blob: 89463eee67914643a02ce711463e62fac0b83c8d [file] [log] [blame]
Dave Kleikamp470decc2006-10-11 01:20:57 -07001/*
Uwe Kleine-König58862692007-05-09 07:51:49 +02002 * linux/fs/jbd2/transaction.c
Dave Kleikamp470decc2006-10-11 01:20:57 -07003 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5 *
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Generic filesystem transaction handling code; part of the ext2fs
13 * journaling system.
14 *
15 * This file manages transactions (compound commits managed by the
16 * journaling code) and handles (individual atomic operations by the
17 * filesystem).
18 */
19
20#include <linux/time.h>
21#include <linux/fs.h>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070022#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070023#include <linux/errno.h>
24#include <linux/slab.h>
25#include <linux/timer.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070026#include <linux/mm.h>
27#include <linux/highmem.h>
Josef Bacike07f7182008-11-26 01:14:26 -050028#include <linux/hrtimer.h>
Theodore Ts'o47def822010-07-27 11:56:05 -040029#include <linux/backing-dev.h>
Randy Dunlap44705752011-10-27 04:05:13 -040030#include <linux/bug.h>
Theodore Ts'o47def822010-07-27 11:56:05 -040031#include <linux/module.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070032
Theodore Ts'o343d9c22013-02-08 13:00:22 -050033#include <trace/events/jbd2.h>
34
Adrian Bunk7ddae862006-12-06 20:38:27 -080035static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
Jan Karade1b7942011-06-13 15:38:22 -040036static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
Adrian Bunk7ddae862006-12-06 20:38:27 -080037
Yongqiang Yang0c2022e2012-02-20 17:53:02 -050038static struct kmem_cache *transaction_cache;
39int __init jbd2_journal_init_transaction_cache(void)
40{
41 J_ASSERT(!transaction_cache);
42 transaction_cache = kmem_cache_create("jbd2_transaction_s",
43 sizeof(transaction_t),
44 0,
45 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
46 NULL);
47 if (transaction_cache)
48 return 0;
49 return -ENOMEM;
50}
51
52void jbd2_journal_destroy_transaction_cache(void)
53{
54 if (transaction_cache) {
55 kmem_cache_destroy(transaction_cache);
56 transaction_cache = NULL;
57 }
58}
59
60void jbd2_journal_free_transaction(transaction_t *transaction)
61{
62 if (unlikely(ZERO_OR_NULL_PTR(transaction)))
63 return;
64 kmem_cache_free(transaction_cache, transaction);
65}
66
Dave Kleikamp470decc2006-10-11 01:20:57 -070067/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -070068 * jbd2_get_transaction: obtain a new transaction_t object.
Dave Kleikamp470decc2006-10-11 01:20:57 -070069 *
70 * Simply allocate and initialise a new transaction. Create it in
71 * RUNNING state and add it to the current journal (which should not
72 * have an existing running transaction: we only make a new transaction
73 * once we have started to commit the old one).
74 *
75 * Preconditions:
76 * The journal MUST be locked. We don't perform atomic mallocs on the
77 * new transaction and we can't block without protecting against other
78 * processes trying to touch the journal while it is in transition.
79 *
Dave Kleikamp470decc2006-10-11 01:20:57 -070080 */
81
82static transaction_t *
Mingming Caof7f4bcc2006-10-11 01:20:59 -070083jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -070084{
85 transaction->t_journal = journal;
86 transaction->t_state = T_RUNNING;
Josef Bacike07f7182008-11-26 01:14:26 -050087 transaction->t_start_time = ktime_get();
Dave Kleikamp470decc2006-10-11 01:20:57 -070088 transaction->t_tid = journal->j_transaction_sequence++;
89 transaction->t_expires = jiffies + journal->j_commit_interval;
90 spin_lock_init(&transaction->t_handle_lock);
Theodore Ts'oa51dca92010-08-02 08:43:25 -040091 atomic_set(&transaction->t_updates, 0);
Jan Kara8f7d89f2013-06-04 12:35:11 -040092 atomic_set(&transaction->t_outstanding_credits,
93 atomic_read(&journal->j_reserved_credits));
Theodore Ts'o8dd42042010-08-03 21:38:29 -040094 atomic_set(&transaction->t_handle_count, 0);
Jan Karac851ed52008-07-11 19:27:31 -040095 INIT_LIST_HEAD(&transaction->t_inode_list);
Theodore Ts'o3e624fc2008-10-16 20:00:24 -040096 INIT_LIST_HEAD(&transaction->t_private_list);
Dave Kleikamp470decc2006-10-11 01:20:57 -070097
98 /* Set up the commit timer for the new transaction. */
Andreas Dilgerb1f485f2009-08-10 22:51:53 -040099 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700100 add_timer(&journal->j_commit_timer);
101
102 J_ASSERT(journal->j_running_transaction == NULL);
103 journal->j_running_transaction = transaction;
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500104 transaction->t_max_wait = 0;
105 transaction->t_start = jiffies;
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500106 transaction->t_requested = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700107
108 return transaction;
109}
110
111/*
112 * Handle management.
113 *
114 * A handle_t is an object which represents a single atomic update to a
115 * filesystem, and which tracks all of the modifications which form part
116 * of that one update.
117 */
118
119/*
Tao Ma28e35e42011-05-22 21:45:26 -0400120 * Update transaction's maximum wait time, if debugging is enabled.
Theodore Ts'o6d0bf002010-08-09 17:28:38 -0400121 *
122 * In order for t_max_wait to be reliable, it must be protected by a
123 * lock. But doing so will mean that start_this_handle() can not be
124 * run in parallel on SMP systems, which limits our scalability. So
125 * unless debugging is enabled, we no longer update t_max_wait, which
126 * means that maximum wait time reported by the jbd2_run_stats
127 * tracepoint will always be zero.
128 */
Tao Ma28e35e42011-05-22 21:45:26 -0400129static inline void update_t_max_wait(transaction_t *transaction,
130 unsigned long ts)
Theodore Ts'o6d0bf002010-08-09 17:28:38 -0400131{
132#ifdef CONFIG_JBD2_DEBUG
Theodore Ts'o6d0bf002010-08-09 17:28:38 -0400133 if (jbd2_journal_enable_debug &&
134 time_after(transaction->t_start, ts)) {
135 ts = jbd2_time_diff(ts, transaction->t_start);
136 spin_lock(&transaction->t_handle_lock);
137 if (ts > transaction->t_max_wait)
138 transaction->t_max_wait = ts;
139 spin_unlock(&transaction->t_handle_lock);
140 }
141#endif
142}
143
144/*
Jan Kara8f7d89f2013-06-04 12:35:11 -0400145 * Wait until running transaction passes T_LOCKED state. Also starts the commit
146 * if needed. The function expects running transaction to exist and releases
147 * j_state_lock.
148 */
149static void wait_transaction_locked(journal_t *journal)
150 __releases(journal->j_state_lock)
151{
152 DEFINE_WAIT(wait);
153 int need_to_start;
154 tid_t tid = journal->j_running_transaction->t_tid;
155
156 prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
157 TASK_UNINTERRUPTIBLE);
158 need_to_start = !tid_geq(journal->j_commit_request, tid);
159 read_unlock(&journal->j_state_lock);
160 if (need_to_start)
161 jbd2_log_start_commit(journal, tid);
162 schedule();
163 finish_wait(&journal->j_wait_transaction_locked, &wait);
164}
165
166static void sub_reserved_credits(journal_t *journal, int blocks)
167{
168 atomic_sub(blocks, &journal->j_reserved_credits);
169 wake_up(&journal->j_wait_reserved);
170}
171
172/*
173 * Wait until we can add credits for handle to the running transaction. Called
174 * with j_state_lock held for reading. Returns 0 if handle joined the running
175 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
176 * caller must retry.
177 */
178static int add_transaction_credits(journal_t *journal, int blocks,
179 int rsv_blocks)
180{
181 transaction_t *t = journal->j_running_transaction;
182 int needed;
183 int total = blocks + rsv_blocks;
184
185 /*
186 * If the current transaction is locked down for commit, wait
187 * for the lock to be released.
188 */
189 if (t->t_state == T_LOCKED) {
190 wait_transaction_locked(journal);
191 return 1;
192 }
193
194 /*
195 * If there is not enough space left in the log to write all
196 * potential buffers requested by this operation, we need to
197 * stall pending a log checkpoint to free some more log space.
198 */
199 needed = atomic_add_return(total, &t->t_outstanding_credits);
200 if (needed > journal->j_max_transaction_buffers) {
201 /*
202 * If the current transaction is already too large,
203 * then start to commit it: we can then go back and
204 * attach this handle to a new transaction.
205 */
206 atomic_sub(total, &t->t_outstanding_credits);
Lukas Czerner6d3ec142015-08-04 11:21:52 -0400207
208 /*
209 * Is the number of reserved credits in the current transaction too
210 * big to fit this handle? Wait until reserved credits are freed.
211 */
212 if (atomic_read(&journal->j_reserved_credits) + total >
213 journal->j_max_transaction_buffers) {
214 read_unlock(&journal->j_state_lock);
215 wait_event(journal->j_wait_reserved,
216 atomic_read(&journal->j_reserved_credits) + total <=
217 journal->j_max_transaction_buffers);
218 return 1;
219 }
220
Jan Kara8f7d89f2013-06-04 12:35:11 -0400221 wait_transaction_locked(journal);
222 return 1;
223 }
224
225 /*
226 * The commit code assumes that it can get enough log space
227 * without forcing a checkpoint. This is *critical* for
228 * correctness: a checkpoint of a buffer which is also
229 * associated with a committing transaction creates a deadlock,
230 * so commit simply cannot force through checkpoints.
231 *
232 * We must therefore ensure the necessary space in the journal
233 * *before* starting to dirty potentially checkpointed buffers
234 * in the new transaction.
235 */
236 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
237 atomic_sub(total, &t->t_outstanding_credits);
238 read_unlock(&journal->j_state_lock);
239 write_lock(&journal->j_state_lock);
240 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
241 __jbd2_log_wait_for_space(journal);
242 write_unlock(&journal->j_state_lock);
243 return 1;
244 }
245
246 /* No reservation? We are done... */
247 if (!rsv_blocks)
248 return 0;
249
250 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
251 /* We allow at most half of a transaction to be reserved */
252 if (needed > journal->j_max_transaction_buffers / 2) {
253 sub_reserved_credits(journal, rsv_blocks);
254 atomic_sub(total, &t->t_outstanding_credits);
255 read_unlock(&journal->j_state_lock);
256 wait_event(journal->j_wait_reserved,
257 atomic_read(&journal->j_reserved_credits) + rsv_blocks
258 <= journal->j_max_transaction_buffers / 2);
259 return 1;
260 }
261 return 0;
262}
263
264/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700265 * start_this_handle: Given a handle, deal with any locking or stalling
266 * needed to make sure that there is enough journal space for the handle
267 * to begin. Attach the handle to a transaction and set up the
268 * transaction's buffer credits.
269 */
270
Theodore Ts'o47def822010-07-27 11:56:05 -0400271static int start_this_handle(journal_t *journal, handle_t *handle,
Dan Carpenterd2159fb2011-09-04 10:20:14 -0400272 gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700273{
Theodore Ts'oe4471832011-02-12 08:18:24 -0500274 transaction_t *transaction, *new_transaction = NULL;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400275 int blocks = handle->h_buffer_credits;
276 int rsv_blocks = 0;
Tao Ma28e35e42011-05-22 21:45:26 -0400277 unsigned long ts = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700278
Jan Kara8f7d89f2013-06-04 12:35:11 -0400279 if (handle->h_rsv_handle)
280 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
281
Lukas Czerner6d3ec142015-08-04 11:21:52 -0400282 /*
283 * Limit the number of reserved credits to 1/2 of maximum transaction
284 * size and limit the number of total credits to not exceed maximum
285 * transaction size per operation.
286 */
287 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
288 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
289 printk(KERN_ERR "JBD2: %s wants too many credits "
290 "credits:%d rsv_credits:%d max:%d\n",
291 current->comm, blocks, rsv_blocks,
292 journal->j_max_transaction_buffers);
293 WARN_ON(1);
294 return -ENOSPC;
295 }
296
Dave Kleikamp470decc2006-10-11 01:20:57 -0700297alloc_transaction:
298 if (!journal->j_running_transaction) {
Michal Hocko6ccaf3e2015-06-08 10:53:10 -0400299 /*
300 * If __GFP_FS is not present, then we may be being called from
301 * inside the fs writeback layer, so we MUST NOT fail.
