blob: 9ad321fd63fdf73b4d7aa99ee5c78c6edbad9da4 [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.
Jan Kara86db97c2009-03-27 17:20:40 -040058 *
59 * Locking rules:
60 * We keep two hash tables of revoke records. One hashtable belongs to the
61 * running transaction (is pointed to by journal->j_revoke), the other one
62 * belongs to the committing transaction. Accesses to the second hash table
63 * happen only from the kjournald and no other thread touches this table. Also
64 * journal_switch_revoke_table() which switches which hashtable belongs to the
65 * running and which to the committing transaction is called only from
66 * kjournald. Therefore we need no locks when accessing the hashtable belonging
67 * to the committing transaction.
68 *
69 * All users operating on the hash table belonging to the running transaction
70 * have a handle to the transaction. Therefore they are safe from kjournald
71 * switching hash tables under them. For operations on the lists of entries in
72 * the hash table j_revoke_lock is used.
73 *
74 * Finally, also replay code uses the hash tables but at this moment noone else
75 * can touch them (filesystem isn't mounted yet) and hence no locking is
76 * needed.
Dave Kleikamp470decc2006-10-11 01:20:57 -070077 */
78
79#ifndef __KERNEL__
80#include "jfs_user.h"
81#else
82#include <linux/time.h>
83#include <linux/fs.h>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070084#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070085#include <linux/errno.h>
86#include <linux/slab.h>
87#include <linux/list.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070088#include <linux/init.h>
Theodore Ts'o67c457a2009-04-14 07:50:56 -040089#include <linux/bio.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070090#endif
vignesh babuf4823942007-07-15 23:41:17 -070091#include <linux/log2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070092
Christoph Lametere18b8902006-12-06 20:33:20 -080093static struct kmem_cache *jbd2_revoke_record_cache;
94static struct kmem_cache *jbd2_revoke_table_cache;
Dave Kleikamp470decc2006-10-11 01:20:57 -070095
96/* Each revoke record represents one single revoked block. During
97 journal replay, this involves recording the transaction ID of the
98 last transaction to revoke this block. */
99
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700100struct jbd2_revoke_record_s
Dave Kleikamp470decc2006-10-11 01:20:57 -0700101{
102 struct list_head hash;
103 tid_t sequence; /* Used for recovery only */
Mingming Cao18eba7a2006-10-11 01:21:13 -0700104 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700105};
106
107
108/* The revoke table is just a simple hash table of revoke records. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700109struct jbd2_revoke_table_s
Dave Kleikamp470decc2006-10-11 01:20:57 -0700110{
111 /* It is conceivable that we might want a larger hash table
112 * for recovery. Must be a power of two. */
113 int hash_size;
114 int hash_shift;
115 struct list_head *hash_table;
116};
117
118
119#ifdef __KERNEL__
120static void write_one_revoke_record(journal_t *, transaction_t *,
121 struct journal_head **, int *,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400122 struct jbd2_revoke_record_s *, int);
123static void flush_descriptor(journal_t *, struct journal_head *, int, int);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700124#endif
125
126/* Utility functions to maintain the revoke table */
127
128/* Borrowed from buffer.c: this is a tried and tested block hash function */
Mingming Cao18eba7a2006-10-11 01:21:13 -0700129static inline int hash(journal_t *journal, unsigned long long block)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700130{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700131 struct jbd2_revoke_table_s *table = journal->j_revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700132 int hash_shift = table->hash_shift;
Mingming Cao29971762006-10-11 01:21:09 -0700133 int hash = (int)block ^ (int)((block >> 31) >> 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700134
Mingming Cao29971762006-10-11 01:21:09 -0700135 return ((hash << (hash_shift - 6)) ^
136 (hash >> 13) ^
137 (hash << (hash_shift - 12))) & (table->hash_size - 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700138}
