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 | ||
340 | #ifdef COMMIT_STATS | |
341 | spin_lock(&journal->j_list_lock); | |
342 | summarise_journal_usage(journal); | |
343 | spin_unlock(&journal->j_list_lock); | |
344 | #endif | |
345 | ||
f7f4bccb MC |
346 | /* Do we need to erase the effects of a prior jbd2_journal_flush? */ |
347 | if (journal->j_flags & JBD2_FLUSHED) { | |
470decc6 | 348 | jbd_debug(3, "super block updated\n"); |
f7f4bccb | 349 | jbd2_journal_update_superblock(journal, 1); |
470decc6 DK |
350 | } else { |
351 | jbd_debug(3, "superblock not updated\n"); | |
352 | } | |
353 | ||
354 | J_ASSERT(journal->j_running_transaction != NULL); | |
355 | J_ASSERT(journal->j_committing_transaction == NULL); | |
356 | ||
357 | commit_transaction = journal->j_running_transaction; | |
358 | J_ASSERT(commit_transaction->t_state == T_RUNNING); | |
359 | ||
879c5e6b | 360 | trace_jbd2_start_commit(journal, commit_transaction); |
470decc6 DK |
361 | jbd_debug(1, "JBD: starting commit of transaction %d\n", |
362 | commit_transaction->t_tid); | |
363 | ||
a931da6a | 364 | write_lock(&journal->j_state_lock); |
470decc6 DK |
365 | commit_transaction->t_state = T_LOCKED; |
366 | ||
879c5e6b | 367 | trace_jbd2_commit_locking(journal, commit_transaction); |
bf699327 TT |
368 | stats.run.rs_wait = commit_transaction->t_max_wait; |
369 | stats.run.rs_locked = jiffies; | |
370 | stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start, | |
371 | stats.run.rs_locked); | |
8e85fb3f | 372 | |
470decc6 | 373 | spin_lock(&commit_transaction->t_handle_lock); |
a51dca9c | 374 | while (atomic_read(&commit_transaction->t_updates)) { |
470decc6 DK |
375 | DEFINE_WAIT(wait); |
376 | ||
377 | prepare_to_wait(&journal->j_wait_updates, &wait, | |
378 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 379 | if (atomic_read(&commit_transaction->t_updates)) { |
470decc6 | 380 | spin_unlock(&commit_transaction->t_handle_lock); |
a931da6a | 381 | write_unlock(&journal->j_state_lock); |
470decc6 | 382 | schedule(); |
a931da6a | 383 | write_lock(&journal->j_state_lock); |
470decc6 DK |
384 | spin_lock(&commit_transaction->t_handle_lock); |
385 | } | |
386 | finish_wait(&journal->j_wait_updates, &wait); | |
387 | } | |
388 | spin_unlock(&commit_transaction->t_handle_lock); | |
389 | ||
a51dca9c | 390 | J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <= |
470decc6 DK |
391 | journal->j_max_transaction_buffers); |
392 | ||
393 | /* | |
394 | * First thing we are allowed to do is to discard any remaining | |
395 | * BJ_Reserved buffers. Note, it is _not_ permissible to assume | |
396 | * that there are no such buffers: if a large filesystem | |
397 | * operation like a truncate needs to split itself over multiple | |
f7f4bccb | 398 | * transactions, then it may try to do a jbd2_journal_restart() while |
470decc6 DK |
399 | * there are still BJ_Reserved buffers outstanding. These must |
400 | * be released cleanly from the current transaction. | |
401 | * | |
402 | * In this case, the filesystem must still reserve write access | |
403 | * again before modifying the buffer in the new transaction, but | |
404 | * we do not require it to remember exactly which old buffers it | |
405 | * has reserved. This is consistent with the existing behaviour | |
f7f4bccb | 406 | * that multiple jbd2_journal_get_write_access() calls to the same |
25985edc | 407 | * buffer are perfectly permissible. |
470decc6 DK |
408 | */ |
409 | while (commit_transaction->t_reserved_list) { | |
410 | jh = commit_transaction->t_reserved_list; | |
411 | JBUFFER_TRACE(jh, "reserved, unused: refile"); | |
412 | /* | |
f7f4bccb | 413 | * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may |
470decc6 DK |
414 | * leave undo-committed data. |