302 */
303 if ((gfp_mask & __GFP_FS) == 0)
304 gfp_mask |= __GFP_NOFAIL;
Wanlong Gaob2f4edb2012-06-01 00:10:32 -0400305 new_transaction = kmem_cache_zalloc(transaction_cache,
306 gfp_mask);
Michal Hocko6ccaf3e2015-06-08 10:53:10 -0400307 if (!new_transaction)
Theodore Ts'o47def822010-07-27 11:56:05 -0400308 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700309 }
310
311 jbd_debug(3, "New handle %p going live.\n", handle);
312
Dave Kleikamp470decc2006-10-11 01:20:57 -0700313 /*
314 * We need to hold j_state_lock until t_updates has been incremented,
315 * for proper journal barrier handling
316 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400317repeat:
318 read_lock(&journal->j_state_lock);
Theodore Ts'o5c2178e2010-10-27 21:30:04 -0400319 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700320 if (is_journal_aborted(journal) ||
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700321 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400322 read_unlock(&journal->j_state_lock);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500323 jbd2_journal_free_transaction(new_transaction);
Theodore Ts'o47def822010-07-27 11:56:05 -0400324 return -EROFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700325 }
326
Jan Kara8f7d89f2013-06-04 12:35:11 -0400327 /*
328 * Wait on the journal's transaction barrier if necessary. Specifically
329 * we allow reserved handles to proceed because otherwise commit could
330 * deadlock on page writeback not being able to complete.
331 */
332 if (!handle->h_reserved && journal->j_barrier_count) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400333 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700334 wait_event(journal->j_wait_transaction_locked,
335 journal->j_barrier_count == 0);
336 goto repeat;
337 }
338
339 if (!journal->j_running_transaction) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400340 read_unlock(&journal->j_state_lock);
341 if (!new_transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700342 goto alloc_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400343 write_lock(&journal->j_state_lock);
Jan Karad7961c72012-12-21 00:15:51 -0500344 if (!journal->j_running_transaction &&
Jan Kara8f7d89f2013-06-04 12:35:11 -0400345 (handle->h_reserved || !journal->j_barrier_count)) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400346 jbd2_get_transaction(journal, new_transaction);
347 new_transaction = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700348 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400349 write_unlock(&journal->j_state_lock);
350 goto repeat;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700351 }
352
353 transaction = journal->j_running_transaction;
354
Jan Kara8f7d89f2013-06-04 12:35:11 -0400355 if (!handle->h_reserved) {
356 /* We may have dropped j_state_lock - restart in that case */
357 if (add_transaction_credits(journal, blocks, rsv_blocks))
358 goto repeat;
359 } else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700360 /*
Jan Kara8f7d89f2013-06-04 12:35:11 -0400361 * We have handle reserved so we are allowed to join T_LOCKED
362 * transaction and we don't have to check for transaction size
363 * and journal space.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700364 */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400365 sub_reserved_credits(journal, blocks);
366 handle->h_reserved = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700367 }
368
369 /* OK, account for the buffers that this operation expects to
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400370 * use and add the handle to the running transaction.
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400371 */
Tao Ma28e35e42011-05-22 21:45:26 -0400372 update_t_max_wait(transaction, ts);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700373 handle->h_transaction = transaction;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400374 handle->h_requested_credits = blocks;
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500375 handle->h_start_jiffies = jiffies;
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400376 atomic_inc(&transaction->t_updates);
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400377 atomic_inc(&transaction->t_handle_count);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400378 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
379 handle, blocks,
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400380 atomic_read(&transaction->t_outstanding_credits),
Jan Kara76c39902013-06-04 12:12:57 -0400381 jbd2_log_space_left(journal));
Theodore Ts'oa931da62010-08-03 21:35:12 -0400382 read_unlock(&journal->j_state_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400383 current->journal_info = handle;
Jan Kara9599b0e2009-08-17 21:23:17 -0400384
385 lock_map_acquire(&handle->h_lockdep_map);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500386 jbd2_journal_free_transaction(new_transaction);
Theodore Ts'o47def822010-07-27 11:56:05 -0400387 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700388}
389
Mingming Cao7b751062008-01-28 23:58:27 -0500390static struct lock_class_key jbd2_handle_key;
391
Dave Kleikamp470decc2006-10-11 01:20:57 -0700392/* Allocate a new handle. This should probably be in a slab... */
393static handle_t *new_handle(int nblocks)
394{
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400395 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700396 if (!handle)
397 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700398 handle->h_buffer_credits = nblocks;
399 handle->h_ref = 1;
400
Mingming Cao7b751062008-01-28 23:58:27 -0500401 lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle",
402 &jbd2_handle_key, 0);
403
Dave Kleikamp470decc2006-10-11 01:20:57 -0700404 return handle;
405}
406
407/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700408 * handle_t *jbd2_journal_start() - Obtain a new handle.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700409 * @journal: Journal to start transaction on.
410 * @nblocks: number of block buffer we might modify
411 *
412 * We make sure that the transaction can guarantee at least nblocks of
413 * modified buffers in the log. We block until the log can guarantee
Jan Kara8f7d89f2013-06-04 12:35:11 -0400414 * that much space. Additionally, if rsv_blocks > 0, we also create another
415 * handle with rsv_blocks reserved blocks in the journal. This handle is
416 * is stored in h_rsv_handle. It is not attached to any particular transaction
417 * and thus doesn't block transaction commit. If the caller uses this reserved
418 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
419 * on the parent handle will dispose the reserved one. Reserved handle has to
420 * be converted to a normal handle using jbd2_journal_start_reserved() before
421 * it can be used.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700422 *
Eryu Guanc8675162011-05-24 17:09:58 -0400423 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
424 * on failure.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700425 */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400426handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
427 gfp_t gfp_mask, unsigned int type,
428 unsigned int line_no)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700429{
430 handle_t *handle = journal_current_handle();
431 int err;
432
433 if (!journal)
434 return ERR_PTR(-EROFS);
435
436 if (handle) {
437 J_ASSERT(handle->h_transaction->t_journal == journal);
438 handle->h_ref++;
439 return handle;
440 }
441
442 handle = new_handle(nblocks);
443 if (!handle)
444 return ERR_PTR(-ENOMEM);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400445 if (rsv_blocks) {
446 handle_t *rsv_handle;
447
448 rsv_handle = new_handle(rsv_blocks);
449 if (!rsv_handle) {
450 jbd2_free_handle(handle);
451 return ERR_PTR(-ENOMEM);
452 }
453 rsv_handle->h_reserved = 1;
454 rsv_handle->h_journal = journal;
455 handle->h_rsv_handle = rsv_handle;
456 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700457
Theodore Ts'o47def822010-07-27 11:56:05 -0400458 err = start_this_handle(journal, handle, gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700459 if (err < 0) {
Jan Kara8f7d89f2013-06-04 12:35:11 -0400460 if (handle->h_rsv_handle)
461 jbd2_free_handle(handle->h_rsv_handle);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400462 jbd2_free_handle(handle);
Dmitry Monakhovdf05c1b82013-03-02 17:08:46 -0500463 return ERR_PTR(err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700464 }
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500465 handle->h_type = type;
466 handle->h_line_no = line_no;
467 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
468 handle->h_transaction->t_tid, type,
469 line_no, nblocks);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700470 return handle;
471}
Theodore Ts'o47def822010-07-27 11:56:05 -0400472EXPORT_SYMBOL(jbd2__journal_start);
473
474
475handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
476{
Jan Kara8f7d89f2013-06-04 12:35:11 -0400477 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
Theodore Ts'o47def822010-07-27 11:56:05 -0400478}
479EXPORT_SYMBOL(jbd2_journal_start);
480
Jan Kara8f7d89f2013-06-04 12:35:11 -0400481void jbd2_journal_free_reserved(handle_t *handle)
482{
483 journal_t *journal = handle->h_journal;
484
485 WARN_ON(!handle->h_reserved);
486 sub_reserved_credits(journal, handle->h_buffer_credits);
487 jbd2_free_handle(handle);
488}
489EXPORT_SYMBOL(jbd2_journal_free_reserved);
490
491/**
492 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
493 * @handle: handle to start
494 *
495 * Start handle that has been previously reserved with jbd2_journal_reserve().
496 * This attaches @handle to the running transaction (or creates one if there's
497 * not transaction running). Unlike jbd2_journal_start() this function cannot
498 * block on journal commit, checkpointing, or similar stuff. It can block on
499 * memory allocation or frozen journal though.
500 *
501 * Return 0 on success, non-zero on error - handle is freed in that case.
502 */
503int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
504 unsigned int line_no)
505{
506 journal_t *journal = handle->h_journal;
507 int ret = -EIO;
508
509 if (WARN_ON(!handle->h_reserved)) {
510 /* Someone passed in normal handle? Just stop it. */
511 jbd2_journal_stop(handle);
512 return ret;
513 }
514 /*
515 * Usefulness of mixing of reserved and unreserved handles is
516 * questionable. So far nobody seems to need it so just error out.
517 */
518 if (WARN_ON(current->journal_info)) {
519 jbd2_journal_free_reserved(handle);
520 return ret;
521 }
522
523 handle->h_journal = NULL;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400524 /*
525 * GFP_NOFS is here because callers are likely from writeback or
526 * similarly constrained call sites
527 */
528 ret = start_this_handle(journal, handle, GFP_NOFS);
Dan Carpenter92e3b402014-02-17 20:33:01 -0500529 if (ret < 0) {
Jan Kara8f7d89f2013-06-04 12:35:11 -0400530 jbd2_journal_free_reserved(handle);
Dan Carpenter92e3b402014-02-17 20:33:01 -0500531 return ret;
532 }
Jan Kara8f7d89f2013-06-04 12:35:11 -0400533 handle->h_type = type;
534 handle->h_line_no = line_no;
Dan Carpenter92e3b402014-02-17 20:33:01 -0500535 return 0;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400536}
537EXPORT_SYMBOL(jbd2_journal_start_reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700538
539/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700540 * int jbd2_journal_extend() - extend buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700541 * @handle: handle to 'extend'
542 * @nblocks: nr blocks to try to extend by.
543 *
544 * Some transactions, such as large extends and truncates, can be done
545 * atomically all at once or in several stages. The operation requests
546 * a credit for a number of buffer modications in advance, but can
547 * extend its credit if it needs more.
548 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700549 * jbd2_journal_extend tries to give the running handle more buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700550 * It does not guarantee that allocation - this is a best-effort only.
551 * The calling process MUST be able to deal cleanly with a failure to
552 * extend here.
553 *
554 * Return 0 on success, non-zero on failure.
555 *
556 * return code < 0 implies an error
557 * return code > 0 implies normal transaction-full status.
558 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700559int jbd2_journal_extend(handle_t *handle, int nblocks)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700560{
561 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400562 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700563 int result;
564 int wanted;
565
Dave Kleikamp470decc2006-10-11 01:20:57 -0700566 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400567 return -EROFS;
568 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700569
570 result = 1;
571
Theodore Ts'oa931da62010-08-03 21:35:12 -0400572 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700573
574 /* Don't extend a locked-down transaction! */
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400575 if (transaction->t_state != T_RUNNING) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700576 jbd_debug(3, "denied handle %p %d blocks: "
577 "transaction not running\n", handle, nblocks);
578 goto error_out;
579 }
580
581 spin_lock(&transaction->t_handle_lock);
Jan Karafe1e8db2013-06-04 12:22:15 -0400582 wanted = atomic_add_return(nblocks,
583 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700584
585 if (wanted > journal->j_max_transaction_buffers) {
586 jbd_debug(3, "denied handle %p %d blocks: "
587 "transaction too large\n", handle, nblocks);
Jan Karafe1e8db2013-06-04 12:22:15 -0400588 atomic_sub(nblocks, &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700589 goto unlock;
590 }
591
Jan Kara76c39902013-06-04 12:12:57 -0400592 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
593 jbd2_log_space_left(journal)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700594 jbd_debug(3, "denied handle %p %d blocks: "
595 "insufficient log space\n", handle, nblocks);
Jan Karafe1e8db2013-06-04 12:22:15 -0400596 atomic_sub(nblocks, &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700597 goto unlock;
598 }
599
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500600 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400601 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500602 handle->h_type, handle->h_line_no,
603 handle->h_buffer_credits,
604 nblocks);
605
Dave Kleikamp470decc2006-10-11 01:20:57 -0700606 handle->h_buffer_credits += nblocks;
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500607 handle->h_requested_credits += nblocks;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700608 result = 0;
609
610 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
611unlock:
612 spin_unlock(&transaction->t_handle_lock);
613error_out:
Theodore Ts'oa931da62010-08-03 21:35:12 -0400614 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700615 return result;
616}
617
618
619/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700620 * int jbd2_journal_restart() - restart a handle .
Dave Kleikamp470decc2006-10-11 01:20:57 -0700621 * @handle: handle to restart
622 * @nblocks: nr credits requested
623 *
624 * Restart a handle for a multi-transaction filesystem
625 * operation.
