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> |
39e3ac25 | 31 | #include <asm/system.h> |
470decc6 DK |
32 | |
33 | /* | |
34 | * Default IO end handler for temporary BJ_IO buffer_heads. | |
35 | */ | |
36 | static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate) | |
37 | { | |
38 | BUFFER_TRACE(bh, ""); | |
39 | if (uptodate) | |
40 | set_buffer_uptodate(bh); | |
41 | else | |
42 | clear_buffer_uptodate(bh); | |
43 | unlock_buffer(bh); | |
44 | } | |
45 | ||
46 | /* | |
87c89c23 JK |
47 | * When an ext4 file is truncated, it is possible that some pages are not |
48 | * successfully freed, because they are attached to a committing transaction. | |
470decc6 DK |
49 | * After the transaction commits, these pages are left on the LRU, with no |
50 | * ->mapping, and with attached buffers. These pages are trivially reclaimable | |
51 | * by the VM, but their apparent absence upsets the VM accounting, and it makes | |
52 | * the numbers in /proc/meminfo look odd. | |
53 | * | |
54 | * So here, we have a buffer which has just come off the forget list. Look to | |
55 | * see if we can strip all buffers from the backing page. | |
56 | * | |
57 | * Called under lock_journal(), and possibly under journal_datalist_lock. The | |
58 | * caller provided us with a ref against the buffer, and we drop that here. | |
59 | */ | |
60 | static void release_buffer_page(struct buffer_head *bh) | |
61 | { | |
62 | struct page *page; | |
63 | ||
64 | if (buffer_dirty(bh)) | |
65 | goto nope; | |
66 | if (atomic_read(&bh->b_count) != 1) | |
67 | goto nope; | |
68 | page = bh->b_page; | |
69 | if (!page) | |
70 | goto nope; | |
71 | if (page->mapping) | |
72 | goto nope; | |
73 | ||
74 | /* OK, it's a truncated page */ | |
529ae9aa | 75 | if (!trylock_page(page)) |
470decc6 DK |
76 | goto nope; |
77 | ||
78 | page_cache_get(page); | |
79 | __brelse(bh); | |
80 | try_to_free_buffers(page); | |
81 | unlock_page(page); | |
82 | page_cache_release(page); | |
83 | return; | |
84 | ||
85 | nope: | |
86 | __brelse(bh); | |
87 | } | |
88 | ||
818d276c GS |
89 | /* |
90 | * Done it all: now submit the commit record. We should have | |
470decc6 DK |
91 | * cleaned up our previous buffers by now, so if we are in abort |
92 | * mode we can now just skip the rest of the journal write | |
93 | * entirely. | |
94 | * | |
95 | * Returns 1 if the journal needs to be aborted or 0 on success | |
96 | */ | |
818d276c GS |
97 | static int journal_submit_commit_record(journal_t *journal, |
98 | transaction_t *commit_transaction, | |
99 | struct buffer_head **cbh, | |
100 | __u32 crc32_sum) | |
470decc6 DK |
101 | { |
102 | struct journal_head *descriptor; | |
818d276c | 103 | struct commit_header *tmp; |
470decc6 | 104 | struct buffer_head *bh; |
818d276c | 105 | int ret; |
736603ab | 106 | struct timespec now = current_kernel_time(); |
470decc6 | 107 | |
6cba611e ZH |
108 | *cbh = NULL; |
109 | ||
470decc6 DK |
110 | if (is_journal_aborted(journal)) |
111 | return 0; | |
112 | ||
f7f4bccb | 113 | descriptor = jbd2_journal_get_descriptor_buffer(journal); |
470decc6 DK |
114 | if (!descriptor) |
115 | return 1; | |
116 | ||
117 | bh = jh2bh(descriptor); | |
118 | ||
818d276c GS |
119 | tmp = (struct commit_header *)bh->b_data; |
120 | tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); | |
121 | tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK); | |
122 | tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid); | |
736603ab TT |
123 | tmp->h_commit_sec = cpu_to_be64(now.tv_sec); |
124 | tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec); | |
818d276c GS |
125 | |
126 | if (JBD2_HAS_COMPAT_FEATURE(journal, | |
127 | JBD2_FEATURE_COMPAT_CHECKSUM)) { | |
128 | tmp->h_chksum_type = JBD2_CRC32_CHKSUM; | |
129 | tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE; | |
130 | tmp->h_chksum[0] = cpu_to_be32(crc32_sum); | |
470decc6 DK |
131 | } |
132 | ||
818d276c GS |
133 | JBUFFER_TRACE(descriptor, "submit commit block"); |
134 | lock_buffer(bh); | |
45a90bfd | 135 | clear_buffer_dirty(bh); |
818d276c GS |
136 | set_buffer_uptodate(bh); |
137 | bh->b_end_io = journal_end_buffer_io_sync; | |
138 | ||
139 | if (journal->j_flags & JBD2_BARRIER && | |
0e3d2a63 | 140 | !JBD2_HAS_INCOMPAT_FEATURE(journal, |
9c35575b | 141 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) |
721a9602 | 142 | ret = submit_bh(WRITE_SYNC | WRITE_FLUSH_FUA, bh); |
9c35575b | 143 | else |
721a9602 | 144 | ret = submit_bh(WRITE_SYNC, bh); |
9c35575b | 145 | |
818d276c GS |
146 | *cbh = bh; |
147 | return ret; | |
148 | } | |
149 | ||
150 | /* | |
151 | * This function along with journal_submit_commit_record | |
152 | * allows to write the commit record asynchronously. | |
153 | */ | |
fd98496f TT |
154 | static int journal_wait_on_commit_record(journal_t *journal, |
155 | struct buffer_head *bh) | |
818d276c GS |
156 | { |
157 | int ret = 0; | |
158 | ||
159 | clear_buffer_dirty(bh); | |
160 | wait_on_buffer(bh); | |
470decc6 | 161 | |
818d276c GS |
162 | if (unlikely(!buffer_uptodate(bh))) |
163 | ret = -EIO; | |
164 | put_bh(bh); /* One for getblk() */ | |
165 | jbd2_journal_put_journal_head(bh2jh(bh)); | |
166 | ||
167 | return ret; | |
470decc6 DK |
168 | } |
169 | ||
cd1aac32 AK |
170 | /* |
171 | * write the filemap data using writepage() address_space_operations. | |
172 | * We don't do block allocation here even for delalloc. We don't | |
