Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
f7f4bccb | 2 | * linux/fs/jbd2/commit.c |
470decc6 DK |
3 | * |
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> | |
f7f4bccb | 18 | #include <linux/jbd2.h> |
470decc6 DK |
19 | #include <linux/errno.h> |
20 | #include <linux/slab.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/pagemap.h> | |
8e85fb3f | 23 | #include <linux/jiffies.h> |
818d276c | 24 | #include <linux/crc32.h> |
cd1aac32 AK |
25 | #include <linux/writeback.h> |
26 | #include <linux/backing-dev.h> | |
fd98496f | 27 | #include <linux/bio.h> |
0e3d2a63 | 28 | #include <linux/blkdev.h> |
39e3ac25 | 29 | #include <linux/bitops.h> |
879c5e6b | 30 | #include <trace/events/jbd2.h> |
470decc6 DK |
31 | |
32 | /* | |
33 | * Default IO end handler for temporary BJ_IO buffer_heads. | |
34 | */ | |
35 | static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate) | |
36 | { | |
37 | BUFFER_TRACE(bh, ""); | |
38 | if (uptodate) | |
39 | set_buffer_uptodate(bh); | |
40 | else | |
41 | clear_buffer_uptodate(bh); | |
42 | unlock_buffer(bh); | |
43 | } | |
44 | ||
45 | /* | |
87c89c23 JK |
46 | * When an ext4 file is truncated, it is possible that some pages are not |
47 | * successfully freed, because they are attached to a committing transaction. | |
470decc6 DK |
48 | * After the transaction commits, these pages are left on the LRU, with no |
49 | * ->mapping, and with attached buffers. These pages are trivially reclaimable | |
50 | * by the VM, but their apparent absence upsets the VM accounting, and it makes | |
51 | * the numbers in /proc/meminfo look odd. | |
52 | * | |
53 | * So here, we have a buffer which has just come off the forget list. Look to | |
54 | * see if we can strip all buffers from the backing page. | |
55 | * | |
56 | * Called under lock_journal(), and possibly under journal_datalist_lock. The | |
57 | * caller provided us with a ref against the buffer, and we drop that here. | |
58 | */ | |
59 | static void release_buffer_page(struct buffer_head *bh) | |
60 | { | |
61 | struct page *page; | |
62 | ||
63 | if (buffer_dirty(bh)) | |
64 | goto nope; | |
65 | if (atomic_read(&bh->b_count) != 1) | |
66 | goto nope; | |
67 | page = bh->b_page; | |
68 | if (!page) | |
69 | goto nope; | |
70 | if (page->mapping) | |
71 | goto nope; | |
72 | ||
73 | /* OK, it's a truncated page */ | |
529ae9aa | 74 | if (!trylock_page(page)) |
470decc6 DK |
75 | goto nope; |
76 | ||
77 | page_cache_get(page); | |
78 | __brelse(bh); | |
79 | try_to_free_buffers(page); | |
80 | unlock_page(page); | |
81 | page_cache_release(page); | |
82 | return; | |
83 | ||
84 | nope: | |
85 | __brelse(bh); | |
86 | } | |
87 | ||
e5a120ae | 88 | static void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh) |
1f56c589 DW |
89 | { |
90 | struct commit_header *h; | |
91 | __u32 csum; | |
92 | ||
93 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
94 | return; | |
95 | ||
e5a120ae | 96 | h = (struct commit_header *)(bh->b_data); |
1f56c589 DW |
97 | h->h_chksum_type = 0; |
98 | h->h_chksum_size = 0; | |
99 | h->h_chksum[0] = 0; | |
e5a120ae | 100 | csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); |
1f56c589 DW |
101 | h->h_chksum[0] = cpu_to_be32(csum); |
102 | } | |
103 | ||
818d276c GS |
104 | /* |
105 | * Done it all: now submit the commit record. We should have | |
470decc6 DK |
106 | * cleaned up our previous buffers by now, so if we are in abort |
107 | * mode we can now just skip the rest of the journal write | |
108 | * entirely. | |
109 | * | |
110 | * Returns 1 if the journal needs to be aborted or 0 on success | |
111 | */ | |
818d276c GS |
112 | static int journal_submit_commit_record(journal_t *journal, |
113 | transaction_t *commit_transaction, | |
114 | struct buffer_head **cbh, | |
115 | __u32 crc32_sum) | |
470decc6 | 116 | { |
818d276c | 117 | struct commit_header *tmp; |
470decc6 | 118 | struct buffer_head *bh; |
818d276c | 119 | int ret; |
736603ab | 120 | struct timespec now = current_kernel_time(); |
470decc6 | 121 | |
6cba611e ZH |
122 | *cbh = NULL; |
123 | ||
470decc6 DK |
124 | if (is_journal_aborted(journal)) |
125 | return 0; | |
126 | ||
e5a120ae JK |
127 | bh = jbd2_journal_get_descriptor_buffer(journal); |
128 | if (!bh) | |
470decc6 DK |
129 | return 1; |
130 | ||
818d276c GS |
131 | tmp = (struct commit_header *)bh->b_data; |
132 | tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); | |
133 | tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK); | |
134 | tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid); | |
736603ab TT |
135 | tmp->h_commit_sec = cpu_to_be64(now.tv_sec); |
136 | tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec); | |
818d276c GS |
137 | |
138 | if (JBD2_HAS_COMPAT_FEATURE(journal, | |
139 | JBD2_FEATURE_COMPAT_CHECKSUM)) { | |
140 | tmp->h_chksum_type = JBD2_CRC32_CHKSUM; | |
141 | tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE; | |
142 | tmp->h_chksum[0] = cpu_to_be32(crc32_sum); | |
470decc6 | 143 | } |
e5a120ae | 144 | jbd2_commit_block_csum_set(journal, bh); |
470decc6 | 145 | |
e5a120ae | 146 | BUFFER_TRACE(bh, "submit commit block"); |
818d276c | 147 | lock_buffer(bh); |
45a90bfd | 148 | clear_buffer_dirty(bh); |
818d276c GS |
149 | set_buffer_uptodate(bh); |
150 | bh->b_end_io = journal_end_buffer_io_sync; | |
151 | ||
152 | if (journal->j_flags & JBD2_BARRIER && | |
0e3d2a63 | 153 | !JBD2_HAS_INCOMPAT_FEATURE(journal, |
9c35575b | 154 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) |
721a9602 | 155 | ret = submit_bh(WRITE_SYNC | WRITE_FLUSH_FUA, bh); |
9c35575b | 156 | else |
721a9602 | 157 | ret = submit_bh(WRITE_SYNC, bh); |
9c35575b | 158 | |
818d276c GS |
159 | *cbh = bh; |
160 | return ret; | |
161 | } | |
162 | ||
163 | /* | |
164 | * This function along with journal_submit_commit_record | |
165 | * allows to write the commit record asynchronously. | |
166 | */ | |
fd98496f TT |
167 | static int journal_wait_on_commit_record(journal_t *journal, |
168 | struct buffer_head *bh) | |
818d276c GS |
169 | { |
170 | int ret = 0; | |
171 | ||
