Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0+
2 : /*
3 : * linux/fs/jbd2/journal.c
4 : *
5 : * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 : *
7 : * Copyright 1998 Red Hat corp --- All Rights Reserved
8 : *
9 : * Generic filesystem journal-writing code; part of the ext2fs
10 : * journaling system.
11 : *
12 : * This file manages journals: areas of disk reserved for logging
13 : * transactional updates. This includes the kernel journaling thread
14 : * which is responsible for scheduling updates to the log.
15 : *
16 : * We do not actually manage the physical storage of the journal in this
17 : * file: that is left to a per-journal policy function, which allows us
18 : * to store the journal within a filesystem-specified area for ext2
19 : * journaling (ext2 can use a reserved inode for storing the log).
20 : */
21 :
22 : #include <linux/module.h>
23 : #include <linux/time.h>
24 : #include <linux/fs.h>
25 : #include <linux/jbd2.h>
26 : #include <linux/errno.h>
27 : #include <linux/slab.h>
28 : #include <linux/init.h>
29 : #include <linux/mm.h>
30 : #include <linux/freezer.h>
31 : #include <linux/pagemap.h>
32 : #include <linux/kthread.h>
33 : #include <linux/poison.h>
34 : #include <linux/proc_fs.h>
35 : #include <linux/seq_file.h>
36 : #include <linux/math64.h>
37 : #include <linux/hash.h>
38 : #include <linux/log2.h>
39 : #include <linux/vmalloc.h>
40 : #include <linux/backing-dev.h>
41 : #include <linux/bitops.h>
42 : #include <linux/ratelimit.h>
43 : #include <linux/sched/mm.h>
44 :
45 : #define CREATE_TRACE_POINTS
46 : #include <trace/events/jbd2.h>
47 :
48 : #include <linux/uaccess.h>
49 : #include <asm/page.h>
50 :
51 : #ifdef CONFIG_JBD2_DEBUG
52 : static ushort jbd2_journal_enable_debug __read_mostly;
53 :
54 : module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
55 : MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
56 : #endif
57 :
58 : EXPORT_SYMBOL(jbd2_journal_extend);
59 : EXPORT_SYMBOL(jbd2_journal_stop);
60 : EXPORT_SYMBOL(jbd2_journal_lock_updates);
61 : EXPORT_SYMBOL(jbd2_journal_unlock_updates);
62 : EXPORT_SYMBOL(jbd2_journal_get_write_access);
63 : EXPORT_SYMBOL(jbd2_journal_get_create_access);
64 : EXPORT_SYMBOL(jbd2_journal_get_undo_access);
65 : EXPORT_SYMBOL(jbd2_journal_set_triggers);
66 : EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
67 : EXPORT_SYMBOL(jbd2_journal_forget);
68 : EXPORT_SYMBOL(jbd2_journal_flush);
69 : EXPORT_SYMBOL(jbd2_journal_revoke);
70 :
71 : EXPORT_SYMBOL(jbd2_journal_init_dev);
72 : EXPORT_SYMBOL(jbd2_journal_init_inode);
73 : EXPORT_SYMBOL(jbd2_journal_check_used_features);
74 : EXPORT_SYMBOL(jbd2_journal_check_available_features);
75 : EXPORT_SYMBOL(jbd2_journal_set_features);
76 : EXPORT_SYMBOL(jbd2_journal_load);
77 : EXPORT_SYMBOL(jbd2_journal_destroy);
78 : EXPORT_SYMBOL(jbd2_journal_abort);
79 : EXPORT_SYMBOL(jbd2_journal_errno);
80 : EXPORT_SYMBOL(jbd2_journal_ack_err);
81 : EXPORT_SYMBOL(jbd2_journal_clear_err);
82 : EXPORT_SYMBOL(jbd2_log_wait_commit);
83 : EXPORT_SYMBOL(jbd2_journal_start_commit);
84 : EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
85 : EXPORT_SYMBOL(jbd2_journal_wipe);
86 : EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
87 : EXPORT_SYMBOL(jbd2_journal_invalidate_folio);
88 : EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
89 : EXPORT_SYMBOL(jbd2_journal_force_commit);
90 : EXPORT_SYMBOL(jbd2_journal_inode_ranged_write);
91 : EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
92 : EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers);
93 : EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
94 : EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
95 : EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
96 : EXPORT_SYMBOL(jbd2_inode_cache);
97 :
98 : static int jbd2_journal_create_slab(size_t slab_size);
99 :
100 : #ifdef CONFIG_JBD2_DEBUG
101 : void __jbd2_debug(int level, const char *file, const char *func,
102 : unsigned int line, const char *fmt, ...)
103 : {
104 : struct va_format vaf;
105 : va_list args;
106 :
107 : if (level > jbd2_journal_enable_debug)
108 : return;
109 : va_start(args, fmt);
110 : vaf.fmt = fmt;
111 : vaf.va = &args;
112 : printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
113 : va_end(args);
114 : }
115 : #endif
116 :
117 : /* Checksumming functions */
118 : static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
119 : {
120 2501 : if (!jbd2_journal_has_csum_v2or3_feature(j))
121 : return 1;
122 :
123 1327 : return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
124 : }
125 :
126 11211 : static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
127 : {
128 11211 : __u32 csum;
129 11211 : __be32 old_csum;
130 :
131 11211 : old_csum = sb->s_checksum;
132 11211 : sb->s_checksum = 0;
133 11211 : csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
134 11211 : sb->s_checksum = old_csum;
135 :
136 11211 : return cpu_to_be32(csum);
137 : }
138 :
139 : /*
140 : * Helper function used to manage commit timeouts
141 : */
142 :
143 204 : static void commit_timeout(struct timer_list *t)
144 : {
145 204 : journal_t *journal = from_timer(journal, t, j_commit_timer);
146 :
147 204 : wake_up_process(journal->j_task);
148 204 : }
149 :
150 : /*
151 : * kjournald2: The main thread function used to manage a logging device
152 : * journal.
153 : *
154 : * This kernel thread is responsible for two things:
155 : *
156 : * 1) COMMIT: Every so often we need to commit the current state of the
157 : * filesystem to disk. The journal thread is responsible for writing
158 : * all of the metadata buffers to disk. If a fast commit is ongoing
159 : * journal thread waits until it's done and then continues from
160 : * there on.
161 : *
162 : * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
163 : * of the data in that part of the log has been rewritten elsewhere on
164 : * the disk. Flushing these old buffers to reclaim space in the log is
165 : * known as checkpointing, and this thread is responsible for that job.
166 : */
167 :
168 2501 : static int kjournald2(void *arg)
169 : {
170 2501 : journal_t *journal = arg;
171 2501 : transaction_t *transaction;
172 :
173 : /*
174 : * Set up an interval timer which can be used to trigger a commit wakeup
175 : * after the commit interval expires
176 : */
177 2501 : timer_setup(&journal->j_commit_timer, commit_timeout, 0);
178 :
179 2501 : set_freezable();
180 :
181 : /* Record that the journal thread is running */
182 2501 : journal->j_task = current;
183 2501 : wake_up(&journal->j_wait_done_commit);
184 :
185 : /*
186 : * Make sure that no allocations from this kernel thread will ever
187 : * recurse to the fs layer because we are responsible for the
188 : * transaction commit and any fs involvement might get stuck waiting for
189 : * the trasn. commit.
190 : */
191 2501 : memalloc_nofs_save();
192 :
193 : /*
194 : * And now, wait forever for commit wakeup events.
195 : */
196 2501 : write_lock(&journal->j_state_lock);
197 :
198 : loop:
199 399147 : if (journal->j_flags & JBD2_UNMOUNT)
200 2501 : goto end_loop;
201 :
202 396646 : jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n",
203 : journal->j_commit_sequence, journal->j_commit_request);
204 :
205 396646 : if (journal->j_commit_sequence != journal->j_commit_request) {
206 201275 : jbd2_debug(1, "OK, requests differ\n");
207 201275 : write_unlock(&journal->j_state_lock);
208 201275 : del_timer_sync(&journal->j_commit_timer);
209 201275 : jbd2_journal_commit_transaction(journal);
210 201275 : write_lock(&journal->j_state_lock);
211 201275 : goto loop;
212 : }
213 :
214 195371 : wake_up(&journal->j_wait_done_commit);
215 195371 : if (freezing(current)) {
216 : /*
217 : * The simpler the better. Flushing journal isn't a
218 : * good idea, because that depends on threads that may
219 : * be already stopped.
220 : */
221 0 : jbd2_debug(1, "Now suspending kjournald2\n");
222 0 : write_unlock(&journal->j_state_lock);
223 0 : try_to_freeze();
224 0 : write_lock(&journal->j_state_lock);
225 : } else {
226 : /*
227 : * We assume on resume that commits are already there,
228 : * so we don't sleep
229 : */
230 195371 : DEFINE_WAIT(wait);
231 195371 : int should_sleep = 1;
232 :
233 195371 : prepare_to_wait(&journal->j_wait_commit, &wait,
234 : TASK_INTERRUPTIBLE);
235 195371 : if (journal->j_commit_sequence != journal->j_commit_request)
236 0 : should_sleep = 0;
237 195371 : transaction = journal->j_running_transaction;
238 195371 : if (transaction && time_after_eq(jiffies,
239 : transaction->t_expires))
240 0 : should_sleep = 0;
241 195371 : if (journal->j_flags & JBD2_UNMOUNT)
242 : should_sleep = 0;
243 195371 : if (should_sleep) {
244 195371 : write_unlock(&journal->j_state_lock);
245 195371 : schedule();
246 195371 : write_lock(&journal->j_state_lock);
247 : }
248 195371 : finish_wait(&journal->j_wait_commit, &wait);
249 : }
250 :
251 195371 : jbd2_debug(1, "kjournald2 wakes\n");
252 :
253 : /*
254 : * Were we woken up by a commit wakeup event?
255 : */
256 195371 : transaction = journal->j_running_transaction;
257 195371 : if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
258 798 : journal->j_commit_request = transaction->t_tid;
259 798 : jbd2_debug(1, "woke because of timeout\n");
260 : }
261 195371 : goto loop;
262 :
263 : end_loop:
264 2501 : del_timer_sync(&journal->j_commit_timer);
265 2501 : journal->j_task = NULL;
266 2501 : wake_up(&journal->j_wait_done_commit);
267 2501 : jbd2_debug(1, "Journal thread exiting.\n");
268 2501 : write_unlock(&journal->j_state_lock);
269 2501 : return 0;
270 : }
271 :
272 2501 : static int jbd2_journal_start_thread(journal_t *journal)
273 : {
274 2501 : struct task_struct *t;
275 :
276 2501 : t = kthread_run(kjournald2, journal, "jbd2/%s",
277 : journal->j_devname);
278 2501 : if (IS_ERR(t))
279 0 : return PTR_ERR(t);
280 :
281 4996 : wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
282 : return 0;
283 : }
284 :
285 2503 : static void journal_kill_thread(journal_t *journal)
286 : {
287 2503 : write_lock(&journal->j_state_lock);
288 2503 : journal->j_flags |= JBD2_UNMOUNT;
289 :
290 5004 : while (journal->j_task) {
291 2501 : write_unlock(&journal->j_state_lock);
292 2501 : wake_up(&journal->j_wait_commit);
293 3768 : wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
294 2501 : write_lock(&journal->j_state_lock);
295 : }
296 2503 : write_unlock(&journal->j_state_lock);
297 2503 : }
298 :
299 : /*
300 : * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
301 : *
302 : * Writes a metadata buffer to a given disk block. The actual IO is not
303 : * performed but a new buffer_head is constructed which labels the data
304 : * to be written with the correct destination disk block.
305 : *
306 : * Any magic-number escaping which needs to be done will cause a
307 : * copy-out here. If the buffer happens to start with the
308 : * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
309 : * magic number is only written to the log for descripter blocks. In
310 : * this case, we copy the data and replace the first word with 0, and we
311 : * return a result code which indicates that this buffer needs to be
312 : * marked as an escaped buffer in the corresponding log descriptor
313 : * block. The missing word can then be restored when the block is read
314 : * during recovery.
315 : *
316 : * If the source buffer has already been modified by a new transaction
317 : * since we took the last commit snapshot, we use the frozen copy of
318 : * that data for IO. If we end up using the existing buffer_head's data
319 : * for the write, then we have to make sure nobody modifies it while the
320 : * IO is in progress. do_get_write_access() handles this.
321 : *
322 : * The function returns a pointer to the buffer_head to be used for IO.
323 : *
324 : *
325 : * Return value:
326 : * <0: Error
327 : * >=0: Finished OK
328 : *
329 : * On success:
330 : * Bit 0 set == escape performed on the data
331 : * Bit 1 set == buffer copy-out performed (kfree the data after IO)
332 : */
333 :
334 3121499 : int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
335 : struct journal_head *jh_in,
336 : struct buffer_head **bh_out,
337 : sector_t blocknr)
338 : {
339 3121499 : int need_copy_out = 0;
340 3121499 : int done_copy_out = 0;
341 3121499 : int do_escape = 0;
342 3121499 : char *mapped_data;
343 3121499 : struct buffer_head *new_bh;
344 3121499 : struct page *new_page;
345 3121499 : unsigned int new_offset;
346 3121499 : struct buffer_head *bh_in = jh2bh(jh_in);
347 3121499 : journal_t *journal = transaction->t_journal;
348 :
349 : /*
350 : * The buffer really shouldn't be locked: only the current committing
351 : * transaction is allowed to write it, so nobody else is allowed
352 : * to do any IO.
353 : *
354 : * akpm: except if we're journalling data, and write() output is
355 : * also part of a shared mapping, and another thread has
356 : * decided to launch a writepage() against this buffer.
357 : */
358 6242998 : J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
359 :
360 3121499 : new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
361 :
362 : /* keep subsequent assertions sane */
363 3121499 : atomic_set(&new_bh->b_count, 1);
364 :
365 3121499 : spin_lock(&jh_in->b_state_lock);
366 3121499 : repeat:
367 : /*
368 : * If a new transaction has already done a buffer copy-out, then
369 : * we use that version of the data for the commit.
370 : */
371 3121499 : if (jh_in->b_frozen_data) {
372 21377 : done_copy_out = 1;
373 21377 : new_page = virt_to_page(jh_in->b_frozen_data);
374 21377 : new_offset = offset_in_page(jh_in->b_frozen_data);
375 : } else {
376 3100122 : new_page = jh2bh(jh_in)->b_page;
377 3100122 : new_offset = offset_in_page(jh2bh(jh_in)->b_data);
378 : }
379 :
380 3121499 : mapped_data = kmap_atomic(new_page);
381 : /*
382 : * Fire data frozen trigger if data already wasn't frozen. Do this
383 : * before checking for escaping, as the trigger may modify the magic
384 : * offset. If a copy-out happens afterwards, it will have the correct
385 : * data in the buffer.
386 : */
387 3121499 : if (!done_copy_out)
388 3100122 : jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
389 : jh_in->b_triggers);
390 :
391 : /*
392 : * Check for escaping
393 : */
394 3121499 : if (*((__be32 *)(mapped_data + new_offset)) ==
395 : cpu_to_be32(JBD2_MAGIC_NUMBER)) {
396 0 : need_copy_out = 1;
397 0 : do_escape = 1;
398 : }
399 3121499 : kunmap_atomic(mapped_data);
400 :
401 : /*
402 : * Do we need to do a data copy?
