blob: 705ae577882b104d438957e0760a4608d93ceeb2 [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 *
Yongqiang Yang1ba37262011-12-28 17:46:46 -050050 * We cache revoke status of a buffer in the current transaction in b_states
51 * bits. As the name says, revokevalid flag indicates that the cached revoke
52 * status of a buffer is valid and we can rely on the cached status.
53 *
Dave Kleikamp470decc2006-10-11 01:20:57 -070054 * Revoke information on buffers is a tri-state value:
55 *
56 * RevokeValid clear: no cached revoke status, need to look it up
57 * RevokeValid set, Revoked clear:
58 * buffer has not been revoked, and cancel_revoke
59 * need do nothing.
60 * RevokeValid set, Revoked set:
61 * buffer has been revoked.
Jan Kara86db97c2009-03-27 17:20:40 -040062 *
63 * Locking rules:
64 * We keep two hash tables of revoke records. One hashtable belongs to the
65 * running transaction (is pointed to by journal->j_revoke), the other one
66 * belongs to the committing transaction. Accesses to the second hash table
67 * happen only from the kjournald and no other thread touches this table. Also
68 * journal_switch_revoke_table() which switches which hashtable belongs to the
69 * running and which to the committing transaction is called only from
70 * kjournald. Therefore we need no locks when accessing the hashtable belonging
71 * to the committing transaction.
72 *
73 * All users operating on the hash table belonging to the running transaction
74 * have a handle to the transaction. Therefore they are safe from kjournald
75 * switching hash tables under them. For operations on the lists of entries in
76 * the hash table j_revoke_lock is used.
77 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -030078 * Finally, also replay code uses the hash tables but at this moment no one else
Jan Kara86db97c2009-03-27 17:20:40 -040079 * can touch them (filesystem isn't mounted yet) and hence no locking is
80 * needed.
Dave Kleikamp470decc2006-10-11 01:20:57 -070081 */
82
83#ifndef __KERNEL__
84#include "jfs_user.h"
85#else
86#include <linux/time.h>
87#include <linux/fs.h>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070088#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070089#include <linux/errno.h>
90#include <linux/slab.h>
91#include <linux/list.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070092#include <linux/init.h>
Theodore Ts'o67c457a2009-04-14 07:50:56 -040093#include <linux/bio.h>
vignesh babuf4823942007-07-15 23:41:17 -070094#include <linux/log2.h>
Theodore Ts'od48458d2014-10-30 10:53:17 -040095#include <linux/hash.h>
Darrick J. Wongdb9ee222014-08-27 18:40:07 -040096#endif
Dave Kleikamp470decc2006-10-11 01:20:57 -070097
Christoph Lametere18b8902006-12-06 20:33:20 -080098static struct kmem_cache *jbd2_revoke_record_cache;
99static struct kmem_cache *jbd2_revoke_table_cache;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700100
101/* Each revoke record represents one single revoked block. During
102 journal replay, this involves recording the transaction ID of the
103 last transaction to revoke this block. */
104
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700105struct jbd2_revoke_record_s
Dave Kleikamp470decc2006-10-11 01:20:57 -0700106{
107 struct list_head hash;
108 tid_t sequence; /* Used for recovery only */
Mingming Cao18eba7a2006-10-11 01:21:13 -0700109 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700110};
111
112
113/* The revoke table is just a simple hash table of revoke records. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700114struct jbd2_revoke_table_s
Dave Kleikamp470decc2006-10-11 01:20:57 -0700115{
116 /* It is conceivable that we might want a larger hash table
117 * for recovery. Must be a power of two. */
118 int hash_size;
119 int hash_shift;
120 struct list_head *hash_table;
121};
122
123
124#ifdef __KERNEL__
125static void write_one_revoke_record(journal_t *, transaction_t *,
Jan Karae5a120a2013-06-04 12:06:01 -0400126 struct list_head *,
127 struct buffer_head **, int *,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400128 struct jbd2_revoke_record_s *, int);
Jan Karae5a120a2013-06-04 12:06:01 -0400129static void flush_descriptor(journal_t *, struct buffer_head *, int, int);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700130#endif
