blob: 67c103867bf8d5d27e84a4eb2ad2c8a09b22d328 [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 printk(KERN_WARNING
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400768 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
Jan Karaf91d1d02009-07-13 16:16:20 -0400769 "There's a risk of filesystem corruption in case of system "
770 "crash.\n",
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400771 bh->b_bdev, (unsigned long long)bh->b_blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700772}
773
Jan Karaee57aba2015-06-08 12:39:07 -0400774/* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
775static void jbd2_freeze_jh_data(struct journal_head *jh)
776{
777 struct page *page;
778 int offset;
779 char *source;
780 struct buffer_head *bh = jh2bh(jh);
781
782 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
783 page = bh->b_page;
784 offset = offset_in_page(bh->b_data);
785 source = kmap_atomic(page);
786 /* Fire data frozen trigger just before we copy the data */
787 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
788 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
789 kunmap_atomic(source);
790
791 /*
792 * Now that the frozen data is saved off, we need to store any matching
793 * triggers.
794 */
795 jh->b_frozen_triggers = jh->b_triggers;
796}
797
Dave Kleikamp470decc2006-10-11 01:20:57 -0700798/*
799 * If the buffer is already part of the current transaction, then there
800 * is nothing we need to do. If it is already part of a prior
801 * transaction which we are still committing to disk, then we need to
802 * make sure that we do not overwrite the old copy: we do copy-out to
803 * preserve the copy going to disk. We also account the buffer against
804 * the handle's metadata buffer credits (unless the buffer is already
805 * part of the transaction, that is).
806 *
807 */
808static int
809do_get_write_access(handle_t *handle, struct journal_head *jh,
810 int force_copy)
811{
812 struct buffer_head *bh;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400813 transaction_t *transaction = handle->h_transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700814 journal_t *journal;
815 int error;
816 char *frozen_buffer = NULL;
Theodore Ts'of783f092013-04-21 16:47:54 -0400817 unsigned long start_lock, time_lock;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700818
819 if (is_handle_aborted(handle))
820 return -EROFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700821 journal = transaction->t_journal;
822
Theodore Ts'ocfef2c62010-12-18 13:07:34 -0500823 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700824
825 JBUFFER_TRACE(jh, "entry");
826repeat:
827 bh = jh2bh(jh);
828
829 /* @@@ Need to check for errors here at some point. */
830
Theodore Ts'of783f092013-04-21 16:47:54 -0400831 start_lock = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700832 lock_buffer(bh);
833 jbd_lock_bh_state(bh);
834
Theodore Ts'of783f092013-04-21 16:47:54 -0400835 /* If it takes too long to lock the buffer, trace it */
836 time_lock = jbd2_time_diff(start_lock, jiffies);
837 if (time_lock > HZ/10)
838 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
839 jiffies_to_msecs(time_lock));
840
Dave Kleikamp470decc2006-10-11 01:20:57 -0700841 /* We now hold the buffer lock so it is safe to query the buffer
842 * state. Is the buffer dirty?
843 *
844 * If so, there are two possibilities. The buffer may be
845 * non-journaled, and undergoing a quite legitimate writeback.
846 * Otherwise, it is journaled, and we don't expect dirty buffers
847 * in that state (the buffers should be marked JBD_Dirty
848 * instead.) So either the IO is being done under our own
849 * control and this is a bug, or it's a third party IO such as
850 * dump(8) (which may leave the buffer scheduled for read ---
851 * ie. locked but not dirty) or tune2fs (which may actually have
852 * the buffer dirtied, ugh.) */
853
854 if (buffer_dirty(bh)) {
855 /*
856 * First question: is this buffer already part of the current
857 * transaction or the existing committing transaction?
858 */
859 if (jh->b_transaction) {
860 J_ASSERT_JH(jh,
861 jh->b_transaction == transaction ||
862 jh->b_transaction ==
863 journal->j_committing_transaction);
864 if (jh->b_next_transaction)
865 J_ASSERT_JH(jh, jh->b_next_transaction ==
866 transaction);
Jan Karaf91d1d02009-07-13 16:16:20 -0400867 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700868 }
869 /*
870 * In any case we need to clean the dirty flag and we must
871 * do it under the buffer lock to be sure we don't race
872 * with running write-out.
873 */
Jan Karaf91d1d02009-07-13 16:16:20 -0400874 JBUFFER_TRACE(jh, "Journalling dirty buffer");
875 clear_buffer_dirty(bh);
876 set_buffer_jbddirty(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700877 }
878
879 unlock_buffer(bh);
880
881 error = -EROFS;
882 if (is_handle_aborted(handle)) {
883 jbd_unlock_bh_state(bh);
884 goto out;
885 }
886 error = 0;
887
888 /*
889 * The buffer is already part of this transaction if b_transaction or
890 * b_next_transaction points to it
891 */
892 if (jh->b_transaction == transaction ||
893 jh->b_next_transaction == transaction)
894 goto done;
895
896 /*
Josef Bacik9fc7c632008-04-17 10:38:59 -0400897 * this is the first time this transaction is touching this buffer,
898 * reset the modified flag
899 */
900 jh->b_modified = 0;
901
902 /*
Jan Kara8b00f402015-06-08 12:44:21 -0400903 * If the buffer is not journaled right now, we need to make sure it
904 * doesn't get written to disk before the caller actually commits the
905 * new data
906 */
907 if (!jh->b_transaction) {
908 JBUFFER_TRACE(jh, "no transaction");
909 J_ASSERT_JH(jh, !jh->b_next_transaction);
910 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Jan Karade92c8c2015-06-08 12:46:37 -0400911 /*
912 * Make sure all stores to jh (b_modified, b_frozen_data) are
913 * visible before attaching it to the running transaction.
914 * Paired with barrier in jbd2_write_access_granted()
915 */
916 smp_wmb();
Jan Kara8b00f402015-06-08 12:44:21 -0400917 spin_lock(&journal->j_list_lock);
918 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
919 spin_unlock(&journal->j_list_lock);
920 goto done;
921 }
922 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700923 * If there is already a copy-out version of this buffer, then we don't
924 * need to make another one
925 */
926 if (jh->b_frozen_data) {
927 JBUFFER_TRACE(jh, "has frozen data");
928 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
Jan Karade92c8c2015-06-08 12:46:37 -0400929 goto attach_next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700930 }
931
Jan Kara8b00f402015-06-08 12:44:21 -0400932 JBUFFER_TRACE(jh, "owned by older transaction");
933 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
934 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700935
936 /*
Jan Kara8b00f402015-06-08 12:44:21 -0400937 * There is one case we have to be very careful about. If the
938 * committing transaction is currently writing this buffer out to disk
939 * and has NOT made a copy-out, then we cannot modify the buffer
940 * contents at all right now. The essence of copy-out is that it is
941 * the extra copy, not the primary copy, which gets journaled. If the
942 * primary copy is already going to disk then we cannot do copy-out
943 * here.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700944 */
Jan Kara8b00f402015-06-08 12:44:21 -0400945 if (buffer_shadow(bh)) {
946 JBUFFER_TRACE(jh, "on shadow: sleep");
947 jbd_unlock_bh_state(bh);
948 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
949 goto repeat;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700950 }
951
Jan Kara8b00f402015-06-08 12:44:21 -0400952 /*
953 * Only do the copy if the currently-owning transaction still needs it.
954 * If buffer isn't on BJ_Metadata list, the committing transaction is
955 * past that stage (here we use the fact that BH_Shadow is set under
956 * bh_state lock together with refiling to BJ_Shadow list and at this
957 * point we know the buffer doesn't have BH_Shadow set).
958 *
959 * Subtle point, though: if this is a get_undo_access, then we will be
960 * relying on the frozen_data to contain the new value of the
961 * committed_data record after the transaction, so we HAVE to force the
962 * frozen_data copy in that case.
