blob: 5f26139a165a7081dd2f835222f26dcbc5d3bf4a [file] [log] [blame]
Dave Kleikamp470decc2006-10-11 01:20:57 -07001/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002 * linux/fs/jbd2/commit.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 * Journal commit routines for the generic filesystem journaling code;
13 * part of the ext2fs journaling system.
14 */
15
16#include <linux/time.h>
17#include <linux/fs.h>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070018#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070019#include <linux/errno.h>
20#include <linux/slab.h>
21#include <linux/mm.h>
22#include <linux/pagemap.h>
Johann Lombardi8e85fb32008-01-28 23:58:27 -050023#include <linux/jiffies.h>
Girish Shilamkar818d2762008-01-28 23:58:27 -050024#include <linux/crc32.h>
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -040025#include <linux/writeback.h>
26#include <linux/backing-dev.h>
Theodore Ts'ofd984962009-01-05 21:34:13 -050027#include <linux/bio.h>
Theodore Ts'o0e3d2a62009-09-11 09:30:12 -040028#include <linux/blkdev.h>
Brian King39e3ac22010-10-27 21:25:12 -040029#include <linux/bitops.h>
Theodore Ts'o879c5e62009-06-17 11:47:48 -040030#include <trace/events/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070031
32/*
Jan Karab34090e2013-06-04 12:08:56 -040033 * IO end handler for temporary buffer_heads handling writes to the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -070034 */
35static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
36{
Jan Karab34090e2013-06-04 12:08:56 -040037 struct buffer_head *orig_bh = bh->b_private;
38
Dave Kleikamp470decc2006-10-11 01:20:57 -070039 BUFFER_TRACE(bh, "");
40 if (uptodate)
41 set_buffer_uptodate(bh);
42 else
43 clear_buffer_uptodate(bh);
Jan Karab34090e2013-06-04 12:08:56 -040044 if (orig_bh) {
45 clear_bit_unlock(BH_Shadow, &orig_bh->b_state);
46 smp_mb__after_clear_bit();
47 wake_up_bit(&orig_bh->b_state, BH_Shadow);
48 }
Dave Kleikamp470decc2006-10-11 01:20:57 -070049 unlock_buffer(bh);
50}
51
52/*
Jan Kara87c89c22008-07-11 19:27:31 -040053 * When an ext4 file is truncated, it is possible that some pages are not
54 * successfully freed, because they are attached to a committing transaction.
Dave Kleikamp470decc2006-10-11 01:20:57 -070055 * After the transaction commits, these pages are left on the LRU, with no
56 * ->mapping, and with attached buffers. These pages are trivially reclaimable
57 * by the VM, but their apparent absence upsets the VM accounting, and it makes
58 * the numbers in /proc/meminfo look odd.
59 *
60 * So here, we have a buffer which has just come off the forget list. Look to
61 * see if we can strip all buffers from the backing page.
62 *
63 * Called under lock_journal(), and possibly under journal_datalist_lock. The
64 * caller provided us with a ref against the buffer, and we drop that here.
65 */
66static void release_buffer_page(struct buffer_head *bh)
67{
68 struct page *page;
69
70 if (buffer_dirty(bh))
71 goto nope;
72 if (atomic_read(&bh->b_count) != 1)
73 goto nope;
74 page = bh->b_page;
75 if (!page)
76 goto nope;
77 if (page->mapping)
78 goto nope;
79
80 /* OK, it's a truncated page */
Nick Piggin529ae9a2008-08-02 12:01:03 +020081 if (!trylock_page(page))
Dave Kleikamp470decc2006-10-11 01:20:57 -070082 goto nope;
83
84 page_cache_get(page);
85 __brelse(bh);
86 try_to_free_buffers(page);
87 unlock_page(page);
88 page_cache_release(page);
89 return;
90
91nope:
92 __brelse(bh);
93}
94
Jan Karae5a120a2013-06-04 12:06:01 -040095static void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
Darrick J. Wong1f56c582012-05-27 08:10:25 -040096{
97 struct commit_header *h;
98 __u32 csum;
99
100 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
101 return;
102
Jan Karae5a120a2013-06-04 12:06:01 -0400103 h = (struct commit_header *)(bh->b_data);
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400104 h->h_chksum_type = 0;
105 h->h_chksum_size = 0;
106 h->h_chksum[0] = 0;
Jan Karae5a120a2013-06-04 12:06:01 -0400107 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400108 h->h_chksum[0] = cpu_to_be32(csum);
109}
110
Dave Kleikamp470decc2006-10-11 01:20:57 -0700111/*
Girish Shilamkar818d2762008-01-28 23:58:27 -0500112 * Done it all: now submit the commit record. We should have
Dave Kleikamp470decc2006-10-11 01:20:57 -0700113 * cleaned up our previous buffers by now, so if we are in abort
114 * mode we can now just skip the rest of the journal write
115 * entirely.
116 *
117 * Returns 1 if the journal needs to be aborted or 0 on success
118 */
Girish Shilamkar818d2762008-01-28 23:58:27 -0500119static int journal_submit_commit_record(journal_t *journal,
120 transaction_t *commit_transaction,
121 struct buffer_head **cbh,
122 __u32 crc32_sum)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700123{
Girish Shilamkar818d2762008-01-28 23:58:27 -0500124 struct commit_header *tmp;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700125 struct buffer_head *bh;
Girish Shilamkar818d2762008-01-28 23:58:27 -0500126 int ret;
Theodore Ts'o736603a2008-07-11 19:27:31 -0400127 struct timespec now = current_kernel_time();
Dave Kleikamp470decc2006-10-11 01:20:57 -0700128
Zhang Huan6cba6112011-04-05 19:16:20 -0400129 *cbh = NULL;
130
Dave Kleikamp470decc2006-10-11 01:20:57 -0700131 if (is_journal_aborted(journal))
132 return 0;
133
Jan Karae5a120a2013-06-04 12:06:01 -0400134 bh = jbd2_journal_get_descriptor_buffer(journal);
135 if (!bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700136 return 1;
137
Girish Shilamkar818d2762008-01-28 23:58:27 -0500138 tmp = (struct commit_header *)bh->b_data;
139 tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
140 tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK);
141 tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
Theodore Ts'o736603a2008-07-11 19:27:31 -0400142 tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
143 tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500144
145 if (JBD2_HAS_COMPAT_FEATURE(journal,
146 JBD2_FEATURE_COMPAT_CHECKSUM)) {
147 tmp->h_chksum_type = JBD2_CRC32_CHKSUM;
148 tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE;
149 tmp->h_chksum[0] = cpu_to_be32(crc32_sum);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700150 }
Jan Karae5a120a2013-06-04 12:06:01 -0400151 jbd2_commit_block_csum_set(journal, bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700152
Jan Karae5a120a2013-06-04 12:06:01 -0400153 BUFFER_TRACE(bh, "submit commit block");
Girish Shilamkar818d2762008-01-28 23:58:27 -0500154 lock_buffer(bh);
Theodore Ts'o45a90bf2008-10-06 12:04:02 -0400155 clear_buffer_dirty(bh);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500156 set_buffer_uptodate(bh);
157 bh->b_end_io = journal_end_buffer_io_sync;
158
159 if (journal->j_flags & JBD2_BARRIER &&
Theodore Ts'o0e3d2a62009-09-11 09:30:12 -0400160 !JBD2_HAS_INCOMPAT_FEATURE(journal,
Christoph Hellwig9c355752010-08-18 05:29:17 -0400161 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT))
Jens Axboe721a9602011-03-09 11:56:30 +0100162 ret = submit_bh(WRITE_SYNC | WRITE_FLUSH_FUA, bh);
Christoph Hellwig9c355752010-08-18 05:29:17 -0400163 else
Jens Axboe721a9602011-03-09 11:56:30 +0100164 ret = submit_bh(WRITE_SYNC, bh);
Christoph Hellwig9c355752010-08-18 05:29:17 -0400165
Girish Shilamkar818d2762008-01-28 23:58:27 -0500166 *cbh = bh;
167 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700168}
169
Girish Shilamkar818d2762008-01-28 23:58:27 -0500170/*
171 * This function along with journal_submit_commit_record
172 * allows to write the commit record asynchronously.
