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 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 folio = bh->b_folio;
72 if (folio->mapping) 72 if (folio->mapping)
73 goto nope; 73 goto nope;
74 74
75 /* OK, it's a truncated page */ 75 /* OK, it's a truncated page */
76 if (!folio_trylock(folio)) 76 if (!folio_trylock(folio))
77 goto nope; 77 goto nope;
78 78
79 folio_get(folio); 79 folio_get(folio);
80 __brelse(bh); 80 __brelse(bh);
81 try_to_free_buffers(folio); 81 try_to_free_buffers(folio);
82 folio_unlock(folio); 82 folio_unlock(folio);
83 folio_put(folio); 83 folio_put(folio);
84 return; 84 return;
85 85
86 nope: 86 nope:
87 __brelse(bh); 87 __brelse(bh);
88 } 88 }
89 89
90 static void jbd2_commit_block_csum_set(journal 90 static void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
91 { 91 {
92 struct commit_header *h; 92 struct commit_header *h;
93 __u32 csum; 93 __u32 csum;
94 94
95 if (!jbd2_journal_has_csum_v2or3(j)) 95 if (!jbd2_journal_has_csum_v2or3(j))
96 return; 96 return;
97 97
98 h = (struct commit_header *)(bh->b_dat 98 h = (struct commit_header *)(bh->b_data);
99 h->h_chksum_type = 0; 99 h->h_chksum_type = 0;
100 h->h_chksum_size = 0; 100 h->h_chksum_size = 0;
101 h->h_chksum[0] = 0; 101 h->h_chksum[0] = 0;
102 csum = jbd2_chksum(j, j->j_csum_seed, 102 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
103 h->h_chksum[0] = cpu_to_be32(csum); 103 h->h_chksum[0] = cpu_to_be32(csum);
104 } 104 }
105 105
106 /* 106 /*
107 * Done it all: now submit the commit record. 107 * Done it all: now submit the commit record. We should have
108 * cleaned up our previous buffers by now, so 108 * cleaned up our previous buffers by now, so if we are in abort
109 * mode we can now just skip the rest of the j 109 * mode we can now just skip the rest of the journal write
110 * entirely. 110 * entirely.
111 * 111 *
112 * Returns 1 if the journal needs to be aborte 112 * Returns 1 if the journal needs to be aborted or 0 on success
113 */ 113 */
114 static int journal_submit_commit_record(journa 114 static int journal_submit_commit_record(journal_t *journal,
115 transa 115 transaction_t *commit_transaction,
116 struct 116 struct buffer_head **cbh,
117 __u32 117 __u32 crc32_sum)
118 { 118 {
119 struct commit_header *tmp; 119 struct commit_header *tmp;
120 struct buffer_head *bh; 120 struct buffer_head *bh;
121 struct timespec64 now; 121 struct timespec64 now;
122 blk_opf_t write_flags = REQ_OP_WRITE | !! 122 blk_opf_t write_flags = REQ_OP_WRITE | REQ_SYNC;
123 123
124 *cbh = NULL; 124 *cbh = NULL;
125 125
126 if (is_journal_aborted(journal)) 126 if (is_journal_aborted(journal))
127 return 0; 127 return 0;
128 128
129 bh = jbd2_journal_get_descriptor_buffe 129 bh = jbd2_journal_get_descriptor_buffer(commit_transaction,
130 130 JBD2_COMMIT_BLOCK);
131 if (!bh) 131 if (!bh)
132 return 1; 132 return 1;
133 133
134 tmp = (struct commit_header *)bh->b_da 134 tmp = (struct commit_header *)bh->b_data;
135 ktime_get_coarse_real_ts64(&now); 135 ktime_get_coarse_real_ts64(&now);
136 tmp->h_commit_sec = cpu_to_be64(now.tv 136 tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
137 tmp->h_commit_nsec = cpu_to_be32(now.t 137 tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
138 138
139 if (jbd2_has_feature_checksum(journal) 139 if (jbd2_has_feature_checksum(journal)) {
140 tmp->h_chksum_type = JBD2 140 tmp->h_chksum_type = JBD2_CRC32_CHKSUM;
141 tmp->h_chksum_size = JBD2 141 tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE;
142 tmp->h_chksum[0] = cpu_ 142 tmp->h_chksum[0] = cpu_to_be32(crc32_sum);
143 } 143 }
144 jbd2_commit_block_csum_set(journal, bh 144 jbd2_commit_block_csum_set(journal, bh);
145 145
146 BUFFER_TRACE(bh, "submit commit block" 146 BUFFER_TRACE(bh, "submit commit block");
147 lock_buffer(bh); 147 lock_buffer(bh);
148 clear_buffer_dirty(bh); 148 clear_buffer_dirty(bh);
149 set_buffer_uptodate(bh); 149 set_buffer_uptodate(bh);
150 bh->b_end_io = journal_end_buffer_io_s 150 bh->b_end_io = journal_end_buffer_io_sync;
151 151
152 if (journal->j_flags & JBD2_BARRIER && 152 if (journal->j_flags & JBD2_BARRIER &&
153 !jbd2_has_feature_async_commit(jou 153 !jbd2_has_feature_async_commit(journal))
154 write_flags |= REQ_PREFLUSH | 154 write_flags |= REQ_PREFLUSH | REQ_FUA;
155 155
156 submit_bh(write_flags, bh); 156 submit_bh(write_flags, bh);
157 *cbh = bh; 157 *cbh = bh;
158 return 0; 158 return 0;
159 } 159 }
160 160
161 /* 161 /*
162 * This function along with journal_submit_com 162 * This function along with journal_submit_commit_record
163 * allows to write the commit record asynchron 163 * allows to write the commit record asynchronously.
164 */ 164 */
165 static int journal_wait_on_commit_record(journ 165 static int journal_wait_on_commit_record(journal_t *journal,
166 struc 166 struct buffer_head *bh)
167 { 167 {
168 int ret = 0; 168 int ret = 0;
169 169
170 clear_buffer_dirty(bh); 170 clear_buffer_dirty(bh);
171 wait_on_buffer(bh); 171 wait_on_buffer(bh);
172 172
173 if (unlikely(!buffer_uptodate(bh))) 173 if (unlikely(!buffer_uptodate(bh)))
174 ret = -EIO; 174 ret = -EIO;
175 put_bh(bh); /* One for getb 175 put_bh(bh); /* One for getblk() */
176 176
177 return ret; 177 return ret;
178 } 178 }
179 179
180 /* Send all the data buffers related to an ino 180 /* Send all the data buffers related to an inode */
181 int jbd2_submit_inode_data(journal_t *journal, 181 int jbd2_submit_inode_data(journal_t *journal, struct jbd2_inode *jinode)
182 { 182 {
183 if (!jinode || !(jinode->i_flags & JI_ 183 if (!jinode || !(jinode->i_flags & JI_WRITE_DATA))
184 return 0; 184 return 0;
185 185
186 trace_jbd2_submit_inode_data(jinode->i 186 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
187 return journal->j_submit_inode_data_bu 187 return journal->j_submit_inode_data_buffers(jinode);
188 188
189 } 189 }
190 EXPORT_SYMBOL(jbd2_submit_inode_data); 190 EXPORT_SYMBOL(jbd2_submit_inode_data);
191 191
192 int jbd2_wait_inode_data(journal_t *journal, s 192 int jbd2_wait_inode_data(journal_t *journal, struct jbd2_inode *jinode)
193 { 193 {
194 if (!jinode || !(jinode->i_flags & JI_ 194 if (!jinode || !(jinode->i_flags & JI_WAIT_DATA) ||
195 !jinode->i_vfs_inode || !jinod 195 !jinode->i_vfs_inode || !jinode->i_vfs_inode->i_mapping)
196 return 0; 196 return 0;
197 return filemap_fdatawait_range_keep_er 197 return filemap_fdatawait_range_keep_errors(
198 jinode->i_vfs_inode->i_mapping 198 jinode->i_vfs_inode->i_mapping, jinode->i_dirty_start,
199 jinode->i_dirty_end); 199 jinode->i_dirty_end);
200 } 200 }
201 EXPORT_SYMBOL(jbd2_wait_inode_data); 201 EXPORT_SYMBOL(jbd2_wait_inode_data);
202 202
203 /* 203 /*
204 * Submit all the data buffers of inode associ 204 * Submit all the data buffers of inode associated with the transaction to
205 * disk. 205 * disk.
206 * 206 *
207 * We are in a committing transaction. Therefo 207 * We are in a committing transaction. Therefore no new inode can be added to
208 * our inode list. We use JI_COMMIT_RUNNING fl 208 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
209 * operate on from being released while we wri 209 * operate on from being released while we write out pages.
