blob: 257ff26257655f4573669a5ab6cb7ee817e0371b [file] [log] [blame]
Dave Kleikamp470decc2006-10-11 01:20:57 -07001/*
Uwe Kleine-König58862692007-05-09 07:51:49 +02002 * linux/fs/jbd2/revoke.c
Dave Kleikamp470decc2006-10-11 01:20:57 -07003 *
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>
Dave Kleikamp470decc2006-10-11 01:20:57 -070069#include <linux/init.h>
70#endif
vignesh babuf4823942007-07-15 23:41:17 -070071#include <linux/log2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070072
Christoph Lametere18b8902006-12-06 20:33:20 -080073static struct kmem_cache *jbd2_revoke_record_cache;
74static struct kmem_cache *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 Cao18eba7a2006-10-11 01:21:13 -070084 unsigned long long 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 Cao18eba7a2006-10-11 01:21:13 -0700109static inline int hash(journal_t *journal, unsigned long long 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 Cao18eba7a2006-10-11 01:21:13 -0700120static int insert_revoke_hash(journal_t *journal, unsigned long long 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;
Harvey Harrison329d2912008-04-17 10:38:59 -0400142 jbd_debug(1, "ENOMEM in %s, retrying\n", __func__);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700143 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 Cao18eba7a2006-10-11 01:21:13 -0700150 unsigned long long 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
Duane Griffin9fa27c82008-04-28 09:40:00 -0400170void jbd2_journal_destroy_revoke_caches(void)
171{
172 if (jbd2_revoke_record_cache) {
173 kmem_cache_destroy(jbd2_revoke_record_cache);
174 jbd2_revoke_record_cache = NULL;
175 }
176 if (jbd2_revoke_table_cache) {
177 kmem_cache_destroy(jbd2_revoke_table_cache);
178 jbd2_revoke_table_cache = NULL;
179 }
180}
181
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700182int __init jbd2_journal_init_revoke_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700183{
Duane Griffin9fa27c82008-04-28 09:40:00 -0400184 J_ASSERT(!jbd2_revoke_record_cache);
185 J_ASSERT(!jbd2_revoke_table_cache);
186
Johann Lombardia920e942006-10-11 01:21:00 -0700187 jbd2_revoke_record_cache = kmem_cache_create("jbd2_revoke_record",
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700188 sizeof(struct jbd2_revoke_record_s),
Mingming Cao77160952008-01-28 23:58:27 -0500189 0,
190 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
191 NULL);
Al Viro1076d172008-03-29 03:07:18 +0000192 if (!jbd2_revoke_record_cache)
Duane Griffin9fa27c82008-04-28 09:40:00 -0400193 goto record_cache_failure;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700194
Johann Lombardia920e942006-10-11 01:21:00 -0700195 jbd2_revoke_table_cache = kmem_cache_create("jbd2_revoke_table",
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700196 sizeof(struct jbd2_revoke_table_s),
Mingming Cao77160952008-01-28 23:58:27 -0500197 0, SLAB_TEMPORARY, NULL);
Duane Griffin9fa27c82008-04-28 09:40:00 -0400198 if (!jbd2_revoke_table_cache)
199 goto table_cache_failure;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700200 return 0;
Duane Griffin9fa27c82008-04-28 09:40:00 -0400201table_cache_failure:
202 jbd2_journal_destroy_revoke_caches();
203record_cache_failure:
204 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700205}
206
Duane Griffin83c49522008-04-17 10:38:59 -0400207static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700208{
Duane Griffin83c49522008-04-17 10:38:59 -0400209 int shift = 0;
210 int tmp = hash_size;
211 struct jbd2_revoke_table_s *table;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700212
Duane Griffin83c49522008-04-17 10:38:59 -0400213 table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
214 if (!table)
215 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700216
Dave Kleikamp470decc2006-10-11 01:20:57 -0700217 while((tmp >>= 1UL) != 0UL)
218 shift++;
219