626 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700627 * If the jbd2_journal_extend() call above fails to grant new buffer credits
628 * to a running handle, a call to jbd2_journal_restart will commit the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700629 * handle's transaction so far and reattach the handle to a new
630 * transaction capabable of guaranteeing the requested number of
Jan Kara8f7d89f2013-06-04 12:35:11 -0400631 * credits. We preserve reserved handle if there's any attached to the
632 * passed in handle.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700633 */
Dan Carpenterd2159fb2011-09-04 10:20:14 -0400634int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700635{
636 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400637 journal_t *journal;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500638 tid_t tid;
639 int need_to_start, ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700640
641 /* If we've had an abort of any type, don't even think about
642 * actually doing the restart! */
643 if (is_handle_aborted(handle))
644 return 0;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400645 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700646
647 /*
648 * First unlink the handle from its current transaction, and start the
649 * commit on that.
650 */
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400651 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700652 J_ASSERT(journal_current_handle() == handle);
653
Theodore Ts'oa931da62010-08-03 21:35:12 -0400654 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700655 spin_lock(&transaction->t_handle_lock);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400656 atomic_sub(handle->h_buffer_credits,
657 &transaction->t_outstanding_credits);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400658 if (handle->h_rsv_handle) {
659 sub_reserved_credits(journal,
660 handle->h_rsv_handle->h_buffer_credits);
661 }
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400662 if (atomic_dec_and_test(&transaction->t_updates))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700663 wake_up(&journal->j_wait_updates);
Theodore Ts'o39c04152013-07-01 08:12:40 -0400664 tid = transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700665 spin_unlock(&transaction->t_handle_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400666 handle->h_transaction = NULL;
667 current->journal_info = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700668
669 jbd_debug(2, "restarting handle %p\n", handle);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500670 need_to_start = !tid_geq(journal->j_commit_request, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400671 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500672 if (need_to_start)
673 jbd2_log_start_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700674
Jan Kara9599b0e2009-08-17 21:23:17 -0400675 lock_map_release(&handle->h_lockdep_map);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700676 handle->h_buffer_credits = nblocks;
Theodore Ts'o47def822010-07-27 11:56:05 -0400677 ret = start_this_handle(journal, handle, gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700678 return ret;
679}
Theodore Ts'o47def822010-07-27 11:56:05 -0400680EXPORT_SYMBOL(jbd2__journal_restart);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700681
682
Theodore Ts'o47def822010-07-27 11:56:05 -0400683int jbd2_journal_restart(handle_t *handle, int nblocks)
684{
685 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
686}
687EXPORT_SYMBOL(jbd2_journal_restart);
688
Dave Kleikamp470decc2006-10-11 01:20:57 -0700689/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700690 * void jbd2_journal_lock_updates () - establish a transaction barrier.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700691 * @journal: Journal to establish a barrier on.
692 *
693 * This locks out any further updates from being started, and blocks
694 * until all existing updates have completed, returning only once the
695 * journal is in a quiescent state with no updates running.
696 *
697 * The journal lock should not be held on entry.
698 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700699void jbd2_journal_lock_updates(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700700{
701 DEFINE_WAIT(wait);
702
Theodore Ts'oa931da62010-08-03 21:35:12 -0400703 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700704 ++journal->j_barrier_count;
705
Jan Kara8f7d89f2013-06-04 12:35:11 -0400706 /* Wait until there are no reserved handles */
707 if (atomic_read(&journal->j_reserved_credits)) {
708 write_unlock(&journal->j_state_lock);
709 wait_event(journal->j_wait_reserved,
710 atomic_read(&journal->j_reserved_credits) == 0);
711 write_lock(&journal->j_state_lock);
712 }
713
Dave Kleikamp470decc2006-10-11 01:20:57 -0700714 /* Wait until there are no running updates */
715 while (1) {
716 transaction_t *transaction = journal->j_running_transaction;
717
718 if (!transaction)
719 break;
720
721 spin_lock(&transaction->t_handle_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700722 prepare_to_wait(&journal->j_wait_updates, &wait,
723 TASK_UNINTERRUPTIBLE);
Jan Kara9837d8e2012-01-04 22:03:11 -0500724 if (!atomic_read(&transaction->t_updates)) {
725 spin_unlock(&transaction->t_handle_lock);
726 finish_wait(&journal->j_wait_updates, &wait);
727 break;
728 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700729 spin_unlock(&transaction->t_handle_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400730 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700731 schedule();
732 finish_wait(&journal->j_wait_updates, &wait);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400733 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700734 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400735 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700736
737 /*
738 * We have now established a barrier against other normal updates, but
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700739 * we also need to barrier against other jbd2_journal_lock_updates() calls
Dave Kleikamp470decc2006-10-11 01:20:57 -0700740 * to make sure that we serialise special journal-locked operations
741 * too.
742 */
743 mutex_lock(&journal->j_barrier);
744}
745
746/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700747 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
Dave Kleikamp470decc2006-10-11 01:20:57 -0700748 * @journal: Journal to release the barrier on.
749 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700750 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700751 *
752 * Should be called without the journal lock held.
753 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700754void jbd2_journal_unlock_updates (journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700755{
756 J_ASSERT(journal->j_barrier_count != 0);
757
758 mutex_unlock(&journal->j_barrier);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400759 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700760 --journal->j_barrier_count;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400761 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700762 wake_up(&journal->j_wait_transaction_locked);
763}
764
Jan Karaf91d1d02009-07-13 16:16:20 -0400765static void warn_dirty_buffer(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700766{
Jan Karaf91d1d02009-07-13 16:16:20 -0400767 char b[BDEVNAME_SIZE];
Dave Kleikamp470decc2006-10-11 01:20:57 -0700768
Jan Karaf91d1d02009-07-13 16:16:20 -0400769 printk(KERN_WARNING
Eryu Guanf2a44522011-11-01 19:09:18 -0400770 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
Jan Karaf91d1d02009-07-13 16:16:20 -0400771 "There's a risk of filesystem corruption in case of system "
772 "crash.\n",
773 bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700774}
775
Jan Karaee57aba2015-06-08 12:39:07 -0400776/* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
777static void jbd2_freeze_jh_data(struct journal_head *jh)
778{
779 struct page *page;
780 int offset;
781 char *source;
782 struct buffer_head *bh = jh2bh(jh);
783
784 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
785 page = bh->b_page;
786 offset = offset_in_page(bh->b_data);
787 source = kmap_atomic(page);
788 /* Fire data frozen trigger just before we copy the data */
789 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
790 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
791 kunmap_atomic(source);
792
793 /*
794 * Now that the frozen data is saved off, we need to store any matching
795 * triggers.
796 */
797 jh->b_frozen_triggers = jh->b_triggers;
798}
799
Dave Kleikamp470decc2006-10-11 01:20:57 -0700800/*
801 * If the buffer is already part of the current transaction, then there
802 * is nothing we need to do. If it is already part of a prior
803 * transaction which we are still committing to disk, then we need to
804 * make sure that we do not overwrite the old copy: we do copy-out to
805 * preserve the copy going to disk. We also account the buffer against
806 * the handle's metadata buffer credits (unless the buffer is already
807 * part of the transaction, that is).
808 *
809 */
810static int
811do_get_write_access(handle_t *handle, struct journal_head *jh,
812 int force_copy)
813{
814 struct buffer_head *bh;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400815 transaction_t *transaction = handle->h_transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700816 journal_t *journal;
817 int error;
818 char *frozen_buffer = NULL;
Theodore Ts'of783f092013-04-21 16:47:54 -0400819 unsigned long start_lock, time_lock;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700820
821 if (is_handle_aborted(handle))
822 return -EROFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700823 journal = transaction->t_journal;
824
Theodore Ts'ocfef2c62010-12-18 13:07:34 -0500825 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700826
827 JBUFFER_TRACE(jh, "entry");
828repeat:
829 bh = jh2bh(jh);
830
831 /* @@@ Need to check for errors here at some point. */
832
Theodore Ts'of783f092013-04-21 16:47:54 -0400833 start_lock = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700834 lock_buffer(bh);
835 jbd_lock_bh_state(bh);
836
Theodore Ts'of783f092013-04-21 16:47:54 -0400837 /* If it takes too long to lock the buffer, trace it */
838 time_lock = jbd2_time_diff(start_lock, jiffies);
839 if (time_lock > HZ/10)
840 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
841 jiffies_to_msecs(time_lock));
842
Dave Kleikamp470decc2006-10-11 01:20:57 -0700843 /* We now hold the buffer lock so it is safe to query the buffer
844 * state. Is the buffer dirty?
845 *
846 * If so, there are two possibilities. The buffer may be
847 * non-journaled, and undergoing a quite legitimate writeback.
848 * Otherwise, it is journaled, and we don't expect dirty buffers
849 * in that state (the buffers should be marked JBD_Dirty
850 * instead.) So either the IO is being done under our own
851 * control and this is a bug, or it's a third party IO such as
852 * dump(8) (which may leave the buffer scheduled for read ---
853 * ie. locked but not dirty) or tune2fs (which may actually have
854 * the buffer dirtied, ugh.) */
855
856 if (buffer_dirty(bh)) {
857 /*
858 * First question: is this buffer already part of the current
859 * transaction or the existing committing transaction?
860 */
861 if (jh->b_transaction) {
862 J_ASSERT_JH(jh,
863 jh->b_transaction == transaction ||
864 jh->b_transaction ==
865 journal->j_committing_transaction);
866 if (jh->b_next_transaction)
867 J_ASSERT_JH(jh, jh->b_next_transaction ==
868 transaction);
Jan Karaf91d1d02009-07-13 16:16:20 -0400869 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700870 }
871 /*
872 * In any case we need to clean the dirty flag and we must
873 * do it under the buffer lock to be sure we don't race
874 * with running write-out.
875 */
Jan Karaf91d1d02009-07-13 16:16:20 -0400876 JBUFFER_TRACE(jh, "Journalling dirty buffer");
877 clear_buffer_dirty(bh);
878 set_buffer_jbddirty(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700879 }
880
881 unlock_buffer(bh);
882
883 error = -EROFS;
884 if (is_handle_aborted(handle)) {
885 jbd_unlock_bh_state(bh);
886 goto out;
887 }
888 error = 0;
889
890 /*
891 * The buffer is already part of this transaction if b_transaction or
892 * b_next_transaction points to it
893 */
894 if (jh->b_transaction == transaction ||
895 jh->b_next_transaction == transaction)
896 goto done;
897
898 /*
Josef Bacik9fc7c632008-04-17 10:38:59 -0400899 * this is the first time this transaction is touching this buffer,
900 * reset the modified flag
901 */
902 jh->b_modified = 0;
903
904 /*
Jan Kara8b00f402015-06-08 12:44:21 -0400905 * If the buffer is not journaled right now, we need to make sure it
906 * doesn't get written to disk before the caller actually commits the
907 * new data
908 */
909 if (!jh->b_transaction) {
910 JBUFFER_TRACE(jh, "no transaction");
911 J_ASSERT_JH(jh, !jh->b_next_transaction);
912 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Jan Karade92c8c2015-06-08 12:46:37 -0400913 /*
914 * Make sure all stores to jh (b_modified, b_frozen_data) are
915 * visible before attaching it to the running transaction.
916 * Paired with barrier in jbd2_write_access_granted()
917 */
918 smp_wmb();
Jan Kara8b00f402015-06-08 12:44:21 -0400919 spin_lock(&journal->j_list_lock);
920 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
921 spin_unlock(&journal->j_list_lock);
922 goto done;
923 }
924 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700925 * If there is already a copy-out version of this buffer, then we don't
926 * need to make another one
927 */
928 if (jh->b_frozen_data) {
929 JBUFFER_TRACE(jh, "has frozen data");
930 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
Jan Karade92c8c2015-06-08 12:46:37 -0400931 goto attach_next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700932 }
933
Jan Kara8b00f402015-06-08 12:44:21 -0400934 JBUFFER_TRACE(jh, "owned by older transaction");
935 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
936 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700937
938 /*
Jan Kara8b00f402015-06-08 12:44:21 -0400939 * There is one case we have to be very careful about. If the
940 * committing transaction is currently writing this buffer out to disk
941 * and has NOT made a copy-out, then we cannot modify the buffer
942 * contents at all right now. The essence of copy-out is that it is
943 * the extra copy, not the primary copy, which gets journaled. If the
944 * primary copy is already going to disk then we cannot do copy-out
945 * here.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700946 */
Jan Kara8b00f402015-06-08 12:44:21 -0400947 if (buffer_shadow(bh)) {
948 JBUFFER_TRACE(jh, "on shadow: sleep");
949 jbd_unlock_bh_state(bh);
950 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
951 goto repeat;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700952 }
953
Jan Kara8b00f402015-06-08 12:44:21 -0400954 /*
955 * Only do the copy if the currently-owning transaction still needs it.