210 */ 210 */
211 static int journal_submit_data_buffers(journal 211 static int journal_submit_data_buffers(journal_t *journal,
212 transaction_t *commit_transact 212 transaction_t *commit_transaction)
213 { 213 {
214 struct jbd2_inode *jinode; 214 struct jbd2_inode *jinode;
215 int err, ret = 0; 215 int err, ret = 0;
216 216
217 spin_lock(&journal->j_list_lock); 217 spin_lock(&journal->j_list_lock);
218 list_for_each_entry(jinode, &commit_tr 218 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
219 if (!(jinode->i_flags & JI_WRI 219 if (!(jinode->i_flags & JI_WRITE_DATA))
220 continue; 220 continue;
221 jinode->i_flags |= JI_COMMIT_R 221 jinode->i_flags |= JI_COMMIT_RUNNING;
222 spin_unlock(&journal->j_list_l 222 spin_unlock(&journal->j_list_lock);
223 /* submit the inode data buffe 223 /* submit the inode data buffers. */
224 trace_jbd2_submit_inode_data(j 224 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
225 if (journal->j_submit_inode_da 225 if (journal->j_submit_inode_data_buffers) {
226 err = journal->j_submi 226 err = journal->j_submit_inode_data_buffers(jinode);
227 if (!ret) 227 if (!ret)
228 ret = err; 228 ret = err;
229 } 229 }
230 spin_lock(&journal->j_list_loc 230 spin_lock(&journal->j_list_lock);
231 J_ASSERT(jinode->i_transaction 231 J_ASSERT(jinode->i_transaction == commit_transaction);
232 jinode->i_flags &= ~JI_COMMIT_ 232 jinode->i_flags &= ~JI_COMMIT_RUNNING;
233 smp_mb(); 233 smp_mb();
234 wake_up_bit(&jinode->i_flags, 234 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
235 } 235 }
236 spin_unlock(&journal->j_list_lock); 236 spin_unlock(&journal->j_list_lock);
237 return ret; 237 return ret;
238 } 238 }
239 239
240 int jbd2_journal_finish_inode_data_buffers(str 240 int jbd2_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
241 { 241 {
242 struct address_space *mapping = jinode 242 struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
243 243
244 return filemap_fdatawait_range_keep_er 244 return filemap_fdatawait_range_keep_errors(mapping,
245 245 jinode->i_dirty_start,
246 246 jinode->i_dirty_end);
247 } 247 }
248 248
249 /* 249 /*
250 * Wait for data submitted for writeout, refil 250 * Wait for data submitted for writeout, refile inodes to proper
251 * transaction if needed. 251 * transaction if needed.
252 * 252 *
253 */ 253 */
254 static int journal_finish_inode_data_buffers(j 254 static int journal_finish_inode_data_buffers(journal_t *journal,
255 transaction_t *commit_transact 255 transaction_t *commit_transaction)
256 { 256 {
257 struct jbd2_inode *jinode, *next_i; 257 struct jbd2_inode *jinode, *next_i;
258 int err, ret = 0; 258 int err, ret = 0;
259 259
260 /* For locking, see the comment in jou 260 /* For locking, see the comment in journal_submit_data_buffers() */
261 spin_lock(&journal->j_list_lock); 261 spin_lock(&journal->j_list_lock);
262 list_for_each_entry(jinode, &commit_tr 262 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
263 if (!(jinode->i_flags & JI_WAI 263 if (!(jinode->i_flags & JI_WAIT_DATA))
264 continue; 264 continue;
265 jinode->i_flags |= JI_COMMIT_R 265 jinode->i_flags |= JI_COMMIT_RUNNING;
266 spin_unlock(&journal->j_list_l 266 spin_unlock(&journal->j_list_lock);
267 /* wait for the inode data buf 267 /* wait for the inode data buffers writeout. */
268 if (journal->j_finish_inode_da 268 if (journal->j_finish_inode_data_buffers) {
269 err = journal->j_finis 269 err = journal->j_finish_inode_data_buffers(jinode);
270 if (!ret) 270 if (!ret)
271 ret = err; 271 ret = err;
272 } 272 }
273 cond_resched(); <<
274 spin_lock(&journal->j_list_loc 273 spin_lock(&journal->j_list_lock);
275 jinode->i_flags &= ~JI_COMMIT_ 274 jinode->i_flags &= ~JI_COMMIT_RUNNING;
276 smp_mb(); 275 smp_mb();
277 wake_up_bit(&jinode->i_flags, 276 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
278 } 277 }
279 278
280 /* Now refile inode to proper lists */ 279 /* Now refile inode to proper lists */
281 list_for_each_entry_safe(jinode, next_ 280 list_for_each_entry_safe(jinode, next_i,
282 &commit_trans 281 &commit_transaction->t_inode_list, i_list) {
283 list_del(&jinode->i_list); 282 list_del(&jinode->i_list);
284 if (jinode->i_next_transaction 283 if (jinode->i_next_transaction) {
285 jinode->i_transaction 284 jinode->i_transaction = jinode->i_next_transaction;
286 jinode->i_next_transac 285 jinode->i_next_transaction = NULL;
287 list_add(&jinode->i_li 286 list_add(&jinode->i_list,
288 &jinode->i_tra 287 &jinode->i_transaction->t_inode_list);
289 } else { 288 } else {
290 jinode->i_transaction 289 jinode->i_transaction = NULL;
291 jinode->i_dirty_start 290 jinode->i_dirty_start = 0;
292 jinode->i_dirty_end = 291 jinode->i_dirty_end = 0;
293 } 292 }
294 } 293 }
295 spin_unlock(&journal->j_list_lock); 294 spin_unlock(&journal->j_list_lock);
296 295
297 return ret; 296 return ret;
298 } 297 }
299 298
300 static __u32 jbd2_checksum_data(__u32 crc32_su 299 static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
301 { 300 {
>> 301 struct page *page = bh->b_page;
302 char *addr; 302 char *addr;
303 __u32 checksum; 303 __u32 checksum;
304 304
305 addr = kmap_local_folio(bh->b_folio, b !! 305 addr = kmap_atomic(page);
306 checksum = crc32_be(crc32_sum, addr, b !! 306 checksum = crc32_be(crc32_sum,
307 kunmap_local(addr); !! 307 (void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
>> 308 kunmap_atomic(addr);
308 309
309 return checksum; 310 return checksum;
310 } 311 }
311 312
312 static void write_tag_block(journal_t *j, jour 313 static void write_tag_block(journal_t *j, journal_block_tag_t *tag,
313 unsigned lo 314 unsigned long long block)
314 { 315 {
315 tag->t_blocknr = cpu_to_be32(block & ( 316 tag->t_blocknr = cpu_to_be32(block & (u32)~0);
316 if (jbd2_has_feature_64bit(j)) 317 if (jbd2_has_feature_64bit(j))
317 tag->t_blocknr_high = cpu_to_b 318 tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
318 } 319 }
319 320
320 static void jbd2_block_tag_csum_set(journal_t 321 static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
321 struct buf 322 struct buffer_head *bh, __u32 sequence)
322 { 323 {
323 journal_block_tag3_t *tag3 = (journal_ 324 journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
>> 325 struct page *page = bh->b_page;
324 __u8 *addr; 326 __u8 *addr;
325 __u32 csum32; 327 __u32 csum32;
326 __be32 seq; 328 __be32 seq;
327 329
328 if (!jbd2_journal_has_csum_v2or3(j)) 330 if (!jbd2_journal_has_csum_v2or3(j))
329 return; 331 return;
330 332
331 seq = cpu_to_be32(sequence); 333 seq = cpu_to_be32(sequence);
332 addr = kmap_local_folio(bh->b_folio, b !! 334 addr = kmap_atomic(page);
333 csum32 = jbd2_chksum(j, j->j_csum_seed 335 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
334 csum32 = jbd2_chksum(j, csum32, addr, !! 336 csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data),
335 kunmap_local(addr); !! 337 bh->b_size);
>> 338 kunmap_atomic(addr);
336 339
337 if (jbd2_has_feature_csum3(j)) 340 if (jbd2_has_feature_csum3(j))
338 tag3->t_checksum = cpu_to_be32 341 tag3->t_checksum = cpu_to_be32(csum32);
339 else 342 else
340 tag->t_checksum = cpu_to_be16( 343 tag->t_checksum = cpu_to_be16(csum32);
341 } 344 }
342 /* 345 /*
343 * jbd2_journal_commit_transaction 346 * jbd2_journal_commit_transaction
344 * 347 *
345 * The primary function for committing a trans 348 * The primary function for committing a transaction to the log. This
346 * function is called by the journal thread to 349 * function is called by the journal thread to begin a complete commit.