131
132/* Utility functions to maintain the revoke table */
133
Mingming Cao18eba7a2006-10-11 01:21:13 -0700134static inline int hash(journal_t *journal, unsigned long long block)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700135{
Theodore Ts'od48458d2014-10-30 10:53:17 -0400136 return hash_64(block, journal->j_revoke->hash_shift);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700137}
138
Mingming Cao18eba7a2006-10-11 01:21:13 -0700139static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700140 tid_t seq)
141{
142 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700143 struct jbd2_revoke_record_s *record;
Michal Hocko7b506b12015-06-15 15:45:58 -0400144 gfp_t gfp_mask = GFP_NOFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700145
Michal Hocko7b506b12015-06-15 15:45:58 -0400146 if (journal_oom_retry)
147 gfp_mask |= __GFP_NOFAIL;
148 record = kmem_cache_alloc(jbd2_revoke_record_cache, gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700149 if (!record)
Michal Hocko7b506b12015-06-15 15:45:58 -0400150 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700151
152 record->sequence = seq;
153 record->blocknr = blocknr;
154 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
155 spin_lock(&journal->j_revoke_lock);
156 list_add(&record->hash, hash_list);
157 spin_unlock(&journal->j_revoke_lock);
158 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700159}
160
161/* Find a revoke record in the journal's hash table. */
162
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700163static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700164 unsigned long long blocknr)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700165{
166 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700167 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700168
169 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
170
171 spin_lock(&journal->j_revoke_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700172 record = (struct jbd2_revoke_record_s *) hash_list->next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700173 while (&(record->hash) != hash_list) {
174 if (record->blocknr == blocknr) {
175 spin_unlock(&journal->j_revoke_lock);
176 return record;
177 }
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700178 record = (struct jbd2_revoke_record_s *) record->hash.next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700179 }
180 spin_unlock(&journal->j_revoke_lock);
181 return NULL;
182}
183
Duane Griffin9fa27c82008-04-28 09:40:00 -0400184void jbd2_journal_destroy_revoke_caches(void)
185{
186 if (jbd2_revoke_record_cache) {
187 kmem_cache_destroy(jbd2_revoke_record_cache);
188 jbd2_revoke_record_cache = NULL;
189 }
190 if (jbd2_revoke_table_cache) {
191 kmem_cache_destroy(jbd2_revoke_table_cache);
192 jbd2_revoke_table_cache = NULL;
193 }
194}
195
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700196int __init jbd2_journal_init_revoke_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700197{
Duane Griffin9fa27c82008-04-28 09:40:00 -0400198 J_ASSERT(!jbd2_revoke_record_cache);
199 J_ASSERT(!jbd2_revoke_table_cache);
200
Yongqiang Yang9c0e00e2012-02-20 17:53:03 -0500201 jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s,
202 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY);
Al Viro1076d172008-03-29 03:07:18 +0000203 if (!jbd2_revoke_record_cache)
Duane Griffin9fa27c82008-04-28 09:40:00 -0400204 goto record_cache_failure;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700205
Yongqiang Yang9c0e00e2012-02-20 17:53:03 -0500206 jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s,
207 SLAB_TEMPORARY);
Duane Griffin9fa27c82008-04-28 09:40:00 -0400208 if (!jbd2_revoke_table_cache)
209 goto table_cache_failure;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700210 return 0;
Duane Griffin9fa27c82008-04-28 09:40:00 -0400211table_cache_failure:
212 jbd2_journal_destroy_revoke_caches();
213record_cache_failure:
214 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700215}
216
Duane Griffin83c49522008-04-17 10:38:59 -0400217static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700218{
Duane Griffin83c49522008-04-17 10:38:59 -0400219 int shift = 0;
220 int tmp = hash_size;
221 struct jbd2_revoke_table_s *table;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700222