139
Mingming Cao18eba7a2006-10-11 01:21:13 -0700140static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700141 tid_t seq)
142{
143 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700144 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700145
146repeat:
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700147 record = kmem_cache_alloc(jbd2_revoke_record_cache, GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700148 if (!record)
149 goto oom;
150
151 record->sequence = seq;
152 record->blocknr = blocknr;
153 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
154 spin_lock(&journal->j_revoke_lock);
155 list_add(&record->hash, hash_list);
156 spin_unlock(&journal->j_revoke_lock);
157 return 0;
158
159oom:
160 if (!journal_oom_retry)
161 return -ENOMEM;
Harvey Harrison329d2912008-04-17 10:38:59 -0400162 jbd_debug(1, "ENOMEM in %s, retrying\n", __func__);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700163 yield();
164 goto repeat;
165}
166
167/* Find a revoke record in the journal's hash table. */
168
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700169static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700170 unsigned long long blocknr)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700171{
172 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700173 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700174
175 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
176
177 spin_lock(&journal->j_revoke_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700178 record = (struct jbd2_revoke_record_s *) hash_list->next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700179 while (&(record->hash) != hash_list) {
180 if (record->blocknr == blocknr) {
181 spin_unlock(&journal->j_revoke_lock);
182 return record;
183 }
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700184 record = (struct jbd2_revoke_record_s *) record->hash.next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700185 }
186 spin_unlock(&journal->j_revoke_lock);
187 return NULL;
188}
189
Duane Griffin9fa27c82008-04-28 09:40:00 -0400190void jbd2_journal_destroy_revoke_caches(void)
191{
192 if (jbd2_revoke_record_cache) {
193 kmem_cache_destroy(jbd2_revoke_record_cache);
194 jbd2_revoke_record_cache = NULL;
195 }
196 if (jbd2_revoke_table_cache) {
197 kmem_cache_destroy(jbd2_revoke_table_cache);
198 jbd2_revoke_table_cache = NULL;
199 }
200}
201
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700202int __init jbd2_journal_init_revoke_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700203{
Duane Griffin9fa27c82008-04-28 09:40:00 -0400204 J_ASSERT(!jbd2_revoke_record_cache);
205 J_ASSERT(!jbd2_revoke_table_cache);
206
Johann Lombardia920e942006-10-11 01:21:00 -0700207 jbd2_revoke_record_cache = kmem_cache_create("jbd2_revoke_record",
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700208 sizeof(struct jbd2_revoke_record_s),
Mingming Cao77160952008-01-28 23:58:27 -0500209 0,
210 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
211 NULL);
Al Viro1076d172008-03-29 03:07:18 +0000212 if (!jbd2_revoke_record_cache)
Duane Griffin9fa27c82008-04-28 09:40:00 -0400213 goto record_cache_failure;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700214
Johann Lombardia920e942006-10-11 01:21:00 -0700215 jbd2_revoke_table_cache = kmem_cache_create("jbd2_revoke_table",
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700216 sizeof(struct jbd2_revoke_table_s),
Mingming Cao77160952008-01-28 23:58:27 -0500217 0, SLAB_TEMPORARY, NULL);
Duane Griffin9fa27c82008-04-28 09:40:00 -0400218 if (!jbd2_revoke_table_cache)
219 goto table_cache_failure;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700220 return 0;
Duane Griffin9fa27c82008-04-28 09:40:00 -0400221table_cache_failure:
222 jbd2_journal_destroy_revoke_caches();
223record_cache_failure:
224 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700225}
226
Duane Griffin83c49522008-04-17 10:38:59 -0400227static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700228{
Duane Griffin83c49522008-04-17 10:38:59 -0400229 int shift = 0;
230 int tmp = hash_size;
231 struct jbd2_revoke_table_s *table;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700232