415 | */ | |
416 | if (jh->b_committed_data) { | |
417 | struct buffer_head *bh = jh2bh(jh); | |
418 | ||
419 | jbd_lock_bh_state(bh); | |
af1e76d6 | 420 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
421 | jh->b_committed_data = NULL; |
422 | jbd_unlock_bh_state(bh); | |
423 | } | |
f7f4bccb | 424 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
425 | } |
426 | ||
427 | /* | |
428 | * Now try to drop any written-back buffers from the journal's | |
429 | * checkpoint lists. We do this *before* commit because it potentially | |
430 | * frees some memory | |
431 | */ | |
432 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 433 | __jbd2_journal_clean_checkpoint_list(journal); |
470decc6 DK |
434 | spin_unlock(&journal->j_list_lock); |
435 | ||
436 | jbd_debug (3, "JBD: commit phase 1\n"); | |
437 | ||
438 | /* | |
439 | * Switch to a new revoke table. | |
440 | */ | |
f7f4bccb | 441 | jbd2_journal_switch_revoke_table(journal); |
470decc6 | 442 | |
879c5e6b | 443 | trace_jbd2_commit_flushing(journal, commit_transaction); |
bf699327 TT |
444 | stats.run.rs_flushing = jiffies; |
445 | stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked, | |
446 | stats.run.rs_flushing); | |
8e85fb3f | 447 | |
470decc6 DK |
448 | commit_transaction->t_state = T_FLUSH; |
449 | journal->j_committing_transaction = commit_transaction; | |
450 | journal->j_running_transaction = NULL; | |
e07f7183 | 451 | start_time = ktime_get(); |
470decc6 DK |
452 | commit_transaction->t_log_start = journal->j_head; |
453 | wake_up(&journal->j_wait_transaction_locked); | |
a931da6a | 454 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
455 | |
456 | jbd_debug (3, "JBD: commit phase 2\n"); | |
457 | ||
470decc6 DK |
458 | /* |
459 | * Now start flushing things to disk, in the order they appear | |
460 | * on the transaction lists. Data blocks go first. | |
461 | */ | |
cd1aac32 | 462 | err = journal_submit_data_buffers(journal, commit_transaction); |
470decc6 | 463 | if (err) |
a7fa2baf | 464 | jbd2_journal_abort(journal, err); |
470decc6 | 465 | |
82f04ab4 | 466 | blk_start_plug(&plug); |
67c457a8 | 467 | jbd2_journal_write_revoke_records(journal, commit_transaction, |
82f04ab4 JA |
468 | WRITE_SYNC); |
469 | blk_finish_plug(&plug); | |
470decc6 DK |
470 | |
471 | jbd_debug(3, "JBD: commit phase 2\n"); | |
472 | ||
470decc6 DK |
473 | /* |
474 | * Way to go: we have now written out all of the data for a | |
475 | * transaction! Now comes the tricky part: we need to write out | |
476 | * metadata. Loop over the transaction's entire buffer list: | |
477 | */ | |
a931da6a | 478 | write_lock(&journal->j_state_lock); |
470decc6 | 479 | commit_transaction->t_state = T_COMMIT; |
a931da6a | 480 | write_unlock(&journal->j_state_lock); |
470decc6 | 481 | |
879c5e6b | 482 | trace_jbd2_commit_logging(journal, commit_transaction); |
bf699327 TT |
483 | stats.run.rs_logging = jiffies; |
484 | stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing, | |
485 | stats.run.rs_logging); | |
a51dca9c TT |
486 | stats.run.rs_blocks = |
487 | atomic_read(&commit_transaction->t_outstanding_credits); | |
bf699327 | 488 | stats.run.rs_blocks_logged = 0; |
8e85fb3f | 489 | |
1dfc3220 | 490 | J_ASSERT(commit_transaction->t_nr_buffers <= |
a51dca9c | 491 | atomic_read(&commit_transaction->t_outstanding_credits)); |
1dfc3220 | 492 | |
87c89c23 | 493 | err = 0; |
470decc6 DK |
494 | descriptor = NULL; |
495 | bufs = 0; | |
82f04ab4 | 496 | blk_start_plug(&plug); |
470decc6 DK |
497 | while (commit_transaction->t_buffers) { |
498 | ||
499 | /* Find the next buffer to be journaled... */ | |
500 | ||
501 | jh = commit_transaction->t_buffers; | |
502 | ||
503 | /* If we're in abort mode, we just un-journal the buffer and | |
