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 | ||
1f56c589 DW |
88 | static void jbd2_commit_block_csum_set(journal_t *j, |
89 | struct journal_head *descriptor) | |
90 | { | |
91 | struct commit_header *h; | |
92 | __u32 csum; | |
93 | ||
94 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
95 | return; | |
96 | ||
97 | h = (struct commit_header *)(jh2bh(descriptor)->b_data); | |
98 | h->h_chksum_type = 0; | |
99 | h->h_chksum_size = 0; | |
100 | h->h_chksum[0] = 0; | |
101 | csum = jbd2_chksum(j, j->j_csum_seed, jh2bh(descriptor)->b_data, | |
102 | j->j_blocksize); | |
103 | h->h_chksum[0] = cpu_to_be32(csum); | |
104 | } | |
105 | ||
818d276c GS |
106 | /* |
107 | * Done it all: now submit the commit record. We should have | |
470decc6 DK |
108 | * cleaned up our previous buffers by now, so if we are in abort |
109 | * mode we can now just skip the rest of the journal write | |
110 | * entirely. | |
111 | * | |
112 | * Returns 1 if the journal needs to be aborted or 0 on success | |
113 | */ | |
818d276c GS |
114 | static int journal_submit_commit_record(journal_t *journal, |
115 | transaction_t *commit_transaction, | |
116 | struct buffer_head **cbh, | |
117 | __u32 crc32_sum) | |
470decc6 DK |
118 | { |
119 | struct journal_head *descriptor; | |
818d276c | 120 | struct commit_header *tmp; |
470decc6 | 121 | struct buffer_head *bh; |
818d276c | 122 | int ret; |
736603ab | 123 | struct timespec now = current_kernel_time(); |
470decc6 | 124 | |
6cba611e ZH |
125 | *cbh = NULL; |
126 | ||
470decc6 DK |
127 | if (is_journal_aborted(journal)) |
128 | return 0; | |
129 | ||
f7f4bccb | 130 | descriptor = jbd2_journal_get_descriptor_buffer(journal); |
470decc6 DK |
131 | if (!descriptor) |
132 | return 1; | |
133 | ||
134 | bh = jh2bh(descriptor); | |
135 | ||
818d276c GS |
136 | tmp = (struct commit_header *)bh->b_data; |
137 | tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); | |
138 | tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK); | |
139 | tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid); | |
736603ab TT |
140 | tmp->h_commit_sec = cpu_to_be64(now.tv_sec); |
141 | tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec); | |
818d276c GS |
142 | |
143 | if (JBD2_HAS_COMPAT_FEATURE(journal, | |
144 | JBD2_FEATURE_COMPAT_CHECKSUM)) { | |
145 | tmp->h_chksum_type = JBD2_CRC32_CHKSUM; | |
146 | tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE; | |
147 | tmp->h_chksum[0] = cpu_to_be32(crc32_sum); | |
470decc6 | 148 | } |
1f56c589 | 149 | jbd2_commit_block_csum_set(journal, descriptor); |
470decc6 | 150 | |
818d276c GS |
151 | JBUFFER_TRACE(descriptor, "submit commit block"); |
152 | lock_buffer(bh); | |
45a90bfd | 153 | clear_buffer_dirty(bh); |
818d276c GS |
154 | set_buffer_uptodate(bh); |
155 | bh->b_end_io = journal_end_buffer_io_sync; | |
156 | ||
157 | if (journal->j_flags & JBD2_BARRIER && | |
0e3d2a63 | 158 | !JBD2_HAS_INCOMPAT_FEATURE(journal, |
9c35575b | 159 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) |
721a9602 | 160 | ret = submit_bh(WRITE_SYNC | WRITE_FLUSH_FUA, bh); |
9c35575b | 161 | else |
721a9602 | 162 | ret = submit_bh(WRITE_SYNC, bh); |
9c35575b | 163 | |
818d276c GS |
164 | *cbh = bh; |
165 | return ret; | |
166 | } | |
167 | ||
168 | /* | |
169 | * This function along with journal_submit_commit_record | |
170 | * allows to write the commit record asynchronously. | |
171 | */ | |
fd98496f TT |
172 | static int journal_wait_on_commit_record(journal_t *journal, |
173 | struct buffer_head *bh) | |
818d276c GS |
174 | { |
175 | int ret = 0; | |
176 | ||
177 | clear_buffer_dirty(bh); | |
178 | wait_on_buffer(bh); | |
470decc6 | 179 | |
818d276c GS |
180 | if (unlikely(!buffer_uptodate(bh))) |
181 | ret = -EIO; | |
182 | put_bh(bh); /* One for getblk() */ | |
183 | jbd2_journal_put_journal_head(bh2jh(bh)); | |
184 | ||
185 | return ret; | |
470decc6 DK |
186 | } |
187 | ||
cd1aac32 AK |
188 | /* |
189 | * write the filemap data using writepage() address_space_operations. | |
190 | * We don't do block allocation here even for delalloc. We don't | |
191 | * use writepages() because with dealyed allocation we may be doing | |
192 | * block allocation in writepages(). | |
193 | */ | |
194 | static int journal_submit_inode_data_buffers(struct address_space *mapping) | |
195 | { | |
196 | int ret; | |
197 | struct writeback_control wbc = { | |
198 | .sync_mode = WB_SYNC_ALL, | |
199 | .nr_to_write = mapping->nrpages * 2, | |
200 | .range_start = 0, | |
201 | .range_end = i_size_read(mapping->host), | |
cd1aac32 AK |
202 | }; |
203 | ||
204 | ret = generic_writepages(mapping, &wbc); | |
205 | return ret; | |
206 | } | |
207 | ||
c851ed54 JK |
208 | /* |
209 | * Submit all the data buffers of inode associated with the transaction to | |
210 | * disk. | |
211 | * | |
212 | * We are in a committing transaction. Therefore no new inode can be added to | |
213 | * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently | |
214 | * operate on from being released while we write out pages. | |
215 | */ | |
cd1aac32 | 216 | static int journal_submit_data_buffers(journal_t *journal, |
c851ed54 JK |
217 | transaction_t *commit_transaction) |
218 | { | |
219 | struct jbd2_inode *jinode; | |
220 | int err, ret = 0; | |
221 | struct address_space *mapping; | |
222 | ||
223 | spin_lock(&journal->j_list_lock); | |
224 | list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) { | |
225 | mapping = jinode->i_vfs_inode->i_mapping; | |
39e3ac25 | 226 | set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
c851ed54 | 227 | spin_unlock(&journal->j_list_lock); |
cd1aac32 AK |
228 | /* |
229 | * submit the inode data buffers. We use writepage | |
230 | * instead of writepages. Because writepages can do | |
231 | * block allocation with delalloc. We need to write | |
232 | * only allocated blocks here. | |
233 | */ | |