173 | * use writepages() because with dealyed allocation we may be doing | |
174 | * block allocation in writepages(). | |
175 | */ | |
176 | static int journal_submit_inode_data_buffers(struct address_space *mapping) | |
177 | { | |
178 | int ret; | |
179 | struct writeback_control wbc = { | |
180 | .sync_mode = WB_SYNC_ALL, | |
181 | .nr_to_write = mapping->nrpages * 2, | |
182 | .range_start = 0, | |
183 | .range_end = i_size_read(mapping->host), | |
cd1aac32 AK |
184 | }; |
185 | ||
186 | ret = generic_writepages(mapping, &wbc); | |
187 | return ret; | |
188 | } | |
189 | ||
c851ed54 JK |
190 | /* |
191 | * Submit all the data buffers of inode associated with the transaction to | |
192 | * disk. | |
193 | * | |
194 | * We are in a committing transaction. Therefore no new inode can be added to | |
195 | * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently | |
196 | * operate on from being released while we write out pages. | |
197 | */ | |
cd1aac32 | 198 | static int journal_submit_data_buffers(journal_t *journal, |
c851ed54 JK |
199 | transaction_t *commit_transaction) |
200 | { | |
201 | struct jbd2_inode *jinode; | |
202 | int err, ret = 0; | |
203 | struct address_space *mapping; | |
204 | ||
205 | spin_lock(&journal->j_list_lock); | |
206 | list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) { | |
207 | mapping = jinode->i_vfs_inode->i_mapping; | |
39e3ac25 | 208 | set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
c851ed54 | 209 | spin_unlock(&journal->j_list_lock); |
cd1aac32 AK |
210 | /* |
211 | * submit the inode data buffers. We use writepage | |
212 | * instead of writepages. Because writepages can do | |
213 | * block allocation with delalloc. We need to write | |
214 | * only allocated blocks here. | |
215 | */ | |
879c5e6b | 216 | trace_jbd2_submit_inode_data(jinode->i_vfs_inode); |
cd1aac32 | 217 | err = journal_submit_inode_data_buffers(mapping); |
c851ed54 JK |
218 | if (!ret) |
219 | ret = err; | |
220 | spin_lock(&journal->j_list_lock); | |
221 | J_ASSERT(jinode->i_transaction == commit_transaction); | |
39e3ac25 BK |
222 | clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
223 | smp_mb__after_clear_bit(); | |
c851ed54 JK |
224 | wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING); |
225 | } | |
226 | spin_unlock(&journal->j_list_lock); | |
227 | return ret; | |
228 | } | |
229 | ||
230 | /* | |
231 | * Wait for data submitted for writeout, refile inodes to proper | |
232 | * transaction if needed. | |
233 | * | |
234 | */ | |
235 | static int journal_finish_inode_data_buffers(journal_t *journal, | |
236 | transaction_t *commit_transaction) | |
237 | { | |
238 | struct jbd2_inode *jinode, *next_i; | |
239 | int err, ret = 0; | |
240 | ||
cd1aac32 | 241 | /* For locking, see the comment in journal_submit_data_buffers() */ |
c851ed54 JK |
242 | spin_lock(&journal->j_list_lock); |
243 | list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) { | |
39e3ac25 | 244 | set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
c851ed54 JK |
245 | spin_unlock(&journal->j_list_lock); |
246 | err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping); | |
e9e34f4e HK |
247 | if (err) { |
248 | /* | |
249 | * Because AS_EIO is cleared by | |
94004ed7 | 250 | * filemap_fdatawait_range(), set it again so |
e9e34f4e HK |
251 | * that user process can get -EIO from fsync(). |
252 | */ | |
253 | set_bit(AS_EIO, | |
254 | &jinode->i_vfs_inode->i_mapping->flags); | |
255 | ||
256 | if (!ret) | |
257 | ret = err; | |
258 | } | |
c851ed54 | 259 | spin_lock(&journal->j_list_lock); |
39e3ac25 BK |
260 | clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
261 | smp_mb__after_clear_bit(); | |
c851ed54 JK |
262 | wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING); |
263 | } | |
264 | ||
265 | /* Now refile inode to proper lists */ | |
266 | list_for_each_entry_safe(jinode, next_i, | |
267 | &commit_transaction->t_inode_list, i_list) { | |
268 | list_del(&jinode->i_list); | |
269 | if (jinode->i_next_transaction) { | |
270 | jinode->i_transaction = jinode->i_next_transaction; | |
271 | jinode->i_next_transaction = NULL; | |
272 | list_add(&jinode->i_list, | |
273 | &jinode->i_transaction->t_inode_list); | |
274 | } else { | |
275 | jinode->i_transaction = NULL; | |
276 | } | |
277 | } | |
278 | spin_unlock(&journal->j_list_lock); | |
279 | ||
280 | return ret; | |
281 | } | |
282 | ||
818d276c GS |
283 | static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh) |
284 | { | |
285 | struct page *page = bh->b_page; | |
286 | char *addr; | |
287 | __u32 checksum; | |
288 | ||
289 | addr = kmap_atomic(page, KM_USER0); | |
290 | checksum = crc32_be(crc32_sum, | |
291 | (void *)(addr + offset_in_page(bh->b_data)), bh->b_size); | |
292 | kunmap_atomic(addr, KM_USER0); | |
293 | ||
294 | return checksum; | |
295 | } | |
296 | ||
297 | static void write_tag_block(int tag_bytes, journal_block_tag_t *tag, | |
18eba7aa | 298 | unsigned long long block) |
b517bea1 ZB |
299 | { |
300 | tag->t_blocknr = cpu_to_be32(block & (u32)~0); | |
cd02ff0b | 301 | if (tag_bytes > JBD2_TAG_SIZE32) |
b517bea1 ZB |
302 | tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1); |
303 | } | |
304 | ||
470decc6 | 305 | /* |
f7f4bccb | 306 | * jbd2_journal_commit_transaction |
470decc6 DK |
307 | * |
308 | * The primary function for committing a transaction to the log. This | |
309 | * function is called by the journal thread to begin a complete commit. | |