7ad7445f | 504 | release it. */ |
470decc6 DK |
505 | |
506 | if (is_journal_aborted(journal)) { | |
7ad7445f | 507 | clear_buffer_jbddirty(jh2bh(jh)); |
470decc6 | 508 | JBUFFER_TRACE(jh, "journal is aborting: refile"); |
e06c8227 JB |
509 | jbd2_buffer_abort_trigger(jh, |
510 | jh->b_frozen_data ? | |
511 | jh->b_frozen_triggers : | |
512 | jh->b_triggers); | |
f7f4bccb | 513 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
514 | /* If that was the last one, we need to clean up |
515 | * any descriptor buffers which may have been | |
516 | * already allocated, even if we are now | |
517 | * aborting. */ | |
518 | if (!commit_transaction->t_buffers) | |
519 | goto start_journal_io; | |
520 | continue; | |
521 | } | |
522 | ||
523 | /* Make sure we have a descriptor block in which to | |
524 | record the metadata buffer. */ | |
525 | ||
526 | if (!descriptor) { | |
527 | struct buffer_head *bh; | |
528 | ||
529 | J_ASSERT (bufs == 0); | |
530 | ||
531 | jbd_debug(4, "JBD: get descriptor\n"); | |
532 | ||
f7f4bccb | 533 | descriptor = jbd2_journal_get_descriptor_buffer(journal); |
470decc6 | 534 | if (!descriptor) { |
a7fa2baf | 535 | jbd2_journal_abort(journal, -EIO); |
470decc6 DK |
536 | continue; |
537 | } | |
538 | ||
539 | bh = jh2bh(descriptor); | |
540 | jbd_debug(4, "JBD: got buffer %llu (%p)\n", | |
541 | (unsigned long long)bh->b_blocknr, bh->b_data); | |
542 | header = (journal_header_t *)&bh->b_data[0]; | |
f7f4bccb MC |
543 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); |
544 | header->h_blocktype = cpu_to_be32(JBD2_DESCRIPTOR_BLOCK); | |
470decc6 DK |
545 | header->h_sequence = cpu_to_be32(commit_transaction->t_tid); |
546 | ||
547 | tagp = &bh->b_data[sizeof(journal_header_t)]; | |
548 | space_left = bh->b_size - sizeof(journal_header_t); | |
549 | first_tag = 1; | |
550 | set_buffer_jwrite(bh); | |
551 | set_buffer_dirty(bh); | |
552 | wbuf[bufs++] = bh; | |
553 | ||
554 | /* Record it so that we can wait for IO | |
555 | completion later */ | |
556 | BUFFER_TRACE(bh, "ph3: file as descriptor"); | |
f7f4bccb | 557 | jbd2_journal_file_buffer(descriptor, commit_transaction, |
470decc6 DK |
558 | BJ_LogCtl); |
559 | } | |
560 | ||
561 | /* Where is the buffer to be written? */ | |
562 | ||
f7f4bccb | 563 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
564 | /* If the block mapping failed, just abandon the buffer |
565 | and repeat this loop: we'll fall into the | |
566 | refile-on-abort condition above. */ | |
567 | if (err) { | |
a7fa2baf | 568 | jbd2_journal_abort(journal, err); |
470decc6 DK |
569 | continue; |
570 | } | |
571 | ||
572 | /* | |
573 | * start_this_handle() uses t_outstanding_credits to determine | |
574 | * the free space in the log, but this counter is changed | |
f7f4bccb | 575 | * by jbd2_journal_next_log_block() also. |
470decc6 | 576 | */ |
a51dca9c | 577 | atomic_dec(&commit_transaction->t_outstanding_credits); |
470decc6 DK |
578 | |
579 | /* Bump b_count to prevent truncate from stumbling over | |
580 | the shadowed buffer! @@@ This can go if we ever get | |
581 | rid of the BJ_IO/BJ_Shadow pairing of buffers. */ | |
582 | atomic_inc(&jh2bh(jh)->b_count); | |
583 | ||
584 | /* Make a temporary IO buffer with which to write it out | |
585 | (this will requeue both the metadata buffer and the | |
586 | temporary IO buffer). new_bh goes on BJ_IO*/ | |
587 | ||
588 | set_bit(BH_JWrite, &jh2bh(jh)->b_state); | |
589 | /* | |
f7f4bccb | 590 | * akpm: jbd2_journal_write_metadata_buffer() sets |
470decc6 DK |
591 | * new_bh->b_transaction to commit_transaction. |
592 | * We need to clean this up before we release new_bh | |
593 | * (which is of type BJ_IO) | |
594 | */ | |
595 | JBUFFER_TRACE(jh, "ph3: write metadata"); | |
