]> git.ipfire.org Git - thirdparty/e2fsprogs.git/blame - e2fsck/revoke.c
e2fsck: if user declines to fix s_inodes_acount, abort
[thirdparty/e2fsprogs.git] / e2fsck / revoke.c
CommitLineData
0e8a9560 1/*
97f168b6 2 * linux/fs/jbd2/revoke.c
efc6f628 3 *
0e8a9560
TT
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:
efc6f628 18 *
0e8a9560
TT
19 * + Commit: during commit we write the entire list of the current
20 * transaction's revoked blocks to the journal
efc6f628 21 *
0e8a9560
TT
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:
efc6f628 32 * The desired end result is the journaling of the new block, so we
0e8a9560
TT
33 * cancel the revoke before the transaction commits.
34 *
35 * Block is journaled and then revoked:
725c474f
TT
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
0e8a9560 38 * later in the log than the log block. In this case, we choose the
725c474f
TT
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.
0e8a9560 43 *
efc6f628 44 * Block is revoked and then written as data:
0e8a9560
TT
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 *
97f168b6
DW
50 * 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 *
0e8a9560
TT
54 * Revoke information on buffers is a tri-state value:
55 *
56 * RevokeValid clear: no cached revoke status, need to look it up
8cf93332 57 * RevokeValid set, Revoked clear:
0e8a9560
TT
58 * buffer has not been revoked, and cancel_revoke
59 * need do nothing.
8cf93332 60 * RevokeValid set, Revoked set:
efc6f628 61 * buffer has been revoked.
97f168b6
DW
62 *
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 *
78 * Finally, also replay code uses the hash tables but at this moment no one else
79 * can touch them (filesystem isn't mounted yet) and hence no locking is
80 * needed.
0e8a9560
TT
81 */
82
83#ifndef __KERNEL__
84#include "jfs_user.h"
85#else
97f168b6 86#include <linux/time.h>
0e8a9560 87#include <linux/fs.h>
97f168b6 88#include <linux/jbd2.h>
0e8a9560
TT
89#include <linux/errno.h>
90#include <linux/slab.h>
0e8a9560 91#include <linux/list.h>
8cf93332 92#include <linux/init.h>
97f168b6
DW
93#include <linux/bio.h>
94#include <linux/log2.h>
0e8a9560
TT
95#endif
96
97f168b6
DW
97static lkmem_cache_t *jbd2_revoke_record_cache;
98static lkmem_cache_t *jbd2_revoke_table_cache;
0e8a9560
TT
99
100/* Each revoke record represents one single revoked block. During
101 journal replay, this involves recording the transaction ID of the
102 last transaction to revoke this block. */
103
97f168b6 104struct jbd2_revoke_record_s
0e8a9560
TT
105{
106 struct list_head hash;
107 tid_t sequence; /* Used for recovery only */
97f168b6 108 unsigned long long blocknr;
0e8a9560
TT
109};
110
111
112/* The revoke table is just a simple hash table of revoke records. */
97f168b6 113struct jbd2_revoke_table_s
0e8a9560
TT
114{
115 /* It is conceivable that we might want a larger hash table
116 * for recovery. Must be a power of two. */
efc6f628
TT
117 int hash_size;
118 int hash_shift;
0e8a9560
TT
119 struct list_head *hash_table;
120};
121
122
123#ifdef __KERNEL__
124static void write_one_revoke_record(journal_t *, transaction_t *,
97f168b6
DW
125 struct list_head *,
126 struct buffer_head **, int *,
127 struct jbd2_revoke_record_s *, int);
128static void flush_descriptor(journal_t *, struct buffer_head *, int, int);
0e8a9560
TT
129#endif
130