173 */
Theodore Ts'ofd984962009-01-05 21:34:13 -0500174static int journal_wait_on_commit_record(journal_t *journal,
175 struct buffer_head *bh)
Girish Shilamkar818d2762008-01-28 23:58:27 -0500176{
177 int ret = 0;
178
179 clear_buffer_dirty(bh);
180 wait_on_buffer(bh);
181
182 if (unlikely(!buffer_uptodate(bh)))
183 ret = -EIO;
184 put_bh(bh); /* One for getblk() */
Girish Shilamkar818d2762008-01-28 23:58:27 -0500185
186 return ret;
187}
188
189/*
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -0400190 * write the filemap data using writepage() address_space_operations.
191 * We don't do block allocation here even for delalloc. We don't
192 * use writepages() because with dealyed allocation we may be doing
193 * block allocation in writepages().
194 */
195static int journal_submit_inode_data_buffers(struct address_space *mapping)
196{
197 int ret;
198 struct writeback_control wbc = {
199 .sync_mode = WB_SYNC_ALL,
200 .nr_to_write = mapping->nrpages * 2,
201 .range_start = 0,
202 .range_end = i_size_read(mapping->host),
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -0400203 };
204
205 ret = generic_writepages(mapping, &wbc);
206 return ret;
207}
208
209/*
Jan Karac851ed52008-07-11 19:27:31 -0400210 * Submit all the data buffers of inode associated with the transaction to
211 * disk.
212 *
213 * We are in a committing transaction. Therefore no new inode can be added to
214 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
215 * operate on from being released while we write out pages.
216 */
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -0400217static int journal_submit_data_buffers(journal_t *journal,
Jan Karac851ed52008-07-11 19:27:31 -0400218 transaction_t *commit_transaction)
219{
220 struct jbd2_inode *jinode;
221 int err, ret = 0;
222 struct address_space *mapping;
223
224 spin_lock(&journal->j_list_lock);
225 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
226 mapping = jinode->i_vfs_inode->i_mapping;
Brian King39e3ac22010-10-27 21:25:12 -0400227 set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
Jan Karac851ed52008-07-11 19:27:31 -0400228 spin_unlock(&journal->j_list_lock);
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -0400229 /*
230 * submit the inode data buffers. We use writepage
231 * instead of writepages. Because writepages can do
232 * block allocation with delalloc. We need to write
233 * only allocated blocks here.
234 */
Theodore Ts'o879c5e62009-06-17 11:47:48 -0400235 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -0400236 err = journal_submit_inode_data_buffers(mapping);
Jan Karac851ed52008-07-11 19:27:31 -0400237 if (!ret)
238 ret = err;
239 spin_lock(&journal->j_list_lock);
240 J_ASSERT(jinode->i_transaction == commit_transaction);
Brian King39e3ac22010-10-27 21:25:12 -0400241 clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
242 smp_mb__after_clear_bit();
Jan Karac851ed52008-07-11 19:27:31 -0400243 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
244 }
245 spin_unlock(&journal->j_list_lock);
246 return ret;
247}
248
249/*
250 * Wait for data submitted for writeout, refile inodes to proper
251 * transaction if needed.
252 *
253 */
254static int journal_finish_inode_data_buffers(journal_t *journal,
255 transaction_t *commit_transaction)
256{
257 struct jbd2_inode *jinode, *next_i;
258 int err, ret = 0;
259
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -0400260 /* For locking, see the comment in journal_submit_data_buffers() */
Jan Karac851ed52008-07-11 19:27:31 -0400261 spin_lock(&journal->j_list_lock);
262 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
Brian King39e3ac22010-10-27 21:25:12 -0400263 set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
Jan Karac851ed52008-07-11 19:27:31 -0400264 spin_unlock(&journal->j_list_lock);
265 err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping);
Hidehiro Kawaie9e34f42008-07-31 22:26:04 -0400266 if (err) {
267 /*
268 * Because AS_EIO is cleared by
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200269 * filemap_fdatawait_range(), set it again so
Hidehiro Kawaie9e34f42008-07-31 22:26:04 -0400270 * that user process can get -EIO from fsync().