f7f4bccb | 596 | flags = jbd2_journal_write_metadata_buffer(commit_transaction, |
470decc6 | 597 | jh, &new_jh, blocknr); |
e6ec116b TT |
598 | if (flags < 0) { |
599 | jbd2_journal_abort(journal, flags); | |
600 | continue; | |
601 | } | |
470decc6 DK |
602 | set_bit(BH_JWrite, &jh2bh(new_jh)->b_state); |
603 | wbuf[bufs++] = jh2bh(new_jh); | |
604 | ||
605 | /* Record the new block's tag in the current descriptor | |
606 | buffer */ | |
607 | ||
608 | tag_flag = 0; | |
609 | if (flags & 1) | |
f7f4bccb | 610 | tag_flag |= JBD2_FLAG_ESCAPE; |
470decc6 | 611 | if (!first_tag) |
f7f4bccb | 612 | tag_flag |= JBD2_FLAG_SAME_UUID; |
470decc6 DK |
613 | |
614 | tag = (journal_block_tag_t *) tagp; | |
b517bea1 | 615 | write_tag_block(tag_bytes, tag, jh2bh(jh)->b_blocknr); |
470decc6 | 616 | tag->t_flags = cpu_to_be32(tag_flag); |
b517bea1 ZB |
617 | tagp += tag_bytes; |
618 | space_left -= tag_bytes; | |
470decc6 DK |
619 | |
620 | if (first_tag) { | |
621 | memcpy (tagp, journal->j_uuid, 16); | |
622 | tagp += 16; | |
623 | space_left -= 16; | |
624 | first_tag = 0; | |
625 | } | |
626 | ||
627 | /* If there's no more to do, or if the descriptor is full, | |
628 | let the IO rip! */ | |
629 | ||
630 | if (bufs == journal->j_wbufsize || | |
631 | commit_transaction->t_buffers == NULL || | |
b517bea1 | 632 | space_left < tag_bytes + 16) { |
470decc6 DK |
633 | |
634 | jbd_debug(4, "JBD: Submit %d IOs\n", bufs); | |
635 | ||
636 | /* Write an end-of-descriptor marker before | |
637 | submitting the IOs. "tag" still points to | |
638 | the last tag we set up. */ | |
639 | ||
f7f4bccb | 640 | tag->t_flags |= cpu_to_be32(JBD2_FLAG_LAST_TAG); |
470decc6 DK |
641 | |
642 | start_journal_io: | |
643 | for (i = 0; i < bufs; i++) { | |
644 | struct buffer_head *bh = wbuf[i]; | |
818d276c GS |
645 | /* |
646 | * Compute checksum. | |
647 | */ | |
648 | if (JBD2_HAS_COMPAT_FEATURE(journal, | |
649 | JBD2_FEATURE_COMPAT_CHECKSUM)) { | |
650 | crc32_sum = | |
651 | jbd2_checksum_data(crc32_sum, bh); | |
652 | } | |
653 | ||
470decc6 DK |
654 | lock_buffer(bh); |
655 | clear_buffer_dirty(bh); | |
656 | set_buffer_uptodate(bh); | |
657 | bh->b_end_io = journal_end_buffer_io_sync; | |
82f04ab4 | 658 | submit_bh(WRITE_SYNC, bh); |
470decc6 DK |
659 | } |
660 | cond_resched(); | |
bf699327 | 661 | stats.run.rs_blocks_logged += bufs; |
470decc6 DK |
662 | |
663 | /* Force a new descriptor to be generated next | |
664 | time round the loop. */ | |
665 | descriptor = NULL; | |
666 | bufs = 0; | |
667 | } | |
668 | } | |
669 | ||
f73bee49 JK |
670 | err = journal_finish_inode_data_buffers(journal, commit_transaction); |
671 | if (err) { | |
672 | printk(KERN_WARNING | |
673 | "JBD2: Detected IO errors while flushing file data " | |
674 | "on %s\n", journal->j_devname); | |
675 | if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR) | |
676 | jbd2_journal_abort(journal, err); | |
677 | err = 0; | |
678 | } | |
679 | ||
cc3e1bea TT |
680 | /* |
681 | * If the journal is not located on the file system device, | |
682 | * then we must flush the file system device before we issue | |
683 | * the commit record | |
684 | */ | |
81be12c8 | 685 | if (commit_transaction->t_need_data_flush && |
cc3e1bea TT |
686 | (journal->j_fs_dev != journal->j_dev) && |
687 | (journal->j_flags & JBD2_BARRIER)) | |
dd3932ed | 688 | blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL); |
818d276c | 689 | |
cc3e1bea | 690 | /* Done it all: now write the commit record asynchronously. */ |
818d276c | 691 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, |
0e3d2a63 | 692 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { |
818d276c GS |
693 | err = journal_submit_commit_record(journal, commit_transaction, |
694 | &cbh, crc32_sum); | |
695 | if (err) | |