131/* Utility functions to maintain the revoke table */
132
133/* Borrowed from buffer.c: this is a tried and tested block hash function */
97f168b6 134static inline int hash(journal_t *journal, unsigned long long block)
0e8a9560 135{
97f168b6 136 struct jbd2_revoke_table_s *table = journal->j_revoke;
efc6f628 137
c8ca2397 138 return (hash_64(block, table->hash_shift));
0e8a9560
TT
139}
140
97f168b6 141static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr,
3e699064 142 tid_t seq)
0e8a9560
TT
143{
144 struct list_head *hash_list;
97f168b6 145 struct jbd2_revoke_record_s *record;
8cf93332
TT
146
147repeat:
97f168b6 148 record = kmem_cache_alloc(jbd2_revoke_record_cache, GFP_NOFS);
0e8a9560 149 if (!record)
8cf93332 150 goto oom;
0e8a9560
TT
151
152 record->sequence = seq;
153 record->blocknr = blocknr;
154 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
97f168b6 155 spin_lock(&journal->j_revoke_lock);
0e8a9560 156 list_add(&record->hash, hash_list);
97f168b6 157 spin_unlock(&journal->j_revoke_lock);
0e8a9560 158 return 0;
8cf93332
TT
159
160oom:
8cf93332
TT
161 if (!journal_oom_retry)
162 return -ENOMEM;
97f168b6
DW
163 jbd_debug(1, "ENOMEM in %s, retrying\n", __func__);
164 yield();
8cf93332 165 goto repeat;
0e8a9560
TT
166}
167
168/* Find a revoke record in the journal's hash table. */
169
97f168b6
DW
170static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal,
171 unsigned long long blocknr)
0e8a9560
TT
172{
173 struct list_head *hash_list;
97f168b6 174 struct jbd2_revoke_record_s *record;
efc6f628 175
0e8a9560
TT
176 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
177
97f168b6
DW
178 spin_lock(&journal->j_revoke_lock);
179 record = (struct jbd2_revoke_record_s *) hash_list->next;
0e8a9560 180 while (&(record->hash) != hash_list) {
97f168b6
DW
181 if (record->blocknr == blocknr) {
182 spin_unlock(&journal->j_revoke_lock);
0e8a9560 183 return record;
97f168b6
DW
184 }
185 record = (struct jbd2_revoke_record_s *) record->hash.next;
0e8a9560 186 }
97f168b6 187 spin_unlock(&journal->j_revoke_lock);
0e8a9560
TT
188 return NULL;
189}
190
97f168b6 191void journal_destroy_revoke_caches(void)
8cf93332 192{
97f168b6
DW
193 if (jbd2_revoke_record_cache) {
194 kmem_cache_destroy(jbd2_revoke_record_cache);
195 jbd2_revoke_record_cache = NULL;
196 }
197 if (jbd2_revoke_table_cache) {
198 kmem_cache_destroy(jbd2_revoke_table_cache);
199 jbd2_revoke_table_cache = NULL;
8cf93332 200 }
efc6f628 201}
8cf93332 202
97f168b6 203int __init journal_init_revoke_caches(void)
8cf93332 204{
97f168b6
DW
205 J_ASSERT(!jbd2_revoke_record_cache);
206 J_ASSERT(!jbd2_revoke_table_cache);
207
208 jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s,
209 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY);
210 if (!jbd2_revoke_record_cache)
211 goto record_cache_failure;
212
213 jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s,
214 SLAB_TEMPORARY);
215 if (!jbd2_revoke_table_cache)
216 goto table_cache_failure;
217 return 0;
218table_cache_failure:
219 journal_destroy_revoke_caches();
220record_cache_failure:
221 return -ENOMEM;
8cf93332 222}
0e8a9560 223
97f168b6 224static struct jbd2_revoke_table_s *journal_init_revoke_table(int hash_size)
0e8a9560 225{
97f168b6
DW
226 int shift = 0;
227 int tmp = hash_size;
228 struct jbd2_revoke_table_s *table;
efc6f628 229
97f168b6
DW
230 table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