Duane Griffin83c49522008-04-17 10:38:59 -0400223 table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
224 if (!table)
225 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700226
Dave Kleikamp470decc2006-10-11 01:20:57 -0700227 while((tmp >>= 1UL) != 0UL)
228 shift++;
229
Duane Griffin83c49522008-04-17 10:38:59 -0400230 table->hash_size = hash_size;
231 table->hash_shift = shift;
232 table->hash_table =
Dave Kleikamp470decc2006-10-11 01:20:57 -0700233 kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
Duane Griffin83c49522008-04-17 10:38:59 -0400234 if (!table->hash_table) {
235 kmem_cache_free(jbd2_revoke_table_cache, table);
236 table = NULL;
237 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700238 }
239
240 for (tmp = 0; tmp < hash_size; tmp++)
Duane Griffin83c49522008-04-17 10:38:59 -0400241 INIT_LIST_HEAD(&table->hash_table[tmp]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700242
Duane Griffin83c49522008-04-17 10:38:59 -0400243out:
244 return table;
245}
246
247static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table)
248{
249 int i;
250 struct list_head *hash_list;
251
252 for (i = 0; i < table->hash_size; i++) {
253 hash_list = &table->hash_table[i];
254 J_ASSERT(list_empty(hash_list));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700255 }
256
Duane Griffin83c49522008-04-17 10:38:59 -0400257 kfree(table->hash_table);
258 kmem_cache_free(jbd2_revoke_table_cache, table);
259}
260
261/* Initialise the revoke table for a given journal to a given size. */
262int jbd2_journal_init_revoke(journal_t *journal, int hash_size)
263{
264 J_ASSERT(journal->j_revoke_table[0] == NULL);
265 J_ASSERT(is_power_of_2(hash_size));
266
267 journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size);
268 if (!journal->j_revoke_table[0])
269 goto fail0;
270
271 journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size);
272 if (!journal->j_revoke_table[1])
273 goto fail1;
274
Dave Kleikamp470decc2006-10-11 01:20:57 -0700275 journal->j_revoke = journal->j_revoke_table[1];
276
Dave Kleikamp470decc2006-10-11 01:20:57 -0700277 spin_lock_init(&journal->j_revoke_lock);
278
279 return 0;
Duane Griffin83c49522008-04-17 10:38:59 -0400280
281fail1:
282 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
283fail0:
284 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700285}
286
Duane Griffin83c49522008-04-17 10:38:59 -0400287/* Destroy a journal's revoke table. The table must already be empty! */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700288void jbd2_journal_destroy_revoke(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700289{
Dave Kleikamp470decc2006-10-11 01:20:57 -0700290 journal->j_revoke = NULL;
Duane Griffin83c49522008-04-17 10:38:59 -0400291 if (journal->j_revoke_table[0])
292 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
293 if (journal->j_revoke_table[1])
294 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700295}
296
297
298#ifdef __KERNEL__
299
300/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700301 * jbd2_journal_revoke: revoke a given buffer_head from the journal. This
Dave Kleikamp470decc2006-10-11 01:20:57 -0700302 * prevents the block from being replayed during recovery if we take a
303 * crash after this current transaction commits. Any subsequent
304 * metadata writes of the buffer in this transaction cancel the
305 * revoke.
306 *
307 * Note that this call may block --- it is up to the caller to make
308 * sure that there are no further calls to journal_write_metadata
309 * before the revoke is complete. In ext3, this implies calling the
310 * revoke before clearing the block bitmap when we are deleting
311 * metadata.
312 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700313 * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a
Dave Kleikamp470decc2006-10-11 01:20:57 -0700314 * parameter, but does _not_ forget the buffer_head if the bh was only
315 * found implicitly.
316 *
317 * bh_in may not be a journalled buffer - it may have come off
318 * the hash tables without an attached journal_head.
319 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700320 * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count
Dave Kleikamp470decc2006-10-11 01:20:57 -0700321 * by one.