696 | __jbd2_journal_abort_hard(journal); | |
e9e34f4e | 697 | } |
c851ed54 | 698 | |
82f04ab4 JA |
699 | blk_finish_plug(&plug); |
700 | ||
470decc6 DK |
701 | /* Lo and behold: we have just managed to send a transaction to |
702 | the log. Before we can commit it, wait for the IO so far to | |
703 | complete. Control buffers being written are on the | |
704 | transaction's t_log_list queue, and metadata buffers are on | |
705 | the t_iobuf_list queue. | |
706 | ||
707 | Wait for the buffers in reverse order. That way we are | |
708 | less likely to be woken up until all IOs have completed, and | |
709 | so we incur less scheduling load. | |
710 | */ | |
711 | ||
87c89c23 | 712 | jbd_debug(3, "JBD: commit phase 3\n"); |
470decc6 DK |
713 | |
714 | /* | |
715 | * akpm: these are BJ_IO, and j_list_lock is not needed. | |
716 | * See __journal_try_to_free_buffer. | |
717 | */ | |
718 | wait_for_iobuf: | |
719 | while (commit_transaction->t_iobuf_list != NULL) { | |
720 | struct buffer_head *bh; | |
721 | ||
722 | jh = commit_transaction->t_iobuf_list->b_tprev; | |
723 | bh = jh2bh(jh); | |
724 | if (buffer_locked(bh)) { | |
725 | wait_on_buffer(bh); | |
726 | goto wait_for_iobuf; | |
727 | } | |
728 | if (cond_resched()) | |
729 | goto wait_for_iobuf; | |
730 | ||
731 | if (unlikely(!buffer_uptodate(bh))) | |
732 | err = -EIO; | |
733 | ||
734 | clear_buffer_jwrite(bh); | |
735 | ||
736 | JBUFFER_TRACE(jh, "ph4: unfile after journal write"); | |
f7f4bccb | 737 | jbd2_journal_unfile_buffer(journal, jh); |
470decc6 DK |
738 | |
739 | /* | |
740 | * ->t_iobuf_list should contain only dummy buffer_heads | |
f7f4bccb | 741 | * which were created by jbd2_journal_write_metadata_buffer(). |
470decc6 DK |
742 | */ |
743 | BUFFER_TRACE(bh, "dumping temporary bh"); | |
f7f4bccb | 744 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
745 | __brelse(bh); |
746 | J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0); | |
747 | free_buffer_head(bh); | |
748 | ||
749 | /* We also have to unlock and free the corresponding | |
750 | shadowed buffer */ | |
751 | jh = commit_transaction->t_shadow_list->b_tprev; | |
752 | bh = jh2bh(jh); | |
753 | clear_bit(BH_JWrite, &bh->b_state); | |
754 | J_ASSERT_BH(bh, buffer_jbddirty(bh)); | |
755 | ||
756 | /* The metadata is now released for reuse, but we need | |
757 | to remember it against this transaction so that when | |
758 | we finally commit, we can do any checkpointing | |
759 | required. */ | |
760 | JBUFFER_TRACE(jh, "file as BJ_Forget"); | |
f7f4bccb | 761 | jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget); |
229309ca JK |
762 | /* |
763 | * Wake up any transactions which were waiting for this IO to | |
764 | * complete. The barrier must be here so that changes by | |
765 | * jbd2_journal_file_buffer() take effect before wake_up_bit() | |
766 | * does the waitqueue check. | |
767 | */ | |
768 | smp_mb(); | |
470decc6 DK |
769 | wake_up_bit(&bh->b_state, BH_Unshadow); |
770 | JBUFFER_TRACE(jh, "brelse shadowed buffer"); | |
771 | __brelse(bh); | |
772 | } | |
773 | ||
774 | J_ASSERT (commit_transaction->t_shadow_list == NULL); | |
775 | ||
87c89c23 | 776 | jbd_debug(3, "JBD: commit phase 4\n"); |
470decc6 DK |
777 | |
778 | /* Here we wait for the revoke record and descriptor record buffers */ | |
779 | wait_for_ctlbuf: | |
780 | while (commit_transaction->t_log_list != NULL) { | |
781 | struct buffer_head *bh; | |
782 | ||
783 | jh = commit_transaction->t_log_list->b_tprev; | |
784 | bh = jh2bh(jh); | |
785 | if (buffer_locked(bh)) { | |
786 | wait_on_buffer(bh); | |
787 | goto wait_for_ctlbuf; | |
788 | } | |
789 | if (cond_resched()) | |
790 | goto wait_for_ctlbuf; | |
791 | ||
792 | if (unlikely(!buffer_uptodate(bh))) | |
793 | err = -EIO; | |
794 | ||