172 | clear_buffer_dirty(bh); | |
173 | wait_on_buffer(bh); | |
470decc6 | 174 | |
818d276c GS |
175 | if (unlikely(!buffer_uptodate(bh))) |
176 | ret = -EIO; | |
177 | put_bh(bh); /* One for getblk() */ | |
818d276c GS |
178 | |
179 | return ret; | |
470decc6 DK |
180 | } |
181 | ||
cd1aac32 AK |
182 | /* |
183 | * write the filemap data using writepage() address_space_operations. | |
184 | * We don't do block allocation here even for delalloc. We don't | |
185 | * use writepages() because with dealyed allocation we may be doing | |
186 | * block allocation in writepages(). | |
187 | */ | |
188 | static int journal_submit_inode_data_buffers(struct address_space *mapping) | |
189 | { | |
190 | int ret; | |
191 | struct writeback_control wbc = { | |
192 | .sync_mode = WB_SYNC_ALL, | |
193 | .nr_to_write = mapping->nrpages * 2, | |
194 | .range_start = 0, | |
195 | .range_end = i_size_read(mapping->host), | |
cd1aac32 AK |
196 | }; |
197 | ||
198 | ret = generic_writepages(mapping, &wbc); | |
199 | return ret; | |
200 | } | |
201 | ||
c851ed54 JK |
202 | /* |
203 | * Submit all the data buffers of inode associated with the transaction to | |
204 | * disk. | |
205 | * | |
206 | * We are in a committing transaction. Therefore no new inode can be added to | |
207 | * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently | |
208 | * operate on from being released while we write out pages. | |
209 | */ | |
cd1aac32 | 210 | static int journal_submit_data_buffers(journal_t *journal, |
c851ed54 JK |
211 | transaction_t *commit_transaction) |
212 | { | |
213 | struct jbd2_inode *jinode; | |
214 | int err, ret = 0; | |
215 | struct address_space *mapping; | |
216 | ||
217 | spin_lock(&journal->j_list_lock); | |
218 | list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) { | |
219 | mapping = jinode->i_vfs_inode->i_mapping; | |
39e3ac25 | 220 | set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
c851ed54 | 221 | spin_unlock(&journal->j_list_lock); |
cd1aac32 AK |
222 | /* |
223 | * submit the inode data buffers. We use writepage | |
224 | * instead of writepages. Because writepages can do | |
225 | * block allocation with delalloc. We need to write | |
226 | * only allocated blocks here. | |
227 | */ | |
879c5e6b | 228 | trace_jbd2_submit_inode_data(jinode->i_vfs_inode); |
cd1aac32 | 229 | err = journal_submit_inode_data_buffers(mapping); |
c851ed54 JK |
230 | if (!ret) |
231 | ret = err; | |
232 | spin_lock(&journal->j_list_lock); | |
233 | J_ASSERT(jinode->i_transaction == commit_transaction); | |
39e3ac25 BK |
234 | clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
235 | smp_mb__after_clear_bit(); | |
c851ed54 JK |
236 | wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING); |
237 | } | |
238 | spin_unlock(&journal->j_list_lock); | |
239 | return ret; | |
240 | } | |
241 | ||
242 | /* | |
243 | * Wait for data submitted for writeout, refile inodes to proper | |
244 | * transaction if needed. | |
245 | * | |
246 | */ | |
247 | static int journal_finish_inode_data_buffers(journal_t *journal, | |
248 | transaction_t *commit_transaction) | |
249 | { | |
250 | struct jbd2_inode *jinode, *next_i; | |
251 | int err, ret = 0; | |
252 | ||
cd1aac32 | 253 | /* For locking, see the comment in journal_submit_data_buffers() */ |
c851ed54 JK |
254 | spin_lock(&journal->j_list_lock); |
255 | list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) { | |
39e3ac25 | 256 | set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
c851ed54 JK |
257 | spin_unlock(&journal->j_list_lock); |
258 | err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping); | |
e9e34f4e HK |
259 | if (err) { |
260 | /* | |
261 | * Because AS_EIO is cleared by | |
94004ed7 | 262 | * filemap_fdatawait_range(), set it again so |
e9e34f4e HK |
263 | * that user process can get -EIO from fsync(). |
264 | */ | |
265 | set_bit(AS_EIO, | |
266 | &jinode->i_vfs_inode->i_mapping->flags); | |
267 | ||
268 | if (!ret) | |
269 | ret = err; | |
270 | } | |
c851ed54 | 271 | spin_lock(&journal->j_list_lock); |
39e3ac25 BK |
272 | clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
273 | smp_mb__after_clear_bit(); | |
c851ed54 JK |
274 | wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING); |
275 | } | |
276 | ||
277 | /* Now refile inode to proper lists */ | |
278 | list_for_each_entry_safe(jinode, next_i, | |
279 | &commit_transaction->t_inode_list, i_list) { | |
280 | list_del(&jinode->i_list); | |
281 | if (jinode->i_next_transaction) { | |
282 | jinode->i_transaction = jinode->i_next_transaction; | |
283 | jinode->i_next_transaction = NULL; | |
284 | list_add(&jinode->i_list, | |
285 | &jinode->i_transaction->t_inode_list); | |
286 | } else { | |
287 | jinode->i_transaction = NULL; | |
288 | } | |
289 | } | |
290 | spin_unlock(&journal->j_list_lock); | |
291 | ||
292 | return ret; | |
293 | } | |
294 | ||
818d276c GS |
295 | static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh) |
296 | { | |
297 | struct page *page = bh->b_page; | |
298 | char *addr; | |
299 | __u32 checksum; | |
300 | ||
303a8f2a | 301 | addr = kmap_atomic(page); |
818d276c GS |
302 | checksum = crc32_be(crc32_sum, |
303 | (void *)(addr + offset_in_page(bh->b_data)), bh->b_size); | |
303a8f2a | 304 | kunmap_atomic(addr); |
818d276c GS |
305 | |
306 | return checksum; | |
307 | } | |
308 | ||
309 | static void write_tag_block(int tag_bytes, journal_block_tag_t *tag, | |
18eba7aa | 310 | unsigned long long block) |
b517bea1 ZB |
311 | { |
312 | tag->t_blocknr = cpu_to_be32(block & (u32)~0); | |
cd02ff0b | 313 | if (tag_bytes > JBD2_TAG_SIZE32) |
b517bea1 ZB |
314 | tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1); |
315 | } | |
316 | ||
3caa487f | 317 | static void jbd2_descr_block_csum_set(journal_t *j, |
e5a120ae | 318 | struct buffer_head *bh) |
3caa487f DW |
319 | { |
320 | struct jbd2_journal_block_tail *tail; | |
321 | __u32 csum; | |
322 | ||
323 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
324 | return; | |
325 | ||
e5a120ae | 326 | tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - |
3caa487f DW |
327 | sizeof(struct jbd2_journal_block_tail)); |
328 | tail->t_checksum = 0; | |
e5a120ae | 329 | csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); |
3caa487f DW |
330 | tail->t_checksum = cpu_to_be32(csum); |
331 | } | |
332 | ||
c3900875 DW |
333 | static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag, |
334 | struct buffer_head *bh, __u32 sequence) | |
335 | { | |
336 | struct page *page = bh->b_page; | |
337 | __u8 *addr; | |
eee06c56 | 338 | __u32 csum32; |
c3900875 DW |
339 | |
340 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
341 | return; | |
342 | ||
343 | sequence = cpu_to_be32(sequence); | |
906adea1 | 344 | addr = kmap_atomic(page); |
eee06c56 DW |
345 | csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&sequence, |
346 | sizeof(sequence)); | |
347 | csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data), | |
348 | bh->b_size); | |
906adea1 | 349 | kunmap_atomic(addr); |
c3900875 | 350 | |
eee06c56 DW |
351 | /* We only have space to store the lower 16 bits of the crc32c. */ |
352 | tag->t_checksum = cpu_to_be16(csum32); | |
c3900875 | 353 | } |
470decc6 | 354 | /* |
f7f4bccb | 355 | * jbd2_journal_commit_transaction |
470decc6 DK |
356 | * |
357 | * The primary function for committing a transaction to the log. This | |
358 | * function is called by the journal thread to begin a complete commit. | |
359 | */ | |
f7f4bccb | 360 | void jbd2_journal_commit_transaction(journal_t *journal) |
470decc6 | 361 | { |
8e85fb3f | 362 | struct transaction_stats_s stats; |
470decc6 | 363 | transaction_t *commit_transaction; |
e5a120ae JK |
364 | struct journal_head *jh; |
365 | struct buffer_head *descriptor; | |
470decc6 DK |
366 | struct buffer_head **wbuf = journal->j_wbuf; |
367 | int bufs; | |
368 | int flags; | |
369 | int err; | |
18eba7aa | 370 | unsigned long long blocknr; |
e07f7183 JB |
371 | ktime_t start_time; |
372 | u64 commit_time; | |
470decc6 DK |
373 | char *tagp = NULL; |
374 | journal_header_t *header; | |
375 | journal_block_tag_t *tag = NULL; | |
376 | int space_left = 0; | |
377 | int first_tag = 0; | |
378 | int tag_flag; | |
794446c6 | 379 | int i; |
b517bea1 | 380 | int tag_bytes = journal_tag_bytes(journal); |
818d276c GS |
381 | struct buffer_head *cbh = NULL; /* For transactional checksums */ |
382 | __u32 crc32_sum = ~0; | |
82f04ab4 | 383 | struct blk_plug plug; |
3339578f JK |
384 | /* Tail of the journal */ |
385 | unsigned long first_block; | |
386 | tid_t first_tid; | |
387 | int update_tail; | |
3caa487f | 388 | int csum_size = 0; |
f5113eff | 389 | LIST_HEAD(io_bufs); |
e5a120ae | 390 | LIST_HEAD(log_bufs); |
3caa487f DW |
391 | |
392 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
393 | csum_size = sizeof(struct jbd2_journal_block_tail); | |
470decc6 DK |
394 | |
395 | /* | |
396 | * First job: lock down the current transaction and wait for | |
397 | * all outstanding updates to complete. | |
398 | */ | |
399 | ||
f7f4bccb MC |
400 | /* Do we need to erase the effects of a prior jbd2_journal_flush? */ |
401 | if (journal->j_flags & JBD2_FLUSHED) { | |
470decc6 | 402 | jbd_debug(3, "super block updated\n"); |
a78bb11d | 403 | mutex_lock(&journal->j_checkpoint_mutex); |
79feb521 JK |
404 | /* |
405 | * We hold j_checkpoint_mutex so tail cannot change under us. | |
406 | * We don't need any special data guarantees for writing sb | |
407 | * since journal is empty and it is ok for write to be | |
408 | * flushed only with transaction commit. | |
409 | */ | |
410 | jbd2_journal_update_sb_log_tail(journal, | |
411 | journal->j_tail_sequence, | |
412 | journal->j_tail, | |
413 | WRITE_SYNC); | |
a78bb11d | 414 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
415 | } else { |
416 | jbd_debug(3, "superblock not updated\n"); | |
417 | } | |
418 | ||
419 | J_ASSERT(journal->j_running_transaction != NULL); | |
420 | J_ASSERT(journal->j_committing_transaction == NULL); | |
421 | ||
422 | commit_transaction = journal->j_running_transaction; | |
423 | J_ASSERT(commit_transaction->t_state == T_RUNNING); | |
424 | ||
879c5e6b | 425 | trace_jbd2_start_commit(journal, commit_transaction); |
f2a44523 | 426 | jbd_debug(1, "JBD2: starting commit of transaction %d\n", |
470decc6 DK |
427 | commit_transaction->t_tid); |
428 | ||
a931da6a | 429 | write_lock(&journal->j_state_lock); |
470decc6 DK |
430 | commit_transaction->t_state = T_LOCKED; |
431 | ||
879c5e6b | 432 | trace_jbd2_commit_locking(journal, commit_transaction); |
bf699327 | 433 | stats.run.rs_wait = commit_transaction->t_max_wait; |
9fff24aa | 434 | stats.run.rs_request_delay = 0; |
bf699327 | 435 | stats.run.rs_locked = jiffies; |
9fff24aa TT |
436 | if (commit_transaction->t_requested) |
437 | stats.run.rs_request_delay = | |
438 | jbd2_time_diff(commit_transaction->t_requested, | |
439 | stats.run.rs_locked); | |
bf699327 TT |
440 | stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start, |
441 | stats.run.rs_locked); | |
8e85fb3f | 442 | |
470decc6 | 443 | spin_lock(&commit_transaction->t_handle_lock); |
a51dca9c | 444 | while (atomic_read(&commit_transaction->t_updates)) { |
470decc6 DK |
445 | DEFINE_WAIT(wait); |
446 | ||
447 | prepare_to_wait(&journal->j_wait_updates, &wait, | |
448 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 449 | if (atomic_read(&commit_transaction->t_updates)) { |
470decc6 | 450 | spin_unlock(&commit_transaction->t_handle_lock); |
a931da6a | 451 | write_unlock(&journal->j_state_lock); |
470decc6 | 452 | schedule(); |
a931da6a | 453 | write_lock(&journal->j_state_lock); |
470decc6 DK |
454 | spin_lock(&commit_transaction->t_handle_lock); |
455 | } | |
456 | finish_wait(&journal->j_wait_updates, &wait); | |
457 | } | |
458 | spin_unlock(&commit_transaction->t_handle_lock); | |
459 | ||
a51dca9c | 460 | J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <= |
470decc6 DK |
461 | journal->j_max_transaction_buffers); |
462 | ||
463 | /* | |
464 | * First thing we are allowed to do is to discard any remaining | |
465 | * BJ_Reserved buffers. Note, it is _not_ permissible to assume | |