322 */
323
Mingming Cao18eba7a2006-10-11 01:21:13 -0700324int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700325 struct buffer_head *bh_in)
326{
327 struct buffer_head *bh = NULL;
328 journal_t *journal;
329 struct block_device *bdev;
330 int err;
331
332 might_sleep();
333 if (bh_in)
334 BUFFER_TRACE(bh_in, "enter");
335
336 journal = handle->h_transaction->t_journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700337 if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){
Dave Kleikamp470decc2006-10-11 01:20:57 -0700338 J_ASSERT (!"Cannot set revoke feature!");
339 return -EINVAL;
340 }
341
342 bdev = journal->j_fs_dev;
343 bh = bh_in;
344
345 if (!bh) {
346 bh = __find_get_block(bdev, blocknr, journal->j_blocksize);
347 if (bh)
348 BUFFER_TRACE(bh, "found on hash");
349 }
Mingming Caocd02ff02007-10-16 18:38:25 -0400350#ifdef JBD2_EXPENSIVE_CHECKING
Dave Kleikamp470decc2006-10-11 01:20:57 -0700351 else {
352 struct buffer_head *bh2;
353
354 /* If there is a different buffer_head lying around in
355 * memory anywhere... */
356 bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize);
357 if (bh2) {
358 /* ... and it has RevokeValid status... */
359 if (bh2 != bh && buffer_revokevalid(bh2))
360 /* ...then it better be revoked too,
361 * since it's illegal to create a revoke
362 * record against a buffer_head which is
363 * not marked revoked --- that would
364 * risk missing a subsequent revoke
365 * cancel. */
366 J_ASSERT_BH(bh2, buffer_revoked(bh2));
367 put_bh(bh2);
368 }
369 }
370#endif
371
372 /* We really ought not ever to revoke twice in a row without
373 first having the revoke cancelled: it's illegal to free a
374 block twice without allocating it in between! */
375 if (bh) {
376 if (!J_EXPECT_BH(bh, !buffer_revoked(bh),
377 "inconsistent data on disk")) {
378 if (!bh_in)
379 brelse(bh);
380 return -EIO;
381 }
382 set_buffer_revoked(bh);
383 set_buffer_revokevalid(bh);
384 if (bh_in) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700385 BUFFER_TRACE(bh_in, "call jbd2_journal_forget");
386 jbd2_journal_forget(handle, bh_in);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700387 } else {
388 BUFFER_TRACE(bh, "call brelse");
389 __brelse(bh);
390 }
391 }
392
Mingming Cao29971762006-10-11 01:21:09 -0700393 jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700394 err = insert_revoke_hash(journal, blocknr,
395 handle->h_transaction->t_tid);
396 BUFFER_TRACE(bh_in, "exit");
397 return err;
398}
399
400/*
401 * Cancel an outstanding revoke. For use only internally by the
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700402 * journaling code (called from jbd2_journal_get_write_access).
Dave Kleikamp470decc2006-10-11 01:20:57 -0700403 *
404 * We trust buffer_revoked() on the buffer if the buffer is already
405 * being journaled: if there is no revoke pending on the buffer, then we
406 * don't do anything here.
407 *
408 * This would break if it were possible for a buffer to be revoked and
409 * discarded, and then reallocated within the same transaction. In such
410 * a case we would have lost the revoked bit, but when we arrived here
411 * the second time we would still have a pending revoke to cancel. So,
412 * do not trust the Revoked bit on buffers unless RevokeValid is also
413 * set.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700414 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700415int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700416{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700417 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700418 journal_t *journal = handle->h_transaction->t_journal;
419 int need_cancel;
420 int did_revoke = 0; /* akpm: debug */
421 struct buffer_head *bh = jh2bh(jh);
422
423 jbd_debug(4, "journal_head %p, cancelling revoke\n", jh);
424
425 /* Is the existing Revoke bit valid? If so, we trust it, and
426 * only perform the full cancel if the revoke bit is set. If
427 * not, we can't trust the revoke bit, and we need to do the
428 * full search for a revoke record. */
429 if (test_set_buffer_revokevalid(bh)) {
430 need_cancel = test_clear_buffer_revoked(bh);
431 } else {
432 need_cancel = 1;
433 clear_buffer_revoked(bh);
434 }
435
436 if (need_cancel) {
437 record = find_revoke_record(journal, bh->b_blocknr);
438 if (record) {
439 jbd_debug(4, "cancelled existing revoke on "
440 "blocknr %llu\n", (unsigned long long)bh->b_blocknr);
441 spin_lock(&journal->j_revoke_lock);
442 list_del(&record->hash);
443 spin_unlock(&journal->j_revoke_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700444 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700445 did_revoke = 1;
446 }
447 }
448
Mingming Caocd02ff02007-10-16 18:38:25 -0400449#ifdef JBD2_EXPENSIVE_CHECKING
Dave Kleikamp470decc2006-10-11 01:20:57 -0700450 /* There better not be one left behind by now! */
451 record = find_revoke_record(journal, bh->b_blocknr);
452 J_ASSERT_JH(jh, record == NULL);
453#endif
454
455 /* Finally, have we just cleared revoke on an unhashed
456 * buffer_head? If so, we'd better make sure we clear the
457 * revoked status on any hashed alias too, otherwise the revoke
458 * state machine will get very upset later on. */
459 if (need_cancel) {
460 struct buffer_head *bh2;
461 bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size);
462 if (bh2) {
463 if (bh2 != bh)
464 clear_buffer_revoked(bh2);
465 __brelse(bh2);
466 }
467 }
468 return did_revoke;
469}
470
Yongqiang Yang1ba37262011-12-28 17:46:46 -0500471/*
472 * journal_clear_revoked_flag clears revoked flag of buffers in
473 * revoke table to reflect there is no revoked buffers in the next
474 * transaction which is going to be started.