795 | BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile"); | |
796 | clear_buffer_jwrite(bh); | |
f7f4bccb MC |
797 | jbd2_journal_unfile_buffer(journal, jh); |
798 | jbd2_journal_put_journal_head(jh); | |
470decc6 DK |
799 | __brelse(bh); /* One for getblk */ |
800 | /* AKPM: bforget here */ | |
801 | } | |
802 | ||
77e841de HK |
803 | if (err) |
804 | jbd2_journal_abort(journal, err); | |
805 | ||
87c89c23 | 806 | jbd_debug(3, "JBD: commit phase 5\n"); |
470decc6 | 807 | |
818d276c | 808 | if (!JBD2_HAS_INCOMPAT_FEATURE(journal, |
0e3d2a63 | 809 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { |
818d276c GS |
810 | err = journal_submit_commit_record(journal, commit_transaction, |
811 | &cbh, crc32_sum); | |
812 | if (err) | |
813 | __jbd2_journal_abort_hard(journal); | |
814 | } | |
6cba611e | 815 | if (cbh) |
fd98496f | 816 | err = journal_wait_on_commit_record(journal, cbh); |
f73bee49 JK |
817 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, |
818 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) && | |
819 | journal->j_flags & JBD2_BARRIER) { | |
dd3932ed | 820 | blkdev_issue_flush(journal->j_dev, GFP_KERNEL, NULL); |
f73bee49 | 821 | } |
470decc6 DK |
822 | |
823 | if (err) | |
a7fa2baf | 824 | jbd2_journal_abort(journal, err); |
470decc6 DK |
825 | |
826 | /* End of a transaction! Finally, we can do checkpoint | |
827 | processing: any buffers committed as a result of this | |
828 | transaction can be removed from any checkpoint list it was on | |
829 | before. */ | |
830 | ||
87c89c23 | 831 | jbd_debug(3, "JBD: commit phase 6\n"); |
470decc6 | 832 | |
c851ed54 | 833 | J_ASSERT(list_empty(&commit_transaction->t_inode_list)); |
470decc6 DK |
834 | J_ASSERT(commit_transaction->t_buffers == NULL); |
835 | J_ASSERT(commit_transaction->t_checkpoint_list == NULL); | |
836 | J_ASSERT(commit_transaction->t_iobuf_list == NULL); | |
837 | J_ASSERT(commit_transaction->t_shadow_list == NULL); | |
838 | J_ASSERT(commit_transaction->t_log_list == NULL); | |
839 | ||
840 | restart_loop: | |
841 | /* | |
842 | * As there are other places (journal_unmap_buffer()) adding buffers | |
843 | * to this list we have to be careful and hold the j_list_lock. | |
844 | */ | |
845 | spin_lock(&journal->j_list_lock); | |
846 | while (commit_transaction->t_forget) { | |
847 | transaction_t *cp_transaction; | |
848 | struct buffer_head *bh; | |
849 | ||
850 | jh = commit_transaction->t_forget; | |
851 | spin_unlock(&journal->j_list_lock); | |
852 | bh = jh2bh(jh); | |
853 | jbd_lock_bh_state(bh); | |
23e2af35 | 854 | J_ASSERT_JH(jh, jh->b_transaction == commit_transaction); |
470decc6 DK |
855 | |
856 | /* | |
857 | * If there is undo-protected committed data against | |
858 | * this buffer, then we can remove it now. If it is a | |
859 | * buffer needing such protection, the old frozen_data | |
860 | * field now points to a committed version of the | |
861 | * buffer, so rotate that field to the new committed | |
862 | * data. | |
863 | * | |
864 | * Otherwise, we can just throw away the frozen data now. | |
e06c8227 JB |
865 | * |
866 | * We also know that the frozen data has already fired | |
867 | * its triggers if they exist, so we can clear that too. | |
470decc6 DK |
868 | */ |
869 | if (jh->b_committed_data) { | |
af1e76d6 | 870 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
871 | jh->b_committed_data = NULL; |
872 | if (jh->b_frozen_data) { | |
873 | jh->b_committed_data = jh->b_frozen_data; | |
874 | jh->b_frozen_data = NULL; | |
e06c8227 | 875 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
876 | } |
877 | } else if (jh->b_frozen_data) { | |
af1e76d6 | 878 | jbd2_free(jh->b_frozen_data, bh->b_size); |
470decc6 | 879 | jh->b_frozen_data = NULL; |