879c5e6b | 234 | trace_jbd2_submit_inode_data(jinode->i_vfs_inode); |
cd1aac32 | 235 | err = journal_submit_inode_data_buffers(mapping); |
c851ed54 JK |
236 | if (!ret) |
237 | ret = err; | |
238 | spin_lock(&journal->j_list_lock); | |
239 | J_ASSERT(jinode->i_transaction == commit_transaction); | |
39e3ac25 BK |
240 | clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
241 | smp_mb__after_clear_bit(); | |
c851ed54 JK |
242 | wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING); |
243 | } | |
244 | spin_unlock(&journal->j_list_lock); | |
245 | return ret; | |
246 | } | |
247 | ||
248 | /* | |
249 | * Wait for data submitted for writeout, refile inodes to proper | |
250 | * transaction if needed. | |
251 | * | |
252 | */ | |
253 | static int journal_finish_inode_data_buffers(journal_t *journal, | |
254 | transaction_t *commit_transaction) | |
255 | { | |
256 | struct jbd2_inode *jinode, *next_i; | |
257 | int err, ret = 0; | |
258 | ||
cd1aac32 | 259 | /* For locking, see the comment in journal_submit_data_buffers() */ |
c851ed54 JK |
260 | spin_lock(&journal->j_list_lock); |
261 | list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) { | |
39e3ac25 | 262 | set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
c851ed54 JK |
263 | spin_unlock(&journal->j_list_lock); |
264 | err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping); | |
e9e34f4e HK |
265 | if (err) { |
266 | /* | |
267 | * Because AS_EIO is cleared by | |
94004ed7 | 268 | * filemap_fdatawait_range(), set it again so |
e9e34f4e HK |
269 | * that user process can get -EIO from fsync(). |
270 | */ | |
271 | set_bit(AS_EIO, | |
272 | &jinode->i_vfs_inode->i_mapping->flags); | |
273 | ||
274 | if (!ret) | |
275 | ret = err; | |
276 | } | |
c851ed54 | 277 | spin_lock(&journal->j_list_lock); |
39e3ac25 BK |
278 | clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags); |
279 | smp_mb__after_clear_bit(); | |
c851ed54 JK |
280 | wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING); |
281 | } | |
282 | ||
283 | /* Now refile inode to proper lists */ | |
284 | list_for_each_entry_safe(jinode, next_i, | |
285 | &commit_transaction->t_inode_list, i_list) { | |
286 | list_del(&jinode->i_list); | |
287 | if (jinode->i_next_transaction) { | |
288 | jinode->i_transaction = jinode->i_next_transaction; | |
289 | jinode->i_next_transaction = NULL; | |
290 | list_add(&jinode->i_list, | |
291 | &jinode->i_transaction->t_inode_list); | |
292 | } else { | |
293 | jinode->i_transaction = NULL; | |
294 | } | |
295 | } | |
296 | spin_unlock(&journal->j_list_lock); | |
297 | ||
298 | return ret; | |
299 | } | |
300 | ||
818d276c GS |
301 | static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh) |
302 | { | |
303 | struct page *page = bh->b_page; | |
304 | char *addr; | |
305 | __u32 checksum; | |
306 | ||
303a8f2a | 307 | addr = kmap_atomic(page); |
818d276c GS |
308 | checksum = crc32_be(crc32_sum, |
309 | (void *)(addr + offset_in_page(bh->b_data)), bh->b_size); | |
303a8f2a | 310 | kunmap_atomic(addr); |
818d276c GS |
311 | |
312 | return checksum; | |
313 | } | |
314 | ||
315 | static void write_tag_block(int tag_bytes, journal_block_tag_t *tag, | |
18eba7aa | 316 | unsigned long long block) |
b517bea1 ZB |
317 | { |
318 | tag->t_blocknr = cpu_to_be32(block & (u32)~0); | |
cd02ff0b | 319 | if (tag_bytes > JBD2_TAG_SIZE32) |
b517bea1 ZB |
320 | tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1); |
321 | } | |
322 | ||
3caa487f DW |
323 | static void jbd2_descr_block_csum_set(journal_t *j, |
324 | struct journal_head *descriptor) | |
325 | { | |
326 | struct jbd2_journal_block_tail *tail; | |
327 | __u32 csum; | |
328 | ||
329 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
330 | return; | |
331 | ||
332 | tail = (struct jbd2_journal_block_tail *) | |
333 | (jh2bh(descriptor)->b_data + j->j_blocksize - | |
334 | sizeof(struct jbd2_journal_block_tail)); | |
335 | tail->t_checksum = 0; | |
336 | csum = jbd2_chksum(j, j->j_csum_seed, jh2bh(descriptor)->b_data, | |
337 | j->j_blocksize); | |
338 | tail->t_checksum = cpu_to_be32(csum); | |
339 | } | |
340 | ||
c3900875 DW |
341 | static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag, |
342 | struct buffer_head *bh, __u32 sequence) | |
343 | { | |
344 | struct page *page = bh->b_page; | |
345 | __u8 *addr; | |
eee06c56 | 346 | __u32 csum32; |
c3900875 DW |
347 | |
348 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
349 | return; | |
350 | ||
351 | sequence = cpu_to_be32(sequence); | |
906adea1 | 352 | addr = kmap_atomic(page); |
eee06c56 DW |
353 | csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&sequence, |
354 | sizeof(sequence)); | |
355 | csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data), | |
356 | bh->b_size); | |
906adea1 | 357 | kunmap_atomic(addr); |
c3900875 | 358 | |
eee06c56 DW |
359 | /* We only have space to store the lower 16 bits of the crc32c. */ |
360 | tag->t_checksum = cpu_to_be16(csum32); | |
c3900875 | 361 | } |
470decc6 | 362 | /* |
f7f4bccb | 363 | * jbd2_journal_commit_transaction |
470decc6 DK |
364 | * |
365 | * The primary function for committing a transaction to the log. This | |
366 | * function is called by the journal thread to begin a complete commit. | |
367 | */ | |
f7f4bccb | 368 | void jbd2_journal_commit_transaction(journal_t *journal) |
470decc6 | 369 | { |
8e85fb3f | 370 | struct transaction_stats_s stats; |
470decc6 DK |
371 | transaction_t *commit_transaction; |
372 | struct journal_head *jh, *new_jh, *descriptor; | |
373 | struct buffer_head **wbuf = journal->j_wbuf; | |
374 | int bufs; | |
375 | int flags; | |
376 | int err; | |
18eba7aa | 377 | unsigned long long blocknr; |
e07f7183 JB |
378 | ktime_t start_time; |
379 | u64 commit_time; | |
470decc6 DK |
380 | char *tagp = NULL; |
381 | journal_header_t *header; | |
382 | journal_block_tag_t *tag = NULL; | |