347 */ 350 */
348 void jbd2_journal_commit_transaction(journal_t 351 void jbd2_journal_commit_transaction(journal_t *journal)
349 { 352 {
350 struct transaction_stats_s stats; 353 struct transaction_stats_s stats;
351 transaction_t *commit_transaction; 354 transaction_t *commit_transaction;
352 struct journal_head *jh; 355 struct journal_head *jh;
353 struct buffer_head *descriptor; 356 struct buffer_head *descriptor;
354 struct buffer_head **wbuf = journal->j 357 struct buffer_head **wbuf = journal->j_wbuf;
355 int bufs; 358 int bufs;
356 int escape; !! 359 int flags;
357 int err; 360 int err;
358 unsigned long long blocknr; 361 unsigned long long blocknr;
359 ktime_t start_time; 362 ktime_t start_time;
360 u64 commit_time; 363 u64 commit_time;
361 char *tagp = NULL; 364 char *tagp = NULL;
362 journal_block_tag_t *tag = NULL; 365 journal_block_tag_t *tag = NULL;
363 int space_left = 0; 366 int space_left = 0;
364 int first_tag = 0; 367 int first_tag = 0;
365 int tag_flag; 368 int tag_flag;
366 int i; 369 int i;
367 int tag_bytes = journal_tag_bytes(jour 370 int tag_bytes = journal_tag_bytes(journal);
368 struct buffer_head *cbh = NULL; /* For 371 struct buffer_head *cbh = NULL; /* For transactional checksums */
369 __u32 crc32_sum = ~0; 372 __u32 crc32_sum = ~0;
370 struct blk_plug plug; 373 struct blk_plug plug;
371 /* Tail of the journal */ 374 /* Tail of the journal */
372 unsigned long first_block; 375 unsigned long first_block;
373 tid_t first_tid; 376 tid_t first_tid;
374 int update_tail; 377 int update_tail;
375 int csum_size = 0; 378 int csum_size = 0;
376 LIST_HEAD(io_bufs); 379 LIST_HEAD(io_bufs);
377 LIST_HEAD(log_bufs); 380 LIST_HEAD(log_bufs);
378 381
379 if (jbd2_journal_has_csum_v2or3(journa 382 if (jbd2_journal_has_csum_v2or3(journal))
380 csum_size = sizeof(struct jbd2 383 csum_size = sizeof(struct jbd2_journal_block_tail);
381 384
382 /* 385 /*
383 * First job: lock down the current tr 386 * First job: lock down the current transaction and wait for
384 * all outstanding updates to complete 387 * all outstanding updates to complete.
385 */ 388 */
386 389
387 /* Do we need to erase the effects of 390 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
388 if (journal->j_flags & JBD2_FLUSHED) { 391 if (journal->j_flags & JBD2_FLUSHED) {
389 jbd2_debug(3, "super block upd 392 jbd2_debug(3, "super block updated\n");
390 mutex_lock_io(&journal->j_chec 393 mutex_lock_io(&journal->j_checkpoint_mutex);
391 /* 394 /*
392 * We hold j_checkpoint_mutex 395 * We hold j_checkpoint_mutex so tail cannot change under us.
393 * We don't need any special d 396 * We don't need any special data guarantees for writing sb
394 * since journal is empty and 397 * since journal is empty and it is ok for write to be
395 * flushed only with transacti 398 * flushed only with transaction commit.
396 */ 399 */
397 jbd2_journal_update_sb_log_tai 400 jbd2_journal_update_sb_log_tail(journal,
398 401 journal->j_tail_sequence,
399 !! 402 journal->j_tail,
>> 403 REQ_SYNC);
400 mutex_unlock(&journal->j_check 404 mutex_unlock(&journal->j_checkpoint_mutex);
401 } else { 405 } else {
402 jbd2_debug(3, "superblock not 406 jbd2_debug(3, "superblock not updated\n");
403 } 407 }
404 408
405 J_ASSERT(journal->j_running_transactio 409 J_ASSERT(journal->j_running_transaction != NULL);
406 J_ASSERT(journal->j_committing_transac 410 J_ASSERT(journal->j_committing_transaction == NULL);
407 411
408 write_lock(&journal->j_state_lock); 412 write_lock(&journal->j_state_lock);
409 journal->j_flags |= JBD2_FULL_COMMIT_O 413 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
410 while (journal->j_flags & JBD2_FAST_CO 414 while (journal->j_flags & JBD2_FAST_COMMIT_ONGOING) {
411 DEFINE_WAIT(wait); 415 DEFINE_WAIT(wait);
412 416
413 prepare_to_wait(&journal->j_fc 417 prepare_to_wait(&journal->j_fc_wait, &wait,
414 TASK_UNINTERRU 418 TASK_UNINTERRUPTIBLE);
415 write_unlock(&journal->j_state 419 write_unlock(&journal->j_state_lock);
416 schedule(); 420 schedule();
417 write_lock(&journal->j_state_l 421 write_lock(&journal->j_state_lock);
418 finish_wait(&journal->j_fc_wai 422 finish_wait(&journal->j_fc_wait, &wait);
419 /* 423 /*
420 * TODO: by blocking fast comm 424 * TODO: by blocking fast commits here, we are increasing
421 * fsync() latency slightly. S 425 * fsync() latency slightly. Strictly speaking, we don't need
422 * to block fast commits until 426 * to block fast commits until the transaction enters T_FLUSH
423 * state. So an optimization i 427 * state. So an optimization is possible where we block new fast
424 * commits here and wait for e 428 * commits here and wait for existing ones to complete
425 * just before we enter T_FLUS 429 * just before we enter T_FLUSH. That way, the existing fast
426 * commits and this full commi 430 * commits and this full commit can proceed parallely.
427 */ 431 */
428 } 432 }
429 write_unlock(&journal->j_state_lock); 433 write_unlock(&journal->j_state_lock);
430 434
431 commit_transaction = journal->j_runnin 435 commit_transaction = journal->j_running_transaction;
432 436
433 trace_jbd2_start_commit(journal, commi 437 trace_jbd2_start_commit(journal, commit_transaction);
434 jbd2_debug(1, "JBD2: starting commit o 438 jbd2_debug(1, "JBD2: starting commit of transaction %d\n",
435 commit_transaction->t_ 439 commit_transaction->t_tid);
436 440
437 write_lock(&journal->j_state_lock); 441 write_lock(&journal->j_state_lock);
438 journal->j_fc_off = 0; 442 journal->j_fc_off = 0;
439 J_ASSERT(commit_transaction->t_state = 443 J_ASSERT(commit_transaction->t_state == T_RUNNING);
440 commit_transaction->t_state = T_LOCKED 444 commit_transaction->t_state = T_LOCKED;
441 445
442 trace_jbd2_commit_locking(journal, com 446 trace_jbd2_commit_locking(journal, commit_transaction);
443 stats.run.rs_wait = commit_transaction 447 stats.run.rs_wait = commit_transaction->t_max_wait;
444 stats.run.rs_request_delay = 0; 448 stats.run.rs_request_delay = 0;
445 stats.run.rs_locked = jiffies; 449 stats.run.rs_locked = jiffies;
446 if (commit_transaction->t_requested) 450 if (commit_transaction->t_requested)
447 stats.run.rs_request_delay = 451 stats.run.rs_request_delay =
448 jbd2_time_diff(commit_ 452 jbd2_time_diff(commit_transaction->t_requested,
449 stats.r 453 stats.run.rs_locked);
450 stats.run.rs_running = jbd2_time_diff( 454 stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
451 455 stats.run.rs_locked);
452 456
453 // waits for any t_updates to finish 457 // waits for any t_updates to finish
454 jbd2_journal_wait_updates(journal); 458 jbd2_journal_wait_updates(journal);
455 459
456 commit_transaction->t_state = T_SWITCH 460 commit_transaction->t_state = T_SWITCH;
457 461
458 J_ASSERT (atomic_read(&commit_transact 462 J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
459 journal->j_max_transac 463 journal->j_max_transaction_buffers);
460 464
461 /* 465 /*
462 * First thing we are allowed to do is 466 * First thing we are allowed to do is to discard any remaining
463 * BJ_Reserved buffers. Note, it is _ 467 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
464 * that there are no such buffers: if 468 * that there are no such buffers: if a large filesystem
465 * operation like a truncate needs to 469 * operation like a truncate needs to split itself over multiple
466 * transactions, then it may try to do 470 * transactions, then it may try to do a jbd2_journal_restart() while
467 * there are still BJ_Reserved buffers 471 * there are still BJ_Reserved buffers outstanding. These must
468 * be released cleanly from the curren 472 * be released cleanly from the current transaction.
469 * 473 *
470 * In this case, the filesystem must s 474 * In this case, the filesystem must still reserve write access
471 * again before modifying the buffer i 475 * again before modifying the buffer in the new transaction, but
472 * we do not require it to remember ex 476 * we do not require it to remember exactly which old buffers it
473 * has reserved. This is consistent w 477 * has reserved. This is consistent with the existing behaviour
474 * that multiple jbd2_journal_get_writ 478 * that multiple jbd2_journal_get_write_access() calls to the same
475 * buffer are perfectly permissible. 479 * buffer are perfectly permissible.
476 * We use journal->j_state_lock here t 480 * We use journal->j_state_lock here to serialize processing of
477 * t_reserved_list with eviction of bu 481 * t_reserved_list with eviction of buffers from journal_unmap_buffer().
478 */ 482 */
479 while (commit_transaction->t_reserved_ 483 while (commit_transaction->t_reserved_list) {
480 jh = commit_transaction->t_res 484 jh = commit_transaction->t_reserved_list;
481 JBUFFER_TRACE(jh, "reserved, u 485 JBUFFER_TRACE(jh, "reserved, unused: refile");
482 /* 486 /*
483 * A jbd2_journal_get_undo_acc 487 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
484 * leave undo-committed data. 488 * leave undo-committed data.
