1 // SPDX-License-Identifier: GPL-2.0+ 1 // SPDX-License-Identifier: GPL-2.0+
2 /* 2 /*
3 * linux/fs/jbd2/commit.c 3 * linux/fs/jbd2/commit.c
4 * 4 *
5 * Written by Stephen C. Tweedie <sct@redhat.c 5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * 6 *
7 * Copyright 1998 Red Hat corp --- All Rights 7 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * 8 *
9 * Journal commit routines for the generic fil 9 * Journal commit routines for the generic filesystem journaling code;
10 * part of the ext2fs journaling system. 10 * part of the ext2fs journaling system.
11 */ 11 */
12 12
13 #include <linux/time.h> 13 #include <linux/time.h>
14 #include <linux/fs.h> 14 #include <linux/fs.h>
15 #include <linux/jbd2.h> 15 #include <linux/jbd2.h>
16 #include <linux/errno.h> 16 #include <linux/errno.h>
17 #include <linux/slab.h> 17 #include <linux/slab.h>
18 #include <linux/mm.h> 18 #include <linux/mm.h>
19 #include <linux/pagemap.h> 19 #include <linux/pagemap.h>
20 #include <linux/jiffies.h> 20 #include <linux/jiffies.h>
21 #include <linux/crc32.h> 21 #include <linux/crc32.h>
22 #include <linux/writeback.h> 22 #include <linux/writeback.h>
23 #include <linux/backing-dev.h> 23 #include <linux/backing-dev.h>
24 #include <linux/bio.h> 24 #include <linux/bio.h>
25 #include <linux/blkdev.h> 25 #include <linux/blkdev.h>
26 #include <linux/bitops.h> 26 #include <linux/bitops.h>
27 #include <trace/events/jbd2.h> 27 #include <trace/events/jbd2.h>
28 28
29 /* 29 /*
30 * IO end handler for temporary buffer_heads h 30 * IO end handler for temporary buffer_heads handling writes to the journal.
31 */ 31 */
32 static void journal_end_buffer_io_sync(struct 32 static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
33 { 33 {
34 struct buffer_head *orig_bh = bh->b_pr 34 struct buffer_head *orig_bh = bh->b_private;
35 35
36 BUFFER_TRACE(bh, ""); 36 BUFFER_TRACE(bh, "");
37 if (uptodate) 37 if (uptodate)
38 set_buffer_uptodate(bh); 38 set_buffer_uptodate(bh);
39 else 39 else
40 clear_buffer_uptodate(bh); 40 clear_buffer_uptodate(bh);
41 if (orig_bh) { 41 if (orig_bh) {
42 clear_bit_unlock(BH_Shadow, &o 42 clear_bit_unlock(BH_Shadow, &orig_bh->b_state);
43 smp_mb__after_atomic(); 43 smp_mb__after_atomic();
44 wake_up_bit(&orig_bh->b_state, 44 wake_up_bit(&orig_bh->b_state, BH_Shadow);
45 } 45 }
46 unlock_buffer(bh); 46 unlock_buffer(bh);
47 } 47 }
48 48
49 /* 49 /*
50 * When an ext4 file is truncated, it is possi 50 * When an ext4 file is truncated, it is possible that some pages are not
51 * successfully freed, because they are attach 51 * successfully freed, because they are attached to a committing transaction.
52 * After the transaction commits, these pages 52 * After the transaction commits, these pages are left on the LRU, with no
53 * ->mapping, and with attached buffers. Thes 53 * ->mapping, and with attached buffers. These pages are trivially reclaimable
54 * by the VM, but their apparent absence upset 54 * by the VM, but their apparent absence upsets the VM accounting, and it makes
55 * the numbers in /proc/meminfo look odd. 55 * the numbers in /proc/meminfo look odd.
56 * 56 *
57 * So here, we have a buffer which has just co 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 ba 58 * see if we can strip all buffers from the backing page.
59 * 59 *
60 * Called under lock_journal(), and possibly u 60 * Called under lock_journal(), and possibly under journal_datalist_lock. The
61 * caller provided us with a ref against the b 61 * caller provided us with a ref against the buffer, and we drop that here.
62 */ 62 */
63 static void release_buffer_page(struct buffer_ 63 static void release_buffer_page(struct buffer_head *bh)
64 { 64 {
65 struct folio *folio; 65 struct folio *folio;
>> 66 struct page *page;
66 67
67 if (buffer_dirty(bh)) 68 if (buffer_dirty(bh))
68 goto nope; 69 goto nope;
69 if (atomic_read(&bh->b_count) != 1) 70 if (atomic_read(&bh->b_count) != 1)
70 goto nope; 71 goto nope;
71 folio = bh->b_folio; !! 72 page = bh->b_page;
>> 73 if (!page)
>> 74 goto nope;
>> 75 folio = page_folio(page);
72 if (folio->mapping) 76 if (folio->mapping)
73 goto nope; 77 goto nope;
74 78
75 /* OK, it's a truncated page */ 79 /* OK, it's a truncated page */
76 if (!folio_trylock(folio)) 80 if (!folio_trylock(folio))
77 goto nope; 81 goto nope;
78 82
79 folio_get(folio); 83 folio_get(folio);
80 __brelse(bh); 84 __brelse(bh);
81 try_to_free_buffers(folio); 85 try_to_free_buffers(folio);
82 folio_unlock(folio); 86 folio_unlock(folio);
83 folio_put(folio); 87 folio_put(folio);
84 return; 88 return;
85 89
86 nope: 90 nope:
87 __brelse(bh); 91 __brelse(bh);
88 } 92 }
89 93
90 static void jbd2_commit_block_csum_set(journal 94 static void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
91 { 95 {
92 struct commit_header *h; 96 struct commit_header *h;
93 __u32 csum; 97 __u32 csum;
94 98
95 if (!jbd2_journal_has_csum_v2or3(j)) 99 if (!jbd2_journal_has_csum_v2or3(j))
96 return; 100 return;
97 101
98 h = (struct commit_header *)(bh->b_dat 102 h = (struct commit_header *)(bh->b_data);
99 h->h_chksum_type = 0; 103 h->h_chksum_type = 0;
100 h->h_chksum_size = 0; 104 h->h_chksum_size = 0;
101 h->h_chksum[0] = 0; 105 h->h_chksum[0] = 0;
102 csum = jbd2_chksum(j, j->j_csum_seed, 106 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
103 h->h_chksum[0] = cpu_to_be32(csum); 107 h->h_chksum[0] = cpu_to_be32(csum);
104 } 108 }
105 109
106 /* 110 /*
107 * Done it all: now submit the commit record. 111 * Done it all: now submit the commit record. We should have
108 * cleaned up our previous buffers by now, so 112 * cleaned up our previous buffers by now, so if we are in abort
109 * mode we can now just skip the rest of the j 113 * mode we can now just skip the rest of the journal write
110 * entirely. 114 * entirely.
111 * 115 *
112 * Returns 1 if the journal needs to be aborte 116 * Returns 1 if the journal needs to be aborted or 0 on success
113 */ 117 */
114 static int journal_submit_commit_record(journa 118 static int journal_submit_commit_record(journal_t *journal,
115 transa 119 transaction_t *commit_transaction,
116 struct 120 struct buffer_head **cbh,
117 __u32 121 __u32 crc32_sum)
118 { 122 {
119 struct commit_header *tmp; 123 struct commit_header *tmp;
120 struct buffer_head *bh; 124 struct buffer_head *bh;
121 struct timespec64 now; 125 struct timespec64 now;
122 blk_opf_t write_flags = REQ_OP_WRITE | 126 blk_opf_t write_flags = REQ_OP_WRITE | JBD2_JOURNAL_REQ_FLAGS;
123 127
124 *cbh = NULL; 128 *cbh = NULL;
125 129
126 if (is_journal_aborted(journal)) 130 if (is_journal_aborted(journal))
127 return 0; 131 return 0;
128 132
129 bh = jbd2_journal_get_descriptor_buffe 133 bh = jbd2_journal_get_descriptor_buffer(commit_transaction,
130 134 JBD2_COMMIT_BLOCK);
131 if (!bh) 135 if (!bh)
132 return 1; 136 return 1;
133 137
134 tmp = (struct commit_header *)bh->b_da 138 tmp = (struct commit_header *)bh->b_data;
135 ktime_get_coarse_real_ts64(&now); 139 ktime_get_coarse_real_ts64(&now);
136 tmp->h_commit_sec = cpu_to_be64(now.tv 140 tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
137 tmp->h_commit_nsec = cpu_to_be32(now.t 141 tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
138 142
139 if (jbd2_has_feature_checksum(journal) 143 if (jbd2_has_feature_checksum(journal)) {
140 tmp->h_chksum_type = JBD2 144 tmp->h_chksum_type = JBD2_CRC32_CHKSUM;
141 tmp->h_chksum_size = JBD2 145 tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE;
142 tmp->h_chksum[0] = cpu_ 146 tmp->h_chksum[0] = cpu_to_be32(crc32_sum);
143 } 147 }
144 jbd2_commit_block_csum_set(journal, bh 148 jbd2_commit_block_csum_set(journal, bh);
145 149
146 BUFFER_TRACE(bh, "submit commit block" 150 BUFFER_TRACE(bh, "submit commit block");
147 lock_buffer(bh); 151 lock_buffer(bh);
148 clear_buffer_dirty(bh); 152 clear_buffer_dirty(bh);
149 set_buffer_uptodate(bh); 153 set_buffer_uptodate(bh);
150 bh->b_end_io = journal_end_buffer_io_s 154 bh->b_end_io = journal_end_buffer_io_sync;
151 155
152 if (journal->j_flags & JBD2_BARRIER && 156 if (journal->j_flags & JBD2_BARRIER &&
153 !jbd2_has_feature_async_commit(jou 157 !jbd2_has_feature_async_commit(journal))
154 write_flags |= REQ_PREFLUSH | 158 write_flags |= REQ_PREFLUSH | REQ_FUA;
155 159
156 submit_bh(write_flags, bh); 160 submit_bh(write_flags, bh);
157 *cbh = bh; 161 *cbh = bh;
158 return 0; 162 return 0;
159 } 163 }
160 164
161 /* 165 /*
162 * This function along with journal_submit_com 166 * This function along with journal_submit_commit_record
163 * allows to write the commit record asynchron 167 * allows to write the commit record asynchronously.
