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 page *page;
66 66
67 if (buffer_dirty(bh)) 67 if (buffer_dirty(bh))
68 goto nope; 68 goto nope;
69 if (atomic_read(&bh->b_count) != 1) 69 if (atomic_read(&bh->b_count) != 1)
70 goto nope; 70 goto nope;
71 folio = bh->b_folio; !! 71 page = bh->b_page;
72 if (folio->mapping) !! 72 if (!page)
>> 73 goto nope;
>> 74 if (page->mapping)
73 goto nope; 75 goto nope;
74 76
75 /* OK, it's a truncated page */ 77 /* OK, it's a truncated page */
76 if (!folio_trylock(folio)) !! 78 if (!trylock_page(page))
77 goto nope; 79 goto nope;
78 80
79 folio_get(folio); !! 81 get_page(page);
80 __brelse(bh); 82 __brelse(bh);
81 try_to_free_buffers(folio); !! 83 try_to_free_buffers(page);
82 folio_unlock(folio); !! 84 unlock_page(page);
83 folio_put(folio); !! 85 put_page(page);
84 return; 86 return;
85 87
86 nope: 88 nope:
87 __brelse(bh); 89 __brelse(bh);
88 } 90 }
89 91
90 static void jbd2_commit_block_csum_set(journal 92 static void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
91 { 93 {
92 struct commit_header *h; 94 struct commit_header *h;
93 __u32 csum; 95 __u32 csum;
94 96
95 if (!jbd2_journal_has_csum_v2or3(j)) 97 if (!jbd2_journal_has_csum_v2or3(j))
96 return; 98 return;
97 99
98 h = (struct commit_header *)(bh->b_dat 100 h = (struct commit_header *)(bh->b_data);
99 h->h_chksum_type = 0; 101 h->h_chksum_type = 0;
100 h->h_chksum_size = 0; 102 h->h_chksum_size = 0;
101 h->h_chksum[0] = 0; 103 h->h_chksum[0] = 0;
102 csum = jbd2_chksum(j, j->j_csum_seed, 104 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
103 h->h_chksum[0] = cpu_to_be32(csum); 105 h->h_chksum[0] = cpu_to_be32(csum);
104 } 106 }
105 107
106 /* 108 /*
107 * Done it all: now submit the commit record. 109 * Done it all: now submit the commit record. We should have
108 * cleaned up our previous buffers by now, so 110 * 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 111 * mode we can now just skip the rest of the journal write
110 * entirely. 112 * entirely.
111 * 113 *
112 * Returns 1 if the journal needs to be aborte 114 * Returns 1 if the journal needs to be aborted or 0 on success
113 */ 115 */
114 static int journal_submit_commit_record(journa 116 static int journal_submit_commit_record(journal_t *journal,
115 transa 117 transaction_t *commit_transaction,
116 struct 118 struct buffer_head **cbh,
117 __u32 119 __u32 crc32_sum)
118 { 120 {
119 struct commit_header *tmp; 121 struct commit_header *tmp;
120 struct buffer_head *bh; 122 struct buffer_head *bh;
>> 123 int ret;
121 struct timespec64 now; 124 struct timespec64 now;
122 blk_opf_t write_flags = REQ_OP_WRITE | <<
123 125
124 *cbh = NULL; 126 *cbh = NULL;
125 127
126 if (is_journal_aborted(journal)) 128 if (is_journal_aborted(journal))
127 return 0; 129 return 0;
128 130
129 bh = jbd2_journal_get_descriptor_buffe 131 bh = jbd2_journal_get_descriptor_buffer(commit_transaction,
130 132 JBD2_COMMIT_BLOCK);
131 if (!bh) 133 if (!bh)
132 return 1; 134 return 1;
133 135
134 tmp = (struct commit_header *)bh->b_da 136 tmp = (struct commit_header *)bh->b_data;
135 ktime_get_coarse_real_ts64(&now); 137 ktime_get_coarse_real_ts64(&now);
136 tmp->h_commit_sec = cpu_to_be64(now.tv 138 tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
137 tmp->h_commit_nsec = cpu_to_be32(now.t 139 tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
138 140
139 if (jbd2_has_feature_checksum(journal) 141 if (jbd2_has_feature_checksum(journal)) {
140 tmp->h_chksum_type = JBD2 142 tmp->h_chksum_type = JBD2_CRC32_CHKSUM;
141 tmp->h_chksum_size = JBD2 143 tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE;
142 tmp->h_chksum[0] = cpu_ 144 tmp->h_chksum[0] = cpu_to_be32(crc32_sum);
143 } 145 }
144 jbd2_commit_block_csum_set(journal, bh 146 jbd2_commit_block_csum_set(journal, bh);
145 147
146 BUFFER_TRACE(bh, "submit commit block" 148 BUFFER_TRACE(bh, "submit commit block");
147 lock_buffer(bh); 149 lock_buffer(bh);
148 clear_buffer_dirty(bh); 150 clear_buffer_dirty(bh);
149 set_buffer_uptodate(bh); 151 set_buffer_uptodate(bh);
150 bh->b_end_io = journal_end_buffer_io_s 152 bh->b_end_io = journal_end_buffer_io_sync;
151 153
152 if (journal->j_flags & JBD2_BARRIER && 154 if (journal->j_flags & JBD2_BARRIER &&
153 !jbd2_has_feature_async_commit(jou 155 !jbd2_has_feature_async_commit(journal))
154 write_flags |= REQ_PREFLUSH | !! 156 ret = submit_bh(REQ_OP_WRITE,
>> 157 REQ_SYNC | REQ_PREFLUSH | REQ_FUA, bh);
>> 158 else
>> 159 ret = submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
155 160
156 submit_bh(write_flags, bh); <<
157 *cbh = bh; 161 *cbh = bh;
158 return 0; !! 162 return ret;
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(); <<
274 spin_lock(&journal->j_list_loc 303 spin_lock(&journal->j_list_lock);
275 jinode->i_flags &= ~JI_COMMIT_ 304 jinode->i_flags &= ~JI_COMMIT_RUNNING;
276 smp_mb(); 305 smp_mb();
277 wake_up_bit(&jinode->i_flags, 306 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
278 } 307 }
279 308
280 /* Now refile inode to proper lists */ 309 /* Now refile inode to proper lists */
281 list_for_each_entry_safe(jinode, next_ 310 list_for_each_entry_safe(jinode, next_i,
282 &commit_trans 311 &commit_transaction->t_inode_list, i_list) {
283 list_del(&jinode->i_list); 312 list_del(&jinode->i_list);
284 if (jinode->i_next_transaction 313 if (jinode->i_next_transaction) {
285 jinode->i_transaction 314 jinode->i_transaction = jinode->i_next_transaction;
286 jinode->i_next_transac 315 jinode->i_next_transaction = NULL;
287 list_add(&jinode->i_li 316 list_add(&jinode->i_list,
288 &jinode->i_tra 317 &jinode->i_transaction->t_inode_list);
289 } else { 318 } else {
290 jinode->i_transaction 319 jinode->i_transaction = NULL;
291 jinode->i_dirty_start 320 jinode->i_dirty_start = 0;
292 jinode->i_dirty_end = 321 jinode->i_dirty_end = 0;
293 } 322 }
294 } 323 }
295 spin_unlock(&journal->j_list_lock); 324 spin_unlock(&journal->j_list_lock);
296 325
297 return ret; 326 return ret;
298 } 327 }
299 328
300 static __u32 jbd2_checksum_data(__u32 crc32_su 329 static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
301 { 330 {
>> 331 struct page *page = bh->b_page;
302 char *addr; 332 char *addr;
303 __u32 checksum; 333 __u32 checksum;
304 334
305 addr = kmap_local_folio(bh->b_folio, b !! 335 addr = kmap_atomic(page);
306 checksum = crc32_be(crc32_sum, addr, b !! 336 checksum = crc32_be(crc32_sum,
307 kunmap_local(addr); !! 337 (void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
>> 338 kunmap_atomic(addr);
308 339
309 return checksum; 340 return checksum;
310 } 341 }
311 342
312 static void write_tag_block(journal_t *j, jour 343 static void write_tag_block(journal_t *j, journal_block_tag_t *tag,
313 unsigned lo 344 unsigned long long block)
314 { 345 {