271 */
272 set_bit(AS_EIO,
273 &jinode->i_vfs_inode->i_mapping->flags);
274
275 if (!ret)
276 ret = err;
277 }
Jan Karac851ed52008-07-11 19:27:31 -0400278 spin_lock(&journal->j_list_lock);
Brian King39e3ac22010-10-27 21:25:12 -0400279 clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
280 smp_mb__after_clear_bit();
Jan Karac851ed52008-07-11 19:27:31 -0400281 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
282 }
283
284 /* Now refile inode to proper lists */
285 list_for_each_entry_safe(jinode, next_i,
286 &commit_transaction->t_inode_list, i_list) {
287 list_del(&jinode->i_list);
288 if (jinode->i_next_transaction) {
289 jinode->i_transaction = jinode->i_next_transaction;
290 jinode->i_next_transaction = NULL;
291 list_add(&jinode->i_list,
292 &jinode->i_transaction->t_inode_list);
293 } else {
294 jinode->i_transaction = NULL;
295 }
296 }
297 spin_unlock(&journal->j_list_lock);
298
299 return ret;
300}
301
Girish Shilamkar818d2762008-01-28 23:58:27 -0500302static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
303{
304 struct page *page = bh->b_page;
305 char *addr;
306 __u32 checksum;
307
Cong Wang303a8f22011-11-25 23:14:31 +0800308 addr = kmap_atomic(page);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500309 checksum = crc32_be(crc32_sum,
310 (void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
Cong Wang303a8f22011-11-25 23:14:31 +0800311 kunmap_atomic(addr);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500312
313 return checksum;
314}
315
316static void write_tag_block(int tag_bytes, journal_block_tag_t *tag,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700317 unsigned long long block)
Zach Brownb517bea2006-10-11 01:21:08 -0700318{
319 tag->t_blocknr = cpu_to_be32(block & (u32)~0);
Mingming Caocd02ff02007-10-16 18:38:25 -0400320 if (tag_bytes > JBD2_TAG_SIZE32)
Zach Brownb517bea2006-10-11 01:21:08 -0700321 tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
322}
323
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400324static void jbd2_descr_block_csum_set(journal_t *j,
Jan Karae5a120a2013-06-04 12:06:01 -0400325 struct buffer_head *bh)
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400326{
327 struct jbd2_journal_block_tail *tail;
328 __u32 csum;
329
330 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
331 return;
332
Jan Karae5a120a2013-06-04 12:06:01 -0400333 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400334 sizeof(struct jbd2_journal_block_tail));
335 tail->t_checksum = 0;
Jan Karae5a120a2013-06-04 12:06:01 -0400336 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400337 tail->t_checksum = cpu_to_be32(csum);
338}
339
Darrick J. Wongc3900872012-05-27 08:12:12 -0400340static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
341 struct buffer_head *bh, __u32 sequence)
342{
343 struct page *page = bh->b_page;
344 __u8 *addr;
Darrick J. Wongeee06c52013-05-28 07:31:59 -0400345 __u32 csum32;
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400346 __be32 seq;
Darrick J. Wongc3900872012-05-27 08:12:12 -0400347
348 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
349 return;
350
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400351 seq = cpu_to_be32(sequence);
Cong Wang906adea2012-06-23 11:24:48 +0800352 addr = kmap_atomic(page);
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400353 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
Darrick J. Wongeee06c52013-05-28 07:31:59 -0400354 csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data),
355 bh->b_size);
Cong Wang906adea2012-06-23 11:24:48 +0800356 kunmap_atomic(addr);
Darrick J. Wongc3900872012-05-27 08:12:12 -0400357
Darrick J. Wongeee06c52013-05-28 07:31:59 -0400358 /* We only have space to store the lower 16 bits of the crc32c. */
359 tag->t_checksum = cpu_to_be16(csum32);
Darrick J. Wongc3900872012-05-27 08:12:12 -0400360}
Dave Kleikamp470decc2006-10-11 01:20:57 -0700361/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700362 * jbd2_journal_commit_transaction
Dave Kleikamp470decc2006-10-11 01:20:57 -0700363 *
364 * The primary function for committing a transaction to the log. This
365 * function is called by the journal thread to begin a complete commit.
366 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700367void jbd2_journal_commit_transaction(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700368{
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500369 struct transaction_stats_s stats;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700370 transaction_t *commit_transaction;
Jan Karae5a120a2013-06-04 12:06:01 -0400371 struct journal_head *jh;
372 struct buffer_head *descriptor;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700373 struct buffer_head **wbuf = journal->j_wbuf;
374 int bufs;
375 int flags;
376 int err;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700377 unsigned long long blocknr;
Josef Bacike07f7182008-11-26 01:14:26 -0500378 ktime_t start_time;
379 u64 commit_time;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700380 char *tagp = NULL;
381 journal_header_t *header;
382 journal_block_tag_t *tag = NULL;
383 int space_left = 0;
384 int first_tag = 0;
385 int tag_flag;
Dmitry Monakhov794446c2013-04-03 22:06:52 -0400386 int i;
Zach Brownb517bea2006-10-11 01:21:08 -0700387 int tag_bytes = journal_tag_bytes(journal);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500388 struct buffer_head *cbh = NULL; /* For transactional checksums */
389 __u32 crc32_sum = ~0;
Jens Axboe82f04ab2011-03-17 11:01:52 +0100390 struct blk_plug plug;
Jan Kara33395782012-03-13 22:45:38 -0400391 /* Tail of the journal */
392 unsigned long first_block;
393 tid_t first_tid;
394 int update_tail;
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400395 int csum_size = 0;
Jan Karaf5113ef2013-06-04 12:01:45 -0400396 LIST_HEAD(io_bufs);
Jan Karae5a120a2013-06-04 12:06:01 -0400397 LIST_HEAD(log_bufs);
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400398
399 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
400 csum_size = sizeof(struct jbd2_journal_block_tail);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700401
402 /*
403 * First job: lock down the current transaction and wait for
404 * all outstanding updates to complete.
405 */
406
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700407 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
408 if (journal->j_flags & JBD2_FLUSHED) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700409 jbd_debug(3, "super block updated\n");
Jan Karaa78bb112012-03-13 15:43:04 -0400410 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara79feb522012-03-13 22:22:54 -0400411 /*
412 * We hold j_checkpoint_mutex so tail cannot change under us.
413 * We don't need any special data guarantees for writing sb
414 * since journal is empty and it is ok for write to be
415 * flushed only with transaction commit.
416 */
417 jbd2_journal_update_sb_log_tail(journal,
418 journal->j_tail_sequence,
419 journal->j_tail,
420 WRITE_SYNC);
Jan Karaa78bb112012-03-13 15:43:04 -0400421 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700422 } else {
423 jbd_debug(3, "superblock not updated\n");
424 }
425
426 J_ASSERT(journal->j_running_transaction != NULL);
427 J_ASSERT(journal->j_committing_transaction == NULL);
428
429 commit_transaction = journal->j_running_transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700430
Theodore Ts'o879c5e62009-06-17 11:47:48 -0400431 trace_jbd2_start_commit(journal, commit_transaction);
Eryu Guanf2a44522011-11-01 19:09:18 -0400432 jbd_debug(1, "JBD2: starting commit of transaction %d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700433 commit_transaction->t_tid);
434
Theodore Ts'oa931da62010-08-03 21:35:12 -0400435 write_lock(&journal->j_state_lock);
Paul Gortmaker3ca841c2013-06-12 22:46:35 -0400436 J_ASSERT(commit_transaction->t_state == T_RUNNING);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700437 commit_transaction->t_state = T_LOCKED;
438
Theodore Ts'o879c5e62009-06-17 11:47:48 -0400439 trace_jbd2_commit_locking(journal, commit_transaction);
Theodore Ts'obf699322009-09-30 00:32:06 -0400440 stats.run.rs_wait = commit_transaction->t_max_wait;
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500441 stats.run.rs_request_delay = 0;
Theodore Ts'obf699322009-09-30 00:32:06 -0400442 stats.run.rs_locked = jiffies;
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500443 if (commit_transaction->t_requested)
444 stats.run.rs_request_delay =
445 jbd2_time_diff(commit_transaction->t_requested,
446 stats.run.rs_locked);
Theodore Ts'obf699322009-09-30 00:32:06 -0400447 stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
448 stats.run.rs_locked);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500449
Dave Kleikamp470decc2006-10-11 01:20:57 -0700450 spin_lock(&commit_transaction->t_handle_lock);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400451 while (atomic_read(&commit_transaction->t_updates)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700452 DEFINE_WAIT(wait);
453
454 prepare_to_wait(&journal->j_wait_updates, &wait,
455 TASK_UNINTERRUPTIBLE);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400456 if (atomic_read(&commit_transaction->t_updates)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700457 spin_unlock(&commit_transaction->t_handle_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400458 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700459 schedule();
Theodore Ts'oa931da62010-08-03 21:35:12 -0400460 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700461 spin_lock(&commit_transaction->t_handle_lock);
462 }
463 finish_wait(&journal->j_wait_updates, &wait);
464 }
465 spin_unlock(&commit_transaction->t_handle_lock);
466
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400467 J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
Dave Kleikamp470decc2006-10-11 01:20:57 -0700468 journal->j_max_transaction_buffers);
469
470 /*
471 * First thing we are allowed to do is to discard any remaining
472 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
473 * that there are no such buffers: if a large filesystem
474 * operation like a truncate needs to split itself over multiple
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700475 * transactions, then it may try to do a jbd2_journal_restart() while
Dave Kleikamp470decc2006-10-11 01:20:57 -0700476 * there are still BJ_Reserved buffers outstanding. These must
477 * be released cleanly from the current transaction.
