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
180 /* Send all the data buffers related to an ino !! 184 /*
181 int jbd2_submit_inode_data(journal_t *journal, !! 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 dealyed allocation we may be doing
>> 188 * block allocation in writepages().
>> 189 */
>> 190 static int journal_submit_inode_data_buffers(struct address_space *mapping,
>> 191 loff_t dirty_start, loff_t dirty_end)
182 { 192 {
183 if (!jinode || !(jinode->i_flags & JI_ !! 193 int ret;
184 return 0; !! 194 struct writeback_control wbc = {
>> 195 .sync_mode = WB_SYNC_ALL,
>> 196 .nr_to_write = mapping->nrpages * 2,
>> 197 .range_start = dirty_start,
>> 198 .range_end = dirty_end,
>> 199 };
185 200
186 trace_jbd2_submit_inode_data(jinode->i !! 201 ret = generic_writepages(mapping, &wbc);
187 return journal->j_submit_inode_data_bu !! 202 return ret;
188 <<
189 } <<
190 EXPORT_SYMBOL(jbd2_submit_inode_data); <<
191 <<
192 int jbd2_wait_inode_data(journal_t *journal, s <<
193 { <<
194 if (!jinode || !(jinode->i_flags & JI_ <<
195 !jinode->i_vfs_inode || !jinod <<
196 return 0; <<
197 return filemap_fdatawait_range_keep_er <<
198 jinode->i_vfs_inode->i_mapping <<
199 jinode->i_dirty_end); <<
200 } 203 }
201 EXPORT_SYMBOL(jbd2_wait_inode_data); <<
202 204
203 /* 205 /*
204 * Submit all the data buffers of inode associ 206 * Submit all the data buffers of inode associated with the transaction to
205 * disk. 207 * disk.
206 * 208 *
207 * We are in a committing transaction. Therefo 209 * We are in a committing transaction. Therefore no new inode can be added to
208 * our inode list. We use JI_COMMIT_RUNNING fl 210 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
209 * operate on from being released while we wri 211 * operate on from being released while we write out pages.
210 */ 212 */
211 static int journal_submit_data_buffers(journal 213 static int journal_submit_data_buffers(journal_t *journal,
212 transaction_t *commit_transact 214 transaction_t *commit_transaction)
213 { 215 {
214 struct jbd2_inode *jinode; 216 struct jbd2_inode *jinode;
215 int err, ret = 0; 217 int err, ret = 0;
>> 218 struct address_space *mapping;
216 219
217 spin_lock(&journal->j_list_lock); 220 spin_lock(&journal->j_list_lock);
218 list_for_each_entry(jinode, &commit_tr 221 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
>> 222 loff_t dirty_start = jinode->i_dirty_start;
>> 223 loff_t dirty_end = jinode->i_dirty_end;
>> 224
219 if (!(jinode->i_flags & JI_WRI 225 if (!(jinode->i_flags & JI_WRITE_DATA))
220 continue; 226 continue;
>> 227 mapping = jinode->i_vfs_inode->i_mapping;
221 jinode->i_flags |= JI_COMMIT_R 228 jinode->i_flags |= JI_COMMIT_RUNNING;
222 spin_unlock(&journal->j_list_l 229 spin_unlock(&journal->j_list_lock);
223 /* submit the inode data buffe !! 230 /*
>> 231 * submit the inode data buffers. We use writepage
>> 232 * instead of writepages. Because writepages can do
>> 233 * block allocation with delalloc. We need to write
>> 234 * only allocated blocks here.
>> 235 */
224 trace_jbd2_submit_inode_data(j 236 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
225 if (journal->j_submit_inode_da !! 237 err = journal_submit_inode_data_buffers(mapping, dirty_start,
226 err = journal->j_submi !! 238 dirty_end);
227 if (!ret) !! 239 if (!ret)
228 ret = err; !! 240 ret = err;
229 } <<
230 spin_lock(&journal->j_list_loc 241 spin_lock(&journal->j_list_lock);
231 J_ASSERT(jinode->i_transaction 242 J_ASSERT(jinode->i_transaction == commit_transaction);
232 jinode->i_flags &= ~JI_COMMIT_ 243 jinode->i_flags &= ~JI_COMMIT_RUNNING;
233 smp_mb(); 244 smp_mb();
234 wake_up_bit(&jinode->i_flags, 245 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
235 } 246 }
236 spin_unlock(&journal->j_list_lock); 247 spin_unlock(&journal->j_list_lock);
237 return ret; 248 return ret;
238 } 249 }
239 250
240 int jbd2_journal_finish_inode_data_buffers(str <<
241 { <<
242 struct address_space *mapping = jinode <<
243 <<
244 return filemap_fdatawait_range_keep_er <<
245 <<
246 <<
247 } <<
248 <<
249 /* 251 /*
250 * Wait for data submitted for writeout, refil 252 * Wait for data submitted for writeout, refile inodes to proper
251 * transaction if needed. 253 * transaction if needed.