315 tag->t_blocknr = cpu_to_be32(block & ( 346 tag->t_blocknr = cpu_to_be32(block & (u32)~0);
316 if (jbd2_has_feature_64bit(j)) 347 if (jbd2_has_feature_64bit(j))
317 tag->t_blocknr_high = cpu_to_b 348 tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
318 } 349 }
319 350
320 static void jbd2_block_tag_csum_set(journal_t 351 static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
321 struct buf 352 struct buffer_head *bh, __u32 sequence)
322 { 353 {
323 journal_block_tag3_t *tag3 = (journal_ 354 journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
>> 355 struct page *page = bh->b_page;
324 __u8 *addr; 356 __u8 *addr;
325 __u32 csum32; 357 __u32 csum32;
326 __be32 seq; 358 __be32 seq;
327 359
328 if (!jbd2_journal_has_csum_v2or3(j)) 360 if (!jbd2_journal_has_csum_v2or3(j))
329 return; 361 return;
330 362
331 seq = cpu_to_be32(sequence); 363 seq = cpu_to_be32(sequence);
332 addr = kmap_local_folio(bh->b_folio, b !! 364 addr = kmap_atomic(page);
333 csum32 = jbd2_chksum(j, j->j_csum_seed 365 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
334 csum32 = jbd2_chksum(j, csum32, addr, !! 366 csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data),
335 kunmap_local(addr); !! 367 bh->b_size);
>> 368 kunmap_atomic(addr);
336 369
337 if (jbd2_has_feature_csum3(j)) 370 if (jbd2_has_feature_csum3(j))
338 tag3->t_checksum = cpu_to_be32 371 tag3->t_checksum = cpu_to_be32(csum32);
339 else 372 else
340 tag->t_checksum = cpu_to_be16( 373 tag->t_checksum = cpu_to_be16(csum32);
341 } 374 }
342 /* 375 /*
343 * jbd2_journal_commit_transaction 376 * jbd2_journal_commit_transaction
344 * 377 *
345 * The primary function for committing a trans 378 * The primary function for committing a transaction to the log. This
346 * function is called by the journal thread to 379 * function is called by the journal thread to begin a complete commit.
347 */ 380 */
348 void jbd2_journal_commit_transaction(journal_t 381 void jbd2_journal_commit_transaction(journal_t *journal)
349 { 382 {
350 struct transaction_stats_s stats; 383 struct transaction_stats_s stats;
351 transaction_t *commit_transaction; 384 transaction_t *commit_transaction;
352 struct journal_head *jh; 385 struct journal_head *jh;
353 struct buffer_head *descriptor; 386 struct buffer_head *descriptor;
354 struct buffer_head **wbuf = journal->j 387 struct buffer_head **wbuf = journal->j_wbuf;
355 int bufs; 388 int bufs;
356 int escape; !! 389 int flags;
357 int err; 390 int err;
358 unsigned long long blocknr; 391 unsigned long long blocknr;
359 ktime_t start_time; 392 ktime_t start_time;
360 u64 commit_time; 393 u64 commit_time;
361 char *tagp = NULL; 394 char *tagp = NULL;
362 journal_block_tag_t *tag = NULL; 395 journal_block_tag_t *tag = NULL;
363 int space_left = 0; 396 int space_left = 0;
364 int first_tag = 0; 397 int first_tag = 0;
365 int tag_flag; 398 int tag_flag;
366 int i; 399 int i;
367 int tag_bytes = journal_tag_bytes(jour 400 int tag_bytes = journal_tag_bytes(journal);
368 struct buffer_head *cbh = NULL; /* For 401 struct buffer_head *cbh = NULL; /* For transactional checksums */
369 __u32 crc32_sum = ~0; 402 __u32 crc32_sum = ~0;
370 struct blk_plug plug; 403 struct blk_plug plug;
371 /* Tail of the journal */ 404 /* Tail of the journal */
372 unsigned long first_block; 405 unsigned long first_block;
373 tid_t first_tid; 406 tid_t first_tid;
374 int update_tail; 407 int update_tail;
375 int csum_size = 0; 408 int csum_size = 0;
376 LIST_HEAD(io_bufs); 409 LIST_HEAD(io_bufs);
377 LIST_HEAD(log_bufs); 410 LIST_HEAD(log_bufs);
378 411
379 if (jbd2_journal_has_csum_v2or3(journa 412 if (jbd2_journal_has_csum_v2or3(journal))
380 csum_size = sizeof(struct jbd2 413 csum_size = sizeof(struct jbd2_journal_block_tail);
381 414
382 /* 415 /*
383 * First job: lock down the current tr 416 * First job: lock down the current transaction and wait for
384 * all outstanding updates to complete 417 * all outstanding updates to complete.
385 */ 418 */
386 419
387 /* Do we need to erase the effects of 420 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
388 if (journal->j_flags & JBD2_FLUSHED) { 421 if (journal->j_flags & JBD2_FLUSHED) {
389 jbd2_debug(3, "super block upd !! 422 jbd_debug(3, "super block updated\n");
390 mutex_lock_io(&journal->j_chec 423 mutex_lock_io(&journal->j_checkpoint_mutex);
391 /* 424 /*
392 * We hold j_checkpoint_mutex 425 * We hold j_checkpoint_mutex so tail cannot change under us.
393 * We don't need any special d 426 * We don't need any special data guarantees for writing sb
394 * since journal is empty and 427 * since journal is empty and it is ok for write to be
395 * flushed only with transacti 428 * flushed only with transaction commit.
396 */ 429 */
397 jbd2_journal_update_sb_log_tai 430 jbd2_journal_update_sb_log_tail(journal,
398 431 journal->j_tail_sequence,
399 !! 432 journal->j_tail,
>> 433 REQ_SYNC);
400 mutex_unlock(&journal->j_check 434 mutex_unlock(&journal->j_checkpoint_mutex);
401 } else { 435 } else {
402 jbd2_debug(3, "superblock not !! 436 jbd_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 jbd_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 spin_lock(&commit_transaction->t_handle_lock);
454 jbd2_journal_wait_updates(journal); !! 488 while (atomic_read(&commit_transaction->t_updates)) {
>> 489 DEFINE_WAIT(wait);
455 490
>> 491 prepare_to_wait(&journal->j_wait_updates, &wait,
>> 492 TASK_UNINTERRUPTIBLE);
>> 493 if (atomic_read(&commit_transaction->t_updates)) {
>> 494 spin_unlock(&commit_transaction->t_handle_lock);
>> 495 write_unlock(&journal->j_state_lock);
>> 496 schedule();
>> 497 write_lock(&journal->j_state_lock);
>> 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);
456 commit_transaction->t_state = T_SWITCH 503 commit_transaction->t_state = T_SWITCH;
>> 504 write_unlock(&journal->j_state_lock);
457 505
458 J_ASSERT (atomic_read(&commit_transact 506 J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
459 journal->j_max_transac 507 journal->j_max_transaction_buffers);
460 508
461 /* 509 /*
462 * First thing we are allowed to do is 510 * First thing we are allowed to do is to discard any remaining
463 * BJ_Reserved buffers. Note, it is _ 511 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
464 * that there are no such buffers: if 512 * that there are no such buffers: if a large filesystem
465 * operation like a truncate needs to 513 * operation like a truncate needs to split itself over multiple
466 * transactions, then it may try to do 514 * transactions, then it may try to do a jbd2_journal_restart() while
467 * there are still BJ_Reserved buffers 515 * there are still BJ_Reserved buffers outstanding. These must
468 * be released cleanly from the curren 516 * be released cleanly from the current transaction.