Duane Griffin83c49522008-04-17 10:38:59 -0400220 table->hash_size = hash_size;
221 table->hash_shift = shift;
222 table->hash_table =
Dave Kleikamp470decc2006-10-11 01:20:57 -0700223 kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
Duane Griffin83c49522008-04-17 10:38:59 -0400224 if (!table->hash_table) {
225 kmem_cache_free(jbd2_revoke_table_cache, table);
226 table = NULL;
227 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700228 }
229
230 for (tmp = 0; tmp < hash_size; tmp++)
Duane Griffin83c49522008-04-17 10:38:59 -0400231 INIT_LIST_HEAD(&table->hash_table[tmp]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700232
Duane Griffin83c49522008-04-17 10:38:59 -0400233out:
234 return table;
235}
236
237static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table)
238{
239 int i;
240 struct list_head *hash_list;
241
242 for (i = 0; i < table->hash_size; i++) {
243 hash_list = &table->hash_table[i];
244 J_ASSERT(list_empty(hash_list));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700245 }
246
Duane Griffin83c49522008-04-17 10:38:59 -0400247 kfree(table->hash_table);
248 kmem_cache_free(jbd2_revoke_table_cache, table);
249}
250
251/* Initialise the revoke table for a given journal to a given size. */
252int jbd2_journal_init_revoke(journal_t *journal, int hash_size)
253{
254 J_ASSERT(journal->j_revoke_table[0] == NULL);
255 J_ASSERT(is_power_of_2(hash_size));
256
257 journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size);
258 if (!journal->j_revoke_table[0])
259 goto fail0;
260
261 journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size);
262 if (!journal->j_revoke_table[1])
263 goto fail1;
264
Dave Kleikamp470decc2006-10-11 01:20:57 -0700265 journal->j_revoke = journal->j_revoke_table[1];
266
Dave Kleikamp470decc2006-10-11 01:20:57 -0700267 spin_lock_init(&journal->j_revoke_lock);
268
269 return 0;
Duane Griffin83c49522008-04-17 10:38:59 -0400270
271fail1:
272 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
273fail0:
274 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700275}
276
Duane Griffin83c49522008-04-17 10:38:59 -0400277/* Destroy a journal's revoke table. The table must already be empty! */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700278void jbd2_journal_destroy_revoke(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700279{
Dave Kleikamp470decc2006-10-11 01:20:57 -0700280 journal->j_revoke = NULL;
Duane Griffin83c49522008-04-17 10:38:59 -0400281 if (journal->j_revoke_table[0])
282 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
283 if (journal->j_revoke_table[1])
284 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700285}
286
287
288#ifdef __KERNEL__
289
290/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700291 * jbd2_journal_revoke: revoke a given buffer_head from the journal. This
Dave Kleikamp470decc2006-10-11 01:20:57 -0700292 * prevents the block from being replayed during recovery if we take a
293 * crash after this current transaction commits. Any subsequent
294 * metadata writes of the buffer in this transaction cancel the
295 * revoke.
296 *
297 * Note that this call may block --- it is up to the caller to make
298 * sure that there are no further calls to journal_write_metadata
299 * before the revoke is complete. In ext3, this implies calling the
300 * revoke before clearing the block bitmap when we are deleting
301 * metadata.
302 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700303 * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a
Dave Kleikamp470decc2006-10-11 01:20:57 -0700304 * parameter, but does _not_ forget the buffer_head if the bh was only
305 * found implicitly.
306 *
307 * bh_in may not be a journalled buffer - it may have come off
308 * the hash tables without an attached journal_head.
309 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700310 * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count
Dave Kleikamp470decc2006-10-11 01:20:57 -0700311 * by one.