485 */ 489 */
486 if (jh->b_committed_data) { 490 if (jh->b_committed_data) {
487 struct buffer_head *bh 491 struct buffer_head *bh = jh2bh(jh);
488 492
489 spin_lock(&jh->b_state 493 spin_lock(&jh->b_state_lock);
490 jbd2_free(jh->b_commit 494 jbd2_free(jh->b_committed_data, bh->b_size);
491 jh->b_committed_data = 495 jh->b_committed_data = NULL;
492 spin_unlock(&jh->b_sta 496 spin_unlock(&jh->b_state_lock);
493 } 497 }
494 jbd2_journal_refile_buffer(jou 498 jbd2_journal_refile_buffer(journal, jh);
495 } 499 }
496 500
497 write_unlock(&journal->j_state_lock); 501 write_unlock(&journal->j_state_lock);
498 /* 502 /*
499 * Now try to drop any written-back bu 503 * Now try to drop any written-back buffers from the journal's
500 * checkpoint lists. We do this *befo 504 * checkpoint lists. We do this *before* commit because it potentially
501 * frees some memory 505 * frees some memory
502 */ 506 */
503 spin_lock(&journal->j_list_lock); 507 spin_lock(&journal->j_list_lock);
504 __jbd2_journal_clean_checkpoint_list(j !! 508 __jbd2_journal_clean_checkpoint_list(journal, false);
505 spin_unlock(&journal->j_list_lock); 509 spin_unlock(&journal->j_list_lock);
506 510
507 jbd2_debug(3, "JBD2: commit phase 1\n" 511 jbd2_debug(3, "JBD2: commit phase 1\n");
508 512
509 /* 513 /*
510 * Clear revoked flag to reflect there 514 * Clear revoked flag to reflect there is no revoked buffers
511 * in the next transaction which is go 515 * in the next transaction which is going to be started.
512 */ 516 */
513 jbd2_clear_buffer_revoked_flags(journa 517 jbd2_clear_buffer_revoked_flags(journal);
514 518
515 /* 519 /*
516 * Switch to a new revoke table. 520 * Switch to a new revoke table.
517 */ 521 */
518 jbd2_journal_switch_revoke_table(journ 522 jbd2_journal_switch_revoke_table(journal);
519 523
520 write_lock(&journal->j_state_lock); 524 write_lock(&journal->j_state_lock);
521 /* 525 /*
522 * Reserved credits cannot be claimed 526 * Reserved credits cannot be claimed anymore, free them
523 */ 527 */
524 atomic_sub(atomic_read(&journal->j_res 528 atomic_sub(atomic_read(&journal->j_reserved_credits),
525 &commit_transaction->t_outs 529 &commit_transaction->t_outstanding_credits);
526 530
527 trace_jbd2_commit_flushing(journal, co 531 trace_jbd2_commit_flushing(journal, commit_transaction);
528 stats.run.rs_flushing = jiffies; 532 stats.run.rs_flushing = jiffies;
529 stats.run.rs_locked = jbd2_time_diff(s 533 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
530 s 534 stats.run.rs_flushing);
531 535
532 commit_transaction->t_state = T_FLUSH; 536 commit_transaction->t_state = T_FLUSH;
533 journal->j_committing_transaction = co 537 journal->j_committing_transaction = commit_transaction;
534 journal->j_running_transaction = NULL; 538 journal->j_running_transaction = NULL;
535 start_time = ktime_get(); 539 start_time = ktime_get();
536 commit_transaction->t_log_start = jour 540 commit_transaction->t_log_start = journal->j_head;
537 wake_up_all(&journal->j_wait_transacti 541 wake_up_all(&journal->j_wait_transaction_locked);
538 write_unlock(&journal->j_state_lock); 542 write_unlock(&journal->j_state_lock);
539 543
540 jbd2_debug(3, "JBD2: commit phase 2a\n 544 jbd2_debug(3, "JBD2: commit phase 2a\n");
541 545
542 /* 546 /*
543 * Now start flushing things to disk, 547 * Now start flushing things to disk, in the order they appear
544 * on the transaction lists. Data blo 548 * on the transaction lists. Data blocks go first.
545 */ 549 */
546 err = journal_submit_data_buffers(jour 550 err = journal_submit_data_buffers(journal, commit_transaction);
547 if (err) 551 if (err)
548 jbd2_journal_abort(journal, er 552 jbd2_journal_abort(journal, err);
549 553
550 blk_start_plug(&plug); 554 blk_start_plug(&plug);
551 jbd2_journal_write_revoke_records(comm 555 jbd2_journal_write_revoke_records(commit_transaction, &log_bufs);
552 556
553 jbd2_debug(3, "JBD2: commit phase 2b\n 557 jbd2_debug(3, "JBD2: commit phase 2b\n");
554 558
555 /* 559 /*
556 * Way to go: we have now written out 560 * Way to go: we have now written out all of the data for a
557 * transaction! Now comes the tricky 561 * transaction! Now comes the tricky part: we need to write out
558 * metadata. Loop over the transactio 562 * metadata. Loop over the transaction's entire buffer list:
559 */ 563 */
560 write_lock(&journal->j_state_lock); 564 write_lock(&journal->j_state_lock);
561 commit_transaction->t_state = T_COMMIT 565 commit_transaction->t_state = T_COMMIT;
562 write_unlock(&journal->j_state_lock); 566 write_unlock(&journal->j_state_lock);
563 567
564 trace_jbd2_commit_logging(journal, com 568 trace_jbd2_commit_logging(journal, commit_transaction);
565 stats.run.rs_logging = jiffies; 569 stats.run.rs_logging = jiffies;
566 stats.run.rs_flushing = jbd2_time_diff 570 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
567 571 stats.run.rs_logging);
568 stats.run.rs_blocks = commit_transacti 572 stats.run.rs_blocks = commit_transaction->t_nr_buffers;
569 stats.run.rs_blocks_logged = 0; 573 stats.run.rs_blocks_logged = 0;
570 574
571 J_ASSERT(commit_transaction->t_nr_buff 575 J_ASSERT(commit_transaction->t_nr_buffers <=
572 atomic_read(&commit_transacti 576 atomic_read(&commit_transaction->t_outstanding_credits));
573 577
>> 578 err = 0;
574 bufs = 0; 579 bufs = 0;
575 descriptor = NULL; 580 descriptor = NULL;
576 while (commit_transaction->t_buffers) 581 while (commit_transaction->t_buffers) {
577 582
578 /* Find the next buffer to be 583 /* Find the next buffer to be journaled... */
579 584
580 jh = commit_transaction->t_buf 585 jh = commit_transaction->t_buffers;
581 586
582 /* If we're in abort mode, we 587 /* If we're in abort mode, we just un-journal the buffer and
583 release it. */ 588 release it. */
584 589
585 if (is_journal_aborted(journal 590 if (is_journal_aborted(journal)) {
586 clear_buffer_jbddirty( 591 clear_buffer_jbddirty(jh2bh(jh));
587 JBUFFER_TRACE(jh, "jou 592 JBUFFER_TRACE(jh, "journal is aborting: refile");
588 jbd2_buffer_abort_trig 593 jbd2_buffer_abort_trigger(jh,
589 594 jh->b_frozen_data ?
590 595 jh->b_frozen_triggers :
591 596 jh->b_triggers);
592 jbd2_journal_refile_bu 597 jbd2_journal_refile_buffer(journal, jh);
593 /* If that was the las 598 /* If that was the last one, we need to clean up
594 * any descriptor buff 599 * any descriptor buffers which may have been
595 * already allocated, 600 * already allocated, even if we are now
596 * aborting. */ 601 * aborting. */
597 if (!commit_transactio 602 if (!commit_transaction->t_buffers)
598 goto start_jou 603 goto start_journal_io;
599 continue; 604 continue;
600 } 605 }
601 606
602 /* Make sure we have a descrip 607 /* Make sure we have a descriptor block in which to
603 record the metadata buffer. 608 record the metadata buffer. */
604 609
605 if (!descriptor) { 610 if (!descriptor) {
606 J_ASSERT (bufs == 0); 611 J_ASSERT (bufs == 0);
607 612
608 jbd2_debug(4, "JBD2: g 613 jbd2_debug(4, "JBD2: get descriptor\n");
609 614
610 descriptor = jbd2_jour 615 descriptor = jbd2_journal_get_descriptor_buffer(
611 616 commit_transaction,
612 617 JBD2_DESCRIPTOR_BLOCK);
613 if (!descriptor) { 618 if (!descriptor) {
614 jbd2_journal_a 619 jbd2_journal_abort(journal, -EIO);
615 continue; 620 continue;
616 } 621 }
617 622
618 jbd2_debug(4, "JBD2: g 623 jbd2_debug(4, "JBD2: got buffer %llu (%p)\n",
619 (unsigned long 624 (unsigned long long)descriptor->b_blocknr,
620 descriptor->b_ 625 descriptor->b_data);
621 tagp = &descriptor->b_ 626 tagp = &descriptor->b_data[sizeof(journal_header_t)];
622 space_left = descripto 627 space_left = descriptor->b_size -
623 628 sizeof(journal_header_t);
624 first_tag = 1; 629 first_tag = 1;
625 set_buffer_jwrite(desc 630 set_buffer_jwrite(descriptor);
626 set_buffer_dirty(descr 631 set_buffer_dirty(descriptor);
627 wbuf[bufs++] = descrip 632 wbuf[bufs++] = descriptor;
628 633
629 /* Record it so that w 634 /* Record it so that we can wait for IO
630 completion later */ 635 completion later */
631 BUFFER_TRACE(descripto 636 BUFFER_TRACE(descriptor, "ph3: file as descriptor");
632 jbd2_file_log_bh(&log_ 637 jbd2_file_log_bh(&log_bufs, descriptor);
633 } 638 }
634 639
635 /* Where is the buffer to be w 640 /* Where is the buffer to be written? */
636 641
637 err = jbd2_journal_next_log_bl 642 err = jbd2_journal_next_log_block(journal, &blocknr);
638 /* If the block mapping failed 643 /* If the block mapping failed, just abandon the buffer
639 and repeat this loop: we'll 644 and repeat this loop: we'll fall into the
640 refile-on-abort condition a 645 refile-on-abort condition above. */
641 if (err) { 646 if (err) {
642 jbd2_journal_abort(jou 647 jbd2_journal_abort(journal, err);
643 continue; 648 continue;
644 } 649 }
645 650
646 /* 651 /*
647 * start_this_handle() uses t_ 652 * start_this_handle() uses t_outstanding_credits to determine
648 * the free space in the log. 653 * the free space in the log.