310 | */ | |
f7f4bccb | 311 | void jbd2_journal_commit_transaction(journal_t *journal) |
470decc6 | 312 | { |
8e85fb3f | 313 | struct transaction_stats_s stats; |
470decc6 DK |
314 | transaction_t *commit_transaction; |
315 | struct journal_head *jh, *new_jh, *descriptor; | |
316 | struct buffer_head **wbuf = journal->j_wbuf; | |
317 | int bufs; | |
318 | int flags; | |
319 | int err; | |
18eba7aa | 320 | unsigned long long blocknr; |
e07f7183 JB |
321 | ktime_t start_time; |
322 | u64 commit_time; | |
470decc6 DK |
323 | char *tagp = NULL; |
324 | journal_header_t *header; | |
325 | journal_block_tag_t *tag = NULL; | |
326 | int space_left = 0; | |
327 | int first_tag = 0; | |
328 | int tag_flag; | |
fb68407b | 329 | int i, to_free = 0; |
b517bea1 | 330 | int tag_bytes = journal_tag_bytes(journal); |
818d276c GS |
331 | struct buffer_head *cbh = NULL; /* For transactional checksums */ |
332 | __u32 crc32_sum = ~0; | |
82f04ab4 | 333 | struct blk_plug plug; |
470decc6 DK |
334 | |
335 | /* | |
336 | * First job: lock down the current transaction and wait for | |
337 | * all outstanding updates to complete. | |
338 | */ | |
339 | ||
f7f4bccb MC |
340 | /* Do we need to erase the effects of a prior jbd2_journal_flush? */ |
341 | if (journal->j_flags & JBD2_FLUSHED) { | |
470decc6 | 342 | jbd_debug(3, "super block updated\n"); |
a78bb11d | 343 | mutex_lock(&journal->j_checkpoint_mutex); |
79feb521 JK |
344 | /* |
345 | * We hold j_checkpoint_mutex so tail cannot change under us. | |
346 | * We don't need any special data guarantees for writing sb | |
347 | * since journal is empty and it is ok for write to be | |
348 | * flushed only with transaction commit. | |
349 | */ | |
350 | jbd2_journal_update_sb_log_tail(journal, | |
351 | journal->j_tail_sequence, | |
352 | journal->j_tail, | |
353 | WRITE_SYNC); | |
a78bb11d | 354 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
355 | } else { |
356 | jbd_debug(3, "superblock not updated\n"); | |
357 | } | |
358 | ||
359 | J_ASSERT(journal->j_running_transaction != NULL); | |
360 | J_ASSERT(journal->j_committing_transaction == NULL); | |
361 | ||
362 | commit_transaction = journal->j_running_transaction; | |
363 | J_ASSERT(commit_transaction->t_state == T_RUNNING); | |
364 | ||
879c5e6b | 365 | trace_jbd2_start_commit(journal, commit_transaction); |
f2a44523 | 366 | jbd_debug(1, "JBD2: starting commit of transaction %d\n", |
470decc6 DK |
367 | commit_transaction->t_tid); |
368 | ||
a931da6a | 369 | write_lock(&journal->j_state_lock); |
470decc6 DK |
370 | commit_transaction->t_state = T_LOCKED; |
371 | ||
879c5e6b | 372 | trace_jbd2_commit_locking(journal, commit_transaction); |
bf699327 TT |
373 | stats.run.rs_wait = commit_transaction->t_max_wait; |
374 | stats.run.rs_locked = jiffies; | |
375 | stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start, | |
376 | stats.run.rs_locked); | |
8e85fb3f | 377 | |
470decc6 | 378 | spin_lock(&commit_transaction->t_handle_lock); |
a51dca9c | 379 | while (atomic_read(&commit_transaction->t_updates)) { |
470decc6 DK |
380 | DEFINE_WAIT(wait); |
381 | ||
382 | prepare_to_wait(&journal->j_wait_updates, &wait, | |
383 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 384 | if (atomic_read(&commit_transaction->t_updates)) { |
470decc6 | 385 | spin_unlock(&commit_transaction->t_handle_lock); |
a931da6a | 386 | write_unlock(&journal->j_state_lock); |
470decc6 | 387 | schedule(); |
a931da6a | 388 | write_lock(&journal->j_state_lock); |
470decc6 DK |
389 | spin_lock(&commit_transaction->t_handle_lock); |
390 | } | |
391 | finish_wait(&journal->j_wait_updates, &wait); | |
392 | } | |
393 | spin_unlock(&commit_transaction->t_handle_lock); | |
394 | ||
a51dca9c | 395 | J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <= |
470decc6 DK |
396 | journal->j_max_transaction_buffers); |
397 | ||
398 | /* | |
399 | * First thing we are allowed to do is to discard any remaining | |
400 | * BJ_Reserved buffers. Note, it is _not_ permissible to assume | |
401 | * that there are no such buffers: if a large filesystem | |
402 | * operation like a truncate needs to split itself over multiple | |
f7f4bccb | 403 | * transactions, then it may try to do a jbd2_journal_restart() while |
470decc6 DK |
404 | * there are still BJ_Reserved buffers outstanding. These must |
405 | * be released cleanly from the current transaction. | |
406 | * | |
407 | * In this case, the filesystem must still reserve write access | |
408 | * again before modifying the buffer in the new transaction, but | |
409 | * we do not require it to remember exactly which old buffers it | |
410 | * has reserved. This is consistent with the existing behaviour | |
f7f4bccb | 411 | * that multiple jbd2_journal_get_write_access() calls to the same |
25985edc | 412 | * buffer are perfectly permissible. |
470decc6 DK |
413 | */ |
414 | while (commit_transaction->t_reserved_list) { | |
415 | jh = commit_transaction->t_reserved_list; | |
416 | JBUFFER_TRACE(jh, "reserved, unused: refile"); | |
417 | /* | |
f7f4bccb | 418 | * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may |
470decc6 DK |
419 | * leave undo-committed data. |
420 | */ | |
421 | if (jh->b_committed_data) { | |
422 | struct buffer_head *bh = jh2bh(jh); | |
423 | ||
424 | jbd_lock_bh_state(bh); | |
af1e76d6 | 425 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
426 | jh->b_committed_data = NULL; |