403 : */
404 3121499 : if (need_copy_out && !done_copy_out) {
405 0 : char *tmp;
406 :
407 0 : spin_unlock(&jh_in->b_state_lock);
408 0 : tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
409 0 : if (!tmp) {
410 0 : brelse(new_bh);
411 0 : return -ENOMEM;
412 : }
413 0 : spin_lock(&jh_in->b_state_lock);
414 0 : if (jh_in->b_frozen_data) {
415 0 : jbd2_free(tmp, bh_in->b_size);
416 0 : goto repeat;
417 : }
418 :
419 0 : jh_in->b_frozen_data = tmp;
420 0 : mapped_data = kmap_atomic(new_page);
421 0 : memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
422 0 : kunmap_atomic(mapped_data);
423 :
424 0 : new_page = virt_to_page(tmp);
425 0 : new_offset = offset_in_page(tmp);
426 0 : done_copy_out = 1;
427 :
428 : /*
429 : * This isn't strictly necessary, as we're using frozen
430 : * data for the escaping, but it keeps consistency with
431 : * b_frozen_data usage.
432 : */
433 0 : jh_in->b_frozen_triggers = jh_in->b_triggers;
434 : }
435 :
436 : /*
437 : * Did we need to do an escaping? Now we've done all the
438 : * copying, we can finally do so.
439 : */
440 3121499 : if (do_escape) {
441 0 : mapped_data = kmap_atomic(new_page);
442 0 : *((unsigned int *)(mapped_data + new_offset)) = 0;
443 0 : kunmap_atomic(mapped_data);
444 : }
445 :
446 3121499 : set_bh_page(new_bh, new_page, new_offset);
447 3121499 : new_bh->b_size = bh_in->b_size;
448 3121499 : new_bh->b_bdev = journal->j_dev;
449 3121499 : new_bh->b_blocknr = blocknr;
450 3121499 : new_bh->b_private = bh_in;
451 3121499 : set_buffer_mapped(new_bh);
452 3121499 : set_buffer_dirty(new_bh);
453 :
454 3121499 : *bh_out = new_bh;
455 :
456 : /*
457 : * The to-be-written buffer needs to get moved to the io queue,
458 : * and the original buffer whose contents we are shadowing or
459 : * copying is moved to the transaction's shadow queue.
460 : */
461 3121499 : JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
462 3121499 : spin_lock(&journal->j_list_lock);
463 3121499 : __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
464 3121499 : spin_unlock(&journal->j_list_lock);
465 3121499 : set_buffer_shadow(bh_in);
466 3121499 : spin_unlock(&jh_in->b_state_lock);
467 :
468 3121499 : return do_escape | (done_copy_out << 1);
469 : }
470 :
471 : /*
472 : * Allocation code for the journal file. Manage the space left in the
473 : * journal, so that we can begin checkpointing when appropriate.
474 : */
475 :
476 : /*
477 : * Called with j_state_lock locked for writing.
478 : * Returns true if a transaction commit was started.
479 : */
480 439510 : static int __jbd2_log_start_commit(journal_t *journal, tid_t target)
481 : {
482 : /* Return if the txn has already requested to be committed */
483 439510 : if (journal->j_commit_request == target)
484 : return 0;
485 :
486 : /*
487 : * The only transaction we can possibly wait upon is the
488 : * currently running transaction (if it exists). Otherwise,
489 : * the target tid must be an old one.
490 : */
491 207246 : if (journal->j_running_transaction &&
492 207246 : journal->j_running_transaction->t_tid == target) {
493 : /*
494 : * We want a new commit: OK, mark the request and wakeup the
495 : * commit thread. We do _not_ do the commit ourselves.
496 : */
497 :
498 201071 : journal->j_commit_request = target;
499 201071 : jbd2_debug(1, "JBD2: requesting commit %u/%u\n",
500 : journal->j_commit_request,
501 : journal->j_commit_sequence);
502 201071 : journal->j_running_transaction->t_requested = jiffies;
503 201071 : wake_up(&journal->j_wait_commit);
504 201071 : return 1;
505 6175 : } else if (!tid_geq(journal->j_commit_request, target))
506 : /* This should never happen, but if it does, preserve
507 : the evidence before kjournald goes into a loop and
508 : increments j_commit_sequence beyond all recognition. */
509 0 : WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
510 : journal->j_commit_request,
511 : journal->j_commit_sequence,
512 : target, journal->j_running_transaction ?
513 : journal->j_running_transaction->t_tid : 0);
514 : return 0;
515 : }
516 :
517 413746 : int jbd2_log_start_commit(journal_t *journal, tid_t tid)
518 : {
519 413746 : int ret;
520 :
521 413746 : write_lock(&journal->j_state_lock);
522 413880 : ret = __jbd2_log_start_commit(journal, tid);
523 413880 : write_unlock(&journal->j_state_lock);
524 413874 : return ret;
525 : }
526 :
527 : /*
528 : * Force and wait any uncommitted transactions. We can only force the running
529 : * transaction if we don't have an active handle, otherwise, we will deadlock.
530 : * Returns: <0 in case of error,
531 : * 0 if nothing to commit,
532 : * 1 if transaction was successfully committed.
533 : */
534 123402 : static int __jbd2_journal_force_commit(journal_t *journal)
535 : {
536 123402 : transaction_t *transaction = NULL;
537 123402 : tid_t tid;
538 123402 : int need_to_start = 0, ret = 0;
539 :
540 123402 : read_lock(&journal->j_state_lock);
541 123432 : if (journal->j_running_transaction && !current->journal_info) {
542 123210 : transaction = journal->j_running_transaction;
543 123210 : if (!tid_geq(journal->j_commit_request, transaction->t_tid))
544 10093 : need_to_start = 1;
545 222 : } else if (journal->j_committing_transaction)
546 : transaction = journal->j_committing_transaction;
547 :
548 123432 : if (!transaction) {
549 : /* Nothing to commit */
550 50 : read_unlock(&journal->j_state_lock);
551 50 : return 0;
552 : }
553 123382 : tid = transaction->t_tid;
554 123382 : read_unlock(&journal->j_state_lock);
555 123354 : if (need_to_start)
556 10091 : jbd2_log_start_commit(journal, tid);
557 123358 : ret = jbd2_log_wait_commit(journal, tid);
558 123175 : if (!ret)
559 123149 : ret = 1;
560 :
561 : return ret;
562 : }
563 :
564 : /**
565 : * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
566 : * calling process is not within transaction.
567 : *
568 : * @journal: journal to force
569 : * Returns true if progress was made.
570 : *
571 : * This is used for forcing out undo-protected data which contains
572 : * bitmaps, when the fs is running out of space.
573 : */
574 120060 : int jbd2_journal_force_commit_nested(journal_t *journal)
575 : {
576 120060 : int ret;
577 :
578 120060 : ret = __jbd2_journal_force_commit(journal);
579 119813 : return ret > 0;
580 : }
581 :
582 : /**
583 : * jbd2_journal_force_commit() - force any uncommitted transactions
584 : * @journal: journal to force
585 : *
586 : * Caller want unconditional commit. We can only force the running transaction
587 : * if we don't have an active handle, otherwise, we will deadlock.
588 : */
589 3352 : int jbd2_journal_force_commit(journal_t *journal)
590 : {
591 3352 : int ret;
592 :
593 3352 : J_ASSERT(!current->journal_info);
594 3352 : ret = __jbd2_journal_force_commit(journal);
595 3350 : if (ret > 0)
596 : ret = 0;
597 3350 : return ret;
598 : }
599 :
600 : /*
601 : * Start a commit of the current running transaction (if any). Returns true
602 : * if a transaction is going to be committed (or is currently already
603 : * committing), and fills its tid in at *ptid
604 : */
605 70614 : int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
606 : {
607 70614 : int ret = 0;
608 :
609 70614 : write_lock(&journal->j_state_lock);
610 70633 : if (journal->j_running_transaction) {
611 25563 : tid_t tid = journal->j_running_transaction->t_tid;
612 :
613 25563 : __jbd2_log_start_commit(journal, tid);
614 : /* There's a running transaction and we've just made sure
615 : * it's commit has been scheduled. */
616 25563 : if (ptid)
617 25563 : *ptid = tid;
618 : ret = 1;
619 45070 : } else if (journal->j_committing_transaction) {
620 : /*
621 : * If commit has been started, then we have to wait for
622 : * completion of that transaction.
623 : */
624 6575 : if (ptid)
625 6575 : *ptid = journal->j_committing_transaction->t_tid;
626 : ret = 1;
627 : }
628 70633 : write_unlock(&journal->j_state_lock);
629 70630 : return ret;
630 : }
631 :
632 : /*
633 : * Return 1 if a given transaction has not yet sent barrier request
634 : * connected with a transaction commit. If 0 is returned, transaction
635 : * may or may not have sent the barrier. Used to avoid sending barrier
636 : * twice in common cases.
637 : */
638 269205 : int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
639 : {
640 269205 : int ret = 0;
641 269205 : transaction_t *commit_trans;
642 :
643 269205 : if (!(journal->j_flags & JBD2_BARRIER))
644 : return 0;
645 269191 : read_lock(&journal->j_state_lock);
646 : /* Transaction already committed? */
647 269192 : if (tid_geq(journal->j_commit_sequence, tid))
648 67163 : goto out;
649 202029 : commit_trans = journal->j_committing_transaction;
650 202029 : if (!commit_trans || commit_trans->t_tid != tid) {
651 189840 : ret = 1;
652 189840 : goto out;
653 : }
654 : /*
655 : * Transaction is being committed and we already proceeded to
656 : * submitting a flush to fs partition?
657 : */
658 12189 : if (journal->j_fs_dev != journal->j_dev) {
659 0 : if (!commit_trans->t_need_data_flush ||
660 0 : commit_trans->t_state >= T_COMMIT_DFLUSH)
661 0 : goto out;
662 : } else {
663 12189 : if (commit_trans->t_state >= T_COMMIT_JFLUSH)
664 556 : goto out;
665 : }
666 : ret = 1;
667 269192 : out:
668 269192 : read_unlock(&journal->j_state_lock);
669 269192 : return ret;
670 : }
671 : EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
672 :
673 : /*
674 : * Wait for a specified commit to complete.
675 : * The caller may not hold the journal lock.
676 : */
677 342138 : int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
678 : {
679 342138 : int err = 0;
680 :
681 342138 : read_lock(&journal->j_state_lock);
682 : #ifdef CONFIG_PROVE_LOCKING
683 : /*
684 : * Some callers make sure transaction is already committing and in that
685 : * case we cannot block on open handles anymore. So don't warn in that
686 : * case.
687 : */
688 : if (tid_gt(tid, journal->j_commit_sequence) &&
689 : (!journal->j_committing_transaction ||
690 : journal->j_committing_transaction->t_tid != tid)) {
691 : read_unlock(&journal->j_state_lock);
692 : jbd2_might_wait_for_commit(journal);
693 : read_lock(&journal->j_state_lock);
694 : }
695 : #endif
696 : #ifdef CONFIG_JBD2_DEBUG
697 : if (!tid_geq(journal->j_commit_request, tid)) {
698 : printk(KERN_ERR
699 : "%s: error: j_commit_request=%u, tid=%u\n",
700 : __func__, journal->j_commit_request, tid);
701 : }
702 : #endif
703 684238 : while (tid_gt(tid, journal->j_commit_sequence)) {
704 342127 : jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
705 : tid, journal->j_commit_sequence);
706 342127 : read_unlock(&journal->j_state_lock);
707 342135 : wake_up(&journal->j_wait_commit);
708 804469 : wait_event(journal->j_wait_done_commit,
709 : !tid_gt(tid, journal->j_commit_sequence));
710 342226 : read_lock(&journal->j_state_lock);
711 : }
712 342127 : read_unlock(&journal->j_state_lock);
713 :
714 342017 : if (unlikely(is_journal_aborted(journal)))
715 3 : err = -EIO;
716 342017 : return err;
717 : }
718 :
719 : /*
720 : * Start a fast commit. If there's an ongoing fast or full commit wait for
721 : * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
722 : * if a fast commit is not needed, either because there's an already a commit
723 : * going on or this tid has already been committed. Returns -EINVAL if no jbd2
724 : * commit has yet been performed.
725 : */
726 0 : int jbd2_fc_begin_commit(journal_t *journal, tid_t tid)
727 : {
728 0 : if (unlikely(is_journal_aborted(journal)))
729 : return -EIO;
730 : /*
731 : * Fast commits only allowed if at least one full commit has
732 : * been processed.
733 : */
734 0 : if (!journal->j_stats.ts_tid)
735 : return -EINVAL;
736 :
737 0 : write_lock(&journal->j_state_lock);
738 0 : if (tid <= journal->j_commit_sequence) {
739 0 : write_unlock(&journal->j_state_lock);
740 0 : return -EALREADY;
741 : }
742 :
743 0 : if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
744 : (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) {
745 0 : DEFINE_WAIT(wait);
746 :
747 0 : prepare_to_wait(&journal->j_fc_wait, &wait,
748 : TASK_UNINTERRUPTIBLE);
749 0 : write_unlock(&journal->j_state_lock);
750 0 : schedule();
751 0 : finish_wait(&journal->j_fc_wait, &wait);
752 0 : return -EALREADY;
753 : }
754 0 : journal->j_flags |= JBD2_FAST_COMMIT_ONGOING;
755 0 : write_unlock(&journal->j_state_lock);
756 0 : jbd2_journal_lock_updates(journal);
757 :
758 0 : return 0;
759 : }
760 : EXPORT_SYMBOL(jbd2_fc_begin_commit);
761 :
762 : /*
763 : * Stop a fast commit. If fallback is set, this function starts commit of
764 : * TID tid before any other fast commit can start.
765 : */
766 0 : static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
767 : {
768 0 : jbd2_journal_unlock_updates(journal);
769 0 : if (journal->j_fc_cleanup_callback)
770 0 : journal->j_fc_cleanup_callback(journal, 0, tid);
771 0 : write_lock(&journal->j_state_lock);
772 0 : journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
773 0 : if (fallback)
774 0 : journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
775 0 : write_unlock(&journal->j_state_lock);
776 0 : wake_up(&journal->j_fc_wait);
777 0 : if (fallback)
778 0 : return jbd2_complete_transaction(journal, tid);
779 : return 0;
780 : }
781 :
782 0 : int jbd2_fc_end_commit(journal_t *journal)
783 : {
784 0 : return __jbd2_fc_end_commit(journal, 0, false);
785 : }
786 : EXPORT_SYMBOL(jbd2_fc_end_commit);
787 :
788 0 : int jbd2_fc_end_commit_fallback(journal_t *journal)
789 : {
790 0 : tid_t tid;
791 :
792 0 : read_lock(&journal->j_state_lock);
793 0 : tid = journal->j_running_transaction ?