252 * 254 *
253 */ 255 */
254 static int journal_finish_inode_data_buffers(j 256 static int journal_finish_inode_data_buffers(journal_t *journal,
255 transaction_t *commit_transact 257 transaction_t *commit_transaction)
256 { 258 {
257 struct jbd2_inode *jinode, *next_i; 259 struct jbd2_inode *jinode, *next_i;
258 int err, ret = 0; 260 int err, ret = 0;
259 261
260 /* For locking, see the comment in jou 262 /* For locking, see the comment in journal_submit_data_buffers() */
261 spin_lock(&journal->j_list_lock); 263 spin_lock(&journal->j_list_lock);
262 list_for_each_entry(jinode, &commit_tr 264 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
>> 265 loff_t dirty_start = jinode->i_dirty_start;
>> 266 loff_t dirty_end = jinode->i_dirty_end;
>> 267
263 if (!(jinode->i_flags & JI_WAI 268 if (!(jinode->i_flags & JI_WAIT_DATA))
264 continue; 269 continue;
265 jinode->i_flags |= JI_COMMIT_R 270 jinode->i_flags |= JI_COMMIT_RUNNING;
266 spin_unlock(&journal->j_list_l 271 spin_unlock(&journal->j_list_lock);
267 /* wait for the inode data buf !! 272 err = filemap_fdatawait_range_keep_errors(
268 if (journal->j_finish_inode_da !! 273 jinode->i_vfs_inode->i_mapping, dirty_start,
269 err = journal->j_finis !! 274 dirty_end);
270 if (!ret) !! 275 if (!ret)
271 ret = err; !! 276 ret = err;
272 } <<
273 cond_resched(); <<
274 spin_lock(&journal->j_list_loc 277 spin_lock(&journal->j_list_lock);
275 jinode->i_flags &= ~JI_COMMIT_ 278 jinode->i_flags &= ~JI_COMMIT_RUNNING;
276 smp_mb(); 279 smp_mb();
277 wake_up_bit(&jinode->i_flags, 280 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
278 } 281 }
279 282
280 /* Now refile inode to proper lists */ 283 /* Now refile inode to proper lists */
281 list_for_each_entry_safe(jinode, next_ 284 list_for_each_entry_safe(jinode, next_i,
282 &commit_trans 285 &commit_transaction->t_inode_list, i_list) {
283 list_del(&jinode->i_list); 286 list_del(&jinode->i_list);
284 if (jinode->i_next_transaction 287 if (jinode->i_next_transaction) {
285 jinode->i_transaction 288 jinode->i_transaction = jinode->i_next_transaction;
286 jinode->i_next_transac 289 jinode->i_next_transaction = NULL;
287 list_add(&jinode->i_li 290 list_add(&jinode->i_list,
288 &jinode->i_tra 291 &jinode->i_transaction->t_inode_list);
289 } else { 292 } else {
290 jinode->i_transaction 293 jinode->i_transaction = NULL;
291 jinode->i_dirty_start 294 jinode->i_dirty_start = 0;
292 jinode->i_dirty_end = 295 jinode->i_dirty_end = 0;
293 } 296 }
294 } 297 }
295 spin_unlock(&journal->j_list_lock); 298 spin_unlock(&journal->j_list_lock);
296 299
297 return ret; 300 return ret;
298 } 301 }
299 302
300 static __u32 jbd2_checksum_data(__u32 crc32_su 303 static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
301 { 304 {
>> 305 struct page *page = bh->b_page;
302 char *addr; 306 char *addr;
303 __u32 checksum; 307 __u32 checksum;
304 308
305 addr = kmap_local_folio(bh->b_folio, b !! 309 addr = kmap_atomic(page);
306 checksum = crc32_be(crc32_sum, addr, b !! 310 checksum = crc32_be(crc32_sum,
307 kunmap_local(addr); !! 311 (void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
>> 312 kunmap_atomic(addr);
308 313
309 return checksum; 314 return checksum;
310 } 315 }
311 316
312 static void write_tag_block(journal_t *j, jour 317 static void write_tag_block(journal_t *j, journal_block_tag_t *tag,
313 unsigned lo 318 unsigned long long block)
314 { 319 {
315 tag->t_blocknr = cpu_to_be32(block & ( 320 tag->t_blocknr = cpu_to_be32(block & (u32)~0);
316 if (jbd2_has_feature_64bit(j)) 321 if (jbd2_has_feature_64bit(j))
317 tag->t_blocknr_high = cpu_to_b 322 tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
318 } 323 }
319 324
320 static void jbd2_block_tag_csum_set(journal_t 325 static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
321 struct buf 326 struct buffer_head *bh, __u32 sequence)
322 { 327 {
323 journal_block_tag3_t *tag3 = (journal_ 328 journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
>> 329 struct page *page = bh->b_page;
324 __u8 *addr; 330 __u8 *addr;
325 __u32 csum32; 331 __u32 csum32;
326 __be32 seq; 332 __be32 seq;
327 333
328 if (!jbd2_journal_has_csum_v2or3(j)) 334 if (!jbd2_journal_has_csum_v2or3(j))
329 return; 335 return;
330 336
331 seq = cpu_to_be32(sequence); 337 seq = cpu_to_be32(sequence);
332 addr = kmap_local_folio(bh->b_folio, b !! 338 addr = kmap_atomic(page);
333 csum32 = jbd2_chksum(j, j->j_csum_seed 339 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
334 csum32 = jbd2_chksum(j, csum32, addr, !! 340 csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data),
335 kunmap_local(addr); !! 341 bh->b_size);
>> 342 kunmap_atomic(addr);
336 343
337 if (jbd2_has_feature_csum3(j)) 344 if (jbd2_has_feature_csum3(j))
338 tag3->t_checksum = cpu_to_be32 345 tag3->t_checksum = cpu_to_be32(csum32);
339 else 346 else
340 tag->t_checksum = cpu_to_be16( 347 tag->t_checksum = cpu_to_be16(csum32);
341 } 348 }
342 /* 349 /*
343 * jbd2_journal_commit_transaction 350 * jbd2_journal_commit_transaction
344 * 351 *
345 * The primary function for committing a trans 352 * The primary function for committing a transaction to the log. This
346 * function is called by the journal thread to 353 * function is called by the journal thread to begin a complete commit.
347 */ 354 */
348 void jbd2_journal_commit_transaction(journal_t 355 void jbd2_journal_commit_transaction(journal_t *journal)
349 { 356 {
350 struct transaction_stats_s stats; 357 struct transaction_stats_s stats;
351 transaction_t *commit_transaction; 358 transaction_t *commit_transaction;
352 struct journal_head *jh; 359 struct journal_head *jh;
353 struct buffer_head *descriptor; 360 struct buffer_head *descriptor;
354 struct buffer_head **wbuf = journal->j 361 struct buffer_head **wbuf = journal->j_wbuf;
355 int bufs; 362 int bufs;
356 int escape; !! 363 int flags;
357 int err; 364 int err;
358 unsigned long long blocknr; 365 unsigned long long blocknr;
359 ktime_t start_time; 366 ktime_t start_time;
360 u64 commit_time; 367 u64 commit_time;
361 char *tagp = NULL; 368 char *tagp = NULL;
362 journal_block_tag_t *tag = NULL; 369 journal_block_tag_t *tag = NULL;
363 int space_left = 0; 370 int space_left = 0;
364 int first_tag = 0; 371 int first_tag = 0;
365 int tag_flag; 372 int tag_flag;
366 int i; 373 int i;
367 int tag_bytes = journal_tag_bytes(jour 374 int tag_bytes = journal_tag_bytes(journal);
368 struct buffer_head *cbh = NULL; /* For 375 struct buffer_head *cbh = NULL; /* For transactional checksums */
369 __u32 crc32_sum = ~0; 376 __u32 crc32_sum = ~0;
370 struct blk_plug plug; 377 struct blk_plug plug;
371 /* Tail of the journal */ 378 /* Tail of the journal */
372 unsigned long first_block; 379 unsigned long first_block;
373 tid_t first_tid; 380 tid_t first_tid;
374 int update_tail; 381 int update_tail;
375 int csum_size = 0; 382 int csum_size = 0;
376 LIST_HEAD(io_bufs); 383 LIST_HEAD(io_bufs);
377 LIST_HEAD(log_bufs); 384 LIST_HEAD(log_bufs);
378 385
379 if (jbd2_journal_has_csum_v2or3(journa 386 if (jbd2_journal_has_csum_v2or3(journal))
380 csum_size = sizeof(struct jbd2 387 csum_size = sizeof(struct jbd2_journal_block_tail);
381 388
382 /* 389 /*
383 * First job: lock down the current tr 390 * First job: lock down the current transaction and wait for
384 * all outstanding updates to complete 391 * all outstanding updates to complete.