427 | jbd_unlock_bh_state(bh); | |
428 | } | |
f7f4bccb | 429 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
430 | } |
431 | ||
432 | /* | |
433 | * Now try to drop any written-back buffers from the journal's | |
434 | * checkpoint lists. We do this *before* commit because it potentially | |
435 | * frees some memory | |
436 | */ | |
437 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 438 | __jbd2_journal_clean_checkpoint_list(journal); |
470decc6 DK |
439 | spin_unlock(&journal->j_list_lock); |
440 | ||
f2a44523 | 441 | jbd_debug(3, "JBD2: commit phase 1\n"); |
470decc6 | 442 | |
1ba37268 YY |
443 | /* |
444 | * Clear revoked flag to reflect there is no revoked buffers | |
445 | * in the next transaction which is going to be started. | |
446 | */ | |
447 | jbd2_clear_buffer_revoked_flags(journal); | |
448 | ||
470decc6 DK |
449 | /* |
450 | * Switch to a new revoke table. | |
451 | */ | |
f7f4bccb | 452 | jbd2_journal_switch_revoke_table(journal); |
470decc6 | 453 | |
879c5e6b | 454 | trace_jbd2_commit_flushing(journal, commit_transaction); |
bf699327 TT |
455 | stats.run.rs_flushing = jiffies; |
456 | stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked, | |
457 | stats.run.rs_flushing); | |
8e85fb3f | 458 | |
470decc6 DK |
459 | commit_transaction->t_state = T_FLUSH; |
460 | journal->j_committing_transaction = commit_transaction; | |
461 | journal->j_running_transaction = NULL; | |
e07f7183 | 462 | start_time = ktime_get(); |
470decc6 DK |
463 | commit_transaction->t_log_start = journal->j_head; |
464 | wake_up(&journal->j_wait_transaction_locked); | |
a931da6a | 465 | write_unlock(&journal->j_state_lock); |
470decc6 | 466 | |
f2a44523 | 467 | jbd_debug(3, "JBD2: commit phase 2\n"); |
470decc6 | 468 | |
470decc6 DK |
469 | /* |
470 | * Now start flushing things to disk, in the order they appear | |
471 | * on the transaction lists. Data blocks go first. | |
472 | */ | |
cd1aac32 | 473 | err = journal_submit_data_buffers(journal, commit_transaction); |
470decc6 | 474 | if (err) |
a7fa2baf | 475 | jbd2_journal_abort(journal, err); |
470decc6 | 476 | |
82f04ab4 | 477 | blk_start_plug(&plug); |
67c457a8 | 478 | jbd2_journal_write_revoke_records(journal, commit_transaction, |
82f04ab4 JA |
479 | WRITE_SYNC); |
480 | blk_finish_plug(&plug); | |
470decc6 | 481 | |
f2a44523 | 482 | jbd_debug(3, "JBD2: commit phase 2\n"); |
470decc6 | 483 | |
470decc6 DK |
484 | /* |
485 | * Way to go: we have now written out all of the data for a | |
486 | * transaction! Now comes the tricky part: we need to write out | |
487 | * metadata. Loop over the transaction's entire buffer list: | |
488 | */ | |
a931da6a | 489 | write_lock(&journal->j_state_lock); |
470decc6 | 490 | commit_transaction->t_state = T_COMMIT; |
a931da6a | 491 | write_unlock(&journal->j_state_lock); |
470decc6 | 492 | |
879c5e6b | 493 | trace_jbd2_commit_logging(journal, commit_transaction); |
bf699327 TT |
494 | stats.run.rs_logging = jiffies; |
495 | stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing, | |
496 | stats.run.rs_logging); | |
a51dca9c TT |
497 | stats.run.rs_blocks = |
498 | atomic_read(&commit_transaction->t_outstanding_credits); | |
bf699327 | 499 | stats.run.rs_blocks_logged = 0; |
8e85fb3f | 500 | |
1dfc3220 | 501 | J_ASSERT(commit_transaction->t_nr_buffers <= |
a51dca9c | 502 | atomic_read(&commit_transaction->t_outstanding_credits)); |
1dfc3220 | 503 | |
87c89c23 | 504 | err = 0; |
470decc6 DK |
505 | descriptor = NULL; |
506 | bufs = 0; | |
82f04ab4 | 507 | blk_start_plug(&plug); |
470decc6 DK |
508 | while (commit_transaction->t_buffers) { |
509 | ||
510 | /* Find the next buffer to be journaled... */ | |
511 | ||
512 | jh = commit_transaction->t_buffers; | |
513 | ||
514 | /* If we're in abort mode, we just un-journal the buffer and | |
7ad7445f | 515 | release it. */ |
470decc6 DK |
516 | |
517 | if (is_journal_aborted(journal)) { | |
7ad7445f | 518 | clear_buffer_jbddirty(jh2bh(jh)); |
470decc6 | 519 | JBUFFER_TRACE(jh, "journal is aborting: refile"); |
e06c8227 JB |
520 | jbd2_buffer_abort_trigger(jh, |
521 | jh->b_frozen_data ? | |
522 | jh->b_frozen_triggers : | |
523 | jh->b_triggers); | |
f7f4bccb | 524 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
525 | /* If that was the last one, we need to clean up |
526 | * any descriptor buffers which may have been | |
527 | * already allocated, even if we are now | |
528 | * aborting. */ | |
529 | if (!commit_transaction->t_buffers) | |
530 | goto start_journal_io; | |
531 | continue; | |
532 | } | |
533 | ||
534 | /* Make sure we have a descriptor block in which to | |
535 | record the metadata buffer. */ | |
536 | ||
537 | if (!descriptor) { | |
538 | struct buffer_head *bh; | |
539 | ||
540 | J_ASSERT (bufs == 0); | |
541 | ||
f2a44523 | 542 | jbd_debug(4, "JBD2: get descriptor\n"); |
470decc6 | 543 | |
f7f4bccb | 544 | descriptor = jbd2_journal_get_descriptor_buffer(journal); |
470decc6 | 545 | if (!descriptor) { |
a7fa2baf | 546 | jbd2_journal_abort(journal, -EIO); |
470decc6 DK |
547 | continue; |
548 | } | |
549 | ||
550 | bh = jh2bh(descriptor); | |
f2a44523 | 551 | jbd_debug(4, "JBD2: got buffer %llu (%p)\n", |
470decc6 DK |
552 | (unsigned long long)bh->b_blocknr, bh->b_data); |
553 | header = (journal_header_t *)&bh->b_data[0]; | |
f7f4bccb MC |
554 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); |
555 | header->h_blocktype = cpu_to_be32(JBD2_DESCRIPTOR_BLOCK); | |