469 * 517 *
470 * In this case, the filesystem must s 518 * In this case, the filesystem must still reserve write access
471 * again before modifying the buffer i 519 * again before modifying the buffer in the new transaction, but
472 * we do not require it to remember ex 520 * we do not require it to remember exactly which old buffers it
473 * has reserved. This is consistent w 521 * has reserved. This is consistent with the existing behaviour
474 * that multiple jbd2_journal_get_writ 522 * that multiple jbd2_journal_get_write_access() calls to the same
475 * buffer are perfectly permissible. 523 * buffer are perfectly permissible.
476 * We use journal->j_state_lock here t <<
477 * t_reserved_list with eviction of bu <<
478 */ 524 */
479 while (commit_transaction->t_reserved_ 525 while (commit_transaction->t_reserved_list) {
480 jh = commit_transaction->t_res 526 jh = commit_transaction->t_reserved_list;
481 JBUFFER_TRACE(jh, "reserved, u 527 JBUFFER_TRACE(jh, "reserved, unused: refile");
482 /* 528 /*
483 * A jbd2_journal_get_undo_acc 529 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
484 * leave undo-committed data. 530 * leave undo-committed data.
485 */ 531 */
486 if (jh->b_committed_data) { 532 if (jh->b_committed_data) {
487 struct buffer_head *bh 533 struct buffer_head *bh = jh2bh(jh);
488 534
489 spin_lock(&jh->b_state 535 spin_lock(&jh->b_state_lock);
490 jbd2_free(jh->b_commit 536 jbd2_free(jh->b_committed_data, bh->b_size);
491 jh->b_committed_data = 537 jh->b_committed_data = NULL;
492 spin_unlock(&jh->b_sta 538 spin_unlock(&jh->b_state_lock);
493 } 539 }
494 jbd2_journal_refile_buffer(jou 540 jbd2_journal_refile_buffer(journal, jh);
495 } 541 }
496 542
497 write_unlock(&journal->j_state_lock); <<
498 /* 543 /*
499 * Now try to drop any written-back bu 544 * Now try to drop any written-back buffers from the journal's
500 * checkpoint lists. We do this *befo 545 * checkpoint lists. We do this *before* commit because it potentially
501 * frees some memory 546 * frees some memory
502 */ 547 */
503 spin_lock(&journal->j_list_lock); 548 spin_lock(&journal->j_list_lock);
504 __jbd2_journal_clean_checkpoint_list(j !! 549 __jbd2_journal_clean_checkpoint_list(journal, false);
505 spin_unlock(&journal->j_list_lock); 550 spin_unlock(&journal->j_list_lock);
506 551
507 jbd2_debug(3, "JBD2: commit phase 1\n" !! 552 jbd_debug(3, "JBD2: commit phase 1\n");
508 553
509 /* 554 /*
510 * Clear revoked flag to reflect there 555 * Clear revoked flag to reflect there is no revoked buffers
511 * in the next transaction which is go 556 * in the next transaction which is going to be started.
512 */ 557 */
513 jbd2_clear_buffer_revoked_flags(journa 558 jbd2_clear_buffer_revoked_flags(journal);
514 559
515 /* 560 /*
516 * Switch to a new revoke table. 561 * Switch to a new revoke table.
517 */ 562 */
518 jbd2_journal_switch_revoke_table(journ 563 jbd2_journal_switch_revoke_table(journal);
519 564
520 write_lock(&journal->j_state_lock); <<
521 /* 565 /*
522 * Reserved credits cannot be claimed 566 * Reserved credits cannot be claimed anymore, free them
523 */ 567 */
524 atomic_sub(atomic_read(&journal->j_res 568 atomic_sub(atomic_read(&journal->j_reserved_credits),
525 &commit_transaction->t_outs 569 &commit_transaction->t_outstanding_credits);
526 570
>> 571 write_lock(&journal->j_state_lock);
527 trace_jbd2_commit_flushing(journal, co 572 trace_jbd2_commit_flushing(journal, commit_transaction);
528 stats.run.rs_flushing = jiffies; 573 stats.run.rs_flushing = jiffies;
529 stats.run.rs_locked = jbd2_time_diff(s 574 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
530 s 575 stats.run.rs_flushing);
531 576
532 commit_transaction->t_state = T_FLUSH; 577 commit_transaction->t_state = T_FLUSH;
533 journal->j_committing_transaction = co 578 journal->j_committing_transaction = commit_transaction;
534 journal->j_running_transaction = NULL; 579 journal->j_running_transaction = NULL;
535 start_time = ktime_get(); 580 start_time = ktime_get();
536 commit_transaction->t_log_start = jour 581 commit_transaction->t_log_start = journal->j_head;
537 wake_up_all(&journal->j_wait_transacti !! 582 wake_up(&journal->j_wait_transaction_locked);
538 write_unlock(&journal->j_state_lock); 583 write_unlock(&journal->j_state_lock);
539 584
540 jbd2_debug(3, "JBD2: commit phase 2a\n !! 585 jbd_debug(3, "JBD2: commit phase 2a\n");
541 586
542 /* 587 /*
543 * Now start flushing things to disk, 588 * Now start flushing things to disk, in the order they appear
544 * on the transaction lists. Data blo 589 * on the transaction lists. Data blocks go first.
545 */ 590 */
546 err = journal_submit_data_buffers(jour 591 err = journal_submit_data_buffers(journal, commit_transaction);
547 if (err) 592 if (err)
548 jbd2_journal_abort(journal, er 593 jbd2_journal_abort(journal, err);
549 594
550 blk_start_plug(&plug); 595 blk_start_plug(&plug);
551 jbd2_journal_write_revoke_records(comm 596 jbd2_journal_write_revoke_records(commit_transaction, &log_bufs);
552 597
553 jbd2_debug(3, "JBD2: commit phase 2b\n !! 598 jbd_debug(3, "JBD2: commit phase 2b\n");
554 599
555 /* 600 /*
556 * Way to go: we have now written out 601 * Way to go: we have now written out all of the data for a
557 * transaction! Now comes the tricky 602 * transaction! Now comes the tricky part: we need to write out
558 * metadata. Loop over the transactio 603 * metadata. Loop over the transaction's entire buffer list:
559 */ 604 */
560 write_lock(&journal->j_state_lock); 605 write_lock(&journal->j_state_lock);
561 commit_transaction->t_state = T_COMMIT 606 commit_transaction->t_state = T_COMMIT;
562 write_unlock(&journal->j_state_lock); 607 write_unlock(&journal->j_state_lock);
563 608
564 trace_jbd2_commit_logging(journal, com 609 trace_jbd2_commit_logging(journal, commit_transaction);
565 stats.run.rs_logging = jiffies; 610 stats.run.rs_logging = jiffies;
566 stats.run.rs_flushing = jbd2_time_diff 611 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
567 612 stats.run.rs_logging);
568 stats.run.rs_blocks = commit_transacti 613 stats.run.rs_blocks = commit_transaction->t_nr_buffers;
569 stats.run.rs_blocks_logged = 0; 614 stats.run.rs_blocks_logged = 0;
570 615
571 J_ASSERT(commit_transaction->t_nr_buff 616 J_ASSERT(commit_transaction->t_nr_buffers <=
572 atomic_read(&commit_transacti 617 atomic_read(&commit_transaction->t_outstanding_credits));
573 618
>> 619 err = 0;
574 bufs = 0; 620 bufs = 0;
575 descriptor = NULL; 621 descriptor = NULL;
576 while (commit_transaction->t_buffers) 622 while (commit_transaction->t_buffers) {
577 623
578 /* Find the next buffer to be 624 /* Find the next buffer to be journaled... */
579 625
580 jh = commit_transaction->t_buf 626 jh = commit_transaction->t_buffers;
581 627
582 /* If we're in abort mode, we 628 /* If we're in abort mode, we just un-journal the buffer and
583 release it. */ 629 release it. */
584 630
585 if (is_journal_aborted(journal 631 if (is_journal_aborted(journal)) {
586 clear_buffer_jbddirty( 632 clear_buffer_jbddirty(jh2bh(jh));
587 JBUFFER_TRACE(jh, "jou 633 JBUFFER_TRACE(jh, "journal is aborting: refile");
588 jbd2_buffer_abort_trig 634 jbd2_buffer_abort_trigger(jh,
589 635 jh->b_frozen_data ?