475 */
476void jbd2_clear_buffer_revoked_flags(journal_t *journal)
477{
478 struct jbd2_revoke_table_s *revoke = journal->j_revoke;
479 int i = 0;
480
481 for (i = 0; i < revoke->hash_size; i++) {
482 struct list_head *hash_list;
483 struct list_head *list_entry;
484 hash_list = &revoke->hash_table[i];
485
486 list_for_each(list_entry, hash_list) {
487 struct jbd2_revoke_record_s *record;
488 struct buffer_head *bh;
489 record = (struct jbd2_revoke_record_s *)list_entry;
490 bh = __find_get_block(journal->j_fs_dev,
491 record->blocknr,
492 journal->j_blocksize);
493 if (bh) {
494 clear_buffer_revoked(bh);
495 __brelse(bh);
496 }
497 }
498 }
499}
500
Dave Kleikamp470decc2006-10-11 01:20:57 -0700501/* journal_switch_revoke table select j_revoke for next transaction
502 * we do not want to suspend any processing until all revokes are
503 * written -bzzz
504 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700505void jbd2_journal_switch_revoke_table(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700506{
507 int i;
508
509 if (journal->j_revoke == journal->j_revoke_table[0])
510 journal->j_revoke = journal->j_revoke_table[1];
511 else
512 journal->j_revoke = journal->j_revoke_table[0];
513
514 for (i = 0; i < journal->j_revoke->hash_size; i++)
515 INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]);
516}
517
518/*
519 * Write revoke records to the journal for all entries in the current
520 * revoke hash, deleting the entries as we go.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700521 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700522void jbd2_journal_write_revoke_records(journal_t *journal,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400523 transaction_t *transaction,
Jan Karae5a120a2013-06-04 12:06:01 -0400524 struct list_head *log_bufs,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400525 int write_op)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700526{
Jan Karae5a120a2013-06-04 12:06:01 -0400527 struct buffer_head *descriptor;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700528 struct jbd2_revoke_record_s *record;
529 struct jbd2_revoke_table_s *revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700530 struct list_head *hash_list;
531 int i, offset, count;
532
533 descriptor = NULL;
534 offset = 0;
535 count = 0;
536
537 /* select revoke table for committing transaction */
538 revoke = journal->j_revoke == journal->j_revoke_table[0] ?
539 journal->j_revoke_table[1] : journal->j_revoke_table[0];
540
541 for (i = 0; i < revoke->hash_size; i++) {
542 hash_list = &revoke->hash_table[i];
543
544 while (!list_empty(hash_list)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700545 record = (struct jbd2_revoke_record_s *)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700546 hash_list->next;
Jan Karae5a120a2013-06-04 12:06:01 -0400547 write_one_revoke_record(journal, transaction, log_bufs,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700548 &descriptor, &offset,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400549 record, write_op);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700550 count++;
551 list_del(&record->hash);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700552 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700553 }
554 }
555 if (descriptor)
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400556 flush_descriptor(journal, descriptor, offset, write_op);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700557 jbd_debug(1, "Wrote %d revoke records\n", count);
558}
559
560/*
561 * Write out one revoke record. We need to create a new descriptor
562 * block if the old one is full or if we have not already created one.