231 if (!table)
232 goto out;
0e8a9560 233
0e8a9560
TT
234 while((tmp >>= 1UL) != 0UL)
235 shift++;
0e8a9560 236
97f168b6
DW
237 table->hash_size = hash_size;
238 table->hash_shift = shift;
239 table->hash_table =
0e8a9560 240 kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
97f168b6
DW
241 if (!table->hash_table) {
242 kmem_cache_free(jbd2_revoke_table_cache, table);
243 table = NULL;
244 goto out;
0e8a9560 245 }
efc6f628 246
0e8a9560 247 for (tmp = 0; tmp < hash_size; tmp++)
97f168b6 248 INIT_LIST_HEAD(&table->hash_table[tmp]);
efc6f628 249
97f168b6
DW
250out:
251 return table;
0e8a9560
TT
252}
253
97f168b6 254static void journal_destroy_revoke_table(struct jbd2_revoke_table_s *table)
0e8a9560 255{
0e8a9560 256 int i;
97f168b6 257 struct list_head *hash_list;
efc6f628 258
97f168b6 259 for (i = 0; i < table->hash_size; i++) {
0e8a9560 260 hash_list = &table->hash_table[i];
97f168b6 261 J_ASSERT(list_empty(hash_list));
0e8a9560 262 }
efc6f628 263
0e8a9560 264 kfree(table->hash_table);
97f168b6
DW
265 kmem_cache_free(jbd2_revoke_table_cache, table);
266}
267
268/* Initialise the revoke table for a given journal to a given size. */
269int journal_init_revoke(journal_t *journal, int hash_size)
270{
271 J_ASSERT(journal->j_revoke_table[0] == NULL);
272 J_ASSERT(is_power_of_2(hash_size));
273
274 journal->j_revoke_table[0] = journal_init_revoke_table(hash_size);
275 if (!journal->j_revoke_table[0])
276 goto fail0;
277
278 journal->j_revoke_table[1] = journal_init_revoke_table(hash_size);
279 if (!journal->j_revoke_table[1])
280 goto fail1;
281
282 journal->j_revoke = journal->j_revoke_table[1];
283
284 spin_lock_init(&journal->j_revoke_lock);
285
286 return 0;
287
288fail1:
289 journal_destroy_revoke_table(journal->j_revoke_table[0]);
290fail0:
291 return -ENOMEM;
292}
293
294/* Destroy a journal's revoke table. The table must already be empty! */
295void journal_destroy_revoke(journal_t *journal)
296{
0e8a9560 297 journal->j_revoke = NULL;
97f168b6
DW
298 if (journal->j_revoke_table[0])
299 journal_destroy_revoke_table(journal->j_revoke_table[0]);
300 if (journal->j_revoke_table[1])
301 journal_destroy_revoke_table(journal->j_revoke_table[1]);
0e8a9560
TT
302}
303
304
305#ifdef __KERNEL__
306
efc6f628 307/*
0e8a9560
TT
308 * journal_revoke: revoke a given buffer_head from the journal. This
309 * prevents the block from being replayed during recovery if we take a
310 * crash after this current transaction commits. Any subsequent
311 * metadata writes of the buffer in this transaction cancel the
efc6f628 312 * revoke.
0e8a9560
TT
313 *
314 * Note that this call may block --- it is up to the caller to make
315 * sure that there are no further calls to journal_write_metadata
316 * before the revoke is complete. In ext3, this implies calling the
317 * revoke before clearing the block bitmap when we are deleting
efc6f628 318 * metadata.
0e8a9560
TT
319 *
320 * Revoke performs a journal_forget on any buffer_head passed in as a
321 * parameter, but does _not_ forget the buffer_head if the bh was only
efc6f628 322 * found implicitly.
0e8a9560 323 *
8cf93332
TT
324 * bh_in may not be a journalled buffer - it may have come off
325 * the hash tables without an attached journal_head.
326 *
327 * If bh_in is non-zero, journal_revoke() will decrement its b_count
328 * by one.