963 */
964 if (jh->b_jlist == BJ_Metadata || force_copy) {
965 JBUFFER_TRACE(jh, "generate frozen data");
966 if (!frozen_buffer) {
967 JBUFFER_TRACE(jh, "allocate memory for buffer");
968 jbd_unlock_bh_state(bh);
Michal Hocko490c1b42016-03-13 17:38:20 -0400969 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
970 GFP_NOFS | __GFP_NOFAIL);
Jan Kara8b00f402015-06-08 12:44:21 -0400971 goto repeat;
972 }
973 jh->b_frozen_data = frozen_buffer;
974 frozen_buffer = NULL;
975 jbd2_freeze_jh_data(jh);
976 }
Jan Karade92c8c2015-06-08 12:46:37 -0400977attach_next:
978 /*
979 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
980 * before attaching it to the running transaction. Paired with barrier
981 * in jbd2_write_access_granted()
982 */
983 smp_wmb();
Jan Kara8b00f402015-06-08 12:44:21 -0400984 jh->b_next_transaction = transaction;
985
Dave Kleikamp470decc2006-10-11 01:20:57 -0700986done:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700987 jbd_unlock_bh_state(bh);
988
989 /*
990 * If we are about to journal a buffer, then any revoke pending on it is
991 * no longer valid
992 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700993 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700994
995out:
996 if (unlikely(frozen_buffer)) /* It's usually NULL */
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400997 jbd2_free(frozen_buffer, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700998
999 JBUFFER_TRACE(jh, "exit");
1000 return error;
1001}
1002
Jan Karade92c8c2015-06-08 12:46:37 -04001003/* Fast check whether buffer is already attached to the required transaction */
Junxiao Bi087ffd42015-12-04 12:29:28 -05001004static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1005 bool undo)
Jan Karade92c8c2015-06-08 12:46:37 -04001006{
1007 struct journal_head *jh;
1008 bool ret = false;
1009
1010 /* Dirty buffers require special handling... */
1011 if (buffer_dirty(bh))
1012 return false;
1013
1014 /*
1015 * RCU protects us from dereferencing freed pages. So the checks we do
1016 * are guaranteed not to oops. However the jh slab object can get freed
1017 * & reallocated while we work with it. So we have to be careful. When
1018 * we see jh attached to the running transaction, we know it must stay
1019 * so until the transaction is committed. Thus jh won't be freed and
1020 * will be attached to the same bh while we run. However it can
1021 * happen jh gets freed, reallocated, and attached to the transaction
1022 * just after we get pointer to it from bh. So we have to be careful
1023 * and recheck jh still belongs to our bh before we return success.
1024 */
1025 rcu_read_lock();
1026 if (!buffer_jbd(bh))
1027 goto out;
1028 /* This should be bh2jh() but that doesn't work with inline functions */
1029 jh = READ_ONCE(bh->b_private);
1030 if (!jh)
1031 goto out;
Junxiao Bi087ffd42015-12-04 12:29:28 -05001032 /* For undo access buffer must have data copied */
1033 if (undo && !jh->b_committed_data)
1034 goto out;
Jan Karade92c8c2015-06-08 12:46:37 -04001035 if (jh->b_transaction != handle->h_transaction &&
1036 jh->b_next_transaction != handle->h_transaction)
1037 goto out;
1038 /*
1039 * There are two reasons for the barrier here:
1040 * 1) Make sure to fetch b_bh after we did previous checks so that we
1041 * detect when jh went through free, realloc, attach to transaction
1042 * while we were checking. Paired with implicit barrier in that path.
1043 * 2) So that access to bh done after jbd2_write_access_granted()
1044 * doesn't get reordered and see inconsistent state of concurrent
1045 * do_get_write_access().
1046 */
1047 smp_mb();
1048 if (unlikely(jh->b_bh != bh))
1049 goto out;
1050 ret = true;
1051out:
1052 rcu_read_unlock();
1053 return ret;
1054}
1055
Dave Kleikamp470decc2006-10-11 01:20:57 -07001056/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001057 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001058 * @handle: transaction to add buffer modifications to
1059 * @bh: bh to be used for metadata writes
Dave Kleikamp470decc2006-10-11 01:20:57 -07001060 *
1061 * Returns an error code or 0 on success.
1062 *
1063 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1064 * because we're write()ing a buffer which is also part of a shared mapping.
1065 */
1066
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001067int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001068{
Jan Karade92c8c2015-06-08 12:46:37 -04001069 struct journal_head *jh;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001070 int rc;
1071
Junxiao Bi087ffd42015-12-04 12:29:28 -05001072 if (jbd2_write_access_granted(handle, bh, false))
Jan Karade92c8c2015-06-08 12:46:37 -04001073 return 0;
1074
1075 jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001076 /* We do not want to get caught playing with fields which the
1077 * log thread also manipulates. Make sure that the buffer
1078 * completes any outstanding IO before proceeding. */
1079 rc = do_get_write_access(handle, jh, 0);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001080 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001081 return rc;
1082}
1083
1084
1085/*
1086 * When the user wants to journal a newly created buffer_head
1087 * (ie. getblk() returned a new buffer and we are going to populate it
1088 * manually rather than reading off disk), then we need to keep the
1089 * buffer_head locked until it has been completely filled with new
1090 * data. In this case, we should be able to make the assertion that
1091 * the bh is not already part of an existing transaction.
1092 *
1093 * The buffer should already be locked by the caller by this point.
1094 * There is no lock ranking violation: it was a newly created,
1095 * unlocked buffer beforehand. */
1096
1097/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001098 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
Dave Kleikamp470decc2006-10-11 01:20:57 -07001099 * @handle: transaction to new buffer to
1100 * @bh: new buffer.
1101 *
1102 * Call this if you create a new bh.
1103 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001104int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001105{
1106 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001107 journal_t *journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001108 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001109 int err;
1110
1111 jbd_debug(5, "journal_head %p\n", jh);
1112 err = -EROFS;
1113 if (is_handle_aborted(handle))
1114 goto out;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001115 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001116 err = 0;
1117
1118 JBUFFER_TRACE(jh, "entry");
1119 /*
1120 * The buffer may already belong to this transaction due to pre-zeroing
1121 * in the filesystem's new_block code. It may also be on the previous,
1122 * committing transaction's lists, but it HAS to be in Forget state in
1123 * that case: the transaction must have deleted the buffer for it to be
1124 * reused here.
1125 */
1126 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001127 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1128 jh->b_transaction == NULL ||
1129 (jh->b_transaction == journal->j_committing_transaction &&
1130 jh->b_jlist == BJ_Forget)));
1131
1132 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1133 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1134
1135 if (jh->b_transaction == NULL) {
Jan Karaf91d1d02009-07-13 16:16:20 -04001136 /*
1137 * Previous jbd2_journal_forget() could have left the buffer
1138 * with jbddirty bit set because it was being committed. When
1139 * the commit finished, we've filed the buffer for
1140 * checkpointing and marked it dirty. Now we are reallocating
1141 * the buffer so the transaction freeing it must have
1142 * committed and so it's safe to clear the dirty bit.
1143 */
1144 clear_buffer_dirty(jh2bh(jh));
Josef Bacik9fc7c632008-04-17 10:38:59 -04001145 /* first access by this transaction */
1146 jh->b_modified = 0;
1147
Dave Kleikamp470decc2006-10-11 01:20:57 -07001148 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Theodore Ts'o6e4862a2014-03-09 00:46:23 -05001149 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001150 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001151 } else if (jh->b_transaction == journal->j_committing_transaction) {
Josef Bacik9fc7c632008-04-17 10:38:59 -04001152 /* first access by this transaction */
1153 jh->b_modified = 0;
1154
Dave Kleikamp470decc2006-10-11 01:20:57 -07001155 JBUFFER_TRACE(jh, "set next transaction");
Theodore Ts'o6e4862a2014-03-09 00:46:23 -05001156 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001157 jh->b_next_transaction = transaction;
1158 }
1159 spin_unlock(&journal->j_list_lock);
1160 jbd_unlock_bh_state(bh);
1161
1162 /*
1163 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1164 * blocks which contain freed but then revoked metadata. We need
1165 * to cancel the revoke in case we end up freeing it yet again
1166 * and the reallocating as data - this would cause a second revoke,
1167 * which hits an assertion error.
1168 */
1169 JBUFFER_TRACE(jh, "cancelling revoke");
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001170 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001171out:
Ding Dinghua3991b402011-05-25 17:43:48 -04001172 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001173 return err;
1174}
1175
1176/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001177 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
Dave Kleikamp470decc2006-10-11 01:20:57 -07001178 * non-rewindable consequences
1179 * @handle: transaction
1180 * @bh: buffer to undo
Dave Kleikamp470decc2006-10-11 01:20:57 -07001181 *
1182 * Sometimes there is a need to distinguish between metadata which has
1183 * been committed to disk and that which has not. The ext3fs code uses
1184 * this for freeing and allocating space, we have to make sure that we
1185 * do not reuse freed space until the deallocation has been committed,
1186 * since if we overwrote that space we would make the delete
1187 * un-rewindable in case of a crash.
1188 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001189 * To deal with that, jbd2_journal_get_undo_access requests write access to a
Dave Kleikamp470decc2006-10-11 01:20:57 -07001190 * buffer for parts of non-rewindable operations such as delete
1191 * operations on the bitmaps. The journaling code must keep a copy of
1192 * the buffer's contents prior to the undo_access call until such time
1193 * as we know that the buffer has definitely been committed to disk.
1194 *
1195 * We never need to know which transaction the committed data is part
1196 * of, buffers touched here are guaranteed to be dirtied later and so
1197 * will be committed to a new transaction in due course, at which point
1198 * we can discard the old committed data pointer.
1199 *
1200 * Returns error number or 0 on success.
1201 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001202int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001203{
1204 int err;
Jan Karade92c8c2015-06-08 12:46:37 -04001205 struct journal_head *jh;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001206 char *committed_data = NULL;
1207
1208 JBUFFER_TRACE(jh, "entry");
Junxiao Bi087ffd42015-12-04 12:29:28 -05001209 if (jbd2_write_access_granted(handle, bh, true))
Jan Karade92c8c2015-06-08 12:46:37 -04001210 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001211
Jan Karade92c8c2015-06-08 12:46:37 -04001212 jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001213 /*
1214 * Do this first --- it can drop the journal lock, so we want to
1215 * make sure that obtaining the committed_data is done
1216 * atomically wrt. completion of any outstanding commits.