590 636 jh->b_frozen_triggers :
591 637 jh->b_triggers);
592 jbd2_journal_refile_bu 638 jbd2_journal_refile_buffer(journal, jh);
593 /* If that was the las 639 /* If that was the last one, we need to clean up
594 * any descriptor buff 640 * any descriptor buffers which may have been
595 * already allocated, 641 * already allocated, even if we are now
596 * aborting. */ 642 * aborting. */
597 if (!commit_transactio 643 if (!commit_transaction->t_buffers)
598 goto start_jou 644 goto start_journal_io;
599 continue; 645 continue;
600 } 646 }
601 647
602 /* Make sure we have a descrip 648 /* Make sure we have a descriptor block in which to
603 record the metadata buffer. 649 record the metadata buffer. */
604 650
605 if (!descriptor) { 651 if (!descriptor) {
606 J_ASSERT (bufs == 0); 652 J_ASSERT (bufs == 0);
607 653
608 jbd2_debug(4, "JBD2: g !! 654 jbd_debug(4, "JBD2: get descriptor\n");
609 655
610 descriptor = jbd2_jour 656 descriptor = jbd2_journal_get_descriptor_buffer(
611 657 commit_transaction,
612 658 JBD2_DESCRIPTOR_BLOCK);
613 if (!descriptor) { 659 if (!descriptor) {
614 jbd2_journal_a 660 jbd2_journal_abort(journal, -EIO);
615 continue; 661 continue;
616 } 662 }
617 663
618 jbd2_debug(4, "JBD2: g !! 664 jbd_debug(4, "JBD2: got buffer %llu (%p)\n",
619 (unsigned long 665 (unsigned long long)descriptor->b_blocknr,
620 descriptor->b_ 666 descriptor->b_data);
621 tagp = &descriptor->b_ 667 tagp = &descriptor->b_data[sizeof(journal_header_t)];
622 space_left = descripto 668 space_left = descriptor->b_size -
623 669 sizeof(journal_header_t);
624 first_tag = 1; 670 first_tag = 1;
625 set_buffer_jwrite(desc 671 set_buffer_jwrite(descriptor);
626 set_buffer_dirty(descr 672 set_buffer_dirty(descriptor);
627 wbuf[bufs++] = descrip 673 wbuf[bufs++] = descriptor;
628 674
629 /* Record it so that w 675 /* Record it so that we can wait for IO
630 completion later */ 676 completion later */
631 BUFFER_TRACE(descripto 677 BUFFER_TRACE(descriptor, "ph3: file as descriptor");
632 jbd2_file_log_bh(&log_ 678 jbd2_file_log_bh(&log_bufs, descriptor);
633 } 679 }
634 680
635 /* Where is the buffer to be w 681 /* Where is the buffer to be written? */
636 682
637 err = jbd2_journal_next_log_bl 683 err = jbd2_journal_next_log_block(journal, &blocknr);
638 /* If the block mapping failed 684 /* If the block mapping failed, just abandon the buffer
639 and repeat this loop: we'll 685 and repeat this loop: we'll fall into the
640 refile-on-abort condition a 686 refile-on-abort condition above. */
641 if (err) { 687 if (err) {
642 jbd2_journal_abort(jou 688 jbd2_journal_abort(journal, err);
643 continue; 689 continue;
644 } 690 }
645 691
646 /* 692 /*
647 * start_this_handle() uses t_ 693 * start_this_handle() uses t_outstanding_credits to determine
648 * the free space in the log. 694 * the free space in the log.
649 */ 695 */
650 atomic_dec(&commit_transaction 696 atomic_dec(&commit_transaction->t_outstanding_credits);
651 697
652 /* Bump b_count to prevent tru 698 /* Bump b_count to prevent truncate from stumbling over
653 the shadowed buffer! @@@ T 699 the shadowed buffer! @@@ This can go if we ever get
654 rid of the shadow pairing o 700 rid of the shadow pairing of buffers. */
655 atomic_inc(&jh2bh(jh)->b_count 701 atomic_inc(&jh2bh(jh)->b_count);
656 702
657 /* 703 /*
658 * Make a temporary IO buffer 704 * Make a temporary IO buffer with which to write it out
659 * (this will requeue the meta 705 * (this will requeue the metadata buffer to BJ_Shadow).
660 */ 706 */
661 set_bit(BH_JWrite, &jh2bh(jh)- 707 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
662 JBUFFER_TRACE(jh, "ph3: write 708 JBUFFER_TRACE(jh, "ph3: write metadata");
663 escape = jbd2_journal_write_me !! 709 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
664 710 jh, &wbuf[bufs], blocknr);
665 if (escape < 0) { !! 711 if (flags < 0) {
666 jbd2_journal_abort(jou !! 712 jbd2_journal_abort(journal, flags);
667 continue; 713 continue;
668 } 714 }
669 jbd2_file_log_bh(&io_bufs, wbu 715 jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
670 716
671 /* Record the new block's tag 717 /* Record the new block's tag in the current descriptor
672 buffer */ 718 buffer */
673 719
674 tag_flag = 0; 720 tag_flag = 0;
675 if (escape) !! 721 if (flags & 1)
676 tag_flag |= JBD2_FLAG_ 722 tag_flag |= JBD2_FLAG_ESCAPE;
677 if (!first_tag) 723 if (!first_tag)
678 tag_flag |= JBD2_FLAG_ 724 tag_flag |= JBD2_FLAG_SAME_UUID;
679 725
680 tag = (journal_block_tag_t *) 726 tag = (journal_block_tag_t *) tagp;
681 write_tag_block(journal, tag, 727 write_tag_block(journal, tag, jh2bh(jh)->b_blocknr);
682 tag->t_flags = cpu_to_be16(tag 728 tag->t_flags = cpu_to_be16(tag_flag);
683 jbd2_block_tag_csum_set(journa 729 jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
684 commit 730 commit_transaction->t_tid);
685 tagp += tag_bytes; 731 tagp += tag_bytes;
686 space_left -= tag_bytes; 732 space_left -= tag_bytes;
687 bufs++; 733 bufs++;
688 734
689 if (first_tag) { 735 if (first_tag) {
690 memcpy (tagp, journal- 736 memcpy (tagp, journal->j_uuid, 16);
691 tagp += 16; 737 tagp += 16;
692 space_left -= 16; 738 space_left -= 16;
693 first_tag = 0; 739 first_tag = 0;
694 } 740 }
695 741
696 /* If there's no more to do, o 742 /* If there's no more to do, or if the descriptor is full,
697 let the IO rip! */ 743 let the IO rip! */
698 744
699 if (bufs == journal->j_wbufsiz 745 if (bufs == journal->j_wbufsize ||
700 commit_transaction->t_buff 746 commit_transaction->t_buffers == NULL ||
701 space_left < tag_bytes + 1 747 space_left < tag_bytes + 16 + csum_size) {
702 748
703 jbd2_debug(4, "JBD2: S !! 749 jbd_debug(4, "JBD2: Submit %d IOs\n", bufs);
704 750
705 /* Write an end-of-des 751 /* Write an end-of-descriptor marker before
706 submitting the IOs. 752 submitting the IOs. "tag" still points to
707 the last tag we set 753 the last tag we set up. */
708 754
709 tag->t_flags |= cpu_to 755 tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
710 start_journal_io: 756 start_journal_io:
711 if (descriptor) 757 if (descriptor)
712 jbd2_descripto 758 jbd2_descriptor_block_csum_set(journal,
713 759 descriptor);
714 760
715 for (i = 0; i < bufs; 761 for (i = 0; i < bufs; i++) {
716 struct buffer_ 762 struct buffer_head *bh = wbuf[i];
717 <<
718 /* 763 /*
719 * Compute che 764 * Compute checksum.