0e8a9560
TT
329 */
330
97f168b6 331int journal_revoke(handle_t *handle, unsigned long long blocknr,
0e8a9560
TT
332 struct buffer_head *bh_in)
333{
8cf93332 334 struct buffer_head *bh = NULL;
0e8a9560 335 journal_t *journal;
97f168b6 336 struct block_device *bdev;
0e8a9560
TT
337 int err;
338
97f168b6 339 might_sleep();
8cf93332
TT
340 if (bh_in)
341 BUFFER_TRACE(bh_in, "enter");
342
0e8a9560 343 journal = handle->h_transaction->t_journal;
1f735038
TT
344 if (!journal_set_features(journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)){
345 J_ASSERT (!"Cannot set revoke feature!");
0e8a9560 346 return -EINVAL;
1f735038 347 }
8cf93332 348
97f168b6 349 bdev = journal->j_fs_dev;
0e8a9560
TT
350 bh = bh_in;
351
8cf93332 352 if (!bh) {
97f168b6 353 bh = __find_get_block(bdev, blocknr, journal->j_blocksize);
8cf93332
TT
354 if (bh)
355 BUFFER_TRACE(bh, "found on hash");
356 }
97f168b6 357#ifdef JFS_EXPENSIVE_CHECKING
8cf93332
TT
358 else {
359 struct buffer_head *bh2;
360
361 /* If there is a different buffer_head lying around in
362 * memory anywhere... */
97f168b6 363 bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize);
8cf93332
TT
364 if (bh2) {
365 /* ... and it has RevokeValid status... */
97f168b6 366 if (bh2 != bh && buffer_revokevalid(bh2))
8cf93332
TT
367 /* ...then it better be revoked too,
368 * since it's illegal to create a revoke
369 * record against a buffer_head which is
370 * not marked revoked --- that would
371 * risk missing a subsequent revoke
372 * cancel. */
97f168b6
DW
373 J_ASSERT_BH(bh2, buffer_revoked(bh2));
374 put_bh(bh2);
8cf93332
TT
375 }
376 }
377#endif
0e8a9560
TT
378
379 /* We really ought not ever to revoke twice in a row without
380 first having the revoke cancelled: it's illegal to free a
381 block twice without allocating it in between! */
382 if (bh) {
97f168b6
DW
383 if (!J_EXPECT_BH(bh, !buffer_revoked(bh),
384 "inconsistent data on disk")) {
385 if (!bh_in)
386 brelse(bh);
387 return -EIO;
388 }
389 set_buffer_revoked(bh);
390 set_buffer_revokevalid(bh);
8cf93332
TT
391 if (bh_in) {
392 BUFFER_TRACE(bh_in, "call journal_forget");
0e8a9560 393 journal_forget(handle, bh_in);
8cf93332
TT
394 } else {
395 BUFFER_TRACE(bh, "call brelse");
396 __brelse(bh);
397 }
0e8a9560
TT
398 }
399
97f168b6 400 jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in);
8cf93332
TT
401 err = insert_revoke_hash(journal, blocknr,
402 handle->h_transaction->t_tid);
8cf93332 403 BUFFER_TRACE(bh_in, "exit");
0e8a9560
TT
404 return err;
405}
406
0e8a9560
TT
407/*
408 * Cancel an outstanding revoke. For use only internally by the
409 * journaling code (called from journal_get_write_access).
410 *
97f168b6 411 * We trust buffer_revoked() on the buffer if the buffer is already
0e8a9560
TT
412 * being journaled: if there is no revoke pending on the buffer, then we
413 * don't do anything here.
414 *
415 * This would break if it were possible for a buffer to be revoked and
416 * discarded, and then reallocated within the same transaction. In such
417 * a case we would have lost the revoked bit, but when we arrived here
418 * the second time we would still have a pending revoke to cancel. So,
419 * do not trust the Revoked bit on buffers unless RevokeValid is also
420 * set.
8cf93332
TT
421 */
422int journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
0e8a9560 423{
97f168b6 424 struct jbd2_revoke_record_s *record;
0e8a9560
TT
425 journal_t *journal = handle->h_transaction->t_journal;
426 int need_cancel;
8cf93332
TT
427 int did_revoke = 0; /* akpm: debug */
428 struct buffer_head *bh = jh2bh(jh);
efc6f628 429
ce20096f 430 jbd_debug(4, "journal_head %p, canceling revoke\n", jh);
8cf93332 431
0e8a9560
TT
432 /* Is the existing Revoke bit valid? If so, we trust it, and
433 * only perform the full cancel if the revoke bit is set. If
434 * not, we can't trust the revoke bit, and we need to do the
435 * full search for a revoke record. */
97f168b6
DW
436 if (test_set_buffer_revokevalid(bh)) {
437 need_cancel = test_clear_buffer_revoked(bh);
438 } else {
0e8a9560 439 need_cancel = 1;
97f168b6 440 clear_buffer_revoked(bh);
0e8a9560 441 }
8cf93332 442
0e8a9560
TT
443 if (need_cancel) {
444 record = find_revoke_record(journal, bh->b_blocknr);
445 if (record) {
