]>
Commit | Line | Data |
---|---|---|
f5166768 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
470decc6 | 2 | /* |
f7f4bccb | 3 | * linux/fs/jbd2/commit.c |
470decc6 DK |
4 | * |
5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 | |
6 | * | |
7 | * Copyright 1998 Red Hat corp --- All Rights Reserved | |
8 | * | |
470decc6 DK |
9 | * Journal commit routines for the generic filesystem journaling code; |
10 | * part of the ext2fs journaling system. | |
11 | */ | |
12 | ||
13 | #include <linux/time.h> | |
14 | #include <linux/fs.h> | |
f7f4bccb | 15 | #include <linux/jbd2.h> |
470decc6 DK |
16 | #include <linux/errno.h> |
17 | #include <linux/slab.h> | |
18 | #include <linux/mm.h> | |
19 | #include <linux/pagemap.h> | |
8e85fb3f | 20 | #include <linux/jiffies.h> |
818d276c | 21 | #include <linux/crc32.h> |
cd1aac32 AK |
22 | #include <linux/writeback.h> |
23 | #include <linux/backing-dev.h> | |
fd98496f | 24 | #include <linux/bio.h> |
0e3d2a63 | 25 | #include <linux/blkdev.h> |
39e3ac25 | 26 | #include <linux/bitops.h> |
879c5e6b | 27 | #include <trace/events/jbd2.h> |
470decc6 DK |
28 | |
29 | /* | |
b34090e5 | 30 | * IO end handler for temporary buffer_heads handling writes to the journal. |
470decc6 DK |
31 | */ |
32 | static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate) | |
33 | { | |
b34090e5 JK |
34 | struct buffer_head *orig_bh = bh->b_private; |
35 | ||
470decc6 DK |
36 | BUFFER_TRACE(bh, ""); |
37 | if (uptodate) | |
38 | set_buffer_uptodate(bh); | |
39 | else | |
40 | clear_buffer_uptodate(bh); | |
b34090e5 JK |
41 | if (orig_bh) { |
42 | clear_bit_unlock(BH_Shadow, &orig_bh->b_state); | |
4e857c58 | 43 | smp_mb__after_atomic(); |
b34090e5 JK |
44 | wake_up_bit(&orig_bh->b_state, BH_Shadow); |
45 | } | |
470decc6 DK |
46 | unlock_buffer(bh); |
47 | } | |
48 | ||
49 | /* | |
87c89c23 JK |
50 | * When an ext4 file is truncated, it is possible that some pages are not |
51 | * successfully freed, because they are attached to a committing transaction. | |
470decc6 DK |
52 | * After the transaction commits, these pages are left on the LRU, with no |
53 | * ->mapping, and with attached buffers. These pages are trivially reclaimable | |
54 | * by the VM, but their apparent absence upsets the VM accounting, and it makes | |
55 | * the numbers in /proc/meminfo look odd. | |
56 | * | |
57 | * So here, we have a buffer which has just come off the forget list. Look to | |
58 | * see if we can strip all buffers from the backing page. | |
59 | * | |
60 | * Called under lock_journal(), and possibly under journal_datalist_lock. The | |
61 | * caller provided us with a ref against the buffer, and we drop that here. | |
62 | */ | |
63 | static void release_buffer_page(struct buffer_head *bh) | |
64 | { | |
65 | struct page *page; | |
66 | ||
67 | if (buffer_dirty(bh)) | |
68 | goto nope; | |
69 | if (atomic_read(&bh->b_count) != 1) | |
70 | goto nope; | |
71 | page = bh->b_page; | |
72 | if (!page) | |
73 | goto nope; | |
74 | if (page->mapping) | |
75 | goto nope; | |
76 | ||
77 | /* OK, it's a truncated page */ | |
529ae9aa | 78 | if (!trylock_page(page)) |
470decc6 DK |
79 | goto nope; |
80 | ||
09cbfeaf | 81 | get_page(page); |
470decc6 DK |
82 | __brelse(bh); |
83 | try_to_free_buffers(page); | |
84 | unlock_page(page); | |
09cbfeaf | 85 | put_page(page); |
470decc6 DK |
86 | return; |
87 | ||
88 | nope: | |
89 | __brelse(bh); | |
90 | } | |
91 | ||
e5a120ae | 92 | static void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh) |
1f56c589 DW |
93 | { |
94 | struct commit_header *h; | |
95 | __u32 csum; | |
96 | ||
db9ee220 | 97 | if (!jbd2_journal_has_csum_v2or3(j)) |
1f56c589 DW |
98 | return; |
99 | ||
e5a120ae | 100 | h = (struct commit_header *)(bh->b_data); |
1f56c589 DW |
101 | h->h_chksum_type = 0; |
102 | h->h_chksum_size = 0; | |
103 | h->h_chksum[0] = 0; | |
e5a120ae | 104 | csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); |
1f56c589 DW |
105 | h->h_chksum[0] = cpu_to_be32(csum); |
106 | } | |
107 | ||
818d276c GS |
108 | /* |
109 | * Done it all: now submit the commit record. We should have | |
470decc6 DK |
110 | * cleaned up our previous buffers by now, so if we are in abort |
111 | * mode we can now just skip the rest of the journal write | |
112 | * entirely. | |
113 | * | |
114 | * Returns 1 if the journal needs to be aborted or 0 on success | |
115 | */ | |
818d276c GS |
116 | static int journal_submit_commit_record(journal_t *journal, |
117 | transaction_t *commit_transaction, | |
118 | struct buffer_head **cbh, | |
119 | __u32 crc32_sum) | |
470decc6 | 120 | { |
818d276c | 121 | struct commit_header *tmp; |
470decc6 | 122 | struct buffer_head *bh; |
818d276c | 123 | int ret; |
b42d1d6b | 124 | struct timespec64 now; |
470decc6 | 125 | |
6cba611e ZH |
126 | *cbh = NULL; |
127 | ||
470decc6 DK |
128 | if (is_journal_aborted(journal)) |
129 | return 0; | |
130 | ||
32ab6715 JK |
131 | bh = jbd2_journal_get_descriptor_buffer(commit_transaction, |
132 | JBD2_COMMIT_BLOCK); | |
e5a120ae | 133 | if (!bh) |
470decc6 DK |
134 | return 1; |
135 | ||
818d276c | 136 | tmp = (struct commit_header *)bh->b_data; |
b42d1d6b | 137 | ktime_get_coarse_real_ts64(&now); |
736603ab TT |
138 | tmp->h_commit_sec = cpu_to_be64(now.tv_sec); |
139 | tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec); | |
818d276c | 140 | |
56316a0d | 141 | if (jbd2_has_feature_checksum(journal)) { |
818d276c GS |
142 | tmp->h_chksum_type = JBD2_CRC32_CHKSUM; |
143 | tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE; | |
144 | tmp->h_chksum[0] = cpu_to_be32(crc32_sum); | |
470decc6 | 145 | } |
e5a120ae | 146 | jbd2_commit_block_csum_set(journal, bh); |
470decc6 | 147 | |
e5a120ae | 148 | BUFFER_TRACE(bh, "submit commit block"); |
818d276c | 149 | lock_buffer(bh); |
45a90bfd | 150 | clear_buffer_dirty(bh); |
818d276c GS |
151 | set_buffer_uptodate(bh); |
152 | bh->b_end_io = journal_end_buffer_io_sync; | |
153 | ||
154 | if (journal->j_flags & JBD2_BARRIER && | |
56316a0d | 155 | !jbd2_has_feature_async_commit(journal)) |
70fd7614 CH |
156 | ret = submit_bh(REQ_OP_WRITE, |
157 | REQ_SYNC | REQ_PREFLUSH | REQ_FUA, bh); | |
9c35575b | 158 | else |
70fd7614 | 159 | ret = submit_bh(REQ_OP_WRITE, REQ_SYNC, bh); |
9c35575b | 160 | |
818d276c GS |
161 | *cbh = bh; |
162 | return ret; | |
163 | } | |
164 | ||
165 | /* | |
166 | * This function along with journal_submit_commit_record | |
167 | * allows to write the commit record asynchronously. | |
