record_cache == 0)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700176 return -ENOMEM;
177
Johann Lombardia920e942006-10-11 01:21:00 -0700178 jbd2_revoke_table_cache = kmem_cache_create("jbd2_revoke_table",
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700179 sizeof(struct jbd2_revoke_table_s),
Dave Kleikamp470decc2006-10-11 01:20:57 -0700180 0, 0, NULL, NULL);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700181 if (jbd2_revoke_table_cache == 0) {
182 kmem_cache_destroy(jbd2_revoke_record_cache);
183 jbd2_revoke_record_cache = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700184 return -ENOMEM;
185 }
186 return 0;
187}
188
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700189void jbd2_journal_destroy_revoke_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700190{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700191 kmem_cache_destroy(jbd2_revoke_record_cache);
192 jbd2_revoke_record_cache = NULL;
193 kmem_cache_destroy(jbd2_revoke_table_cache);
194 jbd2_revoke_table_cache = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700195}
196
197/* Initialise the revoke table for a given journal to a given size. */
198
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700199int jbd2_journal_init_revoke(journal_t *journal, int hash_size)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700200{
201 int shift, tmp;
202
203 J_ASSERT (journal->j_revoke_table[0] == NULL);
204
205 shift = 0;
206 tmp = hash_size;
207 while((tmp >>= 1UL) != 0UL)
208 shift++;
209
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700210 journal->j_revoke_table[0] = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700211 if (!journal->j_revoke_table[0])
212 return -ENOMEM;
213 journal->j_revoke = journal->j_revoke_table[0];
214
215 /* Check that the hash_size is a power of two */
216 J_ASSERT ((hash_size & (hash_size-1)) == 0);
217
218 journal->j_revoke->hash_size = hash_size;
219
220 journal->j_revoke->hash_shift = shift;
221
222 journal->j_revoke->hash_table =
223 kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
224 if (!journal->j_revoke->hash_table) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700225 kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[0]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700226 journal->j_revoke = NULL;
227 return -ENOMEM;
228 }
229
230 for (tmp = 0; tmp < hash_size; tmp++)
231 INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]);
232
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700233 journal->j_revoke_table[1] = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700234 if (!journal->j_revoke_table[1]) {
235 kfree(journal->j_revoke_table[0]->hash_table);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700236 kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[0]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700237 return -ENOMEM;
238 }
239
240 journal->j_revoke = journal->j_revoke_table[1];
241
242 /* Check that the hash_size is a power of two */
243 J_ASSERT ((hash_size & (hash_size-1)) == 0);
244
245 journal->j_revoke->hash_size = hash_size;
246
247 journal->j_revoke->hash_shift = shift;
248
249 journal->j_revoke->hash_table =
250 kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
251 if (!journal->j_revoke->hash_table) {
252 kfree(journal->j_revoke_table[0]->hash_table);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700253 kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[0]);
254 kmem_cache_free(jbd2_revoke_table_cache, journal->j_revoke_table[1]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700255 journal->j_revoke = NULL;
256 return -ENOMEM;
257 }
258
259 for (tmp = 0; tmp < hash_size; tmp++)
260 INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]);
261
262 spin_lock_init(&journal->j_revoke_lock);
263
264 return 0;
265}
266
267/* Destoy a journal's revoke table. The table must already be empty! */
268
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700269void jbd2_journal_destroy_revoke(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700270{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700271 struct jbd2_revoke_table_s *table;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700272 struct list_head *hash_list;
273 int i;
274
275 table = journal->j_revoke_table[0];
276 if (!table)
277 return;
278
279 for (i=0; i<table->hash_size; i++) {
280 hash_list = &table->hash_table[i];
281 J_ASSERT (list_empty(hash_list));
282 }
283
284 kfree(table->hash_table);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700285 kmem_cache_free(jbd2_revoke_table_cache, table);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700286 journal->j_revoke = NULL;
287
288 table = journal->j_revoke_table[1];
289 if (!table)
290 return;
291
292 for (i=0; i<table->hash_size; i++) {
293 hash_list = &table->hash_table[i];
294 J_ASSERT (list_empty(hash_list));
295 }
296
297 kfree(table->hash_table);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700298 kmem_cache_free(jbd2_revoke_table_cache, table);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700299 journal->j_revoke = NULL;
300}
301
302
303#ifdef __KERNEL__
304
305/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700306 * jbd2_journal_revoke: revoke a given buffer_head from the journal. This
Dave Kleikamp470decc2006-10-11 01:20:57 -0700307 * prevents the block from being replayed during recovery if we take a
308 * crash after this current transaction commits. Any subsequent
309 * metadata writes of the buffer in this transaction cancel the
310 * revoke.