383 | int space_left = 0; | |
384 | int first_tag = 0; | |
385 | int tag_flag; | |
794446c6 | 386 | int i; |
b517bea1 | 387 | int tag_bytes = journal_tag_bytes(journal); |
818d276c GS |
388 | struct buffer_head *cbh = NULL; /* For transactional checksums */ |
389 | __u32 crc32_sum = ~0; | |
82f04ab4 | 390 | struct blk_plug plug; |
3339578f JK |
391 | /* Tail of the journal */ |
392 | unsigned long first_block; | |
393 | tid_t first_tid; | |
394 | int update_tail; | |
3caa487f DW |
395 | int csum_size = 0; |
396 | ||
397 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
398 | csum_size = sizeof(struct jbd2_journal_block_tail); | |
470decc6 DK |
399 | |
400 | /* | |
401 | * First job: lock down the current transaction and wait for | |
402 | * all outstanding updates to complete. | |
403 | */ | |
404 | ||
f7f4bccb MC |
405 | /* Do we need to erase the effects of a prior jbd2_journal_flush? */ |
406 | if (journal->j_flags & JBD2_FLUSHED) { | |
470decc6 | 407 | jbd_debug(3, "super block updated\n"); |
a78bb11d | 408 | mutex_lock(&journal->j_checkpoint_mutex); |
79feb521 JK |
409 | /* |
410 | * We hold j_checkpoint_mutex so tail cannot change under us. | |
411 | * We don't need any special data guarantees for writing sb | |
412 | * since journal is empty and it is ok for write to be | |
413 | * flushed only with transaction commit. | |
414 | */ | |
415 | jbd2_journal_update_sb_log_tail(journal, | |
416 | journal->j_tail_sequence, | |
417 | journal->j_tail, | |
418 | WRITE_SYNC); | |
a78bb11d | 419 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
420 | } else { |
421 | jbd_debug(3, "superblock not updated\n"); | |
422 | } | |
423 | ||
424 | J_ASSERT(journal->j_running_transaction != NULL); | |
425 | J_ASSERT(journal->j_committing_transaction == NULL); | |
426 | ||
427 | commit_transaction = journal->j_running_transaction; | |
428 | J_ASSERT(commit_transaction->t_state == T_RUNNING); | |
429 | ||
879c5e6b | 430 | trace_jbd2_start_commit(journal, commit_transaction); |
f2a44523 | 431 | jbd_debug(1, "JBD2: starting commit of transaction %d\n", |
470decc6 DK |
432 | commit_transaction->t_tid); |
433 | ||
a931da6a | 434 | write_lock(&journal->j_state_lock); |
470decc6 DK |
435 | commit_transaction->t_state = T_LOCKED; |
436 | ||
879c5e6b | 437 | trace_jbd2_commit_locking(journal, commit_transaction); |
bf699327 | 438 | stats.run.rs_wait = commit_transaction->t_max_wait; |
9fff24aa | 439 | stats.run.rs_request_delay = 0; |
bf699327 | 440 | stats.run.rs_locked = jiffies; |
9fff24aa TT |
441 | if (commit_transaction->t_requested) |
442 | stats.run.rs_request_delay = | |
443 | jbd2_time_diff(commit_transaction->t_requested, | |
444 | stats.run.rs_locked); | |
bf699327 TT |
445 | stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start, |
446 | stats.run.rs_locked); | |
8e85fb3f | 447 | |
470decc6 | 448 | spin_lock(&commit_transaction->t_handle_lock); |
a51dca9c | 449 | while (atomic_read(&commit_transaction->t_updates)) { |
470decc6 DK |
450 | DEFINE_WAIT(wait); |
451 | ||
452 | prepare_to_wait(&journal->j_wait_updates, &wait, | |
453 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 454 | if (atomic_read(&commit_transaction->t_updates)) { |
470decc6 | 455 | spin_unlock(&commit_transaction->t_handle_lock); |
a931da6a | 456 | write_unlock(&journal->j_state_lock); |
470decc6 | 457 | schedule(); |
a931da6a | 458 | write_lock(&journal->j_state_lock); |
470decc6 DK |
459 | spin_lock(&commit_transaction->t_handle_lock); |
460 | } | |
461 | finish_wait(&journal->j_wait_updates, &wait); | |
462 | } | |
463 | spin_unlock(&commit_transaction->t_handle_lock); | |
464 | ||
a51dca9c | 465 | J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <= |
470decc6 DK |
466 | journal->j_max_transaction_buffers); |
467 | ||
468 | /* | |
469 | * First thing we are allowed to do is to discard any remaining | |
470 | * BJ_Reserved buffers. Note, it is _not_ permissible to assume | |
471 | * that there are no such buffers: if a large filesystem | |
472 | * operation like a truncate needs to split itself over multiple | |
f7f4bccb | 473 | * transactions, then it may try to do a jbd2_journal_restart() while |
470decc6 DK |
474 | * there are still BJ_Reserved buffers outstanding. These must |
475 | * be released cleanly from the current transaction. | |
476 | * | |
477 | * In this case, the filesystem must still reserve write access | |
478 | * again before modifying the buffer in the new transaction, but | |
479 | * we do not require it to remember exactly which old buffers it | |
480 | * has reserved. This is consistent with the existing behaviour | |
f7f4bccb | 481 | * that multiple jbd2_journal_get_write_access() calls to the same |
25985edc | 482 | * buffer are perfectly permissible. |
470decc6 DK |
483 | */ |
484 | while (commit_transaction->t_reserved_list) { | |
485 | jh = commit_transaction->t_reserved_list; | |
486 | JBUFFER_TRACE(jh, "reserved, unused: refile"); | |
487 | /* | |
f7f4bccb | 488 | * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may |
470decc6 DK |
489 | * leave undo-committed data. |
490 | */ | |
491 | if (jh->b_committed_data) { | |
492 | struct buffer_head *bh = jh2bh(jh); | |
493 | ||
494 | jbd_lock_bh_state(bh); | |
af1e76d6 | 495 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
496 | jh->b_committed_data = NULL; |
497 | jbd_unlock_bh_state(bh); | |
498 | } | |
f7f4bccb | 499 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
500 | } |
501 | ||
502 | /* | |
503 | * Now try to drop any written-back buffers from the journal's | |
504 | * checkpoint lists. We do this *before* commit because it potentially | |
505 | * frees some memory | |
506 | */ | |
507 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 508 | __jbd2_journal_clean_checkpoint_list(journal); |
470decc6 DK |
509 | spin_unlock(&journal->j_list_lock); |
510 | ||