Duane Griffin83c49522008-04-17 10:38:59 -0400233 table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
234 if (!table)
235 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700236
Dave Kleikamp470decc2006-10-11 01:20:57 -0700237 while((tmp >>= 1UL) != 0UL)
238 shift++;
239
Duane Griffin83c49522008-04-17 10:38:59 -0400240 table->hash_size = hash_size;
241 table->hash_shift = shift;
242 table->hash_table =
Dave Kleikamp470decc2006-10-11 01:20:57 -0700243 kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
Duane Griffin83c49522008-04-17 10:38:59 -0400244 if (!table->hash_table) {
245 kmem_cache_free(jbd2_revoke_table_cache, table);
246 table = NULL;
247 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700248 }
249
250 for (tmp = 0; tmp < hash_size; tmp++)
Duane Griffin83c49522008-04-17 10:38:59 -0400251 INIT_LIST_HEAD(&table->hash_table[tmp]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700252
Duane Griffin83c49522008-04-17 10:38:59 -0400253out:
254 return table;
255}
256
257static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table)
258{
259 int i;
260 struct list_head *hash_list;
261
262 for (i = 0; i < table->hash_size; i++) {
263 hash_list = &table->hash_table[i];
264 J_ASSERT(list_empty(hash_list));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700265 }
266
Duane Griffin83c49522008-04-17 10:38:59 -0400267 kfree(table->hash_table);
268 kmem_cache_free(jbd2_revoke_table_cache, table);
269}
270
271/* Initialise the revoke table for a given journal to a given size. */
272int jbd2_journal_init_revoke(journal_t *journal, int hash_size)
273{
274 J_ASSERT(journal->j_revoke_table[0] == NULL);
275 J_ASSERT(is_power_of_2(hash_size));
276
277 journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size);
278 if (!journal->j_revoke_table[0])
279 goto fail0;
280
281 journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size);
282 if (!journal->j_revoke_table[1])
283 goto fail1;
284
Dave Kleikamp470decc2006-10-11 01:20:57 -0700285 journal->j_revoke = journal->j_revoke_table[1];
286
Dave Kleikamp470decc2006-10-11 01:20:57 -0700287 spin_lock_init(&journal->j_revoke_lock);
288
289 return 0;
Duane Griffin83c49522008-04-17 10:38:59 -0400290
291fail1:
292 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
293fail0:
294 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700295}
296
Duane Griffin83c49522008-04-17 10:38:59 -0400297/* Destroy a journal's revoke table. The table must already be empty! */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700298void jbd2_journal_destroy_revoke(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700299{
Dave Kleikamp470decc2006-10-11 01:20:57 -0700300 journal->j_revoke = NULL;
Duane Griffin83c49522008-04-17 10:38:59 -0400301 if (journal->j_revoke_table[0])
302 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
303 if (journal->j_revoke_table[1])
304 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700305}
306
307
308#ifdef __KERNEL__
309
310/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700311 * jbd2_journal_revoke: revoke a given buffer_head from the journal. This
Dave Kleikamp470decc2006-10-11 01:20:57 -0700312 * prevents the block from being replayed during recovery if we take a
313 * crash after this current transaction commits. Any subsequent
314 * metadata writes of the buffer in this transaction cancel the
315 * revoke.
316 *
317 * Note that this call may block --- it is up to the caller to make
318 * sure that there are no further calls to journal_write_metadata
319 * before the revoke is complete. In ext3, this implies calling the
320 * revoke before clearing the block bitmap when we are deleting
321 * metadata.
322 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700323 * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a
Dave Kleikamp470decc2006-10-11 01:20:57 -0700324 * parameter, but does _not_ forget the buffer_head if the bh was only
325 * found implicitly.
326 *
327 * bh_in may not be a journalled buffer - it may have come off
328 * the hash tables without an attached journal_head.
329 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700330 * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count
Dave Kleikamp470decc2006-10-11 01:20:57 -0700331 * by one.