385 */ 392 */
386 393
387 /* Do we need to erase the effects of 394 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
388 if (journal->j_flags & JBD2_FLUSHED) { 395 if (journal->j_flags & JBD2_FLUSHED) {
389 jbd2_debug(3, "super block upd !! 396 jbd_debug(3, "super block updated\n");
390 mutex_lock_io(&journal->j_chec 397 mutex_lock_io(&journal->j_checkpoint_mutex);
391 /* 398 /*
392 * We hold j_checkpoint_mutex 399 * We hold j_checkpoint_mutex so tail cannot change under us.
393 * We don't need any special d 400 * We don't need any special data guarantees for writing sb
394 * since journal is empty and 401 * since journal is empty and it is ok for write to be
395 * flushed only with transacti 402 * flushed only with transaction commit.
396 */ 403 */
397 jbd2_journal_update_sb_log_tai 404 jbd2_journal_update_sb_log_tail(journal,
398 405 journal->j_tail_sequence,
399 !! 406 journal->j_tail,
>> 407 REQ_SYNC);
400 mutex_unlock(&journal->j_check 408 mutex_unlock(&journal->j_checkpoint_mutex);
401 } else { 409 } else {
402 jbd2_debug(3, "superblock not !! 410 jbd_debug(3, "superblock not updated\n");
403 } 411 }
404 412
405 J_ASSERT(journal->j_running_transactio 413 J_ASSERT(journal->j_running_transaction != NULL);
406 J_ASSERT(journal->j_committing_transac 414 J_ASSERT(journal->j_committing_transaction == NULL);
407 415
408 write_lock(&journal->j_state_lock); <<
409 journal->j_flags |= JBD2_FULL_COMMIT_O <<
410 while (journal->j_flags & JBD2_FAST_CO <<
411 DEFINE_WAIT(wait); <<
412 <<
413 prepare_to_wait(&journal->j_fc <<
414 TASK_UNINTERRU <<
415 write_unlock(&journal->j_state <<
416 schedule(); <<
417 write_lock(&journal->j_state_l <<
418 finish_wait(&journal->j_fc_wai <<
419 /* <<
420 * TODO: by blocking fast comm <<
421 * fsync() latency slightly. S <<
422 * to block fast commits until <<
423 * state. So an optimization i <<
424 * commits here and wait for e <<
425 * just before we enter T_FLUS <<
426 * commits and this full commi <<
427 */ <<
428 } <<
429 write_unlock(&journal->j_state_lock); <<
430 <<
431 commit_transaction = journal->j_runnin 416 commit_transaction = journal->j_running_transaction;
432 417
433 trace_jbd2_start_commit(journal, commi 418 trace_jbd2_start_commit(journal, commit_transaction);
434 jbd2_debug(1, "JBD2: starting commit o !! 419 jbd_debug(1, "JBD2: starting commit of transaction %d\n",
435 commit_transaction->t_ 420 commit_transaction->t_tid);
436 421
437 write_lock(&journal->j_state_lock); 422 write_lock(&journal->j_state_lock);
438 journal->j_fc_off = 0; <<
439 J_ASSERT(commit_transaction->t_state = 423 J_ASSERT(commit_transaction->t_state == T_RUNNING);
440 commit_transaction->t_state = T_LOCKED 424 commit_transaction->t_state = T_LOCKED;
441 425
442 trace_jbd2_commit_locking(journal, com 426 trace_jbd2_commit_locking(journal, commit_transaction);
443 stats.run.rs_wait = commit_transaction 427 stats.run.rs_wait = commit_transaction->t_max_wait;
444 stats.run.rs_request_delay = 0; 428 stats.run.rs_request_delay = 0;
445 stats.run.rs_locked = jiffies; 429 stats.run.rs_locked = jiffies;
446 if (commit_transaction->t_requested) 430 if (commit_transaction->t_requested)
447 stats.run.rs_request_delay = 431 stats.run.rs_request_delay =
448 jbd2_time_diff(commit_ 432 jbd2_time_diff(commit_transaction->t_requested,
449 stats.r 433 stats.run.rs_locked);
450 stats.run.rs_running = jbd2_time_diff( 434 stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
451 435 stats.run.rs_locked);
452 436
453 // waits for any t_updates to finish !! 437 spin_lock(&commit_transaction->t_handle_lock);
454 jbd2_journal_wait_updates(journal); !! 438 while (atomic_read(&commit_transaction->t_updates)) {
>> 439 DEFINE_WAIT(wait);
455 440
>> 441 prepare_to_wait(&journal->j_wait_updates, &wait,
>> 442 TASK_UNINTERRUPTIBLE);
>> 443 if (atomic_read(&commit_transaction->t_updates)) {
>> 444 spin_unlock(&commit_transaction->t_handle_lock);
>> 445 write_unlock(&journal->j_state_lock);
>> 446 schedule();
>> 447 write_lock(&journal->j_state_lock);
>> 448 spin_lock(&commit_transaction->t_handle_lock);
>> 449 }
>> 450 finish_wait(&journal->j_wait_updates, &wait);
>> 451 }
>> 452 spin_unlock(&commit_transaction->t_handle_lock);
456 commit_transaction->t_state = T_SWITCH 453 commit_transaction->t_state = T_SWITCH;
>> 454 write_unlock(&journal->j_state_lock);
457 455
458 J_ASSERT (atomic_read(&commit_transact 456 J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
459 journal->j_max_transac 457 journal->j_max_transaction_buffers);
460 458
461 /* 459 /*
462 * First thing we are allowed to do is 460 * First thing we are allowed to do is to discard any remaining
463 * BJ_Reserved buffers. Note, it is _ 461 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
464 * that there are no such buffers: if 462 * that there are no such buffers: if a large filesystem
465 * operation like a truncate needs to 463 * operation like a truncate needs to split itself over multiple
466 * transactions, then it may try to do 464 * transactions, then it may try to do a jbd2_journal_restart() while
467 * there are still BJ_Reserved buffers 465 * there are still BJ_Reserved buffers outstanding. These must
468 * be released cleanly from the curren 466 * be released cleanly from the current transaction.