f2a44523 | 511 | jbd_debug(3, "JBD2: commit phase 1\n"); |
470decc6 | 512 | |
1ba37268 YY |
513 | /* |
514 | * Clear revoked flag to reflect there is no revoked buffers | |
515 | * in the next transaction which is going to be started. | |
516 | */ | |
517 | jbd2_clear_buffer_revoked_flags(journal); | |
518 | ||
470decc6 DK |
519 | /* |
520 | * Switch to a new revoke table. | |
521 | */ | |
f7f4bccb | 522 | jbd2_journal_switch_revoke_table(journal); |
470decc6 | 523 | |
879c5e6b | 524 | trace_jbd2_commit_flushing(journal, commit_transaction); |
bf699327 TT |
525 | stats.run.rs_flushing = jiffies; |
526 | stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked, | |
527 | stats.run.rs_flushing); | |
8e85fb3f | 528 | |
470decc6 DK |
529 | commit_transaction->t_state = T_FLUSH; |
530 | journal->j_committing_transaction = commit_transaction; | |
531 | journal->j_running_transaction = NULL; | |
e07f7183 | 532 | start_time = ktime_get(); |
470decc6 DK |
533 | commit_transaction->t_log_start = journal->j_head; |
534 | wake_up(&journal->j_wait_transaction_locked); | |
a931da6a | 535 | write_unlock(&journal->j_state_lock); |
470decc6 | 536 | |
f2a44523 | 537 | jbd_debug(3, "JBD2: commit phase 2\n"); |
470decc6 | 538 | |
470decc6 DK |
539 | /* |
540 | * Now start flushing things to disk, in the order they appear | |
541 | * on the transaction lists. Data blocks go first. | |
542 | */ | |
cd1aac32 | 543 | err = journal_submit_data_buffers(journal, commit_transaction); |
470decc6 | 544 | if (err) |
a7fa2baf | 545 | jbd2_journal_abort(journal, err); |
470decc6 | 546 | |
82f04ab4 | 547 | blk_start_plug(&plug); |
67c457a8 | 548 | jbd2_journal_write_revoke_records(journal, commit_transaction, |
82f04ab4 JA |
549 | WRITE_SYNC); |
550 | blk_finish_plug(&plug); | |
470decc6 | 551 | |
f2a44523 | 552 | jbd_debug(3, "JBD2: commit phase 2\n"); |
470decc6 | 553 | |
470decc6 DK |
554 | /* |
555 | * Way to go: we have now written out all of the data for a | |
556 | * transaction! Now comes the tricky part: we need to write out | |
557 | * metadata. Loop over the transaction's entire buffer list: | |
558 | */ | |
a931da6a | 559 | write_lock(&journal->j_state_lock); |
470decc6 | 560 | commit_transaction->t_state = T_COMMIT; |
a931da6a | 561 | write_unlock(&journal->j_state_lock); |
470decc6 | 562 | |
879c5e6b | 563 | trace_jbd2_commit_logging(journal, commit_transaction); |
bf699327 TT |
564 | stats.run.rs_logging = jiffies; |
565 | stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing, | |
566 | stats.run.rs_logging); | |
a51dca9c TT |
567 | stats.run.rs_blocks = |
568 | atomic_read(&commit_transaction->t_outstanding_credits); | |
bf699327 | 569 | stats.run.rs_blocks_logged = 0; |
8e85fb3f | 570 | |
1dfc3220 | 571 | J_ASSERT(commit_transaction->t_nr_buffers <= |
a51dca9c | 572 | atomic_read(&commit_transaction->t_outstanding_credits)); |
1dfc3220 | 573 | |
87c89c23 | 574 | err = 0; |
470decc6 DK |
575 | descriptor = NULL; |
576 | bufs = 0; | |
82f04ab4 | 577 | blk_start_plug(&plug); |
470decc6 DK |
578 | while (commit_transaction->t_buffers) { |
579 | ||
580 | /* Find the next buffer to be journaled... */ | |
581 | ||
582 | jh = commit_transaction->t_buffers; | |
583 | ||
584 | /* If we're in abort mode, we just un-journal the buffer and | |
7ad7445f | 585 | release it. */ |
470decc6 DK |
586 | |
587 | if (is_journal_aborted(journal)) { | |
7ad7445f | 588 | clear_buffer_jbddirty(jh2bh(jh)); |
470decc6 | 589 | JBUFFER_TRACE(jh, "journal is aborting: refile"); |
e06c8227 JB |
590 | jbd2_buffer_abort_trigger(jh, |
591 | jh->b_frozen_data ? | |
592 | jh->b_frozen_triggers : | |
593 | jh->b_triggers); | |
f7f4bccb | 594 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
595 | /* If that was the last one, we need to clean up |
596 | * any descriptor buffers which may have been | |
597 | * already allocated, even if we are now | |
598 | * aborting. */ | |
599 | if (!commit_transaction->t_buffers) | |
600 | goto start_journal_io; | |
601 | continue; | |
602 | } | |
603 | ||
604 | /* Make sure we have a descriptor block in which to | |
605 | record the metadata buffer. */ | |
606 | ||
607 | if (!descriptor) { | |
608 | struct buffer_head *bh; | |
609 | ||
610 | J_ASSERT (bufs == 0); | |
611 | ||
f2a44523 | 612 | jbd_debug(4, "JBD2: get descriptor\n"); |
470decc6 | 613 | |
f7f4bccb | 614 | descriptor = jbd2_journal_get_descriptor_buffer(journal); |
470decc6 | 615 | if (!descriptor) { |
a7fa2baf | 616 | jbd2_journal_abort(journal, -EIO); |
470decc6 DK |
617 | continue; |
618 | } | |
619 | ||
620 | bh = jh2bh(descriptor); | |
f2a44523 | 621 | jbd_debug(4, "JBD2: got buffer %llu (%p)\n", |
470decc6 DK |
622 | (unsigned long long)bh->b_blocknr, bh->b_data); |
623 | header = (journal_header_t *)&bh->b_data[0]; | |
f7f4bccb MC |
624 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); |
625 | header->h_blocktype = cpu_to_be32(JBD2_DESCRIPTOR_BLOCK); | |
470decc6 DK |
626 | header->h_sequence = cpu_to_be32(commit_transaction->t_tid); |
627 | ||
628 | tagp = &bh->b_data[sizeof(journal_header_t)]; | |
629 | space_left = bh->b_size - sizeof(journal_header_t); | |
630 | first_tag = 1; | |
631 | set_buffer_jwrite(bh); | |
632 | set_buffer_dirty(bh); | |
633 | wbuf[bufs++] = bh; | |
634 | ||
635 | /* Record it so that we can wait for IO | |
636 | completion later */ | |
637 | BUFFER_TRACE(bh, "ph3: file as descriptor"); | |
f7f4bccb | 638 | jbd2_journal_file_buffer(descriptor, commit_transaction, |
470decc6 DK |
639 | BJ_LogCtl); |
640 | } | |
641 | ||
642 | /* Where is the buffer to be written? */ | |
643 | ||
f7f4bccb | 644 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
645 | /* If the block mapping failed, just abandon the buffer |
646 | and repeat this loop: we'll fall into the | |
647 | refile-on-abort condition above. */ | |
648 | if (err) { | |
a7fa2baf | 649 | jbd2_journal_abort(journal, err); |
470decc6 DK |