794 0 : journal->j_running_transaction->t_tid : 0;
795 0 : read_unlock(&journal->j_state_lock);
796 0 : return __jbd2_fc_end_commit(journal, tid, true);
797 : }
798 : EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
799 :
800 : /* Return 1 when transaction with given tid has already committed. */
801 2538565 : int jbd2_transaction_committed(journal_t *journal, tid_t tid)
802 : {
803 2538565 : int ret = 1;
804 :
805 2538565 : read_lock(&journal->j_state_lock);
806 2538783 : if (journal->j_running_transaction &&
807 2394118 : journal->j_running_transaction->t_tid == tid)
808 2080199 : ret = 0;
809 2538783 : if (journal->j_committing_transaction &&
810 55235 : journal->j_committing_transaction->t_tid == tid)
811 6673 : ret = 0;
812 2538783 : read_unlock(&journal->j_state_lock);
813 2538901 : return ret;
814 : }
815 : EXPORT_SYMBOL(jbd2_transaction_committed);
816 :
817 : /*
818 : * When this function returns the transaction corresponding to tid
819 : * will be completed. If the transaction has currently running, start
820 : * committing that transaction before waiting for it to complete. If
821 : * the transaction id is stale, it is by definition already completed,
822 : * so just return SUCCESS.
823 : */
824 233827 : int jbd2_complete_transaction(journal_t *journal, tid_t tid)
825 : {
826 233827 : int need_to_wait = 1;
827 :
828 233827 : read_lock(&journal->j_state_lock);
829 233833 : if (journal->j_running_transaction &&
830 225768 : journal->j_running_transaction->t_tid == tid) {
831 179562 : if (journal->j_commit_request != tid) {
832 : /* transaction not yet started, so request it */
833 175507 : read_unlock(&journal->j_state_lock);
834 175507 : jbd2_log_start_commit(journal, tid);
835 175519 : goto wait_commit;
836 : }
837 54271 : } else if (!(journal->j_committing_transaction &&
838 9747 : journal->j_committing_transaction->t_tid == tid))
839 47867 : need_to_wait = 0;
840 58326 : read_unlock(&journal->j_state_lock);
841 58330 : if (!need_to_wait)
842 : return 0;
843 10461 : wait_commit:
844 185980 : return jbd2_log_wait_commit(journal, tid);
845 : }
846 : EXPORT_SYMBOL(jbd2_complete_transaction);
847 :
848 : /*
849 : * Log buffer allocation routines:
850 : */
851 :
852 3563895 : int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
853 : {
854 3563895 : unsigned long blocknr;
855 :
856 3563895 : write_lock(&journal->j_state_lock);
857 3563895 : J_ASSERT(journal->j_free > 1);
858 :
859 3563895 : blocknr = journal->j_head;
860 3563895 : journal->j_head++;
861 3563895 : journal->j_free--;
862 3563895 : if (journal->j_head == journal->j_last)
863 148 : journal->j_head = journal->j_first;
864 3563895 : write_unlock(&journal->j_state_lock);
865 3563895 : return jbd2_journal_bmap(journal, blocknr, retp);
866 : }
867 :
868 : /* Map one fast commit buffer for use by the file system */
869 0 : int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
870 : {
871 0 : unsigned long long pblock;
872 0 : unsigned long blocknr;
873 0 : int ret = 0;
874 0 : struct buffer_head *bh;
875 0 : int fc_off;
876 :
877 0 : *bh_out = NULL;
878 :
879 0 : if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) {
880 0 : fc_off = journal->j_fc_off;
881 0 : blocknr = journal->j_fc_first + fc_off;
882 0 : journal->j_fc_off++;
883 : } else {
884 : ret = -EINVAL;
885 : }
886 :
887 0 : if (ret)
888 : return ret;
889 :
890 0 : ret = jbd2_journal_bmap(journal, blocknr, &pblock);
891 0 : if (ret)
892 : return ret;
893 :
894 0 : bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
895 0 : if (!bh)
896 : return -ENOMEM;
897 :
898 :
899 0 : journal->j_fc_wbuf[fc_off] = bh;
900 :
901 0 : *bh_out = bh;
902 :
903 0 : return 0;
904 : }
905 : EXPORT_SYMBOL(jbd2_fc_get_buf);
906 :
907 : /*
908 : * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
909 : * for completion.
910 : */
911 0 : int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
912 : {
913 0 : struct buffer_head *bh;
914 0 : int i, j_fc_off;
915 :
916 0 : j_fc_off = journal->j_fc_off;
917 :
918 : /*
919 : * Wait in reverse order to minimize chances of us being woken up before
920 : * all IOs have completed
921 : */
922 0 : for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
923 0 : bh = journal->j_fc_wbuf[i];
924 0 : wait_on_buffer(bh);
925 : /*
926 : * Update j_fc_off so jbd2_fc_release_bufs can release remain
927 : * buffer head.
928 : */
929 0 : if (unlikely(!buffer_uptodate(bh))) {
930 0 : journal->j_fc_off = i + 1;
931 0 : return -EIO;
932 : }
933 0 : put_bh(bh);
934 0 : journal->j_fc_wbuf[i] = NULL;
935 : }
936 :
937 : return 0;
938 : }
939 : EXPORT_SYMBOL(jbd2_fc_wait_bufs);
940 :
941 0 : int jbd2_fc_release_bufs(journal_t *journal)
942 : {
943 0 : struct buffer_head *bh;
944 0 : int i, j_fc_off;
945 :
946 0 : j_fc_off = journal->j_fc_off;
947 :
948 0 : for (i = j_fc_off - 1; i >= 0; i--) {
949 0 : bh = journal->j_fc_wbuf[i];
950 0 : if (!bh)
951 : break;
952 0 : put_bh(bh);
953 0 : journal->j_fc_wbuf[i] = NULL;
954 : }
955 :
956 0 : return 0;
957 : }
958 : EXPORT_SYMBOL(jbd2_fc_release_bufs);
959 :
960 : /*
961 : * Conversion of logical to physical block numbers for the journal
962 : *
963 : * On external journals the journal blocks are identity-mapped, so
964 : * this is a no-op. If needed, we can use j_blk_offset - everything is
965 : * ready.
966 : */
967 3691533 : int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
968 : unsigned long long *retp)
969 : {
970 3691533 : int err = 0;
971 3691533 : unsigned long long ret;
972 3691533 : sector_t block = blocknr;
973 :
974 3691533 : if (journal->j_bmap) {
975 3691533 : err = journal->j_bmap(journal, &block);
976 3691533 : if (err == 0)
977 3691533 : *retp = block;
978 0 : } else if (journal->j_inode) {
979 0 : ret = bmap(journal->j_inode, &block);
980 :
981 0 : if (ret || !block) {
982 0 : printk(KERN_ALERT "%s: journal block not found "
983 : "at offset %lu on %s\n",
984 : __func__, blocknr, journal->j_devname);
985 0 : err = -EIO;
986 0 : jbd2_journal_abort(journal, err);
987 : } else {
988 0 : *retp = block;
989 : }
990 :
991 : } else {
992 0 : *retp = blocknr; /* +journal->j_blk_offset */
993 : }
994 3691533 : return err;
995 : }
996 :
997 : /*
998 : * We play buffer_head aliasing tricks to write data/metadata blocks to
999 : * the journal without copying their contents, but for journal
1000 : * descriptor blocks we do need to generate bona fide buffers.
1001 : *
1002 : * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
1003 : * the buffer's contents they really should run flush_dcache_page(bh->b_page).
1004 : * But we don't bother doing that, so there will be coherency problems with
1005 : * mmaps of blockdevs which hold live JBD-controlled filesystems.
1006 : */
1007 : struct buffer_head *
1008 442396 : jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
1009 : {
1010 442396 : journal_t *journal = transaction->t_journal;
1011 442396 : struct buffer_head *bh;
1012 442396 : unsigned long long blocknr;
1013 442396 : journal_header_t *header;
1014 442396 : int err;
1015 :
1016 442396 : err = jbd2_journal_next_log_block(journal, &blocknr);
1017 :
1018 442396 : if (err)
1019 : return NULL;
1020 :
1021 442396 : bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1022 442396 : if (!bh)
1023 : return NULL;
1024 442396 : atomic_dec(&transaction->t_outstanding_credits);
1025 442396 : lock_buffer(bh);
1026 442396 : memset(bh->b_data, 0, journal->j_blocksize);
1027 442396 : header = (journal_header_t *)bh->b_data;
1028 442396 : header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1029 442396 : header->h_blocktype = cpu_to_be32(type);
1030 442396 : header->h_sequence = cpu_to_be32(transaction->t_tid);
1031 442396 : set_buffer_uptodate(bh);
1032 442396 : unlock_buffer(bh);
1033 442396 : BUFFER_TRACE(bh, "return this buffer");
1034 442396 : return bh;
1035 : }
1036 :
1037 241102 : void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1038 : {
1039 241102 : struct jbd2_journal_block_tail *tail;
1040 241102 : __u32 csum;
1041 :
1042 241102 : if (!jbd2_journal_has_csum_v2or3(j))
1043 : return;
1044 :
1045 240596 : tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1046 : sizeof(struct jbd2_journal_block_tail));
1047 240596 : tail->t_checksum = 0;
1048 240596 : csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1049 240596 : tail->t_checksum = cpu_to_be32(csum);
1050 : }
1051 :
1052 : /*
1053 : * Return tid of the oldest transaction in the journal and block in the journal
1054 : * where the transaction starts.
1055 : *
1056 : * If the journal is now empty, return which will be the next transaction ID
1057 : * we will write and where will that transaction start.
1058 : *
1059 : * The return value is 0 if journal tail cannot be pushed any further, 1 if
1060 : * it can.
1061 : */
1062 245822 : int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1063 : unsigned long *block)
1064 : {
1065 245822 : transaction_t *transaction;
1066 245822 : int ret;
1067 :
1068 245822 : read_lock(&journal->j_state_lock);
1069 245822 : spin_lock(&journal->j_list_lock);
1070 245822 : transaction = journal->j_checkpoint_transactions;
1071 245822 : if (transaction) {
1072 157674 : *tid = transaction->t_tid;
1073 157674 : *block = transaction->t_log_start;
1074 88148 : } else if ((transaction = journal->j_committing_transaction) != NULL) {
1075 86170 : *tid = transaction->t_tid;
1076 86170 : *block = transaction->t_log_start;
1077 1978 : } else if ((transaction = journal->j_running_transaction) != NULL) {
1078 1 : *tid = transaction->t_tid;
1079 1 : *block = journal->j_head;
1080 : } else {
1081 1977 : *tid = journal->j_transaction_sequence;
1082 1977 : *block = journal->j_head;
1083 : }
1084 245822 : ret = tid_gt(*tid, journal->j_tail_sequence);
1085 245822 : spin_unlock(&journal->j_list_lock);
1086 245822 : read_unlock(&journal->j_state_lock);
1087 :
1088 245822 : return ret;
1089 : }
1090 :
1091 : /*
1092 : * Update information in journal structure and in on disk journal superblock
1093 : * about log tail. This function does not check whether information passed in
1094 : * really pushes log tail further. It's responsibility of the caller to make
1095 : * sure provided log tail information is valid (e.g. by holding
1096 : * j_checkpoint_mutex all the time between computing log tail and calling this
1097 : * function as is the case with jbd2_cleanup_journal_tail()).
1098 : *
1099 : * Requires j_checkpoint_mutex
1100 : */
1101 6141 : int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1102 : {
1103 6141 : unsigned long freed;
1104 6141 : int ret;
1105 :
1106 6141 : BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1107 :
1108 : /*
1109 : * We cannot afford for write to remain in drive's caches since as
1110 : * soon as we update j_tail, next transaction can start reusing journal
1111 : * space and if we lose sb update during power failure we'd replay
1112 : * old transaction with possibly newly overwritten data.
1113 : */
1114 6141 : ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
1115 : REQ_SYNC | REQ_FUA);
1116 6141 : if (ret)
1117 4 : goto out;
1118 :
1119 6137 : write_lock(&journal->j_state_lock);
1120 6137 : freed = block - journal->j_tail;
1121 6137 : if (block < journal->j_tail)
1122 148 : freed += journal->j_last - journal->j_first;
1123 :
1124 6137 : trace_jbd2_update_log_tail(journal, tid, block, freed);
1125 6137 : jbd2_debug(1,
1126 : "Cleaning journal tail from %u to %u (offset %lu), "
1127 : "freeing %lu\n",
1128 : journal->j_tail_sequence, tid, block, freed);
1129 :
1130 6137 : journal->j_free += freed;
1131 6137 : journal->j_tail_sequence = tid;
1132 6137 : journal->j_tail = block;
1133 6137 : write_unlock(&journal->j_state_lock);
1134 :
1135 6141 : out:
1136 6141 : return ret;
1137 : }
1138 :
1139 : /*
1140 : * This is a variation of __jbd2_update_log_tail which checks for validity of
1141 : * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1142 : * with other threads updating log tail.