478 *
479 * In this case, the filesystem must still reserve write access
480 * again before modifying the buffer in the new transaction, but
481 * we do not require it to remember exactly which old buffers it
482 * has reserved. This is consistent with the existing behaviour
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700483 * that multiple jbd2_journal_get_write_access() calls to the same
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300484 * buffer are perfectly permissible.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700485 */
486 while (commit_transaction->t_reserved_list) {
487 jh = commit_transaction->t_reserved_list;
488 JBUFFER_TRACE(jh, "reserved, unused: refile");
489 /*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700490 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
Dave Kleikamp470decc2006-10-11 01:20:57 -0700491 * leave undo-committed data.
492 */
493 if (jh->b_committed_data) {
494 struct buffer_head *bh = jh2bh(jh);
495
496 jbd_lock_bh_state(bh);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400497 jbd2_free(jh->b_committed_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700498 jh->b_committed_data = NULL;
499 jbd_unlock_bh_state(bh);
500 }
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700501 jbd2_journal_refile_buffer(journal, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700502 }
503
504 /*
505 * Now try to drop any written-back buffers from the journal's
506 * checkpoint lists. We do this *before* commit because it potentially
507 * frees some memory
508 */
509 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700510 __jbd2_journal_clean_checkpoint_list(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700511 spin_unlock(&journal->j_list_lock);
512
Eryu Guanf2a44522011-11-01 19:09:18 -0400513 jbd_debug(3, "JBD2: commit phase 1\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700514
515 /*
Yongqiang Yang1ba37262011-12-28 17:46:46 -0500516 * Clear revoked flag to reflect there is no revoked buffers
517 * in the next transaction which is going to be started.
518 */
519 jbd2_clear_buffer_revoked_flags(journal);
520
521 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700522 * Switch to a new revoke table.
523 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700524 jbd2_journal_switch_revoke_table(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700525
Jan Kara8f7d89f2013-06-04 12:35:11 -0400526 /*
527 * Reserved credits cannot be claimed anymore, free them
528 */
529 atomic_sub(atomic_read(&journal->j_reserved_credits),
530 &commit_transaction->t_outstanding_credits);
531
Theodore Ts'o879c5e62009-06-17 11:47:48 -0400532 trace_jbd2_commit_flushing(journal, commit_transaction);
Theodore Ts'obf699322009-09-30 00:32:06 -0400533 stats.run.rs_flushing = jiffies;
534 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
535 stats.run.rs_flushing);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500536
Dave Kleikamp470decc2006-10-11 01:20:57 -0700537 commit_transaction->t_state = T_FLUSH;
538 journal->j_committing_transaction = commit_transaction;
539 journal->j_running_transaction = NULL;
Josef Bacike07f7182008-11-26 01:14:26 -0500540 start_time = ktime_get();
Dave Kleikamp470decc2006-10-11 01:20:57 -0700541 commit_transaction->t_log_start = journal->j_head;
542 wake_up(&journal->j_wait_transaction_locked);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400543 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700544
Paul Gortmakercfc7bc82013-06-12 22:56:35 -0400545 jbd_debug(3, "JBD2: commit phase 2a\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700546
547 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700548 * Now start flushing things to disk, in the order they appear
549 * on the transaction lists. Data blocks go first.
550 */
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -0400551 err = journal_submit_data_buffers(journal, commit_transaction);
Jan Karac851ed52008-07-11 19:27:31 -0400552 if (err)
553 jbd2_journal_abort(journal, err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700554
Jens Axboe82f04ab2011-03-17 11:01:52 +0100555 blk_start_plug(&plug);
Theodore Ts'o67c457a2009-04-14 07:50:56 -0400556 jbd2_journal_write_revoke_records(journal, commit_transaction,
Jan Karae5a120a2013-06-04 12:06:01 -0400557 &log_bufs, WRITE_SYNC);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700558
Paul Gortmakercfc7bc82013-06-12 22:56:35 -0400559 jbd_debug(3, "JBD2: commit phase 2b\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700560
561 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700562 * Way to go: we have now written out all of the data for a
563 * transaction! Now comes the tricky part: we need to write out
564 * metadata. Loop over the transaction's entire buffer list:
565 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400566 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700567 commit_transaction->t_state = T_COMMIT;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400568 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700569
Theodore Ts'o879c5e62009-06-17 11:47:48 -0400570 trace_jbd2_commit_logging(journal, commit_transaction);
Theodore Ts'obf699322009-09-30 00:32:06 -0400571 stats.run.rs_logging = jiffies;
572 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
573 stats.run.rs_logging);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400574 stats.run.rs_blocks =
575 atomic_read(&commit_transaction->t_outstanding_credits);
Theodore Ts'obf699322009-09-30 00:32:06 -0400576 stats.run.rs_blocks_logged = 0;
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500577
Josef Bacik1dfc3222008-04-17 10:38:59 -0400578 J_ASSERT(commit_transaction->t_nr_buffers <=
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400579 atomic_read(&commit_transaction->t_outstanding_credits));
Josef Bacik1dfc3222008-04-17 10:38:59 -0400580
Jan Kara87c89c22008-07-11 19:27:31 -0400581 err = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700582 bufs = 0;
Jan Karae5a120a2013-06-04 12:06:01 -0400583 descriptor = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700584 while (commit_transaction->t_buffers) {
585
586 /* Find the next buffer to be journaled... */
587
588 jh = commit_transaction->t_buffers;
589
590 /* If we're in abort mode, we just un-journal the buffer and
Hidehiro Kawai7ad74452008-10-10 20:29:31 -0400591 release it. */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700592
593 if (is_journal_aborted(journal)) {
Hidehiro Kawai7ad74452008-10-10 20:29:31 -0400594 clear_buffer_jbddirty(jh2bh(jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700595 JBUFFER_TRACE(jh, "journal is aborting: refile");
Joel Beckere06c8222008-09-11 15:35:47 -0700596 jbd2_buffer_abort_trigger(jh,
597 jh->b_frozen_data ?