168 | */ | |
fd98496f TT |
169 | static int journal_wait_on_commit_record(journal_t *journal, |
170 | struct buffer_head *bh) | |
818d276c GS |
171 | { |
172 | int ret = 0; | |
173 | ||
174 | clear_buffer_dirty(bh); | |
175 | wait_on_buffer(bh); | |
470decc6 | 176 | |
818d276c GS |
177 | if (unlikely(!buffer_uptodate(bh))) |
178 | ret = -EIO; | |
179 | put_bh(bh); /* One for getblk() */ | |
818d276c GS |
180 | |
181 | return ret; | |
470decc6 DK |
182 | } |
183 | ||
cd1aac32 AK |
184 | /* |
185 | * write the filemap data using writepage() address_space_operations. | |
186 | * We don't do block allocation here even for delalloc. We don't | |
a4977306 | 187 | * use writepages() because with delayed allocation we may be doing |
cd1aac32 AK |
188 | * block allocation in writepages(). |
189 | */ | |
aa3c0c61 | 190 | int jbd2_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) |
cd1aac32 | 191 | { |
aa3c0c61 | 192 | struct address_space *mapping = jinode->i_vfs_inode->i_mapping; |
cd1aac32 AK |
193 | struct writeback_control wbc = { |
194 | .sync_mode = WB_SYNC_ALL, | |
195 | .nr_to_write = mapping->nrpages * 2, | |
aa3c0c61 MFO |
196 | .range_start = jinode->i_dirty_start, |
197 | .range_end = jinode->i_dirty_end, | |
cd1aac32 AK |
198 | }; |
199 | ||
342af94e MFO |
200 | /* |
201 | * submit the inode data buffers. We use writepage | |
202 | * instead of writepages. Because writepages can do | |
203 | * block allocation with delalloc. We need to write | |
204 | * only allocated blocks here. | |
205 | */ | |
aa3c0c61 | 206 | return generic_writepages(mapping, &wbc); |
cd1aac32 AK |
207 | } |
208 | ||
ff780b91 HS |
209 | /* Send all the data buffers related to an inode */ |
210 | int jbd2_submit_inode_data(struct jbd2_inode *jinode) | |
211 | { | |
212 | ||
213 | if (!jinode || !(jinode->i_flags & JI_WRITE_DATA)) | |
214 | return 0; | |
215 | ||
216 | trace_jbd2_submit_inode_data(jinode->i_vfs_inode); | |
217 | return jbd2_journal_submit_inode_data_buffers(jinode); | |
218 | ||
219 | } | |
220 | EXPORT_SYMBOL(jbd2_submit_inode_data); | |
221 | ||
222 | int jbd2_wait_inode_data(journal_t *journal, struct jbd2_inode *jinode) | |
223 | { | |
224 | if (!jinode || !(jinode->i_flags & JI_WAIT_DATA) || | |
225 | !jinode->i_vfs_inode || !jinode->i_vfs_inode->i_mapping) | |
226 | return 0; | |
227 | return filemap_fdatawait_range_keep_errors( | |
228 | jinode->i_vfs_inode->i_mapping, jinode->i_dirty_start, | |
229 | jinode->i_dirty_end); | |
230 | } | |
231 | EXPORT_SYMBOL(jbd2_wait_inode_data); | |
232 | ||
c851ed54 JK |
233 | /* |
234 | * Submit all the data buffers of inode associated with the transaction to | |
235 | * disk. | |
236 | * | |
237 | * We are in a committing transaction. Therefore no new inode can be added to | |
238 | * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently | |
239 | * operate on from being released while we write out pages. | |
240 | */ | |
cd1aac32 | 241 | static int journal_submit_data_buffers(journal_t *journal, |
c851ed54 JK |
242 | transaction_t *commit_transaction) |
243 | { | |
244 | struct jbd2_inode *jinode; | |
245 | int err, ret = 0; | |
c851ed54 JK |
246 | |
247 | spin_lock(&journal->j_list_lock); | |
248 | list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) { | |
41617e1a JK |
249 | if (!(jinode->i_flags & JI_WRITE_DATA)) |
250 | continue; | |
cb0d9d47 | 251 | jinode->i_flags |= JI_COMMIT_RUNNING; |
c851ed54 | 252 | spin_unlock(&journal->j_list_lock); |
342af94e | 253 | /* submit the inode data buffers. */ |
879c5e6b | 254 | trace_jbd2_submit_inode_data(jinode->i_vfs_inode); |
342af94e MFO |
255 | if (journal->j_submit_inode_data_buffers) { |
256 | err = journal->j_submit_inode_data_buffers(jinode); | |
257 | if (!ret) | |
258 | ret = err; | |
259 | } | |
c851ed54 JK |
260 | spin_lock(&journal->j_list_lock); |
261 | J_ASSERT(jinode->i_transaction == commit_transaction); | |
cb0d9d47 JK |
262 | jinode->i_flags &= ~JI_COMMIT_RUNNING; |
263 | smp_mb(); | |
c851ed54 JK |
264 | wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING); |
265 | } | |
266 | spin_unlock(&journal->j_list_lock); | |
267 | return ret; | |
268 | } | |
269 | ||
aa3c0c61 MFO |
270 | int jbd2_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) |
271 | { | |
272 | struct address_space *mapping = jinode->i_vfs_inode->i_mapping; | |
273 | ||
274 | return filemap_fdatawait_range_keep_errors(mapping, | |
275 | jinode->i_dirty_start, | |
276 | jinode->i_dirty_end); | |
277 | } | |
278 | ||
c851ed54 JK |
279 | /* |
280 | * Wait for data submitted for writeout, refile inodes to proper | |
281 | * transaction if needed. | |
282 | * | |
283 | */ | |
284 | static int journal_finish_inode_data_buffers(journal_t *journal, | |
285 | transaction_t *commit_transaction) | |
286 | { | |
287 | struct jbd2_inode *jinode, *next_i; | |
288 | int err, ret = 0; | |
289 | ||
cd1aac32 | 290 | /* For locking, see the comment in journal_submit_data_buffers() */ |
c851ed54 JK |
291 | spin_lock(&journal->j_list_lock); |
292 | list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) { | |
41617e1a JK |
293 | if (!(jinode->i_flags & JI_WAIT_DATA)) |
294 | continue; | |
cb0d9d47 | 295 | jinode->i_flags |= JI_COMMIT_RUNNING; |
c851ed54 | 296 | spin_unlock(&journal->j_list_lock); |
342af94e MFO |
297 | /* wait for the inode data buffers writeout. */ |
298 | if (journal->j_finish_inode_data_buffers) { | |
299 | err = journal->j_finish_inode_data_buffers(jinode); | |
300 | if (!ret) | |
301 | ret = err; | |
302 | } | |
c851ed54 | 303 | spin_lock(&journal->j_list_lock); |
cb0d9d47 JK |
304 | jinode->i_flags &= ~JI_COMMIT_RUNNING; |
305 | smp_mb(); | |
c851ed54 JK |
306 | wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING); |
307 | } | |
308 | ||
309 | /* Now refile inode to proper lists */ | |
310 | list_for_each_entry_safe(jinode, next_i, | |
311 | &commit_transaction->t_inode_list, i_list) { | |
312 | list_del(&jinode->i_list); | |
313 | if (jinode->i_next_transaction) { | |
314 | jinode->i_transaction = jinode->i_next_transaction; | |
315 | jinode->i_next_transaction = NULL; | |
316 | list_add(&jinode->i_list, | |
317 | &jinode->i_transaction->t_inode_list); | |
318 | } else { | |
319 | jinode->i_transaction = NULL; | |
6ba0e7dc RZ |
320 | jinode->i_dirty_start = 0; |
321 | jinode->i_dirty_end = 0; | |
c851ed54 JK |
322 | } |
323 | } | |
324 | spin_unlock(&journal->j_list_lock); | |
325 | ||
326 | return ret; | |