720 */ 765 */
721 if (jbd2_has_f 766 if (jbd2_has_feature_checksum(journal)) {
722 crc32_ 767 crc32_sum =
723 jb 768 jbd2_checksum_data(crc32_sum, bh);
724 } 769 }
725 770
726 lock_buffer(bh 771 lock_buffer(bh);
727 clear_buffer_d 772 clear_buffer_dirty(bh);
728 set_buffer_upt 773 set_buffer_uptodate(bh);
729 bh->b_end_io = 774 bh->b_end_io = journal_end_buffer_io_sync;
730 submit_bh(REQ_ !! 775 submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
731 bh); <<
732 } 776 }
733 cond_resched(); 777 cond_resched();
734 778
735 /* Force a new descrip 779 /* Force a new descriptor to be generated next
736 time round the loop 780 time round the loop. */
737 descriptor = NULL; 781 descriptor = NULL;
738 bufs = 0; 782 bufs = 0;
739 } 783 }
740 } 784 }
741 785
742 err = journal_finish_inode_data_buffer 786 err = journal_finish_inode_data_buffers(journal, commit_transaction);
743 if (err) { 787 if (err) {
744 printk(KERN_WARNING 788 printk(KERN_WARNING
745 "JBD2: Detected IO err 789 "JBD2: Detected IO errors while flushing file data "
746 "on %s\n", journal->j_d 790 "on %s\n", journal->j_devname);
747 if (journal->j_flags & JBD2_AB 791 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
748 jbd2_journal_abort(jou 792 jbd2_journal_abort(journal, err);
749 err = 0; 793 err = 0;
750 } 794 }
751 795
752 /* 796 /*
753 * Get current oldest transaction in t 797 * Get current oldest transaction in the log before we issue flush
754 * to the filesystem device. After the 798 * to the filesystem device. After the flush we can be sure that
755 * blocks of all older transactions ar 799 * blocks of all older transactions are checkpointed to persistent
756 * storage and we will be safe to upda 800 * storage and we will be safe to update journal start in the
757 * superblock with the numbers we get 801 * superblock with the numbers we get here.
758 */ 802 */
759 update_tail = 803 update_tail =
760 jbd2_journal_get_log_tail(jour 804 jbd2_journal_get_log_tail(journal, &first_tid, &first_block);
761 805
762 write_lock(&journal->j_state_lock); 806 write_lock(&journal->j_state_lock);
763 if (update_tail) { 807 if (update_tail) {
764 long freed = first_block - jou 808 long freed = first_block - journal->j_tail;
765 809
766 if (first_block < journal->j_t 810 if (first_block < journal->j_tail)
767 freed += journal->j_la 811 freed += journal->j_last - journal->j_first;
768 /* Update tail only if we free 812 /* Update tail only if we free significant amount of space */
769 if (freed < journal->j_max_tra !! 813 if (freed < jbd2_journal_get_max_txn_bufs(journal))
770 update_tail = 0; 814 update_tail = 0;
771 } 815 }
772 J_ASSERT(commit_transaction->t_state = 816 J_ASSERT(commit_transaction->t_state == T_COMMIT);
773 commit_transaction->t_state = T_COMMIT 817 commit_transaction->t_state = T_COMMIT_DFLUSH;
774 write_unlock(&journal->j_state_lock); 818 write_unlock(&journal->j_state_lock);
775 819
776 /* !! 820 /*
777 * If the journal is not located on th 821 * If the journal is not located on the file system device,
778 * then we must flush the file system 822 * then we must flush the file system device before we issue
779 * the commit record 823 * the commit record
780 */ 824 */
781 if (commit_transaction->t_need_data_fl 825 if (commit_transaction->t_need_data_flush &&
782 (journal->j_fs_dev != journal->j_d 826 (journal->j_fs_dev != journal->j_dev) &&
783 (journal->j_flags & JBD2_BARRIER)) 827 (journal->j_flags & JBD2_BARRIER))
784 blkdev_issue_flush(journal->j_ 828 blkdev_issue_flush(journal->j_fs_dev);
785 829
786 /* Done it all: now write the commit r 830 /* Done it all: now write the commit record asynchronously. */
787 if (jbd2_has_feature_async_commit(jour 831 if (jbd2_has_feature_async_commit(journal)) {
788 err = journal_submit_commit_re 832 err = journal_submit_commit_record(journal, commit_transaction,
789 833 &cbh, crc32_sum);
790 if (err) 834 if (err)
791 jbd2_journal_abort(jou 835 jbd2_journal_abort(journal, err);
792 } 836 }
793 837
794 blk_finish_plug(&plug); 838 blk_finish_plug(&plug);
795 839
796 /* Lo and behold: we have just managed 840 /* Lo and behold: we have just managed to send a transaction to
797 the log. Before we can commit it, 841 the log. Before we can commit it, wait for the IO so far to
798 complete. Control buffers being wr 842 complete. Control buffers being written are on the
799 transaction's t_log_list queue, and 843 transaction's t_log_list queue, and metadata buffers are on
800 the io_bufs list. 844 the io_bufs list.
801 845
802 Wait for the buffers in reverse ord 846 Wait for the buffers in reverse order. That way we are
803 less likely to be woken up until al 847 less likely to be woken up until all IOs have completed, and
804 so we incur less scheduling load. 848 so we incur less scheduling load.
805 */ 849 */
806 850
807 jbd2_debug(3, "JBD2: commit phase 3\n" !! 851 jbd_debug(3, "JBD2: commit phase 3\n");
808 852
809 while (!list_empty(&io_bufs)) { 853 while (!list_empty(&io_bufs)) {
810 struct buffer_head *bh = list_ 854 struct buffer_head *bh = list_entry(io_bufs.prev,
811 855 struct buffer_head,
812 856 b_assoc_buffers);
813 857
814 wait_on_buffer(bh); 858 wait_on_buffer(bh);
815 cond_resched(); 859 cond_resched();
816 860
817 if (unlikely(!buffer_uptodate( 861 if (unlikely(!buffer_uptodate(bh)))
818 err = -EIO; 862 err = -EIO;
819 jbd2_unfile_log_bh(bh); 863 jbd2_unfile_log_bh(bh);
820 stats.run.rs_blocks_logged++; 864 stats.run.rs_blocks_logged++;
821 865
822 /* 866 /*
823 * The list contains temporary 867 * The list contains temporary buffer heads created by
824 * jbd2_journal_write_metadata 868 * jbd2_journal_write_metadata_buffer().