164 */ 168 */
165 static int journal_wait_on_commit_record(journ 169 static int journal_wait_on_commit_record(journal_t *journal,
166 struc 170 struct buffer_head *bh)
167 { 171 {
168 int ret = 0; 172 int ret = 0;
169 173
170 clear_buffer_dirty(bh); 174 clear_buffer_dirty(bh);
171 wait_on_buffer(bh); 175 wait_on_buffer(bh);
172 176
173 if (unlikely(!buffer_uptodate(bh))) 177 if (unlikely(!buffer_uptodate(bh)))
174 ret = -EIO; 178 ret = -EIO;
175 put_bh(bh); /* One for getb 179 put_bh(bh); /* One for getblk() */
176 180
177 return ret; 181 return ret;
178 } 182 }
179 183
>> 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
>> 187 * use writepages() because with delayed allocation we may be doing
>> 188 * block allocation in writepages().
>> 189 */
>> 190 int jbd2_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
>> 191 {
>> 192 struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
>> 193 struct writeback_control wbc = {
>> 194 .sync_mode = WB_SYNC_ALL,
>> 195 .nr_to_write = mapping->nrpages * 2,
>> 196 .range_start = jinode->i_dirty_start,
>> 197 .range_end = jinode->i_dirty_end,
>> 198 };
>> 199
>> 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 */
>> 206 return generic_writepages(mapping, &wbc);
>> 207 }
>> 208
180 /* Send all the data buffers related to an ino 209 /* Send all the data buffers related to an inode */
181 int jbd2_submit_inode_data(journal_t *journal, !! 210 int jbd2_submit_inode_data(struct jbd2_inode *jinode)
182 { 211 {
>> 212
183 if (!jinode || !(jinode->i_flags & JI_ 213 if (!jinode || !(jinode->i_flags & JI_WRITE_DATA))
184 return 0; 214 return 0;
185 215
186 trace_jbd2_submit_inode_data(jinode->i 216 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
187 return journal->j_submit_inode_data_bu !! 217 return jbd2_journal_submit_inode_data_buffers(jinode);
188 218
189 } 219 }
190 EXPORT_SYMBOL(jbd2_submit_inode_data); 220 EXPORT_SYMBOL(jbd2_submit_inode_data);
191 221
192 int jbd2_wait_inode_data(journal_t *journal, s 222 int jbd2_wait_inode_data(journal_t *journal, struct jbd2_inode *jinode)
193 { 223 {
194 if (!jinode || !(jinode->i_flags & JI_ 224 if (!jinode || !(jinode->i_flags & JI_WAIT_DATA) ||
195 !jinode->i_vfs_inode || !jinod 225 !jinode->i_vfs_inode || !jinode->i_vfs_inode->i_mapping)
196 return 0; 226 return 0;
197 return filemap_fdatawait_range_keep_er 227 return filemap_fdatawait_range_keep_errors(
198 jinode->i_vfs_inode->i_mapping 228 jinode->i_vfs_inode->i_mapping, jinode->i_dirty_start,
199 jinode->i_dirty_end); 229 jinode->i_dirty_end);
200 } 230 }
201 EXPORT_SYMBOL(jbd2_wait_inode_data); 231 EXPORT_SYMBOL(jbd2_wait_inode_data);
202 232
203 /* 233 /*
204 * Submit all the data buffers of inode associ 234 * Submit all the data buffers of inode associated with the transaction to
205 * disk. 235 * disk.
206 * 236 *
207 * We are in a committing transaction. Therefo 237 * We are in a committing transaction. Therefore no new inode can be added to
208 * our inode list. We use JI_COMMIT_RUNNING fl 238 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
209 * operate on from being released while we wri 239 * operate on from being released while we write out pages.
210 */ 240 */
211 static int journal_submit_data_buffers(journal 241 static int journal_submit_data_buffers(journal_t *journal,
212 transaction_t *commit_transact 242 transaction_t *commit_transaction)
213 { 243 {
214 struct jbd2_inode *jinode; 244 struct jbd2_inode *jinode;
215 int err, ret = 0; 245 int err, ret = 0;
216 246
217 spin_lock(&journal->j_list_lock); 247 spin_lock(&journal->j_list_lock);
218 list_for_each_entry(jinode, &commit_tr 248 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
219 if (!(jinode->i_flags & JI_WRI 249 if (!(jinode->i_flags & JI_WRITE_DATA))
220 continue; 250 continue;
221 jinode->i_flags |= JI_COMMIT_R 251 jinode->i_flags |= JI_COMMIT_RUNNING;
222 spin_unlock(&journal->j_list_l 252 spin_unlock(&journal->j_list_lock);
223 /* submit the inode data buffe 253 /* submit the inode data buffers. */
224 trace_jbd2_submit_inode_data(j 254 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
225 if (journal->j_submit_inode_da 255 if (journal->j_submit_inode_data_buffers) {
226 err = journal->j_submi 256 err = journal->j_submit_inode_data_buffers(jinode);
227 if (!ret) 257 if (!ret)
228 ret = err; 258 ret = err;
229 } 259 }
230 spin_lock(&journal->j_list_loc 260 spin_lock(&journal->j_list_lock);
231 J_ASSERT(jinode->i_transaction 261 J_ASSERT(jinode->i_transaction == commit_transaction);
232 jinode->i_flags &= ~JI_COMMIT_ 262 jinode->i_flags &= ~JI_COMMIT_RUNNING;
233 smp_mb(); 263 smp_mb();
234 wake_up_bit(&jinode->i_flags, 264 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
235 } 265 }
236 spin_unlock(&journal->j_list_lock); 266 spin_unlock(&journal->j_list_lock);
237 return ret; 267 return ret;
238 } 268 }
239 269
240 int jbd2_journal_finish_inode_data_buffers(str 270 int jbd2_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
241 { 271 {
242 struct address_space *mapping = jinode 272 struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
243 273
244 return filemap_fdatawait_range_keep_er 274 return filemap_fdatawait_range_keep_errors(mapping,
245 275 jinode->i_dirty_start,
246 276 jinode->i_dirty_end);
247 } 277 }
248 278
249 /* 279 /*
250 * Wait for data submitted for writeout, refil 280 * Wait for data submitted for writeout, refile inodes to proper
251 * transaction if needed. 281 * transaction if needed.
252 * 282 *
253 */ 283 */
254 static int journal_finish_inode_data_buffers(j 284 static int journal_finish_inode_data_buffers(journal_t *journal,
255 transaction_t *commit_transact 285 transaction_t *commit_transaction)
256 { 286 {
257 struct jbd2_inode *jinode, *next_i; 287 struct jbd2_inode *jinode, *next_i;
258 int err, ret = 0; 288 int err, ret = 0;
259 289
260 /* For locking, see the comment in jou 290 /* For locking, see the comment in journal_submit_data_buffers() */
261 spin_lock(&journal->j_list_lock); 291 spin_lock(&journal->j_list_lock);
262 list_for_each_entry(jinode, &commit_tr 292 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
263 if (!(jinode->i_flags & JI_WAI 293 if (!(jinode->i_flags & JI_WAIT_DATA))
264 continue; 294 continue;
265 jinode->i_flags |= JI_COMMIT_R 295 jinode->i_flags |= JI_COMMIT_RUNNING;
266 spin_unlock(&journal->j_list_l 296 spin_unlock(&journal->j_list_lock);
267 /* wait for the inode data buf 297 /* wait for the inode data buffers writeout. */
268 if (journal->j_finish_inode_da 298 if (journal->j_finish_inode_data_buffers) {
269 err = journal->j_finis 299 err = journal->j_finish_inode_data_buffers(jinode);
270 if (!ret) 300 if (!ret)
271 ret = err; 301 ret = err;
272 } 302 }
273 cond_resched(); 303 cond_resched();
274 spin_lock(&journal->j_list_loc 304 spin_lock(&journal->j_list_lock);
275 jinode->i_flags &= ~JI_COMMIT_ 305 jinode->i_flags &= ~JI_COMMIT_RUNNING;
276 smp_mb(); 306 smp_mb();
277 wake_up_bit(&jinode->i_flags, 307 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
278 } 308 }
279 309
280 /* Now refile inode to proper lists */ 310 /* Now refile inode to proper lists */
281 list_for_each_entry_safe(jinode, next_ 311 list_for_each_entry_safe(jinode, next_i,
282 &commit_trans 312 &commit_transaction->t_inode_list, i_list) {
283 list_del(&jinode->i_list); 313 list_del(&jinode->i_list);
284 if (jinode->i_next_transaction 314 if (jinode->i_next_transaction) {
285 jinode->i_transaction 315 jinode->i_transaction = jinode->i_next_transaction;
286 jinode->i_next_transac 316 jinode->i_next_transaction = NULL;
287 list_add(&jinode->i_li 317 list_add(&jinode->i_list,
288 &jinode->i_tra 318 &jinode->i_transaction->t_inode_list);
289 } else { 319 } else {
290 jinode->i_transaction 320 jinode->i_transaction = NULL;