598 jh->b_frozen_triggers :
599 jh->b_triggers);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700600 jbd2_journal_refile_buffer(journal, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700601 /* If that was the last one, we need to clean up
602 * any descriptor buffers which may have been
603 * already allocated, even if we are now
604 * aborting. */
605 if (!commit_transaction->t_buffers)
606 goto start_journal_io;
607 continue;
608 }
609
610 /* Make sure we have a descriptor block in which to
611 record the metadata buffer. */
612
613 if (!descriptor) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700614 J_ASSERT (bufs == 0);
615
Eryu Guanf2a44522011-11-01 19:09:18 -0400616 jbd_debug(4, "JBD2: get descriptor\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700617
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700618 descriptor = jbd2_journal_get_descriptor_buffer(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700619 if (!descriptor) {
Jan Karaa7fa2ba2007-10-16 18:38:25 -0400620 jbd2_journal_abort(journal, -EIO);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700621 continue;
622 }
623
Eryu Guanf2a44522011-11-01 19:09:18 -0400624 jbd_debug(4, "JBD2: got buffer %llu (%p)\n",
Jan Karae5a120a2013-06-04 12:06:01 -0400625 (unsigned long long)descriptor->b_blocknr,
626 descriptor->b_data);
627 header = (journal_header_t *)descriptor->b_data;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700628 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
629 header->h_blocktype = cpu_to_be32(JBD2_DESCRIPTOR_BLOCK);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700630 header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
631
Jan Karae5a120a2013-06-04 12:06:01 -0400632 tagp = &descriptor->b_data[sizeof(journal_header_t)];
633 space_left = descriptor->b_size -
634 sizeof(journal_header_t);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700635 first_tag = 1;
Jan Karae5a120a2013-06-04 12:06:01 -0400636 set_buffer_jwrite(descriptor);
637 set_buffer_dirty(descriptor);
638 wbuf[bufs++] = descriptor;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700639
640 /* Record it so that we can wait for IO
641 completion later */
Jan Karae5a120a2013-06-04 12:06:01 -0400642 BUFFER_TRACE(descriptor, "ph3: file as descriptor");
643 jbd2_file_log_bh(&log_bufs, descriptor);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700644 }
645
646 /* Where is the buffer to be written? */
647
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700648 err = jbd2_journal_next_log_block(journal, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700649 /* If the block mapping failed, just abandon the buffer
650 and repeat this loop: we'll fall into the
651 refile-on-abort condition above. */
652 if (err) {
Jan Karaa7fa2ba2007-10-16 18:38:25 -0400653 jbd2_journal_abort(journal, err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700654 continue;
655 }
656
657 /*
658 * start_this_handle() uses t_outstanding_credits to determine
659 * the free space in the log, but this counter is changed
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700660 * by jbd2_journal_next_log_block() also.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700661 */
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400662 atomic_dec(&commit_transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700663
664 /* Bump b_count to prevent truncate from stumbling over
665 the shadowed buffer! @@@ This can go if we ever get
Jan Karaf5113ef2013-06-04 12:01:45 -0400666 rid of the shadow pairing of buffers. */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700667 atomic_inc(&jh2bh(jh)->b_count);
668
Dave Kleikamp470decc2006-10-11 01:20:57 -0700669 /*
Jan Karaf5113ef2013-06-04 12:01:45 -0400670 * Make a temporary IO buffer with which to write it out
671 * (this will requeue the metadata buffer to BJ_Shadow).
Dave Kleikamp470decc2006-10-11 01:20:57 -0700672 */
Jan Karaf5113ef2013-06-04 12:01:45 -0400673 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700674 JBUFFER_TRACE(jh, "ph3: write metadata");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700675 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
Jan Karaf5113ef2013-06-04 12:01:45 -0400676 jh, &wbuf[bufs], blocknr);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500677 if (flags < 0) {
678 jbd2_journal_abort(journal, flags);
679 continue;
680 }
Jan Karaf5113ef2013-06-04 12:01:45 -0400681 jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700682
683 /* Record the new block's tag in the current descriptor
684 buffer */
685
686 tag_flag = 0;
687 if (flags & 1)
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700688 tag_flag |= JBD2_FLAG_ESCAPE;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700689 if (!first_tag)
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700690 tag_flag |= JBD2_FLAG_SAME_UUID;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700691
692 tag = (journal_block_tag_t *) tagp;
Zach Brownb517bea2006-10-11 01:21:08 -0700693 write_tag_block(tag_bytes, tag, jh2bh(jh)->b_blocknr);
Darrick J. Wong8f888ef2012-05-22 22:43:41 -0400694 tag->t_flags = cpu_to_be16(tag_flag);
Jan Karaf5113ef2013-06-04 12:01:45 -0400695 jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
Darrick J. Wongc3900872012-05-27 08:12:12 -0400696 commit_transaction->t_tid);
Zach Brownb517bea2006-10-11 01:21:08 -0700697 tagp += tag_bytes;
698 space_left -= tag_bytes;
Jan Karaf5113ef2013-06-04 12:01:45 -0400699 bufs++;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700700
701 if (first_tag) {
702 memcpy (tagp, journal->j_uuid, 16);
703 tagp += 16;
704 space_left -= 16;
705 first_tag = 0;
706 }
707
708 /* If there's no more to do, or if the descriptor is full,
709 let the IO rip! */
710
711 if (bufs == journal->j_wbufsize ||
712 commit_transaction->t_buffers == NULL ||
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400713 space_left < tag_bytes + 16 + csum_size) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700714
Eryu Guanf2a44522011-11-01 19:09:18 -0400715 jbd_debug(4, "JBD2: Submit %d IOs\n", bufs);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700716
717 /* Write an end-of-descriptor marker before
718 submitting the IOs. "tag" still points to
719 the last tag we set up. */
720
Darrick J. Wong8f888ef2012-05-22 22:43:41 -0400721 tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700722
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400723 jbd2_descr_block_csum_set(journal, descriptor);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700724start_journal_io:
725 for (i = 0; i < bufs; i++) {
726 struct buffer_head *bh = wbuf[i];
Girish Shilamkar818d2762008-01-28 23:58:27 -0500727 /*
728 * Compute checksum.