1217 */
1218 err = do_get_write_access(handle, jh, 1);
1219 if (err)
1220 goto out;
1221
1222repeat:
Michal Hocko490c1b42016-03-13 17:38:20 -04001223 if (!jh->b_committed_data)
1224 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1225 GFP_NOFS|__GFP_NOFAIL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001226
1227 jbd_lock_bh_state(bh);
1228 if (!jh->b_committed_data) {
1229 /* Copy out the current buffer contents into the
1230 * preserved, committed copy. */
1231 JBUFFER_TRACE(jh, "generate b_committed data");
1232 if (!committed_data) {
1233 jbd_unlock_bh_state(bh);
1234 goto repeat;
1235 }
1236
1237 jh->b_committed_data = committed_data;
1238 committed_data = NULL;
1239 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1240 }
1241 jbd_unlock_bh_state(bh);
1242out:
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001243 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001244 if (unlikely(committed_data))
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001245 jbd2_free(committed_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001246 return err;
1247}
1248
1249/**
Joel Beckere06c8222008-09-11 15:35:47 -07001250 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1251 * @bh: buffer to trigger on
1252 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1253 *
1254 * Set any triggers on this journal_head. This is always safe, because
1255 * triggers for a committing buffer will be saved off, and triggers for
1256 * a running transaction will match the buffer in that transaction.
1257 *
1258 * Call with NULL to clear the triggers.
1259 */
1260void jbd2_journal_set_triggers(struct buffer_head *bh,
1261 struct jbd2_buffer_trigger_type *type)
1262{
Jan Karaad56eda2013-03-11 13:24:56 -04001263 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
Joel Beckere06c8222008-09-11 15:35:47 -07001264
Jan Karaad56eda2013-03-11 13:24:56 -04001265 if (WARN_ON(!jh))
1266 return;
Joel Beckere06c8222008-09-11 15:35:47 -07001267 jh->b_triggers = type;
Jan Karaad56eda2013-03-11 13:24:56 -04001268 jbd2_journal_put_journal_head(jh);
Joel Beckere06c8222008-09-11 15:35:47 -07001269}
1270
Jan Kara13ceef02010-07-14 07:56:33 +02001271void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
Joel Beckere06c8222008-09-11 15:35:47 -07001272 struct jbd2_buffer_trigger_type *triggers)
1273{
1274 struct buffer_head *bh = jh2bh(jh);
1275
Jan Kara13ceef02010-07-14 07:56:33 +02001276 if (!triggers || !triggers->t_frozen)
Joel Beckere06c8222008-09-11 15:35:47 -07001277 return;
1278
Jan Kara13ceef02010-07-14 07:56:33 +02001279 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
Joel Beckere06c8222008-09-11 15:35:47 -07001280}
1281
1282void jbd2_buffer_abort_trigger(struct journal_head *jh,
1283 struct jbd2_buffer_trigger_type *triggers)
1284{
1285 if (!triggers || !triggers->t_abort)
1286 return;
1287
1288 triggers->t_abort(triggers, jh2bh(jh));
1289}
1290
Joel Beckere06c8222008-09-11 15:35:47 -07001291/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001292 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
Dave Kleikamp470decc2006-10-11 01:20:57 -07001293 * @handle: transaction to add buffer to.
1294 * @bh: buffer to mark
1295 *
1296 * mark dirty metadata which needs to be journaled as part of the current
1297 * transaction.
1298 *
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001299 * The buffer must have previously had jbd2_journal_get_write_access()
1300 * called so that it has a valid journal_head attached to the buffer
1301 * head.
1302 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001303 * The buffer is placed on the transaction's metadata list and is marked
1304 * as belonging to the transaction.
1305 *
1306 * Returns error number or 0 on success.
1307 *
1308 * Special care needs to be taken if the buffer already belongs to the
1309 * current committing transaction (in which case we should have frozen
1310 * data present for that commit). In that case, we don't relink the
1311 * buffer: that only gets done when the old transaction finally
1312 * completes its commit.
1313 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001314int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001315{
1316 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001317 journal_t *journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001318 struct journal_head *jh;
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001319 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001320
Dave Kleikamp470decc2006-10-11 01:20:57 -07001321 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001322 return -EROFS;
Jan Kara6e06ae82015-07-12 18:11:30 -04001323 if (!buffer_jbd(bh)) {
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001324 ret = -EUCLEAN;
1325 goto out;
1326 }
Jan Kara6e06ae82015-07-12 18:11:30 -04001327 /*
1328 * We don't grab jh reference here since the buffer must be part
1329 * of the running transaction.
1330 */
1331 jh = bh2jh(bh);
1332 /*
1333 * This and the following assertions are unreliable since we may see jh
1334 * in inconsistent state unless we grab bh_state lock. But this is
1335 * crucial to catch bugs so let's do a reliable check until the
1336 * lockless handling is fully proven.
1337 */
1338 if (jh->b_transaction != transaction &&
1339 jh->b_next_transaction != transaction) {
1340 jbd_lock_bh_state(bh);
1341 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1342 jh->b_next_transaction == transaction);
1343 jbd_unlock_bh_state(bh);
1344 }
1345 if (jh->b_modified == 1) {
1346 /* If it's in our transaction it must be in BJ_Metadata list. */
1347 if (jh->b_transaction == transaction &&
1348 jh->b_jlist != BJ_Metadata) {
1349 jbd_lock_bh_state(bh);
1350 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1351 jh->b_jlist == BJ_Metadata);
1352 jbd_unlock_bh_state(bh);
1353 }
1354 goto out;
1355 }
1356
1357 journal = transaction->t_journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001358 jbd_debug(5, "journal_head %p\n", jh);
1359 JBUFFER_TRACE(jh, "entry");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001360
1361 jbd_lock_bh_state(bh);
1362
1363 if (jh->b_modified == 0) {
1364 /*
1365 * This buffer's got modified and becoming part
1366 * of the transaction. This needs to be done
1367 * once a transaction -bzzz
1368 */
1369 jh->b_modified = 1;
Theodore Ts'of6c07ca2013-12-08 21:12:59 -05001370 if (handle->h_buffer_credits <= 0) {
1371 ret = -ENOSPC;
1372 goto out_unlock_bh;
1373 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001374 handle->h_buffer_credits--;
1375 }
1376
1377 /*
1378 * fastpath, to avoid expensive locking. If this buffer is already
1379 * on the running transaction's metadata list there is nothing to do.
1380 * Nobody can take it off again because there is a handle open.
1381 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1382 * result in this test being false, so we go in and take the locks.
1383 */
1384 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1385 JBUFFER_TRACE(jh, "fastpath");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001386 if (unlikely(jh->b_transaction !=
1387 journal->j_running_transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001388 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001389 "jh->b_transaction (%llu, %p, %u) != "
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001390 "journal->j_running_transaction (%p, %u)\n",
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001391 journal->j_devname,
1392 (unsigned long long) bh->b_blocknr,
1393 jh->b_transaction,
1394 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1395 journal->j_running_transaction,
1396 journal->j_running_transaction ?
1397 journal->j_running_transaction->t_tid : 0);
1398 ret = -EINVAL;
1399 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001400 goto out_unlock_bh;
1401 }
1402
1403 set_buffer_jbddirty(bh);
1404
1405 /*
1406 * Metadata already on the current transaction list doesn't
1407 * need to be filed. Metadata on another transaction's list must
1408 * be committing, and will be refiled once the commit completes:
1409 * leave it alone for now.
1410 */
1411 if (jh->b_transaction != transaction) {
1412 JBUFFER_TRACE(jh, "already on other transaction");
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001413 if (unlikely(((jh->b_transaction !=
1414 journal->j_committing_transaction)) ||
1415 (jh->b_next_transaction != transaction))) {
1416 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1417 "bad jh for block %llu: "
1418 "transaction (%p, %u), "
1419 "jh->b_transaction (%p, %u), "
1420 "jh->b_next_transaction (%p, %u), jlist %u\n",
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001421 journal->j_devname,
1422 (unsigned long long) bh->b_blocknr,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001423 transaction, transaction->t_tid,
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001424 jh->b_transaction,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001425 jh->b_transaction ?
1426 jh->b_transaction->t_tid : 0,
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001427 jh->b_next_transaction,
1428 jh->b_next_transaction ?
1429 jh->b_next_transaction->t_tid : 0,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001430 jh->b_jlist);
1431 WARN_ON(1);
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001432 ret = -EINVAL;
1433 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001434 /* And this case is illegal: we can't reuse another
1435 * transaction's data buffer, ever. */
1436 goto out_unlock_bh;
1437 }
1438
1439 /* That test should have eliminated the following case: */
Mingming Cao40191912008-01-28 23:58:27 -05001440 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001441
1442 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1443 spin_lock(&journal->j_list_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001444 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001445 spin_unlock(&journal->j_list_lock);
1446out_unlock_bh:
1447 jbd_unlock_bh_state(bh);
1448out:
1449 JBUFFER_TRACE(jh, "exit");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001450 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001451}
1452
Dave Kleikamp470decc2006-10-11 01:20:57 -07001453/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001454 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001455 * @handle: transaction handle
1456 * @bh: bh to 'forget'
1457 *
1458 * We can only do the bforget if there are no commits pending against the
1459 * buffer. If the buffer is dirty in the current running transaction we
1460 * can safely unlink it.