291 jinode->i_dirty_start 321 jinode->i_dirty_start = 0;
292 jinode->i_dirty_end = 322 jinode->i_dirty_end = 0;
293 } 323 }
294 } 324 }
295 spin_unlock(&journal->j_list_lock); 325 spin_unlock(&journal->j_list_lock);
296 326
297 return ret; 327 return ret;
298 } 328 }
299 329
300 static __u32 jbd2_checksum_data(__u32 crc32_su 330 static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
301 { 331 {
>> 332 struct page *page = bh->b_page;
302 char *addr; 333 char *addr;
303 __u32 checksum; 334 __u32 checksum;
304 335
305 addr = kmap_local_folio(bh->b_folio, b !! 336 addr = kmap_atomic(page);
306 checksum = crc32_be(crc32_sum, addr, b !! 337 checksum = crc32_be(crc32_sum,
307 kunmap_local(addr); !! 338 (void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
>> 339 kunmap_atomic(addr);
308 340
309 return checksum; 341 return checksum;
310 } 342 }
311 343
312 static void write_tag_block(journal_t *j, jour 344 static void write_tag_block(journal_t *j, journal_block_tag_t *tag,
313 unsigned lo 345 unsigned long long block)
314 { 346 {
315 tag->t_blocknr = cpu_to_be32(block & ( 347 tag->t_blocknr = cpu_to_be32(block & (u32)~0);
316 if (jbd2_has_feature_64bit(j)) 348 if (jbd2_has_feature_64bit(j))
317 tag->t_blocknr_high = cpu_to_b 349 tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
318 } 350 }
319 351
320 static void jbd2_block_tag_csum_set(journal_t 352 static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
321 struct buf 353 struct buffer_head *bh, __u32 sequence)
322 { 354 {
323 journal_block_tag3_t *tag3 = (journal_ 355 journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
>> 356 struct page *page = bh->b_page;
324 __u8 *addr; 357 __u8 *addr;
325 __u32 csum32; 358 __u32 csum32;
326 __be32 seq; 359 __be32 seq;
327 360
328 if (!jbd2_journal_has_csum_v2or3(j)) 361 if (!jbd2_journal_has_csum_v2or3(j))
329 return; 362 return;
330 363
331 seq = cpu_to_be32(sequence); 364 seq = cpu_to_be32(sequence);
332 addr = kmap_local_folio(bh->b_folio, b !! 365 addr = kmap_atomic(page);
333 csum32 = jbd2_chksum(j, j->j_csum_seed 366 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
334 csum32 = jbd2_chksum(j, csum32, addr, !! 367 csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data),
335 kunmap_local(addr); !! 368 bh->b_size);
>> 369 kunmap_atomic(addr);
336 370
337 if (jbd2_has_feature_csum3(j)) 371 if (jbd2_has_feature_csum3(j))
338 tag3->t_checksum = cpu_to_be32 372 tag3->t_checksum = cpu_to_be32(csum32);
339 else 373 else
340 tag->t_checksum = cpu_to_be16( 374 tag->t_checksum = cpu_to_be16(csum32);
341 } 375 }
342 /* 376 /*
343 * jbd2_journal_commit_transaction 377 * jbd2_journal_commit_transaction
344 * 378 *
345 * The primary function for committing a trans 379 * The primary function for committing a transaction to the log. This
346 * function is called by the journal thread to 380 * function is called by the journal thread to begin a complete commit.
347 */ 381 */
348 void jbd2_journal_commit_transaction(journal_t 382 void jbd2_journal_commit_transaction(journal_t *journal)
349 { 383 {
350 struct transaction_stats_s stats; 384 struct transaction_stats_s stats;
351 transaction_t *commit_transaction; 385 transaction_t *commit_transaction;
352 struct journal_head *jh; 386 struct journal_head *jh;
353 struct buffer_head *descriptor; 387 struct buffer_head *descriptor;
354 struct buffer_head **wbuf = journal->j 388 struct buffer_head **wbuf = journal->j_wbuf;
355 int bufs; 389 int bufs;
356 int escape; !! 390 int flags;
357 int err; 391 int err;
358 unsigned long long blocknr; 392 unsigned long long blocknr;
359 ktime_t start_time; 393 ktime_t start_time;
360 u64 commit_time; 394 u64 commit_time;
361 char *tagp = NULL; 395 char *tagp = NULL;
362 journal_block_tag_t *tag = NULL; 396 journal_block_tag_t *tag = NULL;
363 int space_left = 0; 397 int space_left = 0;
364 int first_tag = 0; 398 int first_tag = 0;
365 int tag_flag; 399 int tag_flag;
366 int i; 400 int i;
367 int tag_bytes = journal_tag_bytes(jour 401 int tag_bytes = journal_tag_bytes(journal);
368 struct buffer_head *cbh = NULL; /* For 402 struct buffer_head *cbh = NULL; /* For transactional checksums */
369 __u32 crc32_sum = ~0; 403 __u32 crc32_sum = ~0;
370 struct blk_plug plug; 404 struct blk_plug plug;
371 /* Tail of the journal */ 405 /* Tail of the journal */
372 unsigned long first_block; 406 unsigned long first_block;
373 tid_t first_tid; 407 tid_t first_tid;
374 int update_tail; 408 int update_tail;
375 int csum_size = 0; 409 int csum_size = 0;
376 LIST_HEAD(io_bufs); 410 LIST_HEAD(io_bufs);
377 LIST_HEAD(log_bufs); 411 LIST_HEAD(log_bufs);
378 412
379 if (jbd2_journal_has_csum_v2or3(journa 413 if (jbd2_journal_has_csum_v2or3(journal))
380 csum_size = sizeof(struct jbd2 414 csum_size = sizeof(struct jbd2_journal_block_tail);
381 415
382 /* 416 /*
383 * First job: lock down the current tr 417 * First job: lock down the current transaction and wait for
384 * all outstanding updates to complete 418 * all outstanding updates to complete.
385 */ 419 */
386 420
387 /* Do we need to erase the effects of 421 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
388 if (journal->j_flags & JBD2_FLUSHED) { 422 if (journal->j_flags & JBD2_FLUSHED) {
389 jbd2_debug(3, "super block upd 423 jbd2_debug(3, "super block updated\n");
390 mutex_lock_io(&journal->j_chec 424 mutex_lock_io(&journal->j_checkpoint_mutex);
391 /* 425 /*
392 * We hold j_checkpoint_mutex 426 * We hold j_checkpoint_mutex so tail cannot change under us.
393 * We don't need any special d 427 * We don't need any special data guarantees for writing sb
394 * since journal is empty and 428 * since journal is empty and it is ok for write to be
395 * flushed only with transacti 429 * flushed only with transaction commit.
396 */ 430 */
397 jbd2_journal_update_sb_log_tai 431 jbd2_journal_update_sb_log_tail(journal,
398 432 journal->j_tail_sequence,
399 433 journal->j_tail, 0);
400 mutex_unlock(&journal->j_check 434 mutex_unlock(&journal->j_checkpoint_mutex);
401 } else { 435 } else {
402 jbd2_debug(3, "superblock not 436 jbd2_debug(3, "superblock not updated\n");
403 } 437 }
404 438
405 J_ASSERT(journal->j_running_transactio 439 J_ASSERT(journal->j_running_transaction != NULL);
406 J_ASSERT(journal->j_committing_transac 440 J_ASSERT(journal->j_committing_transaction == NULL);
407 441
408 write_lock(&journal->j_state_lock); 442 write_lock(&journal->j_state_lock);
409 journal->j_flags |= JBD2_FULL_COMMIT_O 443 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
410 while (journal->j_flags & JBD2_FAST_CO 444 while (journal->j_flags & JBD2_FAST_COMMIT_ONGOING) {
411 DEFINE_WAIT(wait); 445 DEFINE_WAIT(wait);
412 446
413 prepare_to_wait(&journal->j_fc 447 prepare_to_wait(&journal->j_fc_wait, &wait,
414 TASK_UNINTERRU 448 TASK_UNINTERRUPTIBLE);
415 write_unlock(&journal->j_state 449 write_unlock(&journal->j_state_lock);
416 schedule(); 450 schedule();
417 write_lock(&journal->j_state_l 451 write_lock(&journal->j_state_lock);
418 finish_wait(&journal->j_fc_wai 452 finish_wait(&journal->j_fc_wait, &wait);
419 /* 453 /*
420 * TODO: by blocking fast comm 454 * TODO: by blocking fast commits here, we are increasing
421 * fsync() latency slightly. S 455 * fsync() latency slightly. Strictly speaking, we don't need
422 * to block fast commits until 456 * to block fast commits until the transaction enters T_FLUSH
423 * state. So an optimization i 457 * state. So an optimization is possible where we block new fast
424 * commits here and wait for e 458 * commits here and wait for existing ones to complete
425 * just before we enter T_FLUS 459 * just before we enter T_FLUSH. That way, the existing fast
426 * commits and this full commi 460 * commits and this full commit can proceed parallely.