469 * 467 *
470 * In this case, the filesystem must s 468 * In this case, the filesystem must still reserve write access
471 * again before modifying the buffer i 469 * again before modifying the buffer in the new transaction, but
472 * we do not require it to remember ex 470 * we do not require it to remember exactly which old buffers it
473 * has reserved. This is consistent w 471 * has reserved. This is consistent with the existing behaviour
474 * that multiple jbd2_journal_get_writ 472 * that multiple jbd2_journal_get_write_access() calls to the same
475 * buffer are perfectly permissible. 473 * buffer are perfectly permissible.
476 * We use journal->j_state_lock here t <<
477 * t_reserved_list with eviction of bu <<
478 */ 474 */
479 while (commit_transaction->t_reserved_ 475 while (commit_transaction->t_reserved_list) {
480 jh = commit_transaction->t_res 476 jh = commit_transaction->t_reserved_list;
481 JBUFFER_TRACE(jh, "reserved, u 477 JBUFFER_TRACE(jh, "reserved, unused: refile");
482 /* 478 /*
483 * A jbd2_journal_get_undo_acc 479 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
484 * leave undo-committed data. 480 * leave undo-committed data.
485 */ 481 */
486 if (jh->b_committed_data) { 482 if (jh->b_committed_data) {
487 struct buffer_head *bh 483 struct buffer_head *bh = jh2bh(jh);
488 484
489 spin_lock(&jh->b_state !! 485 jbd_lock_bh_state(bh);
490 jbd2_free(jh->b_commit 486 jbd2_free(jh->b_committed_data, bh->b_size);
491 jh->b_committed_data = 487 jh->b_committed_data = NULL;
492 spin_unlock(&jh->b_sta !! 488 jbd_unlock_bh_state(bh);
493 } 489 }
494 jbd2_journal_refile_buffer(jou 490 jbd2_journal_refile_buffer(journal, jh);
495 } 491 }
496 492
497 write_unlock(&journal->j_state_lock); <<
498 /* 493 /*
499 * Now try to drop any written-back bu 494 * Now try to drop any written-back buffers from the journal's
500 * checkpoint lists. We do this *befo 495 * checkpoint lists. We do this *before* commit because it potentially
501 * frees some memory 496 * frees some memory
502 */ 497 */
503 spin_lock(&journal->j_list_lock); 498 spin_lock(&journal->j_list_lock);
504 __jbd2_journal_clean_checkpoint_list(j !! 499 __jbd2_journal_clean_checkpoint_list(journal, false);
505 spin_unlock(&journal->j_list_lock); 500 spin_unlock(&journal->j_list_lock);
506 501
507 jbd2_debug(3, "JBD2: commit phase 1\n" !! 502 jbd_debug(3, "JBD2: commit phase 1\n");
508 503
509 /* 504 /*
510 * Clear revoked flag to reflect there 505 * Clear revoked flag to reflect there is no revoked buffers
511 * in the next transaction which is go 506 * in the next transaction which is going to be started.
512 */ 507 */
513 jbd2_clear_buffer_revoked_flags(journa 508 jbd2_clear_buffer_revoked_flags(journal);
514 509
515 /* 510 /*
516 * Switch to a new revoke table. 511 * Switch to a new revoke table.
517 */ 512 */
518 jbd2_journal_switch_revoke_table(journ 513 jbd2_journal_switch_revoke_table(journal);
519 514
520 write_lock(&journal->j_state_lock); <<
521 /* 515 /*
522 * Reserved credits cannot be claimed 516 * Reserved credits cannot be claimed anymore, free them
523 */ 517 */
524 atomic_sub(atomic_read(&journal->j_res 518 atomic_sub(atomic_read(&journal->j_reserved_credits),
525 &commit_transaction->t_outs 519 &commit_transaction->t_outstanding_credits);
526 520
>> 521 write_lock(&journal->j_state_lock);
527 trace_jbd2_commit_flushing(journal, co 522 trace_jbd2_commit_flushing(journal, commit_transaction);
528 stats.run.rs_flushing = jiffies; 523 stats.run.rs_flushing = jiffies;
529 stats.run.rs_locked = jbd2_time_diff(s 524 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
530 s 525 stats.run.rs_flushing);
531 526
532 commit_transaction->t_state = T_FLUSH; 527 commit_transaction->t_state = T_FLUSH;
533 journal->j_committing_transaction = co 528 journal->j_committing_transaction = commit_transaction;
534 journal->j_running_transaction = NULL; 529 journal->j_running_transaction = NULL;
535 start_time = ktime_get(); 530 start_time = ktime_get();
536 commit_transaction->t_log_start = jour 531 commit_transaction->t_log_start = journal->j_head;
537 wake_up_all(&journal->j_wait_transacti !! 532 wake_up(&journal->j_wait_transaction_locked);
538 write_unlock(&journal->j_state_lock); 533 write_unlock(&journal->j_state_lock);
539 534
540 jbd2_debug(3, "JBD2: commit phase 2a\n !! 535 jbd_debug(3, "JBD2: commit phase 2a\n");
541 536
542 /* 537 /*
543 * Now start flushing things to disk, 538 * Now start flushing things to disk, in the order they appear
544 * on the transaction lists. Data blo 539 * on the transaction lists. Data blocks go first.