466 | * that there are no such buffers: if a large filesystem | |
467 | * operation like a truncate needs to split itself over multiple | |
f7f4bccb | 468 | * transactions, then it may try to do a jbd2_journal_restart() while |
470decc6 DK |
469 | * there are still BJ_Reserved buffers outstanding. These must |
470 | * be released cleanly from the current transaction. | |
471 | * | |
472 | * In this case, the filesystem must still reserve write access | |
473 | * again before modifying the buffer in the new transaction, but | |
474 | * we do not require it to remember exactly which old buffers it | |
475 | * has reserved. This is consistent with the existing behaviour | |
f7f4bccb | 476 | * that multiple jbd2_journal_get_write_access() calls to the same |
25985edc | 477 | * buffer are perfectly permissible. |
470decc6 DK |
478 | */ |
479 | while (commit_transaction->t_reserved_list) { | |
480 | jh = commit_transaction->t_reserved_list; | |
481 | JBUFFER_TRACE(jh, "reserved, unused: refile"); | |
482 | /* | |
f7f4bccb | 483 | * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may |
470decc6 DK |
484 | * leave undo-committed data. |
485 | */ | |
486 | if (jh->b_committed_data) { | |
487 | struct buffer_head *bh = jh2bh(jh); | |
488 | ||
489 | jbd_lock_bh_state(bh); | |
af1e76d6 | 490 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
491 | jh->b_committed_data = NULL; |
492 | jbd_unlock_bh_state(bh); | |
493 | } | |
f7f4bccb | 494 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
495 | } |
496 | ||
497 | /* | |
498 | * Now try to drop any written-back buffers from the journal's | |
499 | * checkpoint lists. We do this *before* commit because it potentially | |
500 | * frees some memory | |
501 | */ | |
502 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 503 | __jbd2_journal_clean_checkpoint_list(journal); |
470decc6 DK |
504 | spin_unlock(&journal->j_list_lock); |
505 | ||
f2a44523 | 506 | jbd_debug(3, "JBD2: commit phase 1\n"); |
470decc6 | 507 | |
1ba37268 YY |
508 | /* |
509 | * Clear revoked flag to reflect there is no revoked buffers | |
510 | * in the next transaction which is going to be started. | |
511 | */ | |
512 | jbd2_clear_buffer_revoked_flags(journal); | |
513 | ||
470decc6 DK |
514 | /* |
515 | * Switch to a new revoke table. | |
516 | */ | |
f7f4bccb | 517 | jbd2_journal_switch_revoke_table(journal); |
470decc6 | 518 | |
879c5e6b | 519 | trace_jbd2_commit_flushing(journal, commit_transaction); |
bf699327 TT |
520 | stats.run.rs_flushing = jiffies; |
521 | stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked, | |
522 | stats.run.rs_flushing); | |
8e85fb3f | 523 | |
470decc6 DK |
524 | commit_transaction->t_state = T_FLUSH; |
525 | journal->j_committing_transaction = commit_transaction; | |
526 | journal->j_running_transaction = NULL; | |
e07f7183 | 527 | start_time = ktime_get(); |
470decc6 DK |
528 | commit_transaction->t_log_start = journal->j_head; |
529 | wake_up(&journal->j_wait_transaction_locked); | |
a931da6a | 530 | write_unlock(&journal->j_state_lock); |
470decc6 | 531 | |
f2a44523 | 532 | jbd_debug(3, "JBD2: commit phase 2\n"); |
470decc6 | 533 | |
470decc6 DK |
534 | /* |
535 | * Now start flushing things to disk, in the order they appear | |
536 | * on the transaction lists. Data blocks go first. | |
537 | */ | |
cd1aac32 | 538 | err = journal_submit_data_buffers(journal, commit_transaction); |
470decc6 | 539 | if (err) |
a7fa2baf | 540 | jbd2_journal_abort(journal, err); |
470decc6 | 541 | |
82f04ab4 | 542 | blk_start_plug(&plug); |
67c457a8 | 543 | jbd2_journal_write_revoke_records(journal, commit_transaction, |
e5a120ae | 544 | &log_bufs, WRITE_SYNC); |
82f04ab4 | 545 | blk_finish_plug(&plug); |
470decc6 | 546 | |
f2a44523 | 547 | jbd_debug(3, "JBD2: commit phase 2\n"); |
470decc6 | 548 | |
470decc6 DK |
549 | /* |
550 | * Way to go: we have now written out all of the data for a | |
551 | * transaction! Now comes the tricky part: we need to write out | |
552 | * metadata. Loop over the transaction's entire buffer list: | |
553 | */ | |
a931da6a | 554 | write_lock(&journal->j_state_lock); |
470decc6 | 555 | commit_transaction->t_state = T_COMMIT; |
a931da6a | 556 | write_unlock(&journal->j_state_lock); |
470decc6 | 557 | |
879c5e6b | 558 | trace_jbd2_commit_logging(journal, commit_transaction); |
bf699327 TT |
559 | stats.run.rs_logging = jiffies; |
560 | stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing, | |
561 | stats.run.rs_logging); | |
a51dca9c TT |
562 | stats.run.rs_blocks = |
563 | atomic_read(&commit_transaction->t_outstanding_credits); | |
bf699327 | 564 | stats.run.rs_blocks_logged = 0; |
8e85fb3f | 565 | |
1dfc3220 | 566 | J_ASSERT(commit_transaction->t_nr_buffers <= |
a51dca9c | 567 | atomic_read(&commit_transaction->t_outstanding_credits)); |
1dfc3220 | 568 | |
87c89c23 | 569 | err = 0; |
470decc6 | 570 | bufs = 0; |
e5a120ae | 571 | descriptor = NULL; |
82f04ab4 | 572 | blk_start_plug(&plug); |
470decc6 DK |
573 | while (commit_transaction->t_buffers) { |
574 | ||
575 | /* Find the next buffer to be journaled... */ | |
576 | ||
577 | jh = commit_transaction->t_buffers; | |
578 | ||
579 | /* If we're in abort mode, we just un-journal the buffer and | |
7ad7445f | 580 | release it. */ |
470decc6 DK |
581 | |
582 | if (is_journal_aborted(journal)) { | |
7ad7445f | 583 | clear_buffer_jbddirty(jh2bh(jh)); |
470decc6 | 584 | JBUFFER_TRACE(jh, "journal is aborting: refile"); |
e06c8227 JB |
585 | jbd2_buffer_abort_trigger(jh, |
586 | jh->b_frozen_data ? | |
587 | jh->b_frozen_triggers : | |
588 | jh->b_triggers); | |
f7f4bccb | 589 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
590 | /* If that was the last one, we need to clean up |
591 | * any descriptor buffers which may have been | |
592 | * already allocated, even if we are now | |
593 | * aborting. */ | |
594 | if (!commit_transaction->t_buffers) | |
595 | goto start_journal_io; | |
596 | continue; | |
597 | } | |
598 | ||
599 | /* Make sure we have a descriptor block in which to | |
600 | record the metadata buffer. */ | |
601 | ||