427 */ 461 */
428 } 462 }
429 write_unlock(&journal->j_state_lock); 463 write_unlock(&journal->j_state_lock);
430 464
431 commit_transaction = journal->j_runnin 465 commit_transaction = journal->j_running_transaction;
432 466
433 trace_jbd2_start_commit(journal, commi 467 trace_jbd2_start_commit(journal, commit_transaction);
434 jbd2_debug(1, "JBD2: starting commit o 468 jbd2_debug(1, "JBD2: starting commit of transaction %d\n",
435 commit_transaction->t_ 469 commit_transaction->t_tid);
436 470
437 write_lock(&journal->j_state_lock); 471 write_lock(&journal->j_state_lock);
438 journal->j_fc_off = 0; 472 journal->j_fc_off = 0;
439 J_ASSERT(commit_transaction->t_state = 473 J_ASSERT(commit_transaction->t_state == T_RUNNING);
440 commit_transaction->t_state = T_LOCKED 474 commit_transaction->t_state = T_LOCKED;
441 475
442 trace_jbd2_commit_locking(journal, com 476 trace_jbd2_commit_locking(journal, commit_transaction);
443 stats.run.rs_wait = commit_transaction 477 stats.run.rs_wait = commit_transaction->t_max_wait;
444 stats.run.rs_request_delay = 0; 478 stats.run.rs_request_delay = 0;
445 stats.run.rs_locked = jiffies; 479 stats.run.rs_locked = jiffies;
446 if (commit_transaction->t_requested) 480 if (commit_transaction->t_requested)
447 stats.run.rs_request_delay = 481 stats.run.rs_request_delay =
448 jbd2_time_diff(commit_ 482 jbd2_time_diff(commit_transaction->t_requested,
449 stats.r 483 stats.run.rs_locked);
450 stats.run.rs_running = jbd2_time_diff( 484 stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
451 485 stats.run.rs_locked);
452 486
453 // waits for any t_updates to finish 487 // waits for any t_updates to finish
454 jbd2_journal_wait_updates(journal); 488 jbd2_journal_wait_updates(journal);
455 489
456 commit_transaction->t_state = T_SWITCH 490 commit_transaction->t_state = T_SWITCH;
457 491
458 J_ASSERT (atomic_read(&commit_transact 492 J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
459 journal->j_max_transac 493 journal->j_max_transaction_buffers);
460 494
461 /* 495 /*
462 * First thing we are allowed to do is 496 * First thing we are allowed to do is to discard any remaining
463 * BJ_Reserved buffers. Note, it is _ 497 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
464 * that there are no such buffers: if 498 * that there are no such buffers: if a large filesystem
465 * operation like a truncate needs to 499 * operation like a truncate needs to split itself over multiple
466 * transactions, then it may try to do 500 * transactions, then it may try to do a jbd2_journal_restart() while
467 * there are still BJ_Reserved buffers 501 * there are still BJ_Reserved buffers outstanding. These must
468 * be released cleanly from the curren 502 * be released cleanly from the current transaction.
469 * 503 *
470 * In this case, the filesystem must s 504 * In this case, the filesystem must still reserve write access
471 * again before modifying the buffer i 505 * again before modifying the buffer in the new transaction, but
472 * we do not require it to remember ex 506 * we do not require it to remember exactly which old buffers it
473 * has reserved. This is consistent w 507 * has reserved. This is consistent with the existing behaviour
474 * that multiple jbd2_journal_get_writ 508 * that multiple jbd2_journal_get_write_access() calls to the same
475 * buffer are perfectly permissible. 509 * buffer are perfectly permissible.
476 * We use journal->j_state_lock here t 510 * We use journal->j_state_lock here to serialize processing of
477 * t_reserved_list with eviction of bu 511 * t_reserved_list with eviction of buffers from journal_unmap_buffer().
478 */ 512 */
479 while (commit_transaction->t_reserved_ 513 while (commit_transaction->t_reserved_list) {
480 jh = commit_transaction->t_res 514 jh = commit_transaction->t_reserved_list;
481 JBUFFER_TRACE(jh, "reserved, u 515 JBUFFER_TRACE(jh, "reserved, unused: refile");
482 /* 516 /*
483 * A jbd2_journal_get_undo_acc 517 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
484 * leave undo-committed data. 518 * leave undo-committed data.
485 */ 519 */
486 if (jh->b_committed_data) { 520 if (jh->b_committed_data) {
487 struct buffer_head *bh 521 struct buffer_head *bh = jh2bh(jh);
488 522
489 spin_lock(&jh->b_state 523 spin_lock(&jh->b_state_lock);
490 jbd2_free(jh->b_commit 524 jbd2_free(jh->b_committed_data, bh->b_size);
491 jh->b_committed_data = 525 jh->b_committed_data = NULL;
492 spin_unlock(&jh->b_sta 526 spin_unlock(&jh->b_state_lock);
493 } 527 }
494 jbd2_journal_refile_buffer(jou 528 jbd2_journal_refile_buffer(journal, jh);
495 } 529 }
496 530
497 write_unlock(&journal->j_state_lock); 531 write_unlock(&journal->j_state_lock);
498 /* 532 /*
499 * Now try to drop any written-back bu 533 * Now try to drop any written-back buffers from the journal's
500 * checkpoint lists. We do this *befo 534 * checkpoint lists. We do this *before* commit because it potentially
501 * frees some memory 535 * frees some memory
502 */ 536 */
503 spin_lock(&journal->j_list_lock); 537 spin_lock(&journal->j_list_lock);
504 __jbd2_journal_clean_checkpoint_list(j !! 538 __jbd2_journal_clean_checkpoint_list(journal, false);
505 spin_unlock(&journal->j_list_lock); 539 spin_unlock(&journal->j_list_lock);
506 540
507 jbd2_debug(3, "JBD2: commit phase 1\n" 541 jbd2_debug(3, "JBD2: commit phase 1\n");
508 542
509 /* 543 /*
510 * Clear revoked flag to reflect there 544 * Clear revoked flag to reflect there is no revoked buffers
511 * in the next transaction which is go 545 * in the next transaction which is going to be started.
512 */ 546 */
513 jbd2_clear_buffer_revoked_flags(journa 547 jbd2_clear_buffer_revoked_flags(journal);
514 548
515 /* 549 /*
516 * Switch to a new revoke table. 550 * Switch to a new revoke table.
517 */ 551 */
518 jbd2_journal_switch_revoke_table(journ 552 jbd2_journal_switch_revoke_table(journal);
519 553
520 write_lock(&journal->j_state_lock); 554 write_lock(&journal->j_state_lock);
521 /* 555 /*
522 * Reserved credits cannot be claimed 556 * Reserved credits cannot be claimed anymore, free them
523 */ 557 */
524 atomic_sub(atomic_read(&journal->j_res 558 atomic_sub(atomic_read(&journal->j_reserved_credits),
525 &commit_transaction->t_outs 559 &commit_transaction->t_outstanding_credits);
526 560
527 trace_jbd2_commit_flushing(journal, co 561 trace_jbd2_commit_flushing(journal, commit_transaction);
528 stats.run.rs_flushing = jiffies; 562 stats.run.rs_flushing = jiffies;
529 stats.run.rs_locked = jbd2_time_diff(s 563 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
530 s 564 stats.run.rs_flushing);
531 565
532 commit_transaction->t_state = T_FLUSH; 566 commit_transaction->t_state = T_FLUSH;
533 journal->j_committing_transaction = co 567 journal->j_committing_transaction = commit_transaction;
534 journal->j_running_transaction = NULL; 568 journal->j_running_transaction = NULL;
535 start_time = ktime_get(); 569 start_time = ktime_get();
536 commit_transaction->t_log_start = jour 570 commit_transaction->t_log_start = journal->j_head;
537 wake_up_all(&journal->j_wait_transacti 571 wake_up_all(&journal->j_wait_transaction_locked);
538 write_unlock(&journal->j_state_lock); 572 write_unlock(&journal->j_state_lock);
539 573
540 jbd2_debug(3, "JBD2: commit phase 2a\n 574 jbd2_debug(3, "JBD2: commit phase 2a\n");
541 575
542 /* 576 /*
543 * Now start flushing things to disk, 577 * Now start flushing things to disk, in the order they appear
544 * on the transaction lists. Data blo 578 * on the transaction lists. Data blocks go first.
545 */ 579 */
546 err = journal_submit_data_buffers(jour 580 err = journal_submit_data_buffers(journal, commit_transaction);
547 if (err) 581 if (err)
548 jbd2_journal_abort(journal, er 582 jbd2_journal_abort(journal, err);
549 583
550 blk_start_plug(&plug); 584 blk_start_plug(&plug);
551 jbd2_journal_write_revoke_records(comm 585 jbd2_journal_write_revoke_records(commit_transaction, &log_bufs);
552 586
553 jbd2_debug(3, "JBD2: commit phase 2b\n 587 jbd2_debug(3, "JBD2: commit phase 2b\n");
554 588
555 /* 589 /*
556 * Way to go: we have now written out 590 * Way to go: we have now written out all of the data for a
557 * transaction! Now comes the tricky 591 * transaction! Now comes the tricky part: we need to write out
558 * metadata. Loop over the transactio 592 * metadata. Loop over the transaction's entire buffer list:
559 */ 593 */
560 write_lock(&journal->j_state_lock); 594 write_lock(&journal->j_state_lock);
561 commit_transaction->t_state = T_COMMIT 595 commit_transaction->t_state = T_COMMIT;
562 write_unlock(&journal->j_state_lock); 596 write_unlock(&journal->j_state_lock);
563 597
564 trace_jbd2_commit_logging(journal, com 598 trace_jbd2_commit_logging(journal, commit_transaction);
565 stats.run.rs_logging = jiffies; 599 stats.run.rs_logging = jiffies;
566 stats.run.rs_flushing = jbd2_time_diff 600 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
567 601 stats.run.rs_logging);
568 stats.run.rs_blocks = commit_transacti 602 stats.run.rs_blocks = commit_transaction->t_nr_buffers;
569 stats.run.rs_blocks_logged = 0; 603 stats.run.rs_blocks_logged = 0;
570 604
571 J_ASSERT(commit_transaction->t_nr_buff 605 J_ASSERT(commit_transaction->t_nr_buffers <=
572 atomic_read(&commit_transacti 606 atomic_read(&commit_transaction->t_outstanding_credits));
573 607
>> 608 err = 0;
574 bufs = 0; 609 bufs = 0;
575 descriptor = NULL; 610 descriptor = NULL;
576 while (commit_transaction->t_buffers) 611 while (commit_transaction->t_buffers) {
577 612
578 /* Find the next buffer to be 613 /* Find the next buffer to be journaled... */
579 614
580 jh = commit_transaction->t_buf 615 jh = commit_transaction->t_buffers;
581 616
582 /* If we're in abort mode, we 617 /* If we're in abort mode, we just un-journal the buffer and
583 release it. */ 618 release it. */
584 619
585 if (is_journal_aborted(journal 620 if (is_journal_aborted(journal)) {
586 clear_buffer_jbddirty( 621 clear_buffer_jbddirty(jh2bh(jh));
587 JBUFFER_TRACE(jh, "jou 622 JBUFFER_TRACE(jh, "journal is aborting: refile");
588 jbd2_buffer_abort_trig 623 jbd2_buffer_abort_trigger(jh,
589 624 jh->b_frozen_data ?