650 | continue; |
651 | } | |
652 | ||
653 | /* | |
654 | * start_this_handle() uses t_outstanding_credits to determine | |
655 | * the free space in the log, but this counter is changed | |
f7f4bccb | 656 | * by jbd2_journal_next_log_block() also. |
470decc6 | 657 | */ |
a51dca9c | 658 | atomic_dec(&commit_transaction->t_outstanding_credits); |
470decc6 DK |
659 | |
660 | /* Bump b_count to prevent truncate from stumbling over | |
661 | the shadowed buffer! @@@ This can go if we ever get | |
662 | rid of the BJ_IO/BJ_Shadow pairing of buffers. */ | |
663 | atomic_inc(&jh2bh(jh)->b_count); | |
664 | ||
665 | /* Make a temporary IO buffer with which to write it out | |
666 | (this will requeue both the metadata buffer and the | |
667 | temporary IO buffer). new_bh goes on BJ_IO*/ | |
668 | ||
669 | set_bit(BH_JWrite, &jh2bh(jh)->b_state); | |
670 | /* | |
f7f4bccb | 671 | * akpm: jbd2_journal_write_metadata_buffer() sets |
470decc6 DK |
672 | * new_bh->b_transaction to commit_transaction. |
673 | * We need to clean this up before we release new_bh | |
674 | * (which is of type BJ_IO) | |
675 | */ | |
676 | JBUFFER_TRACE(jh, "ph3: write metadata"); | |
f7f4bccb | 677 | flags = jbd2_journal_write_metadata_buffer(commit_transaction, |
470decc6 | 678 | jh, &new_jh, blocknr); |
e6ec116b TT |
679 | if (flags < 0) { |
680 | jbd2_journal_abort(journal, flags); | |
681 | continue; | |
682 | } | |
470decc6 DK |
683 | set_bit(BH_JWrite, &jh2bh(new_jh)->b_state); |
684 | wbuf[bufs++] = jh2bh(new_jh); | |
685 | ||
686 | /* Record the new block's tag in the current descriptor | |
687 | buffer */ | |
688 | ||
689 | tag_flag = 0; | |
690 | if (flags & 1) | |
f7f4bccb | 691 | tag_flag |= JBD2_FLAG_ESCAPE; |
470decc6 | 692 | if (!first_tag) |
f7f4bccb | 693 | tag_flag |= JBD2_FLAG_SAME_UUID; |
470decc6 DK |
694 | |
695 | tag = (journal_block_tag_t *) tagp; | |
b517bea1 | 696 | write_tag_block(tag_bytes, tag, jh2bh(jh)->b_blocknr); |
8f888ef8 | 697 | tag->t_flags = cpu_to_be16(tag_flag); |
c3900875 DW |
698 | jbd2_block_tag_csum_set(journal, tag, jh2bh(new_jh), |
699 | commit_transaction->t_tid); | |
b517bea1 ZB |
700 | tagp += tag_bytes; |
701 | space_left -= tag_bytes; | |
470decc6 DK |
702 | |
703 | if (first_tag) { | |
704 | memcpy (tagp, journal->j_uuid, 16); | |
705 | tagp += 16; | |
706 | space_left -= 16; | |
707 | first_tag = 0; | |
708 | } | |
709 | ||
710 | /* If there's no more to do, or if the descriptor is full, | |
711 | let the IO rip! */ | |
712 | ||
713 | if (bufs == journal->j_wbufsize || | |
714 | commit_transaction->t_buffers == NULL || | |
3caa487f | 715 | space_left < tag_bytes + 16 + csum_size) { |
470decc6 | 716 | |
f2a44523 | 717 | jbd_debug(4, "JBD2: Submit %d IOs\n", bufs); |
470decc6 DK |
718 | |
719 | /* Write an end-of-descriptor marker before | |
720 | submitting the IOs. "tag" still points to | |
721 | the last tag we set up. */ | |
722 | ||
8f888ef8 | 723 | tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG); |
470decc6 | 724 | |
3caa487f | 725 | jbd2_descr_block_csum_set(journal, descriptor); |
470decc6 DK |
726 | start_journal_io: |
727 | for (i = 0; i < bufs; i++) { | |
728 | struct buffer_head *bh = wbuf[i]; | |
818d276c GS |
729 | /* |
730 | * Compute checksum. | |
731 | */ | |
732 | if (JBD2_HAS_COMPAT_FEATURE(journal, | |
733 | JBD2_FEATURE_COMPAT_CHECKSUM)) { | |
734 | crc32_sum = | |
735 | jbd2_checksum_data(crc32_sum, bh); | |
736 | } | |
737 | ||
470decc6 DK |
738 | lock_buffer(bh); |
739 | clear_buffer_dirty(bh); | |
740 | set_buffer_uptodate(bh); | |
741 | bh->b_end_io = journal_end_buffer_io_sync; | |
82f04ab4 | 742 | submit_bh(WRITE_SYNC, bh); |
470decc6 DK |
743 | } |
744 | cond_resched(); | |
bf699327 | 745 | stats.run.rs_blocks_logged += bufs; |
470decc6 DK |
746 | |
747 | /* Force a new descriptor to be generated next | |
748 | time round the loop. */ | |
749 | descriptor = NULL; | |
750 | bufs = 0; | |
751 | } | |
752 | } | |
753 | ||
f73bee49 JK |
754 | err = journal_finish_inode_data_buffers(journal, commit_transaction); |
755 | if (err) { | |
756 | printk(KERN_WARNING | |
757 | "JBD2: Detected IO errors while flushing file data " | |
758 | "on %s\n", journal->j_devname); | |
759 | if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR) | |
760 | jbd2_journal_abort(journal, err); | |
761 | err = 0; | |
762 | } | |
763 | ||
3339578f JK |
764 | /* |
765 | * Get current oldest transaction in the log before we issue flush | |
766 | * to the filesystem device. After the flush we can be sure that | |
767 | * blocks of all older transactions are checkpointed to persistent | |
768 | * storage and we will be safe to update journal start in the | |
769 | * superblock with the numbers we get here. | |
770 | */ | |
771 | update_tail = | |
772 | jbd2_journal_get_log_tail(journal, &first_tid, &first_block); | |
773 | ||
bbd2be36 | 774 | write_lock(&journal->j_state_lock); |
3339578f JK |
775 | if (update_tail) { |
776 | long freed = first_block - journal->j_tail; | |
777 | ||
778 | if (first_block < journal->j_tail) | |
779 | freed += journal->j_last - journal->j_first; | |
780 | /* Update tail only if we free significant amount of space */ | |
781 | if (freed < journal->j_maxlen / 4) | |
782 | update_tail = 0; | |
783 | } | |
bbd2be36 JK |
784 | J_ASSERT(commit_transaction->t_state == T_COMMIT); |
785 | commit_transaction->t_state = T_COMMIT_DFLUSH; | |
786 | write_unlock(&journal->j_state_lock); | |
3339578f | 787 | |
cc3e1bea TT |
788 | /* |
789 | * If the journal is not located on the file system device, | |
790 | * then we must flush the file system device before we issue | |
791 | * the commit record | |
792 | */ | |
81be12c8 | 793 | if (commit_transaction->t_need_data_flush && |
cc3e1bea TT |
794 | (journal->j_fs_dev != journal->j_dev) && |
795 | (journal->j_flags & JBD2_BARRIER)) | |