602 | if (!descriptor) { | |
470decc6 DK |
603 | J_ASSERT (bufs == 0); |
604 | ||
f2a44523 | 605 | jbd_debug(4, "JBD2: get descriptor\n"); |
470decc6 | 606 | |
f7f4bccb | 607 | descriptor = jbd2_journal_get_descriptor_buffer(journal); |
470decc6 | 608 | if (!descriptor) { |
a7fa2baf | 609 | jbd2_journal_abort(journal, -EIO); |
470decc6 DK |
610 | continue; |
611 | } | |
612 | ||
f2a44523 | 613 | jbd_debug(4, "JBD2: got buffer %llu (%p)\n", |
e5a120ae JK |
614 | (unsigned long long)descriptor->b_blocknr, |
615 | descriptor->b_data); | |
616 | header = (journal_header_t *)descriptor->b_data; | |
f7f4bccb MC |
617 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); |
618 | header->h_blocktype = cpu_to_be32(JBD2_DESCRIPTOR_BLOCK); | |
470decc6 DK |
619 | header->h_sequence = cpu_to_be32(commit_transaction->t_tid); |
620 | ||
e5a120ae JK |
621 | tagp = &descriptor->b_data[sizeof(journal_header_t)]; |
622 | space_left = descriptor->b_size - | |
623 | sizeof(journal_header_t); | |
470decc6 | 624 | first_tag = 1; |
e5a120ae JK |
625 | set_buffer_jwrite(descriptor); |
626 | set_buffer_dirty(descriptor); | |
627 | wbuf[bufs++] = descriptor; | |
470decc6 DK |
628 | |
629 | /* Record it so that we can wait for IO | |
630 | completion later */ | |
e5a120ae JK |
631 | BUFFER_TRACE(descriptor, "ph3: file as descriptor"); |
632 | jbd2_file_log_bh(&log_bufs, descriptor); | |
470decc6 DK |
633 | } |
634 | ||
635 | /* Where is the buffer to be written? */ | |
636 | ||
f7f4bccb | 637 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
638 | /* If the block mapping failed, just abandon the buffer |
639 | and repeat this loop: we'll fall into the | |
640 | refile-on-abort condition above. */ | |
641 | if (err) { | |
a7fa2baf | 642 | jbd2_journal_abort(journal, err); |
470decc6 DK |
643 | continue; |
644 | } | |
645 | ||
646 | /* | |
647 | * start_this_handle() uses t_outstanding_credits to determine | |
648 | * the free space in the log, but this counter is changed | |
f7f4bccb | 649 | * by jbd2_journal_next_log_block() also. |
470decc6 | 650 | */ |
a51dca9c | 651 | atomic_dec(&commit_transaction->t_outstanding_credits); |
470decc6 DK |
652 | |
653 | /* Bump b_count to prevent truncate from stumbling over | |
654 | the shadowed buffer! @@@ This can go if we ever get | |
f5113eff | 655 | rid of the shadow pairing of buffers. */ |
470decc6 DK |
656 | atomic_inc(&jh2bh(jh)->b_count); |
657 | ||
470decc6 | 658 | /* |
f5113eff JK |
659 | * Make a temporary IO buffer with which to write it out |
660 | * (this will requeue the metadata buffer to BJ_Shadow). | |
470decc6 | 661 | */ |
f5113eff | 662 | set_bit(BH_JWrite, &jh2bh(jh)->b_state); |
470decc6 | 663 | JBUFFER_TRACE(jh, "ph3: write metadata"); |
f7f4bccb | 664 | flags = jbd2_journal_write_metadata_buffer(commit_transaction, |
f5113eff | 665 | jh, &wbuf[bufs], blocknr); |
e6ec116b TT |
666 | if (flags < 0) { |
667 | jbd2_journal_abort(journal, flags); | |
668 | continue; | |
669 | } | |
f5113eff | 670 | jbd2_file_log_bh(&io_bufs, wbuf[bufs]); |
470decc6 DK |
671 | |
672 | /* Record the new block's tag in the current descriptor | |
673 | buffer */ | |
674 | ||
675 | tag_flag = 0; | |
676 | if (flags & 1) | |
f7f4bccb | 677 | tag_flag |= JBD2_FLAG_ESCAPE; |
470decc6 | 678 | if (!first_tag) |
f7f4bccb | 679 | tag_flag |= JBD2_FLAG_SAME_UUID; |
470decc6 DK |
680 | |
681 | tag = (journal_block_tag_t *) tagp; | |
b517bea1 | 682 | write_tag_block(tag_bytes, tag, jh2bh(jh)->b_blocknr); |
8f888ef8 | 683 | tag->t_flags = cpu_to_be16(tag_flag); |
f5113eff | 684 | jbd2_block_tag_csum_set(journal, tag, wbuf[bufs], |
c3900875 | 685 | commit_transaction->t_tid); |
b517bea1 ZB |
686 | tagp += tag_bytes; |
687 | space_left -= tag_bytes; | |
f5113eff | 688 | bufs++; |
470decc6 DK |
689 | |
690 | if (first_tag) { | |
691 | memcpy (tagp, journal->j_uuid, 16); | |
692 | tagp += 16; | |
693 | space_left -= 16; | |
694 | first_tag = 0; | |
695 | } | |
696 | ||
697 | /* If there's no more to do, or if the descriptor is full, | |
698 | let the IO rip! */ | |
699 | ||
700 | if (bufs == journal->j_wbufsize || | |
701 | commit_transaction->t_buffers == NULL || | |
3caa487f | 702 | space_left < tag_bytes + 16 + csum_size) { |
470decc6 | 703 | |
f2a44523 | 704 | jbd_debug(4, "JBD2: Submit %d IOs\n", bufs); |
470decc6 DK |
705 | |
706 | /* Write an end-of-descriptor marker before | |
707 | submitting the IOs. "tag" still points to | |
708 | the last tag we set up. */ | |
709 | ||
8f888ef8 | 710 | tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG); |
470decc6 | 711 | |
3caa487f | 712 | jbd2_descr_block_csum_set(journal, descriptor); |
470decc6 DK |
713 | start_journal_io: |
714 | for (i = 0; i < bufs; i++) { | |
715 | struct buffer_head *bh = wbuf[i]; | |
818d276c GS |
716 | /* |
717 | * Compute checksum. | |
718 | */ | |
719 | if (JBD2_HAS_COMPAT_FEATURE(journal, | |
720 | JBD2_FEATURE_COMPAT_CHECKSUM)) { | |
721 | crc32_sum = | |
722 | jbd2_checksum_data(crc32_sum, bh); | |
723 | } | |
724 | ||
470decc6 DK |
725 | lock_buffer(bh); |
726 | clear_buffer_dirty(bh); | |
727 | set_buffer_uptodate(bh); | |
728 | bh->b_end_io = journal_end_buffer_io_sync; | |
82f04ab4 | 729 | submit_bh(WRITE_SYNC, bh); |
470decc6 DK |
730 | } |
731 | cond_resched(); | |
bf699327 | 732 | stats.run.rs_blocks_logged += bufs; |
470decc6 DK |
733 | |
734 | /* Force a new descriptor to be generated next | |
735 | time round the loop. */ | |
736 | descriptor = NULL; | |
737 | bufs = 0; | |
738 | } | |
739 | } | |
740 | ||
f73bee49 JK |
741 | err = journal_finish_inode_data_buffers(journal, commit_transaction); |
742 | if (err) { | |
743 | printk(KERN_WARNING | |
744 | "JBD2: Detected IO errors while flushing file data " | |
745 | "on %s\n", journal->j_devname); | |
746 | if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR) | |
747 | jbd2_journal_abort(journal, err); | |
748 | err = 0; | |
749 | } | |
750 | ||
3339578f JK |
751 | /* |
752 | * Get current oldest transaction in the log before we issue flush | |
753 | * to the filesystem device. After the flush we can be sure that | |