311 *
312 * Note that this call may block --- it is up to the caller to make
313 * sure that there are no further calls to journal_write_metadata
314 * before the revoke is complete. In ext3, this implies calling the
315 * revoke before clearing the block bitmap when we are deleting
316 * metadata.
317 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700318 * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a
Dave Kleikamp470decc2006-10-11 01:20:57 -0700319 * parameter, but does _not_ forget the buffer_head if the bh was only
320 * found implicitly.
321 *
322 * bh_in may not be a journalled buffer - it may have come off
323 * the hash tables without an attached journal_head.
324 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700325 * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count
Dave Kleikamp470decc2006-10-11 01:20:57 -0700326 * by one.
327 */
328
Mingming Cao29971762006-10-11 01:21:09 -0700329int jbd2_journal_revoke(handle_t *handle, sector_t blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700330 struct buffer_head *bh_in)
331{
332 struct buffer_head *bh = NULL;
333 journal_t *journal;
334 struct block_device *bdev;
335 int err;
336
337 might_sleep();
338 if (bh_in)
339 BUFFER_TRACE(bh_in, "enter");
340
341 journal = handle->h_transaction->t_journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700342 if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){
Dave Kleikamp470decc2006-10-11 01:20:57 -0700343 J_ASSERT (!"Cannot set revoke feature!");
344 return -EINVAL;
345 }
346
347 bdev = journal->j_fs_dev;
348 bh = bh_in;
349
350 if (!bh) {
351 bh = __find_get_block(bdev, blocknr, journal->j_blocksize);
352 if (bh)
353 BUFFER_TRACE(bh, "found on hash");
354 }
355#ifdef JBD_EXPENSIVE_CHECKING
356 else {
357 struct buffer_head *bh2;
358
359 /* If there is a different buffer_head lying around in
360 * memory anywhere... */
361 bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize);
362 if (bh2) {
363 /* ... and it has RevokeValid status... */
364 if (bh2 != bh && buffer_revokevalid(bh2))
365 /* ...then it better be revoked too,
366 * since it's illegal to create a revoke
367 * record against a buffer_head which is
368 * not marked revoked --- that would
369 * risk missing a subsequent revoke
370 * cancel. */
371 J_ASSERT_BH(bh2, buffer_revoked(bh2));
372 put_bh(bh2);
373 }
374 }
375#endif
376
377 /* We really ought not ever to revoke twice in a row without
378 first having the revoke cancelled: it's illegal to free a
379 block twice without allocating it in between! */
380 if (bh) {
381 if (!J_EXPECT_BH(bh, !buffer_revoked(bh),
382 "inconsistent data on disk")) {
383 if (!bh_in)
384 brelse(bh);
385 return -EIO;
386 }
387 set_buffer_revoked(bh);
388 set_buffer_revokevalid(bh);
389 if (bh_in) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700390 BUFFER_TRACE(bh_in, "call jbd2_journal_forget");
391 jbd2_journal_forget(handle, bh_in);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700392 } else {
393 BUFFER_TRACE(bh, "call brelse");
394 __brelse(bh);
395 }
396 }
397
Mingming Cao29971762006-10-11 01:21:09 -0700398 jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700399 err = insert_revoke_hash(journal, blocknr,
400 handle->h_transaction->t_tid);
401 BUFFER_TRACE(bh_in, "exit");
402 return err;
403}
404
405/*
406 * Cancel an outstanding revoke. For use only internally by the
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700407 * journaling code (called from jbd2_journal_get_write_access).
Dave Kleikamp470decc2006-10-11 01:20:57 -0700408 *
409 * We trust buffer_revoked() on the buffer if the buffer is already
410 * being journaled: if there is no revoke pending on the buffer, then we
411 * don't do anything here.
412 *
413 * This would break if it were possible for a buffer to be revoked and
414 * discarded, and then reallocated within the same transaction. In such
415 * a case we would have lost the revoked bit, but when we arrived here
416 * the second time we would still have a pending revoke to cancel. So,
417 * do not trust the Revoked bit on buffers unless RevokeValid is also
418 * set.
419 *
420 * The caller must have the journal locked.