327 | } | |
328 | ||
818d276c GS |
329 | static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh) |
330 | { | |
331 | struct page *page = bh->b_page; | |
332 | char *addr; | |
333 | __u32 checksum; | |
334 | ||
303a8f2a | 335 | addr = kmap_atomic(page); |
818d276c GS |
336 | checksum = crc32_be(crc32_sum, |
337 | (void *)(addr + offset_in_page(bh->b_data)), bh->b_size); | |
303a8f2a | 338 | kunmap_atomic(addr); |
818d276c GS |
339 | |
340 | return checksum; | |
341 | } | |
342 | ||
db9ee220 | 343 | static void write_tag_block(journal_t *j, journal_block_tag_t *tag, |
18eba7aa | 344 | unsigned long long block) |
b517bea1 ZB |
345 | { |
346 | tag->t_blocknr = cpu_to_be32(block & (u32)~0); | |
56316a0d | 347 | if (jbd2_has_feature_64bit(j)) |
b517bea1 ZB |
348 | tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1); |
349 | } | |
350 | ||
c3900875 DW |
351 | static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag, |
352 | struct buffer_head *bh, __u32 sequence) | |
353 | { | |
db9ee220 | 354 | journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag; |
c3900875 DW |
355 | struct page *page = bh->b_page; |
356 | __u8 *addr; | |
eee06c56 | 357 | __u32 csum32; |
18a6ea1e | 358 | __be32 seq; |
c3900875 | 359 | |
db9ee220 | 360 | if (!jbd2_journal_has_csum_v2or3(j)) |
c3900875 DW |
361 | return; |
362 | ||
18a6ea1e | 363 | seq = cpu_to_be32(sequence); |
906adea1 | 364 | addr = kmap_atomic(page); |
18a6ea1e | 365 | csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq)); |
eee06c56 DW |
366 | csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data), |
367 | bh->b_size); | |
906adea1 | 368 | kunmap_atomic(addr); |
c3900875 | 369 | |
56316a0d | 370 | if (jbd2_has_feature_csum3(j)) |
db9ee220 DW |
371 | tag3->t_checksum = cpu_to_be32(csum32); |
372 | else | |
373 | tag->t_checksum = cpu_to_be16(csum32); | |
c3900875 | 374 | } |
470decc6 | 375 | /* |
f7f4bccb | 376 | * jbd2_journal_commit_transaction |
470decc6 DK |
377 | * |
378 | * The primary function for committing a transaction to the log. This | |
379 | * function is called by the journal thread to begin a complete commit. | |
380 | */ | |
f7f4bccb | 381 | void jbd2_journal_commit_transaction(journal_t *journal) |
470decc6 | 382 | { |
8e85fb3f | 383 | struct transaction_stats_s stats; |
470decc6 | 384 | transaction_t *commit_transaction; |
e5a120ae JK |
385 | struct journal_head *jh; |
386 | struct buffer_head *descriptor; | |
470decc6 DK |
387 | struct buffer_head **wbuf = journal->j_wbuf; |
388 | int bufs; | |
389 | int flags; | |
390 | int err; | |
18eba7aa | 391 | unsigned long long blocknr; |
e07f7183 JB |
392 | ktime_t start_time; |
393 | u64 commit_time; | |
470decc6 | 394 | char *tagp = NULL; |
470decc6 DK |
395 | journal_block_tag_t *tag = NULL; |
396 | int space_left = 0; | |
397 | int first_tag = 0; | |
398 | int tag_flag; | |
794446c6 | 399 | int i; |
b517bea1 | 400 | int tag_bytes = journal_tag_bytes(journal); |
818d276c GS |
401 | struct buffer_head *cbh = NULL; /* For transactional checksums */ |
402 | __u32 crc32_sum = ~0; | |
82f04ab4 | 403 | struct blk_plug plug; |
3339578f JK |
404 | /* Tail of the journal */ |
405 | unsigned long first_block; | |
406 | tid_t first_tid; | |
407 | int update_tail; | |
3caa487f | 408 | int csum_size = 0; |
f5113eff | 409 | LIST_HEAD(io_bufs); |
e5a120ae | 410 | LIST_HEAD(log_bufs); |
3caa487f | 411 | |
db9ee220 | 412 | if (jbd2_journal_has_csum_v2or3(journal)) |
3caa487f | 413 | csum_size = sizeof(struct jbd2_journal_block_tail); |
470decc6 DK |
414 | |
415 | /* | |
416 | * First job: lock down the current transaction and wait for | |
417 | * all outstanding updates to complete. | |
418 | */ | |
419 | ||
f7f4bccb MC |
420 | /* Do we need to erase the effects of a prior jbd2_journal_flush? */ |
421 | if (journal->j_flags & JBD2_FLUSHED) { | |
470decc6 | 422 | jbd_debug(3, "super block updated\n"); |
6fa7aa50 | 423 | mutex_lock_io(&journal->j_checkpoint_mutex); |
79feb521 JK |
424 | /* |
425 | * We hold j_checkpoint_mutex so tail cannot change under us. | |
426 | * We don't need any special data guarantees for writing sb | |
427 | * since journal is empty and it is ok for write to be | |
428 | * flushed only with transaction commit. | |
429 | */ | |
430 | jbd2_journal_update_sb_log_tail(journal, | |
431 | journal->j_tail_sequence, | |
432 | journal->j_tail, | |
70fd7614 | 433 | REQ_SYNC); |
a78bb11d | 434 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
435 | } else { |
436 | jbd_debug(3, "superblock not updated\n"); | |
437 | } | |
438 | ||
439 | J_ASSERT(journal->j_running_transaction != NULL); | |
440 | J_ASSERT(journal->j_committing_transaction == NULL); | |
441 | ||
ff780b91 HS |
442 | write_lock(&journal->j_state_lock); |
443 | journal->j_flags |= JBD2_FULL_COMMIT_ONGOING; | |
444 | while (journal->j_flags & JBD2_FAST_COMMIT_ONGOING) { | |
445 | DEFINE_WAIT(wait); | |
446 | ||
447 | prepare_to_wait(&journal->j_fc_wait, &wait, | |
448 | TASK_UNINTERRUPTIBLE); | |
449 | write_unlock(&journal->j_state_lock); | |
450 | schedule(); | |
451 | write_lock(&journal->j_state_lock); | |
452 | finish_wait(&journal->j_fc_wait, &wait); | |
cc80586a HS |
453 | /* |
454 | * TODO: by blocking fast commits here, we are increasing | |
455 | * fsync() latency slightly. Strictly speaking, we don't need | |
456 | * to block fast commits until the transaction enters T_FLUSH | |
457 | * state. So an optimization is possible where we block new fast | |
458 | * commits here and wait for existing ones to complete | |
459 | * just before we enter T_FLUSH. That way, the existing fast | |
460 | * commits and this full commit can proceed parallely. | |
461 | */ | |
ff780b91 HS |
462 | } |
463 | write_unlock(&journal->j_state_lock); | |
464 | ||
470decc6 | 465 | commit_transaction = journal->j_running_transaction; |
470decc6 | 466 | |
879c5e6b | 467 | trace_jbd2_start_commit(journal, commit_transaction); |
f2a44523 | 468 | jbd_debug(1, "JBD2: starting commit of transaction %d\n", |
470decc6 DK |
469 | commit_transaction->t_tid); |
470 | ||
a931da6a | 471 | write_lock(&journal->j_state_lock); |
ff780b91 | 472 | journal->j_fc_off = 0; |
3ca841c1 | 473 | J_ASSERT(commit_transaction->t_state == T_RUNNING); |
470decc6 DK |
474 | commit_transaction->t_state = T_LOCKED; |
475 | ||
879c5e6b | 476 | trace_jbd2_commit_locking(journal, commit_transaction); |
bf699327 | 477 | stats.run.rs_wait = commit_transaction->t_max_wait; |