956 * If buffer isn't on BJ_Metadata list, the committing transaction is
957 * past that stage (here we use the fact that BH_Shadow is set under
958 * bh_state lock together with refiling to BJ_Shadow list and at this
959 * point we know the buffer doesn't have BH_Shadow set).
960 *
961 * Subtle point, though: if this is a get_undo_access, then we will be
962 * relying on the frozen_data to contain the new value of the
963 * committed_data record after the transaction, so we HAVE to force the
964 * frozen_data copy in that case.
965 */
966 if (jh->b_jlist == BJ_Metadata || force_copy) {
967 JBUFFER_TRACE(jh, "generate frozen data");
968 if (!frozen_buffer) {
969 JBUFFER_TRACE(jh, "allocate memory for buffer");
970 jbd_unlock_bh_state(bh);
971 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS);
972 if (!frozen_buffer) {
973 printk(KERN_ERR "%s: OOM for frozen_buffer\n",
974 __func__);
975 JBUFFER_TRACE(jh, "oom!");
976 error = -ENOMEM;
977 goto out;
978 }
979 goto repeat;
980 }
981 jh->b_frozen_data = frozen_buffer;
982 frozen_buffer = NULL;
983 jbd2_freeze_jh_data(jh);
984 }
Jan Karade92c8c2015-06-08 12:46:37 -0400985attach_next:
986 /*
987 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
988 * before attaching it to the running transaction. Paired with barrier
989 * in jbd2_write_access_granted()
990 */
991 smp_wmb();
Jan Kara8b00f402015-06-08 12:44:21 -0400992 jh->b_next_transaction = transaction;
993
Dave Kleikamp470decc2006-10-11 01:20:57 -0700994done:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700995 jbd_unlock_bh_state(bh);
996
997 /*
998 * If we are about to journal a buffer, then any revoke pending on it is
999 * no longer valid
1000 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001001 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001002
1003out:
1004 if (unlikely(frozen_buffer)) /* It's usually NULL */
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001005 jbd2_free(frozen_buffer, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001006
1007 JBUFFER_TRACE(jh, "exit");
1008 return error;
1009}
1010
Jan Karade92c8c2015-06-08 12:46:37 -04001011/* Fast check whether buffer is already attached to the required transaction */
1012static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh)
1013{
1014 struct journal_head *jh;
1015 bool ret = false;
1016
1017 /* Dirty buffers require special handling... */
1018 if (buffer_dirty(bh))
1019 return false;
1020
1021 /*
1022 * RCU protects us from dereferencing freed pages. So the checks we do
1023 * are guaranteed not to oops. However the jh slab object can get freed
1024 * & reallocated while we work with it. So we have to be careful. When
1025 * we see jh attached to the running transaction, we know it must stay
1026 * so until the transaction is committed. Thus jh won't be freed and
1027 * will be attached to the same bh while we run. However it can
1028 * happen jh gets freed, reallocated, and attached to the transaction
1029 * just after we get pointer to it from bh. So we have to be careful
1030 * and recheck jh still belongs to our bh before we return success.
1031 */
1032 rcu_read_lock();
1033 if (!buffer_jbd(bh))
1034 goto out;
1035 /* This should be bh2jh() but that doesn't work with inline functions */
1036 jh = READ_ONCE(bh->b_private);
1037 if (!jh)
1038 goto out;
1039 if (jh->b_transaction != handle->h_transaction &&
1040 jh->b_next_transaction != handle->h_transaction)
1041 goto out;
1042 /*
1043 * There are two reasons for the barrier here:
1044 * 1) Make sure to fetch b_bh after we did previous checks so that we
1045 * detect when jh went through free, realloc, attach to transaction
1046 * while we were checking. Paired with implicit barrier in that path.
1047 * 2) So that access to bh done after jbd2_write_access_granted()
1048 * doesn't get reordered and see inconsistent state of concurrent
1049 * do_get_write_access().
1050 */
1051 smp_mb();
1052 if (unlikely(jh->b_bh != bh))
1053 goto out;
1054 ret = true;
1055out:
1056 rcu_read_unlock();
1057 return ret;
1058}
1059
Dave Kleikamp470decc2006-10-11 01:20:57 -07001060/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001061 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001062 * @handle: transaction to add buffer modifications to
1063 * @bh: bh to be used for metadata writes
Dave Kleikamp470decc2006-10-11 01:20:57 -07001064 *
1065 * Returns an error code or 0 on success.
1066 *
1067 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1068 * because we're write()ing a buffer which is also part of a shared mapping.
1069 */
1070
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001071int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001072{
Jan Karade92c8c2015-06-08 12:46:37 -04001073 struct journal_head *jh;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001074 int rc;
1075
Jan Karade92c8c2015-06-08 12:46:37 -04001076 if (jbd2_write_access_granted(handle, bh))
1077 return 0;
1078
1079 jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001080 /* We do not want to get caught playing with fields which the
1081 * log thread also manipulates. Make sure that the buffer
1082 * completes any outstanding IO before proceeding. */
1083 rc = do_get_write_access(handle, jh, 0);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001084 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001085 return rc;
1086}
1087
1088
1089/*
1090 * When the user wants to journal a newly created buffer_head
1091 * (ie. getblk() returned a new buffer and we are going to populate it
1092 * manually rather than reading off disk), then we need to keep the
1093 * buffer_head locked until it has been completely filled with new
1094 * data. In this case, we should be able to make the assertion that
1095 * the bh is not already part of an existing transaction.
1096 *
1097 * The buffer should already be locked by the caller by this point.
1098 * There is no lock ranking violation: it was a newly created,
1099 * unlocked buffer beforehand. */
1100
1101/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001102 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
Dave Kleikamp470decc2006-10-11 01:20:57 -07001103 * @handle: transaction to new buffer to
1104 * @bh: new buffer.
1105 *
1106 * Call this if you create a new bh.
1107 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001108int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001109{
1110 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001111 journal_t *journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001112 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001113 int err;
1114
1115 jbd_debug(5, "journal_head %p\n", jh);
1116 err = -EROFS;
1117 if (is_handle_aborted(handle))
1118 goto out;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001119 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001120 err = 0;
1121
1122 JBUFFER_TRACE(jh, "entry");
1123 /*
1124 * The buffer may already belong to this transaction due to pre-zeroing
1125 * in the filesystem's new_block code. It may also be on the previous,
1126 * committing transaction's lists, but it HAS to be in Forget state in
1127 * that case: the transaction must have deleted the buffer for it to be
1128 * reused here.
1129 */
1130 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001131 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1132 jh->b_transaction == NULL ||
1133 (jh->b_transaction == journal->j_committing_transaction &&
1134 jh->b_jlist == BJ_Forget)));
1135
1136 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1137 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1138
1139 if (jh->b_transaction == NULL) {
Jan Karaf91d1d02009-07-13 16:16:20 -04001140 /*
1141 * Previous jbd2_journal_forget() could have left the buffer
1142 * with jbddirty bit set because it was being committed. When
1143 * the commit finished, we've filed the buffer for
1144 * checkpointing and marked it dirty. Now we are reallocating
1145 * the buffer so the transaction freeing it must have
1146 * committed and so it's safe to clear the dirty bit.
1147 */
1148 clear_buffer_dirty(jh2bh(jh));
Josef Bacik9fc7c632008-04-17 10:38:59 -04001149 /* first access by this transaction */
1150 jh->b_modified = 0;
1151
Dave Kleikamp470decc2006-10-11 01:20:57 -07001152 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Theodore Ts'o6e4862a2014-03-09 00:46:23 -05001153 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001154 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001155 } else if (jh->b_transaction == journal->j_committing_transaction) {
Josef Bacik9fc7c632008-04-17 10:38:59 -04001156 /* first access by this transaction */
1157 jh->b_modified = 0;
1158
Dave Kleikamp470decc2006-10-11 01:20:57 -07001159 JBUFFER_TRACE(jh, "set next transaction");
Theodore Ts'o6e4862a2014-03-09 00:46:23 -05001160 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001161 jh->b_next_transaction = transaction;
1162 }
1163 spin_unlock(&journal->j_list_lock);
1164 jbd_unlock_bh_state(bh);
1165
1166 /*
1167 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1168 * blocks which contain freed but then revoked metadata. We need
1169 * to cancel the revoke in case we end up freeing it yet again
1170 * and the reallocating as data - this would cause a second revoke,
1171 * which hits an assertion error.
1172 */
1173 JBUFFER_TRACE(jh, "cancelling revoke");
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001174 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001175out:
Ding Dinghua3991b402011-05-25 17:43:48 -04001176 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001177 return err;
1178}
1179
1180/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001181 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
Dave Kleikamp470decc2006-10-11 01:20:57 -07001182 * non-rewindable consequences
1183 * @handle: transaction
1184 * @bh: buffer to undo
Dave Kleikamp470decc2006-10-11 01:20:57 -07001185 *
1186 * Sometimes there is a need to distinguish between metadata which has
1187 * been committed to disk and that which has not. The ext3fs code uses
1188 * this for freeing and allocating space, we have to make sure that we
1189 * do not reuse freed space until the deallocation has been committed,
1190 * since if we overwrote that space we would make the delete
1191 * un-rewindable in case of a crash.
1192 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001193 * To deal with that, jbd2_journal_get_undo_access requests write access to a
Dave Kleikamp470decc2006-10-11 01:20:57 -07001194 * buffer for parts of non-rewindable operations such as delete
1195 * operations on the bitmaps. The journaling code must keep a copy of
1196 * the buffer's contents prior to the undo_access call until such time
1197 * as we know that the buffer has definitely been committed to disk.
1198 *
1199 * We never need to know which transaction the committed data is part
1200 * of, buffers touched here are guaranteed to be dirtied later and so
1201 * will be committed to a new transaction in due course, at which point
1202 * we can discard the old committed data pointer.
1203 *
1204 * Returns error number or 0 on success.
1205 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001206int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001207{
1208 int err;
Jan Karade92c8c2015-06-08 12:46:37 -04001209 struct journal_head *jh;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001210 char *committed_data = NULL;
1211
1212 JBUFFER_TRACE(jh, "entry");
Jan Karade92c8c2015-06-08 12:46:37 -04001213 if (jbd2_write_access_granted(handle, bh))
1214 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001215
Jan Karade92c8c2015-06-08 12:46:37 -04001216 jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001217 /*
1218 * Do this first --- it can drop the journal lock, so we want to
1219 * make sure that obtaining the committed_data is done
1220 * atomically wrt. completion of any outstanding commits.
1221 */
1222 err = do_get_write_access(handle, jh, 1);
1223 if (err)
1224 goto out;
1225
1226repeat:
1227 if (!jh->b_committed_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001228 committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001229 if (!committed_data) {
Jan Kara75685072013-12-08 21:13:59 -05001230 printk(KERN_ERR "%s: No memory for committed data\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001231 __func__);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001232 err = -ENOMEM;
1233 goto out;
1234 }
1235 }
1236
1237 jbd_lock_bh_state(bh);
1238 if (!jh->b_committed_data) {
1239 /* Copy out the current buffer contents into the
1240 * preserved, committed copy. */
1241 JBUFFER_TRACE(jh, "generate b_committed data");
1242 if (!committed_data) {
1243 jbd_unlock_bh_state(bh);
1244 goto repeat;
1245 }
1246
1247 jh->b_committed_data = committed_data;
1248 committed_data = NULL;
1249 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1250 }
1251 jbd_unlock_bh_state(bh);
1252out:
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001253 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001254 if (unlikely(committed_data))
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001255 jbd2_free(committed_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001256 return err;
1257}
1258
1259/**
Joel Beckere06c8222008-09-11 15:35:47 -07001260 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1261 * @bh: buffer to trigger on
1262 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1263 *
1264 * Set any triggers on this journal_head. This is always safe, because
1265 * triggers for a committing buffer will be saved off, and triggers for
1266 * a running transaction will match the buffer in that transaction.
1267 *
1268 * Call with NULL to clear the triggers.
1269 */
1270void jbd2_journal_set_triggers(struct buffer_head *bh,
1271 struct jbd2_buffer_trigger_type *type)
1272{
Jan Karaad56eda2013-03-11 13:24:56 -04001273 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
Joel Beckere06c8222008-09-11 15:35:47 -07001274
Jan Karaad56eda2013-03-11 13:24:56 -04001275 if (WARN_ON(!jh))
1276 return;
Joel Beckere06c8222008-09-11 15:35:47 -07001277 jh->b_triggers = type;
Jan Karaad56eda2013-03-11 13:24:56 -04001278 jbd2_journal_put_journal_head(jh);
Joel Beckere06c8222008-09-11 15:35:47 -07001279}
1280
Jan Kara13ceef02010-07-14 07:56:33 +02001281void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
Joel Beckere06c8222008-09-11 15:35:47 -07001282 struct jbd2_buffer_trigger_type *triggers)
1283{
1284 struct buffer_head *bh = jh2bh(jh);
1285
Jan Kara13ceef02010-07-14 07:56:33 +02001286 if (!triggers || !triggers->t_frozen)
Joel Beckere06c8222008-09-11 15:35:47 -07001287 return;
1288
Jan Kara13ceef02010-07-14 07:56:33 +02001289 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
Joel Beckere06c8222008-09-11 15:35:47 -07001290}
1291
1292void jbd2_buffer_abort_trigger(struct journal_head *jh,
1293 struct jbd2_buffer_trigger_type *triggers)
1294{
1295 if (!triggers || !triggers->t_abort)
1296 return;
1297
1298 triggers->t_abort(triggers, jh2bh(jh));
1299}
1300
Joel Beckere06c8222008-09-11 15:35:47 -07001301/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001302 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
Dave Kleikamp470decc2006-10-11 01:20:57 -07001303 * @handle: transaction to add buffer to.