99aa7846 | 796 | blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL); |
818d276c | 797 | |
cc3e1bea | 798 | /* Done it all: now write the commit record asynchronously. */ |
818d276c | 799 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, |
0e3d2a63 | 800 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { |
818d276c GS |
801 | err = journal_submit_commit_record(journal, commit_transaction, |
802 | &cbh, crc32_sum); | |
803 | if (err) | |
804 | __jbd2_journal_abort_hard(journal); | |
e9e34f4e | 805 | } |
c851ed54 | 806 | |
82f04ab4 JA |
807 | blk_finish_plug(&plug); |
808 | ||
470decc6 DK |
809 | /* Lo and behold: we have just managed to send a transaction to |
810 | the log. Before we can commit it, wait for the IO so far to | |
811 | complete. Control buffers being written are on the | |
812 | transaction's t_log_list queue, and metadata buffers are on | |
813 | the t_iobuf_list queue. | |
814 | ||
815 | Wait for the buffers in reverse order. That way we are | |
816 | less likely to be woken up until all IOs have completed, and | |
817 | so we incur less scheduling load. | |
818 | */ | |
819 | ||
f2a44523 | 820 | jbd_debug(3, "JBD2: commit phase 3\n"); |
470decc6 DK |
821 | |
822 | /* | |
823 | * akpm: these are BJ_IO, and j_list_lock is not needed. | |
824 | * See __journal_try_to_free_buffer. | |
825 | */ | |
826 | wait_for_iobuf: | |
827 | while (commit_transaction->t_iobuf_list != NULL) { | |
828 | struct buffer_head *bh; | |
829 | ||
830 | jh = commit_transaction->t_iobuf_list->b_tprev; | |
831 | bh = jh2bh(jh); | |
832 | if (buffer_locked(bh)) { | |
833 | wait_on_buffer(bh); | |
834 | goto wait_for_iobuf; | |
835 | } | |
836 | if (cond_resched()) | |
837 | goto wait_for_iobuf; | |
838 | ||
839 | if (unlikely(!buffer_uptodate(bh))) | |
840 | err = -EIO; | |
841 | ||
842 | clear_buffer_jwrite(bh); | |
843 | ||
844 | JBUFFER_TRACE(jh, "ph4: unfile after journal write"); | |
f7f4bccb | 845 | jbd2_journal_unfile_buffer(journal, jh); |
470decc6 DK |
846 | |
847 | /* | |
848 | * ->t_iobuf_list should contain only dummy buffer_heads | |
f7f4bccb | 849 | * which were created by jbd2_journal_write_metadata_buffer(). |
470decc6 DK |
850 | */ |
851 | BUFFER_TRACE(bh, "dumping temporary bh"); | |
f7f4bccb | 852 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
853 | __brelse(bh); |
854 | J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0); | |
855 | free_buffer_head(bh); | |
856 | ||
857 | /* We also have to unlock and free the corresponding | |
858 | shadowed buffer */ | |
859 | jh = commit_transaction->t_shadow_list->b_tprev; | |
860 | bh = jh2bh(jh); | |
861 | clear_bit(BH_JWrite, &bh->b_state); | |
862 | J_ASSERT_BH(bh, buffer_jbddirty(bh)); | |
863 | ||
864 | /* The metadata is now released for reuse, but we need | |
865 | to remember it against this transaction so that when | |
866 | we finally commit, we can do any checkpointing | |
867 | required. */ | |
868 | JBUFFER_TRACE(jh, "file as BJ_Forget"); | |
f7f4bccb | 869 | jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget); |
229309ca JK |
870 | /* |
871 | * Wake up any transactions which were waiting for this IO to | |
872 | * complete. The barrier must be here so that changes by | |
873 | * jbd2_journal_file_buffer() take effect before wake_up_bit() | |
874 | * does the waitqueue check. | |
875 | */ | |
876 | smp_mb(); | |
470decc6 DK |
877 | wake_up_bit(&bh->b_state, BH_Unshadow); |
878 | JBUFFER_TRACE(jh, "brelse shadowed buffer"); | |
879 | __brelse(bh); | |
880 | } | |
881 | ||
882 | J_ASSERT (commit_transaction->t_shadow_list == NULL); | |
883 | ||
f2a44523 | 884 | jbd_debug(3, "JBD2: commit phase 4\n"); |
470decc6 DK |
885 | |
886 | /* Here we wait for the revoke record and descriptor record buffers */ | |
887 | wait_for_ctlbuf: | |
888 | while (commit_transaction->t_log_list != NULL) { | |
889 | struct buffer_head *bh; | |
890 | ||
891 | jh = commit_transaction->t_log_list->b_tprev; | |
892 | bh = jh2bh(jh); | |
893 | if (buffer_locked(bh)) { | |
894 | wait_on_buffer(bh); | |
895 | goto wait_for_ctlbuf; | |
896 | } | |
897 | if (cond_resched()) | |
898 | goto wait_for_ctlbuf; | |
899 | ||
900 | if (unlikely(!buffer_uptodate(bh))) | |
901 | err = -EIO; | |
902 | ||
903 | BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile"); | |
904 | clear_buffer_jwrite(bh); | |
f7f4bccb MC |
905 | jbd2_journal_unfile_buffer(journal, jh); |
906 | jbd2_journal_put_journal_head(jh); | |
470decc6 DK |
907 | __brelse(bh); /* One for getblk */ |
908 | /* AKPM: bforget here */ | |
909 | } | |
910 | ||
77e841de HK |
911 | if (err) |
912 | jbd2_journal_abort(journal, err); | |
913 | ||
f2a44523 | 914 | jbd_debug(3, "JBD2: commit phase 5\n"); |
bbd2be36 JK |
915 | write_lock(&journal->j_state_lock); |
916 | J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH); | |
917 | commit_transaction->t_state = T_COMMIT_JFLUSH; | |
918 | write_unlock(&journal->j_state_lock); | |
470decc6 | 919 | |
818d276c | 920 | if (!JBD2_HAS_INCOMPAT_FEATURE(journal, |
0e3d2a63 | 921 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { |
818d276c GS |
922 | err = journal_submit_commit_record(journal, commit_transaction, |
923 | &cbh, crc32_sum); | |
924 | if (err) | |
925 | __jbd2_journal_abort_hard(journal); | |
926 | } | |
6cba611e | 927 | if (cbh) |
fd98496f | 928 | err = journal_wait_on_commit_record(journal, cbh); |
f73bee49 JK |
929 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, |
930 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) && | |
931 | journal->j_flags & JBD2_BARRIER) { | |
99aa7846 | 932 | blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL); |
f73bee49 | 933 | } |
470decc6 DK |
934 | |
935 | if (err) | |
a7fa2baf | 936 | jbd2_journal_abort(journal, err); |
470decc6 | 937 | |
3339578f JK |
938 | /* |
939 | * Now disk caches for filesystem device are flushed so we are safe to | |