1143 : */
1144 185 : void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1145 : {
1146 185 : mutex_lock_io(&journal->j_checkpoint_mutex);
1147 185 : if (tid_gt(tid, journal->j_tail_sequence))
1148 181 : __jbd2_update_log_tail(journal, tid, block);
1149 185 : mutex_unlock(&journal->j_checkpoint_mutex);
1150 185 : }
1151 :
1152 : struct jbd2_stats_proc_session {
1153 : journal_t *journal;
1154 : struct transaction_stats_s *stats;
1155 : int start;
1156 : int max;
1157 : };
1158 :
1159 0 : static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1160 : {
1161 0 : return *pos ? NULL : SEQ_START_TOKEN;
1162 : }
1163 :
1164 0 : static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1165 : {
1166 0 : (*pos)++;
1167 0 : return NULL;
1168 : }
1169 :
1170 0 : static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1171 : {
1172 0 : struct jbd2_stats_proc_session *s = seq->private;
1173 :
1174 0 : if (v != SEQ_START_TOKEN)
1175 : return 0;
1176 0 : seq_printf(seq, "%lu transactions (%lu requested), "
1177 : "each up to %u blocks\n",
1178 0 : s->stats->ts_tid, s->stats->ts_requested,
1179 0 : s->journal->j_max_transaction_buffers);
1180 0 : if (s->stats->ts_tid == 0)
1181 : return 0;
1182 0 : seq_printf(seq, "average: \n %ums waiting for transaction\n",
1183 0 : jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1184 0 : seq_printf(seq, " %ums request delay\n",
1185 0 : (s->stats->ts_requested == 0) ? 0 :
1186 0 : jiffies_to_msecs(s->stats->run.rs_request_delay /
1187 : s->stats->ts_requested));
1188 0 : seq_printf(seq, " %ums running transaction\n",
1189 0 : jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1190 0 : seq_printf(seq, " %ums transaction was being locked\n",
1191 0 : jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1192 0 : seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1193 0 : jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1194 0 : seq_printf(seq, " %ums logging transaction\n",
1195 0 : jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1196 0 : seq_printf(seq, " %lluus average transaction commit time\n",
1197 0 : div_u64(s->journal->j_average_commit_time, 1000));
1198 0 : seq_printf(seq, " %lu handles per transaction\n",
1199 0 : s->stats->run.rs_handle_count / s->stats->ts_tid);
1200 0 : seq_printf(seq, " %lu blocks per transaction\n",
1201 0 : s->stats->run.rs_blocks / s->stats->ts_tid);
1202 0 : seq_printf(seq, " %lu logged blocks per transaction\n",
1203 0 : s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1204 0 : return 0;
1205 : }
1206 :
1207 0 : static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1208 : {
1209 0 : }
1210 :
1211 : static const struct seq_operations jbd2_seq_info_ops = {
1212 : .start = jbd2_seq_info_start,
1213 : .next = jbd2_seq_info_next,
1214 : .stop = jbd2_seq_info_stop,
1215 : .show = jbd2_seq_info_show,
1216 : };
1217 :
1218 0 : static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1219 : {
1220 0 : journal_t *journal = pde_data(inode);
1221 0 : struct jbd2_stats_proc_session *s;
1222 0 : int rc, size;
1223 :
1224 0 : s = kmalloc(sizeof(*s), GFP_KERNEL);
1225 0 : if (s == NULL)
1226 : return -ENOMEM;
1227 0 : size = sizeof(struct transaction_stats_s);
1228 0 : s->stats = kmalloc(size, GFP_KERNEL);
1229 0 : if (s->stats == NULL) {
1230 0 : kfree(s);
1231 0 : return -ENOMEM;
1232 : }
1233 0 : spin_lock(&journal->j_history_lock);
1234 0 : memcpy(s->stats, &journal->j_stats, size);
1235 0 : s->journal = journal;
1236 0 : spin_unlock(&journal->j_history_lock);
1237 :
1238 0 : rc = seq_open(file, &jbd2_seq_info_ops);
1239 0 : if (rc == 0) {
1240 0 : struct seq_file *m = file->private_data;
1241 0 : m->private = s;
1242 : } else {
1243 0 : kfree(s->stats);
1244 0 : kfree(s);
1245 : }
1246 : return rc;
1247 :
1248 : }
1249 :
1250 0 : static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1251 : {
1252 0 : struct seq_file *seq = file->private_data;
1253 0 : struct jbd2_stats_proc_session *s = seq->private;
1254 0 : kfree(s->stats);
1255 0 : kfree(s);
1256 0 : return seq_release(inode, file);
1257 : }
1258 :
1259 : static const struct proc_ops jbd2_info_proc_ops = {
1260 : .proc_open = jbd2_seq_info_open,
1261 : .proc_read = seq_read,
1262 : .proc_lseek = seq_lseek,
1263 : .proc_release = jbd2_seq_info_release,
1264 : };
1265 :
1266 : static struct proc_dir_entry *proc_jbd2_stats;
1267 :
1268 2503 : static void jbd2_stats_proc_init(journal_t *journal)
1269 : {
1270 2503 : journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1271 2503 : if (journal->j_proc_entry) {
1272 2503 : proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1273 : &jbd2_info_proc_ops, journal);
1274 : }
1275 2503 : }
1276 :
1277 2503 : static void jbd2_stats_proc_exit(journal_t *journal)
1278 : {
1279 2503 : remove_proc_entry("info", journal->j_proc_entry);
1280 2503 : remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1281 2503 : }
1282 :
1283 : /* Minimum size of descriptor tag */
1284 : static int jbd2_min_tag_size(void)
1285 : {
1286 : /*
1287 : * Tag with 32-bit block numbers does not use last four bytes of the
1288 : * structure
1289 : */
1290 : return sizeof(journal_block_tag_t) - 4;
1291 : }
1292 :
1293 : /**
1294 : * jbd2_journal_shrink_scan()
1295 : * @shrink: shrinker to work on
1296 : * @sc: reclaim request to process
1297 : *
1298 : * Scan the checkpointed buffer on the checkpoint list and release the
1299 : * journal_head.
1300 : */
1301 52 : static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1302 : struct shrink_control *sc)
1303 : {
1304 52 : journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1305 52 : unsigned long nr_to_scan = sc->nr_to_scan;
1306 52 : unsigned long nr_shrunk;
1307 52 : unsigned long count;
1308 :
1309 52 : count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1310 52 : trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1311 :
1312 52 : nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1313 :
1314 52 : count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1315 52 : trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1316 :
1317 52 : return nr_shrunk;
1318 : }
1319 :
1320 : /**
1321 : * jbd2_journal_shrink_count()
1322 : * @shrink: shrinker to work on
1323 : * @sc: reclaim request to process
1324 : *
1325 : * Count the number of checkpoint buffers on the checkpoint list.
1326 : */
1327 523 : static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1328 : struct shrink_control *sc)
1329 : {
1330 523 : journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1331 523 : unsigned long count;
1332 :
1333 523 : count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1334 523 : trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1335 :
1336 523 : return count;
1337 : }
1338 :
1339 : /*
1340 : * Management for journal control blocks: functions to create and
1341 : * destroy journal_t structures, and to initialise and read existing
1342 : * journal blocks from disk. */
1343 :
1344 : /* First: create and setup a journal_t object in memory. We initialise
1345 : * very few fields yet: that has to wait until we have created the
1346 : * journal structures from from scratch, or loaded them from disk. */
1347 :
1348 2503 : static journal_t *journal_init_common(struct block_device *bdev,
1349 : struct block_device *fs_dev,
1350 : unsigned long long start, int len, int blocksize)
1351 : {
1352 2503 : static struct lock_class_key jbd2_trans_commit_key;
1353 2503 : journal_t *journal;
1354 2503 : int err;
1355 2503 : struct buffer_head *bh;
1356 2503 : int n;
1357 :
1358 2503 : journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1359 2503 : if (!journal)
1360 : return NULL;
1361 :
1362 2503 : init_waitqueue_head(&journal->j_wait_transaction_locked);
1363 2503 : init_waitqueue_head(&journal->j_wait_done_commit);
1364 2503 : init_waitqueue_head(&journal->j_wait_commit);
1365 2503 : init_waitqueue_head(&journal->j_wait_updates);
1366 2503 : init_waitqueue_head(&journal->j_wait_reserved);
1367 2503 : init_waitqueue_head(&journal->j_fc_wait);
1368 2503 : mutex_init(&journal->j_abort_mutex);
1369 2503 : mutex_init(&journal->j_barrier);
1370 2503 : mutex_init(&journal->j_checkpoint_mutex);
1371 2503 : spin_lock_init(&journal->j_revoke_lock);
1372 2503 : spin_lock_init(&journal->j_list_lock);
1373 2503 : rwlock_init(&journal->j_state_lock);
1374 :
1375 2503 : journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1376 2503 : journal->j_min_batch_time = 0;
1377 2503 : journal->j_max_batch_time = 15000; /* 15ms */
1378 2503 : atomic_set(&journal->j_reserved_credits, 0);
1379 :
1380 : /* The journal is marked for error until we succeed with recovery! */
1381 2503 : journal->j_flags = JBD2_ABORT;
1382 :
1383 : /* Set up a default-sized revoke table for the new mount. */
1384 2503 : err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1385 2503 : if (err)
1386 0 : goto err_cleanup;
1387 :
1388 2503 : spin_lock_init(&journal->j_history_lock);
1389 :
1390 2503 : lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1391 : &jbd2_trans_commit_key, 0);
1392 :
1393 : /* journal descriptor can store up to n blocks -bzzz */
1394 2503 : journal->j_blocksize = blocksize;
1395 2503 : journal->j_dev = bdev;
1396 2503 : journal->j_fs_dev = fs_dev;
1397 2503 : journal->j_blk_offset = start;
1398 2503 : journal->j_total_len = len;
1399 : /* We need enough buffers to write out full descriptor block. */
1400 2503 : n = journal->j_blocksize / jbd2_min_tag_size();
1401 2503 : journal->j_wbufsize = n;
1402 2503 : journal->j_fc_wbuf = NULL;
1403 2503 : journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1404 : GFP_KERNEL);
1405 2503 : if (!journal->j_wbuf)
1406 0 : goto err_cleanup;
1407 :
1408 2503 : bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1409 2503 : if (!bh) {
1410 0 : pr_err("%s: Cannot get buffer for journal superblock\n",
1411 : __func__);
1412 0 : goto err_cleanup;
1413 : }
1414 2503 : journal->j_sb_buffer = bh;
1415 2503 : journal->j_superblock = (journal_superblock_t *)bh->b_data;
1416 :
1417 2503 : journal->j_shrink_transaction = NULL;
1418 2503 : journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan;
1419 2503 : journal->j_shrinker.count_objects = jbd2_journal_shrink_count;
1420 2503 : journal->j_shrinker.seeks = DEFAULT_SEEKS;
1421 2503 : journal->j_shrinker.batch = journal->j_max_transaction_buffers;
1422 :
1423 2503 : if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL))
1424 0 : goto err_cleanup;
1425 :
1426 2503 : if (register_shrinker(&journal->j_shrinker, "jbd2-journal:(%u:%u)",
1427 2503 : MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev))) {
1428 0 : percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1429 0 : goto err_cleanup;
1430 : }
1431 : return journal;
1432 :
1433 0 : err_cleanup:
1434 0 : brelse(journal->j_sb_buffer);
1435 0 : kfree(journal->j_wbuf);
1436 0 : jbd2_journal_destroy_revoke(journal);
1437 0 : kfree(journal);
1438 0 : return NULL;
1439 : }
1440 :
1441 : /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1442 : *
1443 : * Create a journal structure assigned some fixed set of disk blocks to
1444 : * the journal. We don't actually touch those disk blocks yet, but we
1445 : * need to set up all of the mapping information to tell the journaling
1446 : * system where the journal blocks are.
1447 : *
1448 : */
1449 :
1450 : /**
1451 : * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1452 : * @bdev: Block device on which to create the journal
1453 : * @fs_dev: Device which hold journalled filesystem for this journal.
1454 : * @start: Block nr Start of journal.
1455 : * @len: Length of the journal in blocks.
1456 : * @blocksize: blocksize of journalling device
1457 : *
1458 : * Returns: a newly created journal_t *
1459 : *
1460 : * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1461 : * range of blocks on an arbitrary block device.
1462 : *
1463 : */
1464 12 : journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1465 : struct block_device *fs_dev,
1466 : unsigned long long start, int len, int blocksize)
1467 : {
1468 12 : journal_t *journal;
1469 :
1470 12 : journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1471 12 : if (!journal)
1472 : return NULL;
1473 :
1474 12 : snprintf(journal->j_devname, sizeof(journal->j_devname),
1475 : "%pg", journal->j_dev);
1476 12 : strreplace(journal->j_devname, '/', '!');
1477 12 : jbd2_stats_proc_init(journal);
1478 :
1479 12 : return journal;
1480 : }
1481 :
1482 : /**
1483 : * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1484 : * @inode: An inode to create the journal in
1485 : *
1486 : * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1487 : * the journal. The inode must exist already, must support bmap() and
1488 : * must have all data blocks preallocated.
1489 : */
1490 2491 : journal_t *jbd2_journal_init_inode(struct inode *inode)
1491 : {
1492 2491 : journal_t *journal;
1493 2491 : sector_t blocknr;
1494 2491 : int err = 0;
1495 :
1496 2491 : blocknr = 0;
1497 2491 : err = bmap(inode, &blocknr);
1498 :
1499 2491 : if (err || !blocknr) {
1500 0 : pr_err("%s: Cannot locate journal superblock\n",
1501 : __func__);
1502 0 : return NULL;
1503 : }
1504 :
1505 2491 : jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1506 : inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1507 : inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1508 :
1509 2491 : journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1510 2491 : blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1511 2491 : inode->i_sb->s_blocksize);
1512 2491 : if (!journal)
1513 : return NULL;
1514 :
1515 2491 : journal->j_inode = inode;
1516 2491 : snprintf(journal->j_devname, sizeof(journal->j_devname),
1517 : "%pg-%lu", journal->j_dev, journal->j_inode->i_ino);
1518 2491 : strreplace(journal->j_devname, '/', '!');
1519 2491 : jbd2_stats_proc_init(journal);
1520 :
1521 2491 : return journal;
1522 : }
1523 :
1524 : /*
1525 : * If the journal init or create aborts, we need to mark the journal
1526 : * superblock as being NULL to prevent the journal destroy from writing
1527 : * back a bogus superblock.
1528 : */
1529 : static void journal_fail_superblock(journal_t *journal)
1530 : {
1531 1 : struct buffer_head *bh = journal->j_sb_buffer;
1532 1 : brelse(bh);
1533 1 : journal->j_sb_buffer = NULL;
1534 : }
1535 :
1536 : /*
1537 : * Given a journal_t structure, initialise the various fields for
1538 : * startup of a new journaling session. We use this both when creating
1539 : * a journal, and after recovering an old journal to reset it for
1540 : * subsequent use.
1541 : */
1542 :
1543 2501 : static int journal_reset(journal_t *journal)
1544 : {
1545 2501 : journal_superblock_t *sb = journal->j_superblock;
1546 2501 : unsigned long long first, last;
1547 :
1548 2501 : first = be32_to_cpu(sb->s_first);
1549 2501 : last = be32_to_cpu(sb->s_maxlen);
1550 2501 : if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1551 0 : printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1552 : first, last);
1553 0 : journal_fail_superblock(journal);
1554 0 : return -EINVAL;
1555 : }
1556 :
1557 2501 : journal->j_first = first;
1558 2501 : journal->j_last = last;
1559 :
1560 2501 : if (journal->j_head != 0 && journal->j_flags & JBD2_CYCLE_RECORD) {
1561 : /*
1562 : * Disable the cycled recording mode if the journal head block
1563 : * number is not correct.
1564 : */
1565 1336 : if (journal->j_head < first || journal->j_head >= last) {
1566 0 : printk(KERN_WARNING "JBD2: Incorrect Journal head block %lu, "
1567 : "disable journal_cycle_record\n",
1568 : journal->j_head);
1569 0 : journal->j_head = journal->j_first;
1570 : }
1571 : } else {
1572 1165 : journal->j_head = journal->j_first;
1573 : }
1574 2501 : journal->j_tail = journal->j_head;
1575 2501 : journal->j_free = journal->j_last - journal->j_first;
1576 :
1577 2501 : journal->j_tail_sequence = journal->j_transaction_sequence;
1578 2501 : journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1579 2501 : journal->j_commit_request = journal->j_commit_sequence;
1580 :
1581 2501 : journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1582 :
1583 : /*
1584 : * Now that journal recovery is done, turn fast commits off here. This
1585 : * way, if fast commit was enabled before the crash but if now FS has
1586 : * disabled it, we don't enable fast commits.
1587 : */
1588 2501 : jbd2_clear_feature_fast_commit(journal);
1589 :
1590 : /*
1591 : * As a special case, if the on-disk copy is already marked as needing
1592 : * no recovery (s_start == 0), then we can safely defer the superblock
1593 : * update until the next commit by setting JBD2_FLUSHED. This avoids
1594 : * attempting a write to a potential-readonly device.