470decc6 DK |
556 | header->h_sequence = cpu_to_be32(commit_transaction->t_tid); |
557 | ||
558 | tagp = &bh->b_data[sizeof(journal_header_t)]; | |
559 | space_left = bh->b_size - sizeof(journal_header_t); | |
560 | first_tag = 1; | |
561 | set_buffer_jwrite(bh); | |
562 | set_buffer_dirty(bh); | |
563 | wbuf[bufs++] = bh; | |
564 | ||
565 | /* Record it so that we can wait for IO | |
566 | completion later */ | |
567 | BUFFER_TRACE(bh, "ph3: file as descriptor"); | |
f7f4bccb | 568 | jbd2_journal_file_buffer(descriptor, commit_transaction, |
470decc6 DK |
569 | BJ_LogCtl); |
570 | } | |
571 | ||
572 | /* Where is the buffer to be written? */ | |
573 | ||
f7f4bccb | 574 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
575 | /* If the block mapping failed, just abandon the buffer |
576 | and repeat this loop: we'll fall into the | |
577 | refile-on-abort condition above. */ | |
578 | if (err) { | |
a7fa2baf | 579 | jbd2_journal_abort(journal, err); |
470decc6 DK |
580 | continue; |
581 | } | |
582 | ||
583 | /* | |
584 | * start_this_handle() uses t_outstanding_credits to determine | |
585 | * the free space in the log, but this counter is changed | |
f7f4bccb | 586 | * by jbd2_journal_next_log_block() also. |
470decc6 | 587 | */ |
a51dca9c | 588 | atomic_dec(&commit_transaction->t_outstanding_credits); |
470decc6 DK |
589 | |
590 | /* Bump b_count to prevent truncate from stumbling over | |
591 | the shadowed buffer! @@@ This can go if we ever get | |
592 | rid of the BJ_IO/BJ_Shadow pairing of buffers. */ | |
593 | atomic_inc(&jh2bh(jh)->b_count); | |
594 | ||
595 | /* Make a temporary IO buffer with which to write it out | |
596 | (this will requeue both the metadata buffer and the | |
597 | temporary IO buffer). new_bh goes on BJ_IO*/ | |
598 | ||
599 | set_bit(BH_JWrite, &jh2bh(jh)->b_state); | |
600 | /* | |
f7f4bccb | 601 | * akpm: jbd2_journal_write_metadata_buffer() sets |
470decc6 DK |
602 | * new_bh->b_transaction to commit_transaction. |
603 | * We need to clean this up before we release new_bh | |
604 | * (which is of type BJ_IO) | |
605 | */ | |
606 | JBUFFER_TRACE(jh, "ph3: write metadata"); | |
f7f4bccb | 607 | flags = jbd2_journal_write_metadata_buffer(commit_transaction, |
470decc6 | 608 | jh, &new_jh, blocknr); |
e6ec116b TT |
609 | if (flags < 0) { |
610 | jbd2_journal_abort(journal, flags); | |
611 | continue; | |
612 | } | |
470decc6 DK |
613 | set_bit(BH_JWrite, &jh2bh(new_jh)->b_state); |
614 | wbuf[bufs++] = jh2bh(new_jh); | |
615 | ||
616 | /* Record the new block's tag in the current descriptor | |
617 | buffer */ | |
618 | ||
619 | tag_flag = 0; | |
620 | if (flags & 1) | |
f7f4bccb | 621 | tag_flag |= JBD2_FLAG_ESCAPE; |
470decc6 | 622 | if (!first_tag) |
f7f4bccb | 623 | tag_flag |= JBD2_FLAG_SAME_UUID; |
470decc6 DK |
624 | |
625 | tag = (journal_block_tag_t *) tagp; | |
b517bea1 | 626 | write_tag_block(tag_bytes, tag, jh2bh(jh)->b_blocknr); |
470decc6 | 627 | tag->t_flags = cpu_to_be32(tag_flag); |
b517bea1 ZB |
628 | tagp += tag_bytes; |
629 | space_left -= tag_bytes; | |
470decc6 DK |
630 | |
631 | if (first_tag) { | |
632 | memcpy (tagp, journal->j_uuid, 16); | |
633 | tagp += 16; | |
634 | space_left -= 16; | |
635 | first_tag = 0; | |
636 | } | |
637 | ||
638 | /* If there's no more to do, or if the descriptor is full, | |
639 | let the IO rip! */ | |
640 | ||
641 | if (bufs == journal->j_wbufsize || | |
642 | commit_transaction->t_buffers == NULL || | |
b517bea1 | 643 | space_left < tag_bytes + 16) { |
470decc6 | 644 | |
f2a44523 | 645 | jbd_debug(4, "JBD2: Submit %d IOs\n", bufs); |
470decc6 DK |
646 | |
647 | /* Write an end-of-descriptor marker before | |
648 | submitting the IOs. "tag" still points to | |
649 | the last tag we set up. */ | |
650 | ||
f7f4bccb | 651 | tag->t_flags |= cpu_to_be32(JBD2_FLAG_LAST_TAG); |
470decc6 DK |
652 | |
653 | start_journal_io: | |
654 | for (i = 0; i < bufs; i++) { | |
655 | struct buffer_head *bh = wbuf[i]; | |
818d276c GS |
656 | /* |
657 | * Compute checksum. | |
658 | */ | |
659 | if (JBD2_HAS_COMPAT_FEATURE(journal, | |
660 | JBD2_FEATURE_COMPAT_CHECKSUM)) { | |
661 | crc32_sum = | |
662 | jbd2_checksum_data(crc32_sum, bh); | |
663 | } | |
664 | ||
470decc6 DK |
665 | lock_buffer(bh); |
666 | clear_buffer_dirty(bh); | |
667 | set_buffer_uptodate(bh); | |
668 | bh->b_end_io = journal_end_buffer_io_sync; | |
82f04ab4 | 669 | submit_bh(WRITE_SYNC, bh); |
470decc6 DK |
670 | } |
671 | cond_resched(); | |
bf699327 | 672 | stats.run.rs_blocks_logged += bufs; |
470decc6 DK |
673 | |
674 | /* Force a new descriptor to be generated next | |
675 | time round the loop. */ | |
676 | descriptor = NULL; | |
677 | bufs = 0; | |
678 | } | |
679 | } | |
680 | ||
f73bee49 JK |
681 | err = journal_finish_inode_data_buffers(journal, commit_transaction); |
682 | if (err) { | |
683 | printk(KERN_WARNING | |
684 | "JBD2: Detected IO errors while flushing file data " | |
685 | "on %s\n", journal->j_devname); | |
686 | if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR) | |
687 | jbd2_journal_abort(journal, err); | |
688 | err = 0; | |
689 | } | |
690 | ||
bbd2be36 JK |
691 | write_lock(&journal->j_state_lock); |
692 | J_ASSERT(commit_transaction->t_state == T_COMMIT); | |
693 | commit_transaction->t_state = T_COMMIT_DFLUSH; | |
694 | write_unlock(&journal->j_state_lock); | |
cc3e1bea TT |
695 | /* |