940 | * erase checkpointed transactions from the log by updating journal | |
941 | * superblock. | |
942 | */ | |
943 | if (update_tail) | |
944 | jbd2_update_log_tail(journal, first_tid, first_block); | |
945 | ||
470decc6 DK |
946 | /* End of a transaction! Finally, we can do checkpoint |
947 | processing: any buffers committed as a result of this | |
948 | transaction can be removed from any checkpoint list it was on | |
949 | before. */ | |
950 | ||
f2a44523 | 951 | jbd_debug(3, "JBD2: commit phase 6\n"); |
470decc6 | 952 | |
c851ed54 | 953 | J_ASSERT(list_empty(&commit_transaction->t_inode_list)); |
470decc6 DK |
954 | J_ASSERT(commit_transaction->t_buffers == NULL); |
955 | J_ASSERT(commit_transaction->t_checkpoint_list == NULL); | |
956 | J_ASSERT(commit_transaction->t_iobuf_list == NULL); | |
957 | J_ASSERT(commit_transaction->t_shadow_list == NULL); | |
958 | J_ASSERT(commit_transaction->t_log_list == NULL); | |
959 | ||
960 | restart_loop: | |
961 | /* | |
962 | * As there are other places (journal_unmap_buffer()) adding buffers | |
963 | * to this list we have to be careful and hold the j_list_lock. | |
964 | */ | |
965 | spin_lock(&journal->j_list_lock); | |
966 | while (commit_transaction->t_forget) { | |
967 | transaction_t *cp_transaction; | |
968 | struct buffer_head *bh; | |
de1b7941 | 969 | int try_to_free = 0; |
470decc6 DK |
970 | |
971 | jh = commit_transaction->t_forget; | |
972 | spin_unlock(&journal->j_list_lock); | |
973 | bh = jh2bh(jh); | |
de1b7941 JK |
974 | /* |
975 | * Get a reference so that bh cannot be freed before we are | |
976 | * done with it. | |
977 | */ | |
978 | get_bh(bh); | |
470decc6 | 979 | jbd_lock_bh_state(bh); |
23e2af35 | 980 | J_ASSERT_JH(jh, jh->b_transaction == commit_transaction); |
470decc6 DK |
981 | |
982 | /* | |
983 | * If there is undo-protected committed data against | |
984 | * this buffer, then we can remove it now. If it is a | |
985 | * buffer needing such protection, the old frozen_data | |
986 | * field now points to a committed version of the | |
987 | * buffer, so rotate that field to the new committed | |
988 | * data. | |
989 | * | |
990 | * Otherwise, we can just throw away the frozen data now. | |
e06c8227 JB |
991 | * |
992 | * We also know that the frozen data has already fired | |
993 | * its triggers if they exist, so we can clear that too. | |
470decc6 DK |
994 | */ |
995 | if (jh->b_committed_data) { | |
af1e76d6 | 996 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
997 | jh->b_committed_data = NULL; |
998 | if (jh->b_frozen_data) { | |
999 | jh->b_committed_data = jh->b_frozen_data; | |
1000 | jh->b_frozen_data = NULL; | |
e06c8227 | 1001 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
1002 | } |
1003 | } else if (jh->b_frozen_data) { | |
af1e76d6 | 1004 | jbd2_free(jh->b_frozen_data, bh->b_size); |
470decc6 | 1005 | jh->b_frozen_data = NULL; |
e06c8227 | 1006 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
1007 | } |
1008 | ||
1009 | spin_lock(&journal->j_list_lock); | |
1010 | cp_transaction = jh->b_cp_transaction; | |
1011 | if (cp_transaction) { | |
1012 | JBUFFER_TRACE(jh, "remove from old cp transaction"); | |
8e85fb3f | 1013 | cp_transaction->t_chp_stats.cs_dropped++; |
f7f4bccb | 1014 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
1015 | } |
1016 | ||
1017 | /* Only re-checkpoint the buffer_head if it is marked | |
1018 | * dirty. If the buffer was added to the BJ_Forget list | |
f7f4bccb | 1019 | * by jbd2_journal_forget, it may no longer be dirty and |
470decc6 DK |
1020 | * there's no point in keeping a checkpoint record for |
1021 | * it. */ | |
1022 | ||
b794e7a6 JK |
1023 | /* |
1024 | * A buffer which has been freed while still being journaled by | |
1025 | * a previous transaction. | |
1026 | */ | |
1027 | if (buffer_freed(bh)) { | |
1028 | /* | |
1029 | * If the running transaction is the one containing | |
1030 | * "add to orphan" operation (b_next_transaction != | |
1031 | * NULL), we have to wait for that transaction to | |
1032 | * commit before we can really get rid of the buffer. | |
1033 | * So just clear b_modified to not confuse transaction | |
1034 | * credit accounting and refile the buffer to | |
1035 | * BJ_Forget of the running transaction. If the just | |
1036 | * committed transaction contains "add to orphan" | |
1037 | * operation, we can completely invalidate the buffer | |
1038 | * now. We are rather through in that since the | |
1039 | * buffer may be still accessible when blocksize < | |
1040 | * pagesize and it is attached to the last partial | |
1041 | * page. | |
1042 | */ | |
1043 | jh->b_modified = 0; | |
1044 | if (!jh->b_next_transaction) { | |
1045 | clear_buffer_freed(bh); | |
1046 | clear_buffer_jbddirty(bh); | |
1047 | clear_buffer_mapped(bh); | |
1048 | clear_buffer_new(bh); | |
1049 | clear_buffer_req(bh); | |
1050 | bh->b_bdev = NULL; | |
1051 | } | |
470decc6 DK |
1052 | } |
1053 | ||
1054 | if (buffer_jbddirty(bh)) { | |
1055 | JBUFFER_TRACE(jh, "add to new checkpointing trans"); | |
f7f4bccb | 1056 | __jbd2_journal_insert_checkpoint(jh, commit_transaction); |
7ad7445f HK |
1057 | if (is_journal_aborted(journal)) |
1058 | clear_buffer_jbddirty(bh); | |
470decc6 DK |
1059 | } else { |
1060 | J_ASSERT_BH(bh, !buffer_dirty(bh)); | |
de1b7941 JK |
1061 | /* |
1062 | * The buffer on BJ_Forget list and not jbddirty means | |
470decc6 DK |
1063 | * it has been freed by this transaction and hence it |
1064 | * could not have been reallocated until this | |
1065 | * transaction has committed. *BUT* it could be | |
1066 | * reallocated once we have written all the data to | |
1067 | * disk and before we process the buffer on BJ_Forget | |
de1b7941 JK |
1068 | * list. |
1069 | */ | |
1070 | if (!jh->b_next_transaction) | |
1071 | try_to_free = 1; | |
470decc6 | 1072 | } |
de1b7941 JK |
1073 | JBUFFER_TRACE(jh, "refile or unfile buffer"); |