9fff24aa | 478 | stats.run.rs_request_delay = 0; |
bf699327 | 479 | stats.run.rs_locked = jiffies; |
9fff24aa TT |
480 | if (commit_transaction->t_requested) |
481 | stats.run.rs_request_delay = | |
482 | jbd2_time_diff(commit_transaction->t_requested, | |
483 | stats.run.rs_locked); | |
bf699327 TT |
484 | stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start, |
485 | stats.run.rs_locked); | |
8e85fb3f | 486 | |
470decc6 | 487 | spin_lock(&commit_transaction->t_handle_lock); |
a51dca9c | 488 | while (atomic_read(&commit_transaction->t_updates)) { |
470decc6 DK |
489 | DEFINE_WAIT(wait); |
490 | ||
491 | prepare_to_wait(&journal->j_wait_updates, &wait, | |
492 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 493 | if (atomic_read(&commit_transaction->t_updates)) { |
470decc6 | 494 | spin_unlock(&commit_transaction->t_handle_lock); |
a931da6a | 495 | write_unlock(&journal->j_state_lock); |
470decc6 | 496 | schedule(); |
a931da6a | 497 | write_lock(&journal->j_state_lock); |
470decc6 DK |
498 | spin_lock(&commit_transaction->t_handle_lock); |
499 | } | |
500 | finish_wait(&journal->j_wait_updates, &wait); | |
501 | } | |
502 | spin_unlock(&commit_transaction->t_handle_lock); | |
96f1e097 JK |
503 | commit_transaction->t_state = T_SWITCH; |
504 | write_unlock(&journal->j_state_lock); | |
470decc6 | 505 | |
a51dca9c | 506 | J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <= |
470decc6 DK |
507 | journal->j_max_transaction_buffers); |
508 | ||
509 | /* | |
510 | * First thing we are allowed to do is to discard any remaining | |
511 | * BJ_Reserved buffers. Note, it is _not_ permissible to assume | |
512 | * that there are no such buffers: if a large filesystem | |
513 | * operation like a truncate needs to split itself over multiple | |
f7f4bccb | 514 | * transactions, then it may try to do a jbd2_journal_restart() while |
470decc6 DK |
515 | * there are still BJ_Reserved buffers outstanding. These must |
516 | * be released cleanly from the current transaction. | |
517 | * | |
518 | * In this case, the filesystem must still reserve write access | |
519 | * again before modifying the buffer in the new transaction, but | |
520 | * we do not require it to remember exactly which old buffers it | |
521 | * has reserved. This is consistent with the existing behaviour | |
f7f4bccb | 522 | * that multiple jbd2_journal_get_write_access() calls to the same |
25985edc | 523 | * buffer are perfectly permissible. |
470decc6 DK |
524 | */ |
525 | while (commit_transaction->t_reserved_list) { | |
526 | jh = commit_transaction->t_reserved_list; | |
527 | JBUFFER_TRACE(jh, "reserved, unused: refile"); | |
528 | /* | |
f7f4bccb | 529 | * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may |
470decc6 DK |
530 | * leave undo-committed data. |
531 | */ | |
532 | if (jh->b_committed_data) { | |
533 | struct buffer_head *bh = jh2bh(jh); | |
534 | ||
46417064 | 535 | spin_lock(&jh->b_state_lock); |
af1e76d6 | 536 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 | 537 | jh->b_committed_data = NULL; |
46417064 | 538 | spin_unlock(&jh->b_state_lock); |
470decc6 | 539 | } |
f7f4bccb | 540 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
541 | } |
542 | ||
543 | /* | |
544 | * Now try to drop any written-back buffers from the journal's | |
545 | * checkpoint lists. We do this *before* commit because it potentially | |
546 | * frees some memory | |
547 | */ | |
548 | spin_lock(&journal->j_list_lock); | |
841df7df | 549 | __jbd2_journal_clean_checkpoint_list(journal, false); |
470decc6 DK |
550 | spin_unlock(&journal->j_list_lock); |
551 | ||
f2a44523 | 552 | jbd_debug(3, "JBD2: commit phase 1\n"); |
470decc6 | 553 | |
1ba37268 YY |
554 | /* |
555 | * Clear revoked flag to reflect there is no revoked buffers | |
556 | * in the next transaction which is going to be started. | |
557 | */ | |
558 | jbd2_clear_buffer_revoked_flags(journal); | |
559 | ||
470decc6 DK |
560 | /* |
561 | * Switch to a new revoke table. | |
562 | */ | |
f7f4bccb | 563 | jbd2_journal_switch_revoke_table(journal); |
470decc6 | 564 | |
8f7d89f3 JK |
565 | /* |
566 | * Reserved credits cannot be claimed anymore, free them | |
567 | */ | |
568 | atomic_sub(atomic_read(&journal->j_reserved_credits), | |
569 | &commit_transaction->t_outstanding_credits); | |
570 | ||
96f1e097 | 571 | write_lock(&journal->j_state_lock); |
879c5e6b | 572 | trace_jbd2_commit_flushing(journal, commit_transaction); |
bf699327 TT |
573 | stats.run.rs_flushing = jiffies; |
574 | stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked, | |
575 | stats.run.rs_flushing); | |
8e85fb3f | 576 | |
470decc6 DK |
577 | commit_transaction->t_state = T_FLUSH; |
578 | journal->j_committing_transaction = commit_transaction; | |
579 | journal->j_running_transaction = NULL; | |
e07f7183 | 580 | start_time = ktime_get(); |
470decc6 DK |
581 | commit_transaction->t_log_start = journal->j_head; |
582 | wake_up(&journal->j_wait_transaction_locked); | |
a931da6a | 583 | write_unlock(&journal->j_state_lock); |
470decc6 | 584 | |
cfc7bc89 | 585 | jbd_debug(3, "JBD2: commit phase 2a\n"); |
470decc6 | 586 | |
470decc6 DK |
587 | /* |
588 | * Now start flushing things to disk, in the order they appear | |
589 | * on the transaction lists. Data blocks go first. | |
590 | */ | |
cd1aac32 | 591 | err = journal_submit_data_buffers(journal, commit_transaction); |
470decc6 | 592 | if (err) |
a7fa2baf | 593 | jbd2_journal_abort(journal, err); |
470decc6 | 594 | |
82f04ab4 | 595 | blk_start_plug(&plug); |
9bcf976c | 596 | jbd2_journal_write_revoke_records(commit_transaction, &log_bufs); |
470decc6 | 597 | |
cfc7bc89 | 598 | jbd_debug(3, "JBD2: commit phase 2b\n"); |
470decc6 | 599 | |
470decc6 DK |
600 | /* |
601 | * Way to go: we have now written out all of the data for a | |
602 | * transaction! Now comes the tricky part: we need to write out | |
603 | * metadata. Loop over the transaction's entire buffer list: | |
604 | */ | |
a931da6a | 605 | write_lock(&journal->j_state_lock); |
470decc6 | 606 | commit_transaction->t_state = T_COMMIT; |
a931da6a | 607 | write_unlock(&journal->j_state_lock); |
470decc6 | 608 | |
879c5e6b | 609 | trace_jbd2_commit_logging(journal, commit_transaction); |
bf699327 TT |
610 | stats.run.rs_logging = jiffies; |
611 | stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing, | |
612 | stats.run.rs_logging); | |
9f356e5a | 613 | stats.run.rs_blocks = commit_transaction->t_nr_buffers; |