421 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700422int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700423{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700424 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700425 journal_t *journal = handle->h_transaction->t_journal;
426 int need_cancel;
427 int did_revoke = 0; /* akpm: debug */
428 struct buffer_head *bh = jh2bh(jh);
429
430 jbd_debug(4, "journal_head %p, cancelling revoke\n", jh);
431
432 /* Is the existing Revoke bit valid? If so, we trust it, and
433 * only perform the full cancel if the revoke bit is set. If
434 * not, we can't trust the revoke bit, and we need to do the
435 * full search for a revoke record. */
436 if (test_set_buffer_revokevalid(bh)) {
437 need_cancel = test_clear_buffer_revoked(bh);
438 } else {
439 need_cancel = 1;
440 clear_buffer_revoked(bh);
441 }
442
443 if (need_cancel) {
444 record = find_revoke_record(journal, bh->b_blocknr);
445 if (record) {
446 jbd_debug(4, "cancelled existing revoke on "
447 "blocknr %llu\n", (unsigned long long)bh->b_blocknr);
448 spin_lock(&journal->j_revoke_lock);
449 list_del(&record->hash);
450 spin_unlock(&journal->j_revoke_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700451 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700452 did_revoke = 1;
453 }
454 }
455
456#ifdef JBD_EXPENSIVE_CHECKING
457 /* There better not be one left behind by now! */
458 record = find_revoke_record(journal, bh->b_blocknr);
459 J_ASSERT_JH(jh, record == NULL);
460#endif
461
462 /* Finally, have we just cleared revoke on an unhashed
463 * buffer_head? If so, we'd better make sure we clear the
464 * revoked status on any hashed alias too, otherwise the revoke
465 * state machine will get very upset later on. */
466 if (need_cancel) {
467 struct buffer_head *bh2;
468 bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size);
469 if (bh2) {
470 if (bh2 != bh)
471 clear_buffer_revoked(bh2);
472 __brelse(bh2);
473 }
474 }
475 return did_revoke;
476}
477
478/* journal_switch_revoke table select j_revoke for next transaction
479 * we do not want to suspend any processing until all revokes are
480 * written -bzzz
481 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700482void jbd2_journal_switch_revoke_table(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700483{
484 int i;
485
486 if (journal->j_revoke == journal->j_revoke_table[0])
487 journal->j_revoke = journal->j_revoke_table[1];
488 else
489 journal->j_revoke = journal->j_revoke_table[0];
490
491 for (i = 0; i < journal->j_revoke->hash_size; i++)
492 INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]);
493}
494
495/*
496 * Write revoke records to the journal for all entries in the current
497 * revoke hash, deleting the entries as we go.
498 *
499 * Called with the journal lock held.
500 */
501
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700502void jbd2_journal_write_revoke_records(journal_t *journal,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700503 transaction_t *transaction)
504{
505 struct journal_head *descriptor;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700506 struct jbd2_revoke_record_s *record;
507 struct jbd2_revoke_table_s *revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700508 struct list_head *hash_list;
509 int i, offset, count;
510
511 descriptor = NULL;
512 offset = 0;
513 count = 0;
514
515 /* select revoke table for committing transaction */
516 revoke = journal->j_revoke == journal->j_revoke_table[0] ?
517 journal->j_revoke_table[1] : journal->j_revoke_table[0];
518
519 for (i = 0; i < revoke->hash_size; i++) {
520 hash_list = &revoke->hash_table[i];
521
522 while (!list_empty(hash_list)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700523 record = (struct jbd2_revoke_record_s *)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700524 hash_list->next;
525 write_one_revoke_record(journal, transaction,
526 &descriptor, &offset,
527 record);
528 count++;
529 list_del(&record->hash);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700530 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700531 }
532 }
533 if (descriptor)
534 flush_descriptor(journal, descriptor, offset);
535 jbd_debug(1, "Wrote %d revoke records\n", count);
536}
537
538/*
539 * Write out one revoke record. We need to create a new descriptor
540 * block if the old one is full or if we have not already created one.