1304 * @bh: buffer to mark
1305 *
1306 * mark dirty metadata which needs to be journaled as part of the current
1307 * transaction.
1308 *
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001309 * The buffer must have previously had jbd2_journal_get_write_access()
1310 * called so that it has a valid journal_head attached to the buffer
1311 * head.
1312 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001313 * The buffer is placed on the transaction's metadata list and is marked
1314 * as belonging to the transaction.
1315 *
1316 * Returns error number or 0 on success.
1317 *
1318 * Special care needs to be taken if the buffer already belongs to the
1319 * current committing transaction (in which case we should have frozen
1320 * data present for that commit). In that case, we don't relink the
1321 * buffer: that only gets done when the old transaction finally
1322 * completes its commit.
1323 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001324int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001325{
1326 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001327 journal_t *journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001328 struct journal_head *jh;
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001329 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001330
Dave Kleikamp470decc2006-10-11 01:20:57 -07001331 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001332 return -EROFS;
Jan Kara6e06ae82015-07-12 18:11:30 -04001333 if (!buffer_jbd(bh)) {
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001334 ret = -EUCLEAN;
1335 goto out;
1336 }
Jan Kara6e06ae82015-07-12 18:11:30 -04001337 /*
1338 * We don't grab jh reference here since the buffer must be part
1339 * of the running transaction.
1340 */
1341 jh = bh2jh(bh);
1342 /*
1343 * This and the following assertions are unreliable since we may see jh
1344 * in inconsistent state unless we grab bh_state lock. But this is
1345 * crucial to catch bugs so let's do a reliable check until the
1346 * lockless handling is fully proven.
1347 */
1348 if (jh->b_transaction != transaction &&
1349 jh->b_next_transaction != transaction) {
1350 jbd_lock_bh_state(bh);
1351 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1352 jh->b_next_transaction == transaction);
1353 jbd_unlock_bh_state(bh);
1354 }
1355 if (jh->b_modified == 1) {
1356 /* If it's in our transaction it must be in BJ_Metadata list. */
1357 if (jh->b_transaction == transaction &&
1358 jh->b_jlist != BJ_Metadata) {
1359 jbd_lock_bh_state(bh);
1360 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1361 jh->b_jlist == BJ_Metadata);
1362 jbd_unlock_bh_state(bh);
1363 }
1364 goto out;
1365 }
1366
1367 journal = transaction->t_journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001368 jbd_debug(5, "journal_head %p\n", jh);
1369 JBUFFER_TRACE(jh, "entry");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001370
1371 jbd_lock_bh_state(bh);
1372
1373 if (jh->b_modified == 0) {
1374 /*
1375 * This buffer's got modified and becoming part
1376 * of the transaction. This needs to be done
1377 * once a transaction -bzzz
1378 */
1379 jh->b_modified = 1;
Theodore Ts'of6c07ca2013-12-08 21:12:59 -05001380 if (handle->h_buffer_credits <= 0) {
1381 ret = -ENOSPC;
1382 goto out_unlock_bh;
1383 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001384 handle->h_buffer_credits--;
1385 }
1386
1387 /*
1388 * fastpath, to avoid expensive locking. If this buffer is already
1389 * on the running transaction's metadata list there is nothing to do.
1390 * Nobody can take it off again because there is a handle open.
1391 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1392 * result in this test being false, so we go in and take the locks.
1393 */
1394 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1395 JBUFFER_TRACE(jh, "fastpath");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001396 if (unlikely(jh->b_transaction !=
1397 journal->j_running_transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001398 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001399 "jh->b_transaction (%llu, %p, %u) != "
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001400 "journal->j_running_transaction (%p, %u)\n",
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001401 journal->j_devname,
1402 (unsigned long long) bh->b_blocknr,
1403 jh->b_transaction,
1404 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1405 journal->j_running_transaction,
1406 journal->j_running_transaction ?
1407 journal->j_running_transaction->t_tid : 0);
1408 ret = -EINVAL;
1409 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001410 goto out_unlock_bh;
1411 }
1412
1413 set_buffer_jbddirty(bh);
1414
1415 /*
1416 * Metadata already on the current transaction list doesn't
1417 * need to be filed. Metadata on another transaction's list must
1418 * be committing, and will be refiled once the commit completes:
1419 * leave it alone for now.
1420 */
1421 if (jh->b_transaction != transaction) {
1422 JBUFFER_TRACE(jh, "already on other transaction");
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001423 if (unlikely(((jh->b_transaction !=
1424 journal->j_committing_transaction)) ||
1425 (jh->b_next_transaction != transaction))) {
1426 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1427 "bad jh for block %llu: "
1428 "transaction (%p, %u), "
1429 "jh->b_transaction (%p, %u), "
1430 "jh->b_next_transaction (%p, %u), jlist %u\n",
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001431 journal->j_devname,
1432 (unsigned long long) bh->b_blocknr,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001433 transaction, transaction->t_tid,
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001434 jh->b_transaction,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001435 jh->b_transaction ?
1436 jh->b_transaction->t_tid : 0,
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001437 jh->b_next_transaction,
1438 jh->b_next_transaction ?
1439 jh->b_next_transaction->t_tid : 0,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001440 jh->b_jlist);
1441 WARN_ON(1);
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001442 ret = -EINVAL;
1443 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001444 /* And this case is illegal: we can't reuse another
1445 * transaction's data buffer, ever. */
1446 goto out_unlock_bh;
1447 }
1448
1449 /* That test should have eliminated the following case: */
Mingming Cao40191912008-01-28 23:58:27 -05001450 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001451
1452 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1453 spin_lock(&journal->j_list_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001454 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001455 spin_unlock(&journal->j_list_lock);
1456out_unlock_bh:
1457 jbd_unlock_bh_state(bh);
1458out:
1459 JBUFFER_TRACE(jh, "exit");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001460 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001461}
1462
Dave Kleikamp470decc2006-10-11 01:20:57 -07001463/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001464 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001465 * @handle: transaction handle
1466 * @bh: bh to 'forget'
1467 *
1468 * We can only do the bforget if there are no commits pending against the
1469 * buffer. If the buffer is dirty in the current running transaction we
1470 * can safely unlink it.
1471 *
1472 * bh may not be a journalled buffer at all - it may be a non-JBD
1473 * buffer which came off the hashtable. Check for this.
1474 *
1475 * Decrements bh->b_count by one.
1476 *
1477 * Allow this call even if the handle has aborted --- it may be part of
1478 * the caller's cleanup after an abort.
1479 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001480int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001481{
1482 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001483 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001484 struct journal_head *jh;
1485 int drop_reserve = 0;
1486 int err = 0;
Josef Bacik1dfc3222008-04-17 10:38:59 -04001487 int was_modified = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001488
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001489 if (is_handle_aborted(handle))
1490 return -EROFS;
1491 journal = transaction->t_journal;
1492
Dave Kleikamp470decc2006-10-11 01:20:57 -07001493 BUFFER_TRACE(bh, "entry");
1494
1495 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001496
1497 if (!buffer_jbd(bh))
1498 goto not_jbd;
1499 jh = bh2jh(bh);
1500
1501 /* Critical error: attempting to delete a bitmap buffer, maybe?
1502 * Don't do any jbd operations, and return an error. */
1503 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1504 "inconsistent data on disk")) {
1505 err = -EIO;
1506 goto not_jbd;
1507 }
1508
Adam Buchbinder48fc7f72012-09-19 21:48:00 -04001509 /* keep track of whether or not this transaction modified us */
Josef Bacik1dfc3222008-04-17 10:38:59 -04001510 was_modified = jh->b_modified;
1511
Dave Kleikamp470decc2006-10-11 01:20:57 -07001512 /*
1513 * The buffer's going from the transaction, we must drop
1514 * all references -bzzz
1515 */
1516 jh->b_modified = 0;
1517
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001518 if (jh->b_transaction == transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001519 J_ASSERT_JH(jh, !jh->b_frozen_data);
1520
1521 /* If we are forgetting a buffer which is already part
1522 * of this transaction, then we can just drop it from
1523 * the transaction immediately. */
1524 clear_buffer_dirty(bh);
1525 clear_buffer_jbddirty(bh);
1526
1527 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1528
Josef Bacik1dfc3222008-04-17 10:38:59 -04001529 /*
1530 * we only want to drop a reference if this transaction
1531 * modified the buffer
1532 */
1533 if (was_modified)
1534 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001535
1536 /*
1537 * We are no longer going to journal this buffer.
1538 * However, the commit of this transaction is still
1539 * important to the buffer: the delete that we are now
1540 * processing might obsolete an old log entry, so by
1541 * committing, we can satisfy the buffer's checkpoint.
1542 *
1543 * So, if we have a checkpoint on the buffer, we should
1544 * now refile the buffer on our BJ_Forget list so that
1545 * we know to remove the checkpoint after we commit.
1546 */
1547
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001548 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001549 if (jh->b_cp_transaction) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001550 __jbd2_journal_temp_unlink_buffer(jh);
1551 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001552 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001553 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001554 if (!buffer_jbd(bh)) {
1555 spin_unlock(&journal->j_list_lock);
1556 jbd_unlock_bh_state(bh);
1557 __bforget(bh);
1558 goto drop;
1559 }
1560 }
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001561 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001562 } else if (jh->b_transaction) {
1563 J_ASSERT_JH(jh, (jh->b_transaction ==
1564 journal->j_committing_transaction));
1565 /* However, if the buffer is still owned by a prior
1566 * (committing) transaction, we can't drop it yet... */
1567 JBUFFER_TRACE(jh, "belongs to older transaction");
1568 /* ... but we CAN drop it from the new transaction if we
1569 * have also modified it since the original commit. */
1570
1571 if (jh->b_next_transaction) {
1572 J_ASSERT(jh->b_next_transaction == transaction);
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001573 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001574 jh->b_next_transaction = NULL;
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001575 spin_unlock(&journal->j_list_lock);
Josef Bacik1dfc3222008-04-17 10:38:59 -04001576
1577 /*
1578 * only drop a reference if this transaction modified
1579 * the buffer
1580 */
1581 if (was_modified)
1582 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001583 }
1584 }
1585
1586not_jbd:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001587 jbd_unlock_bh_state(bh);
1588 __brelse(bh);
1589drop:
1590 if (drop_reserve) {
1591 /* no need to reserve log space for this block -bzzz */
1592 handle->h_buffer_credits++;
1593 }
1594 return err;
1595}
1596
1597/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001598 * int jbd2_journal_stop() - complete a transaction
Dave Kleikamp470decc2006-10-11 01:20:57 -07001599 * @handle: tranaction to complete.
1600 *
1601 * All done for a particular handle.
1602 *
1603 * There is not much action needed here. We just return any remaining
1604 * buffer credits to the transaction and remove the handle. The only
1605 * complication is that we need to start a commit operation if the
1606 * filesystem is marked for synchronous update.
1607 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001608 * jbd2_journal_stop itself will not usually return an error, but it may
Dave Kleikamp470decc2006-10-11 01:20:57 -07001609 * do so in unusual circumstances. In particular, expect it to
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001610 * return -EIO if a jbd2_journal_abort has been executed since the
Dave Kleikamp470decc2006-10-11 01:20:57 -07001611 * transaction began.