bf699327 | 614 | stats.run.rs_blocks_logged = 0; |
8e85fb3f | 615 | |
1dfc3220 | 616 | J_ASSERT(commit_transaction->t_nr_buffers <= |
a51dca9c | 617 | atomic_read(&commit_transaction->t_outstanding_credits)); |
1dfc3220 | 618 | |
87c89c23 | 619 | err = 0; |
470decc6 | 620 | bufs = 0; |
e5a120ae | 621 | descriptor = NULL; |
470decc6 DK |
622 | while (commit_transaction->t_buffers) { |
623 | ||
624 | /* Find the next buffer to be journaled... */ | |
625 | ||
626 | jh = commit_transaction->t_buffers; | |
627 | ||
628 | /* If we're in abort mode, we just un-journal the buffer and | |
7ad7445f | 629 | release it. */ |
470decc6 DK |
630 | |
631 | if (is_journal_aborted(journal)) { | |
7ad7445f | 632 | clear_buffer_jbddirty(jh2bh(jh)); |
470decc6 | 633 | JBUFFER_TRACE(jh, "journal is aborting: refile"); |
e06c8227 JB |
634 | jbd2_buffer_abort_trigger(jh, |
635 | jh->b_frozen_data ? | |
636 | jh->b_frozen_triggers : | |
637 | jh->b_triggers); | |
f7f4bccb | 638 | jbd2_journal_refile_buffer(journal, jh); |
470decc6 DK |
639 | /* If that was the last one, we need to clean up |
640 | * any descriptor buffers which may have been | |
641 | * already allocated, even if we are now | |
642 | * aborting. */ | |
643 | if (!commit_transaction->t_buffers) | |
644 | goto start_journal_io; | |
645 | continue; | |
646 | } | |
647 | ||
648 | /* Make sure we have a descriptor block in which to | |
649 | record the metadata buffer. */ | |
650 | ||
651 | if (!descriptor) { | |
470decc6 DK |
652 | J_ASSERT (bufs == 0); |
653 | ||
f2a44523 | 654 | jbd_debug(4, "JBD2: get descriptor\n"); |
470decc6 | 655 | |
32ab6715 JK |
656 | descriptor = jbd2_journal_get_descriptor_buffer( |
657 | commit_transaction, | |
658 | JBD2_DESCRIPTOR_BLOCK); | |
470decc6 | 659 | if (!descriptor) { |
a7fa2baf | 660 | jbd2_journal_abort(journal, -EIO); |
470decc6 DK |
661 | continue; |
662 | } | |
663 | ||
f2a44523 | 664 | jbd_debug(4, "JBD2: got buffer %llu (%p)\n", |
e5a120ae JK |
665 | (unsigned long long)descriptor->b_blocknr, |
666 | descriptor->b_data); | |
e5a120ae JK |
667 | tagp = &descriptor->b_data[sizeof(journal_header_t)]; |
668 | space_left = descriptor->b_size - | |
669 | sizeof(journal_header_t); | |
470decc6 | 670 | first_tag = 1; |
e5a120ae JK |
671 | set_buffer_jwrite(descriptor); |
672 | set_buffer_dirty(descriptor); | |
673 | wbuf[bufs++] = descriptor; | |
470decc6 DK |
674 | |
675 | /* Record it so that we can wait for IO | |
676 | completion later */ | |
e5a120ae JK |
677 | BUFFER_TRACE(descriptor, "ph3: file as descriptor"); |
678 | jbd2_file_log_bh(&log_bufs, descriptor); | |
470decc6 DK |
679 | } |
680 | ||
681 | /* Where is the buffer to be written? */ | |
682 | ||
f7f4bccb | 683 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
684 | /* If the block mapping failed, just abandon the buffer |
685 | and repeat this loop: we'll fall into the | |
686 | refile-on-abort condition above. */ | |
687 | if (err) { | |
a7fa2baf | 688 | jbd2_journal_abort(journal, err); |
470decc6 DK |
689 | continue; |
690 | } | |
691 | ||
692 | /* | |
693 | * start_this_handle() uses t_outstanding_credits to determine | |
0db45889 | 694 | * the free space in the log. |
470decc6 | 695 | */ |
a51dca9c | 696 | atomic_dec(&commit_transaction->t_outstanding_credits); |
470decc6 DK |
697 | |
698 | /* Bump b_count to prevent truncate from stumbling over | |
699 | the shadowed buffer! @@@ This can go if we ever get | |
f5113eff | 700 | rid of the shadow pairing of buffers. */ |
470decc6 DK |
701 | atomic_inc(&jh2bh(jh)->b_count); |
702 | ||
470decc6 | 703 | /* |
f5113eff JK |
704 | * Make a temporary IO buffer with which to write it out |
705 | * (this will requeue the metadata buffer to BJ_Shadow). | |
470decc6 | 706 | */ |
f5113eff | 707 | set_bit(BH_JWrite, &jh2bh(jh)->b_state); |
470decc6 | 708 | JBUFFER_TRACE(jh, "ph3: write metadata"); |
f7f4bccb | 709 | flags = jbd2_journal_write_metadata_buffer(commit_transaction, |
f5113eff | 710 | jh, &wbuf[bufs], blocknr); |
e6ec116b TT |
711 | if (flags < 0) { |
712 | jbd2_journal_abort(journal, flags); | |
713 | continue; | |
714 | } | |
f5113eff | 715 | jbd2_file_log_bh(&io_bufs, wbuf[bufs]); |
470decc6 DK |
716 | |
717 | /* Record the new block's tag in the current descriptor | |
718 | buffer */ | |
719 | ||
720 | tag_flag = 0; | |
721 | if (flags & 1) | |
f7f4bccb | 722 | tag_flag |= JBD2_FLAG_ESCAPE; |
470decc6 | 723 | if (!first_tag) |
f7f4bccb | 724 | tag_flag |= JBD2_FLAG_SAME_UUID; |
470decc6 DK |
725 | |
726 | tag = (journal_block_tag_t *) tagp; | |
db9ee220 | 727 | write_tag_block(journal, tag, jh2bh(jh)->b_blocknr); |
8f888ef8 | 728 | tag->t_flags = cpu_to_be16(tag_flag); |
f5113eff | 729 | jbd2_block_tag_csum_set(journal, tag, wbuf[bufs], |
c3900875 | 730 | commit_transaction->t_tid); |
b517bea1 ZB |
731 | tagp += tag_bytes; |
732 | space_left -= tag_bytes; | |
f5113eff | 733 | bufs++; |
470decc6 DK |
734 | |
735 | if (first_tag) { | |
736 | memcpy (tagp, journal->j_uuid, 16); | |
737 | tagp += 16; | |
738 | space_left -= 16; | |
739 | first_tag = 0; | |
740 | } | |
741 | ||
742 | /* If there's no more to do, or if the descriptor is full, | |
743 | let the IO rip! */ | |
744 | ||
745 | if (bufs == journal->j_wbufsize || | |
746 | commit_transaction->t_buffers == NULL || | |
3caa487f | 747 | space_left < tag_bytes + 16 + csum_size) { |
470decc6 | 748 | |
f2a44523 | 749 | jbd_debug(4, "JBD2: Submit %d IOs\n", bufs); |
470decc6 DK |
750 | |
751 | /* Write an end-of-descriptor marker before | |
752 | submitting the IOs. "tag" still points to | |
753 | the last tag we set up. */ | |
754 | ||
8f888ef8 | 755 | tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG); |
470decc6 | 756 | start_journal_io: |
6e876c3d | 757 | if (descriptor) |
758 | jbd2_descriptor_block_csum_set(journal, | |
759 | descriptor); | |
760 | ||
470decc6 DK |
761 | for (i = 0; i < bufs; i++) { |
762 | struct buffer_head *bh = wbuf[i]; | |
818d276c GS |
763 | /* |
764 | * Compute checksum. | |
765 | */ | |
56316a0d | 766 | if (jbd2_has_feature_checksum(journal)) { |
818d276c GS |
767 | crc32_sum = |
768 | jbd2_checksum_data(crc32_sum, bh); | |
769 | } | |
770 | ||
470decc6 DK |
771 | lock_buffer(bh); |
772 | clear_buffer_dirty(bh); | |
773 | set_buffer_uptodate(bh); | |
774 | bh->b_end_io = journal_end_buffer_io_sync; | |
70fd7614 | 775 | submit_bh(REQ_OP_WRITE, REQ_SYNC, bh); |
470decc6 DK |
776 | } |
777 | cond_resched(); | |
778 | ||
779 | /* Force a new descriptor to be generated next | |