729 */
730 if (JBD2_HAS_COMPAT_FEATURE(journal,
731 JBD2_FEATURE_COMPAT_CHECKSUM)) {
732 crc32_sum =
733 jbd2_checksum_data(crc32_sum, bh);
734 }
735
Dave Kleikamp470decc2006-10-11 01:20:57 -0700736 lock_buffer(bh);
737 clear_buffer_dirty(bh);
738 set_buffer_uptodate(bh);
739 bh->b_end_io = journal_end_buffer_io_sync;
Jens Axboe82f04ab2011-03-17 11:01:52 +0100740 submit_bh(WRITE_SYNC, bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700741 }
742 cond_resched();
Theodore Ts'obf699322009-09-30 00:32:06 -0400743 stats.run.rs_blocks_logged += bufs;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700744
745 /* Force a new descriptor to be generated next
746 time round the loop. */
747 descriptor = NULL;
748 bufs = 0;
749 }
750 }
751
Jan Karac851ed52008-07-11 19:27:31 -0400752 err = journal_finish_inode_data_buffers(journal, commit_transaction);
Hidehiro Kawaie9e34f42008-07-31 22:26:04 -0400753 if (err) {
Hidehiro Kawaie9e34f42008-07-31 22:26:04 -0400754 printk(KERN_WARNING
755 "JBD2: Detected IO errors while flushing file data "
Theodore Ts'o05496762008-09-16 14:36:17 -0400756 "on %s\n", journal->j_devname);
Hidehiro Kawai5bf56832008-10-10 22:12:43 -0400757 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
758 jbd2_journal_abort(journal, err);
Hidehiro Kawaie9e34f42008-07-31 22:26:04 -0400759 err = 0;
760 }
Jan Karac851ed52008-07-11 19:27:31 -0400761
Jan Kara33395782012-03-13 22:45:38 -0400762 /*
763 * Get current oldest transaction in the log before we issue flush
764 * to the filesystem device. After the flush we can be sure that
765 * blocks of all older transactions are checkpointed to persistent
766 * storage and we will be safe to update journal start in the
767 * superblock with the numbers we get here.
768 */
769 update_tail =
770 jbd2_journal_get_log_tail(journal, &first_tid, &first_block);
771
Jan Karabbd2be32011-05-24 11:59:18 -0400772 write_lock(&journal->j_state_lock);
Jan Kara33395782012-03-13 22:45:38 -0400773 if (update_tail) {
774 long freed = first_block - journal->j_tail;
775
776 if (first_block < journal->j_tail)
777 freed += journal->j_last - journal->j_first;
778 /* Update tail only if we free significant amount of space */
779 if (freed < journal->j_maxlen / 4)
780 update_tail = 0;
781 }
Jan Karabbd2be32011-05-24 11:59:18 -0400782 J_ASSERT(commit_transaction->t_state == T_COMMIT);
783 commit_transaction->t_state = T_COMMIT_DFLUSH;
784 write_unlock(&journal->j_state_lock);
Jan Kara33395782012-03-13 22:45:38 -0400785
Girish Shilamkar818d2762008-01-28 23:58:27 -0500786 /*
787 * If the journal is not located on the file system device,
788 * then we must flush the file system device before we issue
789 * the commit record
790 */
Jan Kara81be12c2011-05-24 11:52:40 -0400791 if (commit_transaction->t_need_data_flush &&
Girish Shilamkar818d2762008-01-28 23:58:27 -0500792 (journal->j_fs_dev != journal->j_dev) &&
793 (journal->j_flags & JBD2_BARRIER))
Shaohua Li99aa7842012-04-13 10:27:35 +0800794 blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500795
796 /* Done it all: now write the commit record asynchronously. */
797 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
798 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
799 err = journal_submit_commit_record(journal, commit_transaction,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700800 &cbh, crc32_sum);
801 if (err)
802 __jbd2_journal_abort_hard(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700803 }
804
Jens Axboe82f04ab2011-03-17 11:01:52 +0100805 blk_finish_plug(&plug);
806
Dave Kleikamp470decc2006-10-11 01:20:57 -0700807 /* Lo and behold: we have just managed to send a transaction to
808 the log. Before we can commit it, wait for the IO so far to
809 complete. Control buffers being written are on the
810 transaction's t_log_list queue, and metadata buffers are on
Jan Karaf5113ef2013-06-04 12:01:45 -0400811 the io_bufs list.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700812
813 Wait for the buffers in reverse order. That way we are
814 less likely to be woken up until all IOs have completed, and
815 so we incur less scheduling load.
816 */
817
Eryu Guanf2a44522011-11-01 19:09:18 -0400818 jbd_debug(3, "JBD2: commit phase 3\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700819
Jan Karaf5113ef2013-06-04 12:01:45 -0400820 while (!list_empty(&io_bufs)) {
821 struct buffer_head *bh = list_entry(io_bufs.prev,
822 struct buffer_head,
823 b_assoc_buffers);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700824
Jan Karaf5113ef2013-06-04 12:01:45 -0400825 wait_on_buffer(bh);
826 cond_resched();
Dave Kleikamp470decc2006-10-11 01:20:57 -0700827
828 if (unlikely(!buffer_uptodate(bh)))
829 err = -EIO;
Jan Karaf5113ef2013-06-04 12:01:45 -0400830 jbd2_unfile_log_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700831
832 /*
Jan Karaf5113ef2013-06-04 12:01:45 -0400833 * The list contains temporary buffer heads created by
834 * jbd2_journal_write_metadata_buffer().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700835 */
836 BUFFER_TRACE(bh, "dumping temporary bh");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700837 __brelse(bh);
838 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
839 free_buffer_head(bh);
840
Jan Karaf5113ef2013-06-04 12:01:45 -0400841 /* We also have to refile the corresponding shadowed buffer */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700842 jh = commit_transaction->t_shadow_list->b_tprev;
843 bh = jh2bh(jh);
Jan Karaf5113ef2013-06-04 12:01:45 -0400844 clear_buffer_jwrite(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700845 J_ASSERT_BH(bh, buffer_jbddirty(bh));
Jan Karab34090e2013-06-04 12:08:56 -0400846 J_ASSERT_BH(bh, !buffer_shadow(bh));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700847
848 /* The metadata is now released for reuse, but we need
849 to remember it against this transaction so that when
850 we finally commit, we can do any checkpointing
851 required. */
852 JBUFFER_TRACE(jh, "file as BJ_Forget");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700853 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700854 JBUFFER_TRACE(jh, "brelse shadowed buffer");
855 __brelse(bh);
856 }
857
858 J_ASSERT (commit_transaction->t_shadow_list == NULL);
859
Eryu Guanf2a44522011-11-01 19:09:18 -0400860 jbd_debug(3, "JBD2: commit phase 4\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700861
862 /* Here we wait for the revoke record and descriptor record buffers */
Jan Karae5a120a2013-06-04 12:06:01 -0400863 while (!list_empty(&log_bufs)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700864 struct buffer_head *bh;
865
Jan Karae5a120a2013-06-04 12:06:01 -0400866 bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
867 wait_on_buffer(bh);
868 cond_resched();
Dave Kleikamp470decc2006-10-11 01:20:57 -0700869
870 if (unlikely(!buffer_uptodate(bh)))
871 err = -EIO;
872
873 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
874 clear_buffer_jwrite(bh);
Jan Karae5a120a2013-06-04 12:06:01 -0400875 jbd2_unfile_log_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700876 __brelse(bh); /* One for getblk */
877 /* AKPM: bforget here */
878 }
879
Hidehiro Kawai77e841d2008-10-12 16:39:16 -0400880 if (err)
881 jbd2_journal_abort(journal, err);
882
Eryu Guanf2a44522011-11-01 19:09:18 -0400883 jbd_debug(3, "JBD2: commit phase 5\n");
Jan Karabbd2be32011-05-24 11:59:18 -0400884 write_lock(&journal->j_state_lock);
885 J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
886 commit_transaction->t_state = T_COMMIT_JFLUSH;
887 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700888
Girish Shilamkar818d2762008-01-28 23:58:27 -0500889 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
Theodore Ts'o0e3d2a62009-09-11 09:30:12 -0400890 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
Girish Shilamkar818d2762008-01-28 23:58:27 -0500891 err = journal_submit_commit_record(journal, commit_transaction,
892 &cbh, crc32_sum);
893 if (err)
894 __jbd2_journal_abort_hard(journal);
895 }
Zhang Huan6cba6112011-04-05 19:16:20 -0400896 if (cbh)
Theodore Ts'ofd984962009-01-05 21:34:13 -0500897 err = journal_wait_on_commit_record(journal, cbh);
Jan Karaf73bee42010-08-18 15:56:56 +0200898 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
899 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) &&
900 journal->j_flags & JBD2_BARRIER) {
Shaohua Li99aa7842012-04-13 10:27:35 +0800901 blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL);
Jan Karaf73bee42010-08-18 15:56:56 +0200902 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700903
904 if (err)
Jan Karaa7fa2ba2007-10-16 18:38:25 -0400905 jbd2_journal_abort(journal, err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700906
Jan Kara33395782012-03-13 22:45:38 -0400907 /*
908 * Now disk caches for filesystem device are flushed so we are safe to
909 * erase checkpointed transactions from the log by updating journal
910 * superblock.