1595 : */
1596 2501 : if (sb->s_start == 0) {
1597 2253 : jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb "
1598 : "(start %ld, seq %u, errno %d)\n",
1599 : journal->j_tail, journal->j_tail_sequence,
1600 : journal->j_errno);
1601 2253 : journal->j_flags |= JBD2_FLUSHED;
1602 : } else {
1603 : /* Lock here to make assertions happy... */
1604 248 : mutex_lock_io(&journal->j_checkpoint_mutex);
1605 : /*
1606 : * Update log tail information. We use REQ_FUA since new
1607 : * transaction will start reusing journal space and so we
1608 : * must make sure information about current log tail is on
1609 : * disk before that.
1610 : */
1611 248 : jbd2_journal_update_sb_log_tail(journal,
1612 : journal->j_tail_sequence,
1613 : journal->j_tail,
1614 : REQ_SYNC | REQ_FUA);
1615 248 : mutex_unlock(&journal->j_checkpoint_mutex);
1616 : }
1617 2501 : return jbd2_journal_start_thread(journal);
1618 : }
1619 :
1620 : /*
1621 : * This function expects that the caller will have locked the journal
1622 : * buffer head, and will return with it unlocked
1623 : */
1624 9953 : static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags)
1625 : {
1626 9953 : struct buffer_head *bh = journal->j_sb_buffer;
1627 9953 : journal_superblock_t *sb = journal->j_superblock;
1628 9953 : int ret = 0;
1629 :
1630 : /* Buffer got discarded which means block device got invalidated */
1631 19906 : if (!buffer_mapped(bh)) {
1632 0 : unlock_buffer(bh);
1633 0 : return -EIO;
1634 : }
1635 :
1636 9953 : trace_jbd2_write_superblock(journal, write_flags);
1637 9953 : if (!(journal->j_flags & JBD2_BARRIER))
1638 0 : write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1639 19906 : if (buffer_write_io_error(bh)) {
1640 : /*
1641 : * Oh, dear. A previous attempt to write the journal
1642 : * superblock failed. This could happen because the
1643 : * USB device was yanked out. Or it could happen to
1644 : * be a transient write error and maybe the block will
1645 : * be remapped. Nothing we can do but to retry the
1646 : * write and hope for the best.
1647 : */
1648 0 : printk(KERN_ERR "JBD2: previous I/O error detected "
1649 : "for journal superblock update for %s.\n",
1650 : journal->j_devname);
1651 0 : clear_buffer_write_io_error(bh);
1652 0 : set_buffer_uptodate(bh);
1653 : }
1654 9953 : if (jbd2_journal_has_csum_v2or3(journal))
1655 9884 : sb->s_checksum = jbd2_superblock_csum(journal, sb);
1656 9953 : get_bh(bh);
1657 9953 : bh->b_end_io = end_buffer_write_sync;
1658 9953 : submit_bh(REQ_OP_WRITE | write_flags, bh);
1659 9953 : wait_on_buffer(bh);
1660 19906 : if (buffer_write_io_error(bh)) {
1661 11 : clear_buffer_write_io_error(bh);
1662 11 : set_buffer_uptodate(bh);
1663 11 : ret = -EIO;
1664 : }
1665 11 : if (ret) {
1666 11 : printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n",
1667 : journal->j_devname);
1668 11 : if (!is_journal_aborted(journal))
1669 5 : jbd2_journal_abort(journal, ret);
1670 : }
1671 :
1672 : return ret;
1673 : }
1674 :
1675 : /**
1676 : * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1677 : * @journal: The journal to update.
1678 : * @tail_tid: TID of the new transaction at the tail of the log
1679 : * @tail_block: The first block of the transaction at the tail of the log
1680 : * @write_flags: Flags for the journal sb write operation
1681 : *
1682 : * Update a journal's superblock information about log tail and write it to
1683 : * disk, waiting for the IO to complete.
1684 : */
1685 8032 : int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1686 : unsigned long tail_block,
1687 : blk_opf_t write_flags)
1688 : {
1689 8032 : journal_superblock_t *sb = journal->j_superblock;
1690 8032 : int ret;
1691 :
1692 8032 : if (is_journal_aborted(journal))
1693 : return -EIO;
1694 16062 : if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) {
1695 2 : jbd2_journal_abort(journal, -EIO);
1696 2 : return -EIO;
1697 : }
1698 :
1699 8029 : BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1700 8029 : jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1701 : tail_block, tail_tid);
1702 :
1703 8029 : lock_buffer(journal->j_sb_buffer);
1704 8029 : sb->s_sequence = cpu_to_be32(tail_tid);
1705 8029 : sb->s_start = cpu_to_be32(tail_block);
1706 :
1707 8029 : ret = jbd2_write_superblock(journal, write_flags);
1708 8029 : if (ret)
1709 5 : goto out;
1710 :
1711 : /* Log is no longer empty */
1712 8024 : write_lock(&journal->j_state_lock);
1713 8024 : WARN_ON(!sb->s_sequence);
1714 8024 : journal->j_flags &= ~JBD2_FLUSHED;
1715 8024 : write_unlock(&journal->j_state_lock);
1716 :
1717 : out:
1718 : return ret;
1719 : }
1720 :
1721 : /**
1722 : * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1723 : * @journal: The journal to update.
1724 : * @write_flags: Flags for the journal sb write operation
1725 : *
1726 : * Update a journal's dynamic superblock fields to show that journal is empty.
1727 : * Write updated superblock to disk waiting for IO to complete.
1728 : */
1729 2808 : static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags)
1730 : {
1731 2808 : journal_superblock_t *sb = journal->j_superblock;
1732 2808 : bool had_fast_commit = false;
1733 :
1734 2808 : BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1735 2808 : lock_buffer(journal->j_sb_buffer);
1736 2808 : if (sb->s_start == 0) { /* Is it already empty? */
1737 1039 : unlock_buffer(journal->j_sb_buffer);
1738 1039 : return;
1739 : }
1740 :
1741 1769 : jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1742 : journal->j_tail_sequence);
1743 :
1744 1769 : sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1745 1769 : sb->s_start = cpu_to_be32(0);
1746 1769 : sb->s_head = cpu_to_be32(journal->j_head);
1747 3538 : if (jbd2_has_feature_fast_commit(journal)) {
1748 : /*
1749 : * When journal is clean, no need to commit fast commit flag and
1750 : * make file system incompatible with older kernels.
1751 : */
1752 0 : jbd2_clear_feature_fast_commit(journal);
1753 0 : had_fast_commit = true;
1754 : }
1755 :
1756 1769 : jbd2_write_superblock(journal, write_flags);
1757 :
1758 1769 : if (had_fast_commit)
1759 0 : jbd2_set_feature_fast_commit(journal);
1760 :
1761 : /* Log is no longer empty */
1762 1769 : write_lock(&journal->j_state_lock);
1763 1769 : journal->j_flags |= JBD2_FLUSHED;
1764 1769 : write_unlock(&journal->j_state_lock);
1765 : }
1766 :
1767 : /**
1768 : * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1769 : * @journal: The journal to erase.
1770 : * @flags: A discard/zeroout request is sent for each physically contigous
1771 : * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1772 : * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1773 : * to perform.
1774 : *
1775 : * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1776 : * will be explicitly written if no hardware offload is available, see
1777 : * blkdev_issue_zeroout for more details.
1778 : */
1779 1 : static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1780 : {
1781 1 : int err = 0;
1782 1 : unsigned long block, log_offset; /* logical */
1783 1 : unsigned long long phys_block, block_start, block_stop; /* physical */
1784 1 : loff_t byte_start, byte_stop, byte_count;
1785 :
1786 : /* flags must be set to either discard or zeroout */
1787 1 : if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1788 1 : ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1789 : (flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1790 : return -EINVAL;
1791 :
1792 1 : if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1793 0 : !bdev_max_discard_sectors(journal->j_dev))
1794 : return -EOPNOTSUPP;
1795 :
1796 : /*
1797 : * lookup block mapping and issue discard/zeroout for each
1798 : * contiguous region
1799 : */
1800 1 : log_offset = be32_to_cpu(journal->j_superblock->s_first);
1801 1 : block_start = ~0ULL;
1802 1026 : for (block = log_offset; block < journal->j_total_len; block++) {
1803 1025 : err = jbd2_journal_bmap(journal, block, &phys_block);
1804 1025 : if (err) {
1805 0 : pr_err("JBD2: bad block at offset %lu", block);
1806 0 : return err;
1807 : }
1808 :
1809 1025 : if (block_start == ~0ULL) {
1810 3 : block_start = phys_block;
1811 3 : block_stop = block_start - 1;
1812 : }
1813 :
1814 : /*
1815 : * last block not contiguous with current block,
1816 : * process last contiguous region and return to this block on
1817 : * next loop
1818 : */
1819 1025 : if (phys_block != block_stop + 1) {
1820 2 : block--;
1821 : } else {
1822 1023 : block_stop++;
1823 : /*
1824 : * if this isn't the last block of journal,
1825 : * no need to process now because next block may also
1826 : * be part of this contiguous region
1827 : */
1828 1023 : if (block != journal->j_total_len - 1)
1829 1022 : continue;
1830 : }
1831 :
1832 : /*
1833 : * end of contiguous region or this is last block of journal,
1834 : * take care of the region
1835 : */
1836 3 : byte_start = block_start * journal->j_blocksize;
1837 3 : byte_stop = block_stop * journal->j_blocksize;
1838 3 : byte_count = (block_stop - block_start + 1) *
1839 : journal->j_blocksize;
1840 :
1841 3 : truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping,
1842 : byte_start, byte_stop);
1843 :
1844 3 : if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
1845 0 : err = blkdev_issue_discard(journal->j_dev,
1846 0 : byte_start >> SECTOR_SHIFT,
1847 0 : byte_count >> SECTOR_SHIFT,
1848 : GFP_NOFS);
1849 3 : } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
1850 3 : err = blkdev_issue_zeroout(journal->j_dev,
1851 3 : byte_start >> SECTOR_SHIFT,
1852 3 : byte_count >> SECTOR_SHIFT,
1853 : GFP_NOFS, 0);
1854 : }
1855 :
1856 3 : if (unlikely(err != 0)) {
1857 0 : pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
1858 : err, block_start, block_stop);
1859 0 : return err;
1860 : }
1861 :
1862 : /* reset start and stop after processing a region */
1863 : block_start = ~0ULL;
1864 : }
1865 :
1866 1 : return blkdev_issue_flush(journal->j_dev);
1867 : }
1868 :
1869 : /**
1870 : * jbd2_journal_update_sb_errno() - Update error in the journal.
1871 : * @journal: The journal to update.
1872 : *
1873 : * Update a journal's errno. Write updated superblock to disk waiting for IO
1874 : * to complete.
1875 : */
1876 155 : void jbd2_journal_update_sb_errno(journal_t *journal)
1877 : {
1878 155 : journal_superblock_t *sb = journal->j_superblock;
1879 155 : int errcode;
1880 :
1881 155 : lock_buffer(journal->j_sb_buffer);
1882 155 : errcode = journal->j_errno;
1883 155 : if (errcode == -ESHUTDOWN)
1884 148 : errcode = 0;
1885 155 : jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1886 155 : sb->s_errno = cpu_to_be32(errcode);
1887 :
1888 155 : jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1889 155 : }
1890 : EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1891 :
1892 13158 : static int journal_revoke_records_per_block(journal_t *journal)
1893 : {
1894 13158 : int record_size;
1895 13158 : int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1896 :
1897 26316 : if (jbd2_has_feature_64bit(journal))
1898 : record_size = 8;
1899 : else
1900 2705 : record_size = 4;
1901 :
1902 13158 : if (jbd2_journal_has_csum_v2or3(journal))
1903 8457 : space -= sizeof(struct jbd2_journal_block_tail);
1904 13158 : return space / record_size;
1905 : }
1906 :
1907 : /*
1908 : * Read the superblock for a given journal, performing initial
1909 : * validation of the format.
1910 : */
1911 285828 : static int journal_get_superblock(journal_t *journal)
1912 : {
1913 285828 : struct buffer_head *bh;
1914 285828 : journal_superblock_t *sb;
1915 285828 : int err;
1916 :
1917 285828 : bh = journal->j_sb_buffer;
1918 :
1919 285828 : J_ASSERT(bh != NULL);
1920 571656 : if (buffer_verified(bh))
1921 : return 0;
1922 :
1923 2502 : err = bh_read(bh, 0);
1924 2502 : if (err < 0) {
1925 0 : printk(KERN_ERR
1926 : "JBD2: IO error reading journal superblock\n");
1927 0 : goto out;
1928 : }
1929 :
1930 2502 : sb = journal->j_superblock;
1931 :
1932 2502 : err = -EINVAL;
1933 :
1934 2502 : if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1935 2501 : sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1936 1 : printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1937 1 : goto out;
1938 : }
1939 :
1940 2501 : if (be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V1 &&
1941 : be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V2) {
1942 0 : printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1943 0 : goto out;
1944 : }
1945 :
1946 2501 : if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1947 0 : printk(KERN_WARNING "JBD2: journal file too short\n");
1948 0 : goto out;
1949 : }
1950 :
1951 2501 : if (be32_to_cpu(sb->s_first) == 0 ||
1952 : be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1953 0 : printk(KERN_WARNING
1954 : "JBD2: Invalid start block of journal: %u\n",
1955 : be32_to_cpu(sb->s_first));
1956 0 : goto out;
1957 : }
1958 :
1959 5002 : if (jbd2_has_feature_csum2(journal) &&
1960 : jbd2_has_feature_csum3(journal)) {
1961 : /* Can't have checksum v2 and v3 at the same time! */
1962 0 : printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1963 : "at the same time!\n");
1964 0 : goto out;
1965 : }
1966 :
1967 3828 : if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1968 : jbd2_has_feature_checksum(journal)) {
1969 : /* Can't have checksum v1 and v2 on at the same time! */
1970 0 : printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1971 : "at the same time!\n");
1972 0 : goto out;
1973 : }
1974 :
1975 2501 : if (!jbd2_verify_csum_type(journal, sb)) {
1976 0 : printk(KERN_ERR "JBD2: Unknown checksum type\n");
1977 0 : goto out;
1978 : }
1979 :
1980 : /* Load the checksum driver */
1981 2501 : if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1982 1327 : journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1983 1327 : if (IS_ERR(journal->j_chksum_driver)) {
1984 0 : printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1985 0 : err = PTR_ERR(journal->j_chksum_driver);
1986 0 : journal->j_chksum_driver = NULL;
1987 0 : goto out;
1988 : }
1989 : /* Check superblock checksum */
1990 1327 : if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1991 0 : printk(KERN_ERR "JBD2: journal checksum error\n");
1992 0 : err = -EFSBADCRC;
1993 0 : goto out;
1994 : }
1995 : }
1996 2501 : set_buffer_verified(bh);
1997 : return 0;
1998 :
1999 1 : out:
2000 1 : journal_fail_superblock(journal);
2001 1 : return err;
2002 : }
2003 :
2004 : /*
2005 : * Load the on-disk journal superblock and read the key fields into the
2006 : * journal_t.