332 */
333
Mingming Cao18eba7a2006-10-11 01:21:13 -0700334int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700335 struct buffer_head *bh_in)
336{
337 struct buffer_head *bh = NULL;
338 journal_t *journal;
339 struct block_device *bdev;
340 int err;
341
342 might_sleep();
343 if (bh_in)
344 BUFFER_TRACE(bh_in, "enter");
345
346 journal = handle->h_transaction->t_journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700347 if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){
Dave Kleikamp470decc2006-10-11 01:20:57 -0700348 J_ASSERT (!"Cannot set revoke feature!");
349 return -EINVAL;
350 }
351
352 bdev = journal->j_fs_dev;
353 bh = bh_in;
354
355 if (!bh) {
356 bh = __find_get_block(bdev, blocknr, journal->j_blocksize);
357 if (bh)
358 BUFFER_TRACE(bh, "found on hash");
359 }
Mingming Caocd02ff02007-10-16 18:38:25 -0400360#ifdef JBD2_EXPENSIVE_CHECKING
Dave Kleikamp470decc2006-10-11 01:20:57 -0700361 else {
362 struct buffer_head *bh2;
363
364 /* If there is a different buffer_head lying around in
365 * memory anywhere... */
366 bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize);
367 if (bh2) {
368 /* ... and it has RevokeValid status... */
369 if (bh2 != bh && buffer_revokevalid(bh2))
370 /* ...then it better be revoked too,
371 * since it's illegal to create a revoke
372 * record against a buffer_head which is
373 * not marked revoked --- that would
374 * risk missing a subsequent revoke
375 * cancel. */
376 J_ASSERT_BH(bh2, buffer_revoked(bh2));
377 put_bh(bh2);
378 }
379 }
380#endif
381
382 /* We really ought not ever to revoke twice in a row without
383 first having the revoke cancelled: it's illegal to free a
384 block twice without allocating it in between! */
385 if (bh) {
386 if (!J_EXPECT_BH(bh, !buffer_revoked(bh),
387 "inconsistent data on disk")) {
388 if (!bh_in)
389 brelse(bh);
390 return -EIO;
391 }
392 set_buffer_revoked(bh);
393 set_buffer_revokevalid(bh);
394 if (bh_in) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700395 BUFFER_TRACE(bh_in, "call jbd2_journal_forget");
396 jbd2_journal_forget(handle, bh_in);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700397 } else {
398 BUFFER_TRACE(bh, "call brelse");
399 __brelse(bh);
400 }
401 }
402
Mingming Cao29971762006-10-11 01:21:09 -0700403 jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700404 err = insert_revoke_hash(journal, blocknr,
405 handle->h_transaction->t_tid);
406 BUFFER_TRACE(bh_in, "exit");
407 return err;
408}
409
410/*
411 * Cancel an outstanding revoke. For use only internally by the
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700412 * journaling code (called from jbd2_journal_get_write_access).
Dave Kleikamp470decc2006-10-11 01:20:57 -0700413 *
414 * We trust buffer_revoked() on the buffer if the buffer is already
415 * being journaled: if there is no revoke pending on the buffer, then we
416 * don't do anything here.
417 *
418 * This would break if it were possible for a buffer to be revoked and
419 * discarded, and then reallocated within the same transaction. In such
420 * a case we would have lost the revoked bit, but when we arrived here
421 * the second time we would still have a pending revoke to cancel. So,
422 * do not trust the Revoked bit on buffers unless RevokeValid is also
423 * set.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700424 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700425int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700426{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700427 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700428 journal_t *journal = handle->h_transaction->t_journal;
429 int need_cancel;
430 int did_revoke = 0; /* akpm: debug */
431 struct buffer_head *bh = jh2bh(jh);
432
433 jbd_debug(4, "journal_head %p, cancelling revoke\n", jh);
434
435 /* Is the existing Revoke bit valid? If so, we trust it, and
436 * only perform the full cancel if the revoke bit is set. If
437 * not, we can't trust the revoke bit, and we need to do the
438 * full search for a revoke record. */
439 if (test_set_buffer_revokevalid(bh)) {
440 need_cancel = test_clear_buffer_revoked(bh);
441 } else {
442 need_cancel = 1;
443 clear_buffer_revoked(bh);
444 }
445
446 if (need_cancel) {
447 record = find_revoke_record(journal, bh->b_blocknr);
448 if (record) {
449 jbd_debug(4, "cancelled existing revoke on "
450 "blocknr %llu\n", (unsigned long long)bh->b_blocknr);
451 spin_lock(&journal->j_revoke_lock);
452 list_del(&record->hash);
453 spin_unlock(&journal->j_revoke_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700454 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700455 did_revoke = 1;
456 }
457 }
458
Mingming Caocd02ff02007-10-16 18:38:25 -0400459#ifdef JBD2_EXPENSIVE_CHECKING
Dave Kleikamp470decc2006-10-11 01:20:57 -0700460 /* There better not be one left behind by now! */
461 record = find_revoke_record(journal, bh->b_blocknr);
462 J_ASSERT_JH(jh, record == NULL);
463#endif
464
465 /* Finally, have we just cleared revoke on an unhashed
466 * buffer_head? If so, we'd better make sure we clear the
467 * revoked status on any hashed alias too, otherwise the revoke
468 * state machine will get very upset later on. */
469 if (need_cancel) {
470 struct buffer_head *bh2;
471 bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size);
472 if (bh2) {
473 if (bh2 != bh)
474 clear_buffer_revoked(bh2);
475 __brelse(bh2);
476 }
477 }
478 return did_revoke;
479}
480
481/* journal_switch_revoke table select j_revoke for next transaction