911 */
912 if (update_tail)
913 jbd2_update_log_tail(journal, first_tid, first_block);
914
Dave Kleikamp470decc2006-10-11 01:20:57 -0700915 /* End of a transaction! Finally, we can do checkpoint
916 processing: any buffers committed as a result of this
917 transaction can be removed from any checkpoint list it was on
918 before. */
919
Eryu Guanf2a44522011-11-01 19:09:18 -0400920 jbd_debug(3, "JBD2: commit phase 6\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700921
Jan Karac851ed52008-07-11 19:27:31 -0400922 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700923 J_ASSERT(commit_transaction->t_buffers == NULL);
924 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700925 J_ASSERT(commit_transaction->t_shadow_list == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700926
927restart_loop:
928 /*
929 * As there are other places (journal_unmap_buffer()) adding buffers
930 * to this list we have to be careful and hold the j_list_lock.
931 */
932 spin_lock(&journal->j_list_lock);
933 while (commit_transaction->t_forget) {
934 transaction_t *cp_transaction;
935 struct buffer_head *bh;
Jan Karade1b7942011-06-13 15:38:22 -0400936 int try_to_free = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700937
938 jh = commit_transaction->t_forget;
939 spin_unlock(&journal->j_list_lock);
940 bh = jh2bh(jh);
Jan Karade1b7942011-06-13 15:38:22 -0400941 /*
942 * Get a reference so that bh cannot be freed before we are
943 * done with it.
944 */
945 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700946 jbd_lock_bh_state(bh);
dingdinghua23e2af32010-02-24 12:11:20 -0500947 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700948
949 /*
950 * If there is undo-protected committed data against
951 * this buffer, then we can remove it now. If it is a
952 * buffer needing such protection, the old frozen_data
953 * field now points to a committed version of the
954 * buffer, so rotate that field to the new committed
955 * data.
956 *
957 * Otherwise, we can just throw away the frozen data now.
Joel Beckere06c8222008-09-11 15:35:47 -0700958 *
959 * We also know that the frozen data has already fired
960 * its triggers if they exist, so we can clear that too.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700961 */
962 if (jh->b_committed_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400963 jbd2_free(jh->b_committed_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700964 jh->b_committed_data = NULL;
965 if (jh->b_frozen_data) {
966 jh->b_committed_data = jh->b_frozen_data;
967 jh->b_frozen_data = NULL;
Joel Beckere06c8222008-09-11 15:35:47 -0700968 jh->b_frozen_triggers = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700969 }
970 } else if (jh->b_frozen_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400971 jbd2_free(jh->b_frozen_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700972 jh->b_frozen_data = NULL;
Joel Beckere06c8222008-09-11 15:35:47 -0700973 jh->b_frozen_triggers = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700974 }
975
976 spin_lock(&journal->j_list_lock);
977 cp_transaction = jh->b_cp_transaction;
978 if (cp_transaction) {
979 JBUFFER_TRACE(jh, "remove from old cp transaction");
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500980 cp_transaction->t_chp_stats.cs_dropped++;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700981 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700982 }
983
984 /* Only re-checkpoint the buffer_head if it is marked
985 * dirty. If the buffer was added to the BJ_Forget list
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700986 * by jbd2_journal_forget, it may no longer be dirty and
Dave Kleikamp470decc2006-10-11 01:20:57 -0700987 * there's no point in keeping a checkpoint record for
988 * it. */
989
Jan Karab794e7a2012-09-26 23:11:13 -0400990 /*
991 * A buffer which has been freed while still being journaled by
992 * a previous transaction.
993 */
994 if (buffer_freed(bh)) {
995 /*
996 * If the running transaction is the one containing
997 * "add to orphan" operation (b_next_transaction !=
998 * NULL), we have to wait for that transaction to
999 * commit before we can really get rid of the buffer.
1000 * So just clear b_modified to not confuse transaction
1001 * credit accounting and refile the buffer to
1002 * BJ_Forget of the running transaction. If the just
1003 * committed transaction contains "add to orphan"
1004 * operation, we can completely invalidate the buffer
1005 * now. We are rather through in that since the
1006 * buffer may be still accessible when blocksize <
1007 * pagesize and it is attached to the last partial
1008 * page.
1009 */
1010 jh->b_modified = 0;
1011 if (!jh->b_next_transaction) {
1012 clear_buffer_freed(bh);
1013 clear_buffer_jbddirty(bh);
1014 clear_buffer_mapped(bh);
1015 clear_buffer_new(bh);
1016 clear_buffer_req(bh);
1017 bh->b_bdev = NULL;
1018 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001019 }
1020
1021 if (buffer_jbddirty(bh)) {
1022 JBUFFER_TRACE(jh, "add to new checkpointing trans");
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001023 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
Hidehiro Kawai7ad74452008-10-10 20:29:31 -04001024 if (is_journal_aborted(journal))
1025 clear_buffer_jbddirty(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001026 } else {
1027 J_ASSERT_BH(bh, !buffer_dirty(bh));
Jan Karade1b7942011-06-13 15:38:22 -04001028 /*
1029 * The buffer on BJ_Forget list and not jbddirty means
Dave Kleikamp470decc2006-10-11 01:20:57 -07001030 * it has been freed by this transaction and hence it
1031 * could not have been reallocated until this
1032 * transaction has committed. *BUT* it could be
1033 * reallocated once we have written all the data to
1034 * disk and before we process the buffer on BJ_Forget
Jan Karade1b7942011-06-13 15:38:22 -04001035 * list.