825 */ 869 */
826 BUFFER_TRACE(bh, "dumping temp 870 BUFFER_TRACE(bh, "dumping temporary bh");
827 __brelse(bh); 871 __brelse(bh);
828 J_ASSERT_BH(bh, atomic_read(&b 872 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
829 free_buffer_head(bh); 873 free_buffer_head(bh);
830 874
831 /* We also have to refile the 875 /* We also have to refile the corresponding shadowed buffer */
832 jh = commit_transaction->t_sha 876 jh = commit_transaction->t_shadow_list->b_tprev;
833 bh = jh2bh(jh); 877 bh = jh2bh(jh);
834 clear_buffer_jwrite(bh); 878 clear_buffer_jwrite(bh);
835 J_ASSERT_BH(bh, buffer_jbddirt 879 J_ASSERT_BH(bh, buffer_jbddirty(bh));
836 J_ASSERT_BH(bh, !buffer_shadow 880 J_ASSERT_BH(bh, !buffer_shadow(bh));
837 881
838 /* The metadata is now release 882 /* The metadata is now released for reuse, but we need
839 to remember it against this 883 to remember it against this transaction so that when
840 we finally commit, we can d 884 we finally commit, we can do any checkpointing
841 required. */ 885 required. */
842 JBUFFER_TRACE(jh, "file as BJ_ 886 JBUFFER_TRACE(jh, "file as BJ_Forget");
843 jbd2_journal_file_buffer(jh, c 887 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
844 JBUFFER_TRACE(jh, "brelse shad 888 JBUFFER_TRACE(jh, "brelse shadowed buffer");
845 __brelse(bh); 889 __brelse(bh);
846 } 890 }
847 891
848 J_ASSERT (commit_transaction->t_shadow 892 J_ASSERT (commit_transaction->t_shadow_list == NULL);
849 893
850 jbd2_debug(3, "JBD2: commit phase 4\n" !! 894 jbd_debug(3, "JBD2: commit phase 4\n");
851 895
852 /* Here we wait for the revoke record 896 /* Here we wait for the revoke record and descriptor record buffers */
853 while (!list_empty(&log_bufs)) { 897 while (!list_empty(&log_bufs)) {
854 struct buffer_head *bh; 898 struct buffer_head *bh;
855 899
856 bh = list_entry(log_bufs.prev, 900 bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
857 wait_on_buffer(bh); 901 wait_on_buffer(bh);
858 cond_resched(); 902 cond_resched();
859 903
860 if (unlikely(!buffer_uptodate( 904 if (unlikely(!buffer_uptodate(bh)))
861 err = -EIO; 905 err = -EIO;
862 906
863 BUFFER_TRACE(bh, "ph5: control 907 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
864 clear_buffer_jwrite(bh); 908 clear_buffer_jwrite(bh);
865 jbd2_unfile_log_bh(bh); 909 jbd2_unfile_log_bh(bh);
866 stats.run.rs_blocks_logged++; 910 stats.run.rs_blocks_logged++;
867 __brelse(bh); /* One 911 __brelse(bh); /* One for getblk */
868 /* AKPM: bforget here */ 912 /* AKPM: bforget here */
869 } 913 }
870 914
871 if (err) 915 if (err)
872 jbd2_journal_abort(journal, er 916 jbd2_journal_abort(journal, err);
873 917
874 jbd2_debug(3, "JBD2: commit phase 5\n" !! 918 jbd_debug(3, "JBD2: commit phase 5\n");
875 write_lock(&journal->j_state_lock); 919 write_lock(&journal->j_state_lock);
876 J_ASSERT(commit_transaction->t_state = 920 J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
877 commit_transaction->t_state = T_COMMIT 921 commit_transaction->t_state = T_COMMIT_JFLUSH;
878 write_unlock(&journal->j_state_lock); 922 write_unlock(&journal->j_state_lock);
879 923
880 if (!jbd2_has_feature_async_commit(jou 924 if (!jbd2_has_feature_async_commit(journal)) {
881 err = journal_submit_commit_re 925 err = journal_submit_commit_record(journal, commit_transaction,
882 926 &cbh, crc32_sum);
883 if (err) 927 if (err)
884 jbd2_journal_abort(jou 928 jbd2_journal_abort(journal, err);
885 } 929 }
886 if (cbh) 930 if (cbh)
887 err = journal_wait_on_commit_r 931 err = journal_wait_on_commit_record(journal, cbh);
888 stats.run.rs_blocks_logged++; 932 stats.run.rs_blocks_logged++;
889 if (jbd2_has_feature_async_commit(jour 933 if (jbd2_has_feature_async_commit(journal) &&
890 journal->j_flags & JBD2_BARRIER) { 934 journal->j_flags & JBD2_BARRIER) {
891 blkdev_issue_flush(journal->j_ 935 blkdev_issue_flush(journal->j_dev);
892 } 936 }
893 937
894 if (err) 938 if (err)
895 jbd2_journal_abort(journal, er 939 jbd2_journal_abort(journal, err);
896 940
897 WARN_ON_ONCE( 941 WARN_ON_ONCE(
898 atomic_read(&commit_transactio 942 atomic_read(&commit_transaction->t_outstanding_credits) < 0);
899 943
900 /* 944 /*
901 * Now disk caches for filesystem devi 945 * Now disk caches for filesystem device are flushed so we are safe to
902 * erase checkpointed transactions fro 946 * erase checkpointed transactions from the log by updating journal
903 * superblock. 947 * superblock.
904 */ 948 */
905 if (update_tail) 949 if (update_tail)
906 jbd2_update_log_tail(journal, 950 jbd2_update_log_tail(journal, first_tid, first_block);
907 951
908 /* End of a transaction! Finally, we 952 /* End of a transaction! Finally, we can do checkpoint
909 processing: any buffers committed a 953 processing: any buffers committed as a result of this
910 transaction can be removed from any 954 transaction can be removed from any checkpoint list it was on
911 before. */ 955 before. */
912 956
913 jbd2_debug(3, "JBD2: commit phase 6\n" !! 957 jbd_debug(3, "JBD2: commit phase 6\n");
914 958
915 J_ASSERT(list_empty(&commit_transactio 959 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
916 J_ASSERT(commit_transaction->t_buffers 960 J_ASSERT(commit_transaction->t_buffers == NULL);
917 J_ASSERT(commit_transaction->t_checkpo 961 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
918 J_ASSERT(commit_transaction->t_shadow_ 962 J_ASSERT(commit_transaction->t_shadow_list == NULL);
919 963
920 restart_loop: 964 restart_loop:
921 /* 965 /*
922 * As there are other places (journal_ 966 * As there are other places (journal_unmap_buffer()) adding buffers
923 * to this list we have to be careful 967 * to this list we have to be careful and hold the j_list_lock.
924 */ 968 */
925 spin_lock(&journal->j_list_lock); 969 spin_lock(&journal->j_list_lock);
926 while (commit_transaction->t_forget) { 970 while (commit_transaction->t_forget) {
927 transaction_t *cp_transaction; 971 transaction_t *cp_transaction;
928 struct buffer_head *bh; 972 struct buffer_head *bh;
929 int try_to_free = 0; 973 int try_to_free = 0;
930 bool drop_ref; 974 bool drop_ref;
931 975
932 jh = commit_transaction->t_for 976 jh = commit_transaction->t_forget;
933 spin_unlock(&journal->j_list_l 977 spin_unlock(&journal->j_list_lock);
934 bh = jh2bh(jh); 978 bh = jh2bh(jh);
935 /* 979 /*
936 * Get a reference so that bh 980 * Get a reference so that bh cannot be freed before we are
937 * done with it. 981 * done with it.
938 */ 982 */
939 get_bh(bh); 983 get_bh(bh);
940 spin_lock(&jh->b_state_lock); 984 spin_lock(&jh->b_state_lock);
941 J_ASSERT_JH(jh, jh->b_transact 985 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
942 986
943 /* 987 /*
944 * If there is undo-protected 988 * If there is undo-protected committed data against
945 * this buffer, then we can re 989 * this buffer, then we can remove it now. If it is a
946 * buffer needing such protect 990 * buffer needing such protection, the old frozen_data
947 * field now points to a commi 991 * field now points to a committed version of the
948 * buffer, so rotate that fiel 992 * buffer, so rotate that field to the new committed
949 * data. 993 * data.
950 * 994 *
951 * Otherwise, we can just thro 995 * Otherwise, we can just throw away the frozen data now.
952 * 996 *
953 * We also know that the froze 997 * We also know that the frozen data has already fired
954 * its triggers if they exist, 998 * its triggers if they exist, so we can clear that too.