2007 : */
2008 :
2009 4753 : static int load_superblock(journal_t *journal)
2010 : {
2011 4753 : int err;
2012 4753 : journal_superblock_t *sb;
2013 4753 : int num_fc_blocks;
2014 :
2015 4753 : err = journal_get_superblock(journal);
2016 4753 : if (err)
2017 : return err;
2018 :
2019 4752 : sb = journal->j_superblock;
2020 :
2021 4752 : journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
2022 4752 : journal->j_tail = be32_to_cpu(sb->s_start);
2023 4752 : journal->j_first = be32_to_cpu(sb->s_first);
2024 4752 : journal->j_errno = be32_to_cpu(sb->s_errno);
2025 4752 : journal->j_last = be32_to_cpu(sb->s_maxlen);
2026 :
2027 4752 : if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
2028 0 : journal->j_total_len = be32_to_cpu(sb->s_maxlen);
2029 : /* Precompute checksum seed for all metadata */
2030 4752 : if (jbd2_journal_has_csum_v2or3(journal))
2031 2416 : journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2032 : sizeof(sb->s_uuid));
2033 9504 : journal->j_revoke_records_per_block =
2034 4752 : journal_revoke_records_per_block(journal);
2035 :
2036 9504 : if (jbd2_has_feature_fast_commit(journal)) {
2037 0 : journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
2038 0 : num_fc_blocks = jbd2_journal_get_num_fc_blks(sb);
2039 0 : if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
2040 0 : journal->j_last = journal->j_fc_last - num_fc_blocks;
2041 0 : journal->j_fc_first = journal->j_last + 1;
2042 0 : journal->j_fc_off = 0;
2043 : }
2044 :
2045 : return 0;
2046 : }
2047 :
2048 :
2049 : /**
2050 : * jbd2_journal_load() - Read journal from disk.
2051 : * @journal: Journal to act on.
2052 : *
2053 : * Given a journal_t structure which tells us which disk blocks contain
2054 : * a journal, read the journal from disk to initialise the in-memory
2055 : * structures.
2056 : */
2057 2501 : int jbd2_journal_load(journal_t *journal)
2058 : {
2059 2501 : int err;
2060 2501 : journal_superblock_t *sb;
2061 :
2062 2501 : err = load_superblock(journal);
2063 2501 : if (err)
2064 : return err;
2065 :
2066 2501 : sb = journal->j_superblock;
2067 :
2068 : /*
2069 : * If this is a V2 superblock, then we have to check the
2070 : * features flags on it.
2071 : */
2072 2501 : if (jbd2_format_support_feature(journal)) {
2073 2501 : if ((sb->s_feature_ro_compat &
2074 2501 : ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
2075 2501 : (sb->s_feature_incompat &
2076 : ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
2077 0 : printk(KERN_WARNING
2078 : "JBD2: Unrecognised features on journal\n");
2079 0 : return -EINVAL;
2080 : }
2081 : }
2082 :
2083 : /*
2084 : * Create a slab for this blocksize
2085 : */
2086 2501 : err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2087 2501 : if (err)
2088 : return err;
2089 :
2090 : /* Let the recovery code check whether it needs to recover any
2091 : * data from the journal. */
2092 2501 : if (jbd2_journal_recover(journal))
2093 0 : goto recovery_error;
2094 :
2095 2501 : if (journal->j_failed_commit) {
2096 0 : printk(KERN_ERR "JBD2: journal transaction %u on %s "
2097 : "is corrupt.\n", journal->j_failed_commit,
2098 : journal->j_devname);
2099 0 : return -EFSCORRUPTED;
2100 : }
2101 : /*
2102 : * clear JBD2_ABORT flag initialized in journal_init_common
2103 : * here to update log tail information with the newest seq.
2104 : */
2105 2501 : journal->j_flags &= ~JBD2_ABORT;
2106 :
2107 : /* OK, we've finished with the dynamic journal bits:
2108 : * reinitialise the dynamic contents of the superblock in memory
2109 : * and reset them on disk. */
2110 2501 : if (journal_reset(journal))
2111 0 : goto recovery_error;
2112 :
2113 2501 : journal->j_flags |= JBD2_LOADED;
2114 2501 : return 0;
2115 :
2116 0 : recovery_error:
2117 0 : printk(KERN_WARNING "JBD2: recovery failed\n");
2118 0 : return -EIO;
2119 : }
2120 :
2121 : /**
2122 : * jbd2_journal_destroy() - Release a journal_t structure.
2123 : * @journal: Journal to act on.
2124 : *
2125 : * Release a journal_t structure once it is no longer in use by the
2126 : * journaled object.
2127 : * Return <0 if we couldn't clean up the journal.
2128 : */
2129 2503 : int jbd2_journal_destroy(journal_t *journal)
2130 : {
2131 2503 : int err = 0;
2132 :
2133 : /* Wait for the commit thread to wake up and die. */
2134 2503 : journal_kill_thread(journal);
2135 :
2136 : /* Force a final log commit */
2137 2503 : if (journal->j_running_transaction)
2138 32 : jbd2_journal_commit_transaction(journal);
2139 :
2140 : /* Force any old transactions to disk */
2141 :
2142 : /* Totally anal locking here... */
2143 2503 : spin_lock(&journal->j_list_lock);
2144 4924 : while (journal->j_checkpoint_transactions != NULL) {
2145 2539 : spin_unlock(&journal->j_list_lock);
2146 2539 : mutex_lock_io(&journal->j_checkpoint_mutex);
2147 2539 : err = jbd2_log_do_checkpoint(journal);
2148 2539 : mutex_unlock(&journal->j_checkpoint_mutex);
2149 : /*
2150 : * If checkpointing failed, just free the buffers to avoid
2151 : * looping forever
2152 : */
2153 2539 : if (err) {
2154 118 : jbd2_journal_destroy_checkpoint(journal);
2155 118 : spin_lock(&journal->j_list_lock);
2156 : break;
2157 : }
2158 2421 : spin_lock(&journal->j_list_lock);
2159 : }
2160 :
2161 2503 : J_ASSERT(journal->j_running_transaction == NULL);
2162 2503 : J_ASSERT(journal->j_committing_transaction == NULL);
2163 2503 : J_ASSERT(journal->j_checkpoint_transactions == NULL);
2164 2503 : spin_unlock(&journal->j_list_lock);
2165 :
2166 : /*
2167 : * OK, all checkpoint transactions have been checked, now check the
2168 : * write out io error flag and abort the journal if some buffer failed
2169 : * to write back to the original location, otherwise the filesystem
2170 : * may become inconsistent.
2171 : */
2172 4849 : if (!is_journal_aborted(journal) &&
2173 2346 : test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags))
2174 0 : jbd2_journal_abort(journal, -EIO);
2175 :
2176 2503 : if (journal->j_sb_buffer) {
2177 2502 : if (!is_journal_aborted(journal)) {
2178 2346 : mutex_lock_io(&journal->j_checkpoint_mutex);
2179 :
2180 2346 : write_lock(&journal->j_state_lock);
2181 2346 : journal->j_tail_sequence =
2182 2346 : ++journal->j_transaction_sequence;
2183 2346 : write_unlock(&journal->j_state_lock);
2184 :
2185 2346 : jbd2_mark_journal_empty(journal,
2186 : REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
2187 2346 : mutex_unlock(&journal->j_checkpoint_mutex);
2188 : } else
2189 : err = -EIO;
2190 2502 : brelse(journal->j_sb_buffer);
2191 : }
2192 :
2193 2503 : if (journal->j_shrinker.flags & SHRINKER_REGISTERED) {
2194 2503 : percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2195 2503 : unregister_shrinker(&journal->j_shrinker);
2196 : }
2197 2503 : if (journal->j_proc_entry)
2198 2503 : jbd2_stats_proc_exit(journal);
2199 2503 : iput(journal->j_inode);
2200 2503 : if (journal->j_revoke)
2201 2503 : jbd2_journal_destroy_revoke(journal);
2202 2503 : if (journal->j_chksum_driver)
2203 2288 : crypto_free_shash(journal->j_chksum_driver);
2204 2503 : kfree(journal->j_fc_wbuf);
2205 2503 : kfree(journal->j_wbuf);
2206 2503 : kfree(journal);
2207 :
2208 2503 : return err;
2209 : }
2210 :
2211 :
2212 : /**
2213 : * jbd2_journal_check_used_features() - Check if features specified are used.
2214 : * @journal: Journal to check.
2215 : * @compat: bitmask of compatible features
2216 : * @ro: bitmask of features that force read-only mount
2217 : * @incompat: bitmask of incompatible features
2218 : *
2219 : * Check whether the journal uses all of a given set of
2220 : * features. Return true (non-zero) if it does.
2221 : **/
2222 :
2223 281085 : int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2224 : unsigned long ro, unsigned long incompat)
2225 : {
2226 281085 : journal_superblock_t *sb;
2227 :
2228 281085 : if (!compat && !ro && !incompat)
2229 : return 1;
2230 281063 : if (journal_get_superblock(journal))
2231 : return 0;
2232 281077 : if (!jbd2_format_support_feature(journal))
2233 : return 0;
2234 :
2235 281077 : sb = journal->j_superblock;
2236 :
2237 281077 : if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2238 281083 : ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2239 281082 : ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2240 277677 : return 1;
2241 :
2242 : return 0;
2243 : }
2244 :
2245 : /**
2246 : * jbd2_journal_check_available_features() - Check feature set in journalling layer
2247 : * @journal: Journal to check.
2248 : * @compat: bitmask of compatible features
2249 : * @ro: bitmask of features that force read-only mount
2250 : * @incompat: bitmask of incompatible features
2251 : *
2252 : * Check whether the journaling code supports the use of
2253 : * all of a given set of features on this journal. Return true
2254 : * (non-zero) if it can. */
2255 :
2256 5886 : int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2257 : unsigned long ro, unsigned long incompat)
2258 : {
2259 5886 : if (!compat && !ro && !incompat)
2260 : return 1;
2261 :
2262 5886 : if (!jbd2_format_support_feature(journal))
2263 : return 0;
2264 :
2265 5886 : if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2266 5886 : (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2267 5886 : (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2268 5886 : return 1;
2269 :
2270 : return 0;
2271 : }
2272 :
2273 : static int
2274 0 : jbd2_journal_initialize_fast_commit(journal_t *journal)
2275 : {
2276 0 : journal_superblock_t *sb = journal->j_superblock;
2277 0 : unsigned long long num_fc_blks;
2278 :
2279 0 : num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2280 0 : if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2281 : return -ENOSPC;
2282 :
2283 : /* Are we called twice? */
2284 0 : WARN_ON(journal->j_fc_wbuf != NULL);
2285 0 : journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2286 : sizeof(struct buffer_head *), GFP_KERNEL);
2287 0 : if (!journal->j_fc_wbuf)
2288 : return -ENOMEM;
2289 :
2290 0 : journal->j_fc_wbufsize = num_fc_blks;
2291 0 : journal->j_fc_last = journal->j_last;
2292 0 : journal->j_last = journal->j_fc_last - num_fc_blks;
2293 0 : journal->j_fc_first = journal->j_last + 1;
2294 0 : journal->j_fc_off = 0;
2295 0 : journal->j_free = journal->j_last - journal->j_first;
2296 0 : journal->j_max_transaction_buffers =
2297 : jbd2_journal_get_max_txn_bufs(journal);
2298 :
2299 0 : return 0;
2300 : }
2301 :
2302 : /**
2303 : * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2304 : * @journal: Journal to act on.
2305 : * @compat: bitmask of compatible features
2306 : * @ro: bitmask of features that force read-only mount
2307 : * @incompat: bitmask of incompatible features
2308 : *
2309 : * Mark a given journal feature as present on the
2310 : * superblock. Returns true if the requested features could be set.
2311 : *
2312 : */
2313 :
2314 281097 : int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2315 : unsigned long ro, unsigned long incompat)
2316 : {
2317 : #define INCOMPAT_FEATURE_ON(f) \
2318 : ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2319 : #define COMPAT_FEATURE_ON(f) \
2320 : ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2321 281097 : journal_superblock_t *sb;
2322 :
2323 281097 : if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2324 : return 1;
2325 :
2326 3405 : if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2327 : return 0;
2328 :
2329 : /* If enabling v2 checksums, turn on v3 instead */
2330 3405 : if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2331 0 : incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2332 0 : incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2333 : }
2334 :
2335 : /* Asking for checksumming v3 and v1? Only give them v3. */
2336 3405 : if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2337 2288 : compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2338 0 : compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2339 :
2340 3405 : jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2341 : compat, ro, incompat);
2342 :
2343 3405 : sb = journal->j_superblock;
2344 :
2345 3405 : if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2346 0 : if (jbd2_journal_initialize_fast_commit(journal)) {
2347 0 : pr_err("JBD2: Cannot enable fast commits.\n");
2348 0 : return 0;
2349 : }
2350 : }
2351 :
2352 : /* Load the checksum driver if necessary */
2353 3405 : if ((journal->j_chksum_driver == NULL) &&
2354 961 : INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2355 961 : journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2356 961 : if (IS_ERR(journal->j_chksum_driver)) {
2357 0 : printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2358 0 : journal->j_chksum_driver = NULL;
2359 0 : return 0;
2360 : }
2361 : /* Precompute checksum seed for all metadata */
2362 961 : journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2363 : sizeof(sb->s_uuid));
2364 : }
2365 :
2366 3405 : lock_buffer(journal->j_sb_buffer);
2367 :
2368 : /* If enabling v3 checksums, update superblock */
2369 3405 : if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2370 2288 : sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2371 2288 : sb->s_feature_compat &=
2372 : ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2373 : }
2374 :
2375 : /* If enabling v1 checksums, downgrade superblock */
2376 3405 : if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2377 0 : sb->s_feature_incompat &=
2378 : ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2379 : JBD2_FEATURE_INCOMPAT_CSUM_V3);
2380 :
2381 3405 : sb->s_feature_compat |= cpu_to_be32(compat);
2382 3405 : sb->s_feature_ro_compat |= cpu_to_be32(ro);
2383 3405 : sb->s_feature_incompat |= cpu_to_be32(incompat);
2384 3405 : unlock_buffer(journal->j_sb_buffer);
2385 6810 : journal->j_revoke_records_per_block =
2386 3405 : journal_revoke_records_per_block(journal);
2387 :
2388 3405 : return 1;
2389 : #undef COMPAT_FEATURE_ON
2390 : #undef INCOMPAT_FEATURE_ON
2391 : }
2392 :
2393 : /*
2394 : * jbd2_journal_clear_features() - Clear a given journal feature in the
2395 : * superblock
2396 : * @journal: Journal to act on.
2397 : * @compat: bitmask of compatible features
2398 : * @ro: bitmask of features that force read-only mount
2399 : * @incompat: bitmask of incompatible features
2400 : *
2401 : * Clear a given journal feature as present on the
2402 : * superblock.