e06c8227 | 880 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
881 | } |
882 | ||
883 | spin_lock(&journal->j_list_lock); | |
884 | cp_transaction = jh->b_cp_transaction; | |
885 | if (cp_transaction) { | |
886 | JBUFFER_TRACE(jh, "remove from old cp transaction"); | |
8e85fb3f | 887 | cp_transaction->t_chp_stats.cs_dropped++; |
f7f4bccb | 888 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
889 | } |
890 | ||
891 | /* Only re-checkpoint the buffer_head if it is marked | |
892 | * dirty. If the buffer was added to the BJ_Forget list | |
f7f4bccb | 893 | * by jbd2_journal_forget, it may no longer be dirty and |
470decc6 DK |
894 | * there's no point in keeping a checkpoint record for |
895 | * it. */ | |
896 | ||
897 | /* A buffer which has been freed while still being | |
898 | * journaled by a previous transaction may end up still | |
899 | * being dirty here, but we want to avoid writing back | |
ba869023 | 900 | * that buffer in the future after the "add to orphan" |
901 | * operation been committed, That's not only a performance | |
902 | * gain, it also stops aliasing problems if the buffer is | |
903 | * left behind for writeback and gets reallocated for another | |
470decc6 | 904 | * use in a different page. */ |
ba869023 | 905 | if (buffer_freed(bh) && !jh->b_next_transaction) { |
470decc6 DK |
906 | clear_buffer_freed(bh); |
907 | clear_buffer_jbddirty(bh); | |
908 | } | |
909 | ||
910 | if (buffer_jbddirty(bh)) { | |
911 | JBUFFER_TRACE(jh, "add to new checkpointing trans"); | |
f7f4bccb | 912 | __jbd2_journal_insert_checkpoint(jh, commit_transaction); |
7ad7445f HK |
913 | if (is_journal_aborted(journal)) |
914 | clear_buffer_jbddirty(bh); | |
470decc6 | 915 | JBUFFER_TRACE(jh, "refile for checkpoint writeback"); |
f7f4bccb | 916 | __jbd2_journal_refile_buffer(jh); |
470decc6 DK |
917 | jbd_unlock_bh_state(bh); |
918 | } else { | |
919 | J_ASSERT_BH(bh, !buffer_dirty(bh)); | |
920 | /* The buffer on BJ_Forget list and not jbddirty means | |
921 | * it has been freed by this transaction and hence it | |
922 | * could not have been reallocated until this | |
923 | * transaction has committed. *BUT* it could be | |
924 | * reallocated once we have written all the data to | |
925 | * disk and before we process the buffer on BJ_Forget | |
926 | * list. */ | |
927 | JBUFFER_TRACE(jh, "refile or unfile freed buffer"); | |
f7f4bccb | 928 | __jbd2_journal_refile_buffer(jh); |
470decc6 DK |
929 | if (!jh->b_transaction) { |
930 | jbd_unlock_bh_state(bh); | |
931 | /* needs a brelse */ | |
f7f4bccb | 932 | jbd2_journal_remove_journal_head(bh); |
470decc6 DK |
933 | release_buffer_page(bh); |
934 | } else | |
935 | jbd_unlock_bh_state(bh); | |
936 | } | |
937 | cond_resched_lock(&journal->j_list_lock); | |
938 | } | |
939 | spin_unlock(&journal->j_list_lock); | |
940 | /* | |
f5a7a6b0 JK |
941 | * This is a bit sleazy. We use j_list_lock to protect transition |
942 | * of a transaction into T_FINISHED state and calling | |
943 | * __jbd2_journal_drop_transaction(). Otherwise we could race with | |
944 | * other checkpointing code processing the transaction... | |
470decc6 | 945 | */ |
a931da6a | 946 | write_lock(&journal->j_state_lock); |
470decc6 DK |
947 | spin_lock(&journal->j_list_lock); |
948 | /* | |
949 | * Now recheck if some buffers did not get attached to the transaction | |
950 | * while the lock was dropped... | |
951 | */ | |
952 | if (commit_transaction->t_forget) { | |
953 | spin_unlock(&journal->j_list_lock); | |
a931da6a | 954 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
955 | goto restart_loop; |
956 | } | |
957 | ||
958 | /* Done with this transaction! */ | |
959 | ||
87c89c23 | 960 | jbd_debug(3, "JBD: commit phase 7\n"); |
470decc6 DK |
961 | |
962 | J_ASSERT(commit_transaction->t_state == T_COMMIT); | |