312 */
313
Mingming Cao18eba7a2006-10-11 01:21:13 -0700314int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700315 struct buffer_head *bh_in)
316{
317 struct buffer_head *bh = NULL;
318 journal_t *journal;
319 struct block_device *bdev;
320 int err;
321
322 might_sleep();
323 if (bh_in)
324 BUFFER_TRACE(bh_in, "enter");
325
326 journal = handle->h_transaction->t_journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700327 if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){
Dave Kleikamp470decc2006-10-11 01:20:57 -0700328 J_ASSERT (!"Cannot set revoke feature!");
329 return -EINVAL;
330 }
331
332 bdev = journal->j_fs_dev;
333 bh = bh_in;
334
335 if (!bh) {
336 bh = __find_get_block(bdev, blocknr, journal->j_blocksize);
337 if (bh)
338 BUFFER_TRACE(bh, "found on hash");
339 }
Mingming Caocd02ff02007-10-16 18:38:25 -0400340#ifdef JBD2_EXPENSIVE_CHECKING
Dave Kleikamp470decc2006-10-11 01:20:57 -0700341 else {
342 struct buffer_head *bh2;
343
344 /* If there is a different buffer_head lying around in
345 * memory anywhere... */
346 bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize);
347 if (bh2) {
348 /* ... and it has RevokeValid status... */
349 if (bh2 != bh && buffer_revokevalid(bh2))
350 /* ...then it better be revoked too,
351 * since it's illegal to create a revoke
352 * record against a buffer_head which is
353 * not marked revoked --- that would
354 * risk missing a subsequent revoke
355 * cancel. */
356 J_ASSERT_BH(bh2, buffer_revoked(bh2));
357 put_bh(bh2);
358 }
359 }
360#endif
361
362 /* We really ought not ever to revoke twice in a row without
363 first having the revoke cancelled: it's illegal to free a
364 block twice without allocating it in between! */
365 if (bh) {
366 if (!J_EXPECT_BH(bh, !buffer_revoked(bh),
367 "inconsistent data on disk")) {
368 if (!bh_in)
369 brelse(bh);
370 return -EIO;
371 }
372 set_buffer_revoked(bh);
373 set_buffer_revokevalid(bh);
374 if (bh_in) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700375 BUFFER_TRACE(bh_in, "call jbd2_journal_forget");
376 jbd2_journal_forget(handle, bh_in);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700377 } else {
378 BUFFER_TRACE(bh, "call brelse");
379 __brelse(bh);
380 }
381 }
382
Mingming Cao29971762006-10-11 01:21:09 -0700383 jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700384 err = insert_revoke_hash(journal, blocknr,
385 handle->h_transaction->t_tid);
386 BUFFER_TRACE(bh_in, "exit");
387 return err;
388}
389
390/*
391 * Cancel an outstanding revoke. For use only internally by the
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700392 * journaling code (called from jbd2_journal_get_write_access).
Dave Kleikamp470decc2006-10-11 01:20:57 -0700393 *
394 * We trust buffer_revoked() on the buffer if the buffer is already
395 * being journaled: if there is no revoke pending on the buffer, then we
396 * don't do anything here.
397 *
398 * This would break if it were possible for a buffer to be revoked and
399 * discarded, and then reallocated within the same transaction. In such
400 * a case we would have lost the revoked bit, but when we arrived here
401 * the second time we would still have a pending revoke to cancel. So,
402 * do not trust the Revoked bit on buffers unless RevokeValid is also
403 * set.
404 *
405 * The caller must have the journal locked.
406 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700407int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700408{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700409 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700410 journal_t *journal = handle->h_transaction->t_journal;
411 int need_cancel;
412 int did_revoke = 0; /* akpm: debug */
413 struct buffer_head *bh = jh2bh(jh);
414
415 jbd_debug(4, "journal_head %p, cancelling revoke\n", jh);
416
417 /* Is the existing Revoke bit valid? If so, we trust it, and
418 * only perform the full cancel if the revoke bit is set. If
419 * not, we can't trust the revoke bit, and we need to do the
420 * full search for a revoke record. */
421 if (test_set_buffer_revokevalid(bh)) {
422 need_cancel = test_clear_buffer_revoked(bh);
423 } else {
424 need_cancel = 1;
425 clear_buffer_revoked(bh);
426 }
427
428 if (need_cancel) {
429 record = find_revoke_record(journal, bh->b_blocknr);
430 if (record) {
431 jbd_debug(4, "cancelled existing revoke on "
432 "blocknr %llu\n", (unsigned long long)bh->b_blocknr);
433 spin_lock(&journal->j_revoke_lock);
434 list_del(&record->hash);
435 spin_unlock(&journal->j_revoke_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700436 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700437 did_revoke = 1;
438 }
439 }
440
Mingming Caocd02ff02007-10-16 18:38:25 -0400441#ifdef JBD2_EXPENSIVE_CHECKING
Dave Kleikamp470decc2006-10-11 01:20:57 -0700442 /* There better not be one left behind by now! */
443 record = find_revoke_record(journal, bh->b_blocknr);
444 J_ASSERT_JH(jh, record == NULL);
445#endif
446
447 /* Finally, have we just cleared revoke on an unhashed
448 * buffer_head? If so, we'd better make sure we clear the
449 * revoked status on any hashed alias too, otherwise the revoke
450 * state machine will get very upset later on. */
451 if (need_cancel) {
452 struct buffer_head *bh2;
453 bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size);