1612 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001613int jbd2_journal_stop(handle_t *handle)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001614{
1615 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001616 journal_t *journal;
1617 int err = 0, wait_for_commit = 0;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001618 tid_t tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001619 pid_t pid;
1620
Lukas Czerner9d506592015-05-14 18:55:18 -04001621 if (!transaction) {
1622 /*
1623 * Handle is already detached from the transaction so
1624 * there is nothing to do other than decrease a refcount,
1625 * or free the handle if refcount drops to zero
1626 */
1627 if (--handle->h_ref > 0) {
1628 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1629 handle->h_ref);
1630 return err;
1631 } else {
1632 if (handle->h_rsv_handle)
1633 jbd2_free_handle(handle->h_rsv_handle);
1634 goto free_and_exit;
1635 }
1636 }
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001637 journal = transaction->t_journal;
1638
Dave Kleikamp470decc2006-10-11 01:20:57 -07001639 J_ASSERT(journal_current_handle() == handle);
1640
1641 if (is_handle_aborted(handle))
1642 err = -EIO;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001643 else
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001644 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001645
1646 if (--handle->h_ref > 0) {
1647 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1648 handle->h_ref);
1649 return err;
1650 }
1651
1652 jbd_debug(4, "Handle %p going down\n", handle);
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001653 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001654 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001655 handle->h_type, handle->h_line_no,
1656 jiffies - handle->h_start_jiffies,
1657 handle->h_sync, handle->h_requested_credits,
1658 (handle->h_requested_credits -
1659 handle->h_buffer_credits));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001660
1661 /*
1662 * Implement synchronous transaction batching. If the handle
1663 * was synchronous, don't force a commit immediately. Let's
Josef Bacike07f7182008-11-26 01:14:26 -05001664 * yield and let another thread piggyback onto this
1665 * transaction. Keep doing that while new threads continue to
1666 * arrive. It doesn't cost much - we're about to run a commit
1667 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1668 * operations by 30x or more...
Dave Kleikamp470decc2006-10-11 01:20:57 -07001669 *
Josef Bacike07f7182008-11-26 01:14:26 -05001670 * We try and optimize the sleep time against what the
1671 * underlying disk can do, instead of having a static sleep
1672 * time. This is useful for the case where our storage is so
1673 * fast that it is more optimal to go ahead and force a flush
1674 * and wait for the transaction to be committed than it is to
1675 * wait for an arbitrary amount of time for new writers to
1676 * join the transaction. We achieve this by measuring how
1677 * long it takes to commit a transaction, and compare it with
1678 * how long this transaction has been running, and if run time
1679 * < commit time then we sleep for the delta and commit. This
1680 * greatly helps super fast disks that would see slowdowns as
1681 * more threads started doing fsyncs.
1682 *
1683 * But don't do this if this process was the most recent one
1684 * to perform a synchronous write. We do this to detect the
1685 * case where a single process is doing a stream of sync
1686 * writes. No point in waiting for joiners in that case.
Eric Sandeen5dd21422014-07-05 19:18:22 -04001687 *
1688 * Setting max_batch_time to 0 disables this completely.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001689 */
1690 pid = current->pid;
Eric Sandeen5dd21422014-07-05 19:18:22 -04001691 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1692 journal->j_max_batch_time) {
Josef Bacike07f7182008-11-26 01:14:26 -05001693 u64 commit_time, trans_time;
1694
Dave Kleikamp470decc2006-10-11 01:20:57 -07001695 journal->j_last_sync_writer = pid;
Josef Bacike07f7182008-11-26 01:14:26 -05001696
Theodore Ts'oa931da62010-08-03 21:35:12 -04001697 read_lock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001698 commit_time = journal->j_average_commit_time;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001699 read_unlock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001700
1701 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1702 transaction->t_start_time));
1703
Theodore Ts'o30773842009-01-03 20:27:38 -05001704 commit_time = max_t(u64, commit_time,
1705 1000*journal->j_min_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001706 commit_time = min_t(u64, commit_time,
Theodore Ts'o30773842009-01-03 20:27:38 -05001707 1000*journal->j_max_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001708
1709 if (trans_time < commit_time) {
1710 ktime_t expires = ktime_add_ns(ktime_get(),
1711 commit_time);
1712 set_current_state(TASK_UNINTERRUPTIBLE);
1713 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1714 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001715 }
1716
Theodore Ts'o70585482009-03-25 23:35:46 -04001717 if (handle->h_sync)
1718 transaction->t_synchronous_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001719 current->journal_info = NULL;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001720 atomic_sub(handle->h_buffer_credits,
1721 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001722
1723 /*
1724 * If the handle is marked SYNC, we need to set another commit
1725 * going! We also want to force a commit if the current
1726 * transaction is occupying too much of the log, or if the
1727 * transaction is too old now.
1728 */
1729 if (handle->h_sync ||
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001730 (atomic_read(&transaction->t_outstanding_credits) >
1731 journal->j_max_transaction_buffers) ||
1732 time_after_eq(jiffies, transaction->t_expires)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001733 /* Do this even for aborted journals: an abort still
1734 * completes the commit thread, it just doesn't write
1735 * anything to disk. */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001736
Dave Kleikamp470decc2006-10-11 01:20:57 -07001737 jbd_debug(2, "transaction too old, requesting commit for "
1738 "handle %p\n", handle);
1739 /* This is non-blocking */
Theodore Ts'oc35a56a2010-05-16 05:00:00 -04001740 jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001741
1742 /*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001743 * Special case: JBD2_SYNC synchronous updates require us
Dave Kleikamp470decc2006-10-11 01:20:57 -07001744 * to wait for the commit to complete.
1745 */
1746 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001747 wait_for_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001748 }
1749
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001750 /*
1751 * Once we drop t_updates, if it goes to zero the transaction
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001752 * could start committing on us and eventually disappear. So
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001753 * once we do this, we must not dereference transaction
1754 * pointer again.
1755 */
1756 tid = transaction->t_tid;
1757 if (atomic_dec_and_test(&transaction->t_updates)) {
1758 wake_up(&journal->j_wait_updates);
1759 if (journal->j_barrier_count)
1760 wake_up(&journal->j_wait_transaction_locked);
1761 }
1762
1763 if (wait_for_commit)
1764 err = jbd2_log_wait_commit(journal, tid);
1765
Ingo Molnar3295f0e2008-08-11 10:30:30 +02001766 lock_map_release(&handle->h_lockdep_map);
Mingming Cao7b751062008-01-28 23:58:27 -05001767
Jan Kara8f7d89f2013-06-04 12:35:11 -04001768 if (handle->h_rsv_handle)
1769 jbd2_journal_free_reserved(handle->h_rsv_handle);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001770free_and_exit:
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001771 jbd2_free_handle(handle);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001772 return err;
1773}
1774
Dave Kleikamp470decc2006-10-11 01:20:57 -07001775/*
1776 *
1777 * List management code snippets: various functions for manipulating the
1778 * transaction buffer lists.
1779 *
1780 */
1781
1782/*
1783 * Append a buffer to a transaction list, given the transaction's list head
1784 * pointer.
1785 *
1786 * j_list_lock is held.
1787 *
1788 * jbd_lock_bh_state(jh2bh(jh)) is held.
1789 */
1790
1791static inline void
1792__blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1793{
1794 if (!*list) {
1795 jh->b_tnext = jh->b_tprev = jh;
1796 *list = jh;
1797 } else {
1798 /* Insert at the tail of the list to preserve order */
1799 struct journal_head *first = *list, *last = first->b_tprev;
1800 jh->b_tprev = last;
1801 jh->b_tnext = first;
1802 last->b_tnext = first->b_tprev = jh;
1803 }
1804}
1805
1806/*
1807 * Remove a buffer from a transaction list, given the transaction's list
1808 * head pointer.
1809 *
1810 * Called with j_list_lock held, and the journal may not be locked.
1811 *
1812 * jbd_lock_bh_state(jh2bh(jh)) is held.
1813 */
1814
1815static inline void
1816__blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1817{
1818 if (*list == jh) {
1819 *list = jh->b_tnext;
1820 if (*list == jh)
1821 *list = NULL;
1822 }
1823 jh->b_tprev->b_tnext = jh->b_tnext;
1824 jh->b_tnext->b_tprev = jh->b_tprev;
1825}
1826
1827/*
1828 * Remove a buffer from the appropriate transaction list.
1829 *
1830 * Note that this function can *change* the value of
Jan Karaf5113ef2013-06-04 12:01:45 -04001831 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1832 * t_reserved_list. If the caller is holding onto a copy of one of these
1833 * pointers, it could go bad. Generally the caller needs to re-read the
1834 * pointer from the transaction_t.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001835 *
Jan Kara5bebccf2012-03-13 22:25:06 -04001836 * Called under j_list_lock.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001837 */
Jan Kara5bebccf2012-03-13 22:25:06 -04001838static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001839{
1840 struct journal_head **list = NULL;
1841 transaction_t *transaction;
1842 struct buffer_head *bh = jh2bh(jh);
1843
1844 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1845 transaction = jh->b_transaction;
1846 if (transaction)
1847 assert_spin_locked(&transaction->t_journal->j_list_lock);
1848
1849 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1850 if (jh->b_jlist != BJ_None)
Mingming Cao40191912008-01-28 23:58:27 -05001851 J_ASSERT_JH(jh, transaction != NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001852
1853 switch (jh->b_jlist) {
1854 case BJ_None:
1855 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001856 case BJ_Metadata:
1857 transaction->t_nr_buffers--;
1858 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1859 list = &transaction->t_buffers;
1860 break;
1861 case BJ_Forget:
1862 list = &transaction->t_forget;
1863 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001864 case BJ_Shadow:
1865 list = &transaction->t_shadow_list;
1866 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001867 case BJ_Reserved:
1868 list = &transaction->t_reserved_list;
1869 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001870 }
1871
1872 __blist_del_buffer(list, jh);
1873 jh->b_jlist = BJ_None;
1874 if (test_clear_buffer_jbddirty(bh))
1875 mark_buffer_dirty(bh); /* Expose it to the VM */
1876}
1877
Jan Karade1b7942011-06-13 15:38:22 -04001878/*
1879 * Remove buffer from all transactions.
1880 *
1881 * Called with bh_state lock and j_list_lock
1882 *
1883 * jh and bh may be already freed when this function returns.
1884 */
1885static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001886{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001887 __jbd2_journal_temp_unlink_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001888 jh->b_transaction = NULL;
Jan Karade1b7942011-06-13 15:38:22 -04001889 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001890}
1891
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001892void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001893{
Jan Karade1b7942011-06-13 15:38:22 -04001894 struct buffer_head *bh = jh2bh(jh);
1895
1896 /* Get reference so that buffer cannot be freed before we unlock it */
1897 get_bh(bh);
1898 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001899 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001900 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001901 spin_unlock(&journal->j_list_lock);
Jan Karade1b7942011-06-13 15:38:22 -04001902 jbd_unlock_bh_state(bh);
1903 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001904}
1905
1906/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001907 * Called from jbd2_journal_try_to_free_buffers().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001908 *
1909 * Called under jbd_lock_bh_state(bh)
1910 */
1911static void
1912__journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1913{
1914 struct journal_head *jh;
1915
1916 jh = bh2jh(bh);
1917
1918 if (buffer_locked(bh) || buffer_dirty(bh))
1919 goto out;
1920
Theodore Ts'od2eb0b92014-03-09 00:07:19 -05001921 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001922 goto out;
1923
1924 spin_lock(&journal->j_list_lock);
Theodore Ts'od2eb0b92014-03-09 00:07:19 -05001925 if (jh->b_cp_transaction != NULL) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001926 /* written-back checkpointed metadata buffer */
Jan Karac254c9e2012-03-13 22:27:44 -04001927 JBUFFER_TRACE(jh, "remove from checkpoint list");
1928 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001929 }
1930 spin_unlock(&journal->j_list_lock);
1931out:
1932 return;
1933}
1934
Dave Kleikamp470decc2006-10-11 01:20:57 -07001935/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001936 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001937 * @journal: journal for operation
1938 * @page: to try and free
Mingming Cao530576b2008-07-13 21:06:39 -04001939 * @gfp_mask: we use the mask to detect how hard should we try to release
Mel Gormand0164ad2015-11-06 16:28:21 -08001940 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1941 * code to release the buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001942 *
1943 *
1944 * For all the buffers on this page,
1945 * if they are fully written out ordered data, move them onto BUF_CLEAN
1946 * so try_to_free_buffers() can reap them.
1947 *
1948 * This function returns non-zero if we wish try_to_free_buffers()
1949 * to be called. We do this if the page is releasable by try_to_free_buffers().
1950 * We also do it if the page has locked or dirty buffers and the caller wants
1951 * us to perform sync or async writeout.
1952 *
1953 * This complicates JBD locking somewhat. We aren't protected by the
1954 * BKL here. We wish to remove the buffer from its committing or
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001955 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001956 *
1957 * This may *change* the value of transaction_t->t_datalist, so anyone
1958 * who looks at t_datalist needs to lock against this function.
1959 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001960 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1961 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
Dave Kleikamp470decc2006-10-11 01:20:57 -07001962 * will come out of the lock with the buffer dirty, which makes it
1963 * ineligible for release here.
1964 *
1965 * Who else is affected by this? hmm... Really the only contender
1966 * is do_get_write_access() - it could be looking at the buffer while
1967 * journal_try_to_free_buffer() is changing its state. But that
1968 * cannot happen because we never reallocate freed data as metadata
1969 * while the data is part of a transaction. Yes?