8cf93332 446 jbd_debug(4, "cancelled existing revoke on "
97f168b6
DW
447 "blocknr %llu\n", (unsigned long long)bh->b_blocknr);
448 spin_lock(&journal->j_revoke_lock);
0e8a9560 449 list_del(&record->hash);
97f168b6
DW
450 spin_unlock(&journal->j_revoke_lock);
451 kmem_cache_free(jbd2_revoke_record_cache, record);
8cf93332 452 did_revoke = 1;
0e8a9560
TT
453 }
454 }
8cf93332 455
97f168b6 456#ifdef JFS_EXPENSIVE_CHECKING
8cf93332
TT
457 /* There better not be one left behind by now! */
458 record = find_revoke_record(journal, bh->b_blocknr);
459 J_ASSERT_JH(jh, record == NULL);
460#endif
461
462 /* Finally, have we just cleared revoke on an unhashed
463 * buffer_head? If so, we'd better make sure we clear the
464 * revoked status on any hashed alias too, otherwise the revoke
465 * state machine will get very upset later on. */
97f168b6 466 if (need_cancel) {
8cf93332 467 struct buffer_head *bh2;
97f168b6 468 bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size);
8cf93332 469 if (bh2) {
97f168b6
DW
470 if (bh2 != bh)
471 clear_buffer_revoked(bh2);
8cf93332
TT
472 __brelse(bh2);
473 }
474 }
8cf93332 475 return did_revoke;
0e8a9560
TT
476}
477
97f168b6
DW
478/*
479 * journal_clear_revoked_flag clears revoked flag of buffers in
480 * revoke table to reflect there is no revoked buffers in the next
481 * transaction which is going to be started.
482 */
483void jbd2_clear_buffer_revoked_flags(journal_t *journal)
484{
485 struct jbd2_revoke_table_s *revoke = journal->j_revoke;
486 int i = 0;
487
488 for (i = 0; i < revoke->hash_size; i++) {
489 struct list_head *hash_list;
490 struct list_head *list_entry;
491 hash_list = &revoke->hash_table[i];
492
493 list_for_each(list_entry, hash_list) {
494 struct jbd2_revoke_record_s *record;
495 struct buffer_head *bh;
496 record = (struct jbd2_revoke_record_s *)list_entry;
497 bh = __find_get_block(journal->j_fs_dev,
498 record->blocknr,
499 journal->j_blocksize);
500 if (bh) {
501 clear_buffer_revoked(bh);
502 __brelse(bh);
503 }
504 }
505 }
506}
507
508/* journal_switch_revoke table select j_revoke for next transaction
509 * we do not want to suspend any processing until all revokes are
510 * written -bzzz
511 */
512void journal_switch_revoke_table(journal_t *journal)
513{
514 int i;
515
516 if (journal->j_revoke == journal->j_revoke_table[0])
517 journal->j_revoke = journal->j_revoke_table[1];
518 else
519 journal->j_revoke = journal->j_revoke_table[0];
520
521 for (i = 0; i < journal->j_revoke->hash_size; i++)
522 INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]);
523}
0e8a9560
TT
524
525/*
526 * Write revoke records to the journal for all entries in the current
527 * revoke hash, deleting the entries as we go.
0e8a9560 528 */
efc6f628 529void journal_write_revoke_records(journal_t *journal,
97f168b6
DW
530 transaction_t *transaction,
531 struct list_head *log_bufs,
532 int write_op)
0e8a9560 533{
97f168b6
DW
534 struct buffer_head *descriptor;
535 struct jbd2_revoke_record_s *record;
536 struct jbd2_revoke_table_s *revoke;
0e8a9560 537 struct list_head *hash_list;
1f735038 538 int i, offset, count;
8cf93332 539
efc6f628 540 descriptor = NULL;
0e8a9560 541 offset = 0;
1f735038 542 count = 0;
97f168b6
DW
543
544 /* select revoke table for committing transaction */
545 revoke = journal->j_revoke == journal->j_revoke_table[0] ?
546 journal->j_revoke_table[1] : journal->j_revoke_table[0];
efc6f628 547
0e8a9560
TT
548 for (i = 0; i < revoke->hash_size; i++) {
549 hash_list = &revoke->hash_table[i];
550
551 while (!list_empty(hash_list)) {
97f168b6 552 record = (struct jbd2_revoke_record_s *)
0e8a9560 553 hash_list->next;
97f168b6 554 write_one_revoke_record(journal, transaction, log_bufs,
efc6f628 555 &descriptor, &offset,
97f168b6 556 record, write_op);
1f735038 557 count++;
0e8a9560 558 list_del(&record->hash);
97f168b6 559 kmem_cache_free(jbd2_revoke_record_cache, record);
0e8a9560
TT
560 }
561 }
efc6f628 562 if (descriptor)
97f168b6 563 flush_descriptor(journal, descriptor, offset, write_op);
8cf93332 564 jbd_debug(1, "Wrote %d revoke records\n", count);
0e8a9560
TT
565}
566
efc6f628 567/*
0e8a9560 568 * Write out one revoke record. We need to create a new descriptor
efc6f628 569 * block if the old one is full or if we have not already created one.