454 if (bh2) {
455 if (bh2 != bh)
456 clear_buffer_revoked(bh2);
457 __brelse(bh2);
458 }
459 }
460 return did_revoke;
461}
462
463/* journal_switch_revoke table select j_revoke for next transaction
464 * we do not want to suspend any processing until all revokes are
465 * written -bzzz
466 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700467void jbd2_journal_switch_revoke_table(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700468{
469 int i;
470
471 if (journal->j_revoke == journal->j_revoke_table[0])
472 journal->j_revoke = journal->j_revoke_table[1];
473 else
474 journal->j_revoke = journal->j_revoke_table[0];
475
476 for (i = 0; i < journal->j_revoke->hash_size; i++)
477 INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]);
478}
479
480/*
481 * Write revoke records to the journal for all entries in the current
482 * revoke hash, deleting the entries as we go.
483 *
484 * Called with the journal lock held.
485 */
486
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700487void jbd2_journal_write_revoke_records(journal_t *journal,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700488 transaction_t *transaction)
489{
490 struct journal_head *descriptor;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700491 struct jbd2_revoke_record_s *record;
492 struct jbd2_revoke_table_s *revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700493 struct list_head *hash_list;
494 int i, offset, count;
495
496 descriptor = NULL;
497 offset = 0;
498 count = 0;
499
500 /* select revoke table for committing transaction */
501 revoke = journal->j_revoke == journal->j_revoke_table[0] ?
502 journal->j_revoke_table[1] : journal->j_revoke_table[0];
503
504 for (i = 0; i < revoke->hash_size; i++) {
505 hash_list = &revoke->hash_table[i];
506
507 while (!list_empty(hash_list)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700508 record = (struct jbd2_revoke_record_s *)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700509 hash_list->next;
510 write_one_revoke_record(journal, transaction,
511 &descriptor, &offset,
512 record);
513 count++;
514 list_del(&record->hash);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700515 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700516 }
517 }
518 if (descriptor)
519 flush_descriptor(journal, descriptor, offset);
520 jbd_debug(1, "Wrote %d revoke records\n", count);
521}
522
523/*
524 * Write out one revoke record. We need to create a new descriptor
525 * block if the old one is full or if we have not already created one.
526 */
527
528static void write_one_revoke_record(journal_t *journal,
529 transaction_t *transaction,
530 struct journal_head **descriptorp,
531 int *offsetp,
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700532 struct jbd2_revoke_record_s *record)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700533{
534 struct journal_head *descriptor;
535 int offset;
536 journal_header_t *header;
537
538 /* If we are already aborting, this all becomes a noop. We
539 still need to go round the loop in
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700540 jbd2_journal_write_revoke_records in order to free all of the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700541 revoke records: only the IO to the journal is omitted. */
542 if (is_journal_aborted(journal))
543 return;
544
545 descriptor = *descriptorp;
546 offset = *offsetp;
547
548 /* Make sure we have a descriptor with space left for the record */
549 if (descriptor) {
550 if (offset == journal->j_blocksize) {
551 flush_descriptor(journal, descriptor, offset);
552 descriptor = NULL;
553 }
554 }
555
556 if (!descriptor) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700557 descriptor = jbd2_journal_get_descriptor_buffer(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700558 if (!descriptor)
559 return;
560 header = (journal_header_t *) &jh2bh(descriptor)->b_data[0];
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700561 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
562 header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700563 header->h_sequence = cpu_to_be32(transaction->t_tid);
564
565 /* Record it so that we can wait for IO completion later */
566 JBUFFER_TRACE(descriptor, "file as BJ_LogCtl");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700567 jbd2_journal_file_buffer(descriptor, transaction, BJ_LogCtl);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700568
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700569 offset = sizeof(jbd2_journal_revoke_header_t);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700570 *descriptorp = descriptor;
571 }
572
Zach Brownb517bea2006-10-11 01:21:08 -0700573 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) {
574 * ((__be64 *)(&jh2bh(descriptor)->b_data[offset])) =
575 cpu_to_be64(record->blocknr);
576 offset += 8;
577
578 } else {
579 * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) =
580 cpu_to_be32(record->blocknr);
581 offset += 4;
582 }
583
Dave Kleikamp470decc2006-10-11 01:20:57 -0700584 *offsetp = offset;
585}
586
587/*
588 * Flush a revoke descriptor out to the journal. If we are aborting,
589 * this is a noop; otherwise we are generating a buffer which needs to
590 * be waited for during commit, so it has to go onto the appropriate
591 * journal buffer list.