545 */ 540 */
546 err = journal_submit_data_buffers(jour 541 err = journal_submit_data_buffers(journal, commit_transaction);
547 if (err) 542 if (err)
548 jbd2_journal_abort(journal, er 543 jbd2_journal_abort(journal, err);
549 544
550 blk_start_plug(&plug); 545 blk_start_plug(&plug);
551 jbd2_journal_write_revoke_records(comm 546 jbd2_journal_write_revoke_records(commit_transaction, &log_bufs);
552 547
553 jbd2_debug(3, "JBD2: commit phase 2b\n !! 548 jbd_debug(3, "JBD2: commit phase 2b\n");
554 549
555 /* 550 /*
556 * Way to go: we have now written out 551 * Way to go: we have now written out all of the data for a
557 * transaction! Now comes the tricky 552 * transaction! Now comes the tricky part: we need to write out
558 * metadata. Loop over the transactio 553 * metadata. Loop over the transaction's entire buffer list:
559 */ 554 */
560 write_lock(&journal->j_state_lock); 555 write_lock(&journal->j_state_lock);
561 commit_transaction->t_state = T_COMMIT 556 commit_transaction->t_state = T_COMMIT;
562 write_unlock(&journal->j_state_lock); 557 write_unlock(&journal->j_state_lock);
563 558
564 trace_jbd2_commit_logging(journal, com 559 trace_jbd2_commit_logging(journal, commit_transaction);
565 stats.run.rs_logging = jiffies; 560 stats.run.rs_logging = jiffies;
566 stats.run.rs_flushing = jbd2_time_diff 561 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
567 562 stats.run.rs_logging);
568 stats.run.rs_blocks = commit_transacti !! 563 stats.run.rs_blocks =
>> 564 atomic_read(&commit_transaction->t_outstanding_credits);
569 stats.run.rs_blocks_logged = 0; 565 stats.run.rs_blocks_logged = 0;
570 566
571 J_ASSERT(commit_transaction->t_nr_buff 567 J_ASSERT(commit_transaction->t_nr_buffers <=
572 atomic_read(&commit_transacti 568 atomic_read(&commit_transaction->t_outstanding_credits));
573 569
>> 570 err = 0;
574 bufs = 0; 571 bufs = 0;
575 descriptor = NULL; 572 descriptor = NULL;
576 while (commit_transaction->t_buffers) 573 while (commit_transaction->t_buffers) {
577 574
578 /* Find the next buffer to be 575 /* Find the next buffer to be journaled... */
579 576
580 jh = commit_transaction->t_buf 577 jh = commit_transaction->t_buffers;
581 578
582 /* If we're in abort mode, we 579 /* If we're in abort mode, we just un-journal the buffer and
583 release it. */ 580 release it. */
584 581
585 if (is_journal_aborted(journal 582 if (is_journal_aborted(journal)) {
586 clear_buffer_jbddirty( 583 clear_buffer_jbddirty(jh2bh(jh));
587 JBUFFER_TRACE(jh, "jou 584 JBUFFER_TRACE(jh, "journal is aborting: refile");
588 jbd2_buffer_abort_trig 585 jbd2_buffer_abort_trigger(jh,
589 586 jh->b_frozen_data ?
590 587 jh->b_frozen_triggers :
591 588 jh->b_triggers);
592 jbd2_journal_refile_bu 589 jbd2_journal_refile_buffer(journal, jh);
593 /* If that was the las 590 /* If that was the last one, we need to clean up
594 * any descriptor buff 591 * any descriptor buffers which may have been
595 * already allocated, 592 * already allocated, even if we are now
596 * aborting. */ 593 * aborting. */
597 if (!commit_transactio 594 if (!commit_transaction->t_buffers)
598 goto start_jou 595 goto start_journal_io;
599 continue; 596 continue;
600 } 597 }
601 598
602 /* Make sure we have a descrip 599 /* Make sure we have a descriptor block in which to
603 record the metadata buffer. 600 record the metadata buffer. */
604 601
605 if (!descriptor) { 602 if (!descriptor) {
606 J_ASSERT (bufs == 0); 603 J_ASSERT (bufs == 0);
607 604
608 jbd2_debug(4, "JBD2: g !! 605 jbd_debug(4, "JBD2: get descriptor\n");
609 606
610 descriptor = jbd2_jour 607 descriptor = jbd2_journal_get_descriptor_buffer(
611 608 commit_transaction,
612 609 JBD2_DESCRIPTOR_BLOCK);
613 if (!descriptor) { 610 if (!descriptor) {
614 jbd2_journal_a 611 jbd2_journal_abort(journal, -EIO);
615 continue; 612 continue;
616 } 613 }
617 614
618 jbd2_debug(4, "JBD2: g !! 615 jbd_debug(4, "JBD2: got buffer %llu (%p)\n",
619 (unsigned long 616 (unsigned long long)descriptor->b_blocknr,
620 descriptor->b_ 617 descriptor->b_data);
621 tagp = &descriptor->b_ 618 tagp = &descriptor->b_data[sizeof(journal_header_t)];
622 space_left = descripto 619 space_left = descriptor->b_size -
623 620 sizeof(journal_header_t);
624 first_tag = 1; 621 first_tag = 1;
625 set_buffer_jwrite(desc 622 set_buffer_jwrite(descriptor);
626 set_buffer_dirty(descr 623 set_buffer_dirty(descriptor);
627 wbuf[bufs++] = descrip 624 wbuf[bufs++] = descriptor;
628 625
629 /* Record it so that w 626 /* Record it so that we can wait for IO
630 completion later */ 627 completion later */
631 BUFFER_TRACE(descripto 628 BUFFER_TRACE(descriptor, "ph3: file as descriptor");
632 jbd2_file_log_bh(&log_ 629 jbd2_file_log_bh(&log_bufs, descriptor);
633 } 630 }
634 631
635 /* Where is the buffer to be w 632 /* Where is the buffer to be written? */
636 633
637 err = jbd2_journal_next_log_bl 634 err = jbd2_journal_next_log_block(journal, &blocknr);
638 /* If the block mapping failed 635 /* If the block mapping failed, just abandon the buffer
639 and repeat this loop: we'll 636 and repeat this loop: we'll fall into the
640 refile-on-abort condition a 637 refile-on-abort condition above. */
641 if (err) { 638 if (err) {
642 jbd2_journal_abort(jou 639 jbd2_journal_abort(journal, err);
643 continue; 640 continue;
644 } 641 }
645 642
646 /* 643 /*
647 * start_this_handle() uses t_ 644 * start_this_handle() uses t_outstanding_credits to determine
648 * the free space in the log. !! 645 * the free space in the log, but this counter is changed
>> 646 * by jbd2_journal_next_log_block() also.
649 */ 647 */
650 atomic_dec(&commit_transaction 648 atomic_dec(&commit_transaction->t_outstanding_credits);
651 649
652 /* Bump b_count to prevent tru 650 /* Bump b_count to prevent truncate from stumbling over
653 the shadowed buffer! @@@ T 651 the shadowed buffer! @@@ This can go if we ever get
654 rid of the shadow pairing o 652 rid of the shadow pairing of buffers. */
655 atomic_inc(&jh2bh(jh)->b_count 653 atomic_inc(&jh2bh(jh)->b_count);
656 654
657 /* 655 /*
658 * Make a temporary IO buffer 656 * Make a temporary IO buffer with which to write it out
659 * (this will requeue the meta 657 * (this will requeue the metadata buffer to BJ_Shadow).