541 */
542
543static void write_one_revoke_record(journal_t *journal,
544 transaction_t *transaction,
545 struct journal_head **descriptorp,
546 int *offsetp,
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700547 struct jbd2_revoke_record_s *record)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700548{
549 struct journal_head *descriptor;
550 int offset;
551 journal_header_t *header;
552
553 /* If we are already aborting, this all becomes a noop. We
554 still need to go round the loop in
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700555 jbd2_journal_write_revoke_records in order to free all of the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700556 revoke records: only the IO to the journal is omitted. */
557 if (is_journal_aborted(journal))
558 return;
559
560 descriptor = *descriptorp;
561 offset = *offsetp;
562
563 /* Make sure we have a descriptor with space left for the record */
564 if (descriptor) {
565 if (offset == journal->j_blocksize) {
566 flush_descriptor(journal, descriptor, offset);
567 descriptor = NULL;
568 }
569 }
570
571 if (!descriptor) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700572 descriptor = jbd2_journal_get_descriptor_buffer(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700573 if (!descriptor)
574 return;
575 header = (journal_header_t *) &jh2bh(descriptor)->b_data[0];
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700576 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
577 header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700578 header->h_sequence = cpu_to_be32(transaction->t_tid);
579
580 /* Record it so that we can wait for IO completion later */
581 JBUFFER_TRACE(descriptor, "file as BJ_LogCtl");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700582 jbd2_journal_file_buffer(descriptor, transaction, BJ_LogCtl);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700583
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700584 offset = sizeof(jbd2_journal_revoke_header_t);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700585 *descriptorp = descriptor;
586 }
587
Zach Brownb517bea2006-10-11 01:21:08 -0700588 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) {
589 * ((__be64 *)(&jh2bh(descriptor)->b_data[offset])) =
590 cpu_to_be64(record->blocknr);
591 offset += 8;
592
593 } else {
594 * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) =
595 cpu_to_be32(record->blocknr);
596 offset += 4;
597 }
598
Dave Kleikamp470decc2006-10-11 01:20:57 -0700599 *offsetp = offset;
600}
601
602/*
603 * Flush a revoke descriptor out to the journal. If we are aborting,
604 * this is a noop; otherwise we are generating a buffer which needs to
605 * be waited for during commit, so it has to go onto the appropriate
606 * journal buffer list.
607 */
608
609static void flush_descriptor(journal_t *journal,
610 struct journal_head *descriptor,
611 int offset)
612{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700613 jbd2_journal_revoke_header_t *header;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700614 struct buffer_head *bh = jh2bh(descriptor);
615
616 if (is_journal_aborted(journal)) {
617 put_bh(bh);
618 return;
619 }
620
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700621 header = (jbd2_journal_revoke_header_t *) jh2bh(descriptor)->b_data;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700622 header->r_count = cpu_to_be32(offset);
623 set_buffer_jwrite(bh);
624 BUFFER_TRACE(bh, "write");
625 set_buffer_dirty(bh);
626 ll_rw_block(SWRITE, 1, &bh);
627}
628#endif
629
630/*
631 * Revoke support for recovery.
632 *
633 * Recovery needs to be able to:
634 *
635 * record all revoke records, including the tid of the latest instance
636 * of each revoke in the journal
637 *
638 * check whether a given block in a given transaction should be replayed
639 * (ie. has not been revoked by a revoke record in that or a subsequent
640 * transaction)
641 *
642 * empty the revoke table after recovery.
643 */
644
645/*
646 * First, setting revoke records. We create a new revoke record for
647 * every block ever revoked in the log as we scan it for recovery, and
648 * we update the existing records if we find multiple revokes for a
649 * single block.
650 */
651
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700652int jbd2_journal_set_revoke(journal_t *journal,
Mingming Cao29971762006-10-11 01:21:09 -0700653 sector_t blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700654 tid_t sequence)
655{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700656 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700657
658 record = find_revoke_record(journal, blocknr);
659 if (record) {
660 /* If we have multiple occurrences, only record the
661 * latest sequence number in the hashed record */
662 if (tid_gt(sequence, record->sequence))
663 record->sequence = sequence;
664 return 0;
665 }
666 return insert_revoke_hash(journal, blocknr, sequence);
667}
668
669/*
670 * Test revoke records. For a given block referenced in the log, has
671 * that block been revoked? A revoke record with a given transaction
672 * sequence number revokes all blocks in that transaction and earlier
673 * ones, but later transactions still need replayed.
674 */
675
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700676int jbd2_journal_test_revoke(journal_t *journal,
Mingming Cao29971762006-10-11 01:21:09 -0700677 sector_t blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700678 tid_t sequence)
679{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700680 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700681
682 record = find_revoke_record(journal, blocknr);
683 if (!record)
684 return 0;
685 if (tid_gt(sequence, record->sequence))
686 return 0;
687 return 1;
688}
689
690/*
691 * Finally, once recovery is over, we need to clear the revoke table so
692 * that it can be reused by the running filesystem.
693 */
694
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700695void jbd2_journal_clear_revoke(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700696{
697 int i;
698 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700699 struct jbd2_revoke_record_s *record;
700 struct jbd2_revoke_table_s *revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700701
702 revoke = journal->j_revoke;
703
704 for (i = 0; i < revoke->hash_size; i++) {
705 hash_list = &revoke->hash_table[i];
706 while (!list_empty(hash_list)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700707 record = (struct jbd2_revoke_record_s*) hash_list->next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700708 list_del(&record->hash);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700709 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700710 }
711 }
712}