563 */
564
565static void write_one_revoke_record(journal_t *journal,
566 transaction_t *transaction,
Jan Karae5a120a2013-06-04 12:06:01 -0400567 struct list_head *log_bufs,
568 struct buffer_head **descriptorp,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700569 int *offsetp,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400570 struct jbd2_revoke_record_s *record,
571 int write_op)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700572{
Darrick J. Wong42a71062012-05-27 08:08:24 -0400573 int csum_size = 0;
Jan Karae5a120a2013-06-04 12:06:01 -0400574 struct buffer_head *descriptor;
Darrick J. Wonge531d0b2015-05-14 19:11:50 -0400575 int sz, offset;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700576 journal_header_t *header;
577
578 /* If we are already aborting, this all becomes a noop. We
579 still need to go round the loop in
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700580 jbd2_journal_write_revoke_records in order to free all of the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700581 revoke records: only the IO to the journal is omitted. */
582 if (is_journal_aborted(journal))
583 return;
584
585 descriptor = *descriptorp;
586 offset = *offsetp;
587
Darrick J. Wong42a71062012-05-27 08:08:24 -0400588 /* Do we need to leave space at the end for a checksum? */
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400589 if (jbd2_journal_has_csum_v2or3(journal))
Darrick J. Wong42a71062012-05-27 08:08:24 -0400590 csum_size = sizeof(struct jbd2_journal_revoke_tail);
591
Darrick J. Wong56316a02015-10-17 16:18:45 -0400592 if (jbd2_has_feature_64bit(journal))
Darrick J. Wonge531d0b2015-05-14 19:11:50 -0400593 sz = 8;
594 else
595 sz = 4;
596
Dave Kleikamp470decc2006-10-11 01:20:57 -0700597 /* Make sure we have a descriptor with space left for the record */
598 if (descriptor) {
Darrick J. Wonge531d0b2015-05-14 19:11:50 -0400599 if (offset + sz > journal->j_blocksize - csum_size) {
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400600 flush_descriptor(journal, descriptor, offset, write_op);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700601 descriptor = NULL;
602 }
603 }
604
605 if (!descriptor) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700606 descriptor = jbd2_journal_get_descriptor_buffer(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700607 if (!descriptor)
608 return;
Jan Karae5a120a2013-06-04 12:06:01 -0400609 header = (journal_header_t *)descriptor->b_data;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700610 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
611 header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700612 header->h_sequence = cpu_to_be32(transaction->t_tid);
613
614 /* Record it so that we can wait for IO completion later */
Jan Karae5a120a2013-06-04 12:06:01 -0400615 BUFFER_TRACE(descriptor, "file in log_bufs");
616 jbd2_file_log_bh(log_bufs, descriptor);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700617
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700618 offset = sizeof(jbd2_journal_revoke_header_t);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700619 *descriptorp = descriptor;
620 }
621
Darrick J. Wong56316a02015-10-17 16:18:45 -0400622 if (jbd2_has_feature_64bit(journal))
Jan Karae5a120a2013-06-04 12:06:01 -0400623 * ((__be64 *)(&descriptor->b_data[offset])) =
Zach Brownb517bea2006-10-11 01:21:08 -0700624 cpu_to_be64(record->blocknr);
Darrick J. Wonge531d0b2015-05-14 19:11:50 -0400625 else
Jan Karae5a120a2013-06-04 12:06:01 -0400626 * ((__be32 *)(&descriptor->b_data[offset])) =
Zach Brownb517bea2006-10-11 01:21:08 -0700627 cpu_to_be32(record->blocknr);
Darrick J. Wonge531d0b2015-05-14 19:11:50 -0400628 offset += sz;
Zach Brownb517bea2006-10-11 01:21:08 -0700629
Dave Kleikamp470decc2006-10-11 01:20:57 -0700630 *offsetp = offset;
631}
632
Jan Karae5a120a2013-06-04 12:06:01 -0400633static void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh)
Darrick J. Wong42a71062012-05-27 08:08:24 -0400634{
635 struct jbd2_journal_revoke_tail *tail;
636 __u32 csum;
637
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400638 if (!jbd2_journal_has_csum_v2or3(j))
Darrick J. Wong42a71062012-05-27 08:08:24 -0400639 return;
640
Jan Karae5a120a2013-06-04 12:06:01 -0400641 tail = (struct jbd2_journal_revoke_tail *)(bh->b_data + j->j_blocksize -
Darrick J. Wong42a71062012-05-27 08:08:24 -0400642 sizeof(struct jbd2_journal_revoke_tail));
643 tail->r_checksum = 0;
Jan Karae5a120a2013-06-04 12:06:01 -0400644 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
Darrick J. Wong42a71062012-05-27 08:08:24 -0400645 tail->r_checksum = cpu_to_be32(csum);
646}
647
Dave Kleikamp470decc2006-10-11 01:20:57 -0700648/*
649 * Flush a revoke descriptor out to the journal. If we are aborting,
650 * this is a noop; otherwise we are generating a buffer which needs to
651 * be waited for during commit, so it has to go onto the appropriate
652 * journal buffer list.