592 */
593
594static void flush_descriptor(journal_t *journal,
595 struct journal_head *descriptor,
596 int offset)
597{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700598 jbd2_journal_revoke_header_t *header;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700599 struct buffer_head *bh = jh2bh(descriptor);
600
601 if (is_journal_aborted(journal)) {
602 put_bh(bh);
603 return;
604 }
605
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700606 header = (jbd2_journal_revoke_header_t *) jh2bh(descriptor)->b_data;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700607 header->r_count = cpu_to_be32(offset);
608 set_buffer_jwrite(bh);
609 BUFFER_TRACE(bh, "write");
610 set_buffer_dirty(bh);
611 ll_rw_block(SWRITE, 1, &bh);
612}
613#endif
614
615/*
616 * Revoke support for recovery.
617 *
618 * Recovery needs to be able to:
619 *
620 * record all revoke records, including the tid of the latest instance
621 * of each revoke in the journal
622 *
623 * check whether a given block in a given transaction should be replayed
624 * (ie. has not been revoked by a revoke record in that or a subsequent
625 * transaction)
626 *
627 * empty the revoke table after recovery.
628 */
629
630/*
631 * First, setting revoke records. We create a new revoke record for
632 * every block ever revoked in the log as we scan it for recovery, and
633 * we update the existing records if we find multiple revokes for a
634 * single block.
635 */
636
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700637int jbd2_journal_set_revoke(journal_t *journal,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700638 unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700639 tid_t sequence)
640{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700641 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700642
643 record = find_revoke_record(journal, blocknr);
644 if (record) {
645 /* If we have multiple occurrences, only record the
646 * latest sequence number in the hashed record */
647 if (tid_gt(sequence, record->sequence))
648 record->sequence = sequence;
649 return 0;
650 }
651 return insert_revoke_hash(journal, blocknr, sequence);
652}
653
654/*
655 * Test revoke records. For a given block referenced in the log, has
656 * that block been revoked? A revoke record with a given transaction
657 * sequence number revokes all blocks in that transaction and earlier
658 * ones, but later transactions still need replayed.
659 */
660
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700661int jbd2_journal_test_revoke(journal_t *journal,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700662 unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700663 tid_t sequence)
664{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700665 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700666
667 record = find_revoke_record(journal, blocknr);
668 if (!record)
669 return 0;
670 if (tid_gt(sequence, record->sequence))
671 return 0;
672 return 1;
673}
674
675/*
676 * Finally, once recovery is over, we need to clear the revoke table so
677 * that it can be reused by the running filesystem.
678 */
679
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700680void jbd2_journal_clear_revoke(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700681{
682 int i;
683 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700684 struct jbd2_revoke_record_s *record;
685 struct jbd2_revoke_table_s *revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700686
687 revoke = journal->j_revoke;
688
689 for (i = 0; i < revoke->hash_size; i++) {
690 hash_list = &revoke->hash_table[i];
691 while (!list_empty(hash_list)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700692 record = (struct jbd2_revoke_record_s*) hash_list->next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700693 list_del(&record->hash);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700694 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700695 }
696 }
697}