1461 *
1462 * bh may not be a journalled buffer at all - it may be a non-JBD
1463 * buffer which came off the hashtable. Check for this.
1464 *
1465 * Decrements bh->b_count by one.
1466 *
1467 * Allow this call even if the handle has aborted --- it may be part of
1468 * the caller's cleanup after an abort.
1469 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001470int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001471{
1472 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001473 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001474 struct journal_head *jh;
1475 int drop_reserve = 0;
1476 int err = 0;
Josef Bacik1dfc3222008-04-17 10:38:59 -04001477 int was_modified = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001478
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001479 if (is_handle_aborted(handle))
1480 return -EROFS;
1481 journal = transaction->t_journal;
1482
Dave Kleikamp470decc2006-10-11 01:20:57 -07001483 BUFFER_TRACE(bh, "entry");
1484
1485 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001486
1487 if (!buffer_jbd(bh))
1488 goto not_jbd;
1489 jh = bh2jh(bh);
1490
1491 /* Critical error: attempting to delete a bitmap buffer, maybe?
1492 * Don't do any jbd operations, and return an error. */
1493 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1494 "inconsistent data on disk")) {
1495 err = -EIO;
1496 goto not_jbd;
1497 }
1498
Adam Buchbinder48fc7f72012-09-19 21:48:00 -04001499 /* keep track of whether or not this transaction modified us */
Josef Bacik1dfc3222008-04-17 10:38:59 -04001500 was_modified = jh->b_modified;
1501
Dave Kleikamp470decc2006-10-11 01:20:57 -07001502 /*
1503 * The buffer's going from the transaction, we must drop
1504 * all references -bzzz
1505 */
1506 jh->b_modified = 0;
1507
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001508 if (jh->b_transaction == transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001509 J_ASSERT_JH(jh, !jh->b_frozen_data);
1510
1511 /* If we are forgetting a buffer which is already part
1512 * of this transaction, then we can just drop it from
1513 * the transaction immediately. */
1514 clear_buffer_dirty(bh);
1515 clear_buffer_jbddirty(bh);
1516
1517 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1518
Josef Bacik1dfc3222008-04-17 10:38:59 -04001519 /*
1520 * we only want to drop a reference if this transaction
1521 * modified the buffer
1522 */
1523 if (was_modified)
1524 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001525
1526 /*
1527 * We are no longer going to journal this buffer.
1528 * However, the commit of this transaction is still
1529 * important to the buffer: the delete that we are now
1530 * processing might obsolete an old log entry, so by
1531 * committing, we can satisfy the buffer's checkpoint.
1532 *
1533 * So, if we have a checkpoint on the buffer, we should
1534 * now refile the buffer on our BJ_Forget list so that
1535 * we know to remove the checkpoint after we commit.
1536 */
1537
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001538 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001539 if (jh->b_cp_transaction) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001540 __jbd2_journal_temp_unlink_buffer(jh);
1541 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001542 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001543 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001544 if (!buffer_jbd(bh)) {
1545 spin_unlock(&journal->j_list_lock);
1546 jbd_unlock_bh_state(bh);
1547 __bforget(bh);
1548 goto drop;
1549 }
1550 }
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001551 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001552 } else if (jh->b_transaction) {
1553 J_ASSERT_JH(jh, (jh->b_transaction ==
1554 journal->j_committing_transaction));
1555 /* However, if the buffer is still owned by a prior
1556 * (committing) transaction, we can't drop it yet... */
1557 JBUFFER_TRACE(jh, "belongs to older transaction");
1558 /* ... but we CAN drop it from the new transaction if we
1559 * have also modified it since the original commit. */
1560
1561 if (jh->b_next_transaction) {
1562 J_ASSERT(jh->b_next_transaction == transaction);
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001563 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001564 jh->b_next_transaction = NULL;
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001565 spin_unlock(&journal->j_list_lock);
Josef Bacik1dfc3222008-04-17 10:38:59 -04001566
1567 /*
1568 * only drop a reference if this transaction modified
1569 * the buffer
1570 */
1571 if (was_modified)
1572 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001573 }
1574 }
1575
1576not_jbd:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001577 jbd_unlock_bh_state(bh);
1578 __brelse(bh);
1579drop:
1580 if (drop_reserve) {
1581 /* no need to reserve log space for this block -bzzz */
1582 handle->h_buffer_credits++;
1583 }
1584 return err;
1585}
1586
1587/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001588 * int jbd2_journal_stop() - complete a transaction
Dave Kleikamp470decc2006-10-11 01:20:57 -07001589 * @handle: tranaction to complete.
1590 *
1591 * All done for a particular handle.
1592 *
1593 * There is not much action needed here. We just return any remaining
1594 * buffer credits to the transaction and remove the handle. The only
1595 * complication is that we need to start a commit operation if the
1596 * filesystem is marked for synchronous update.
1597 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001598 * jbd2_journal_stop itself will not usually return an error, but it may
Dave Kleikamp470decc2006-10-11 01:20:57 -07001599 * do so in unusual circumstances. In particular, expect it to
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001600 * return -EIO if a jbd2_journal_abort has been executed since the
Dave Kleikamp470decc2006-10-11 01:20:57 -07001601 * transaction began.
1602 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001603int jbd2_journal_stop(handle_t *handle)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001604{
1605 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001606 journal_t *journal;
1607 int err = 0, wait_for_commit = 0;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001608 tid_t tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001609 pid_t pid;
1610
Lukas Czerner9d506592015-05-14 18:55:18 -04001611 if (!transaction) {
1612 /*
1613 * Handle is already detached from the transaction so
1614 * there is nothing to do other than decrease a refcount,
1615 * or free the handle if refcount drops to zero
1616 */
1617 if (--handle->h_ref > 0) {
1618 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1619 handle->h_ref);
1620 return err;
1621 } else {
1622 if (handle->h_rsv_handle)
1623 jbd2_free_handle(handle->h_rsv_handle);
1624 goto free_and_exit;
1625 }
1626 }
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001627 journal = transaction->t_journal;
1628
Dave Kleikamp470decc2006-10-11 01:20:57 -07001629 J_ASSERT(journal_current_handle() == handle);
1630
1631 if (is_handle_aborted(handle))
1632 err = -EIO;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001633 else
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001634 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001635
1636 if (--handle->h_ref > 0) {
1637 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1638 handle->h_ref);
1639 return err;
1640 }
1641
1642 jbd_debug(4, "Handle %p going down\n", handle);
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001643 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001644 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001645 handle->h_type, handle->h_line_no,
1646 jiffies - handle->h_start_jiffies,
1647 handle->h_sync, handle->h_requested_credits,
1648 (handle->h_requested_credits -
1649 handle->h_buffer_credits));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001650
1651 /*
1652 * Implement synchronous transaction batching. If the handle
1653 * was synchronous, don't force a commit immediately. Let's
Josef Bacike07f7182008-11-26 01:14:26 -05001654 * yield and let another thread piggyback onto this
1655 * transaction. Keep doing that while new threads continue to
1656 * arrive. It doesn't cost much - we're about to run a commit
1657 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1658 * operations by 30x or more...
Dave Kleikamp470decc2006-10-11 01:20:57 -07001659 *
Josef Bacike07f7182008-11-26 01:14:26 -05001660 * We try and optimize the sleep time against what the
1661 * underlying disk can do, instead of having a static sleep
1662 * time. This is useful for the case where our storage is so
1663 * fast that it is more optimal to go ahead and force a flush
1664 * and wait for the transaction to be committed than it is to
1665 * wait for an arbitrary amount of time for new writers to
1666 * join the transaction. We achieve this by measuring how
1667 * long it takes to commit a transaction, and compare it with
1668 * how long this transaction has been running, and if run time
1669 * < commit time then we sleep for the delta and commit. This
1670 * greatly helps super fast disks that would see slowdowns as
1671 * more threads started doing fsyncs.
1672 *
1673 * But don't do this if this process was the most recent one
1674 * to perform a synchronous write. We do this to detect the
1675 * case where a single process is doing a stream of sync
1676 * writes. No point in waiting for joiners in that case.