955 */ 999 */
956 if (jh->b_committed_data) { 1000 if (jh->b_committed_data) {
957 jbd2_free(jh->b_commit 1001 jbd2_free(jh->b_committed_data, bh->b_size);
958 jh->b_committed_data = 1002 jh->b_committed_data = NULL;
959 if (jh->b_frozen_data) 1003 if (jh->b_frozen_data) {
960 jh->b_committe 1004 jh->b_committed_data = jh->b_frozen_data;
961 jh->b_frozen_d 1005 jh->b_frozen_data = NULL;
962 jh->b_frozen_t 1006 jh->b_frozen_triggers = NULL;
963 } 1007 }
964 } else if (jh->b_frozen_data) 1008 } else if (jh->b_frozen_data) {
965 jbd2_free(jh->b_frozen 1009 jbd2_free(jh->b_frozen_data, bh->b_size);
966 jh->b_frozen_data = NU 1010 jh->b_frozen_data = NULL;
967 jh->b_frozen_triggers 1011 jh->b_frozen_triggers = NULL;
968 } 1012 }
969 1013
970 spin_lock(&journal->j_list_loc 1014 spin_lock(&journal->j_list_lock);
971 cp_transaction = jh->b_cp_tran 1015 cp_transaction = jh->b_cp_transaction;
972 if (cp_transaction) { 1016 if (cp_transaction) {
973 JBUFFER_TRACE(jh, "rem 1017 JBUFFER_TRACE(jh, "remove from old cp transaction");
974 cp_transaction->t_chp_ 1018 cp_transaction->t_chp_stats.cs_dropped++;
975 __jbd2_journal_remove_ 1019 __jbd2_journal_remove_checkpoint(jh);
976 } 1020 }
977 1021
978 /* Only re-checkpoint the buff 1022 /* Only re-checkpoint the buffer_head if it is marked
979 * dirty. If the buffer was a 1023 * dirty. If the buffer was added to the BJ_Forget list
980 * by jbd2_journal_forget, it 1024 * by jbd2_journal_forget, it may no longer be dirty and
981 * there's no point in keeping 1025 * there's no point in keeping a checkpoint record for
982 * it. */ 1026 * it. */
983 1027
984 /* 1028 /*
985 * A buffer which has been fre 1029 * A buffer which has been freed while still being journaled
986 * by a previous transaction, 1030 * by a previous transaction, refile the buffer to BJ_Forget of
987 * the running transaction. If 1031 * the running transaction. If the just committed transaction
988 * contains "add to orphan" op 1032 * contains "add to orphan" operation, we can completely
989 * invalidate the buffer now. 1033 * invalidate the buffer now. We are rather through in that
990 * since the buffer may be sti 1034 * since the buffer may be still accessible when blocksize <
991 * pagesize and it is attached 1035 * pagesize and it is attached to the last partial page.
992 */ 1036 */
993 if (buffer_freed(bh) && !jh->b 1037 if (buffer_freed(bh) && !jh->b_next_transaction) {
994 struct address_space * 1038 struct address_space *mapping;
995 1039
996 clear_buffer_freed(bh) 1040 clear_buffer_freed(bh);
997 clear_buffer_jbddirty( 1041 clear_buffer_jbddirty(bh);
998 1042
999 /* 1043 /*
1000 * Block device buffe 1044 * Block device buffers need to stay mapped all the
1001 * time, so it is eno 1045 * time, so it is enough to clear buffer_jbddirty and
1002 * buffer_freed bits. 1046 * buffer_freed bits. For the file mapping buffers (i.e.
1003 * journalled data) w 1047 * journalled data) we need to unmap buffer and clear
1004 * more bits. We also 1048 * more bits. We also need to be careful about the check
1005 * because the data p 1049 * because the data page mapping can get cleared under
1006 * our hands. Note th 1050 * our hands. Note that if mapping == NULL, we don't
1007 * need to make buffe 1051 * need to make buffer unmapped because the page is
1008 * already detached f 1052 * already detached from the mapping and buffers cannot
1009 * get reused. 1053 * get reused.
1010 */ 1054 */
1011 mapping = READ_ONCE(b !! 1055 mapping = READ_ONCE(bh->b_page->mapping);
1012 if (mapping && !sb_is 1056 if (mapping && !sb_is_blkdev_sb(mapping->host->i_sb)) {
1013 clear_buffer_ 1057 clear_buffer_mapped(bh);
1014 clear_buffer_ 1058 clear_buffer_new(bh);
1015 clear_buffer_ 1059 clear_buffer_req(bh);
1016 bh->b_bdev = 1060 bh->b_bdev = NULL;
1017 } 1061 }
1018 } 1062 }
1019 1063
1020 if (buffer_jbddirty(bh)) { 1064 if (buffer_jbddirty(bh)) {
1021 JBUFFER_TRACE(jh, "ad 1065 JBUFFER_TRACE(jh, "add to new checkpointing trans");
1022 __jbd2_journal_insert 1066 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
1023 if (is_journal_aborte 1067 if (is_journal_aborted(journal))
1024 clear_buffer_ 1068 clear_buffer_jbddirty(bh);
1025 } else { 1069 } else {
1026 J_ASSERT_BH(bh, !buff 1070 J_ASSERT_BH(bh, !buffer_dirty(bh));
1027 /* 1071 /*
1028 * The buffer on BJ_F 1072 * The buffer on BJ_Forget list and not jbddirty means
1029 * it has been freed 1073 * it has been freed by this transaction and hence it
1030 * could not have bee 1074 * could not have been reallocated until this
1031 * transaction has co 1075 * transaction has committed. *BUT* it could be
1032 * reallocated once w 1076 * reallocated once we have written all the data to
1033 * disk and before we 1077 * disk and before we process the buffer on BJ_Forget
1034 * list. 1078 * list.
1035 */ 1079 */
1036 if (!jh->b_next_trans 1080 if (!jh->b_next_transaction)
1037 try_to_free = 1081 try_to_free = 1;
1038 } 1082 }
1039 JBUFFER_TRACE(jh, "refile or 1083 JBUFFER_TRACE(jh, "refile or unfile buffer");
1040 drop_ref = __jbd2_journal_ref 1084 drop_ref = __jbd2_journal_refile_buffer(jh);
1041 spin_unlock(&jh->b_state_lock 1085 spin_unlock(&jh->b_state_lock);
1042 if (drop_ref) 1086 if (drop_ref)
1043 jbd2_journal_put_jour 1087 jbd2_journal_put_journal_head(jh);
1044 if (try_to_free) 1088 if (try_to_free)
1045 release_buffer_page(b 1089 release_buffer_page(bh); /* Drops bh reference */
1046 else 1090 else
1047 __brelse(bh); 1091 __brelse(bh);
1048 cond_resched_lock(&journal->j 1092 cond_resched_lock(&journal->j_list_lock);
1049 } 1093 }
1050 spin_unlock(&journal->j_list_lock); 1094 spin_unlock(&journal->j_list_lock);
1051 /* 1095 /*
1052 * This is a bit sleazy. We use j_li 1096 * This is a bit sleazy. We use j_list_lock to protect transition
1053 * of a transaction into T_FINISHED s 1097 * of a transaction into T_FINISHED state and calling
1054 * __jbd2_journal_drop_transaction(). 1098 * __jbd2_journal_drop_transaction(). Otherwise we could race with
1055 * other checkpointing code processin 1099 * other checkpointing code processing the transaction...
1056 */ 1100 */
1057 write_lock(&journal->j_state_lock); 1101 write_lock(&journal->j_state_lock);
1058 spin_lock(&journal->j_list_lock); 1102 spin_lock(&journal->j_list_lock);
1059 /* 1103 /*
1060 * Now recheck if some buffers did no 1104 * Now recheck if some buffers did not get attached to the transaction
1061 * while the lock was dropped... 1105 * while the lock was dropped...