696 | * If the journal is not located on the file system device, | |
697 | * then we must flush the file system device before we issue | |
698 | * the commit record | |
699 | */ | |
81be12c8 | 700 | if (commit_transaction->t_need_data_flush && |
cc3e1bea TT |
701 | (journal->j_fs_dev != journal->j_dev) && |
702 | (journal->j_flags & JBD2_BARRIER)) | |
dd3932ed | 703 | blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL); |
818d276c | 704 | |
cc3e1bea | 705 | /* Done it all: now write the commit record asynchronously. */ |
818d276c | 706 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, |
0e3d2a63 | 707 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { |
818d276c GS |
708 | err = journal_submit_commit_record(journal, commit_transaction, |
709 | &cbh, crc32_sum); | |
710 | if (err) | |
711 | __jbd2_journal_abort_hard(journal); | |
e9e34f4e | 712 | } |
c851ed54 | 713 | |
82f04ab4 JA |
714 | blk_finish_plug(&plug); |
715 | ||
470decc6 DK |
716 | /* Lo and behold: we have just managed to send a transaction to |
717 | the log. Before we can commit it, wait for the IO so far to | |
718 | complete. Control buffers being written are on the | |
719 | transaction's t_log_list queue, and metadata buffers are on | |
720 | the t_iobuf_list queue. | |
721 | ||
722 | Wait for the buffers in reverse order. That way we are | |
723 | less likely to be woken up until all IOs have completed, and | |
724 | so we incur less scheduling load. | |
725 | */ | |
726 | ||
f2a44523 | 727 | jbd_debug(3, "JBD2: commit phase 3\n"); |
470decc6 DK |
728 | |
729 | /* | |
730 | * akpm: these are BJ_IO, and j_list_lock is not needed. | |
731 | * See __journal_try_to_free_buffer. | |
732 | */ | |
733 | wait_for_iobuf: | |
734 | while (commit_transaction->t_iobuf_list != NULL) { | |
735 | struct buffer_head *bh; | |
736 | ||
737 | jh = commit_transaction->t_iobuf_list->b_tprev; | |
738 | bh = jh2bh(jh); | |
739 | if (buffer_locked(bh)) { | |
740 | wait_on_buffer(bh); | |
741 | goto wait_for_iobuf; | |
742 | } | |
743 | if (cond_resched()) | |
744 | goto wait_for_iobuf; | |
745 | ||
746 | if (unlikely(!buffer_uptodate(bh))) | |
747 | err = -EIO; | |
748 | ||
749 | clear_buffer_jwrite(bh); | |
750 | ||
751 | JBUFFER_TRACE(jh, "ph4: unfile after journal write"); | |
f7f4bccb | 752 | jbd2_journal_unfile_buffer(journal, jh); |
470decc6 DK |
753 | |
754 | /* | |
755 | * ->t_iobuf_list should contain only dummy buffer_heads | |
f7f4bccb | 756 | * which were created by jbd2_journal_write_metadata_buffer(). |
470decc6 DK |
757 | */ |
758 | BUFFER_TRACE(bh, "dumping temporary bh"); | |
f7f4bccb | 759 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
760 | __brelse(bh); |
761 | J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0); | |
762 | free_buffer_head(bh); | |
763 | ||
764 | /* We also have to unlock and free the corresponding | |
765 | shadowed buffer */ | |
766 | jh = commit_transaction->t_shadow_list->b_tprev; | |
767 | bh = jh2bh(jh); | |
768 | clear_bit(BH_JWrite, &bh->b_state); | |
769 | J_ASSERT_BH(bh, buffer_jbddirty(bh)); | |
770 | ||
771 | /* The metadata is now released for reuse, but we need | |
772 | to remember it against this transaction so that when | |
773 | we finally commit, we can do any checkpointing | |
774 | required. */ | |
775 | JBUFFER_TRACE(jh, "file as BJ_Forget"); | |
f7f4bccb | 776 | jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget); |
229309ca JK |
777 | /* |
778 | * Wake up any transactions which were waiting for this IO to | |
779 | * complete. The barrier must be here so that changes by | |
780 | * jbd2_journal_file_buffer() take effect before wake_up_bit() | |
781 | * does the waitqueue check. | |
782 | */ | |
783 | smp_mb(); | |
470decc6 DK |
784 | wake_up_bit(&bh->b_state, BH_Unshadow); |
785 | JBUFFER_TRACE(jh, "brelse shadowed buffer"); | |
786 | __brelse(bh); | |
787 | } | |
788 | ||
789 | J_ASSERT (commit_transaction->t_shadow_list == NULL); | |
790 | ||
f2a44523 | 791 | jbd_debug(3, "JBD2: commit phase 4\n"); |
470decc6 DK |
792 | |
793 | /* Here we wait for the revoke record and descriptor record buffers */ | |
794 | wait_for_ctlbuf: | |
795 | while (commit_transaction->t_log_list != NULL) { | |
796 | struct buffer_head *bh; | |
797 | ||
798 | jh = commit_transaction->t_log_list->b_tprev; | |
799 | bh = jh2bh(jh); | |
800 | if (buffer_locked(bh)) { | |
801 | wait_on_buffer(bh); | |
802 | goto wait_for_ctlbuf; | |
803 | } | |
804 | if (cond_resched()) | |
805 | goto wait_for_ctlbuf; | |
806 | ||
807 | if (unlikely(!buffer_uptodate(bh))) | |
808 | err = -EIO; | |
809 | ||
810 | BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile"); | |
811 | clear_buffer_jwrite(bh); | |
f7f4bccb MC |
812 | jbd2_journal_unfile_buffer(journal, jh); |
813 | jbd2_journal_put_journal_head(jh); | |
470decc6 DK |
814 | __brelse(bh); /* One for getblk */ |
815 | /* AKPM: bforget here */ | |
816 | } | |
817 | ||
77e841de HK |
818 | if (err) |
819 | jbd2_journal_abort(journal, err); | |
820 | ||
f2a44523 | 821 | jbd_debug(3, "JBD2: commit phase 5\n"); |
bbd2be36 JK |
822 | write_lock(&journal->j_state_lock); |
823 | J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH); | |
824 | commit_transaction->t_state = T_COMMIT_JFLUSH; | |
825 | write_unlock(&journal->j_state_lock); | |
470decc6 | 826 | |
818d276c | 827 | if (!JBD2_HAS_INCOMPAT_FEATURE(journal, |
0e3d2a63 | 828 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { |
818d276c GS |