590 625 jh->b_frozen_triggers :
591 626 jh->b_triggers);
592 jbd2_journal_refile_bu 627 jbd2_journal_refile_buffer(journal, jh);
593 /* If that was the las 628 /* If that was the last one, we need to clean up
594 * any descriptor buff 629 * any descriptor buffers which may have been
595 * already allocated, 630 * already allocated, even if we are now
596 * aborting. */ 631 * aborting. */
597 if (!commit_transactio 632 if (!commit_transaction->t_buffers)
598 goto start_jou 633 goto start_journal_io;
599 continue; 634 continue;
600 } 635 }
601 636
602 /* Make sure we have a descrip 637 /* Make sure we have a descriptor block in which to
603 record the metadata buffer. 638 record the metadata buffer. */
604 639
605 if (!descriptor) { 640 if (!descriptor) {
606 J_ASSERT (bufs == 0); 641 J_ASSERT (bufs == 0);
607 642
608 jbd2_debug(4, "JBD2: g 643 jbd2_debug(4, "JBD2: get descriptor\n");
609 644
610 descriptor = jbd2_jour 645 descriptor = jbd2_journal_get_descriptor_buffer(
611 646 commit_transaction,
612 647 JBD2_DESCRIPTOR_BLOCK);
613 if (!descriptor) { 648 if (!descriptor) {
614 jbd2_journal_a 649 jbd2_journal_abort(journal, -EIO);
615 continue; 650 continue;
616 } 651 }
617 652
618 jbd2_debug(4, "JBD2: g 653 jbd2_debug(4, "JBD2: got buffer %llu (%p)\n",
619 (unsigned long 654 (unsigned long long)descriptor->b_blocknr,
620 descriptor->b_ 655 descriptor->b_data);
621 tagp = &descriptor->b_ 656 tagp = &descriptor->b_data[sizeof(journal_header_t)];
622 space_left = descripto 657 space_left = descriptor->b_size -
623 658 sizeof(journal_header_t);
624 first_tag = 1; 659 first_tag = 1;
625 set_buffer_jwrite(desc 660 set_buffer_jwrite(descriptor);
626 set_buffer_dirty(descr 661 set_buffer_dirty(descriptor);
627 wbuf[bufs++] = descrip 662 wbuf[bufs++] = descriptor;
628 663
629 /* Record it so that w 664 /* Record it so that we can wait for IO
630 completion later */ 665 completion later */
631 BUFFER_TRACE(descripto 666 BUFFER_TRACE(descriptor, "ph3: file as descriptor");
632 jbd2_file_log_bh(&log_ 667 jbd2_file_log_bh(&log_bufs, descriptor);
633 } 668 }
634 669
635 /* Where is the buffer to be w 670 /* Where is the buffer to be written? */
636 671
637 err = jbd2_journal_next_log_bl 672 err = jbd2_journal_next_log_block(journal, &blocknr);
638 /* If the block mapping failed 673 /* If the block mapping failed, just abandon the buffer
639 and repeat this loop: we'll 674 and repeat this loop: we'll fall into the
640 refile-on-abort condition a 675 refile-on-abort condition above. */
641 if (err) { 676 if (err) {
642 jbd2_journal_abort(jou 677 jbd2_journal_abort(journal, err);
643 continue; 678 continue;
644 } 679 }
645 680
646 /* 681 /*
647 * start_this_handle() uses t_ 682 * start_this_handle() uses t_outstanding_credits to determine
648 * the free space in the log. 683 * the free space in the log.
649 */ 684 */
650 atomic_dec(&commit_transaction 685 atomic_dec(&commit_transaction->t_outstanding_credits);
651 686
652 /* Bump b_count to prevent tru 687 /* Bump b_count to prevent truncate from stumbling over
653 the shadowed buffer! @@@ T 688 the shadowed buffer! @@@ This can go if we ever get
654 rid of the shadow pairing o 689 rid of the shadow pairing of buffers. */
655 atomic_inc(&jh2bh(jh)->b_count 690 atomic_inc(&jh2bh(jh)->b_count);
656 691
657 /* 692 /*
658 * Make a temporary IO buffer 693 * Make a temporary IO buffer with which to write it out
659 * (this will requeue the meta 694 * (this will requeue the metadata buffer to BJ_Shadow).
660 */ 695 */
661 set_bit(BH_JWrite, &jh2bh(jh)- 696 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
662 JBUFFER_TRACE(jh, "ph3: write 697 JBUFFER_TRACE(jh, "ph3: write metadata");
663 escape = jbd2_journal_write_me !! 698 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
664 699 jh, &wbuf[bufs], blocknr);
665 if (escape < 0) { !! 700 if (flags < 0) {
666 jbd2_journal_abort(jou !! 701 jbd2_journal_abort(journal, flags);
667 continue; 702 continue;
668 } 703 }
669 jbd2_file_log_bh(&io_bufs, wbu 704 jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
670 705
671 /* Record the new block's tag 706 /* Record the new block's tag in the current descriptor
672 buffer */ 707 buffer */
673 708
674 tag_flag = 0; 709 tag_flag = 0;
675 if (escape) !! 710 if (flags & 1)
676 tag_flag |= JBD2_FLAG_ 711 tag_flag |= JBD2_FLAG_ESCAPE;
677 if (!first_tag) 712 if (!first_tag)
678 tag_flag |= JBD2_FLAG_ 713 tag_flag |= JBD2_FLAG_SAME_UUID;
679 714
680 tag = (journal_block_tag_t *) 715 tag = (journal_block_tag_t *) tagp;
681 write_tag_block(journal, tag, 716 write_tag_block(journal, tag, jh2bh(jh)->b_blocknr);
682 tag->t_flags = cpu_to_be16(tag 717 tag->t_flags = cpu_to_be16(tag_flag);
683 jbd2_block_tag_csum_set(journa 718 jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
684 commit 719 commit_transaction->t_tid);
685 tagp += tag_bytes; 720 tagp += tag_bytes;
686 space_left -= tag_bytes; 721 space_left -= tag_bytes;
687 bufs++; 722 bufs++;
688 723
689 if (first_tag) { 724 if (first_tag) {
690 memcpy (tagp, journal- 725 memcpy (tagp, journal->j_uuid, 16);
691 tagp += 16; 726 tagp += 16;
692 space_left -= 16; 727 space_left -= 16;
693 first_tag = 0; 728 first_tag = 0;
694 } 729 }
695 730
696 /* If there's no more to do, o 731 /* If there's no more to do, or if the descriptor is full,
697 let the IO rip! */ 732 let the IO rip! */
698 733
699 if (bufs == journal->j_wbufsiz 734 if (bufs == journal->j_wbufsize ||
700 commit_transaction->t_buff 735 commit_transaction->t_buffers == NULL ||
701 space_left < tag_bytes + 1 736 space_left < tag_bytes + 16 + csum_size) {
702 737
703 jbd2_debug(4, "JBD2: S 738 jbd2_debug(4, "JBD2: Submit %d IOs\n", bufs);
704 739
705 /* Write an end-of-des 740 /* Write an end-of-descriptor marker before
706 submitting the IOs. 741 submitting the IOs. "tag" still points to
707 the last tag we set 742 the last tag we set up. */
708 743
709 tag->t_flags |= cpu_to 744 tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
710 start_journal_io: 745 start_journal_io:
711 if (descriptor) 746 if (descriptor)
712 jbd2_descripto 747 jbd2_descriptor_block_csum_set(journal,
713 748 descriptor);
714 749
715 for (i = 0; i < bufs; 750 for (i = 0; i < bufs; i++) {
716 struct buffer_ 751 struct buffer_head *bh = wbuf[i];
717 752
718 /* 753 /*
719 * Compute che 754 * Compute checksum.
720 */ 755 */
721 if (jbd2_has_f 756 if (jbd2_has_feature_checksum(journal)) {
722 crc32_ 757 crc32_sum =
723 jb 758 jbd2_checksum_data(crc32_sum, bh);
724 } 759 }
725 760
726 lock_buffer(bh 761 lock_buffer(bh);
727 clear_buffer_d 762 clear_buffer_dirty(bh);
728 set_buffer_upt 763 set_buffer_uptodate(bh);
729 bh->b_end_io = 764 bh->b_end_io = journal_end_buffer_io_sync;
730 submit_bh(REQ_ 765 submit_bh(REQ_OP_WRITE | JBD2_JOURNAL_REQ_FLAGS,
731 bh); 766 bh);
732 } 767 }
733 cond_resched(); 768 cond_resched();
734 769
735 /* Force a new descrip 770 /* Force a new descriptor to be generated next
736 time round the loop 771 time round the loop. */
737 descriptor = NULL; 772 descriptor = NULL;
738 bufs = 0; 773 bufs = 0;
739 } 774 }
740 } 775 }
741 776
742 err = journal_finish_inode_data_buffer 777 err = journal_finish_inode_data_buffers(journal, commit_transaction);
743 if (err) { 778 if (err) {
744 printk(KERN_WARNING 779 printk(KERN_WARNING
745 "JBD2: Detected IO err 780 "JBD2: Detected IO errors while flushing file data "
746 "on %s\n", journal->j_d 781 "on %s\n", journal->j_devname);
747 if (journal->j_flags & JBD2_AB 782 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
748 jbd2_journal_abort(jou 783 jbd2_journal_abort(journal, err);
749 err = 0; 784 err = 0;
750 } 785 }
751 786
752 /* 787 /*
753 * Get current oldest transaction in t 788 * Get current oldest transaction in the log before we issue flush
754 * to the filesystem device. After the 789 * to the filesystem device. After the flush we can be sure that
755 * blocks of all older transactions ar 790 * blocks of all older transactions are checkpointed to persistent
756 * storage and we will be safe to upda 791 * storage and we will be safe to update journal start in the
757 * superblock with the numbers we get 792 * superblock with the numbers we get here.