1062 */ 1106 */
1063 if (commit_transaction->t_forget) { 1107 if (commit_transaction->t_forget) {
1064 spin_unlock(&journal->j_list_ 1108 spin_unlock(&journal->j_list_lock);
1065 write_unlock(&journal->j_stat 1109 write_unlock(&journal->j_state_lock);
1066 goto restart_loop; 1110 goto restart_loop;
1067 } 1111 }
1068 1112
1069 /* Add the transaction to the checkpo 1113 /* Add the transaction to the checkpoint list
1070 * __journal_remove_checkpoint() can 1114 * __journal_remove_checkpoint() can not destroy transaction
1071 * under us because it is not marked 1115 * under us because it is not marked as T_FINISHED yet */
1072 if (journal->j_checkpoint_transaction 1116 if (journal->j_checkpoint_transactions == NULL) {
1073 journal->j_checkpoint_transac 1117 journal->j_checkpoint_transactions = commit_transaction;
1074 commit_transaction->t_cpnext 1118 commit_transaction->t_cpnext = commit_transaction;
1075 commit_transaction->t_cpprev 1119 commit_transaction->t_cpprev = commit_transaction;
1076 } else { 1120 } else {
1077 commit_transaction->t_cpnext 1121 commit_transaction->t_cpnext =
1078 journal->j_checkpoint 1122 journal->j_checkpoint_transactions;
1079 commit_transaction->t_cpprev 1123 commit_transaction->t_cpprev =
1080 commit_transaction->t 1124 commit_transaction->t_cpnext->t_cpprev;
1081 commit_transaction->t_cpnext- 1125 commit_transaction->t_cpnext->t_cpprev =
1082 commit_transaction; 1126 commit_transaction;
1083 commit_transaction->t_cpprev- 1127 commit_transaction->t_cpprev->t_cpnext =
1084 commit_transa 1128 commit_transaction;
1085 } 1129 }
1086 spin_unlock(&journal->j_list_lock); 1130 spin_unlock(&journal->j_list_lock);
1087 1131
1088 /* Done with this transaction! */ 1132 /* Done with this transaction! */
1089 1133
1090 jbd2_debug(3, "JBD2: commit phase 7\n !! 1134 jbd_debug(3, "JBD2: commit phase 7\n");
1091 1135
1092 J_ASSERT(commit_transaction->t_state 1136 J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
1093 1137
1094 commit_transaction->t_start = jiffies 1138 commit_transaction->t_start = jiffies;
1095 stats.run.rs_logging = jbd2_time_diff 1139 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
1096 1140 commit_transaction->t_start);
1097 1141
1098 /* 1142 /*
1099 * File the transaction statistics 1143 * File the transaction statistics
1100 */ 1144 */
1101 stats.ts_tid = commit_transaction->t_ 1145 stats.ts_tid = commit_transaction->t_tid;
1102 stats.run.rs_handle_count = 1146 stats.run.rs_handle_count =
1103 atomic_read(&commit_transacti 1147 atomic_read(&commit_transaction->t_handle_count);
1104 trace_jbd2_run_stats(journal->j_fs_de 1148 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
1105 commit_transacti 1149 commit_transaction->t_tid, &stats.run);
1106 stats.ts_requested = (commit_transact 1150 stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0;
1107 1151
1108 commit_transaction->t_state = T_COMMI 1152 commit_transaction->t_state = T_COMMIT_CALLBACK;
1109 J_ASSERT(commit_transaction == journa 1153 J_ASSERT(commit_transaction == journal->j_committing_transaction);
1110 WRITE_ONCE(journal->j_commit_sequence !! 1154 journal->j_commit_sequence = commit_transaction->t_tid;
1111 journal->j_committing_transaction = N 1155 journal->j_committing_transaction = NULL;
1112 commit_time = ktime_to_ns(ktime_sub(k 1156 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1113 1157
1114 /* 1158 /*
1115 * weight the commit time higher than 1159 * weight the commit time higher than the average time so we don't
1116 * react too strongly to vast changes 1160 * react too strongly to vast changes in the commit time
1117 */ 1161 */
1118 if (likely(journal->j_average_commit_ 1162 if (likely(journal->j_average_commit_time))
1119 journal->j_average_commit_tim 1163 journal->j_average_commit_time = (commit_time +
1120 journal->j_av 1164 journal->j_average_commit_time*3) / 4;
1121 else 1165 else
1122 journal->j_average_commit_tim 1166 journal->j_average_commit_time = commit_time;
1123 1167
1124 write_unlock(&journal->j_state_lock); 1168 write_unlock(&journal->j_state_lock);
1125 1169
1126 if (journal->j_commit_callback) 1170 if (journal->j_commit_callback)
1127 journal->j_commit_callback(jo 1171 journal->j_commit_callback(journal, commit_transaction);
1128 if (journal->j_fc_cleanup_callback) 1172 if (journal->j_fc_cleanup_callback)
1129 journal->j_fc_cleanup_callbac !! 1173 journal->j_fc_cleanup_callback(journal, 1);
1130 1174
1131 trace_jbd2_end_commit(journal, commit 1175 trace_jbd2_end_commit(journal, commit_transaction);
1132 jbd2_debug(1, "JBD2: commit %d comple !! 1176 jbd_debug(1, "JBD2: commit %d complete, head %d\n",
1133 journal->j_commit_sequence, 1177 journal->j_commit_sequence, journal->j_tail_sequence);
1134 1178
1135 write_lock(&journal->j_state_lock); 1179 write_lock(&journal->j_state_lock);
1136 journal->j_flags &= ~JBD2_FULL_COMMIT 1180 journal->j_flags &= ~JBD2_FULL_COMMIT_ONGOING;
1137 journal->j_flags &= ~JBD2_FAST_COMMIT 1181 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
1138 spin_lock(&journal->j_list_lock); 1182 spin_lock(&journal->j_list_lock);
1139 commit_transaction->t_state = T_FINIS 1183 commit_transaction->t_state = T_FINISHED;
1140 /* Check if the transaction can be dr 1184 /* Check if the transaction can be dropped now that we are finished */
1141 if (commit_transaction->t_checkpoint_ !! 1185 if (commit_transaction->t_checkpoint_list == NULL &&
>> 1186 commit_transaction->t_checkpoint_io_list == NULL) {
1142 __jbd2_journal_drop_transacti 1187 __jbd2_journal_drop_transaction(journal, commit_transaction);
1143 jbd2_journal_free_transaction 1188 jbd2_journal_free_transaction(commit_transaction);
1144 } 1189 }
1145 spin_unlock(&journal->j_list_lock); 1190 spin_unlock(&journal->j_list_lock);
1146 write_unlock(&journal->j_state_lock); 1191 write_unlock(&journal->j_state_lock);
1147 wake_up(&journal->j_wait_done_commit) 1192 wake_up(&journal->j_wait_done_commit);
1148 wake_up(&journal->j_fc_wait); 1193 wake_up(&journal->j_fc_wait);
1149 1194
1150 /* 1195 /*
1151 * Calculate overall stats 1196 * Calculate overall stats
1152 */ 1197 */
1153 spin_lock(&journal->j_history_lock); 1198 spin_lock(&journal->j_history_lock);
1154 journal->j_stats.ts_tid++; 1199 journal->j_stats.ts_tid++;
1155 journal->j_stats.ts_requested += stat 1200 journal->j_stats.ts_requested += stats.ts_requested;
1156 journal->j_stats.run.rs_wait += stats 1201 journal->j_stats.run.rs_wait += stats.run.rs_wait;
1157 journal->j_stats.run.rs_request_delay 1202 journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
1158 journal->j_stats.run.rs_running += st 1203 journal->j_stats.run.rs_running += stats.run.rs_running;
1159 journal->j_stats.run.rs_locked += sta 1204 journal->j_stats.run.rs_locked += stats.run.rs_locked;
1160 journal->j_stats.run.rs_flushing += s 1205 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
1161 journal->j_stats.run.rs_logging += st 1206 journal->j_stats.run.rs_logging += stats.run.rs_logging;
1162 journal->j_stats.run.rs_handle_count 1207 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
1163 journal->j_stats.run.rs_blocks += sta 1208 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
1164 journal->j_stats.run.rs_blocks_logged 1209 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
1165 spin_unlock(&journal->j_history_lock) 1210 spin_unlock(&journal->j_history_lock);
1166 } 1211 }
1167 1212
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.