963 | ||
8e85fb3f | 964 | commit_transaction->t_start = jiffies; |
bf699327 TT |
965 | stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging, |
966 | commit_transaction->t_start); | |
8e85fb3f JL |
967 | |
968 | /* | |
bf699327 | 969 | * File the transaction statistics |
8e85fb3f | 970 | */ |
8e85fb3f | 971 | stats.ts_tid = commit_transaction->t_tid; |
8dd42046 TT |
972 | stats.run.rs_handle_count = |
973 | atomic_read(&commit_transaction->t_handle_count); | |
bf699327 TT |
974 | trace_jbd2_run_stats(journal->j_fs_dev->bd_dev, |
975 | commit_transaction->t_tid, &stats.run); | |
8e85fb3f JL |
976 | |
977 | /* | |
978 | * Calculate overall stats | |
979 | */ | |
bf699327 | 980 | spin_lock(&journal->j_history_lock); |
8e85fb3f | 981 | journal->j_stats.ts_tid++; |
bf699327 TT |
982 | journal->j_stats.run.rs_wait += stats.run.rs_wait; |
983 | journal->j_stats.run.rs_running += stats.run.rs_running; | |
984 | journal->j_stats.run.rs_locked += stats.run.rs_locked; | |
985 | journal->j_stats.run.rs_flushing += stats.run.rs_flushing; | |
986 | journal->j_stats.run.rs_logging += stats.run.rs_logging; | |
987 | journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count; | |
988 | journal->j_stats.run.rs_blocks += stats.run.rs_blocks; | |
989 | journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged; | |
8e85fb3f JL |
990 | spin_unlock(&journal->j_history_lock); |
991 | ||
470decc6 DK |
992 | commit_transaction->t_state = T_FINISHED; |
993 | J_ASSERT(commit_transaction == journal->j_committing_transaction); | |
994 | journal->j_commit_sequence = commit_transaction->t_tid; | |
995 | journal->j_committing_transaction = NULL; | |
e07f7183 | 996 | commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); |
470decc6 | 997 | |
e07f7183 JB |
998 | /* |
999 | * weight the commit time higher than the average time so we don't | |
1000 | * react too strongly to vast changes in the commit time | |
1001 | */ | |
1002 | if (likely(journal->j_average_commit_time)) | |
1003 | journal->j_average_commit_time = (commit_time + | |
1004 | journal->j_average_commit_time*3) / 4; | |
1005 | else | |
1006 | journal->j_average_commit_time = commit_time; | |
a931da6a | 1007 | write_unlock(&journal->j_state_lock); |
6c20ec85 | 1008 | |
f89b7795 JK |
1009 | if (commit_transaction->t_checkpoint_list == NULL && |
1010 | commit_transaction->t_checkpoint_io_list == NULL) { | |
f7f4bccb | 1011 | __jbd2_journal_drop_transaction(journal, commit_transaction); |
fb68407b | 1012 | to_free = 1; |
470decc6 DK |
1013 | } else { |
1014 | if (journal->j_checkpoint_transactions == NULL) { | |
1015 | journal->j_checkpoint_transactions = commit_transaction; | |
1016 | commit_transaction->t_cpnext = commit_transaction; | |
1017 | commit_transaction->t_cpprev = commit_transaction; | |
1018 | } else { | |
1019 | commit_transaction->t_cpnext = | |
1020 | journal->j_checkpoint_transactions; | |
1021 | commit_transaction->t_cpprev = | |
1022 | commit_transaction->t_cpnext->t_cpprev; | |
1023 | commit_transaction->t_cpnext->t_cpprev = | |
1024 | commit_transaction; | |
1025 | commit_transaction->t_cpprev->t_cpnext = | |
1026 | commit_transaction; | |
1027 | } | |
1028 | } | |
1029 | spin_unlock(&journal->j_list_lock); | |
1030 | ||
fb68407b AK |
1031 | if (journal->j_commit_callback) |
1032 | journal->j_commit_callback(journal, commit_transaction); | |
1033 | ||
879c5e6b | 1034 | trace_jbd2_end_commit(journal, commit_transaction); |
470decc6 DK |
1035 | jbd_debug(1, "JBD: commit %d complete, head %d\n", |
1036 | journal->j_commit_sequence, journal->j_tail_sequence); | |
fb68407b AK |
1037 | if (to_free) |
1038 | kfree(commit_transaction); | |
470decc6 DK |
1039 | |
1040 | wake_up(&journal->j_wait_done_commit); | |
1041 | } |