829 | err = journal_submit_commit_record(journal, commit_transaction, |
830 | &cbh, crc32_sum); | |
831 | if (err) | |
832 | __jbd2_journal_abort_hard(journal); | |
833 | } | |
6cba611e | 834 | if (cbh) |
fd98496f | 835 | err = journal_wait_on_commit_record(journal, cbh); |
f73bee49 JK |
836 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, |
837 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) && | |
838 | journal->j_flags & JBD2_BARRIER) { | |
dd3932ed | 839 | blkdev_issue_flush(journal->j_dev, GFP_KERNEL, NULL); |
f73bee49 | 840 | } |
470decc6 DK |
841 | |
842 | if (err) | |
a7fa2baf | 843 | jbd2_journal_abort(journal, err); |
470decc6 DK |
844 | |
845 | /* End of a transaction! Finally, we can do checkpoint | |
846 | processing: any buffers committed as a result of this | |
847 | transaction can be removed from any checkpoint list it was on | |
848 | before. */ | |
849 | ||
f2a44523 | 850 | jbd_debug(3, "JBD2: commit phase 6\n"); |
470decc6 | 851 | |
c851ed54 | 852 | J_ASSERT(list_empty(&commit_transaction->t_inode_list)); |
470decc6 DK |
853 | J_ASSERT(commit_transaction->t_buffers == NULL); |
854 | J_ASSERT(commit_transaction->t_checkpoint_list == NULL); | |
855 | J_ASSERT(commit_transaction->t_iobuf_list == NULL); | |
856 | J_ASSERT(commit_transaction->t_shadow_list == NULL); | |
857 | J_ASSERT(commit_transaction->t_log_list == NULL); | |
858 | ||
859 | restart_loop: | |
860 | /* | |
861 | * As there are other places (journal_unmap_buffer()) adding buffers | |
862 | * to this list we have to be careful and hold the j_list_lock. | |
863 | */ | |
864 | spin_lock(&journal->j_list_lock); | |
865 | while (commit_transaction->t_forget) { | |
866 | transaction_t *cp_transaction; | |
867 | struct buffer_head *bh; | |
de1b7941 | 868 | int try_to_free = 0; |
470decc6 DK |
869 | |
870 | jh = commit_transaction->t_forget; | |
871 | spin_unlock(&journal->j_list_lock); | |
872 | bh = jh2bh(jh); | |
de1b7941 JK |
873 | /* |
874 | * Get a reference so that bh cannot be freed before we are | |
875 | * done with it. | |
876 | */ | |
877 | get_bh(bh); | |
470decc6 | 878 | jbd_lock_bh_state(bh); |
23e2af35 | 879 | J_ASSERT_JH(jh, jh->b_transaction == commit_transaction); |
470decc6 DK |
880 | |
881 | /* | |
882 | * If there is undo-protected committed data against | |
883 | * this buffer, then we can remove it now. If it is a | |
884 | * buffer needing such protection, the old frozen_data | |
885 | * field now points to a committed version of the | |
886 | * buffer, so rotate that field to the new committed | |
887 | * data. | |
888 | * | |
889 | * Otherwise, we can just throw away the frozen data now. | |
e06c8227 JB |
890 | * |
891 | * We also know that the frozen data has already fired | |
892 | * its triggers if they exist, so we can clear that too. | |
470decc6 DK |
893 | */ |
894 | if (jh->b_committed_data) { | |
af1e76d6 | 895 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
896 | jh->b_committed_data = NULL; |
897 | if (jh->b_frozen_data) { | |
898 | jh->b_committed_data = jh->b_frozen_data; | |
899 | jh->b_frozen_data = NULL; | |
e06c8227 | 900 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
901 | } |
902 | } else if (jh->b_frozen_data) { | |
af1e76d6 | 903 | jbd2_free(jh->b_frozen_data, bh->b_size); |
470decc6 | 904 | jh->b_frozen_data = NULL; |
e06c8227 | 905 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
906 | } |
907 | ||
908 | spin_lock(&journal->j_list_lock); | |
909 | cp_transaction = jh->b_cp_transaction; | |
910 | if (cp_transaction) { | |
911 | JBUFFER_TRACE(jh, "remove from old cp transaction"); | |
8e85fb3f | 912 | cp_transaction->t_chp_stats.cs_dropped++; |
f7f4bccb | 913 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
914 | } |
915 | ||
916 | /* Only re-checkpoint the buffer_head if it is marked | |
917 | * dirty. If the buffer was added to the BJ_Forget list | |
f7f4bccb | 918 | * by jbd2_journal_forget, it may no longer be dirty and |
470decc6 DK |
919 | * there's no point in keeping a checkpoint record for |
920 | * it. */ | |
921 | ||
922 | /* A buffer which has been freed while still being | |
923 | * journaled by a previous transaction may end up still | |
924 | * being dirty here, but we want to avoid writing back | |
ba869023 | 925 | * that buffer in the future after the "add to orphan" |
926 | * operation been committed, That's not only a performance | |
927 | * gain, it also stops aliasing problems if the buffer is | |
928 | * left behind for writeback and gets reallocated for another | |
470decc6 | 929 | * use in a different page. */ |
ba869023 | 930 | if (buffer_freed(bh) && !jh->b_next_transaction) { |
470decc6 DK |
931 | clear_buffer_freed(bh); |
932 | clear_buffer_jbddirty(bh); | |
933 | } | |
934 | ||
935 | if (buffer_jbddirty(bh)) { | |
936 | JBUFFER_TRACE(jh, "add to new checkpointing trans"); | |
f7f4bccb | 937 | __jbd2_journal_insert_checkpoint(jh, commit_transaction); |
7ad7445f HK |
938 | if (is_journal_aborted(journal)) |
939 | clear_buffer_jbddirty(bh); | |
470decc6 DK |
940 | } else { |
941 | J_ASSERT_BH(bh, !buffer_dirty(bh)); | |
de1b7941 JK |
942 | /* |
943 | * The buffer on BJ_Forget list and not jbddirty means | |
470decc6 DK |
944 | * it has been freed by this transaction and hence it |
945 | * could not have been reallocated until this | |
946 | * transaction has committed. *BUT* it could be | |
947 | * reallocated once we have written all the data to | |