754 | * blocks of all older transactions are checkpointed to persistent | |
755 | * storage and we will be safe to update journal start in the | |
756 | * superblock with the numbers we get here. | |
757 | */ | |
758 | update_tail = | |
759 | jbd2_journal_get_log_tail(journal, &first_tid, &first_block); | |
760 | ||
bbd2be36 | 761 | write_lock(&journal->j_state_lock); |
3339578f JK |
762 | if (update_tail) { |
763 | long freed = first_block - journal->j_tail; | |
764 | ||
765 | if (first_block < journal->j_tail) | |
766 | freed += journal->j_last - journal->j_first; | |
767 | /* Update tail only if we free significant amount of space */ | |
768 | if (freed < journal->j_maxlen / 4) | |
769 | update_tail = 0; | |
770 | } | |
bbd2be36 JK |
771 | J_ASSERT(commit_transaction->t_state == T_COMMIT); |
772 | commit_transaction->t_state = T_COMMIT_DFLUSH; | |
773 | write_unlock(&journal->j_state_lock); | |
3339578f | 774 | |
cc3e1bea TT |
775 | /* |
776 | * If the journal is not located on the file system device, | |
777 | * then we must flush the file system device before we issue | |
778 | * the commit record | |
779 | */ | |
81be12c8 | 780 | if (commit_transaction->t_need_data_flush && |
cc3e1bea TT |
781 | (journal->j_fs_dev != journal->j_dev) && |
782 | (journal->j_flags & JBD2_BARRIER)) | |
99aa7846 | 783 | blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL); |
818d276c | 784 | |
cc3e1bea | 785 | /* Done it all: now write the commit record asynchronously. */ |
818d276c | 786 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, |
0e3d2a63 | 787 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { |
818d276c GS |
788 | err = journal_submit_commit_record(journal, commit_transaction, |
789 | &cbh, crc32_sum); | |
790 | if (err) | |
791 | __jbd2_journal_abort_hard(journal); | |
e9e34f4e | 792 | } |
c851ed54 | 793 | |
82f04ab4 JA |
794 | blk_finish_plug(&plug); |
795 | ||
470decc6 DK |
796 | /* Lo and behold: we have just managed to send a transaction to |
797 | the log. Before we can commit it, wait for the IO so far to | |
798 | complete. Control buffers being written are on the | |
799 | transaction's t_log_list queue, and metadata buffers are on | |
f5113eff | 800 | the io_bufs list. |
470decc6 DK |
801 | |
802 | Wait for the buffers in reverse order. That way we are | |
803 | less likely to be woken up until all IOs have completed, and | |
804 | so we incur less scheduling load. | |
805 | */ | |
806 | ||
f2a44523 | 807 | jbd_debug(3, "JBD2: commit phase 3\n"); |
470decc6 | 808 | |
f5113eff JK |
809 | while (!list_empty(&io_bufs)) { |
810 | struct buffer_head *bh = list_entry(io_bufs.prev, | |
811 | struct buffer_head, | |
812 | b_assoc_buffers); | |
470decc6 | 813 | |
f5113eff JK |
814 | wait_on_buffer(bh); |
815 | cond_resched(); | |
470decc6 DK |
816 | |
817 | if (unlikely(!buffer_uptodate(bh))) | |
818 | err = -EIO; | |
f5113eff | 819 | jbd2_unfile_log_bh(bh); |
470decc6 DK |
820 | |
821 | /* | |
f5113eff JK |
822 | * The list contains temporary buffer heads created by |
823 | * jbd2_journal_write_metadata_buffer(). | |
470decc6 DK |
824 | */ |
825 | BUFFER_TRACE(bh, "dumping temporary bh"); | |
470decc6 DK |
826 | __brelse(bh); |
827 | J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0); | |
828 | free_buffer_head(bh); | |
829 | ||
f5113eff | 830 | /* We also have to refile the corresponding shadowed buffer */ |
470decc6 DK |
831 | jh = commit_transaction->t_shadow_list->b_tprev; |
832 | bh = jh2bh(jh); | |
f5113eff | 833 | clear_buffer_jwrite(bh); |
470decc6 DK |
834 | J_ASSERT_BH(bh, buffer_jbddirty(bh)); |
835 | ||
836 | /* The metadata is now released for reuse, but we need | |
837 | to remember it against this transaction so that when | |
838 | we finally commit, we can do any checkpointing | |
839 | required. */ | |
840 | JBUFFER_TRACE(jh, "file as BJ_Forget"); | |
f7f4bccb | 841 | jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget); |
229309ca JK |
842 | /* |
843 | * Wake up any transactions which were waiting for this IO to | |
844 | * complete. The barrier must be here so that changes by | |
845 | * jbd2_journal_file_buffer() take effect before wake_up_bit() | |
846 | * does the waitqueue check. | |
847 | */ | |
848 | smp_mb(); | |
470decc6 DK |
849 | wake_up_bit(&bh->b_state, BH_Unshadow); |
850 | JBUFFER_TRACE(jh, "brelse shadowed buffer"); | |
851 | __brelse(bh); | |
852 | } | |
853 | ||
854 | J_ASSERT (commit_transaction->t_shadow_list == NULL); | |
855 | ||
f2a44523 | 856 | jbd_debug(3, "JBD2: commit phase 4\n"); |
470decc6 DK |
857 | |
858 | /* Here we wait for the revoke record and descriptor record buffers */ | |
e5a120ae | 859 | while (!list_empty(&log_bufs)) { |
470decc6 DK |
860 | struct buffer_head *bh; |
861 | ||
e5a120ae JK |
862 | bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers); |
863 | wait_on_buffer(bh); | |
864 | cond_resched(); | |
470decc6 DK |
865 | |
866 | if (unlikely(!buffer_uptodate(bh))) | |
867 | err = -EIO; | |
868 | ||
869 | BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile"); | |
870 | clear_buffer_jwrite(bh); | |
e5a120ae | 871 | jbd2_unfile_log_bh(bh); |
470decc6 DK |
872 | __brelse(bh); /* One for getblk */ |
873 | /* AKPM: bforget here */ | |
874 | } | |
875 | ||
77e841de HK |
876 | if (err) |
877 | jbd2_journal_abort(journal, err); | |
878 | ||
f2a44523 | 879 | jbd_debug(3, "JBD2: commit phase 5\n"); |
bbd2be36 JK |
880 | write_lock(&journal->j_state_lock); |
881 | J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH); | |
882 | commit_transaction->t_state = T_COMMIT_JFLUSH; | |
883 | write_unlock(&journal->j_state_lock); | |
470decc6 | 884 | |
818d276c | 885 | if (!JBD2_HAS_INCOMPAT_FEATURE(journal, |
0e3d2a63 | 886 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { |
818d276c GS |
887 | err = journal_submit_commit_record(journal, commit_transaction, |
888 | &cbh, crc32_sum); | |
889 | if (err) | |
890 | __jbd2_journal_abort_hard(journal); | |
891 | } | |
6cba611e | 892 | if (cbh) |
fd98496f | 893 | err = journal_wait_on_commit_record(journal, cbh); |