758 */ 793 */
759 update_tail = 794 update_tail =
760 jbd2_journal_get_log_tail(jour 795 jbd2_journal_get_log_tail(journal, &first_tid, &first_block);
761 796
762 write_lock(&journal->j_state_lock); 797 write_lock(&journal->j_state_lock);
763 if (update_tail) { 798 if (update_tail) {
764 long freed = first_block - jou 799 long freed = first_block - journal->j_tail;
765 800
766 if (first_block < journal->j_t 801 if (first_block < journal->j_tail)
767 freed += journal->j_la 802 freed += journal->j_last - journal->j_first;
768 /* Update tail only if we free 803 /* Update tail only if we free significant amount of space */
769 if (freed < journal->j_max_tra 804 if (freed < journal->j_max_transaction_buffers)
770 update_tail = 0; 805 update_tail = 0;
771 } 806 }
772 J_ASSERT(commit_transaction->t_state = 807 J_ASSERT(commit_transaction->t_state == T_COMMIT);
773 commit_transaction->t_state = T_COMMIT 808 commit_transaction->t_state = T_COMMIT_DFLUSH;
774 write_unlock(&journal->j_state_lock); 809 write_unlock(&journal->j_state_lock);
775 810
776 /* 811 /*
777 * If the journal is not located on th 812 * If the journal is not located on the file system device,
778 * then we must flush the file system 813 * then we must flush the file system device before we issue
779 * the commit record 814 * the commit record
780 */ 815 */
781 if (commit_transaction->t_need_data_fl 816 if (commit_transaction->t_need_data_flush &&
782 (journal->j_fs_dev != journal->j_d 817 (journal->j_fs_dev != journal->j_dev) &&
783 (journal->j_flags & JBD2_BARRIER)) 818 (journal->j_flags & JBD2_BARRIER))
784 blkdev_issue_flush(journal->j_ 819 blkdev_issue_flush(journal->j_fs_dev);
785 820
786 /* Done it all: now write the commit r 821 /* Done it all: now write the commit record asynchronously. */
787 if (jbd2_has_feature_async_commit(jour 822 if (jbd2_has_feature_async_commit(journal)) {
788 err = journal_submit_commit_re 823 err = journal_submit_commit_record(journal, commit_transaction,
789 824 &cbh, crc32_sum);
790 if (err) 825 if (err)
791 jbd2_journal_abort(jou 826 jbd2_journal_abort(journal, err);
792 } 827 }
793 828
794 blk_finish_plug(&plug); 829 blk_finish_plug(&plug);
795 830
796 /* Lo and behold: we have just managed 831 /* Lo and behold: we have just managed to send a transaction to
797 the log. Before we can commit it, 832 the log. Before we can commit it, wait for the IO so far to
798 complete. Control buffers being wr 833 complete. Control buffers being written are on the
799 transaction's t_log_list queue, and 834 transaction's t_log_list queue, and metadata buffers are on
800 the io_bufs list. 835 the io_bufs list.
801 836
802 Wait for the buffers in reverse ord 837 Wait for the buffers in reverse order. That way we are
803 less likely to be woken up until al 838 less likely to be woken up until all IOs have completed, and
804 so we incur less scheduling load. 839 so we incur less scheduling load.
805 */ 840 */
806 841
807 jbd2_debug(3, "JBD2: commit phase 3\n" 842 jbd2_debug(3, "JBD2: commit phase 3\n");
808 843
809 while (!list_empty(&io_bufs)) { 844 while (!list_empty(&io_bufs)) {
810 struct buffer_head *bh = list_ 845 struct buffer_head *bh = list_entry(io_bufs.prev,
811 846 struct buffer_head,
812 847 b_assoc_buffers);
813 848
814 wait_on_buffer(bh); 849 wait_on_buffer(bh);
815 cond_resched(); 850 cond_resched();
816 851
817 if (unlikely(!buffer_uptodate( 852 if (unlikely(!buffer_uptodate(bh)))
818 err = -EIO; 853 err = -EIO;
819 jbd2_unfile_log_bh(bh); 854 jbd2_unfile_log_bh(bh);
820 stats.run.rs_blocks_logged++; 855 stats.run.rs_blocks_logged++;
821 856
822 /* 857 /*
823 * The list contains temporary 858 * The list contains temporary buffer heads created by
824 * jbd2_journal_write_metadata 859 * jbd2_journal_write_metadata_buffer().
825 */ 860 */
826 BUFFER_TRACE(bh, "dumping temp 861 BUFFER_TRACE(bh, "dumping temporary bh");
827 __brelse(bh); 862 __brelse(bh);
828 J_ASSERT_BH(bh, atomic_read(&b 863 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
829 free_buffer_head(bh); 864 free_buffer_head(bh);
830 865
831 /* We also have to refile the 866 /* We also have to refile the corresponding shadowed buffer */
832 jh = commit_transaction->t_sha 867 jh = commit_transaction->t_shadow_list->b_tprev;
833 bh = jh2bh(jh); 868 bh = jh2bh(jh);
834 clear_buffer_jwrite(bh); 869 clear_buffer_jwrite(bh);
835 J_ASSERT_BH(bh, buffer_jbddirt 870 J_ASSERT_BH(bh, buffer_jbddirty(bh));
836 J_ASSERT_BH(bh, !buffer_shadow 871 J_ASSERT_BH(bh, !buffer_shadow(bh));
837 872
838 /* The metadata is now release 873 /* The metadata is now released for reuse, but we need
839 to remember it against this 874 to remember it against this transaction so that when
840 we finally commit, we can d 875 we finally commit, we can do any checkpointing
841 required. */ 876 required. */
842 JBUFFER_TRACE(jh, "file as BJ_ 877 JBUFFER_TRACE(jh, "file as BJ_Forget");
843 jbd2_journal_file_buffer(jh, c 878 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
844 JBUFFER_TRACE(jh, "brelse shad 879 JBUFFER_TRACE(jh, "brelse shadowed buffer");
845 __brelse(bh); 880 __brelse(bh);
846 } 881 }
847 882
848 J_ASSERT (commit_transaction->t_shadow 883 J_ASSERT (commit_transaction->t_shadow_list == NULL);
849 884
850 jbd2_debug(3, "JBD2: commit phase 4\n" 885 jbd2_debug(3, "JBD2: commit phase 4\n");
851 886
852 /* Here we wait for the revoke record 887 /* Here we wait for the revoke record and descriptor record buffers */
853 while (!list_empty(&log_bufs)) { 888 while (!list_empty(&log_bufs)) {
854 struct buffer_head *bh; 889 struct buffer_head *bh;
855 890
856 bh = list_entry(log_bufs.prev, 891 bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
857 wait_on_buffer(bh); 892 wait_on_buffer(bh);
858 cond_resched(); 893 cond_resched();
859 894
860 if (unlikely(!buffer_uptodate( 895 if (unlikely(!buffer_uptodate(bh)))
861 err = -EIO; 896 err = -EIO;
862 897
863 BUFFER_TRACE(bh, "ph5: control 898 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
864 clear_buffer_jwrite(bh); 899 clear_buffer_jwrite(bh);
865 jbd2_unfile_log_bh(bh); 900 jbd2_unfile_log_bh(bh);
866 stats.run.rs_blocks_logged++; 901 stats.run.rs_blocks_logged++;
867 __brelse(bh); /* One 902 __brelse(bh); /* One for getblk */
868 /* AKPM: bforget here */ 903 /* AKPM: bforget here */
869 } 904 }
870 905
871 if (err) 906 if (err)
872 jbd2_journal_abort(journal, er 907 jbd2_journal_abort(journal, err);
873 908
874 jbd2_debug(3, "JBD2: commit phase 5\n" 909 jbd2_debug(3, "JBD2: commit phase 5\n");
875 write_lock(&journal->j_state_lock); 910 write_lock(&journal->j_state_lock);
876 J_ASSERT(commit_transaction->t_state = 911 J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
877 commit_transaction->t_state = T_COMMIT 912 commit_transaction->t_state = T_COMMIT_JFLUSH;
878 write_unlock(&journal->j_state_lock); 913 write_unlock(&journal->j_state_lock);
879 914
880 if (!jbd2_has_feature_async_commit(jou 915 if (!jbd2_has_feature_async_commit(journal)) {
881 err = journal_submit_commit_re 916 err = journal_submit_commit_record(journal, commit_transaction,
882 917 &cbh, crc32_sum);
883 if (err) 918 if (err)
884 jbd2_journal_abort(jou 919 jbd2_journal_abort(journal, err);
885 } 920 }
886 if (cbh) 921 if (cbh)
887 err = journal_wait_on_commit_r 922 err = journal_wait_on_commit_record(journal, cbh);
888 stats.run.rs_blocks_logged++; 923 stats.run.rs_blocks_logged++;
889 if (jbd2_has_feature_async_commit(jour 924 if (jbd2_has_feature_async_commit(journal) &&
890 journal->j_flags & JBD2_BARRIER) { 925 journal->j_flags & JBD2_BARRIER) {
891 blkdev_issue_flush(journal->j_ 926 blkdev_issue_flush(journal->j_dev);
892 } 927 }
893 928
894 if (err) 929 if (err)
895 jbd2_journal_abort(journal, er 930 jbd2_journal_abort(journal, err);
896 931
897 WARN_ON_ONCE( 932 WARN_ON_ONCE(
898 atomic_read(&commit_transactio 933 atomic_read(&commit_transaction->t_outstanding_credits) < 0);
899 934
900 /* 935 /*
901 * Now disk caches for filesystem devi 936 * Now disk caches for filesystem device are flushed so we are safe to
902 * erase checkpointed transactions fro 937 * erase checkpointed transactions from the log by updating journal
903 * superblock. 938 * superblock.