Eric Sandeen5dd21422014-07-05 19:18:22 -04001677 *
1678 * Setting max_batch_time to 0 disables this completely.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001679 */
1680 pid = current->pid;
Eric Sandeen5dd21422014-07-05 19:18:22 -04001681 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1682 journal->j_max_batch_time) {
Josef Bacike07f7182008-11-26 01:14:26 -05001683 u64 commit_time, trans_time;
1684
Dave Kleikamp470decc2006-10-11 01:20:57 -07001685 journal->j_last_sync_writer = pid;
Josef Bacike07f7182008-11-26 01:14:26 -05001686
Theodore Ts'oa931da62010-08-03 21:35:12 -04001687 read_lock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001688 commit_time = journal->j_average_commit_time;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001689 read_unlock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001690
1691 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1692 transaction->t_start_time));
1693
Theodore Ts'o30773842009-01-03 20:27:38 -05001694 commit_time = max_t(u64, commit_time,
1695 1000*journal->j_min_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001696 commit_time = min_t(u64, commit_time,
Theodore Ts'o30773842009-01-03 20:27:38 -05001697 1000*journal->j_max_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001698
1699 if (trans_time < commit_time) {
1700 ktime_t expires = ktime_add_ns(ktime_get(),
1701 commit_time);
1702 set_current_state(TASK_UNINTERRUPTIBLE);
1703 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1704 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001705 }
1706
Theodore Ts'o70585482009-03-25 23:35:46 -04001707 if (handle->h_sync)
1708 transaction->t_synchronous_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001709 current->journal_info = NULL;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001710 atomic_sub(handle->h_buffer_credits,
1711 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001712
1713 /*
1714 * If the handle is marked SYNC, we need to set another commit
1715 * going! We also want to force a commit if the current
1716 * transaction is occupying too much of the log, or if the
1717 * transaction is too old now.
1718 */
1719 if (handle->h_sync ||
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001720 (atomic_read(&transaction->t_outstanding_credits) >
1721 journal->j_max_transaction_buffers) ||
1722 time_after_eq(jiffies, transaction->t_expires)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001723 /* Do this even for aborted journals: an abort still
1724 * completes the commit thread, it just doesn't write
1725 * anything to disk. */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001726
Dave Kleikamp470decc2006-10-11 01:20:57 -07001727 jbd_debug(2, "transaction too old, requesting commit for "
1728 "handle %p\n", handle);
1729 /* This is non-blocking */
Theodore Ts'oc35a56a2010-05-16 05:00:00 -04001730 jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001731
1732 /*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001733 * Special case: JBD2_SYNC synchronous updates require us
Dave Kleikamp470decc2006-10-11 01:20:57 -07001734 * to wait for the commit to complete.
1735 */
1736 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001737 wait_for_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001738 }
1739
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001740 /*
1741 * Once we drop t_updates, if it goes to zero the transaction
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001742 * could start committing on us and eventually disappear. So
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001743 * once we do this, we must not dereference transaction
1744 * pointer again.
1745 */
1746 tid = transaction->t_tid;
1747 if (atomic_dec_and_test(&transaction->t_updates)) {
1748 wake_up(&journal->j_wait_updates);
1749 if (journal->j_barrier_count)
1750 wake_up(&journal->j_wait_transaction_locked);
1751 }
1752
1753 if (wait_for_commit)
1754 err = jbd2_log_wait_commit(journal, tid);
1755
Ingo Molnar3295f0e2008-08-11 10:30:30 +02001756 lock_map_release(&handle->h_lockdep_map);
Mingming Cao7b751062008-01-28 23:58:27 -05001757
Jan Kara8f7d89f2013-06-04 12:35:11 -04001758 if (handle->h_rsv_handle)
1759 jbd2_journal_free_reserved(handle->h_rsv_handle);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001760free_and_exit:
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001761 jbd2_free_handle(handle);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001762 return err;
1763}
1764
Dave Kleikamp470decc2006-10-11 01:20:57 -07001765/*
1766 *
1767 * List management code snippets: various functions for manipulating the
1768 * transaction buffer lists.
1769 *
1770 */
1771
1772/*
1773 * Append a buffer to a transaction list, given the transaction's list head
1774 * pointer.
1775 *
1776 * j_list_lock is held.
1777 *
1778 * jbd_lock_bh_state(jh2bh(jh)) is held.
1779 */
1780
1781static inline void
1782__blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1783{
1784 if (!*list) {
1785 jh->b_tnext = jh->b_tprev = jh;
1786 *list = jh;
1787 } else {
1788 /* Insert at the tail of the list to preserve order */
1789 struct journal_head *first = *list, *last = first->b_tprev;
1790 jh->b_tprev = last;
1791 jh->b_tnext = first;
1792 last->b_tnext = first->b_tprev = jh;
1793 }
1794}
1795
1796/*
1797 * Remove a buffer from a transaction list, given the transaction's list
1798 * head pointer.
1799 *
1800 * Called with j_list_lock held, and the journal may not be locked.
1801 *
1802 * jbd_lock_bh_state(jh2bh(jh)) is held.
1803 */
1804
1805static inline void
1806__blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1807{
1808 if (*list == jh) {
1809 *list = jh->b_tnext;
1810 if (*list == jh)
1811 *list = NULL;
1812 }
1813 jh->b_tprev->b_tnext = jh->b_tnext;
1814 jh->b_tnext->b_tprev = jh->b_tprev;
1815}
1816
1817/*
1818 * Remove a buffer from the appropriate transaction list.
1819 *
1820 * Note that this function can *change* the value of
Jan Karaf5113ef2013-06-04 12:01:45 -04001821 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1822 * t_reserved_list. If the caller is holding onto a copy of one of these
1823 * pointers, it could go bad. Generally the caller needs to re-read the
1824 * pointer from the transaction_t.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001825 *
Jan Kara5bebccf2012-03-13 22:25:06 -04001826 * Called under j_list_lock.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001827 */
Jan Kara5bebccf2012-03-13 22:25:06 -04001828static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001829{
1830 struct journal_head **list = NULL;
1831 transaction_t *transaction;
1832 struct buffer_head *bh = jh2bh(jh);
1833
1834 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1835 transaction = jh->b_transaction;
1836 if (transaction)
1837 assert_spin_locked(&transaction->t_journal->j_list_lock);
1838
1839 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1840 if (jh->b_jlist != BJ_None)
Mingming Cao40191912008-01-28 23:58:27 -05001841 J_ASSERT_JH(jh, transaction != NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001842
1843 switch (jh->b_jlist) {
1844 case BJ_None:
1845 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001846 case BJ_Metadata:
1847 transaction->t_nr_buffers--;
1848 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1849 list = &transaction->t_buffers;
1850 break;
1851 case BJ_Forget:
1852 list = &transaction->t_forget;
1853 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001854 case BJ_Shadow:
1855 list = &transaction->t_shadow_list;
1856 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001857 case BJ_Reserved:
1858 list = &transaction->t_reserved_list;
1859 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001860 }
1861
1862 __blist_del_buffer(list, jh);
1863 jh->b_jlist = BJ_None;
1864 if (test_clear_buffer_jbddirty(bh))
1865 mark_buffer_dirty(bh); /* Expose it to the VM */
1866}
1867
Jan Karade1b7942011-06-13 15:38:22 -04001868/*
1869 * Remove buffer from all transactions.
1870 *
1871 * Called with bh_state lock and j_list_lock
1872 *
1873 * jh and bh may be already freed when this function returns.
1874 */
1875static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001876{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001877 __jbd2_journal_temp_unlink_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001878 jh->b_transaction = NULL;
Jan Karade1b7942011-06-13 15:38:22 -04001879 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001880}
1881
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001882void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001883{
Jan Karade1b7942011-06-13 15:38:22 -04001884 struct buffer_head *bh = jh2bh(jh);
1885
1886 /* Get reference so that buffer cannot be freed before we unlock it */
1887 get_bh(bh);
1888 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001889 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001890 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001891 spin_unlock(&journal->j_list_lock);
Jan Karade1b7942011-06-13 15:38:22 -04001892 jbd_unlock_bh_state(bh);
1893 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001894}
1895
1896/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001897 * Called from jbd2_journal_try_to_free_buffers().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001898 *
1899 * Called under jbd_lock_bh_state(bh)
1900 */
1901static void
1902__journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1903{
1904 struct journal_head *jh;
1905
1906 jh = bh2jh(bh);
1907
1908 if (buffer_locked(bh) || buffer_dirty(bh))
1909 goto out;
1910
Theodore Ts'od2eb0b92014-03-09 00:07:19 -05001911 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001912 goto out;
1913
1914 spin_lock(&journal->j_list_lock);
Theodore Ts'od2eb0b92014-03-09 00:07:19 -05001915 if (jh->b_cp_transaction != NULL) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001916 /* written-back checkpointed metadata buffer */
Jan Karac254c9e2012-03-13 22:27:44 -04001917 JBUFFER_TRACE(jh, "remove from checkpoint list");
1918 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001919 }
1920 spin_unlock(&journal->j_list_lock);
1921out:
1922 return;
1923}
1924
Dave Kleikamp470decc2006-10-11 01:20:57 -07001925/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001926 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001927 * @journal: journal for operation
1928 * @page: to try and free
Mingming Cao530576b2008-07-13 21:06:39 -04001929 * @gfp_mask: we use the mask to detect how hard should we try to release
Mel Gormand0164ad2015-11-06 16:28:21 -08001930 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1931 * code to release the buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001932 *
1933 *
1934 * For all the buffers on this page,
1935 * if they are fully written out ordered data, move them onto BUF_CLEAN
1936 * so try_to_free_buffers() can reap them.