1036 */
1037 if (!jh->b_next_transaction)
1038 try_to_free = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001039 }
Jan Karade1b7942011-06-13 15:38:22 -04001040 JBUFFER_TRACE(jh, "refile or unfile buffer");
1041 __jbd2_journal_refile_buffer(jh);
1042 jbd_unlock_bh_state(bh);
1043 if (try_to_free)
1044 release_buffer_page(bh); /* Drops bh reference */
1045 else
1046 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001047 cond_resched_lock(&journal->j_list_lock);
1048 }
1049 spin_unlock(&journal->j_list_lock);
1050 /*
Jan Karaf5a7a6b2008-01-28 23:58:27 -05001051 * This is a bit sleazy. We use j_list_lock to protect transition
1052 * of a transaction into T_FINISHED state and calling
1053 * __jbd2_journal_drop_transaction(). Otherwise we could race with
1054 * other checkpointing code processing the transaction...
Dave Kleikamp470decc2006-10-11 01:20:57 -07001055 */
Theodore Ts'oa931da62010-08-03 21:35:12 -04001056 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001057 spin_lock(&journal->j_list_lock);
1058 /*
1059 * Now recheck if some buffers did not get attached to the transaction
1060 * while the lock was dropped...
1061 */
1062 if (commit_transaction->t_forget) {
1063 spin_unlock(&journal->j_list_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001064 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001065 goto restart_loop;
1066 }
1067
Theodore Ts'od4e839d2014-03-08 22:34:10 -05001068 /* Add the transaction to the checkpoint list
1069 * __journal_remove_checkpoint() can not destroy transaction
1070 * under us because it is not marked as T_FINISHED yet */
1071 if (journal->j_checkpoint_transactions == NULL) {
1072 journal->j_checkpoint_transactions = commit_transaction;
1073 commit_transaction->t_cpnext = commit_transaction;
1074 commit_transaction->t_cpprev = commit_transaction;
1075 } else {
1076 commit_transaction->t_cpnext =
1077 journal->j_checkpoint_transactions;
1078 commit_transaction->t_cpprev =
1079 commit_transaction->t_cpnext->t_cpprev;
1080 commit_transaction->t_cpnext->t_cpprev =
1081 commit_transaction;
1082 commit_transaction->t_cpprev->t_cpnext =
1083 commit_transaction;
1084 }
1085 spin_unlock(&journal->j_list_lock);
1086
Dave Kleikamp470decc2006-10-11 01:20:57 -07001087 /* Done with this transaction! */
1088
Eryu Guanf2a44522011-11-01 19:09:18 -04001089 jbd_debug(3, "JBD2: commit phase 7\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001090
Jan Karabbd2be32011-05-24 11:59:18 -04001091 J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001092
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001093 commit_transaction->t_start = jiffies;
Theodore Ts'obf699322009-09-30 00:32:06 -04001094 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
1095 commit_transaction->t_start);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001096
1097 /*
Theodore Ts'obf699322009-09-30 00:32:06 -04001098 * File the transaction statistics
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001099 */
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001100 stats.ts_tid = commit_transaction->t_tid;
Theodore Ts'o8dd42042010-08-03 21:38:29 -04001101 stats.run.rs_handle_count =
1102 atomic_read(&commit_transaction->t_handle_count);
Theodore Ts'obf699322009-09-30 00:32:06 -04001103 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
1104 commit_transaction->t_tid, &stats.run);
Theodore Ts'o42cf3452014-03-08 19:51:16 -05001105 stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0;
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001106
Dmitry Monakhov794446c2013-04-03 22:06:52 -04001107 commit_transaction->t_state = T_COMMIT_CALLBACK;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001108 J_ASSERT(commit_transaction == journal->j_committing_transaction);
1109 journal->j_commit_sequence = commit_transaction->t_tid;
1110 journal->j_committing_transaction = NULL;
Josef Bacike07f7182008-11-26 01:14:26 -05001111 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001112
Josef Bacike07f7182008-11-26 01:14:26 -05001113 /*
1114 * weight the commit time higher than the average time so we don't
1115 * react too strongly to vast changes in the commit time
1116 */
1117 if (likely(journal->j_average_commit_time))
1118 journal->j_average_commit_time = (commit_time +
1119 journal->j_average_commit_time*3) / 4;
1120 else
1121 journal->j_average_commit_time = commit_time;
Dmitry Monakhov794446c2013-04-03 22:06:52 -04001122
Theodore Ts'oa931da62010-08-03 21:35:12 -04001123 write_unlock(&journal->j_state_lock);
Theodore Ts'o6c20ec82008-10-28 21:08:20 -04001124
Aneesh Kumar K.Vfb684072008-11-06 17:50:21 -05001125 if (journal->j_commit_callback)
1126 journal->j_commit_callback(journal, commit_transaction);
1127
Theodore Ts'o879c5e62009-06-17 11:47:48 -04001128 trace_jbd2_end_commit(journal, commit_transaction);
Eryu Guanf2a44522011-11-01 19:09:18 -04001129 jbd_debug(1, "JBD2: commit %d complete, head %d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -07001130 journal->j_commit_sequence, journal->j_tail_sequence);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001131
Dmitry Monakhov794446c2013-04-03 22:06:52 -04001132 write_lock(&journal->j_state_lock);
1133 spin_lock(&journal->j_list_lock);
1134 commit_transaction->t_state = T_FINISHED;
Theodore Ts'od4e839d2014-03-08 22:34:10 -05001135 /* Check if the transaction can be dropped now that we are finished */
Dmitry Monakhov794446c2013-04-03 22:06:52 -04001136 if (commit_transaction->t_checkpoint_list == NULL &&
1137 commit_transaction->t_checkpoint_io_list == NULL) {
1138 __jbd2_journal_drop_transaction(journal, commit_transaction);
1139 jbd2_journal_free_transaction(commit_transaction);
1140 }
1141 spin_unlock(&journal->j_list_lock);
1142 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001143 wake_up(&journal->j_wait_done_commit);
Theodore Ts'o42cf3452014-03-08 19:51:16 -05001144
1145 /*
1146 * Calculate overall stats
1147 */
1148 spin_lock(&journal->j_history_lock);
1149 journal->j_stats.ts_tid++;
1150 journal->j_stats.ts_requested += stats.ts_requested;
1151 journal->j_stats.run.rs_wait += stats.run.rs_wait;
1152 journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
1153 journal->j_stats.run.rs_running += stats.run.rs_running;
1154 journal->j_stats.run.rs_locked += stats.run.rs_locked;
1155 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
1156 journal->j_stats.run.rs_logging += stats.run.rs_logging;
1157 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
1158 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
1159 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
1160 spin_unlock(&journal->j_history_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001161}