653 */
654
655static void flush_descriptor(journal_t *journal,
Jan Karae5a120a2013-06-04 12:06:01 -0400656 struct buffer_head *descriptor,
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400657 int offset, int write_op)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700658{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700659 jbd2_journal_revoke_header_t *header;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700660
661 if (is_journal_aborted(journal)) {
Jan Karae5a120a2013-06-04 12:06:01 -0400662 put_bh(descriptor);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700663 return;
664 }
665
Jan Karae5a120a2013-06-04 12:06:01 -0400666 header = (jbd2_journal_revoke_header_t *)descriptor->b_data;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700667 header->r_count = cpu_to_be32(offset);
Darrick J. Wong42a71062012-05-27 08:08:24 -0400668 jbd2_revoke_csum_set(journal, descriptor);
669
Jan Karae5a120a2013-06-04 12:06:01 -0400670 set_buffer_jwrite(descriptor);
671 BUFFER_TRACE(descriptor, "write");
672 set_buffer_dirty(descriptor);
673 write_dirty_buffer(descriptor, write_op);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700674}
675#endif
676
677/*
678 * Revoke support for recovery.
679 *
680 * Recovery needs to be able to:
681 *
682 * record all revoke records, including the tid of the latest instance
683 * of each revoke in the journal
684 *
685 * check whether a given block in a given transaction should be replayed
686 * (ie. has not been revoked by a revoke record in that or a subsequent
687 * transaction)
688 *
689 * empty the revoke table after recovery.
690 */
691
692/*
693 * First, setting revoke records. We create a new revoke record for
694 * every block ever revoked in the log as we scan it for recovery, and
695 * we update the existing records if we find multiple revokes for a
696 * single block.
697 */
698
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700699int jbd2_journal_set_revoke(journal_t *journal,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700700 unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700701 tid_t sequence)
702{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700703 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700704
705 record = find_revoke_record(journal, blocknr);
706 if (record) {
707 /* If we have multiple occurrences, only record the
708 * latest sequence number in the hashed record */
709 if (tid_gt(sequence, record->sequence))
710 record->sequence = sequence;
711 return 0;
712 }
713 return insert_revoke_hash(journal, blocknr, sequence);
714}
715
716/*
717 * Test revoke records. For a given block referenced in the log, has
718 * that block been revoked? A revoke record with a given transaction
719 * sequence number revokes all blocks in that transaction and earlier
720 * ones, but later transactions still need replayed.
721 */
722
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700723int jbd2_journal_test_revoke(journal_t *journal,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700724 unsigned long long blocknr,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700725 tid_t sequence)
726{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700727 struct jbd2_revoke_record_s *record;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700728
729 record = find_revoke_record(journal, blocknr);
730 if (!record)
731 return 0;
732 if (tid_gt(sequence, record->sequence))
733 return 0;
734 return 1;
735}
736
737/*
738 * Finally, once recovery is over, we need to clear the revoke table so
739 * that it can be reused by the running filesystem.
740 */
741
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700742void jbd2_journal_clear_revoke(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700743{
744 int i;
745 struct list_head *hash_list;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700746 struct jbd2_revoke_record_s *record;
747 struct jbd2_revoke_table_s *revoke;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700748
749 revoke = journal->j_revoke;
750
751 for (i = 0; i < revoke->hash_size; i++) {
752 hash_list = &revoke->hash_table[i];
753 while (!list_empty(hash_list)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700754 record = (struct jbd2_revoke_record_s*) hash_list->next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700755 list_del(&record->hash);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700756 kmem_cache_free(jbd2_revoke_record_cache, record);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700757 }
758 }
759}