482 * we do not want to suspend any processing until all revokes are
483 * written -bzzz
484 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700485void jbd2_journal_switch_revoke_table(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700486{
487 int i;
488
489 if (journal->j_revoke == journal->j_revoke_table[0])
490 journal->j_revoke = journal->j_revoke_table[1];
491 else
492 journal->j_revoke = journal->j_revoke_table[0];
493
494 for (i = 0; i < journal->j_revoke->hash_size; i++)
495 INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]);
496}
497
498/*
499 * Write revoke records to the journal for all entries in the current
500 * revoke hash, deleting the entries as we go.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700501 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700502void jbd2_journal_write_revoke_records(journal_t *journal,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400503 transaction_t *transaction,
504 int write_op)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700505{
506 struct journal_head *descriptor;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700507 struct jbd2_revoke_record_s *record;
508 struct jbd2_revoke_table_s *revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700509 struct list_head *hash_list;
510 int i, offset, count;
511
512 descriptor = NULL;
513 offset = 0;
514 count = 0;
515
516 /* select revoke table for committing transaction */
517 revoke = journal->j_revoke == journal->j_revoke_table[0] ?
518 journal->j_revoke_table[1] : journal->j_revoke_table[0];
519
520 for (i = 0; i < revoke->hash_size; i++) {
521 hash_list = &revoke->hash_table[i];
522
523 while (!list_empty(hash_list)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700524 record = (struct jbd2_revoke_record_s *)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700525 hash_list->next;
526 write_one_revoke_record(journal, transaction,
527 &descriptor, &offset,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400528 record, write_op);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700529 count++;
530 list_del(&record->hash);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700531 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700532 }
533 }
534 if (descriptor)
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400535 flush_descriptor(journal, descriptor, offset, write_op);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700536 jbd_debug(1, "Wrote %d revoke records\n", count);
537}
538
539/*
540 * Write out one revoke record. We need to create a new descriptor
541 * block if the old one is full or if we have not already created one.
542 */
543
544static void write_one_revoke_record(journal_t *journal,
545 transaction_t *transaction,
546 struct journal_head **descriptorp,
547 int *offsetp,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400548 struct jbd2_revoke_record_s *record,
549 int write_op)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700550{
551 struct journal_head *descriptor;
552 int offset;
553 journal_header_t *header;
554
555 /* If we are already aborting, this all becomes a noop. We
556 still need to go round the loop in
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700557 jbd2_journal_write_revoke_records in order to free all of the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700558 revoke records: only the IO to the journal is omitted. */
559 if (is_journal_aborted(journal))
560 return;
561
562 descriptor = *descriptorp;
563 offset = *offsetp;
564
565 /* Make sure we have a descriptor with space left for the record */
566 if (descriptor) {
567 if (offset == journal->j_blocksize) {
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400568 flush_descriptor(journal, descriptor, offset, write_op);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700569 descriptor = NULL;
570 }
571 }
572
573 if (!descriptor) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700574 descriptor = jbd2_journal_get_descriptor_buffer(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700575 if (!descriptor)
576 return;
577 header = (journal_header_t *) &jh2bh(descriptor)->b_data[0];
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700578 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
579 header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700580 header->h_sequence = cpu_to_be32(transaction->t_tid);
581
582 /* Record it so that we can wait for IO completion later */
583 JBUFFER_TRACE(descriptor, "file as BJ_LogCtl");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700584 jbd2_journal_file_buffer(descriptor, transaction, BJ_LogCtl);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700585
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700586 offset = sizeof(jbd2_journal_revoke_header_t);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700587 *descriptorp = descriptor;
588 }
589
Zach Brownb517bea2006-10-11 01:21:08 -0700590 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) {
591 * ((__be64 *)(&jh2bh(descriptor)->b_data[offset])) =
592 cpu_to_be64(record->blocknr);
593 offset += 8;
594
595 } else {
596 * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) =
597 cpu_to_be32(record->blocknr);
598 offset += 4;
599 }
600
Dave Kleikamp470decc2006-10-11 01:20:57 -0700601 *offsetp = offset;
602}
603
604/*
605 * Flush a revoke descriptor out to the journal. If we are aborting,
606 * this is a noop; otherwise we are generating a buffer which needs to
607 * be waited for during commit, so it has to go onto the appropriate
608 * journal buffer list.