649 */ 654 */
650 atomic_dec(&commit_transaction 655 atomic_dec(&commit_transaction->t_outstanding_credits);
651 656
652 /* Bump b_count to prevent tru 657 /* Bump b_count to prevent truncate from stumbling over
653 the shadowed buffer! @@@ T 658 the shadowed buffer! @@@ This can go if we ever get
654 rid of the shadow pairing o 659 rid of the shadow pairing of buffers. */
655 atomic_inc(&jh2bh(jh)->b_count 660 atomic_inc(&jh2bh(jh)->b_count);
656 661
657 /* 662 /*
658 * Make a temporary IO buffer 663 * Make a temporary IO buffer with which to write it out
659 * (this will requeue the meta 664 * (this will requeue the metadata buffer to BJ_Shadow).
660 */ 665 */
661 set_bit(BH_JWrite, &jh2bh(jh)- 666 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
662 JBUFFER_TRACE(jh, "ph3: write 667 JBUFFER_TRACE(jh, "ph3: write metadata");
663 escape = jbd2_journal_write_me !! 668 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
664 669 jh, &wbuf[bufs], blocknr);
665 if (escape < 0) { !! 670 if (flags < 0) {
666 jbd2_journal_abort(jou !! 671 jbd2_journal_abort(journal, flags);
667 continue; 672 continue;
668 } 673 }
669 jbd2_file_log_bh(&io_bufs, wbu 674 jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
670 675
671 /* Record the new block's tag 676 /* Record the new block's tag in the current descriptor
672 buffer */ 677 buffer */
673 678
674 tag_flag = 0; 679 tag_flag = 0;
675 if (escape) !! 680 if (flags & 1)
676 tag_flag |= JBD2_FLAG_ 681 tag_flag |= JBD2_FLAG_ESCAPE;
677 if (!first_tag) 682 if (!first_tag)
678 tag_flag |= JBD2_FLAG_ 683 tag_flag |= JBD2_FLAG_SAME_UUID;
679 684
680 tag = (journal_block_tag_t *) 685 tag = (journal_block_tag_t *) tagp;
681 write_tag_block(journal, tag, 686 write_tag_block(journal, tag, jh2bh(jh)->b_blocknr);
682 tag->t_flags = cpu_to_be16(tag 687 tag->t_flags = cpu_to_be16(tag_flag);
683 jbd2_block_tag_csum_set(journa 688 jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
684 commit 689 commit_transaction->t_tid);
685 tagp += tag_bytes; 690 tagp += tag_bytes;
686 space_left -= tag_bytes; 691 space_left -= tag_bytes;
687 bufs++; 692 bufs++;
688 693
689 if (first_tag) { 694 if (first_tag) {
690 memcpy (tagp, journal- 695 memcpy (tagp, journal->j_uuid, 16);
691 tagp += 16; 696 tagp += 16;
692 space_left -= 16; 697 space_left -= 16;
693 first_tag = 0; 698 first_tag = 0;
694 } 699 }
695 700
696 /* If there's no more to do, o 701 /* If there's no more to do, or if the descriptor is full,
697 let the IO rip! */ 702 let the IO rip! */
698 703
699 if (bufs == journal->j_wbufsiz 704 if (bufs == journal->j_wbufsize ||
700 commit_transaction->t_buff 705 commit_transaction->t_buffers == NULL ||
701 space_left < tag_bytes + 1 706 space_left < tag_bytes + 16 + csum_size) {
702 707
703 jbd2_debug(4, "JBD2: S 708 jbd2_debug(4, "JBD2: Submit %d IOs\n", bufs);
704 709
705 /* Write an end-of-des 710 /* Write an end-of-descriptor marker before
706 submitting the IOs. 711 submitting the IOs. "tag" still points to
707 the last tag we set 712 the last tag we set up. */
708 713
709 tag->t_flags |= cpu_to 714 tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
710 start_journal_io: 715 start_journal_io:
711 if (descriptor) 716 if (descriptor)
712 jbd2_descripto 717 jbd2_descriptor_block_csum_set(journal,
713 718 descriptor);
714 719
715 for (i = 0; i < bufs; 720 for (i = 0; i < bufs; i++) {
716 struct buffer_ 721 struct buffer_head *bh = wbuf[i];
717 <<
718 /* 722 /*
719 * Compute che 723 * Compute checksum.
720 */ 724 */
721 if (jbd2_has_f 725 if (jbd2_has_feature_checksum(journal)) {
722 crc32_ 726 crc32_sum =
723 jb 727 jbd2_checksum_data(crc32_sum, bh);
724 } 728 }
725 729
726 lock_buffer(bh 730 lock_buffer(bh);
727 clear_buffer_d 731 clear_buffer_dirty(bh);
728 set_buffer_upt 732 set_buffer_uptodate(bh);
729 bh->b_end_io = 733 bh->b_end_io = journal_end_buffer_io_sync;
730 submit_bh(REQ_ !! 734 submit_bh(REQ_OP_WRITE | REQ_SYNC, bh);
731 bh); <<
732 } 735 }
733 cond_resched(); 736 cond_resched();
734 737
735 /* Force a new descrip 738 /* Force a new descriptor to be generated next
736 time round the loop 739 time round the loop. */
737 descriptor = NULL; 740 descriptor = NULL;
738 bufs = 0; 741 bufs = 0;
739 } 742 }
740 } 743 }
741 744
742 err = journal_finish_inode_data_buffer 745 err = journal_finish_inode_data_buffers(journal, commit_transaction);
743 if (err) { 746 if (err) {
744 printk(KERN_WARNING 747 printk(KERN_WARNING
745 "JBD2: Detected IO err 748 "JBD2: Detected IO errors while flushing file data "
746 "on %s\n", journal->j_d 749 "on %s\n", journal->j_devname);
747 if (journal->j_flags & JBD2_AB 750 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
748 jbd2_journal_abort(jou 751 jbd2_journal_abort(journal, err);
749 err = 0; 752 err = 0;
750 } 753 }
751 754
752 /* 755 /*
753 * Get current oldest transaction in t 756 * Get current oldest transaction in the log before we issue flush
754 * to the filesystem device. After the 757 * to the filesystem device. After the flush we can be sure that
755 * blocks of all older transactions ar 758 * blocks of all older transactions are checkpointed to persistent
756 * storage and we will be safe to upda 759 * storage and we will be safe to update journal start in the
757 * superblock with the numbers we get 760 * superblock with the numbers we get here.
758 */ 761 */
759 update_tail = 762 update_tail =
760 jbd2_journal_get_log_tail(jour 763 jbd2_journal_get_log_tail(journal, &first_tid, &first_block);
761 764
762 write_lock(&journal->j_state_lock); 765 write_lock(&journal->j_state_lock);
763 if (update_tail) { 766 if (update_tail) {
764 long freed = first_block - jou 767 long freed = first_block - journal->j_tail;
765 768
766 if (first_block < journal->j_t 769 if (first_block < journal->j_tail)
767 freed += journal->j_la 770 freed += journal->j_last - journal->j_first;
768 /* Update tail only if we free 771 /* Update tail only if we free significant amount of space */
769 if (freed < journal->j_max_tra !! 772 if (freed < jbd2_journal_get_max_txn_bufs(journal))
770 update_tail = 0; 773 update_tail = 0;
771 } 774 }
772 J_ASSERT(commit_transaction->t_state = 775 J_ASSERT(commit_transaction->t_state == T_COMMIT);
773 commit_transaction->t_state = T_COMMIT 776 commit_transaction->t_state = T_COMMIT_DFLUSH;
774 write_unlock(&journal->j_state_lock); 777 write_unlock(&journal->j_state_lock);
775 778
776 /* 779 /*
777 * If the journal is not located on th 780 * If the journal is not located on the file system device,
778 * then we must flush the file system 781 * then we must flush the file system device before we issue
779 * the commit record 782 * the commit record
780 */ 783 */
781 if (commit_transaction->t_need_data_fl 784 if (commit_transaction->t_need_data_flush &&
782 (journal->j_fs_dev != journal->j_d 785 (journal->j_fs_dev != journal->j_dev) &&
783 (journal->j_flags & JBD2_BARRIER)) 786 (journal->j_flags & JBD2_BARRIER))
784 blkdev_issue_flush(journal->j_ 787 blkdev_issue_flush(journal->j_fs_dev);
785 788
786 /* Done it all: now write the commit r 789 /* Done it all: now write the commit record asynchronously. */
787 if (jbd2_has_feature_async_commit(jour 790 if (jbd2_has_feature_async_commit(journal)) {
788 err = journal_submit_commit_re 791 err = journal_submit_commit_record(journal, commit_transaction,
789 792 &cbh, crc32_sum);
790 if (err) 793 if (err)
791 jbd2_journal_abort(jou 794 jbd2_journal_abort(journal, err);
792 } 795 }
793 796
794 blk_finish_plug(&plug); 797 blk_finish_plug(&plug);
795 798
796 /* Lo and behold: we have just managed 799 /* Lo and behold: we have just managed to send a transaction to
797 the log. Before we can commit it, 800 the log. Before we can commit it, wait for the IO so far to
798 complete. Control buffers being wr 801 complete. Control buffers being written are on the
799 transaction's t_log_list queue, and 802 transaction's t_log_list queue, and metadata buffers are on
800 the io_bufs list. 803 the io_bufs list.