f73bee49 JK |
894 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, |
895 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) && | |
896 | journal->j_flags & JBD2_BARRIER) { | |
99aa7846 | 897 | blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL); |
f73bee49 | 898 | } |
470decc6 DK |
899 | |
900 | if (err) | |
a7fa2baf | 901 | jbd2_journal_abort(journal, err); |
470decc6 | 902 | |
3339578f JK |
903 | /* |
904 | * Now disk caches for filesystem device are flushed so we are safe to | |
905 | * erase checkpointed transactions from the log by updating journal | |
906 | * superblock. | |
907 | */ | |
908 | if (update_tail) | |
909 | jbd2_update_log_tail(journal, first_tid, first_block); | |
910 | ||
470decc6 DK |
911 | /* End of a transaction! Finally, we can do checkpoint |
912 | processing: any buffers committed as a result of this | |
913 | transaction can be removed from any checkpoint list it was on | |
914 | before. */ | |
915 | ||
f2a44523 | 916 | jbd_debug(3, "JBD2: commit phase 6\n"); |
470decc6 | 917 | |
c851ed54 | 918 | J_ASSERT(list_empty(&commit_transaction->t_inode_list)); |
470decc6 DK |
919 | J_ASSERT(commit_transaction->t_buffers == NULL); |
920 | J_ASSERT(commit_transaction->t_checkpoint_list == NULL); | |
470decc6 | 921 | J_ASSERT(commit_transaction->t_shadow_list == NULL); |
470decc6 DK |
922 | |
923 | restart_loop: | |
924 | /* | |
925 | * As there are other places (journal_unmap_buffer()) adding buffers | |
926 | * to this list we have to be careful and hold the j_list_lock. | |
927 | */ | |
928 | spin_lock(&journal->j_list_lock); | |
929 | while (commit_transaction->t_forget) { | |
930 | transaction_t *cp_transaction; | |
931 | struct buffer_head *bh; | |
de1b7941 | 932 | int try_to_free = 0; |
470decc6 DK |
933 | |
934 | jh = commit_transaction->t_forget; | |
935 | spin_unlock(&journal->j_list_lock); | |
936 | bh = jh2bh(jh); | |
de1b7941 JK |
937 | /* |
938 | * Get a reference so that bh cannot be freed before we are | |
939 | * done with it. | |
940 | */ | |
941 | get_bh(bh); | |
470decc6 | 942 | jbd_lock_bh_state(bh); |
23e2af35 | 943 | J_ASSERT_JH(jh, jh->b_transaction == commit_transaction); |
470decc6 DK |
944 | |
945 | /* | |
946 | * If there is undo-protected committed data against | |
947 | * this buffer, then we can remove it now. If it is a | |
948 | * buffer needing such protection, the old frozen_data | |
949 | * field now points to a committed version of the | |
950 | * buffer, so rotate that field to the new committed | |
951 | * data. | |
952 | * | |
953 | * Otherwise, we can just throw away the frozen data now. | |
e06c8227 JB |
954 | * |
955 | * We also know that the frozen data has already fired | |
956 | * its triggers if they exist, so we can clear that too. | |
470decc6 DK |
957 | */ |
958 | if (jh->b_committed_data) { | |
af1e76d6 | 959 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
960 | jh->b_committed_data = NULL; |
961 | if (jh->b_frozen_data) { | |
962 | jh->b_committed_data = jh->b_frozen_data; | |
963 | jh->b_frozen_data = NULL; | |
e06c8227 | 964 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
965 | } |
966 | } else if (jh->b_frozen_data) { | |
af1e76d6 | 967 | jbd2_free(jh->b_frozen_data, bh->b_size); |
470decc6 | 968 | jh->b_frozen_data = NULL; |
e06c8227 | 969 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
970 | } |
971 | ||
972 | spin_lock(&journal->j_list_lock); | |
973 | cp_transaction = jh->b_cp_transaction; | |
974 | if (cp_transaction) { | |
975 | JBUFFER_TRACE(jh, "remove from old cp transaction"); | |
8e85fb3f | 976 | cp_transaction->t_chp_stats.cs_dropped++; |
f7f4bccb | 977 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
978 | } |
979 | ||
980 | /* Only re-checkpoint the buffer_head if it is marked | |
981 | * dirty. If the buffer was added to the BJ_Forget list | |
f7f4bccb | 982 | * by jbd2_journal_forget, it may no longer be dirty and |
470decc6 DK |
983 | * there's no point in keeping a checkpoint record for |
984 | * it. */ | |
985 | ||
b794e7a6 JK |
986 | /* |
987 | * A buffer which has been freed while still being journaled by | |
988 | * a previous transaction. | |
989 | */ | |
990 | if (buffer_freed(bh)) { | |
991 | /* | |
992 | * If the running transaction is the one containing | |
993 | * "add to orphan" operation (b_next_transaction != | |
994 | * NULL), we have to wait for that transaction to | |
995 | * commit before we can really get rid of the buffer. | |
996 | * So just clear b_modified to not confuse transaction | |
997 | * credit accounting and refile the buffer to | |
998 | * BJ_Forget of the running transaction. If the just | |
999 | * committed transaction contains "add to orphan" | |
1000 | * operation, we can completely invalidate the buffer | |
1001 | * now. We are rather through in that since the | |
1002 | * buffer may be still accessible when blocksize < | |
1003 | * pagesize and it is attached to the last partial | |
1004 | * page. | |
1005 | */ | |
1006 | jh->b_modified = 0; | |
1007 | if (!jh->b_next_transaction) { | |
1008 | clear_buffer_freed(bh); | |
1009 | clear_buffer_jbddirty(bh); | |
1010 | clear_buffer_mapped(bh); | |
1011 | clear_buffer_new(bh); | |
1012 | clear_buffer_req(bh); | |
1013 | bh->b_bdev = NULL; | |
1014 | } | |
470decc6 DK |
1015 | } |
1016 | ||
1017 | if (buffer_jbddirty(bh)) { | |
1018 | JBUFFER_TRACE(jh, "add to new checkpointing trans"); | |
f7f4bccb | 1019 | __jbd2_journal_insert_checkpoint(jh, commit_transaction); |
7ad7445f HK |
1020 | if (is_journal_aborted(journal)) |
1021 | clear_buffer_jbddirty(bh); | |
470decc6 DK |
1022 | } else { |
1023 | J_ASSERT_BH(bh, !buffer_dirty(bh)); | |
de1b7941 JK |
1024 | /* |
1025 | * The buffer on BJ_Forget list and not jbddirty means | |
470decc6 DK |
1026 | * it has been freed by this transaction and hence it |
1027 | * could not have been reallocated until this | |
1028 | * transaction has committed. *BUT* it could be | |
1029 | * reallocated once we have written all the data to | |
1030 | * disk and before we process the buffer on BJ_Forget | |
de1b7941 JK |