1937 *
1938 * This function returns non-zero if we wish try_to_free_buffers()
1939 * to be called. We do this if the page is releasable by try_to_free_buffers().
1940 * We also do it if the page has locked or dirty buffers and the caller wants
1941 * us to perform sync or async writeout.
1942 *
1943 * This complicates JBD locking somewhat. We aren't protected by the
1944 * BKL here. We wish to remove the buffer from its committing or
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001945 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001946 *
1947 * This may *change* the value of transaction_t->t_datalist, so anyone
1948 * who looks at t_datalist needs to lock against this function.
1949 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001950 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1951 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
Dave Kleikamp470decc2006-10-11 01:20:57 -07001952 * will come out of the lock with the buffer dirty, which makes it
1953 * ineligible for release here.
1954 *
1955 * Who else is affected by this? hmm... Really the only contender
1956 * is do_get_write_access() - it could be looking at the buffer while
1957 * journal_try_to_free_buffer() is changing its state. But that
1958 * cannot happen because we never reallocate freed data as metadata
1959 * while the data is part of a transaction. Yes?
Mingming Cao530576b2008-07-13 21:06:39 -04001960 *
1961 * Return 0 on failure, 1 on success
Dave Kleikamp470decc2006-10-11 01:20:57 -07001962 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001963int jbd2_journal_try_to_free_buffers(journal_t *journal,
Mingming Cao530576b2008-07-13 21:06:39 -04001964 struct page *page, gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001965{
1966 struct buffer_head *head;
1967 struct buffer_head *bh;
1968 int ret = 0;
1969
1970 J_ASSERT(PageLocked(page));
1971
1972 head = page_buffers(page);
1973 bh = head;
1974 do {
1975 struct journal_head *jh;
1976
1977 /*
1978 * We take our own ref against the journal_head here to avoid
1979 * having to add tons of locking around each instance of
Mingming Cao530576b2008-07-13 21:06:39 -04001980 * jbd2_journal_put_journal_head().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001981 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001982 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001983 if (!jh)
1984 continue;
1985
1986 jbd_lock_bh_state(bh);
1987 __journal_try_to_free_buffer(journal, bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001988 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001989 jbd_unlock_bh_state(bh);
1990 if (buffer_jbd(bh))
1991 goto busy;
1992 } while ((bh = bh->b_this_page) != head);
Mingming Cao530576b2008-07-13 21:06:39 -04001993
Dave Kleikamp470decc2006-10-11 01:20:57 -07001994 ret = try_to_free_buffers(page);
Mingming Cao530576b2008-07-13 21:06:39 -04001995
Dave Kleikamp470decc2006-10-11 01:20:57 -07001996busy:
1997 return ret;
1998}
1999
2000/*
2001 * This buffer is no longer needed. If it is on an older transaction's
2002 * checkpoint list we need to record it on this transaction's forget list
2003 * to pin this buffer (and hence its checkpointing transaction) down until
2004 * this transaction commits. If the buffer isn't on a checkpoint list, we
2005 * release it.
2006 * Returns non-zero if JBD no longer has an interest in the buffer.
2007 *
2008 * Called under j_list_lock.
2009 *
2010 * Called under jbd_lock_bh_state(bh).
2011 */
2012static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2013{
2014 int may_free = 1;
2015 struct buffer_head *bh = jh2bh(jh);
2016
Dave Kleikamp470decc2006-10-11 01:20:57 -07002017 if (jh->b_cp_transaction) {
2018 JBUFFER_TRACE(jh, "on running+cp transaction");
Jan Karade1b7942011-06-13 15:38:22 -04002019 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karaf91d1d02009-07-13 16:16:20 -04002020 /*
2021 * We don't want to write the buffer anymore, clear the
2022 * bit so that we don't confuse checks in
2023 * __journal_file_buffer
2024 */
2025 clear_buffer_dirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002026 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002027 may_free = 0;
2028 } else {
2029 JBUFFER_TRACE(jh, "on running transaction");
Jan Karade1b7942011-06-13 15:38:22 -04002030 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002031 }
2032 return may_free;
2033}
2034
2035/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002036 * jbd2_journal_invalidatepage
Dave Kleikamp470decc2006-10-11 01:20:57 -07002037 *
2038 * This code is tricky. It has a number of cases to deal with.
2039 *
2040 * There are two invariants which this code relies on:
2041 *
2042 * i_size must be updated on disk before we start calling invalidatepage on the
2043 * data.
2044 *
2045 * This is done in ext3 by defining an ext3_setattr method which
2046 * updates i_size before truncate gets going. By maintaining this
2047 * invariant, we can be sure that it is safe to throw away any buffers
2048 * attached to the current transaction: once the transaction commits,
2049 * we know that the data will not be needed.
2050 *
2051 * Note however that we can *not* throw away data belonging to the
2052 * previous, committing transaction!
2053 *
2054 * Any disk blocks which *are* part of the previous, committing
2055 * transaction (and which therefore cannot be discarded immediately) are
2056 * not going to be reused in the new running transaction
2057 *
2058 * The bitmap committed_data images guarantee this: any block which is
2059 * allocated in one transaction and removed in the next will be marked
2060 * as in-use in the committed_data bitmap, so cannot be reused until
2061 * the next transaction to delete the block commits. This means that
2062 * leaving committing buffers dirty is quite safe: the disk blocks
2063 * cannot be reallocated to a different file and so buffer aliasing is
2064 * not possible.
2065 *
2066 *
2067 * The above applies mainly to ordered data mode. In writeback mode we
2068 * don't make guarantees about the order in which data hits disk --- in
2069 * particular we don't guarantee that new dirty data is flushed before
2070 * transaction commit --- so it is always safe just to discard data
2071 * immediately in that mode. --sct
2072 */
2073
2074/*
2075 * The journal_unmap_buffer helper function returns zero if the buffer
2076 * concerned remains pinned as an anonymous buffer belonging to an older
2077 * transaction.
2078 *
2079 * We're outside-transaction here. Either or both of j_running_transaction
2080 * and j_committing_transaction may be NULL.
2081 */
Jan Karab794e7a2012-09-26 23:11:13 -04002082static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2083 int partial_page)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002084{
2085 transaction_t *transaction;
2086 struct journal_head *jh;
2087 int may_free = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002088
2089 BUFFER_TRACE(bh, "entry");
2090
2091 /*
2092 * It is safe to proceed here without the j_list_lock because the
2093 * buffers cannot be stolen by try_to_free_buffers as long as we are
2094 * holding the page lock. --sct
2095 */
2096
2097 if (!buffer_jbd(bh))
2098 goto zap_buffer_unlocked;
2099
Jan Kara87c89c22008-07-11 19:27:31 -04002100 /* OK, we have data buffer in journaled mode */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002101 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002102 jbd_lock_bh_state(bh);
2103 spin_lock(&journal->j_list_lock);
2104
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002105 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002106 if (!jh)
2107 goto zap_buffer_no_jh;
2108
dingdinghuaba869022010-02-15 16:35:42 -05002109 /*
2110 * We cannot remove the buffer from checkpoint lists until the
2111 * transaction adding inode to orphan list (let's call it T)
2112 * is committed. Otherwise if the transaction changing the
2113 * buffer would be cleaned from the journal before T is
2114 * committed, a crash will cause that the correct contents of
2115 * the buffer will be lost. On the other hand we have to
2116 * clear the buffer dirty bit at latest at the moment when the
2117 * transaction marking the buffer as freed in the filesystem
2118 * structures is committed because from that moment on the
Jan Karab794e7a2012-09-26 23:11:13 -04002119 * block can be reallocated and used by a different page.
dingdinghuaba869022010-02-15 16:35:42 -05002120 * Since the block hasn't been freed yet but the inode has
2121 * already been added to orphan list, it is safe for us to add
2122 * the buffer to BJ_Forget list of the newest transaction.