801 804
802 Wait for the buffers in reverse ord 805 Wait for the buffers in reverse order. That way we are
803 less likely to be woken up until al 806 less likely to be woken up until all IOs have completed, and
804 so we incur less scheduling load. 807 so we incur less scheduling load.
805 */ 808 */
806 809
807 jbd2_debug(3, "JBD2: commit phase 3\n" 810 jbd2_debug(3, "JBD2: commit phase 3\n");
808 811
809 while (!list_empty(&io_bufs)) { 812 while (!list_empty(&io_bufs)) {
810 struct buffer_head *bh = list_ 813 struct buffer_head *bh = list_entry(io_bufs.prev,
811 814 struct buffer_head,
812 815 b_assoc_buffers);
813 816
814 wait_on_buffer(bh); 817 wait_on_buffer(bh);
815 cond_resched(); 818 cond_resched();
816 819
817 if (unlikely(!buffer_uptodate( 820 if (unlikely(!buffer_uptodate(bh)))
818 err = -EIO; 821 err = -EIO;
819 jbd2_unfile_log_bh(bh); 822 jbd2_unfile_log_bh(bh);
820 stats.run.rs_blocks_logged++; 823 stats.run.rs_blocks_logged++;
821 824
822 /* 825 /*
823 * The list contains temporary 826 * The list contains temporary buffer heads created by
824 * jbd2_journal_write_metadata 827 * jbd2_journal_write_metadata_buffer().
825 */ 828 */
826 BUFFER_TRACE(bh, "dumping temp 829 BUFFER_TRACE(bh, "dumping temporary bh");
827 __brelse(bh); 830 __brelse(bh);
828 J_ASSERT_BH(bh, atomic_read(&b 831 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
829 free_buffer_head(bh); 832 free_buffer_head(bh);
830 833
831 /* We also have to refile the 834 /* We also have to refile the corresponding shadowed buffer */
832 jh = commit_transaction->t_sha 835 jh = commit_transaction->t_shadow_list->b_tprev;
833 bh = jh2bh(jh); 836 bh = jh2bh(jh);
834 clear_buffer_jwrite(bh); 837 clear_buffer_jwrite(bh);
835 J_ASSERT_BH(bh, buffer_jbddirt 838 J_ASSERT_BH(bh, buffer_jbddirty(bh));
836 J_ASSERT_BH(bh, !buffer_shadow 839 J_ASSERT_BH(bh, !buffer_shadow(bh));
837 840
838 /* The metadata is now release 841 /* The metadata is now released for reuse, but we need
839 to remember it against this 842 to remember it against this transaction so that when
840 we finally commit, we can d 843 we finally commit, we can do any checkpointing
841 required. */ 844 required. */
842 JBUFFER_TRACE(jh, "file as BJ_ 845 JBUFFER_TRACE(jh, "file as BJ_Forget");
843 jbd2_journal_file_buffer(jh, c 846 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
844 JBUFFER_TRACE(jh, "brelse shad 847 JBUFFER_TRACE(jh, "brelse shadowed buffer");
845 __brelse(bh); 848 __brelse(bh);
846 } 849 }
847 850
848 J_ASSERT (commit_transaction->t_shadow 851 J_ASSERT (commit_transaction->t_shadow_list == NULL);
849 852
850 jbd2_debug(3, "JBD2: commit phase 4\n" 853 jbd2_debug(3, "JBD2: commit phase 4\n");
851 854
852 /* Here we wait for the revoke record 855 /* Here we wait for the revoke record and descriptor record buffers */
853 while (!list_empty(&log_bufs)) { 856 while (!list_empty(&log_bufs)) {
854 struct buffer_head *bh; 857 struct buffer_head *bh;
855 858
856 bh = list_entry(log_bufs.prev, 859 bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
857 wait_on_buffer(bh); 860 wait_on_buffer(bh);
858 cond_resched(); 861 cond_resched();
859 862
860 if (unlikely(!buffer_uptodate( 863 if (unlikely(!buffer_uptodate(bh)))
861 err = -EIO; 864 err = -EIO;
862 865
863 BUFFER_TRACE(bh, "ph5: control 866 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
864 clear_buffer_jwrite(bh); 867 clear_buffer_jwrite(bh);
865 jbd2_unfile_log_bh(bh); 868 jbd2_unfile_log_bh(bh);
866 stats.run.rs_blocks_logged++; 869 stats.run.rs_blocks_logged++;
867 __brelse(bh); /* One 870 __brelse(bh); /* One for getblk */
868 /* AKPM: bforget here */ 871 /* AKPM: bforget here */
869 } 872 }
870 873
871 if (err) 874 if (err)
872 jbd2_journal_abort(journal, er 875 jbd2_journal_abort(journal, err);
873 876
874 jbd2_debug(3, "JBD2: commit phase 5\n" 877 jbd2_debug(3, "JBD2: commit phase 5\n");
875 write_lock(&journal->j_state_lock); 878 write_lock(&journal->j_state_lock);
876 J_ASSERT(commit_transaction->t_state = 879 J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
877 commit_transaction->t_state = T_COMMIT 880 commit_transaction->t_state = T_COMMIT_JFLUSH;
878 write_unlock(&journal->j_state_lock); 881 write_unlock(&journal->j_state_lock);
879 882
880 if (!jbd2_has_feature_async_commit(jou 883 if (!jbd2_has_feature_async_commit(journal)) {
881 err = journal_submit_commit_re 884 err = journal_submit_commit_record(journal, commit_transaction,
882 885 &cbh, crc32_sum);
883 if (err) 886 if (err)
884 jbd2_journal_abort(jou 887 jbd2_journal_abort(journal, err);
885 } 888 }
886 if (cbh) 889 if (cbh)
887 err = journal_wait_on_commit_r 890 err = journal_wait_on_commit_record(journal, cbh);
888 stats.run.rs_blocks_logged++; 891 stats.run.rs_blocks_logged++;
889 if (jbd2_has_feature_async_commit(jour 892 if (jbd2_has_feature_async_commit(journal) &&
890 journal->j_flags & JBD2_BARRIER) { 893 journal->j_flags & JBD2_BARRIER) {
891 blkdev_issue_flush(journal->j_ 894 blkdev_issue_flush(journal->j_dev);
892 } 895 }
893 896
894 if (err) 897 if (err)
895 jbd2_journal_abort(journal, er 898 jbd2_journal_abort(journal, err);
896 899
897 WARN_ON_ONCE( 900 WARN_ON_ONCE(
898 atomic_read(&commit_transactio 901 atomic_read(&commit_transaction->t_outstanding_credits) < 0);
899 902
900 /* 903 /*
901 * Now disk caches for filesystem devi 904 * Now disk caches for filesystem device are flushed so we are safe to
902 * erase checkpointed transactions fro 905 * erase checkpointed transactions from the log by updating journal
903 * superblock. 906 * superblock.
904 */ 907 */
905 if (update_tail) 908 if (update_tail)
906 jbd2_update_log_tail(journal, 909 jbd2_update_log_tail(journal, first_tid, first_block);
907 910
908 /* End of a transaction! Finally, we 911 /* End of a transaction! Finally, we can do checkpoint
909 processing: any buffers committed a 912 processing: any buffers committed as a result of this
910 transaction can be removed from any 913 transaction can be removed from any checkpoint list it was on
911 before. */ 914 before. */
912 915
913 jbd2_debug(3, "JBD2: commit phase 6\n" 916 jbd2_debug(3, "JBD2: commit phase 6\n");
914 917
915 J_ASSERT(list_empty(&commit_transactio 918 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
916 J_ASSERT(commit_transaction->t_buffers 919 J_ASSERT(commit_transaction->t_buffers == NULL);
917 J_ASSERT(commit_transaction->t_checkpo 920 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
918 J_ASSERT(commit_transaction->t_shadow_ 921 J_ASSERT(commit_transaction->t_shadow_list == NULL);
919 922
920 restart_loop: 923 restart_loop:
921 /* 924 /*
922 * As there are other places (journal_ 925 * As there are other places (journal_unmap_buffer()) adding buffers
923 * to this list we have to be careful 926 * to this list we have to be careful and hold the j_list_lock.