1031 | * list. |
1032 | */ | |
1033 | if (!jh->b_next_transaction) | |
1034 | try_to_free = 1; | |
470decc6 | 1035 | } |
de1b7941 JK |
1036 | JBUFFER_TRACE(jh, "refile or unfile buffer"); |
1037 | __jbd2_journal_refile_buffer(jh); | |
1038 | jbd_unlock_bh_state(bh); | |
1039 | if (try_to_free) | |
1040 | release_buffer_page(bh); /* Drops bh reference */ | |
1041 | else | |
1042 | __brelse(bh); | |
470decc6 DK |
1043 | cond_resched_lock(&journal->j_list_lock); |
1044 | } | |
1045 | spin_unlock(&journal->j_list_lock); | |
1046 | /* | |
f5a7a6b0 JK |
1047 | * This is a bit sleazy. We use j_list_lock to protect transition |
1048 | * of a transaction into T_FINISHED state and calling | |
1049 | * __jbd2_journal_drop_transaction(). Otherwise we could race with | |
1050 | * other checkpointing code processing the transaction... | |
470decc6 | 1051 | */ |
a931da6a | 1052 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1053 | spin_lock(&journal->j_list_lock); |
1054 | /* | |
1055 | * Now recheck if some buffers did not get attached to the transaction | |
1056 | * while the lock was dropped... | |
1057 | */ | |
1058 | if (commit_transaction->t_forget) { | |
1059 | spin_unlock(&journal->j_list_lock); | |
a931da6a | 1060 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1061 | goto restart_loop; |
1062 | } | |
1063 | ||
1064 | /* Done with this transaction! */ | |
1065 | ||
f2a44523 | 1066 | jbd_debug(3, "JBD2: commit phase 7\n"); |
470decc6 | 1067 | |
bbd2be36 | 1068 | J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH); |
470decc6 | 1069 | |
8e85fb3f | 1070 | commit_transaction->t_start = jiffies; |
bf699327 TT |
1071 | stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging, |
1072 | commit_transaction->t_start); | |
8e85fb3f JL |
1073 | |
1074 | /* | |
bf699327 | 1075 | * File the transaction statistics |
8e85fb3f | 1076 | */ |
8e85fb3f | 1077 | stats.ts_tid = commit_transaction->t_tid; |
8dd42046 TT |
1078 | stats.run.rs_handle_count = |
1079 | atomic_read(&commit_transaction->t_handle_count); | |
bf699327 TT |
1080 | trace_jbd2_run_stats(journal->j_fs_dev->bd_dev, |
1081 | commit_transaction->t_tid, &stats.run); | |
8e85fb3f JL |
1082 | |
1083 | /* | |
1084 | * Calculate overall stats | |
1085 | */ | |
bf699327 | 1086 | spin_lock(&journal->j_history_lock); |
8e85fb3f | 1087 | journal->j_stats.ts_tid++; |
9fff24aa TT |
1088 | if (commit_transaction->t_requested) |
1089 | journal->j_stats.ts_requested++; | |
bf699327 | 1090 | journal->j_stats.run.rs_wait += stats.run.rs_wait; |
9fff24aa | 1091 | journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay; |
bf699327 TT |
1092 | journal->j_stats.run.rs_running += stats.run.rs_running; |
1093 | journal->j_stats.run.rs_locked += stats.run.rs_locked; | |
1094 | journal->j_stats.run.rs_flushing += stats.run.rs_flushing; | |
1095 | journal->j_stats.run.rs_logging += stats.run.rs_logging; | |
1096 | journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count; | |
1097 | journal->j_stats.run.rs_blocks += stats.run.rs_blocks; | |
1098 | journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged; | |
8e85fb3f JL |
1099 | spin_unlock(&journal->j_history_lock); |
1100 | ||
794446c6 | 1101 | commit_transaction->t_state = T_COMMIT_CALLBACK; |
470decc6 DK |
1102 | J_ASSERT(commit_transaction == journal->j_committing_transaction); |
1103 | journal->j_commit_sequence = commit_transaction->t_tid; | |
1104 | journal->j_committing_transaction = NULL; | |
e07f7183 | 1105 | commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); |
470decc6 | 1106 | |
e07f7183 JB |
1107 | /* |
1108 | * weight the commit time higher than the average time so we don't | |
1109 | * react too strongly to vast changes in the commit time | |
1110 | */ | |
1111 | if (likely(journal->j_average_commit_time)) | |
1112 | journal->j_average_commit_time = (commit_time + | |
1113 | journal->j_average_commit_time*3) / 4; | |
1114 | else | |
1115 | journal->j_average_commit_time = commit_time; | |
794446c6 | 1116 | |
a931da6a | 1117 | write_unlock(&journal->j_state_lock); |
6c20ec85 | 1118 | |
794446c6 DM |
1119 | if (journal->j_checkpoint_transactions == NULL) { |
1120 | journal->j_checkpoint_transactions = commit_transaction; | |
1121 | commit_transaction->t_cpnext = commit_transaction; | |
1122 | commit_transaction->t_cpprev = commit_transaction; | |
470decc6 | 1123 | } else { |
794446c6 DM |
1124 | commit_transaction->t_cpnext = |
1125 | journal->j_checkpoint_transactions; | |
1126 | commit_transaction->t_cpprev = | |
1127 | commit_transaction->t_cpnext->t_cpprev; | |
1128 | commit_transaction->t_cpnext->t_cpprev = | |
1129 | commit_transaction; | |
1130 | commit_transaction->t_cpprev->t_cpnext = | |
470decc6 | 1131 | commit_transaction; |
470decc6 DK |
1132 | } |
1133 | spin_unlock(&journal->j_list_lock); | |
794446c6 DM |
1134 | /* Drop all spin_locks because commit_callback may be block. |
1135 | * __journal_remove_checkpoint() can not destroy transaction | |
1136 | * under us because it is not marked as T_FINISHED yet */ | |
fb68407b AK |
1137 | if (journal->j_commit_callback) |
1138 | journal->j_commit_callback(journal, commit_transaction); | |
1139 | ||
879c5e6b | 1140 | trace_jbd2_end_commit(journal, commit_transaction); |
f2a44523 | 1141 | jbd_debug(1, "JBD2: commit %d complete, head %d\n", |
470decc6 DK |
1142 | journal->j_commit_sequence, journal->j_tail_sequence); |
1143 | ||
794446c6 DM |
1144 | write_lock(&journal->j_state_lock); |
1145 | spin_lock(&journal->j_list_lock); | |
1146 | commit_transaction->t_state = T_FINISHED; | |
1147 | /* Recheck checkpoint lists after j_list_lock was dropped */ | |
1148 | if (commit_transaction->t_checkpoint_list == NULL && | |
1149 | commit_transaction->t_checkpoint_io_list == NULL) { | |
1150 | __jbd2_journal_drop_transaction(journal, commit_transaction); | |
1151 | jbd2_journal_free_transaction(commit_transaction); | |
1152 | } | |
1153 | spin_unlock(&journal->j_list_lock); | |
1154 | write_unlock(&journal->j_state_lock); | |
470decc6 DK |
1155 | wake_up(&journal->j_wait_done_commit); |
1156 | } |