Mingming Cao530576b2008-07-13 21:06:39 -04001970 *
1971 * Return 0 on failure, 1 on success
Dave Kleikamp470decc2006-10-11 01:20:57 -07001972 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001973int jbd2_journal_try_to_free_buffers(journal_t *journal,
Mingming Cao530576b2008-07-13 21:06:39 -04001974 struct page *page, gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001975{
1976 struct buffer_head *head;
1977 struct buffer_head *bh;
1978 int ret = 0;
1979
1980 J_ASSERT(PageLocked(page));
1981
1982 head = page_buffers(page);
1983 bh = head;
1984 do {
1985 struct journal_head *jh;
1986
1987 /*
1988 * We take our own ref against the journal_head here to avoid
1989 * having to add tons of locking around each instance of
Mingming Cao530576b2008-07-13 21:06:39 -04001990 * jbd2_journal_put_journal_head().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001991 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001992 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001993 if (!jh)
1994 continue;
1995
1996 jbd_lock_bh_state(bh);
1997 __journal_try_to_free_buffer(journal, bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001998 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001999 jbd_unlock_bh_state(bh);
2000 if (buffer_jbd(bh))
2001 goto busy;
2002 } while ((bh = bh->b_this_page) != head);
Mingming Cao530576b2008-07-13 21:06:39 -04002003
Dave Kleikamp470decc2006-10-11 01:20:57 -07002004 ret = try_to_free_buffers(page);
Mingming Cao530576b2008-07-13 21:06:39 -04002005
Dave Kleikamp470decc2006-10-11 01:20:57 -07002006busy:
2007 return ret;
2008}
2009
2010/*
2011 * This buffer is no longer needed. If it is on an older transaction's
2012 * checkpoint list we need to record it on this transaction's forget list
2013 * to pin this buffer (and hence its checkpointing transaction) down until
2014 * this transaction commits. If the buffer isn't on a checkpoint list, we
2015 * release it.
2016 * Returns non-zero if JBD no longer has an interest in the buffer.
2017 *
2018 * Called under j_list_lock.
2019 *
2020 * Called under jbd_lock_bh_state(bh).
2021 */
2022static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2023{
2024 int may_free = 1;
2025 struct buffer_head *bh = jh2bh(jh);
2026
Dave Kleikamp470decc2006-10-11 01:20:57 -07002027 if (jh->b_cp_transaction) {
2028 JBUFFER_TRACE(jh, "on running+cp transaction");
Jan Karade1b7942011-06-13 15:38:22 -04002029 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karaf91d1d02009-07-13 16:16:20 -04002030 /*
2031 * We don't want to write the buffer anymore, clear the
2032 * bit so that we don't confuse checks in
2033 * __journal_file_buffer
2034 */
2035 clear_buffer_dirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002036 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002037 may_free = 0;
2038 } else {
2039 JBUFFER_TRACE(jh, "on running transaction");
Jan Karade1b7942011-06-13 15:38:22 -04002040 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002041 }
2042 return may_free;
2043}
2044
2045/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002046 * jbd2_journal_invalidatepage
Dave Kleikamp470decc2006-10-11 01:20:57 -07002047 *
2048 * This code is tricky. It has a number of cases to deal with.
2049 *
2050 * There are two invariants which this code relies on:
2051 *
2052 * i_size must be updated on disk before we start calling invalidatepage on the
2053 * data.
2054 *
2055 * This is done in ext3 by defining an ext3_setattr method which
2056 * updates i_size before truncate gets going. By maintaining this
2057 * invariant, we can be sure that it is safe to throw away any buffers
2058 * attached to the current transaction: once the transaction commits,
2059 * we know that the data will not be needed.
2060 *
2061 * Note however that we can *not* throw away data belonging to the
2062 * previous, committing transaction!
2063 *
2064 * Any disk blocks which *are* part of the previous, committing
2065 * transaction (and which therefore cannot be discarded immediately) are
2066 * not going to be reused in the new running transaction
2067 *
2068 * The bitmap committed_data images guarantee this: any block which is
2069 * allocated in one transaction and removed in the next will be marked
2070 * as in-use in the committed_data bitmap, so cannot be reused until
2071 * the next transaction to delete the block commits. This means that
2072 * leaving committing buffers dirty is quite safe: the disk blocks
2073 * cannot be reallocated to a different file and so buffer aliasing is
2074 * not possible.
2075 *
2076 *
2077 * The above applies mainly to ordered data mode. In writeback mode we
2078 * don't make guarantees about the order in which data hits disk --- in
2079 * particular we don't guarantee that new dirty data is flushed before
2080 * transaction commit --- so it is always safe just to discard data
2081 * immediately in that mode. --sct
2082 */
2083
2084/*
2085 * The journal_unmap_buffer helper function returns zero if the buffer
2086 * concerned remains pinned as an anonymous buffer belonging to an older
2087 * transaction.
2088 *
2089 * We're outside-transaction here. Either or both of j_running_transaction
2090 * and j_committing_transaction may be NULL.
2091 */
Jan Karab794e7a2012-09-26 23:11:13 -04002092static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2093 int partial_page)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002094{
2095 transaction_t *transaction;
2096 struct journal_head *jh;
2097 int may_free = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002098
2099 BUFFER_TRACE(bh, "entry");
2100
2101 /*
2102 * It is safe to proceed here without the j_list_lock because the
2103 * buffers cannot be stolen by try_to_free_buffers as long as we are
2104 * holding the page lock. --sct
2105 */
2106
2107 if (!buffer_jbd(bh))
2108 goto zap_buffer_unlocked;
2109
Jan Kara87c89c22008-07-11 19:27:31 -04002110 /* OK, we have data buffer in journaled mode */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002111 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002112 jbd_lock_bh_state(bh);
2113 spin_lock(&journal->j_list_lock);
2114
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002115 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002116 if (!jh)
2117 goto zap_buffer_no_jh;
2118
dingdinghuaba869022010-02-15 16:35:42 -05002119 /*
2120 * We cannot remove the buffer from checkpoint lists until the
2121 * transaction adding inode to orphan list (let's call it T)
2122 * is committed. Otherwise if the transaction changing the
2123 * buffer would be cleaned from the journal before T is
2124 * committed, a crash will cause that the correct contents of
2125 * the buffer will be lost. On the other hand we have to
2126 * clear the buffer dirty bit at latest at the moment when the
2127 * transaction marking the buffer as freed in the filesystem
2128 * structures is committed because from that moment on the
Jan Karab794e7a2012-09-26 23:11:13 -04002129 * block can be reallocated and used by a different page.
dingdinghuaba869022010-02-15 16:35:42 -05002130 * Since the block hasn't been freed yet but the inode has
2131 * already been added to orphan list, it is safe for us to add
2132 * the buffer to BJ_Forget list of the newest transaction.
Jan Karab794e7a2012-09-26 23:11:13 -04002133 *
2134 * Also we have to clear buffer_mapped flag of a truncated buffer
2135 * because the buffer_head may be attached to the page straddling
2136 * i_size (can happen only when blocksize < pagesize) and thus the
2137 * buffer_head can be reused when the file is extended again. So we end
2138 * up keeping around invalidated buffers attached to transactions'
2139 * BJ_Forget list just to stop checkpointing code from cleaning up
2140 * the transaction this buffer was modified in.
dingdinghuaba869022010-02-15 16:35:42 -05002141 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002142 transaction = jh->b_transaction;
2143 if (transaction == NULL) {
2144 /* First case: not on any transaction. If it
2145 * has no checkpoint link, then we can zap it:
2146 * it's a writeback-mode buffer so we don't care
2147 * if it hits disk safely. */
2148 if (!jh->b_cp_transaction) {
2149 JBUFFER_TRACE(jh, "not on any transaction: zap");
2150 goto zap_buffer;
2151 }
2152
2153 if (!buffer_dirty(bh)) {
2154 /* bdflush has written it. We can drop it now */
2155 goto zap_buffer;
2156 }
2157
2158 /* OK, it must be in the journal but still not
2159 * written fully to disk: it's metadata or
2160 * journaled data... */
2161
2162 if (journal->j_running_transaction) {
2163 /* ... and once the current transaction has
2164 * committed, the buffer won't be needed any
2165 * longer. */
2166 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
Jan Karab794e7a2012-09-26 23:11:13 -04002167 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002168 journal->j_running_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002169 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002170 } else {
2171 /* There is no currently-running transaction. So the
2172 * orphan record which we wrote for this file must have
2173 * passed into commit. We must attach this buffer to
2174 * the committing transaction, if it exists. */
2175 if (journal->j_committing_transaction) {
2176 JBUFFER_TRACE(jh, "give to committing trans");
Jan Karab794e7a2012-09-26 23:11:13 -04002177 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002178 journal->j_committing_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002179 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002180 } else {
2181 /* The orphan record's transaction has
2182 * committed. We can cleanse this buffer */
2183 clear_buffer_jbddirty(bh);
2184 goto zap_buffer;
2185 }
2186 }
2187 } else if (transaction == journal->j_committing_transaction) {
Eric Sandeen9b579882006-10-28 10:38:28 -07002188 JBUFFER_TRACE(jh, "on committing transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002189 /*
dingdinghuaba869022010-02-15 16:35:42 -05002190 * The buffer is committing, we simply cannot touch
Jan Karab794e7a2012-09-26 23:11:13 -04002191 * it. If the page is straddling i_size we have to wait
2192 * for commit and try again.
2193 */
2194 if (partial_page) {
Jan Karab794e7a2012-09-26 23:11:13 -04002195 jbd2_journal_put_journal_head(jh);
2196 spin_unlock(&journal->j_list_lock);
2197 jbd_unlock_bh_state(bh);
2198 write_unlock(&journal->j_state_lock);
Jan Kara53e87262012-12-25 13:29:52 -05002199 return -EBUSY;
Jan Karab794e7a2012-09-26 23:11:13 -04002200 }
2201 /*
2202 * OK, buffer won't be reachable after truncate. We just set
2203 * j_next_transaction to the running transaction (if there is
2204 * one) and mark buffer as freed so that commit code knows it
2205 * should clear dirty bits when it is done with the buffer.
dingdinghuaba869022010-02-15 16:35:42 -05002206 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002207 set_buffer_freed(bh);
dingdinghuaba869022010-02-15 16:35:42 -05002208 if (journal->j_running_transaction && buffer_jbddirty(bh))
2209 jh->b_next_transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002210 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002211 spin_unlock(&journal->j_list_lock);
2212 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002213 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002214 return 0;
2215 } else {
2216 /* Good, the buffer belongs to the running transaction.
2217 * We are writing our own transaction's data, not any
2218 * previous one's, so it is safe to throw it away
2219 * (remember that we expect the filesystem to have set
2220 * i_size already for this truncate so recovery will not
2221 * expose the disk blocks we are discarding here.) */
2222 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
Eric Sandeen9b579882006-10-28 10:38:28 -07002223 JBUFFER_TRACE(jh, "on running transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002224 may_free = __dispose_buffer(jh, transaction);
2225 }
2226
2227zap_buffer:
Jan Karab794e7a2012-09-26 23:11:13 -04002228 /*
2229 * This is tricky. Although the buffer is truncated, it may be reused
2230 * if blocksize < pagesize and it is attached to the page straddling
2231 * EOF. Since the buffer might have been added to BJ_Forget list of the
2232 * running transaction, journal_get_write_access() won't clear
2233 * b_modified and credit accounting gets confused. So clear b_modified
2234 * here.
2235 */
2236 jh->b_modified = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002237 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002238zap_buffer_no_jh:
2239 spin_unlock(&journal->j_list_lock);
2240 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002241 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002242zap_buffer_unlocked:
2243 clear_buffer_dirty(bh);
2244 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2245 clear_buffer_mapped(bh);
2246 clear_buffer_req(bh);
2247 clear_buffer_new(bh);
Eric Sandeen15291162012-02-20 17:53:01 -05002248 clear_buffer_delay(bh);
2249 clear_buffer_unwritten(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002250 bh->b_bdev = NULL;
2251 return may_free;
2252}
2253
2254/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002255 * void jbd2_journal_invalidatepage()
Dave Kleikamp470decc2006-10-11 01:20:57 -07002256 * @journal: journal to use for flush...
2257 * @page: page to flush
Lukas Czerner259709b2013-05-21 23:20:03 -04002258 * @offset: start of the range to invalidate
2259 * @length: length of the range to invalidate
Dave Kleikamp470decc2006-10-11 01:20:57 -07002260 *
Lukas Czerner259709b2013-05-21 23:20:03 -04002261 * Reap page buffers containing data after in the specified range in page.