609 */
610
611static void flush_descriptor(journal_t *journal,
612 struct journal_head *descriptor,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400613 int offset, int write_op)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700614{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700615 jbd2_journal_revoke_header_t *header;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700616 struct buffer_head *bh = jh2bh(descriptor);
617
618 if (is_journal_aborted(journal)) {
619 put_bh(bh);
620 return;
621 }
622
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700623 header = (jbd2_journal_revoke_header_t *) jh2bh(descriptor)->b_data;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700624 header->r_count = cpu_to_be32(offset);
625 set_buffer_jwrite(bh);
626 BUFFER_TRACE(bh, "write");
627 set_buffer_dirty(bh);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200628 write_dirty_buffer(bh, write_op);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700629}
630#endif
631
632/*
633 * Revoke support for recovery.
634 *
635 * Recovery needs to be able to:
636 *
637 * record all revoke records, including the tid of the latest instance
638 * of each revoke in the journal
639 *
640 * check whether a given block in a given transaction should be replayed
641 * (ie. has not been revoked by a revoke record in that or a subsequent
642 * transaction)
643 *
644 * empty the revoke table after recovery.
645 */
646
647/*
648 * First, setting revoke records. We create a new revoke record for
649 * every block ever revoked in the log as we scan it for recovery, and
650 * we update the existing records if we find multiple revokes for a
651 * single block.
652 */
653
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700654int jbd2_journal_set_revoke(journal_t *journal,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700655 unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700656 tid_t sequence)
657{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700658 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700659
660 record = find_revoke_record(journal, blocknr);
661 if (record) {
662 /* If we have multiple occurrences, only record the
663 * latest sequence number in the hashed record */
664 if (tid_gt(sequence, record->sequence))
665 record->sequence = sequence;
666 return 0;
667 }
668 return insert_revoke_hash(journal, blocknr, sequence);
669}
670
671/*
672 * Test revoke records. For a given block referenced in the log, has
673 * that block been revoked? A revoke record with a given transaction
674 * sequence number revokes all blocks in that transaction and earlier
675 * ones, but later transactions still need replayed.
676 */
677
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700678int jbd2_journal_test_revoke(journal_t *journal,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700679 unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700680 tid_t sequence)
681{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700682 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700683
684 record = find_revoke_record(journal, blocknr);
685 if (!record)
686 return 0;
687 if (tid_gt(sequence, record->sequence))
688 return 0;
689 return 1;
690}
691
692/*
693 * Finally, once recovery is over, we need to clear the revoke table so
694 * that it can be reused by the running filesystem.
695 */
696
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700697void jbd2_journal_clear_revoke(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700698{
699 int i;
700 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700701 struct jbd2_revoke_record_s *record;
702 struct jbd2_revoke_table_s *revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700703
704 revoke = journal->j_revoke;
705
706 for (i = 0; i < revoke->hash_size; i++) {
707 hash_list = &revoke->hash_table[i];
708 while (!list_empty(hash_list)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700709 record = (struct jbd2_revoke_record_s*) hash_list->next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700710 list_del(&record->hash);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700711 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700712 }
713 }
714}