904 */ 939 */
905 if (update_tail) 940 if (update_tail)
906 jbd2_update_log_tail(journal, 941 jbd2_update_log_tail(journal, first_tid, first_block);
907 942
908 /* End of a transaction! Finally, we 943 /* End of a transaction! Finally, we can do checkpoint
909 processing: any buffers committed a 944 processing: any buffers committed as a result of this
910 transaction can be removed from any 945 transaction can be removed from any checkpoint list it was on
911 before. */ 946 before. */
912 947
913 jbd2_debug(3, "JBD2: commit phase 6\n" 948 jbd2_debug(3, "JBD2: commit phase 6\n");
914 949
915 J_ASSERT(list_empty(&commit_transactio 950 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
916 J_ASSERT(commit_transaction->t_buffers 951 J_ASSERT(commit_transaction->t_buffers == NULL);
917 J_ASSERT(commit_transaction->t_checkpo 952 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
918 J_ASSERT(commit_transaction->t_shadow_ 953 J_ASSERT(commit_transaction->t_shadow_list == NULL);
919 954
920 restart_loop: 955 restart_loop:
921 /* 956 /*
922 * As there are other places (journal_ 957 * As there are other places (journal_unmap_buffer()) adding buffers
923 * to this list we have to be careful 958 * to this list we have to be careful and hold the j_list_lock.
924 */ 959 */
925 spin_lock(&journal->j_list_lock); 960 spin_lock(&journal->j_list_lock);
926 while (commit_transaction->t_forget) { 961 while (commit_transaction->t_forget) {
927 transaction_t *cp_transaction; 962 transaction_t *cp_transaction;
928 struct buffer_head *bh; 963 struct buffer_head *bh;
929 int try_to_free = 0; 964 int try_to_free = 0;
930 bool drop_ref; 965 bool drop_ref;
931 966
932 jh = commit_transaction->t_for 967 jh = commit_transaction->t_forget;
933 spin_unlock(&journal->j_list_l 968 spin_unlock(&journal->j_list_lock);
934 bh = jh2bh(jh); 969 bh = jh2bh(jh);
935 /* 970 /*
936 * Get a reference so that bh 971 * Get a reference so that bh cannot be freed before we are
937 * done with it. 972 * done with it.
938 */ 973 */
939 get_bh(bh); 974 get_bh(bh);
940 spin_lock(&jh->b_state_lock); 975 spin_lock(&jh->b_state_lock);
941 J_ASSERT_JH(jh, jh->b_transact 976 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
942 977
943 /* 978 /*
944 * If there is undo-protected 979 * If there is undo-protected committed data against
945 * this buffer, then we can re 980 * this buffer, then we can remove it now. If it is a
946 * buffer needing such protect 981 * buffer needing such protection, the old frozen_data
947 * field now points to a commi 982 * field now points to a committed version of the
948 * buffer, so rotate that fiel 983 * buffer, so rotate that field to the new committed
949 * data. 984 * data.
950 * 985 *
951 * Otherwise, we can just thro 986 * Otherwise, we can just throw away the frozen data now.
952 * 987 *
953 * We also know that the froze 988 * We also know that the frozen data has already fired
954 * its triggers if they exist, 989 * its triggers if they exist, so we can clear that too.
955 */ 990 */
956 if (jh->b_committed_data) { 991 if (jh->b_committed_data) {
957 jbd2_free(jh->b_commit 992 jbd2_free(jh->b_committed_data, bh->b_size);
958 jh->b_committed_data = 993 jh->b_committed_data = NULL;
959 if (jh->b_frozen_data) 994 if (jh->b_frozen_data) {
960 jh->b_committe 995 jh->b_committed_data = jh->b_frozen_data;
961 jh->b_frozen_d 996 jh->b_frozen_data = NULL;
962 jh->b_frozen_t 997 jh->b_frozen_triggers = NULL;
963 } 998 }
964 } else if (jh->b_frozen_data) 999 } else if (jh->b_frozen_data) {
965 jbd2_free(jh->b_frozen 1000 jbd2_free(jh->b_frozen_data, bh->b_size);
966 jh->b_frozen_data = NU 1001 jh->b_frozen_data = NULL;
967 jh->b_frozen_triggers 1002 jh->b_frozen_triggers = NULL;
968 } 1003 }
969 1004
970 spin_lock(&journal->j_list_loc 1005 spin_lock(&journal->j_list_lock);
971 cp_transaction = jh->b_cp_tran 1006 cp_transaction = jh->b_cp_transaction;
972 if (cp_transaction) { 1007 if (cp_transaction) {
973 JBUFFER_TRACE(jh, "rem 1008 JBUFFER_TRACE(jh, "remove from old cp transaction");
974 cp_transaction->t_chp_ 1009 cp_transaction->t_chp_stats.cs_dropped++;
975 __jbd2_journal_remove_ 1010 __jbd2_journal_remove_checkpoint(jh);
976 } 1011 }
977 1012
978 /* Only re-checkpoint the buff 1013 /* Only re-checkpoint the buffer_head if it is marked
979 * dirty. If the buffer was a 1014 * dirty. If the buffer was added to the BJ_Forget list
980 * by jbd2_journal_forget, it 1015 * by jbd2_journal_forget, it may no longer be dirty and
981 * there's no point in keeping 1016 * there's no point in keeping a checkpoint record for
982 * it. */ 1017 * it. */
983 1018
984 /* 1019 /*
985 * A buffer which has been fre 1020 * A buffer which has been freed while still being journaled
986 * by a previous transaction, 1021 * by a previous transaction, refile the buffer to BJ_Forget of
987 * the running transaction. If 1022 * the running transaction. If the just committed transaction
988 * contains "add to orphan" op 1023 * contains "add to orphan" operation, we can completely
989 * invalidate the buffer now. 1024 * invalidate the buffer now. We are rather through in that
990 * since the buffer may be sti 1025 * since the buffer may be still accessible when blocksize <
991 * pagesize and it is attached 1026 * pagesize and it is attached to the last partial page.
992 */ 1027 */
993 if (buffer_freed(bh) && !jh->b 1028 if (buffer_freed(bh) && !jh->b_next_transaction) {
994 struct address_space * 1029 struct address_space *mapping;
995 1030
996 clear_buffer_freed(bh) 1031 clear_buffer_freed(bh);
997 clear_buffer_jbddirty( 1032 clear_buffer_jbddirty(bh);
998 1033
999 /* 1034 /*
1000 * Block device buffe 1035 * Block device buffers need to stay mapped all the
1001 * time, so it is eno 1036 * time, so it is enough to clear buffer_jbddirty and
1002 * buffer_freed bits. 1037 * buffer_freed bits. For the file mapping buffers (i.e.
1003 * journalled data) w 1038 * journalled data) we need to unmap buffer and clear
1004 * more bits. We also 1039 * more bits. We also need to be careful about the check
1005 * because the data p 1040 * because the data page mapping can get cleared under
1006 * our hands. Note th 1041 * our hands. Note that if mapping == NULL, we don't
1007 * need to make buffe 1042 * need to make buffer unmapped because the page is
1008 * already detached f 1043 * already detached from the mapping and buffers cannot
1009 * get reused. 1044 * get reused.
1010 */ 1045 */
1011 mapping = READ_ONCE(b !! 1046 mapping = READ_ONCE(bh->b_page->mapping);
1012 if (mapping && !sb_is 1047 if (mapping && !sb_is_blkdev_sb(mapping->host->i_sb)) {
1013 clear_buffer_ 1048 clear_buffer_mapped(bh);
1014 clear_buffer_ 1049 clear_buffer_new(bh);
1015 clear_buffer_ 1050 clear_buffer_req(bh);
1016 bh->b_bdev = 1051 bh->b_bdev = NULL;
1017 } 1052 }
1018 } 1053 }
1019 1054
1020 if (buffer_jbddirty(bh)) { 1055 if (buffer_jbddirty(bh)) {
1021 JBUFFER_TRACE(jh, "ad 1056 JBUFFER_TRACE(jh, "add to new checkpointing trans");
1022 __jbd2_journal_insert 1057 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
1023 if (is_journal_aborte 1058 if (is_journal_aborted(journal))
1024 clear_buffer_ 1059 clear_buffer_jbddirty(bh);
1025 } else { 1060 } else {
1026 J_ASSERT_BH(bh, !buff 1061 J_ASSERT_BH(bh, !buffer_dirty(bh));
1027 /* 1062 /*
1028 * The buffer on BJ_F 1063 * The buffer on BJ_Forget list and not jbddirty means
1029 * it has been freed 1064 * it has been freed by this transaction and hence it
1030 * could not have bee 1065 * could not have been reallocated until this
1031 * transaction has co 1066 * transaction has committed. *BUT* it could be
1032 * reallocated once w 1067 * reallocated once we have written all the data to
1033 * disk and before we 1068 * disk and before we process the buffer on BJ_Forget
1034 * list. 1069 * list.