0e8a9560
TT
570 */
571
efc6f628 572static void write_one_revoke_record(journal_t *journal,
0e8a9560 573 transaction_t *transaction,
97f168b6
DW
574 struct list_head *log_bufs,
575 struct buffer_head **descriptorp,
0e8a9560 576 int *offsetp,
97f168b6
DW
577 struct jbd2_revoke_record_s *record,
578 int write_op)
0e8a9560 579{
97f168b6
DW
580 int csum_size = 0;
581 struct buffer_head *descriptor;
04c66cb2 582 int sz, offset;
0e8a9560 583 journal_header_t *header;
8cf93332 584
0e8a9560
TT
585 /* If we are already aborting, this all becomes a noop. We
586 still need to go round the loop in
587 journal_write_revoke_records in order to free all of the
588 revoke records: only the IO to the journal is omitted. */
8cf93332 589 if (is_journal_aborted(journal))
0e8a9560
TT
590 return;
591
592 descriptor = *descriptorp;
593 offset = *offsetp;
8cf93332 594
97f168b6 595 /* Do we need to leave space at the end for a checksum? */
38d5adf3 596 if (journal_has_csum_v2or3(journal))
97f168b6
DW
597 csum_size = sizeof(struct journal_revoke_tail);
598
04c66cb2
DW
599 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
600 sz = 8;
601 else
602 sz = 4;
603
0e8a9560
TT
604 /* Make sure we have a descriptor with space left for the record */
605 if (descriptor) {
04c66cb2 606 if (offset + sz > journal->j_blocksize - csum_size) {
97f168b6 607 flush_descriptor(journal, descriptor, offset, write_op);
0e8a9560
TT
608 descriptor = NULL;
609 }
610 }
efc6f628 611
0e8a9560
TT
612 if (!descriptor) {
613 descriptor = journal_get_descriptor_buffer(journal);
8cf93332
TT
614 if (!descriptor)
615 return;
97f168b6
DW
616 header = (journal_header_t *)descriptor->b_data;
617 header->h_magic = ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER);
618 header->h_blocktype = ext2fs_cpu_to_be32(JFS_REVOKE_BLOCK);
619 header->h_sequence = ext2fs_cpu_to_be32(transaction->t_tid);
0e8a9560
TT
620
621 /* Record it so that we can wait for IO completion later */
97f168b6
DW
622 BUFFER_TRACE(descriptor, "file in log_bufs");
623 jbd2_file_log_bh(log_bufs, descriptor);
8cf93332 624
0e8a9560
TT
625 offset = sizeof(journal_revoke_header_t);
626 *descriptorp = descriptor;
627 }
efc6f628 628
04c66cb2
DW
629 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) {
630 * ((__be64 *)(&descriptor->b_data[offset])) =
631 cpu_to_be64(record->blocknr);
632 else
633 * ((__be32 *)(&descriptor->b_data[offset])) =
634 cpu_to_be32(record->blocknr);
635 offset += sz;
97f168b6 636
0e8a9560
TT
637 *offsetp = offset;
638}
639
97f168b6
DW
640static void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh)
641{
642 struct journal_revoke_tail *tail;
643 __u32 csum;
644
38d5adf3 645 if (!journal_has_csum_v2or3(j))
97f168b6
DW
646 return;
647
648 tail = (struct journal_revoke_tail *)(bh->b_data + j->j_blocksize -
649 sizeof(struct journal_revoke_tail));
650 tail->r_checksum = 0;
651 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
652 tail->r_checksum = ext2fs_cpu_to_be32(csum);
653}
654
efc6f628 655/*
0e8a9560
TT
656 * Flush a revoke descriptor out to the journal. If we are aborting,
657 * this is a noop; otherwise we are generating a buffer which needs to
658 * be waited for during commit, so it has to go onto the appropriate
659 * journal buffer list.