780 | time round the loop. */ | |
781 | descriptor = NULL; | |
782 | bufs = 0; | |
783 | } | |
784 | } | |
785 | ||
f73bee49 JK |
786 | err = journal_finish_inode_data_buffers(journal, commit_transaction); |
787 | if (err) { | |
788 | printk(KERN_WARNING | |
789 | "JBD2: Detected IO errors while flushing file data " | |
790 | "on %s\n", journal->j_devname); | |
791 | if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR) | |
792 | jbd2_journal_abort(journal, err); | |
793 | err = 0; | |
794 | } | |
795 | ||
3339578f JK |
796 | /* |
797 | * Get current oldest transaction in the log before we issue flush | |
798 | * to the filesystem device. After the flush we can be sure that | |
799 | * blocks of all older transactions are checkpointed to persistent | |
800 | * storage and we will be safe to update journal start in the | |
801 | * superblock with the numbers we get here. | |
802 | */ | |
803 | update_tail = | |
804 | jbd2_journal_get_log_tail(journal, &first_tid, &first_block); | |
805 | ||
bbd2be36 | 806 | write_lock(&journal->j_state_lock); |
3339578f JK |
807 | if (update_tail) { |
808 | long freed = first_block - journal->j_tail; | |
809 | ||
810 | if (first_block < journal->j_tail) | |
811 | freed += journal->j_last - journal->j_first; | |
812 | /* Update tail only if we free significant amount of space */ | |
ede7dc7f | 813 | if (freed < jbd2_journal_get_max_txn_bufs(journal)) |
3339578f JK |
814 | update_tail = 0; |
815 | } | |
bbd2be36 JK |
816 | J_ASSERT(commit_transaction->t_state == T_COMMIT); |
817 | commit_transaction->t_state = T_COMMIT_DFLUSH; | |
818 | write_unlock(&journal->j_state_lock); | |
3339578f | 819 | |
cc3e1bea TT |
820 | /* |
821 | * If the journal is not located on the file system device, | |
822 | * then we must flush the file system device before we issue | |
823 | * the commit record | |
824 | */ | |
81be12c8 | 825 | if (commit_transaction->t_need_data_flush && |
cc3e1bea TT |
826 | (journal->j_fs_dev != journal->j_dev) && |
827 | (journal->j_flags & JBD2_BARRIER)) | |
c6bf3f0e | 828 | blkdev_issue_flush(journal->j_fs_dev); |
818d276c | 829 | |
cc3e1bea | 830 | /* Done it all: now write the commit record asynchronously. */ |
56316a0d | 831 | if (jbd2_has_feature_async_commit(journal)) { |
818d276c GS |
832 | err = journal_submit_commit_record(journal, commit_transaction, |
833 | &cbh, crc32_sum); | |
834 | if (err) | |
d0a186e0 | 835 | jbd2_journal_abort(journal, err); |
e9e34f4e | 836 | } |
c851ed54 | 837 | |
82f04ab4 JA |
838 | blk_finish_plug(&plug); |
839 | ||
470decc6 DK |
840 | /* Lo and behold: we have just managed to send a transaction to |
841 | the log. Before we can commit it, wait for the IO so far to | |
842 | complete. Control buffers being written are on the | |
843 | transaction's t_log_list queue, and metadata buffers are on | |
f5113eff | 844 | the io_bufs list. |
470decc6 DK |
845 | |
846 | Wait for the buffers in reverse order. That way we are | |
847 | less likely to be woken up until all IOs have completed, and | |
848 | so we incur less scheduling load. | |
849 | */ | |
850 | ||
f2a44523 | 851 | jbd_debug(3, "JBD2: commit phase 3\n"); |
470decc6 | 852 | |
f5113eff JK |
853 | while (!list_empty(&io_bufs)) { |
854 | struct buffer_head *bh = list_entry(io_bufs.prev, | |
855 | struct buffer_head, | |
856 | b_assoc_buffers); | |
470decc6 | 857 | |
f5113eff JK |
858 | wait_on_buffer(bh); |
859 | cond_resched(); | |
470decc6 DK |
860 | |
861 | if (unlikely(!buffer_uptodate(bh))) | |
862 | err = -EIO; | |
f5113eff | 863 | jbd2_unfile_log_bh(bh); |
015c6033 | 864 | stats.run.rs_blocks_logged++; |
470decc6 DK |
865 | |
866 | /* | |
f5113eff JK |
867 | * The list contains temporary buffer heads created by |
868 | * jbd2_journal_write_metadata_buffer(). | |
470decc6 DK |
869 | */ |
870 | BUFFER_TRACE(bh, "dumping temporary bh"); | |
470decc6 DK |
871 | __brelse(bh); |
872 | J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0); | |
873 | free_buffer_head(bh); | |
874 | ||
f5113eff | 875 | /* We also have to refile the corresponding shadowed buffer */ |
470decc6 DK |
876 | jh = commit_transaction->t_shadow_list->b_tprev; |
877 | bh = jh2bh(jh); | |
f5113eff | 878 | clear_buffer_jwrite(bh); |
470decc6 | 879 | J_ASSERT_BH(bh, buffer_jbddirty(bh)); |
b34090e5 | 880 | J_ASSERT_BH(bh, !buffer_shadow(bh)); |
470decc6 DK |
881 | |
882 | /* The metadata is now released for reuse, but we need | |
883 | to remember it against this transaction so that when | |
884 | we finally commit, we can do any checkpointing | |
885 | required. */ | |
886 | JBUFFER_TRACE(jh, "file as BJ_Forget"); | |
f7f4bccb | 887 | jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget); |
470decc6 DK |
888 | JBUFFER_TRACE(jh, "brelse shadowed buffer"); |
889 | __brelse(bh); | |
890 | } | |
891 | ||
892 | J_ASSERT (commit_transaction->t_shadow_list == NULL); | |
893 | ||
f2a44523 | 894 | jbd_debug(3, "JBD2: commit phase 4\n"); |
470decc6 DK |
895 | |
896 | /* Here we wait for the revoke record and descriptor record buffers */ | |
e5a120ae | 897 | while (!list_empty(&log_bufs)) { |
470decc6 DK |
898 | struct buffer_head *bh; |
899 | ||
e5a120ae JK |
900 | bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers); |
901 | wait_on_buffer(bh); | |
902 | cond_resched(); | |
470decc6 DK |
903 | |
904 | if (unlikely(!buffer_uptodate(bh))) | |
905 | err = -EIO; | |
906 | ||
907 | BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile"); | |
908 | clear_buffer_jwrite(bh); | |
e5a120ae | 909 | jbd2_unfile_log_bh(bh); |
015c6033 | 910 | stats.run.rs_blocks_logged++; |
470decc6 DK |
911 | __brelse(bh); /* One for getblk */ |
912 | /* AKPM: bforget here */ | |
913 | } | |
914 | ||
77e841de HK |
915 | if (err) |
916 | jbd2_journal_abort(journal, err); | |
917 | ||
f2a44523 | 918 | jbd_debug(3, "JBD2: commit phase 5\n"); |
bbd2be36 JK |
919 | write_lock(&journal->j_state_lock); |
920 | J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH); | |
921 | commit_transaction->t_state = T_COMMIT_JFLUSH; | |
922 | write_unlock(&journal->j_state_lock); | |
470decc6 | 923 | |
56316a0d | 924 | if (!jbd2_has_feature_async_commit(journal)) { |
818d276c GS |
925 | err = journal_submit_commit_record(journal, commit_transaction, |
926 | &cbh, crc32_sum); | |
927 | if (err) | |
d0a186e0 | 928 | jbd2_journal_abort(journal, err); |
818d276c | 929 | } |
6cba611e | 930 | if (cbh) |
fd98496f | 931 | err = journal_wait_on_commit_record(journal, cbh); |