1035 */ 1070 */
1036 if (!jh->b_next_trans 1071 if (!jh->b_next_transaction)
1037 try_to_free = 1072 try_to_free = 1;
1038 } 1073 }
1039 JBUFFER_TRACE(jh, "refile or 1074 JBUFFER_TRACE(jh, "refile or unfile buffer");
1040 drop_ref = __jbd2_journal_ref 1075 drop_ref = __jbd2_journal_refile_buffer(jh);
1041 spin_unlock(&jh->b_state_lock 1076 spin_unlock(&jh->b_state_lock);
1042 if (drop_ref) 1077 if (drop_ref)
1043 jbd2_journal_put_jour 1078 jbd2_journal_put_journal_head(jh);
1044 if (try_to_free) 1079 if (try_to_free)
1045 release_buffer_page(b 1080 release_buffer_page(bh); /* Drops bh reference */
1046 else 1081 else
1047 __brelse(bh); 1082 __brelse(bh);
1048 cond_resched_lock(&journal->j 1083 cond_resched_lock(&journal->j_list_lock);
1049 } 1084 }
1050 spin_unlock(&journal->j_list_lock); 1085 spin_unlock(&journal->j_list_lock);
1051 /* 1086 /*
1052 * This is a bit sleazy. We use j_li 1087 * This is a bit sleazy. We use j_list_lock to protect transition
1053 * of a transaction into T_FINISHED s 1088 * of a transaction into T_FINISHED state and calling
1054 * __jbd2_journal_drop_transaction(). 1089 * __jbd2_journal_drop_transaction(). Otherwise we could race with
1055 * other checkpointing code processin 1090 * other checkpointing code processing the transaction...
1056 */ 1091 */
1057 write_lock(&journal->j_state_lock); 1092 write_lock(&journal->j_state_lock);
1058 spin_lock(&journal->j_list_lock); 1093 spin_lock(&journal->j_list_lock);
1059 /* 1094 /*
1060 * Now recheck if some buffers did no 1095 * Now recheck if some buffers did not get attached to the transaction
1061 * while the lock was dropped... 1096 * while the lock was dropped...
1062 */ 1097 */
1063 if (commit_transaction->t_forget) { 1098 if (commit_transaction->t_forget) {
1064 spin_unlock(&journal->j_list_ 1099 spin_unlock(&journal->j_list_lock);
1065 write_unlock(&journal->j_stat 1100 write_unlock(&journal->j_state_lock);
1066 goto restart_loop; 1101 goto restart_loop;
1067 } 1102 }
1068 1103
1069 /* Add the transaction to the checkpo 1104 /* Add the transaction to the checkpoint list
1070 * __journal_remove_checkpoint() can 1105 * __journal_remove_checkpoint() can not destroy transaction
1071 * under us because it is not marked 1106 * under us because it is not marked as T_FINISHED yet */
1072 if (journal->j_checkpoint_transaction 1107 if (journal->j_checkpoint_transactions == NULL) {
1073 journal->j_checkpoint_transac 1108 journal->j_checkpoint_transactions = commit_transaction;
1074 commit_transaction->t_cpnext 1109 commit_transaction->t_cpnext = commit_transaction;
1075 commit_transaction->t_cpprev 1110 commit_transaction->t_cpprev = commit_transaction;
1076 } else { 1111 } else {
1077 commit_transaction->t_cpnext 1112 commit_transaction->t_cpnext =
1078 journal->j_checkpoint 1113 journal->j_checkpoint_transactions;
1079 commit_transaction->t_cpprev 1114 commit_transaction->t_cpprev =
1080 commit_transaction->t 1115 commit_transaction->t_cpnext->t_cpprev;
1081 commit_transaction->t_cpnext- 1116 commit_transaction->t_cpnext->t_cpprev =
1082 commit_transaction; 1117 commit_transaction;
1083 commit_transaction->t_cpprev- 1118 commit_transaction->t_cpprev->t_cpnext =
1084 commit_transa 1119 commit_transaction;
1085 } 1120 }
1086 spin_unlock(&journal->j_list_lock); 1121 spin_unlock(&journal->j_list_lock);
1087 1122
1088 /* Done with this transaction! */ 1123 /* Done with this transaction! */
1089 1124
1090 jbd2_debug(3, "JBD2: commit phase 7\n 1125 jbd2_debug(3, "JBD2: commit phase 7\n");
1091 1126
1092 J_ASSERT(commit_transaction->t_state 1127 J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
1093 1128
1094 commit_transaction->t_start = jiffies 1129 commit_transaction->t_start = jiffies;
1095 stats.run.rs_logging = jbd2_time_diff 1130 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
1096 1131 commit_transaction->t_start);
1097 1132
1098 /* 1133 /*
1099 * File the transaction statistics 1134 * File the transaction statistics
1100 */ 1135 */
1101 stats.ts_tid = commit_transaction->t_ 1136 stats.ts_tid = commit_transaction->t_tid;
1102 stats.run.rs_handle_count = 1137 stats.run.rs_handle_count =
1103 atomic_read(&commit_transacti 1138 atomic_read(&commit_transaction->t_handle_count);
1104 trace_jbd2_run_stats(journal->j_fs_de 1139 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
1105 commit_transacti 1140 commit_transaction->t_tid, &stats.run);
1106 stats.ts_requested = (commit_transact 1141 stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0;
1107 1142
1108 commit_transaction->t_state = T_COMMI 1143 commit_transaction->t_state = T_COMMIT_CALLBACK;
1109 J_ASSERT(commit_transaction == journa 1144 J_ASSERT(commit_transaction == journal->j_committing_transaction);
1110 WRITE_ONCE(journal->j_commit_sequence !! 1145 journal->j_commit_sequence = commit_transaction->t_tid;
1111 journal->j_committing_transaction = N 1146 journal->j_committing_transaction = NULL;
1112 commit_time = ktime_to_ns(ktime_sub(k 1147 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1113 1148
1114 /* 1149 /*
1115 * weight the commit time higher than 1150 * weight the commit time higher than the average time so we don't
1116 * react too strongly to vast changes 1151 * react too strongly to vast changes in the commit time
1117 */ 1152 */
1118 if (likely(journal->j_average_commit_ 1153 if (likely(journal->j_average_commit_time))
1119 journal->j_average_commit_tim 1154 journal->j_average_commit_time = (commit_time +
1120 journal->j_av 1155 journal->j_average_commit_time*3) / 4;
1121 else 1156 else
1122 journal->j_average_commit_tim 1157 journal->j_average_commit_time = commit_time;
1123 1158
1124 write_unlock(&journal->j_state_lock); 1159 write_unlock(&journal->j_state_lock);
1125 1160
1126 if (journal->j_commit_callback) 1161 if (journal->j_commit_callback)
1127 journal->j_commit_callback(jo 1162 journal->j_commit_callback(journal, commit_transaction);
1128 if (journal->j_fc_cleanup_callback) 1163 if (journal->j_fc_cleanup_callback)
1129 journal->j_fc_cleanup_callbac 1164 journal->j_fc_cleanup_callback(journal, 1, commit_transaction->t_tid);
1130 1165
1131 trace_jbd2_end_commit(journal, commit 1166 trace_jbd2_end_commit(journal, commit_transaction);
1132 jbd2_debug(1, "JBD2: commit %d comple 1167 jbd2_debug(1, "JBD2: commit %d complete, head %d\n",
1133 journal->j_commit_sequence, 1168 journal->j_commit_sequence, journal->j_tail_sequence);
1134 1169
1135 write_lock(&journal->j_state_lock); 1170 write_lock(&journal->j_state_lock);
1136 journal->j_flags &= ~JBD2_FULL_COMMIT 1171 journal->j_flags &= ~JBD2_FULL_COMMIT_ONGOING;
1137 journal->j_flags &= ~JBD2_FAST_COMMIT 1172 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
1138 spin_lock(&journal->j_list_lock); 1173 spin_lock(&journal->j_list_lock);
1139 commit_transaction->t_state = T_FINIS 1174 commit_transaction->t_state = T_FINISHED;
1140 /* Check if the transaction can be dr 1175 /* Check if the transaction can be dropped now that we are finished */
1141 if (commit_transaction->t_checkpoint_ 1176 if (commit_transaction->t_checkpoint_list == NULL) {
1142 __jbd2_journal_drop_transacti 1177 __jbd2_journal_drop_transaction(journal, commit_transaction);
1143 jbd2_journal_free_transaction 1178 jbd2_journal_free_transaction(commit_transaction);
1144 } 1179 }
1145 spin_unlock(&journal->j_list_lock); 1180 spin_unlock(&journal->j_list_lock);
1146 write_unlock(&journal->j_state_lock); 1181 write_unlock(&journal->j_state_lock);
1147 wake_up(&journal->j_wait_done_commit) 1182 wake_up(&journal->j_wait_done_commit);
1148 wake_up(&journal->j_fc_wait); 1183 wake_up(&journal->j_fc_wait);
1149 1184
1150 /* 1185 /*
1151 * Calculate overall stats 1186 * Calculate overall stats
1152 */ 1187 */
1153 spin_lock(&journal->j_history_lock); 1188 spin_lock(&journal->j_history_lock);
1154 journal->j_stats.ts_tid++; 1189 journal->j_stats.ts_tid++;
1155 journal->j_stats.ts_requested += stat 1190 journal->j_stats.ts_requested += stats.ts_requested;
1156 journal->j_stats.run.rs_wait += stats 1191 journal->j_stats.run.rs_wait += stats.run.rs_wait;
1157 journal->j_stats.run.rs_request_delay 1192 journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
1158 journal->j_stats.run.rs_running += st 1193 journal->j_stats.run.rs_running += stats.run.rs_running;
1159 journal->j_stats.run.rs_locked += sta 1194 journal->j_stats.run.rs_locked += stats.run.rs_locked;
1160 journal->j_stats.run.rs_flushing += s 1195 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
1161 journal->j_stats.run.rs_logging += st 1196 journal->j_stats.run.rs_logging += stats.run.rs_logging;
1162 journal->j_stats.run.rs_handle_count 1197 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
1163 journal->j_stats.run.rs_blocks += sta 1198 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
1164 journal->j_stats.run.rs_blocks_logged 1199 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
1165 spin_unlock(&journal->j_history_lock) 1200 spin_unlock(&journal->j_history_lock);
1166 } 1201 }
1167 1202
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.