660 */
661
efc6f628 662static void flush_descriptor(journal_t *journal,
97f168b6
DW
663 struct buffer_head *descriptor,
664 int offset, int write_op)
0e8a9560
TT
665{
666 journal_revoke_header_t *header;
8cf93332
TT
667
668 if (is_journal_aborted(journal)) {
97f168b6 669 put_bh(descriptor);
0e8a9560
TT
670 return;
671 }
efc6f628 672
97f168b6
DW
673 header = (journal_revoke_header_t *)descriptor->b_data;
674 header->r_count = ext2fs_cpu_to_be32(offset);
675 jbd2_revoke_csum_set(journal, descriptor);
0e8a9560 676
97f168b6
DW
677 set_buffer_jwrite(descriptor);
678 BUFFER_TRACE(descriptor, "write");
679 set_buffer_dirty(descriptor);
680 write_dirty_buffer(descriptor, write_op);
681}
0e8a9560
TT
682#endif
683
efc6f628 684/*
0e8a9560
TT
685 * Revoke support for recovery.
686 *
687 * Recovery needs to be able to:
688 *
689 * record all revoke records, including the tid of the latest instance
690 * of each revoke in the journal
691 *
692 * check whether a given block in a given transaction should be replayed
693 * (ie. has not been revoked by a revoke record in that or a subsequent
694 * transaction)
efc6f628 695 *
0e8a9560
TT
696 * empty the revoke table after recovery.
697 */
698
699/*
700 * First, setting revoke records. We create a new revoke record for
701 * every block ever revoked in the log as we scan it for recovery, and
702 * we update the existing records if we find multiple revokes for a
efc6f628 703 * single block.
0e8a9560
TT
704 */
705
efc6f628 706int journal_set_revoke(journal_t *journal,
97f168b6 707 unsigned long long blocknr,
0e8a9560
TT
708 tid_t sequence)
709{
97f168b6 710 struct jbd2_revoke_record_s *record;
efc6f628 711
0e8a9560
TT
712 record = find_revoke_record(journal, blocknr);
713 if (record) {
97f168b6 714 /* If we have multiple occurrences, only record the
0e8a9560 715 * latest sequence number in the hashed record */
725c474f 716 if (tid_gt(sequence, record->sequence))
0e8a9560
TT
717 record->sequence = sequence;
718 return 0;
efc6f628 719 }
0e8a9560
TT
720 return insert_revoke_hash(journal, blocknr, sequence);
721}
722
efc6f628 723/*
0e8a9560
TT
724 * Test revoke records. For a given block referenced in the log, has
725 * that block been revoked? A revoke record with a given transaction
726 * sequence number revokes all blocks in that transaction and earlier
727 * ones, but later transactions still need replayed.
728 */
729
efc6f628 730int journal_test_revoke(journal_t *journal,
97f168b6 731 unsigned long long blocknr,
0e8a9560
TT
732 tid_t sequence)
733{
97f168b6 734 struct jbd2_revoke_record_s *record;
efc6f628 735
0e8a9560
TT
736 record = find_revoke_record(journal, blocknr);
737 if (!record)
738 return 0;
725c474f 739 if (tid_gt(sequence, record->sequence))
0e8a9560
TT
740 return 0;
741 return 1;
742}
743
744/*
745 * Finally, once recovery is over, we need to clear the revoke table so
746 * that it can be reused by the running filesystem.
747 */
748
749void journal_clear_revoke(journal_t *journal)
750{
751 int i;
752 struct list_head *hash_list;
97f168b6
DW
753 struct jbd2_revoke_record_s *record;
754 struct jbd2_revoke_table_s *revoke;
efc6f628 755
0e8a9560 756 revoke = journal->j_revoke;
efc6f628 757
0e8a9560
TT
758 for (i = 0; i < revoke->hash_size; i++) {
759 hash_list = &revoke->hash_table[i];
760 while (!list_empty(hash_list)) {
97f168b6 761 record = (struct jbd2_revoke_record_s*) hash_list->next;
0e8a9560 762 list_del(&record->hash);
97f168b6 763 kmem_cache_free(jbd2_revoke_record_cache, record);
0e8a9560
TT
764 }
765 }
766}