924 */ 927 */
925 spin_lock(&journal->j_list_lock); 928 spin_lock(&journal->j_list_lock);
926 while (commit_transaction->t_forget) { 929 while (commit_transaction->t_forget) {
927 transaction_t *cp_transaction; 930 transaction_t *cp_transaction;
928 struct buffer_head *bh; 931 struct buffer_head *bh;
929 int try_to_free = 0; 932 int try_to_free = 0;
930 bool drop_ref; 933 bool drop_ref;
931 934
932 jh = commit_transaction->t_for 935 jh = commit_transaction->t_forget;
933 spin_unlock(&journal->j_list_l 936 spin_unlock(&journal->j_list_lock);
934 bh = jh2bh(jh); 937 bh = jh2bh(jh);
935 /* 938 /*
936 * Get a reference so that bh 939 * Get a reference so that bh cannot be freed before we are
937 * done with it. 940 * done with it.
938 */ 941 */
939 get_bh(bh); 942 get_bh(bh);
940 spin_lock(&jh->b_state_lock); 943 spin_lock(&jh->b_state_lock);
941 J_ASSERT_JH(jh, jh->b_transact 944 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
942 945
943 /* 946 /*
944 * If there is undo-protected 947 * If there is undo-protected committed data against
945 * this buffer, then we can re 948 * this buffer, then we can remove it now. If it is a
946 * buffer needing such protect 949 * buffer needing such protection, the old frozen_data
947 * field now points to a commi 950 * field now points to a committed version of the
948 * buffer, so rotate that fiel 951 * buffer, so rotate that field to the new committed
949 * data. 952 * data.
950 * 953 *
951 * Otherwise, we can just thro 954 * Otherwise, we can just throw away the frozen data now.
952 * 955 *
953 * We also know that the froze 956 * We also know that the frozen data has already fired
954 * its triggers if they exist, 957 * its triggers if they exist, so we can clear that too.
955 */ 958 */
956 if (jh->b_committed_data) { 959 if (jh->b_committed_data) {
957 jbd2_free(jh->b_commit 960 jbd2_free(jh->b_committed_data, bh->b_size);
958 jh->b_committed_data = 961 jh->b_committed_data = NULL;
959 if (jh->b_frozen_data) 962 if (jh->b_frozen_data) {
960 jh->b_committe 963 jh->b_committed_data = jh->b_frozen_data;
961 jh->b_frozen_d 964 jh->b_frozen_data = NULL;
962 jh->b_frozen_t 965 jh->b_frozen_triggers = NULL;
963 } 966 }
964 } else if (jh->b_frozen_data) 967 } else if (jh->b_frozen_data) {
965 jbd2_free(jh->b_frozen 968 jbd2_free(jh->b_frozen_data, bh->b_size);
966 jh->b_frozen_data = NU 969 jh->b_frozen_data = NULL;
967 jh->b_frozen_triggers 970 jh->b_frozen_triggers = NULL;
968 } 971 }
969 972
970 spin_lock(&journal->j_list_loc 973 spin_lock(&journal->j_list_lock);
971 cp_transaction = jh->b_cp_tran 974 cp_transaction = jh->b_cp_transaction;
972 if (cp_transaction) { 975 if (cp_transaction) {
973 JBUFFER_TRACE(jh, "rem 976 JBUFFER_TRACE(jh, "remove from old cp transaction");
974 cp_transaction->t_chp_ 977 cp_transaction->t_chp_stats.cs_dropped++;
975 __jbd2_journal_remove_ 978 __jbd2_journal_remove_checkpoint(jh);
976 } 979 }
977 980
978 /* Only re-checkpoint the buff 981 /* Only re-checkpoint the buffer_head if it is marked
979 * dirty. If the buffer was a 982 * dirty. If the buffer was added to the BJ_Forget list
980 * by jbd2_journal_forget, it 983 * by jbd2_journal_forget, it may no longer be dirty and
981 * there's no point in keeping 984 * there's no point in keeping a checkpoint record for
982 * it. */ 985 * it. */
983 986
984 /* 987 /*
985 * A buffer which has been fre 988 * A buffer which has been freed while still being journaled
986 * by a previous transaction, 989 * by a previous transaction, refile the buffer to BJ_Forget of
987 * the running transaction. If 990 * the running transaction. If the just committed transaction
988 * contains "add to orphan" op 991 * contains "add to orphan" operation, we can completely
989 * invalidate the buffer now. 992 * invalidate the buffer now. We are rather through in that
990 * since the buffer may be sti 993 * since the buffer may be still accessible when blocksize <
991 * pagesize and it is attached 994 * pagesize and it is attached to the last partial page.
992 */ 995 */
993 if (buffer_freed(bh) && !jh->b 996 if (buffer_freed(bh) && !jh->b_next_transaction) {
994 struct address_space * 997 struct address_space *mapping;
995 998
996 clear_buffer_freed(bh) 999 clear_buffer_freed(bh);
997 clear_buffer_jbddirty( 1000 clear_buffer_jbddirty(bh);
998 1001
999 /* 1002 /*
1000 * Block device buffe 1003 * Block device buffers need to stay mapped all the
1001 * time, so it is eno 1004 * time, so it is enough to clear buffer_jbddirty and
1002 * buffer_freed bits. 1005 * buffer_freed bits. For the file mapping buffers (i.e.
1003 * journalled data) w 1006 * journalled data) we need to unmap buffer and clear
1004 * more bits. We also 1007 * more bits. We also need to be careful about the check
1005 * because the data p 1008 * because the data page mapping can get cleared under
1006 * our hands. Note th 1009 * our hands. Note that if mapping == NULL, we don't
1007 * need to make buffe 1010 * need to make buffer unmapped because the page is
1008 * already detached f 1011 * already detached from the mapping and buffers cannot
1009 * get reused. 1012 * get reused.
1010 */ 1013 */
1011 mapping = READ_ONCE(b 1014 mapping = READ_ONCE(bh->b_folio->mapping);
1012 if (mapping && !sb_is 1015 if (mapping && !sb_is_blkdev_sb(mapping->host->i_sb)) {
1013 clear_buffer_ 1016 clear_buffer_mapped(bh);
1014 clear_buffer_ 1017 clear_buffer_new(bh);
1015 clear_buffer_ 1018 clear_buffer_req(bh);
1016 bh->b_bdev = 1019 bh->b_bdev = NULL;
1017 } 1020 }
1018 } 1021 }
1019 1022
1020 if (buffer_jbddirty(bh)) { 1023 if (buffer_jbddirty(bh)) {
1021 JBUFFER_TRACE(jh, "ad 1024 JBUFFER_TRACE(jh, "add to new checkpointing trans");
1022 __jbd2_journal_insert 1025 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
1023 if (is_journal_aborte 1026 if (is_journal_aborted(journal))
1024 clear_buffer_ 1027 clear_buffer_jbddirty(bh);
1025 } else { 1028 } else {
1026 J_ASSERT_BH(bh, !buff 1029 J_ASSERT_BH(bh, !buffer_dirty(bh));
1027 /* 1030 /*
1028 * The buffer on BJ_F 1031 * The buffer on BJ_Forget list and not jbddirty means
1029 * it has been freed 1032 * it has been freed by this transaction and hence it
1030 * could not have bee 1033 * could not have been reallocated until this
1031 * transaction has co 1034 * transaction has committed. *BUT* it could be
1032 * reallocated once w 1035 * reallocated once we have written all the data to
1033 * disk and before we 1036 * disk and before we process the buffer on BJ_Forget
1034 * list. 1037 * list.
1035 */ 1038 */
1036 if (!jh->b_next_trans 1039 if (!jh->b_next_transaction)
1037 try_to_free = 1040 try_to_free = 1;
1038 } 1041 }
1039 JBUFFER_TRACE(jh, "refile or 1042 JBUFFER_TRACE(jh, "refile or unfile buffer");
1040 drop_ref = __jbd2_journal_ref 1043 drop_ref = __jbd2_journal_refile_buffer(jh);
1041 spin_unlock(&jh->b_state_lock 1044 spin_unlock(&jh->b_state_lock);
1042 if (drop_ref) 1045 if (drop_ref)
1043 jbd2_journal_put_jour 1046 jbd2_journal_put_journal_head(jh);
1044 if (try_to_free) 1047 if (try_to_free)
1045 release_buffer_page(b 1048 release_buffer_page(bh); /* Drops bh reference */
1046 else 1049 else
1047 __brelse(bh); 1050 __brelse(bh);
1048 cond_resched_lock(&journal->j 1051 cond_resched_lock(&journal->j_list_lock);
1049 } 1052 }
1050 spin_unlock(&journal->j_list_lock); 1053 spin_unlock(&journal->j_list_lock);
1051 /* 1054 /*
1052 * This is a bit sleazy. We use j_li 1055 * This is a bit sleazy. We use j_list_lock to protect transition
1053 * of a transaction into T_FINISHED s 1056 * of a transaction into T_FINISHED state and calling
1054 * __jbd2_journal_drop_transaction(). 1057 * __jbd2_journal_drop_transaction(). Otherwise we could race with
1055 * other checkpointing code processin 1058 * other checkpointing code processing the transaction...