Jan Karab794e7a2012-09-26 23:11:13 -04002123 *
2124 * Also we have to clear buffer_mapped flag of a truncated buffer
2125 * because the buffer_head may be attached to the page straddling
2126 * i_size (can happen only when blocksize < pagesize) and thus the
2127 * buffer_head can be reused when the file is extended again. So we end
2128 * up keeping around invalidated buffers attached to transactions'
2129 * BJ_Forget list just to stop checkpointing code from cleaning up
2130 * the transaction this buffer was modified in.
dingdinghuaba869022010-02-15 16:35:42 -05002131 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002132 transaction = jh->b_transaction;
2133 if (transaction == NULL) {
2134 /* First case: not on any transaction. If it
2135 * has no checkpoint link, then we can zap it:
2136 * it's a writeback-mode buffer so we don't care
2137 * if it hits disk safely. */
2138 if (!jh->b_cp_transaction) {
2139 JBUFFER_TRACE(jh, "not on any transaction: zap");
2140 goto zap_buffer;
2141 }
2142
2143 if (!buffer_dirty(bh)) {
2144 /* bdflush has written it. We can drop it now */
Jan Karabc23f0c2015-11-24 15:34:35 -05002145 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002146 goto zap_buffer;
2147 }
2148
2149 /* OK, it must be in the journal but still not
2150 * written fully to disk: it's metadata or
2151 * journaled data... */
2152
2153 if (journal->j_running_transaction) {
2154 /* ... and once the current transaction has
2155 * committed, the buffer won't be needed any
2156 * longer. */
2157 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
Jan Karab794e7a2012-09-26 23:11:13 -04002158 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002159 journal->j_running_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002160 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002161 } else {
2162 /* There is no currently-running transaction. So the
2163 * orphan record which we wrote for this file must have
2164 * passed into commit. We must attach this buffer to
2165 * the committing transaction, if it exists. */
2166 if (journal->j_committing_transaction) {
2167 JBUFFER_TRACE(jh, "give to committing trans");
Jan Karab794e7a2012-09-26 23:11:13 -04002168 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002169 journal->j_committing_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002170 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002171 } else {
2172 /* The orphan record's transaction has
2173 * committed. We can cleanse this buffer */
2174 clear_buffer_jbddirty(bh);
Jan Karabc23f0c2015-11-24 15:34:35 -05002175 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002176 goto zap_buffer;
2177 }
2178 }
2179 } else if (transaction == journal->j_committing_transaction) {
Eric Sandeen9b579882006-10-28 10:38:28 -07002180 JBUFFER_TRACE(jh, "on committing transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002181 /*
dingdinghuaba869022010-02-15 16:35:42 -05002182 * The buffer is committing, we simply cannot touch
Jan Karab794e7a2012-09-26 23:11:13 -04002183 * it. If the page is straddling i_size we have to wait
2184 * for commit and try again.
2185 */
2186 if (partial_page) {
Jan Karab794e7a2012-09-26 23:11:13 -04002187 jbd2_journal_put_journal_head(jh);
2188 spin_unlock(&journal->j_list_lock);
2189 jbd_unlock_bh_state(bh);
2190 write_unlock(&journal->j_state_lock);
Jan Kara53e87262012-12-25 13:29:52 -05002191 return -EBUSY;
Jan Karab794e7a2012-09-26 23:11:13 -04002192 }
2193 /*
2194 * OK, buffer won't be reachable after truncate. We just set
2195 * j_next_transaction to the running transaction (if there is
2196 * one) and mark buffer as freed so that commit code knows it
2197 * should clear dirty bits when it is done with the buffer.
dingdinghuaba869022010-02-15 16:35:42 -05002198 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002199 set_buffer_freed(bh);
dingdinghuaba869022010-02-15 16:35:42 -05002200 if (journal->j_running_transaction && buffer_jbddirty(bh))
2201 jh->b_next_transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002202 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002203 spin_unlock(&journal->j_list_lock);
2204 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002205 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002206 return 0;
2207 } else {
2208 /* Good, the buffer belongs to the running transaction.
2209 * We are writing our own transaction's data, not any
2210 * previous one's, so it is safe to throw it away
2211 * (remember that we expect the filesystem to have set
2212 * i_size already for this truncate so recovery will not
2213 * expose the disk blocks we are discarding here.) */
2214 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
Eric Sandeen9b579882006-10-28 10:38:28 -07002215 JBUFFER_TRACE(jh, "on running transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002216 may_free = __dispose_buffer(jh, transaction);
2217 }
2218
2219zap_buffer:
Jan Karab794e7a2012-09-26 23:11:13 -04002220 /*
2221 * This is tricky. Although the buffer is truncated, it may be reused
2222 * if blocksize < pagesize and it is attached to the page straddling
2223 * EOF. Since the buffer might have been added to BJ_Forget list of the
2224 * running transaction, journal_get_write_access() won't clear
2225 * b_modified and credit accounting gets confused. So clear b_modified
2226 * here.
2227 */
2228 jh->b_modified = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002229 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002230zap_buffer_no_jh:
2231 spin_unlock(&journal->j_list_lock);
2232 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002233 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002234zap_buffer_unlocked:
2235 clear_buffer_dirty(bh);
2236 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2237 clear_buffer_mapped(bh);
2238 clear_buffer_req(bh);
2239 clear_buffer_new(bh);
Eric Sandeen15291162012-02-20 17:53:01 -05002240 clear_buffer_delay(bh);
2241 clear_buffer_unwritten(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002242 bh->b_bdev = NULL;
2243 return may_free;
2244}
2245
2246/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002247 * void jbd2_journal_invalidatepage()
Dave Kleikamp470decc2006-10-11 01:20:57 -07002248 * @journal: journal to use for flush...
2249 * @page: page to flush
Lukas Czerner259709b2013-05-21 23:20:03 -04002250 * @offset: start of the range to invalidate
2251 * @length: length of the range to invalidate
Dave Kleikamp470decc2006-10-11 01:20:57 -07002252 *
Lukas Czerner259709b2013-05-21 23:20:03 -04002253 * Reap page buffers containing data after in the specified range in page.
2254 * Can return -EBUSY if buffers are part of the committing transaction and
2255 * the page is straddling i_size. Caller then has to wait for current commit
2256 * and try again.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002257 */
Jan Kara53e87262012-12-25 13:29:52 -05002258int jbd2_journal_invalidatepage(journal_t *journal,
2259 struct page *page,
Lukas Czerner259709b2013-05-21 23:20:03 -04002260 unsigned int offset,
2261 unsigned int length)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002262{
2263 struct buffer_head *head, *bh, *next;
Lukas Czerner259709b2013-05-21 23:20:03 -04002264 unsigned int stop = offset + length;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002265 unsigned int curr_off = 0;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002266 int partial_page = (offset || length < PAGE_SIZE);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002267 int may_free = 1;
Jan Kara53e87262012-12-25 13:29:52 -05002268 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002269
2270 if (!PageLocked(page))
2271 BUG();
2272 if (!page_has_buffers(page))
Jan Kara53e87262012-12-25 13:29:52 -05002273 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002274
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002275 BUG_ON(stop > PAGE_SIZE || stop < length);
Lukas Czerner259709b2013-05-21 23:20:03 -04002276
Dave Kleikamp470decc2006-10-11 01:20:57 -07002277 /* We will potentially be playing with lists other than just the
2278 * data lists (especially for journaled data mode), so be
2279 * cautious in our locking. */
2280
2281 head = bh = page_buffers(page);
2282 do {
2283 unsigned int next_off = curr_off + bh->b_size;
2284 next = bh->b_this_page;
2285
Lukas Czerner259709b2013-05-21 23:20:03 -04002286 if (next_off > stop)
2287 return 0;
2288
Dave Kleikamp470decc2006-10-11 01:20:57 -07002289 if (offset <= curr_off) {
2290 /* This block is wholly outside the truncation point */
2291 lock_buffer(bh);
Lukas Czerner259709b2013-05-21 23:20:03 -04002292 ret = journal_unmap_buffer(journal, bh, partial_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002293 unlock_buffer(bh);
Jan Kara53e87262012-12-25 13:29:52 -05002294 if (ret < 0)
2295 return ret;
2296 may_free &= ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002297 }
2298 curr_off = next_off;
2299 bh = next;
2300
2301 } while (bh != head);
2302
Lukas Czerner259709b2013-05-21 23:20:03 -04002303 if (!partial_page) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002304 if (may_free && try_to_free_buffers(page))
2305 J_ASSERT(!page_has_buffers(page));
2306 }
Jan Kara53e87262012-12-25 13:29:52 -05002307 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002308}
2309
2310/*
2311 * File a buffer on the given transaction list.
2312 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002313void __jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002314 transaction_t *transaction, int jlist)
2315{
2316 struct journal_head **list = NULL;
2317 int was_dirty = 0;
2318 struct buffer_head *bh = jh2bh(jh);
2319
2320 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2321 assert_spin_locked(&transaction->t_journal->j_list_lock);
2322
2323 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2324 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
Mingming Cao40191912008-01-28 23:58:27 -05002325 jh->b_transaction == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002326
2327 if (jh->b_transaction && jh->b_jlist == jlist)
2328 return;
2329
Dave Kleikamp470decc2006-10-11 01:20:57 -07002330 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2331 jlist == BJ_Shadow || jlist == BJ_Forget) {
Jan Karaf91d1d02009-07-13 16:16:20 -04002332 /*
2333 * For metadata buffers, we track dirty bit in buffer_jbddirty
2334 * instead of buffer_dirty. We should not see a dirty bit set
2335 * here because we clear it in do_get_write_access but e.g.
2336 * tune2fs can modify the sb and set the dirty bit at any time
2337 * so we try to gracefully handle that.