1074 | __jbd2_journal_refile_buffer(jh); | |
1075 | jbd_unlock_bh_state(bh); | |
1076 | if (try_to_free) | |
1077 | release_buffer_page(bh); /* Drops bh reference */ | |
1078 | else | |
1079 | __brelse(bh); | |
470decc6 DK |
1080 | cond_resched_lock(&journal->j_list_lock); |
1081 | } | |
1082 | spin_unlock(&journal->j_list_lock); | |
1083 | /* | |
f5a7a6b0 JK |
1084 | * This is a bit sleazy. We use j_list_lock to protect transition |
1085 | * of a transaction into T_FINISHED state and calling | |
1086 | * __jbd2_journal_drop_transaction(). Otherwise we could race with | |
1087 | * other checkpointing code processing the transaction... | |
470decc6 | 1088 | */ |
a931da6a | 1089 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1090 | spin_lock(&journal->j_list_lock); |
1091 | /* | |
1092 | * Now recheck if some buffers did not get attached to the transaction | |
1093 | * while the lock was dropped... | |
1094 | */ | |
1095 | if (commit_transaction->t_forget) { | |
1096 | spin_unlock(&journal->j_list_lock); | |
a931da6a | 1097 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1098 | goto restart_loop; |
1099 | } | |
1100 | ||
1101 | /* Done with this transaction! */ | |
1102 | ||
f2a44523 | 1103 | jbd_debug(3, "JBD2: commit phase 7\n"); |
470decc6 | 1104 | |
bbd2be36 | 1105 | J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH); |
470decc6 | 1106 | |
8e85fb3f | 1107 | commit_transaction->t_start = jiffies; |
bf699327 TT |
1108 | stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging, |
1109 | commit_transaction->t_start); | |
8e85fb3f JL |
1110 | |
1111 | /* | |
bf699327 | 1112 | * File the transaction statistics |
8e85fb3f | 1113 | */ |
8e85fb3f | 1114 | stats.ts_tid = commit_transaction->t_tid; |
8dd42046 TT |
1115 | stats.run.rs_handle_count = |
1116 | atomic_read(&commit_transaction->t_handle_count); | |
bf699327 TT |
1117 | trace_jbd2_run_stats(journal->j_fs_dev->bd_dev, |
1118 | commit_transaction->t_tid, &stats.run); | |
8e85fb3f JL |
1119 | |
1120 | /* | |
1121 | * Calculate overall stats | |
1122 | */ | |
bf699327 | 1123 | spin_lock(&journal->j_history_lock); |
8e85fb3f | 1124 | journal->j_stats.ts_tid++; |
9fff24aa TT |
1125 | if (commit_transaction->t_requested) |
1126 | journal->j_stats.ts_requested++; | |
bf699327 | 1127 | journal->j_stats.run.rs_wait += stats.run.rs_wait; |
9fff24aa | 1128 | journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay; |
bf699327 TT |
1129 | journal->j_stats.run.rs_running += stats.run.rs_running; |
1130 | journal->j_stats.run.rs_locked += stats.run.rs_locked; | |
1131 | journal->j_stats.run.rs_flushing += stats.run.rs_flushing; | |
1132 | journal->j_stats.run.rs_logging += stats.run.rs_logging; | |
1133 | journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count; | |
1134 | journal->j_stats.run.rs_blocks += stats.run.rs_blocks; | |
1135 | journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged; | |
8e85fb3f JL |
1136 | spin_unlock(&journal->j_history_lock); |
1137 | ||
794446c6 | 1138 | commit_transaction->t_state = T_COMMIT_CALLBACK; |
470decc6 DK |
1139 | J_ASSERT(commit_transaction == journal->j_committing_transaction); |
1140 | journal->j_commit_sequence = commit_transaction->t_tid; | |
1141 | journal->j_committing_transaction = NULL; | |
e07f7183 | 1142 | commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); |
470decc6 | 1143 | |
e07f7183 JB |
1144 | /* |
1145 | * weight the commit time higher than the average time so we don't | |
1146 | * react too strongly to vast changes in the commit time | |
1147 | */ | |
1148 | if (likely(journal->j_average_commit_time)) | |
1149 | journal->j_average_commit_time = (commit_time + | |
1150 | journal->j_average_commit_time*3) / 4; | |
1151 | else | |
1152 | journal->j_average_commit_time = commit_time; | |
794446c6 | 1153 | |
a931da6a | 1154 | write_unlock(&journal->j_state_lock); |
6c20ec85 | 1155 | |
794446c6 DM |
1156 | if (journal->j_checkpoint_transactions == NULL) { |
1157 | journal->j_checkpoint_transactions = commit_transaction; | |
1158 | commit_transaction->t_cpnext = commit_transaction; | |
1159 | commit_transaction->t_cpprev = commit_transaction; | |
470decc6 | 1160 | } else { |
794446c6 DM |
1161 | commit_transaction->t_cpnext = |
1162 | journal->j_checkpoint_transactions; | |
1163 | commit_transaction->t_cpprev = | |
1164 | commit_transaction->t_cpnext->t_cpprev; | |
1165 | commit_transaction->t_cpnext->t_cpprev = | |
1166 | commit_transaction; | |
1167 | commit_transaction->t_cpprev->t_cpnext = | |
470decc6 | 1168 | commit_transaction; |
470decc6 DK |
1169 | } |
1170 | spin_unlock(&journal->j_list_lock); | |
794446c6 DM |
1171 | /* Drop all spin_locks because commit_callback may be block. |
1172 | * __journal_remove_checkpoint() can not destroy transaction | |
1173 | * under us because it is not marked as T_FINISHED yet */ | |
fb68407b AK |
1174 | if (journal->j_commit_callback) |
1175 | journal->j_commit_callback(journal, commit_transaction); | |
1176 | ||
879c5e6b | 1177 | trace_jbd2_end_commit(journal, commit_transaction); |
f2a44523 | 1178 | jbd_debug(1, "JBD2: commit %d complete, head %d\n", |
470decc6 DK |
1179 | journal->j_commit_sequence, journal->j_tail_sequence); |
1180 | ||
794446c6 DM |
1181 | write_lock(&journal->j_state_lock); |
1182 | spin_lock(&journal->j_list_lock); | |
1183 | commit_transaction->t_state = T_FINISHED; | |
1184 | /* Recheck checkpoint lists after j_list_lock was dropped */ | |
1185 | if (commit_transaction->t_checkpoint_list == NULL && | |
1186 | commit_transaction->t_checkpoint_io_list == NULL) { | |
1187 | __jbd2_journal_drop_transaction(journal, commit_transaction); | |
1188 | jbd2_journal_free_transaction(commit_transaction); | |
1189 | } | |
1190 | spin_unlock(&journal->j_list_lock); | |
1191 | write_unlock(&journal->j_state_lock); | |
470decc6 DK |
1192 | wake_up(&journal->j_wait_done_commit); |
1193 | } |