2262 * Can return -EBUSY if buffers are part of the committing transaction and
2263 * the page is straddling i_size. Caller then has to wait for current commit
2264 * and try again.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002265 */
Jan Kara53e87262012-12-25 13:29:52 -05002266int jbd2_journal_invalidatepage(journal_t *journal,
2267 struct page *page,
Lukas Czerner259709b2013-05-21 23:20:03 -04002268 unsigned int offset,
2269 unsigned int length)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002270{
2271 struct buffer_head *head, *bh, *next;
Lukas Czerner259709b2013-05-21 23:20:03 -04002272 unsigned int stop = offset + length;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002273 unsigned int curr_off = 0;
Lukas Czerner259709b2013-05-21 23:20:03 -04002274 int partial_page = (offset || length < PAGE_CACHE_SIZE);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002275 int may_free = 1;
Jan Kara53e87262012-12-25 13:29:52 -05002276 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002277
2278 if (!PageLocked(page))
2279 BUG();
2280 if (!page_has_buffers(page))
Jan Kara53e87262012-12-25 13:29:52 -05002281 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002282
Lukas Czerner259709b2013-05-21 23:20:03 -04002283 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
2284
Dave Kleikamp470decc2006-10-11 01:20:57 -07002285 /* We will potentially be playing with lists other than just the
2286 * data lists (especially for journaled data mode), so be
2287 * cautious in our locking. */
2288
2289 head = bh = page_buffers(page);
2290 do {
2291 unsigned int next_off = curr_off + bh->b_size;
2292 next = bh->b_this_page;
2293
Lukas Czerner259709b2013-05-21 23:20:03 -04002294 if (next_off > stop)
2295 return 0;
2296
Dave Kleikamp470decc2006-10-11 01:20:57 -07002297 if (offset <= curr_off) {
2298 /* This block is wholly outside the truncation point */
2299 lock_buffer(bh);
Lukas Czerner259709b2013-05-21 23:20:03 -04002300 ret = journal_unmap_buffer(journal, bh, partial_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002301 unlock_buffer(bh);
Jan Kara53e87262012-12-25 13:29:52 -05002302 if (ret < 0)
2303 return ret;
2304 may_free &= ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002305 }
2306 curr_off = next_off;
2307 bh = next;
2308
2309 } while (bh != head);
2310
Lukas Czerner259709b2013-05-21 23:20:03 -04002311 if (!partial_page) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002312 if (may_free && try_to_free_buffers(page))
2313 J_ASSERT(!page_has_buffers(page));
2314 }
Jan Kara53e87262012-12-25 13:29:52 -05002315 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002316}
2317
2318/*
2319 * File a buffer on the given transaction list.
2320 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002321void __jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002322 transaction_t *transaction, int jlist)
2323{
2324 struct journal_head **list = NULL;
2325 int was_dirty = 0;
2326 struct buffer_head *bh = jh2bh(jh);
2327
2328 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2329 assert_spin_locked(&transaction->t_journal->j_list_lock);
2330
2331 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2332 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
Mingming Cao40191912008-01-28 23:58:27 -05002333 jh->b_transaction == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002334
2335 if (jh->b_transaction && jh->b_jlist == jlist)
2336 return;
2337
Dave Kleikamp470decc2006-10-11 01:20:57 -07002338 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2339 jlist == BJ_Shadow || jlist == BJ_Forget) {
Jan Karaf91d1d02009-07-13 16:16:20 -04002340 /*
2341 * For metadata buffers, we track dirty bit in buffer_jbddirty
2342 * instead of buffer_dirty. We should not see a dirty bit set
2343 * here because we clear it in do_get_write_access but e.g.
2344 * tune2fs can modify the sb and set the dirty bit at any time
2345 * so we try to gracefully handle that.
2346 */
2347 if (buffer_dirty(bh))
2348 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002349 if (test_clear_buffer_dirty(bh) ||
2350 test_clear_buffer_jbddirty(bh))
2351 was_dirty = 1;
2352 }
2353
2354 if (jh->b_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002355 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002356 else
2357 jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002358 jh->b_transaction = transaction;
2359
2360 switch (jlist) {
2361 case BJ_None:
2362 J_ASSERT_JH(jh, !jh->b_committed_data);
2363 J_ASSERT_JH(jh, !jh->b_frozen_data);
2364 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002365 case BJ_Metadata:
2366 transaction->t_nr_buffers++;
2367 list = &transaction->t_buffers;
2368 break;
2369 case BJ_Forget:
2370 list = &transaction->t_forget;
2371 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002372 case BJ_Shadow:
2373 list = &transaction->t_shadow_list;
2374 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002375 case BJ_Reserved:
2376 list = &transaction->t_reserved_list;
2377 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002378 }
2379
2380 __blist_add_buffer(list, jh);
2381 jh->b_jlist = jlist;
2382
2383 if (was_dirty)
2384 set_buffer_jbddirty(bh);
2385}
2386
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002387void jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002388 transaction_t *transaction, int jlist)
2389{
2390 jbd_lock_bh_state(jh2bh(jh));
2391 spin_lock(&transaction->t_journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002392 __jbd2_journal_file_buffer(jh, transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002393 spin_unlock(&transaction->t_journal->j_list_lock);
2394 jbd_unlock_bh_state(jh2bh(jh));
2395}
2396
2397/*
2398 * Remove a buffer from its current buffer list in preparation for
2399 * dropping it from its current transaction entirely. If the buffer has
2400 * already started to be used by a subsequent transaction, refile the
2401 * buffer on that transaction's metadata list.
2402 *
Jan Karade1b7942011-06-13 15:38:22 -04002403 * Called under j_list_lock
Dave Kleikamp470decc2006-10-11 01:20:57 -07002404 * Called under jbd_lock_bh_state(jh2bh(jh))
Jan Karade1b7942011-06-13 15:38:22 -04002405 *
2406 * jh and bh may be already free when this function returns
Dave Kleikamp470decc2006-10-11 01:20:57 -07002407 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002408void __jbd2_journal_refile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002409{
dingdinghuaba869022010-02-15 16:35:42 -05002410 int was_dirty, jlist;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002411 struct buffer_head *bh = jh2bh(jh);
2412
2413 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2414 if (jh->b_transaction)
2415 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2416
2417 /* If the buffer is now unused, just drop it. */
2418 if (jh->b_next_transaction == NULL) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002419 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002420 return;
2421 }
2422
2423 /*
2424 * It has been modified by a later transaction: add it to the new
2425 * transaction's metadata list.
2426 */
2427
2428 was_dirty = test_clear_buffer_jbddirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002429 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002430 /*
2431 * We set b_transaction here because b_next_transaction will inherit
2432 * our jh reference and thus __jbd2_journal_file_buffer() must not
2433 * take a new one.
2434 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002435 jh->b_transaction = jh->b_next_transaction;
2436 jh->b_next_transaction = NULL;
dingdinghuaba869022010-02-15 16:35:42 -05002437 if (buffer_freed(bh))
2438 jlist = BJ_Forget;
2439 else if (jh->b_modified)
2440 jlist = BJ_Metadata;
2441 else
2442 jlist = BJ_Reserved;
2443 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002444 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2445
2446 if (was_dirty)
2447 set_buffer_jbddirty(bh);
2448}
2449
2450/*
Jan Karade1b7942011-06-13 15:38:22 -04002451 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2452 * bh reference so that we can safely unlock bh.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002453 *
Jan Karade1b7942011-06-13 15:38:22 -04002454 * The jh and bh may be freed by this call.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002455 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002456void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002457{
2458 struct buffer_head *bh = jh2bh(jh);
2459
Jan Karade1b7942011-06-13 15:38:22 -04002460 /* Get reference so that buffer cannot be freed before we unlock it */
2461 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002462 jbd_lock_bh_state(bh);
2463 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002464 __jbd2_journal_refile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002465 jbd_unlock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002466 spin_unlock(&journal->j_list_lock);
2467 __brelse(bh);
2468}
Jan Karac851ed52008-07-11 19:27:31 -04002469
2470/*
2471 * File inode in the inode list of the handle's transaction
2472 */
2473int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
2474{
2475 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002476 journal_t *journal;
Jan Karac851ed52008-07-11 19:27:31 -04002477
2478 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002479 return -EROFS;
2480 journal = transaction->t_journal;
Jan Karac851ed52008-07-11 19:27:31 -04002481
2482 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2483 transaction->t_tid);
2484
2485 /*
2486 * First check whether inode isn't already on the transaction's
2487 * lists without taking the lock. Note that this check is safe
2488 * without the lock as we cannot race with somebody removing inode
2489 * from the transaction. The reason is that we remove inode from the
2490 * transaction only in journal_release_jbd_inode() and when we commit
2491 * the transaction. We are guarded from the first case by holding
2492 * a reference to the inode. We are safe against the second case
2493 * because if jinode->i_transaction == transaction, commit code
2494 * cannot touch the transaction because we hold reference to it,
2495 * and if jinode->i_next_transaction == transaction, commit code
2496 * will only file the inode where we want it.
2497 */
2498 if (jinode->i_transaction == transaction ||
2499 jinode->i_next_transaction == transaction)
2500 return 0;
2501
2502 spin_lock(&journal->j_list_lock);
2503
2504 if (jinode->i_transaction == transaction ||
2505 jinode->i_next_transaction == transaction)
2506 goto done;
2507
Jan Kara81be12c2011-05-24 11:52:40 -04002508 /*
2509 * We only ever set this variable to 1 so the test is safe. Since
2510 * t_need_data_flush is likely to be set, we do the test to save some
2511 * cacheline bouncing
2512 */
2513 if (!transaction->t_need_data_flush)
2514 transaction->t_need_data_flush = 1;
Jan Karac851ed52008-07-11 19:27:31 -04002515 /* On some different transaction's list - should be
2516 * the committing one */
2517 if (jinode->i_transaction) {
2518 J_ASSERT(jinode->i_next_transaction == NULL);
2519 J_ASSERT(jinode->i_transaction ==
2520 journal->j_committing_transaction);
2521 jinode->i_next_transaction = transaction;
2522 goto done;
2523 }
2524 /* Not on any transaction list... */
2525 J_ASSERT(!jinode->i_next_transaction);
2526 jinode->i_transaction = transaction;
2527 list_add(&jinode->i_list, &transaction->t_inode_list);
2528done:
2529 spin_unlock(&journal->j_list_lock);
2530
2531 return 0;
2532}
2533
2534/*
Jan Kara7f5aa212009-02-10 11:15:34 -05002535 * File truncate and transaction commit interact with each other in a
2536 * non-trivial way. If a transaction writing data block A is
2537 * committing, we cannot discard the data by truncate until we have
2538 * written them. Otherwise if we crashed after the transaction with
2539 * write has committed but before the transaction with truncate has
2540 * committed, we could see stale data in block A. This function is a
2541 * helper to solve this problem. It starts writeout of the truncated
2542 * part in case it is in the committing transaction.
2543 *
2544 * Filesystem code must call this function when inode is journaled in
2545 * ordered mode before truncation happens and after the inode has been
2546 * placed on orphan list with the new inode size. The second condition
2547 * avoids the race that someone writes new data and we start
2548 * committing the transaction after this function has been called but
2549 * before a transaction for truncate is started (and furthermore it
2550 * allows us to optimize the case where the addition to orphan list
2551 * happens in the same transaction as write --- we don't have to write
2552 * any data in such case).
Jan Karac851ed52008-07-11 19:27:31 -04002553 */
Jan Kara7f5aa212009-02-10 11:15:34 -05002554int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2555 struct jbd2_inode *jinode,
Jan Karac851ed52008-07-11 19:27:31 -04002556 loff_t new_size)
2557{
Jan Kara7f5aa212009-02-10 11:15:34 -05002558 transaction_t *inode_trans, *commit_trans;
Jan Karac851ed52008-07-11 19:27:31 -04002559 int ret = 0;
2560
Jan Kara7f5aa212009-02-10 11:15:34 -05002561 /* This is a quick check to avoid locking if not necessary */
2562 if (!jinode->i_transaction)
Jan Karac851ed52008-07-11 19:27:31 -04002563 goto out;
Jan Kara7f5aa212009-02-10 11:15:34 -05002564 /* Locks are here just to force reading of recent values, it is
2565 * enough that the transaction was not committing before we started
2566 * a transaction adding the inode to orphan list */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002567 read_lock(&journal->j_state_lock);
Jan Karac851ed52008-07-11 19:27:31 -04002568 commit_trans = journal->j_committing_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002569 read_unlock(&journal->j_state_lock);
Jan Kara7f5aa212009-02-10 11:15:34 -05002570 spin_lock(&journal->j_list_lock);
2571 inode_trans = jinode->i_transaction;
2572 spin_unlock(&journal->j_list_lock);
2573 if (inode_trans == commit_trans) {
2574 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
Jan Karac851ed52008-07-11 19:27:31 -04002575 new_size, LLONG_MAX);
2576 if (ret)
2577 jbd2_journal_abort(journal, ret);
2578 }
2579out:
2580 return ret;
2581}