2338 */
2339 if (buffer_dirty(bh))
2340 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002341 if (test_clear_buffer_dirty(bh) ||
2342 test_clear_buffer_jbddirty(bh))
2343 was_dirty = 1;
2344 }
2345
2346 if (jh->b_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002347 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002348 else
2349 jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002350 jh->b_transaction = transaction;
2351
2352 switch (jlist) {
2353 case BJ_None:
2354 J_ASSERT_JH(jh, !jh->b_committed_data);
2355 J_ASSERT_JH(jh, !jh->b_frozen_data);
2356 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002357 case BJ_Metadata:
2358 transaction->t_nr_buffers++;
2359 list = &transaction->t_buffers;
2360 break;
2361 case BJ_Forget:
2362 list = &transaction->t_forget;
2363 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002364 case BJ_Shadow:
2365 list = &transaction->t_shadow_list;
2366 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002367 case BJ_Reserved:
2368 list = &transaction->t_reserved_list;
2369 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002370 }
2371
2372 __blist_add_buffer(list, jh);
2373 jh->b_jlist = jlist;
2374
2375 if (was_dirty)
2376 set_buffer_jbddirty(bh);
2377}
2378
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002379void jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002380 transaction_t *transaction, int jlist)
2381{
2382 jbd_lock_bh_state(jh2bh(jh));
2383 spin_lock(&transaction->t_journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002384 __jbd2_journal_file_buffer(jh, transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002385 spin_unlock(&transaction->t_journal->j_list_lock);
2386 jbd_unlock_bh_state(jh2bh(jh));
2387}
2388
2389/*
2390 * Remove a buffer from its current buffer list in preparation for
2391 * dropping it from its current transaction entirely. If the buffer has
2392 * already started to be used by a subsequent transaction, refile the
2393 * buffer on that transaction's metadata list.
2394 *
Jan Karade1b7942011-06-13 15:38:22 -04002395 * Called under j_list_lock
Dave Kleikamp470decc2006-10-11 01:20:57 -07002396 * Called under jbd_lock_bh_state(jh2bh(jh))
Jan Karade1b7942011-06-13 15:38:22 -04002397 *
2398 * jh and bh may be already free when this function returns
Dave Kleikamp470decc2006-10-11 01:20:57 -07002399 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002400void __jbd2_journal_refile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002401{
dingdinghuaba869022010-02-15 16:35:42 -05002402 int was_dirty, jlist;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002403 struct buffer_head *bh = jh2bh(jh);
2404
2405 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2406 if (jh->b_transaction)
2407 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2408
2409 /* If the buffer is now unused, just drop it. */
2410 if (jh->b_next_transaction == NULL) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002411 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002412 return;
2413 }
2414
2415 /*
2416 * It has been modified by a later transaction: add it to the new
2417 * transaction's metadata list.
2418 */
2419
2420 was_dirty = test_clear_buffer_jbddirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002421 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002422 /*
2423 * We set b_transaction here because b_next_transaction will inherit
2424 * our jh reference and thus __jbd2_journal_file_buffer() must not
2425 * take a new one.
2426 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002427 jh->b_transaction = jh->b_next_transaction;
2428 jh->b_next_transaction = NULL;
dingdinghuaba869022010-02-15 16:35:42 -05002429 if (buffer_freed(bh))
2430 jlist = BJ_Forget;
2431 else if (jh->b_modified)
2432 jlist = BJ_Metadata;
2433 else
2434 jlist = BJ_Reserved;
2435 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002436 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2437
2438 if (was_dirty)
2439 set_buffer_jbddirty(bh);
2440}
2441
2442/*
Jan Karade1b7942011-06-13 15:38:22 -04002443 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2444 * bh reference so that we can safely unlock bh.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002445 *
Jan Karade1b7942011-06-13 15:38:22 -04002446 * The jh and bh may be freed by this call.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002447 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002448void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002449{
2450 struct buffer_head *bh = jh2bh(jh);
2451
Jan Karade1b7942011-06-13 15:38:22 -04002452 /* Get reference so that buffer cannot be freed before we unlock it */
2453 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002454 jbd_lock_bh_state(bh);
2455 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002456 __jbd2_journal_refile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002457 jbd_unlock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002458 spin_unlock(&journal->j_list_lock);
2459 __brelse(bh);
2460}
Jan Karac851ed52008-07-11 19:27:31 -04002461
2462/*
2463 * File inode in the inode list of the handle's transaction
2464 */
2465int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
2466{
2467 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002468 journal_t *journal;
Jan Karac851ed52008-07-11 19:27:31 -04002469
2470 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002471 return -EROFS;
2472 journal = transaction->t_journal;
Jan Karac851ed52008-07-11 19:27:31 -04002473
2474 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2475 transaction->t_tid);
2476
2477 /*
2478 * First check whether inode isn't already on the transaction's
2479 * lists without taking the lock. Note that this check is safe
2480 * without the lock as we cannot race with somebody removing inode
2481 * from the transaction. The reason is that we remove inode from the
2482 * transaction only in journal_release_jbd_inode() and when we commit
2483 * the transaction. We are guarded from the first case by holding
2484 * a reference to the inode. We are safe against the second case
2485 * because if jinode->i_transaction == transaction, commit code
2486 * cannot touch the transaction because we hold reference to it,
2487 * and if jinode->i_next_transaction == transaction, commit code
2488 * will only file the inode where we want it.
2489 */
2490 if (jinode->i_transaction == transaction ||
2491 jinode->i_next_transaction == transaction)
2492 return 0;
2493
2494 spin_lock(&journal->j_list_lock);
2495
2496 if (jinode->i_transaction == transaction ||
2497 jinode->i_next_transaction == transaction)
2498 goto done;
2499
Jan Kara81be12c2011-05-24 11:52:40 -04002500 /*
2501 * We only ever set this variable to 1 so the test is safe. Since
2502 * t_need_data_flush is likely to be set, we do the test to save some
2503 * cacheline bouncing
2504 */
2505 if (!transaction->t_need_data_flush)
2506 transaction->t_need_data_flush = 1;
Jan Karac851ed52008-07-11 19:27:31 -04002507 /* On some different transaction's list - should be
2508 * the committing one */
2509 if (jinode->i_transaction) {
2510 J_ASSERT(jinode->i_next_transaction == NULL);
2511 J_ASSERT(jinode->i_transaction ==
2512 journal->j_committing_transaction);
2513 jinode->i_next_transaction = transaction;
2514 goto done;
2515 }
2516 /* Not on any transaction list... */
2517 J_ASSERT(!jinode->i_next_transaction);
2518 jinode->i_transaction = transaction;
2519 list_add(&jinode->i_list, &transaction->t_inode_list);
2520done:
2521 spin_unlock(&journal->j_list_lock);
2522
2523 return 0;
2524}
2525
2526/*
Jan Kara7f5aa212009-02-10 11:15:34 -05002527 * File truncate and transaction commit interact with each other in a
2528 * non-trivial way. If a transaction writing data block A is
2529 * committing, we cannot discard the data by truncate until we have
2530 * written them. Otherwise if we crashed after the transaction with
2531 * write has committed but before the transaction with truncate has
2532 * committed, we could see stale data in block A. This function is a
2533 * helper to solve this problem. It starts writeout of the truncated
2534 * part in case it is in the committing transaction.
2535 *
2536 * Filesystem code must call this function when inode is journaled in
2537 * ordered mode before truncation happens and after the inode has been
2538 * placed on orphan list with the new inode size. The second condition
2539 * avoids the race that someone writes new data and we start
2540 * committing the transaction after this function has been called but
2541 * before a transaction for truncate is started (and furthermore it
2542 * allows us to optimize the case where the addition to orphan list
2543 * happens in the same transaction as write --- we don't have to write
2544 * any data in such case).
Jan Karac851ed52008-07-11 19:27:31 -04002545 */
Jan Kara7f5aa212009-02-10 11:15:34 -05002546int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2547 struct jbd2_inode *jinode,
Jan Karac851ed52008-07-11 19:27:31 -04002548 loff_t new_size)
2549{
Jan Kara7f5aa212009-02-10 11:15:34 -05002550 transaction_t *inode_trans, *commit_trans;
Jan Karac851ed52008-07-11 19:27:31 -04002551 int ret = 0;
2552
Jan Kara7f5aa212009-02-10 11:15:34 -05002553 /* This is a quick check to avoid locking if not necessary */
2554 if (!jinode->i_transaction)
Jan Karac851ed52008-07-11 19:27:31 -04002555 goto out;
Jan Kara7f5aa212009-02-10 11:15:34 -05002556 /* Locks are here just to force reading of recent values, it is
2557 * enough that the transaction was not committing before we started
2558 * a transaction adding the inode to orphan list */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002559 read_lock(&journal->j_state_lock);
Jan Karac851ed52008-07-11 19:27:31 -04002560 commit_trans = journal->j_committing_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002561 read_unlock(&journal->j_state_lock);
Jan Kara7f5aa212009-02-10 11:15:34 -05002562 spin_lock(&journal->j_list_lock);
2563 inode_trans = jinode->i_transaction;
2564 spin_unlock(&journal->j_list_lock);
2565 if (inode_trans == commit_trans) {
2566 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
Jan Karac851ed52008-07-11 19:27:31 -04002567 new_size, LLONG_MAX);
2568 if (ret)
2569 jbd2_journal_abort(journal, ret);
2570 }
2571out:
2572 return ret;
2573}