660 */ 658 */
661 set_bit(BH_JWrite, &jh2bh(jh)- 659 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
662 JBUFFER_TRACE(jh, "ph3: write 660 JBUFFER_TRACE(jh, "ph3: write metadata");
663 escape = jbd2_journal_write_me !! 661 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
664 662 jh, &wbuf[bufs], blocknr);
665 if (escape < 0) { !! 663 if (flags < 0) {
666 jbd2_journal_abort(jou !! 664 jbd2_journal_abort(journal, flags);
667 continue; 665 continue;
668 } 666 }
669 jbd2_file_log_bh(&io_bufs, wbu 667 jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
670 668
671 /* Record the new block's tag 669 /* Record the new block's tag in the current descriptor
672 buffer */ 670 buffer */
673 671
674 tag_flag = 0; 672 tag_flag = 0;
675 if (escape) !! 673 if (flags & 1)
676 tag_flag |= JBD2_FLAG_ 674 tag_flag |= JBD2_FLAG_ESCAPE;
677 if (!first_tag) 675 if (!first_tag)
678 tag_flag |= JBD2_FLAG_ 676 tag_flag |= JBD2_FLAG_SAME_UUID;
679 677
680 tag = (journal_block_tag_t *) 678 tag = (journal_block_tag_t *) tagp;
681 write_tag_block(journal, tag, 679 write_tag_block(journal, tag, jh2bh(jh)->b_blocknr);
682 tag->t_flags = cpu_to_be16(tag 680 tag->t_flags = cpu_to_be16(tag_flag);
683 jbd2_block_tag_csum_set(journa 681 jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
684 commit 682 commit_transaction->t_tid);
685 tagp += tag_bytes; 683 tagp += tag_bytes;
686 space_left -= tag_bytes; 684 space_left -= tag_bytes;
687 bufs++; 685 bufs++;
688 686
689 if (first_tag) { 687 if (first_tag) {
690 memcpy (tagp, journal- 688 memcpy (tagp, journal->j_uuid, 16);
691 tagp += 16; 689 tagp += 16;
692 space_left -= 16; 690 space_left -= 16;
693 first_tag = 0; 691 first_tag = 0;
694 } 692 }
695 693
696 /* If there's no more to do, o 694 /* If there's no more to do, or if the descriptor is full,
697 let the IO rip! */ 695 let the IO rip! */
698 696
699 if (bufs == journal->j_wbufsiz 697 if (bufs == journal->j_wbufsize ||
700 commit_transaction->t_buff 698 commit_transaction->t_buffers == NULL ||
701 space_left < tag_bytes + 1 699 space_left < tag_bytes + 16 + csum_size) {
702 700
703 jbd2_debug(4, "JBD2: S !! 701 jbd_debug(4, "JBD2: Submit %d IOs\n", bufs);
704 702
705 /* Write an end-of-des 703 /* Write an end-of-descriptor marker before
706 submitting the IOs. 704 submitting the IOs. "tag" still points to
707 the last tag we set 705 the last tag we set up. */
708 706
709 tag->t_flags |= cpu_to 707 tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
710 start_journal_io: 708 start_journal_io:
711 if (descriptor) 709 if (descriptor)
712 jbd2_descripto 710 jbd2_descriptor_block_csum_set(journal,
713 711 descriptor);
714 712
715 for (i = 0; i < bufs; 713 for (i = 0; i < bufs; i++) {
716 struct buffer_ 714 struct buffer_head *bh = wbuf[i];
717 <<
718 /* 715 /*
719 * Compute che 716 * Compute checksum.
720 */ 717 */
721 if (jbd2_has_f 718 if (jbd2_has_feature_checksum(journal)) {
722 crc32_ 719 crc32_sum =
723 jb 720 jbd2_checksum_data(crc32_sum, bh);
724 } 721 }
725 722
726 lock_buffer(bh 723 lock_buffer(bh);
727 clear_buffer_d 724 clear_buffer_dirty(bh);
728 set_buffer_upt 725 set_buffer_uptodate(bh);
729 bh->b_end_io = 726 bh->b_end_io = journal_end_buffer_io_sync;
730 submit_bh(REQ_ !! 727 submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
731 bh); <<
732 } 728 }
733 cond_resched(); 729 cond_resched();
>> 730 stats.run.rs_blocks_logged += bufs;
734 731
735 /* Force a new descrip 732 /* Force a new descriptor to be generated next
736 time round the loop 733 time round the loop. */
737 descriptor = NULL; 734 descriptor = NULL;
738 bufs = 0; 735 bufs = 0;
739 } 736 }
740 } 737 }
741 738
742 err = journal_finish_inode_data_buffer 739 err = journal_finish_inode_data_buffers(journal, commit_transaction);
743 if (err) { 740 if (err) {
744 printk(KERN_WARNING 741 printk(KERN_WARNING
745 "JBD2: Detected IO err 742 "JBD2: Detected IO errors while flushing file data "
746 "on %s\n", journal->j_d 743 "on %s\n", journal->j_devname);
747 if (journal->j_flags & JBD2_AB 744 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
748 jbd2_journal_abort(jou 745 jbd2_journal_abort(journal, err);
749 err = 0; 746 err = 0;
750 } 747 }
751 748
752 /* 749 /*
753 * Get current oldest transaction in t 750 * Get current oldest transaction in the log before we issue flush
754 * to the filesystem device. After the 751 * to the filesystem device. After the flush we can be sure that
755 * blocks of all older transactions ar 752 * blocks of all older transactions are checkpointed to persistent
756 * storage and we will be safe to upda 753 * storage and we will be safe to update journal start in the
757 * superblock with the numbers we get 754 * superblock with the numbers we get here.