2403 : */
2404 5001 : void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2405 : unsigned long ro, unsigned long incompat)
2406 : {
2407 5001 : journal_superblock_t *sb;
2408 :
2409 5001 : jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2410 : compat, ro, incompat);
2411 :
2412 5001 : sb = journal->j_superblock;
2413 :
2414 5001 : sb->s_feature_compat &= ~cpu_to_be32(compat);
2415 5001 : sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2416 5001 : sb->s_feature_incompat &= ~cpu_to_be32(incompat);
2417 10002 : journal->j_revoke_records_per_block =
2418 5001 : journal_revoke_records_per_block(journal);
2419 5001 : }
2420 : EXPORT_SYMBOL(jbd2_journal_clear_features);
2421 :
2422 : /**
2423 : * jbd2_journal_flush() - Flush journal
2424 : * @journal: Journal to act on.
2425 : * @flags: optional operation on the journal blocks after the flush (see below)
2426 : *
2427 : * Flush all data for a given journal to disk and empty the journal.
2428 : * Filesystems can use this when remounting readonly to ensure that
2429 : * recovery does not need to happen on remount. Optionally, a discard or zeroout
2430 : * can be issued on the journal blocks after flushing.
2431 : *
2432 : * flags:
2433 : * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2434 : * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2435 : */
2436 462 : int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2437 : {
2438 462 : int err = 0;
2439 462 : transaction_t *transaction = NULL;
2440 :
2441 462 : write_lock(&journal->j_state_lock);
2442 :
2443 : /* Force everything buffered to the log... */
2444 462 : if (journal->j_running_transaction) {
2445 54 : transaction = journal->j_running_transaction;
2446 54 : __jbd2_log_start_commit(journal, transaction->t_tid);
2447 408 : } else if (journal->j_committing_transaction)
2448 : transaction = journal->j_committing_transaction;
2449 :
2450 : /* Wait for the log commit to complete... */
2451 462 : if (transaction) {
2452 54 : tid_t tid = transaction->t_tid;
2453 :
2454 54 : write_unlock(&journal->j_state_lock);
2455 54 : jbd2_log_wait_commit(journal, tid);
2456 : } else {
2457 408 : write_unlock(&journal->j_state_lock);
2458 : }
2459 :
2460 : /* ...and flush everything in the log out to disk. */
2461 462 : spin_lock(&journal->j_list_lock);
2462 810 : while (!err && journal->j_checkpoint_transactions != NULL) {
2463 348 : spin_unlock(&journal->j_list_lock);
2464 348 : mutex_lock_io(&journal->j_checkpoint_mutex);
2465 348 : err = jbd2_log_do_checkpoint(journal);
2466 348 : mutex_unlock(&journal->j_checkpoint_mutex);
2467 348 : spin_lock(&journal->j_list_lock);
2468 : }
2469 462 : spin_unlock(&journal->j_list_lock);
2470 :
2471 462 : if (is_journal_aborted(journal))
2472 : return -EIO;
2473 :
2474 462 : mutex_lock_io(&journal->j_checkpoint_mutex);
2475 462 : if (!err) {
2476 462 : err = jbd2_cleanup_journal_tail(journal);
2477 462 : if (err < 0) {
2478 0 : mutex_unlock(&journal->j_checkpoint_mutex);
2479 0 : goto out;
2480 : }
2481 : err = 0;
2482 : }
2483 :
2484 : /* Finally, mark the journal as really needing no recovery.
2485 : * This sets s_start==0 in the underlying superblock, which is
2486 : * the magic code for a fully-recovered superblock. Any future
2487 : * commits of data to the journal will restore the current
2488 : * s_start value. */
2489 462 : jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2490 :
2491 462 : if (flags)
2492 1 : err = __jbd2_journal_erase(journal, flags);
2493 :
2494 462 : mutex_unlock(&journal->j_checkpoint_mutex);
2495 462 : write_lock(&journal->j_state_lock);
2496 462 : J_ASSERT(!journal->j_running_transaction);
2497 462 : J_ASSERT(!journal->j_committing_transaction);
2498 462 : J_ASSERT(!journal->j_checkpoint_transactions);
2499 462 : J_ASSERT(journal->j_head == journal->j_tail);
2500 462 : J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2501 462 : write_unlock(&journal->j_state_lock);
2502 : out:
2503 : return err;
2504 : }
2505 :
2506 : /**
2507 : * jbd2_journal_wipe() - Wipe journal contents
2508 : * @journal: Journal to act on.
2509 : * @write: flag (see below)
2510 : *
2511 : * Wipe out all of the contents of a journal, safely. This will produce
2512 : * a warning if the journal contains any valid recovery information.
2513 : * Must be called between journal_init_*() and jbd2_journal_load().
2514 : *
2515 : * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2516 : * we merely suppress recovery.
2517 : */
2518 :
2519 2252 : int jbd2_journal_wipe(journal_t *journal, int write)
2520 : {
2521 2252 : int err = 0;
2522 :
2523 2252 : J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2524 :
2525 2252 : err = load_superblock(journal);
2526 2252 : if (err)
2527 : return err;
2528 :
2529 2251 : if (!journal->j_tail)
2530 2251 : goto no_recovery;
2531 :
2532 0 : printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2533 : write ? "Clearing" : "Ignoring");
2534 :
2535 0 : err = jbd2_journal_skip_recovery(journal);
2536 0 : if (write) {
2537 : /* Lock to make assertions happy... */
2538 0 : mutex_lock_io(&journal->j_checkpoint_mutex);
2539 0 : jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2540 0 : mutex_unlock(&journal->j_checkpoint_mutex);
2541 : }
2542 :
2543 0 : no_recovery:
2544 : return err;
2545 : }
2546 :
2547 : /**
2548 : * jbd2_journal_abort () - Shutdown the journal immediately.
2549 : * @journal: the journal to shutdown.
2550 : * @errno: an error number to record in the journal indicating
2551 : * the reason for the shutdown.
2552 : *
2553 : * Perform a complete, immediate shutdown of the ENTIRE
2554 : * journal (not of a single transaction). This operation cannot be
2555 : * undone without closing and reopening the journal.
2556 : *
2557 : * The jbd2_journal_abort function is intended to support higher level error
2558 : * recovery mechanisms such as the ext2/ext3 remount-readonly error
2559 : * mode.
2560 : *
2561 : * Journal abort has very specific semantics. Any existing dirty,
2562 : * unjournaled buffers in the main filesystem will still be written to
2563 : * disk by bdflush, but the journaling mechanism will be suspended
2564 : * immediately and no further transaction commits will be honoured.
2565 : *
2566 : * Any dirty, journaled buffers will be written back to disk without
2567 : * hitting the journal. Atomicity cannot be guaranteed on an aborted
2568 : * filesystem, but we _do_ attempt to leave as much data as possible
2569 : * behind for fsck to use for cleanup.
2570 : *
2571 : * Any attempt to get a new transaction handle on a journal which is in
2572 : * ABORT state will just result in an -EROFS error return. A
2573 : * jbd2_journal_stop on an existing handle will return -EIO if we have
2574 : * entered abort state during the update.
2575 : *
2576 : * Recursive transactions are not disturbed by journal abort until the
2577 : * final jbd2_journal_stop, which will receive the -EIO error.
2578 : *
2579 : * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2580 : * which will be recorded (if possible) in the journal superblock. This
2581 : * allows a client to record failure conditions in the middle of a
2582 : * transaction without having to complete the transaction to record the
2583 : * failure to disk. ext3_error, for example, now uses this
2584 : * functionality.
2585 : *
2586 : */
2587 :
2588 156 : void jbd2_journal_abort(journal_t *journal, int errno)
2589 : {
2590 156 : transaction_t *transaction;
2591 :
2592 : /*
2593 : * Lock the aborting procedure until everything is done, this avoid
2594 : * races between filesystem's error handling flow (e.g. ext4_abort()),
2595 : * ensure panic after the error info is written into journal's
2596 : * superblock.
2597 : */
2598 156 : mutex_lock(&journal->j_abort_mutex);
2599 : /*
2600 : * ESHUTDOWN always takes precedence because a file system check
2601 : * caused by any other journal abort error is not required after
2602 : * a shutdown triggered.
2603 : */
2604 156 : write_lock(&journal->j_state_lock);
2605 156 : if (journal->j_flags & JBD2_ABORT) {
2606 1 : int old_errno = journal->j_errno;
2607 :
2608 1 : write_unlock(&journal->j_state_lock);
2609 1 : if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2610 0 : journal->j_errno = errno;
2611 0 : jbd2_journal_update_sb_errno(journal);
2612 : }
2613 1 : mutex_unlock(&journal->j_abort_mutex);
2614 1 : return;
2615 : }
2616 :
2617 : /*
2618 : * Mark the abort as occurred and start current running transaction
2619 : * to release all journaled buffer.
2620 : */
2621 155 : pr_err("Aborting journal on device %s.\n", journal->j_devname);
2622 :
2623 155 : journal->j_flags |= JBD2_ABORT;
2624 155 : journal->j_errno = errno;
2625 155 : transaction = journal->j_running_transaction;
2626 155 : if (transaction)
2627 13 : __jbd2_log_start_commit(journal, transaction->t_tid);
2628 155 : write_unlock(&journal->j_state_lock);
2629 :
2630 : /*
2631 : * Record errno to the journal super block, so that fsck and jbd2
2632 : * layer could realise that a filesystem check is needed.
2633 : */
2634 155 : jbd2_journal_update_sb_errno(journal);
2635 155 : mutex_unlock(&journal->j_abort_mutex);
2636 : }
2637 :
2638 : /**
2639 : * jbd2_journal_errno() - returns the journal's error state.
2640 : * @journal: journal to examine.
2641 : *
2642 : * This is the errno number set with jbd2_journal_abort(), the last
2643 : * time the journal was mounted - if the journal was stopped
2644 : * without calling abort this will be 0.
2645 : *
2646 : * If the journal has been aborted on this mount time -EROFS will
2647 : * be returned.
2648 : */
2649 2544 : int jbd2_journal_errno(journal_t *journal)
2650 : {
2651 2544 : int err;
2652 :
2653 2544 : read_lock(&journal->j_state_lock);
2654 2544 : if (journal->j_flags & JBD2_ABORT)
2655 : err = -EROFS;
2656 : else
2657 2544 : err = journal->j_errno;
2658 2544 : read_unlock(&journal->j_state_lock);
2659 2544 : return err;
2660 : }
2661 :
2662 : /**
2663 : * jbd2_journal_clear_err() - clears the journal's error state
2664 : * @journal: journal to act on.
2665 : *
2666 : * An error must be cleared or acked to take a FS out of readonly
2667 : * mode.
2668 : */
2669 0 : int jbd2_journal_clear_err(journal_t *journal)
2670 : {
2671 0 : int err = 0;
2672 :
2673 0 : write_lock(&journal->j_state_lock);
2674 0 : if (journal->j_flags & JBD2_ABORT)
2675 : err = -EROFS;
2676 : else
2677 0 : journal->j_errno = 0;
2678 0 : write_unlock(&journal->j_state_lock);
2679 0 : return err;
2680 : }
2681 :
2682 : /**
2683 : * jbd2_journal_ack_err() - Ack journal err.
2684 : * @journal: journal to act on.
2685 : *
2686 : * An error must be cleared or acked to take a FS out of readonly
2687 : * mode.
2688 : */
2689 0 : void jbd2_journal_ack_err(journal_t *journal)
2690 : {
2691 0 : write_lock(&journal->j_state_lock);
2692 0 : if (journal->j_errno)
2693 0 : journal->j_flags |= JBD2_ACK_ERR;
2694 0 : write_unlock(&journal->j_state_lock);
2695 0 : }
2696 :
2697 6556787 : int jbd2_journal_blocks_per_page(struct inode *inode)
2698 : {
2699 6556787 : return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2700 : }
2701 :
2702 : /*
2703 : * helper functions to deal with 32 or 64bit block numbers.
2704 : */
2705 409691 : size_t journal_tag_bytes(journal_t *journal)
2706 : {
2707 409691 : size_t sz;
2708 :
2709 819382 : if (jbd2_has_feature_csum3(journal))
2710 : return sizeof(journal_block_tag3_t);
2711 :
2712 1826 : sz = sizeof(journal_block_tag_t);
2713 :
2714 3652 : if (jbd2_has_feature_csum2(journal))
2715 0 : sz += sizeof(__u16);
2716 :
2717 3652 : if (jbd2_has_feature_64bit(journal))
2718 : return sz;
2719 : else
2720 100 : return sz - sizeof(__u32);
2721 : }
2722 :
2723 : /*
2724 : * JBD memory management
2725 : *
2726 : * These functions are used to allocate block-sized chunks of memory
2727 : * used for making copies of buffer_head data. Very often it will be
2728 : * page-sized chunks of data, but sometimes it will be in
2729 : * sub-page-size chunks. (For example, 16k pages on Power systems
2730 : * with a 4k block file system.) For blocks smaller than a page, we
2731 : * use a SLAB allocator. There are slab caches for each block size,
2732 : * which are allocated at mount time, if necessary, and we only free
2733 : * (all of) the slab caches when/if the jbd2 module is unloaded. For
2734 : * this reason we don't need to a mutex to protect access to
2735 : * jbd2_slab[] allocating or releasing memory; only in
2736 : * jbd2_journal_create_slab().