1056 */ 1059 */
1057 write_lock(&journal->j_state_lock); 1060 write_lock(&journal->j_state_lock);
1058 spin_lock(&journal->j_list_lock); 1061 spin_lock(&journal->j_list_lock);
1059 /* 1062 /*
1060 * Now recheck if some buffers did no 1063 * Now recheck if some buffers did not get attached to the transaction
1061 * while the lock was dropped... 1064 * while the lock was dropped...
1062 */ 1065 */
1063 if (commit_transaction->t_forget) { 1066 if (commit_transaction->t_forget) {
1064 spin_unlock(&journal->j_list_ 1067 spin_unlock(&journal->j_list_lock);
1065 write_unlock(&journal->j_stat 1068 write_unlock(&journal->j_state_lock);
1066 goto restart_loop; 1069 goto restart_loop;
1067 } 1070 }
1068 1071
1069 /* Add the transaction to the checkpo 1072 /* Add the transaction to the checkpoint list
1070 * __journal_remove_checkpoint() can 1073 * __journal_remove_checkpoint() can not destroy transaction
1071 * under us because it is not marked 1074 * under us because it is not marked as T_FINISHED yet */
1072 if (journal->j_checkpoint_transaction 1075 if (journal->j_checkpoint_transactions == NULL) {
1073 journal->j_checkpoint_transac 1076 journal->j_checkpoint_transactions = commit_transaction;
1074 commit_transaction->t_cpnext 1077 commit_transaction->t_cpnext = commit_transaction;
1075 commit_transaction->t_cpprev 1078 commit_transaction->t_cpprev = commit_transaction;
1076 } else { 1079 } else {
1077 commit_transaction->t_cpnext 1080 commit_transaction->t_cpnext =
1078 journal->j_checkpoint 1081 journal->j_checkpoint_transactions;
1079 commit_transaction->t_cpprev 1082 commit_transaction->t_cpprev =
1080 commit_transaction->t 1083 commit_transaction->t_cpnext->t_cpprev;
1081 commit_transaction->t_cpnext- 1084 commit_transaction->t_cpnext->t_cpprev =
1082 commit_transaction; 1085 commit_transaction;
1083 commit_transaction->t_cpprev- 1086 commit_transaction->t_cpprev->t_cpnext =
1084 commit_transa 1087 commit_transaction;
1085 } 1088 }
1086 spin_unlock(&journal->j_list_lock); 1089 spin_unlock(&journal->j_list_lock);
1087 1090
1088 /* Done with this transaction! */ 1091 /* Done with this transaction! */
1089 1092
1090 jbd2_debug(3, "JBD2: commit phase 7\n 1093 jbd2_debug(3, "JBD2: commit phase 7\n");
1091 1094
1092 J_ASSERT(commit_transaction->t_state 1095 J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
1093 1096
1094 commit_transaction->t_start = jiffies 1097 commit_transaction->t_start = jiffies;
1095 stats.run.rs_logging = jbd2_time_diff 1098 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
1096 1099 commit_transaction->t_start);
1097 1100
1098 /* 1101 /*
1099 * File the transaction statistics 1102 * File the transaction statistics
1100 */ 1103 */
1101 stats.ts_tid = commit_transaction->t_ 1104 stats.ts_tid = commit_transaction->t_tid;
1102 stats.run.rs_handle_count = 1105 stats.run.rs_handle_count =
1103 atomic_read(&commit_transacti 1106 atomic_read(&commit_transaction->t_handle_count);
1104 trace_jbd2_run_stats(journal->j_fs_de 1107 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
1105 commit_transacti 1108 commit_transaction->t_tid, &stats.run);
1106 stats.ts_requested = (commit_transact 1109 stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0;
1107 1110
1108 commit_transaction->t_state = T_COMMI 1111 commit_transaction->t_state = T_COMMIT_CALLBACK;
1109 J_ASSERT(commit_transaction == journa 1112 J_ASSERT(commit_transaction == journal->j_committing_transaction);
1110 WRITE_ONCE(journal->j_commit_sequence !! 1113 journal->j_commit_sequence = commit_transaction->t_tid;
1111 journal->j_committing_transaction = N 1114 journal->j_committing_transaction = NULL;
1112 commit_time = ktime_to_ns(ktime_sub(k 1115 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1113 1116
1114 /* 1117 /*
1115 * weight the commit time higher than 1118 * weight the commit time higher than the average time so we don't
1116 * react too strongly to vast changes 1119 * react too strongly to vast changes in the commit time
1117 */ 1120 */
1118 if (likely(journal->j_average_commit_ 1121 if (likely(journal->j_average_commit_time))
1119 journal->j_average_commit_tim 1122 journal->j_average_commit_time = (commit_time +
1120 journal->j_av 1123 journal->j_average_commit_time*3) / 4;
1121 else 1124 else
1122 journal->j_average_commit_tim 1125 journal->j_average_commit_time = commit_time;
1123 1126
1124 write_unlock(&journal->j_state_lock); 1127 write_unlock(&journal->j_state_lock);
1125 1128
1126 if (journal->j_commit_callback) 1129 if (journal->j_commit_callback)
1127 journal->j_commit_callback(jo 1130 journal->j_commit_callback(journal, commit_transaction);
1128 if (journal->j_fc_cleanup_callback) 1131 if (journal->j_fc_cleanup_callback)
1129 journal->j_fc_cleanup_callbac 1132 journal->j_fc_cleanup_callback(journal, 1, commit_transaction->t_tid);
1130 1133
1131 trace_jbd2_end_commit(journal, commit 1134 trace_jbd2_end_commit(journal, commit_transaction);
1132 jbd2_debug(1, "JBD2: commit %d comple 1135 jbd2_debug(1, "JBD2: commit %d complete, head %d\n",
1133 journal->j_commit_sequence, 1136 journal->j_commit_sequence, journal->j_tail_sequence);
1134 1137
1135 write_lock(&journal->j_state_lock); 1138 write_lock(&journal->j_state_lock);
1136 journal->j_flags &= ~JBD2_FULL_COMMIT 1139 journal->j_flags &= ~JBD2_FULL_COMMIT_ONGOING;
1137 journal->j_flags &= ~JBD2_FAST_COMMIT 1140 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
1138 spin_lock(&journal->j_list_lock); 1141 spin_lock(&journal->j_list_lock);
1139 commit_transaction->t_state = T_FINIS 1142 commit_transaction->t_state = T_FINISHED;
1140 /* Check if the transaction can be dr 1143 /* Check if the transaction can be dropped now that we are finished */
1141 if (commit_transaction->t_checkpoint_ 1144 if (commit_transaction->t_checkpoint_list == NULL) {
1142 __jbd2_journal_drop_transacti 1145 __jbd2_journal_drop_transaction(journal, commit_transaction);
1143 jbd2_journal_free_transaction 1146 jbd2_journal_free_transaction(commit_transaction);
1144 } 1147 }
1145 spin_unlock(&journal->j_list_lock); 1148 spin_unlock(&journal->j_list_lock);
1146 write_unlock(&journal->j_state_lock); 1149 write_unlock(&journal->j_state_lock);
1147 wake_up(&journal->j_wait_done_commit) 1150 wake_up(&journal->j_wait_done_commit);
1148 wake_up(&journal->j_fc_wait); 1151 wake_up(&journal->j_fc_wait);
1149 1152
1150 /* 1153 /*
1151 * Calculate overall stats 1154 * Calculate overall stats
1152 */ 1155 */
1153 spin_lock(&journal->j_history_lock); 1156 spin_lock(&journal->j_history_lock);
1154 journal->j_stats.ts_tid++; 1157 journal->j_stats.ts_tid++;
1155 journal->j_stats.ts_requested += stat 1158 journal->j_stats.ts_requested += stats.ts_requested;
1156 journal->j_stats.run.rs_wait += stats 1159 journal->j_stats.run.rs_wait += stats.run.rs_wait;
1157 journal->j_stats.run.rs_request_delay 1160 journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
1158 journal->j_stats.run.rs_running += st 1161 journal->j_stats.run.rs_running += stats.run.rs_running;
1159 journal->j_stats.run.rs_locked += sta 1162 journal->j_stats.run.rs_locked += stats.run.rs_locked;
1160 journal->j_stats.run.rs_flushing += s 1163 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
1161 journal->j_stats.run.rs_logging += st 1164 journal->j_stats.run.rs_logging += stats.run.rs_logging;
1162 journal->j_stats.run.rs_handle_count 1165 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
1163 journal->j_stats.run.rs_blocks += sta 1166 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
1164 journal->j_stats.run.rs_blocks_logged 1167 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
1165 spin_unlock(&journal->j_history_lock) 1168 spin_unlock(&journal->j_history_lock);
1166 } 1169 }
1167 1170
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.