758 */ 755 */
759 update_tail = 756 update_tail =
760 jbd2_journal_get_log_tail(jour 757 jbd2_journal_get_log_tail(journal, &first_tid, &first_block);
761 758
762 write_lock(&journal->j_state_lock); 759 write_lock(&journal->j_state_lock);
763 if (update_tail) { 760 if (update_tail) {
764 long freed = first_block - jou 761 long freed = first_block - journal->j_tail;
765 762
766 if (first_block < journal->j_t 763 if (first_block < journal->j_tail)
767 freed += journal->j_la 764 freed += journal->j_last - journal->j_first;
768 /* Update tail only if we free 765 /* Update tail only if we free significant amount of space */
769 if (freed < journal->j_max_tra !! 766 if (freed < journal->j_maxlen / 4)
770 update_tail = 0; 767 update_tail = 0;
771 } 768 }
772 J_ASSERT(commit_transaction->t_state = 769 J_ASSERT(commit_transaction->t_state == T_COMMIT);
773 commit_transaction->t_state = T_COMMIT 770 commit_transaction->t_state = T_COMMIT_DFLUSH;
774 write_unlock(&journal->j_state_lock); 771 write_unlock(&journal->j_state_lock);
775 772
776 /* !! 773 /*
777 * If the journal is not located on th 774 * If the journal is not located on the file system device,
778 * then we must flush the file system 775 * then we must flush the file system device before we issue
779 * the commit record 776 * the commit record
780 */ 777 */
781 if (commit_transaction->t_need_data_fl 778 if (commit_transaction->t_need_data_flush &&
782 (journal->j_fs_dev != journal->j_d 779 (journal->j_fs_dev != journal->j_dev) &&
783 (journal->j_flags & JBD2_BARRIER)) 780 (journal->j_flags & JBD2_BARRIER))
784 blkdev_issue_flush(journal->j_ !! 781 blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL);
785 782
786 /* Done it all: now write the commit r 783 /* Done it all: now write the commit record asynchronously. */
787 if (jbd2_has_feature_async_commit(jour 784 if (jbd2_has_feature_async_commit(journal)) {
788 err = journal_submit_commit_re 785 err = journal_submit_commit_record(journal, commit_transaction,
789 786 &cbh, crc32_sum);
790 if (err) 787 if (err)
791 jbd2_journal_abort(jou !! 788 __jbd2_journal_abort_hard(journal);
792 } 789 }
793 790
794 blk_finish_plug(&plug); 791 blk_finish_plug(&plug);
795 792
796 /* Lo and behold: we have just managed 793 /* Lo and behold: we have just managed to send a transaction to
797 the log. Before we can commit it, 794 the log. Before we can commit it, wait for the IO so far to
798 complete. Control buffers being wr 795 complete. Control buffers being written are on the
799 transaction's t_log_list queue, and 796 transaction's t_log_list queue, and metadata buffers are on
800 the io_bufs list. 797 the io_bufs list.
801 798
802 Wait for the buffers in reverse ord 799 Wait for the buffers in reverse order. That way we are
803 less likely to be woken up until al 800 less likely to be woken up until all IOs have completed, and
804 so we incur less scheduling load. 801 so we incur less scheduling load.
805 */ 802 */
806 803
807 jbd2_debug(3, "JBD2: commit phase 3\n" !! 804 jbd_debug(3, "JBD2: commit phase 3\n");
808 805
809 while (!list_empty(&io_bufs)) { 806 while (!list_empty(&io_bufs)) {
810 struct buffer_head *bh = list_ 807 struct buffer_head *bh = list_entry(io_bufs.prev,
811 808 struct buffer_head,
812 809 b_assoc_buffers);
813 810
814 wait_on_buffer(bh); 811 wait_on_buffer(bh);
815 cond_resched(); 812 cond_resched();
816 813
817 if (unlikely(!buffer_uptodate( 814 if (unlikely(!buffer_uptodate(bh)))
818 err = -EIO; 815 err = -EIO;
819 jbd2_unfile_log_bh(bh); 816 jbd2_unfile_log_bh(bh);
820 stats.run.rs_blocks_logged++; <<
821 817
822 /* 818 /*
823 * The list contains temporary 819 * The list contains temporary buffer heads created by
824 * jbd2_journal_write_metadata 820 * jbd2_journal_write_metadata_buffer().
825 */ 821 */
826 BUFFER_TRACE(bh, "dumping temp 822 BUFFER_TRACE(bh, "dumping temporary bh");
827 __brelse(bh); 823 __brelse(bh);
828 J_ASSERT_BH(bh, atomic_read(&b 824 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
829 free_buffer_head(bh); 825 free_buffer_head(bh);
830 826
831 /* We also have to refile the 827 /* We also have to refile the corresponding shadowed buffer */
832 jh = commit_transaction->t_sha 828 jh = commit_transaction->t_shadow_list->b_tprev;
833 bh = jh2bh(jh); 829 bh = jh2bh(jh);
834 clear_buffer_jwrite(bh); 830 clear_buffer_jwrite(bh);
835 J_ASSERT_BH(bh, buffer_jbddirt 831 J_ASSERT_BH(bh, buffer_jbddirty(bh));
836 J_ASSERT_BH(bh, !buffer_shadow 832 J_ASSERT_BH(bh, !buffer_shadow(bh));
837 833
838 /* The metadata is now release 834 /* The metadata is now released for reuse, but we need
839 to remember it against this 835 to remember it against this transaction so that when
840 we finally commit, we can d 836 we finally commit, we can do any checkpointing
841 required. */ 837 required. */
842 JBUFFER_TRACE(jh, "file as BJ_ 838 JBUFFER_TRACE(jh, "file as BJ_Forget");
843 jbd2_journal_file_buffer(jh, c 839 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
844 JBUFFER_TRACE(jh, "brelse shad 840 JBUFFER_TRACE(jh, "brelse shadowed buffer");
845 __brelse(bh); 841 __brelse(bh);
846 } 842 }
847 843
848 J_ASSERT (commit_transaction->t_shadow 844 J_ASSERT (commit_transaction->t_shadow_list == NULL);
849 845
850 jbd2_debug(3, "JBD2: commit phase 4\n" !! 846 jbd_debug(3, "JBD2: commit phase 4\n");
851 847
852 /* Here we wait for the revoke record 848 /* Here we wait for the revoke record and descriptor record buffers */
853 while (!list_empty(&log_bufs)) { 849 while (!list_empty(&log_bufs)) {
854 struct buffer_head *bh; 850 struct buffer_head *bh;
855 851
856 bh = list_entry(log_bufs.prev, 852 bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
857 wait_on_buffer(bh); 853 wait_on_buffer(bh);
858 cond_resched(); 854 cond_resched();
859 855
860 if (unlikely(!buffer_uptodate( 856 if (unlikely(!buffer_uptodate(bh)))