948 | * disk and before we process the buffer on BJ_Forget | |
de1b7941 JK |
949 | * list. |
950 | */ | |
951 | if (!jh->b_next_transaction) | |
952 | try_to_free = 1; | |
470decc6 | 953 | } |
de1b7941 JK |
954 | JBUFFER_TRACE(jh, "refile or unfile buffer"); |
955 | __jbd2_journal_refile_buffer(jh); | |
956 | jbd_unlock_bh_state(bh); | |
957 | if (try_to_free) | |
958 | release_buffer_page(bh); /* Drops bh reference */ | |
959 | else | |
960 | __brelse(bh); | |
470decc6 DK |
961 | cond_resched_lock(&journal->j_list_lock); |
962 | } | |
963 | spin_unlock(&journal->j_list_lock); | |
964 | /* | |
f5a7a6b0 JK |
965 | * This is a bit sleazy. We use j_list_lock to protect transition |
966 | * of a transaction into T_FINISHED state and calling | |
967 | * __jbd2_journal_drop_transaction(). Otherwise we could race with | |
968 | * other checkpointing code processing the transaction... | |
470decc6 | 969 | */ |
a931da6a | 970 | write_lock(&journal->j_state_lock); |
470decc6 DK |
971 | spin_lock(&journal->j_list_lock); |
972 | /* | |
973 | * Now recheck if some buffers did not get attached to the transaction | |
974 | * while the lock was dropped... | |
975 | */ | |
976 | if (commit_transaction->t_forget) { | |
977 | spin_unlock(&journal->j_list_lock); | |
a931da6a | 978 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
979 | goto restart_loop; |
980 | } | |
981 | ||
982 | /* Done with this transaction! */ | |
983 | ||
f2a44523 | 984 | jbd_debug(3, "JBD2: commit phase 7\n"); |
470decc6 | 985 | |
bbd2be36 | 986 | J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH); |
470decc6 | 987 | |
8e85fb3f | 988 | commit_transaction->t_start = jiffies; |
bf699327 TT |
989 | stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging, |
990 | commit_transaction->t_start); | |
8e85fb3f JL |
991 | |
992 | /* | |
bf699327 | 993 | * File the transaction statistics |
8e85fb3f | 994 | */ |
8e85fb3f | 995 | stats.ts_tid = commit_transaction->t_tid; |
8dd42046 TT |
996 | stats.run.rs_handle_count = |
997 | atomic_read(&commit_transaction->t_handle_count); | |
bf699327 TT |
998 | trace_jbd2_run_stats(journal->j_fs_dev->bd_dev, |
999 | commit_transaction->t_tid, &stats.run); | |
8e85fb3f JL |
1000 | |
1001 | /* | |
1002 | * Calculate overall stats | |
1003 | */ | |
bf699327 | 1004 | spin_lock(&journal->j_history_lock); |
8e85fb3f | 1005 | journal->j_stats.ts_tid++; |
bf699327 TT |
1006 | journal->j_stats.run.rs_wait += stats.run.rs_wait; |
1007 | journal->j_stats.run.rs_running += stats.run.rs_running; | |
1008 | journal->j_stats.run.rs_locked += stats.run.rs_locked; | |
1009 | journal->j_stats.run.rs_flushing += stats.run.rs_flushing; | |
1010 | journal->j_stats.run.rs_logging += stats.run.rs_logging; | |
1011 | journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count; | |
1012 | journal->j_stats.run.rs_blocks += stats.run.rs_blocks; | |
1013 | journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged; | |
8e85fb3f JL |
1014 | spin_unlock(&journal->j_history_lock); |
1015 | ||
470decc6 DK |
1016 | commit_transaction->t_state = T_FINISHED; |
1017 | J_ASSERT(commit_transaction == journal->j_committing_transaction); | |
1018 | journal->j_commit_sequence = commit_transaction->t_tid; | |
1019 | journal->j_committing_transaction = NULL; | |
e07f7183 | 1020 | commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); |
470decc6 | 1021 | |
e07f7183 JB |
1022 | /* |
1023 | * weight the commit time higher than the average time so we don't | |
1024 | * react too strongly to vast changes in the commit time | |
1025 | */ | |
1026 | if (likely(journal->j_average_commit_time)) | |
1027 | journal->j_average_commit_time = (commit_time + | |
1028 | journal->j_average_commit_time*3) / 4; | |
1029 | else | |
1030 | journal->j_average_commit_time = commit_time; | |
a931da6a | 1031 | write_unlock(&journal->j_state_lock); |
6c20ec85 | 1032 | |
f89b7795 JK |
1033 | if (commit_transaction->t_checkpoint_list == NULL && |
1034 | commit_transaction->t_checkpoint_io_list == NULL) { | |
f7f4bccb | 1035 | __jbd2_journal_drop_transaction(journal, commit_transaction); |
fb68407b | 1036 | to_free = 1; |
470decc6 DK |
1037 | } else { |
1038 | if (journal->j_checkpoint_transactions == NULL) { | |
1039 | journal->j_checkpoint_transactions = commit_transaction; | |
1040 | commit_transaction->t_cpnext = commit_transaction; | |
1041 | commit_transaction->t_cpprev = commit_transaction; | |
1042 | } else { | |
1043 | commit_transaction->t_cpnext = | |
1044 | journal->j_checkpoint_transactions; | |
1045 | commit_transaction->t_cpprev = | |
1046 | commit_transaction->t_cpnext->t_cpprev; | |
1047 | commit_transaction->t_cpnext->t_cpprev = | |
1048 | commit_transaction; | |
1049 | commit_transaction->t_cpprev->t_cpnext = | |
1050 | commit_transaction; | |
1051 | } | |
1052 | } | |
1053 | spin_unlock(&journal->j_list_lock); | |
1054 | ||
fb68407b AK |
1055 | if (journal->j_commit_callback) |
1056 | journal->j_commit_callback(journal, commit_transaction); | |
1057 | ||
879c5e6b | 1058 | trace_jbd2_end_commit(journal, commit_transaction); |
f2a44523 | 1059 | jbd_debug(1, "JBD2: commit %d complete, head %d\n", |
470decc6 | 1060 | journal->j_commit_sequence, journal->j_tail_sequence); |
fb68407b | 1061 | if (to_free) |
0c2022ec | 1062 | jbd2_journal_free_transaction(commit_transaction); |
470decc6 DK |
1063 | |
1064 | wake_up(&journal->j_wait_done_commit); | |
1065 | } |