015c6033 | 932 | stats.run.rs_blocks_logged++; |
56316a0d | 933 | if (jbd2_has_feature_async_commit(journal) && |
f73bee49 | 934 | journal->j_flags & JBD2_BARRIER) { |
c6bf3f0e | 935 | blkdev_issue_flush(journal->j_dev); |
f73bee49 | 936 | } |
470decc6 DK |
937 | |
938 | if (err) | |
a7fa2baf | 939 | jbd2_journal_abort(journal, err); |
470decc6 | 940 | |
9f356e5a JK |
941 | WARN_ON_ONCE( |
942 | atomic_read(&commit_transaction->t_outstanding_credits) < 0); | |
943 | ||
3339578f JK |
944 | /* |
945 | * Now disk caches for filesystem device are flushed so we are safe to | |
946 | * erase checkpointed transactions from the log by updating journal | |
947 | * superblock. | |
948 | */ | |
949 | if (update_tail) | |
950 | jbd2_update_log_tail(journal, first_tid, first_block); | |
951 | ||
470decc6 DK |
952 | /* End of a transaction! Finally, we can do checkpoint |
953 | processing: any buffers committed as a result of this | |
954 | transaction can be removed from any checkpoint list it was on | |
955 | before. */ | |
956 | ||
f2a44523 | 957 | jbd_debug(3, "JBD2: commit phase 6\n"); |
470decc6 | 958 | |
c851ed54 | 959 | J_ASSERT(list_empty(&commit_transaction->t_inode_list)); |
470decc6 DK |
960 | J_ASSERT(commit_transaction->t_buffers == NULL); |
961 | J_ASSERT(commit_transaction->t_checkpoint_list == NULL); | |
470decc6 | 962 | J_ASSERT(commit_transaction->t_shadow_list == NULL); |
470decc6 DK |
963 | |
964 | restart_loop: | |
965 | /* | |
966 | * As there are other places (journal_unmap_buffer()) adding buffers | |
967 | * to this list we have to be careful and hold the j_list_lock. | |
968 | */ | |
969 | spin_lock(&journal->j_list_lock); | |
970 | while (commit_transaction->t_forget) { | |
971 | transaction_t *cp_transaction; | |
972 | struct buffer_head *bh; | |
de1b7941 | 973 | int try_to_free = 0; |
93108ebb | 974 | bool drop_ref; |
470decc6 DK |
975 | |
976 | jh = commit_transaction->t_forget; | |
977 | spin_unlock(&journal->j_list_lock); | |
978 | bh = jh2bh(jh); | |
de1b7941 JK |
979 | /* |
980 | * Get a reference so that bh cannot be freed before we are | |
981 | * done with it. | |
982 | */ | |
983 | get_bh(bh); | |
46417064 | 984 | spin_lock(&jh->b_state_lock); |
23e2af35 | 985 | J_ASSERT_JH(jh, jh->b_transaction == commit_transaction); |
470decc6 DK |
986 | |
987 | /* | |
988 | * If there is undo-protected committed data against | |
989 | * this buffer, then we can remove it now. If it is a | |
990 | * buffer needing such protection, the old frozen_data | |
991 | * field now points to a committed version of the | |
992 | * buffer, so rotate that field to the new committed | |
993 | * data. | |
994 | * | |
995 | * Otherwise, we can just throw away the frozen data now. | |
e06c8227 JB |
996 | * |
997 | * We also know that the frozen data has already fired | |
998 | * its triggers if they exist, so we can clear that too. | |
470decc6 DK |
999 | */ |
1000 | if (jh->b_committed_data) { | |
af1e76d6 | 1001 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
1002 | jh->b_committed_data = NULL; |
1003 | if (jh->b_frozen_data) { | |
1004 | jh->b_committed_data = jh->b_frozen_data; | |
1005 | jh->b_frozen_data = NULL; | |
e06c8227 | 1006 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
1007 | } |
1008 | } else if (jh->b_frozen_data) { | |
af1e76d6 | 1009 | jbd2_free(jh->b_frozen_data, bh->b_size); |
470decc6 | 1010 | jh->b_frozen_data = NULL; |
e06c8227 | 1011 | jh->b_frozen_triggers = NULL; |
470decc6 DK |
1012 | } |
1013 | ||
1014 | spin_lock(&journal->j_list_lock); | |
1015 | cp_transaction = jh->b_cp_transaction; | |
1016 | if (cp_transaction) { | |
1017 | JBUFFER_TRACE(jh, "remove from old cp transaction"); | |
8e85fb3f | 1018 | cp_transaction->t_chp_stats.cs_dropped++; |
f7f4bccb | 1019 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
1020 | } |
1021 | ||
1022 | /* Only re-checkpoint the buffer_head if it is marked | |
1023 | * dirty. If the buffer was added to the BJ_Forget list | |
f7f4bccb | 1024 | * by jbd2_journal_forget, it may no longer be dirty and |
470decc6 DK |
1025 | * there's no point in keeping a checkpoint record for |
1026 | * it. */ | |
1027 | ||
b794e7a6 | 1028 | /* |
6a66a7de | 1029 | * A buffer which has been freed while still being journaled |
1030 | * by a previous transaction, refile the buffer to BJ_Forget of | |
1031 | * the running transaction. If the just committed transaction | |
1032 | * contains "add to orphan" operation, we can completely | |
1033 | * invalidate the buffer now. We are rather through in that | |
1034 | * since the buffer may be still accessible when blocksize < | |
1035 | * pagesize and it is attached to the last partial page. | |
1036 | */ | |
1037 | if (buffer_freed(bh) && !jh->b_next_transaction) { | |
c96dceea | 1038 | struct address_space *mapping; |
1039 | ||
6a66a7de | 1040 | clear_buffer_freed(bh); |
1041 | clear_buffer_jbddirty(bh); | |
c96dceea | 1042 | |
1043 | /* | |
1044 | * Block device buffers need to stay mapped all the | |
1045 | * time, so it is enough to clear buffer_jbddirty and | |
1046 | * buffer_freed bits. For the file mapping buffers (i.e. | |
1047 | * journalled data) we need to unmap buffer and clear | |
1048 | * more bits. We also need to be careful about the check | |
1049 | * because the data page mapping can get cleared under | |
780f66e5 | 1050 | * our hands. Note that if mapping == NULL, we don't |
1051 | * need to make buffer unmapped because the page is | |
1052 | * already detached from the mapping and buffers cannot | |
1053 | * get reused. | |
c96dceea | 1054 | */ |
1055 | mapping = READ_ONCE(bh->b_page->mapping); | |
1056 | if (mapping && !sb_is_blkdev_sb(mapping->host->i_sb)) { | |
1057 | clear_buffer_mapped(bh); | |
1058 | clear_buffer_new(bh); | |
1059 | clear_buffer_req(bh); | |
1060 | bh->b_bdev = NULL; | |
1061 | } | |
470decc6 DK |
1062 | } |
1063 | ||
1064 | if (buffer_jbddirty(bh)) { | |
1065 | JBUFFER_TRACE(jh, "add to new checkpointing trans"); | |
f7f4bccb | 1066 | __jbd2_journal_insert_checkpoint(jh, commit_transaction); |
7ad7445f HK |
1067 | if (is_journal_aborted(journal)) |
1068 | clear_buffer_jbddirty(bh); | |
470decc6 DK |
1069 | } else { |
1070 | J_ASSERT_BH(bh, !buffer_dirty(bh)); | |
de1b7941 JK |
1071 | /* |
1072 | * The buffer on BJ_Forget list and not jbddirty means | |
470decc6 DK |
1073 | * it has been freed by this transaction and hence it |
1074 | * could not have been reallocated until this | |