2737 : */
2738 : #define JBD2_MAX_SLABS 8
2739 : static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2740 :
2741 : static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2742 : "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2743 : "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2744 : };
2745 :
2746 :
2747 0 : static void jbd2_journal_destroy_slabs(void)
2748 : {
2749 0 : int i;
2750 :
2751 0 : for (i = 0; i < JBD2_MAX_SLABS; i++) {
2752 0 : kmem_cache_destroy(jbd2_slab[i]);
2753 0 : jbd2_slab[i] = NULL;
2754 : }
2755 0 : }
2756 :
2757 2501 : static int jbd2_journal_create_slab(size_t size)
2758 : {
2759 2501 : static DEFINE_MUTEX(jbd2_slab_create_mutex);
2760 2501 : int i = order_base_2(size) - 10;
2761 2501 : size_t slab_size;
2762 :
2763 2501 : if (size == PAGE_SIZE)
2764 : return 0;
2765 :
2766 7 : if (i >= JBD2_MAX_SLABS)
2767 : return -EINVAL;
2768 :
2769 7 : if (unlikely(i < 0))
2770 0 : i = 0;
2771 7 : mutex_lock(&jbd2_slab_create_mutex);
2772 7 : if (jbd2_slab[i]) {
2773 5 : mutex_unlock(&jbd2_slab_create_mutex);
2774 5 : return 0; /* Already created */
2775 : }
2776 :
2777 2 : slab_size = 1 << (i+10);
2778 2 : jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2779 : slab_size, 0, NULL);
2780 2 : mutex_unlock(&jbd2_slab_create_mutex);
2781 2 : if (!jbd2_slab[i]) {
2782 0 : printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2783 0 : return -ENOMEM;
2784 : }
2785 : return 0;
2786 : }
2787 :
2788 0 : static struct kmem_cache *get_slab(size_t size)
2789 : {
2790 0 : int i = order_base_2(size) - 10;
2791 :
2792 0 : BUG_ON(i >= JBD2_MAX_SLABS);
2793 0 : if (unlikely(i < 0))
2794 0 : i = 0;
2795 0 : BUG_ON(jbd2_slab[i] == NULL);
2796 0 : return jbd2_slab[i];
2797 : }
2798 :
2799 21938 : void *jbd2_alloc(size_t size, gfp_t flags)
2800 : {
2801 21938 : void *ptr;
2802 :
2803 21938 : BUG_ON(size & (size-1)); /* Must be a power of 2 */
2804 :
2805 21938 : if (size < PAGE_SIZE)
2806 0 : ptr = kmem_cache_alloc(get_slab(size), flags);
2807 : else
2808 43876 : ptr = (void *)__get_free_pages(flags, get_order(size));
2809 :
2810 : /* Check alignment; SLUB has gotten this wrong in the past,
2811 : * and this can lead to user data corruption! */
2812 21913 : BUG_ON(((unsigned long) ptr) & (size-1));
2813 :
2814 21913 : return ptr;
2815 : }
2816 :
2817 21986 : void jbd2_free(void *ptr, size_t size)
2818 : {
2819 21986 : if (size < PAGE_SIZE)
2820 0 : kmem_cache_free(get_slab(size), ptr);
2821 : else
2822 43972 : free_pages((unsigned long)ptr, get_order(size));
2823 21986 : };
2824 :
2825 : /*
2826 : * Journal_head storage management
2827 : */
2828 : static struct kmem_cache *jbd2_journal_head_cache;
2829 : #ifdef CONFIG_JBD2_DEBUG
2830 : static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2831 : #endif
2832 :
2833 12 : static int __init jbd2_journal_init_journal_head_cache(void)
2834 : {
2835 12 : J_ASSERT(!jbd2_journal_head_cache);
2836 12 : jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2837 : sizeof(struct journal_head),
2838 : 0, /* offset */
2839 : SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2840 : NULL); /* ctor */
2841 12 : if (!jbd2_journal_head_cache) {
2842 0 : printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2843 0 : return -ENOMEM;
2844 : }
2845 : return 0;
2846 : }
2847 :
2848 : static void jbd2_journal_destroy_journal_head_cache(void)
2849 : {
2850 0 : kmem_cache_destroy(jbd2_journal_head_cache);
2851 0 : jbd2_journal_head_cache = NULL;
2852 : }
2853 :
2854 : /*
2855 : * journal_head splicing and dicing
2856 : */
2857 1461167 : static struct journal_head *journal_alloc_journal_head(void)
2858 : {
2859 1461167 : struct journal_head *ret;
2860 :
2861 : #ifdef CONFIG_JBD2_DEBUG
2862 : atomic_inc(&nr_journal_heads);
2863 : #endif
2864 1461167 : ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2865 1461780 : if (!ret) {
2866 0 : jbd2_debug(1, "out of memory for journal_head\n");
2867 0 : pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2868 0 : ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2869 : GFP_NOFS | __GFP_NOFAIL);
2870 : }
2871 1461780 : if (ret)
2872 1461780 : spin_lock_init(&ret->b_state_lock);
2873 1461756 : return ret;
2874 : }
2875 :
2876 : static void journal_free_journal_head(struct journal_head *jh)
2877 : {
2878 : #ifdef CONFIG_JBD2_DEBUG
2879 : atomic_dec(&nr_journal_heads);
2880 : memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2881 : #endif
2882 1462308 : kmem_cache_free(jbd2_journal_head_cache, jh);
2883 58 : }
2884 :
2885 : /*
2886 : * A journal_head is attached to a buffer_head whenever JBD has an
2887 : * interest in the buffer.
2888 : *
2889 : * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2890 : * is set. This bit is tested in core kernel code where we need to take
2891 : * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2892 : * there.
2893 : *
2894 : * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2895 : *
2896 : * When a buffer has its BH_JBD bit set it is immune from being released by
2897 : * core kernel code, mainly via ->b_count.
2898 : *
2899 : * A journal_head is detached from its buffer_head when the journal_head's
2900 : * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2901 : * transaction (b_cp_transaction) hold their references to b_jcount.
2902 : *
2903 : * Various places in the kernel want to attach a journal_head to a buffer_head
2904 : * _before_ attaching the journal_head to a transaction. To protect the
2905 : * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2906 : * journal_head's b_jcount refcount by one. The caller must call
2907 : * jbd2_journal_put_journal_head() to undo this.
2908 : *
2909 : * So the typical usage would be:
2910 : *
2911 : * (Attach a journal_head if needed. Increments b_jcount)
2912 : * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2913 : * ...
2914 : * (Get another reference for transaction)
2915 : * jbd2_journal_grab_journal_head(bh);
2916 : * jh->b_transaction = xxx;
2917 : * (Put original reference)
2918 : * jbd2_journal_put_journal_head(jh);
2919 : */
2920 :
2921 : /*
2922 : * Give a buffer_head a journal_head.
2923 : *
2924 : * May sleep.
2925 : */
2926 3321034 : struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2927 : {
2928 3321034 : struct journal_head *jh;
2929 3321034 : struct journal_head *new_jh = NULL;
2930 :
2931 3320634 : repeat:
2932 6641268 : if (!buffer_jbd(bh))
2933 1461204 : new_jh = journal_alloc_journal_head();
2934 :
2935 3321175 : jbd_lock_bh_journal_head(bh);
2936 6644050 : if (buffer_jbd(bh)) {
2937 1860429 : jh = bh2jh(bh);
2938 : } else {
2939 1461596 : J_ASSERT_BH(bh,
2940 : (atomic_read(&bh->b_count) > 0) ||
2941 : (bh->b_folio && bh->b_folio->mapping));
2942 :
2943 1461596 : if (!new_jh) {
2944 0 : jbd_unlock_bh_journal_head(bh);
2945 0 : goto repeat;
2946 : }
2947 :
2948 1461716 : jh = new_jh;
2949 1461716 : new_jh = NULL; /* We consumed it */
2950 1461716 : set_buffer_jbd(bh);
2951 1461877 : bh->b_private = jh;
2952 1461877 : jh->b_bh = bh;
2953 1461877 : get_bh(bh);
2954 3322353 : BUFFER_TRACE(bh, "added journal_head");
2955 : }
2956 3322353 : jh->b_jcount++;
2957 3322353 : jbd_unlock_bh_journal_head(bh);
2958 3322308 : if (new_jh)
2959 58 : journal_free_journal_head(new_jh);
2960 3322308 : return bh->b_private;
2961 : }
2962 :
2963 : /*
2964 : * Grab a ref against this buffer_head's journal_head. If it ended up not
2965 : * having a journal_head, return NULL
2966 : */
2967 40052669 : struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2968 : {
2969 40052669 : struct journal_head *jh = NULL;
2970 :
2971 40052669 : jbd_lock_bh_journal_head(bh);
2972 80111362 : if (buffer_jbd(bh)) {
2973 6388254 : jh = bh2jh(bh);
2974 6388254 : jh->b_jcount++;
2975 : }
2976 40055681 : jbd_unlock_bh_journal_head(bh);
2977 40054810 : return jh;
2978 : }
2979 : EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2980 :
2981 1462253 : static void __journal_remove_journal_head(struct buffer_head *bh)
2982 : {
2983 1462253 : struct journal_head *jh = bh2jh(bh);
2984 :
2985 1462253 : J_ASSERT_JH(jh, jh->b_transaction == NULL);
2986 1462253 : J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2987 1462253 : J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2988 1462253 : J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2989 2924506 : J_ASSERT_BH(bh, buffer_jbd(bh));
2990 1462253 : J_ASSERT_BH(bh, jh2bh(jh) == bh);
2991 1462253 : BUFFER_TRACE(bh, "remove journal_head");
2992 :
2993 : /* Unlink before dropping the lock */
2994 1462253 : bh->b_private = NULL;
2995 1462253 : jh->b_bh = NULL; /* debug, really */
2996 1462253 : clear_buffer_jbd(bh);
2997 1462251 : }
2998 :
2999 1462250 : static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
3000 : {
3001 1462250 : if (jh->b_frozen_data) {
3002 0 : printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
3003 0 : jbd2_free(jh->b_frozen_data, b_size);
3004 : }
3005 1462250 : if (jh->b_committed_data) {
3006 0 : printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
3007 0 : jbd2_free(jh->b_committed_data, b_size);
3008 : }
3009 1462250 : journal_free_journal_head(jh);
3010 1462249 : }
3011 :
3012 : /*
3013 : * Drop a reference on the passed journal_head. If it fell to zero then
3014 : * release the journal_head from the buffer_head.
3015 : */
3016 9710530 : void jbd2_journal_put_journal_head(struct journal_head *jh)
3017 : {
3018 9710530 : struct buffer_head *bh = jh2bh(jh);
3019 :
3020 9710530 : jbd_lock_bh_journal_head(bh);
3021 9710777 : J_ASSERT_JH(jh, jh->b_jcount > 0);
3022 9710777 : --jh->b_jcount;
3023 9710777 : if (!jh->b_jcount) {
3024 1462253 : __journal_remove_journal_head(bh);
3025 1462251 : jbd_unlock_bh_journal_head(bh);
3026 1462250 : journal_release_journal_head(jh, bh->b_size);
3027 1462250 : __brelse(bh);
3028 : } else {
3029 8248524 : jbd_unlock_bh_journal_head(bh);
3030 : }
3031 9710673 : }
3032 : EXPORT_SYMBOL(jbd2_journal_put_journal_head);
3033 :
3034 : /*
3035 : * Initialize jbd inode head
3036 : */
3037 1999264 : void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3038 : {
3039 1999264 : jinode->i_transaction = NULL;
3040 1999264 : jinode->i_next_transaction = NULL;
3041 1999264 : jinode->i_vfs_inode = inode;
3042 1999264 : jinode->i_flags = 0;
3043 1999264 : jinode->i_dirty_start = 0;
3044 1999264 : jinode->i_dirty_end = 0;
3045 1999264 : INIT_LIST_HEAD(&jinode->i_list);
3046 1999264 : }
3047 :
3048 : /*
3049 : * Function to be called before we start removing inode from memory (i.e.,
3050 : * clear_inode() is a fine place to be called from). It removes inode from
3051 : * transaction's lists.
3052 : */
3053 2005071 : void jbd2_journal_release_jbd_inode(journal_t *journal,
3054 : struct jbd2_inode *jinode)
3055 : {
3056 2005071 : if (!journal)
3057 : return;
3058 2005071 : restart:
3059 2005071 : spin_lock(&journal->j_list_lock);
3060 : /* Is commit writing out inode - we have to wait */
3061 2005506 : if (jinode->i_flags & JI_COMMIT_RUNNING) {
3062 0 : wait_queue_head_t *wq;
3063 0 : DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3064 0 : wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3065 0 : prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3066 0 : spin_unlock(&journal->j_list_lock);
3067 0 : schedule();
3068 0 : finish_wait(wq, &wait.wq_entry);
3069 0 : goto restart;
3070 : }
3071 :
3072 2005506 : if (jinode->i_transaction) {
3073 192 : list_del(&jinode->i_list);
3074 192 : jinode->i_transaction = NULL;
3075 : }
3076 2005506 : spin_unlock(&journal->j_list_lock);
3077 : }
3078 :
3079 :
3080 : #ifdef CONFIG_PROC_FS
3081 :
3082 : #define JBD2_STATS_PROC_NAME "fs/jbd2"
3083 :
3084 12 : static void __init jbd2_create_jbd_stats_proc_entry(void)
3085 : {
3086 12 : proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3087 12 : }
3088 :
3089 0 : static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3090 : {
3091 0 : if (proc_jbd2_stats)
3092 0 : remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3093 0 : }
3094 :
3095 : #else
3096 :
3097 : #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3098 : #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3099 :
3100 : #endif
3101 :
3102 : struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3103 :
3104 12 : static int __init jbd2_journal_init_inode_cache(void)
3105 : {
3106 12 : J_ASSERT(!jbd2_inode_cache);
3107 12 : jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3108 12 : if (!jbd2_inode_cache) {
3109 0 : pr_emerg("JBD2: failed to create inode cache\n");
3110 0 : return -ENOMEM;
3111 : }
3112 : return 0;
3113 : }
3114 :
3115 12 : static int __init jbd2_journal_init_handle_cache(void)
3116 : {
3117 12 : J_ASSERT(!jbd2_handle_cache);
3118 12 : jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3119 12 : if (!jbd2_handle_cache) {
3120 0 : printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3121 0 : return -ENOMEM;
3122 : }
3123 : return 0;
3124 : }
3125 :
3126 : static void jbd2_journal_destroy_inode_cache(void)
3127 : {
3128 0 : kmem_cache_destroy(jbd2_inode_cache);
3129 0 : jbd2_inode_cache = NULL;
3130 : }
3131 :
3132 : static void jbd2_journal_destroy_handle_cache(void)
3133 : {
3134 0 : kmem_cache_destroy(jbd2_handle_cache);
3135 0 : jbd2_handle_cache = NULL;
3136 : }
3137 :
3138 : /*
3139 : * Module startup and shutdown
3140 : */
3141 :
3142 12 : static int __init journal_init_caches(void)
3143 : {
3144 12 : int ret;
3145 :
3146 12 : ret = jbd2_journal_init_revoke_record_cache();
3147 12 : if (ret == 0)
3148 12 : ret = jbd2_journal_init_revoke_table_cache();
3149 12 : if (ret == 0)
3150 12 : ret = jbd2_journal_init_journal_head_cache();
3151 12 : if (ret == 0)
3152 12 : ret = jbd2_journal_init_handle_cache();
3153 12 : if (ret == 0)
3154 12 : ret = jbd2_journal_init_inode_cache();
3155 12 : if (ret == 0)
3156 12 : ret = jbd2_journal_init_transaction_cache();
3157 12 : return ret;
3158 : }
3159 :
3160 0 : static void jbd2_journal_destroy_caches(void)
3161 : {
3162 0 : jbd2_journal_destroy_revoke_record_cache();
3163 0 : jbd2_journal_destroy_revoke_table_cache();
3164 0 : jbd2_journal_destroy_journal_head_cache();
3165 0 : jbd2_journal_destroy_handle_cache();
3166 0 : jbd2_journal_destroy_inode_cache();
3167 0 : jbd2_journal_destroy_transaction_cache();
3168 0 : jbd2_journal_destroy_slabs();
3169 0 : }
3170 :
3171 12 : static int __init journal_init(void)
3172 : {
3173 12 : int ret;
3174 :
3175 12 : BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3176 :
3177 12 : ret = journal_init_caches();
3178 12 : if (ret == 0) {
3179 12 : jbd2_create_jbd_stats_proc_entry();
3180 : } else {
3181 0 : jbd2_journal_destroy_caches();
3182 : }
3183 12 : return ret;
3184 : }
3185 :
3186 0 : static void __exit journal_exit(void)
3187 : {
3188 : #ifdef CONFIG_JBD2_DEBUG
3189 : int n = atomic_read(&nr_journal_heads);
3190 : if (n)
3191 : printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3192 : #endif
3193 0 : jbd2_remove_jbd_stats_proc_entry();
3194 0 : jbd2_journal_destroy_caches();
3195 0 : }
3196 :
3197 : MODULE_LICENSE("GPL");
3198 : module_init(journal_init);
3199 : module_exit(journal_exit);
3200 :
|