1075 | * transaction has committed. *BUT* it could be | |
1076 | * reallocated once we have written all the data to | |
1077 | * disk and before we process the buffer on BJ_Forget | |
de1b7941 JK |
1078 | * list. |
1079 | */ | |
1080 | if (!jh->b_next_transaction) | |
1081 | try_to_free = 1; | |
470decc6 | 1082 | } |
de1b7941 | 1083 | JBUFFER_TRACE(jh, "refile or unfile buffer"); |
93108ebb | 1084 | drop_ref = __jbd2_journal_refile_buffer(jh); |
46417064 | 1085 | spin_unlock(&jh->b_state_lock); |
93108ebb JK |
1086 | if (drop_ref) |
1087 | jbd2_journal_put_journal_head(jh); | |
de1b7941 JK |
1088 | if (try_to_free) |
1089 | release_buffer_page(bh); /* Drops bh reference */ | |
1090 | else | |
1091 | __brelse(bh); | |
470decc6 DK |
1092 | cond_resched_lock(&journal->j_list_lock); |
1093 | } | |
1094 | spin_unlock(&journal->j_list_lock); | |
1095 | /* | |
f5a7a6b0 JK |
1096 | * This is a bit sleazy. We use j_list_lock to protect transition |
1097 | * of a transaction into T_FINISHED state and calling | |
1098 | * __jbd2_journal_drop_transaction(). Otherwise we could race with | |
1099 | * other checkpointing code processing the transaction... | |
470decc6 | 1100 | */ |
a931da6a | 1101 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1102 | spin_lock(&journal->j_list_lock); |
1103 | /* | |
1104 | * Now recheck if some buffers did not get attached to the transaction | |
1105 | * while the lock was dropped... | |
1106 | */ | |
1107 | if (commit_transaction->t_forget) { | |
1108 | spin_unlock(&journal->j_list_lock); | |
a931da6a | 1109 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1110 | goto restart_loop; |
1111 | } | |
1112 | ||
d4e839d4 TT |
1113 | /* Add the transaction to the checkpoint list |
1114 | * __journal_remove_checkpoint() can not destroy transaction | |
1115 | * under us because it is not marked as T_FINISHED yet */ | |
1116 | if (journal->j_checkpoint_transactions == NULL) { | |
1117 | journal->j_checkpoint_transactions = commit_transaction; | |
1118 | commit_transaction->t_cpnext = commit_transaction; | |
1119 | commit_transaction->t_cpprev = commit_transaction; | |
1120 | } else { | |
1121 | commit_transaction->t_cpnext = | |
1122 | journal->j_checkpoint_transactions; | |
1123 | commit_transaction->t_cpprev = | |
1124 | commit_transaction->t_cpnext->t_cpprev; | |
1125 | commit_transaction->t_cpnext->t_cpprev = | |
1126 | commit_transaction; | |
1127 | commit_transaction->t_cpprev->t_cpnext = | |
1128 | commit_transaction; | |
1129 | } | |
1130 | spin_unlock(&journal->j_list_lock); | |
1131 | ||
470decc6 DK |
1132 | /* Done with this transaction! */ |
1133 | ||
f2a44523 | 1134 | jbd_debug(3, "JBD2: commit phase 7\n"); |
470decc6 | 1135 | |
bbd2be36 | 1136 | J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH); |
470decc6 | 1137 | |
8e85fb3f | 1138 | commit_transaction->t_start = jiffies; |
bf699327 TT |
1139 | stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging, |
1140 | commit_transaction->t_start); | |
8e85fb3f JL |
1141 | |
1142 | /* | |
bf699327 | 1143 | * File the transaction statistics |
8e85fb3f | 1144 | */ |
8e85fb3f | 1145 | stats.ts_tid = commit_transaction->t_tid; |
8dd42046 TT |
1146 | stats.run.rs_handle_count = |
1147 | atomic_read(&commit_transaction->t_handle_count); | |
bf699327 TT |
1148 | trace_jbd2_run_stats(journal->j_fs_dev->bd_dev, |
1149 | commit_transaction->t_tid, &stats.run); | |
42cf3452 | 1150 | stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0; |
8e85fb3f | 1151 | |
794446c6 | 1152 | commit_transaction->t_state = T_COMMIT_CALLBACK; |
470decc6 DK |
1153 | J_ASSERT(commit_transaction == journal->j_committing_transaction); |
1154 | journal->j_commit_sequence = commit_transaction->t_tid; | |
1155 | journal->j_committing_transaction = NULL; | |
e07f7183 | 1156 | commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); |
470decc6 | 1157 | |
e07f7183 JB |
1158 | /* |
1159 | * weight the commit time higher than the average time so we don't | |
1160 | * react too strongly to vast changes in the commit time | |
1161 | */ | |
1162 | if (likely(journal->j_average_commit_time)) | |
1163 | journal->j_average_commit_time = (commit_time + | |
1164 | journal->j_average_commit_time*3) / 4; | |
1165 | else | |
1166 | journal->j_average_commit_time = commit_time; | |
794446c6 | 1167 | |
a931da6a | 1168 | write_unlock(&journal->j_state_lock); |
6c20ec85 | 1169 | |
fb68407b AK |
1170 | if (journal->j_commit_callback) |
1171 | journal->j_commit_callback(journal, commit_transaction); | |
ff780b91 | 1172 | if (journal->j_fc_cleanup_callback) |
e5aed39e | 1173 | journal->j_fc_cleanup_callback(journal, 1, commit_transaction->t_tid); |
fb68407b | 1174 | |
879c5e6b | 1175 | trace_jbd2_end_commit(journal, commit_transaction); |
f2a44523 | 1176 | jbd_debug(1, "JBD2: commit %d complete, head %d\n", |
470decc6 DK |
1177 | journal->j_commit_sequence, journal->j_tail_sequence); |
1178 | ||
794446c6 | 1179 | write_lock(&journal->j_state_lock); |
ff780b91 HS |
1180 | journal->j_flags &= ~JBD2_FULL_COMMIT_ONGOING; |
1181 | journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING; | |
794446c6 DM |
1182 | spin_lock(&journal->j_list_lock); |
1183 | commit_transaction->t_state = T_FINISHED; | |
d4e839d4 | 1184 | /* Check if the transaction can be dropped now that we are finished */ |
794446c6 DM |
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 | 1192 | wake_up(&journal->j_wait_done_commit); |
ff780b91 | 1193 | wake_up(&journal->j_fc_wait); |
42cf3452 TT |
1194 | |
1195 | /* | |
1196 | * Calculate overall stats | |
1197 | */ | |
1198 | spin_lock(&journal->j_history_lock); | |
1199 | journal->j_stats.ts_tid++; | |
1200 | journal->j_stats.ts_requested += stats.ts_requested; | |
1201 | journal->j_stats.run.rs_wait += stats.run.rs_wait; | |
1202 | journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay; | |
1203 | journal->j_stats.run.rs_running += stats.run.rs_running; | |
1204 | journal->j_stats.run.rs_locked += stats.run.rs_locked; | |
1205 | journal->j_stats.run.rs_flushing += stats.run.rs_flushing; | |
1206 | journal->j_stats.run.rs_logging += stats.run.rs_logging; | |
1207 | journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count; | |
1208 | journal->j_stats.run.rs_blocks += stats.run.rs_blocks; | |
1209 | journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged; | |
1210 | spin_unlock(&journal->j_history_lock); | |
470decc6 | 1211 | } |