861 err = -EIO; 857 err = -EIO;
862 858
863 BUFFER_TRACE(bh, "ph5: control 859 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
864 clear_buffer_jwrite(bh); 860 clear_buffer_jwrite(bh);
865 jbd2_unfile_log_bh(bh); 861 jbd2_unfile_log_bh(bh);
866 stats.run.rs_blocks_logged++; <<
867 __brelse(bh); /* One 862 __brelse(bh); /* One for getblk */
868 /* AKPM: bforget here */ 863 /* AKPM: bforget here */
869 } 864 }
870 865
871 if (err) 866 if (err)
872 jbd2_journal_abort(journal, er 867 jbd2_journal_abort(journal, err);
873 868
874 jbd2_debug(3, "JBD2: commit phase 5\n" !! 869 jbd_debug(3, "JBD2: commit phase 5\n");
875 write_lock(&journal->j_state_lock); 870 write_lock(&journal->j_state_lock);
876 J_ASSERT(commit_transaction->t_state = 871 J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
877 commit_transaction->t_state = T_COMMIT 872 commit_transaction->t_state = T_COMMIT_JFLUSH;
878 write_unlock(&journal->j_state_lock); 873 write_unlock(&journal->j_state_lock);
879 874
880 if (!jbd2_has_feature_async_commit(jou 875 if (!jbd2_has_feature_async_commit(journal)) {
881 err = journal_submit_commit_re 876 err = journal_submit_commit_record(journal, commit_transaction,
882 877 &cbh, crc32_sum);
883 if (err) 878 if (err)
884 jbd2_journal_abort(jou !! 879 __jbd2_journal_abort_hard(journal);
885 } 880 }
886 if (cbh) 881 if (cbh)
887 err = journal_wait_on_commit_r 882 err = journal_wait_on_commit_record(journal, cbh);
888 stats.run.rs_blocks_logged++; <<
889 if (jbd2_has_feature_async_commit(jour 883 if (jbd2_has_feature_async_commit(journal) &&
890 journal->j_flags & JBD2_BARRIER) { 884 journal->j_flags & JBD2_BARRIER) {
891 blkdev_issue_flush(journal->j_ !! 885 blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL);
892 } 886 }
893 887
894 if (err) 888 if (err)
895 jbd2_journal_abort(journal, er 889 jbd2_journal_abort(journal, err);
896 890
897 WARN_ON_ONCE( <<
898 atomic_read(&commit_transactio <<
899 <<
900 /* 891 /*
901 * Now disk caches for filesystem devi 892 * Now disk caches for filesystem device are flushed so we are safe to
902 * erase checkpointed transactions fro 893 * erase checkpointed transactions from the log by updating journal
903 * superblock. 894 * superblock.
904 */ 895 */
905 if (update_tail) 896 if (update_tail)
906 jbd2_update_log_tail(journal, 897 jbd2_update_log_tail(journal, first_tid, first_block);
907 898
908 /* End of a transaction! Finally, we 899 /* End of a transaction! Finally, we can do checkpoint
909 processing: any buffers committed a 900 processing: any buffers committed as a result of this
910 transaction can be removed from any 901 transaction can be removed from any checkpoint list it was on
911 before. */ 902 before. */
912 903
913 jbd2_debug(3, "JBD2: commit phase 6\n" !! 904 jbd_debug(3, "JBD2: commit phase 6\n");
914 905
915 J_ASSERT(list_empty(&commit_transactio 906 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
916 J_ASSERT(commit_transaction->t_buffers 907 J_ASSERT(commit_transaction->t_buffers == NULL);
917 J_ASSERT(commit_transaction->t_checkpo 908 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
918 J_ASSERT(commit_transaction->t_shadow_ 909 J_ASSERT(commit_transaction->t_shadow_list == NULL);
919 910
920 restart_loop: 911 restart_loop:
921 /* 912 /*
922 * As there are other places (journal_ 913 * As there are other places (journal_unmap_buffer()) adding buffers
923 * to this list we have to be careful 914 * to this list we have to be careful and hold the j_list_lock.
924 */ 915 */
925 spin_lock(&journal->j_list_lock); 916 spin_lock(&journal->j_list_lock);
926 while (commit_transaction->t_forget) { 917 while (commit_transaction->t_forget) {
927 transaction_t *cp_transaction; 918 transaction_t *cp_transaction;
928 struct buffer_head *bh; 919 struct buffer_head *bh;
929 int try_to_free = 0; 920 int try_to_free = 0;
930 bool drop_ref; <<
931 921
932 jh = commit_transaction->t_for 922 jh = commit_transaction->t_forget;
933 spin_unlock(&journal->j_list_l 923 spin_unlock(&journal->j_list_lock);
934 bh = jh2bh(jh); 924 bh = jh2bh(jh);
935 /* 925 /*
936 * Get a reference so that bh 926 * Get a reference so that bh cannot be freed before we are
937 * done with it. 927 * done with it.
938 */ 928 */
939 get_bh(bh); 929 get_bh(bh);
940 spin_lock(&jh->b_state_lock); !! 930 jbd_lock_bh_state(bh);
941 J_ASSERT_JH(jh, jh->b_transact 931 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
942 932
943 /* 933 /*
944 * If there is undo-protected 934 * If there is undo-protected committed data against
945 * this buffer, then we can re 935 * this buffer, then we can remove it now. If it is a
946 * buffer needing such protect 936 * buffer needing such protection, the old frozen_data
947 * field now points to a commi 937 * field now points to a committed version of the
948 * buffer, so rotate that fiel 938 * buffer, so rotate that field to the new committed
949 * data. 939 * data.
950 * 940 *
951 * Otherwise, we can just thro 941 * Otherwise, we can just throw away the frozen data now.
952 * 942 *
953 * We also know that the froze 943 * We also know that the frozen data has already fired
954 * its triggers if they exist, 944 * its triggers if they exist, so we can clear that too.
955 */ 945 */
956 if (jh->b_committed_data) { 946 if (jh->b_committed_data) {
957 jbd2_free(jh->b_commit 947 jbd2_free(jh->b_committed_data, bh->b_size);
958 jh->b_committed_data = 948 jh->b_committed_data = NULL;
959 if (jh->b_frozen_data) 949 if (jh->b_frozen_data) {
960 jh->b_committe 950 jh->b_committed_data = jh->b_frozen_data;
961 jh->b_frozen_d 951 jh->b_frozen_data = NULL;
962 jh->b_frozen_t 952 jh->b_frozen_triggers = NULL;
963 } 953 }
964 } else if (jh->b_frozen_data) 954 } else if (jh->b_frozen_data) {
965 jbd2_free(jh->b_frozen 955 jbd2_free(jh->b_frozen_data, bh->b_size);
966 jh->b_frozen_data = NU 956 jh->b_frozen_data = NULL;
967 jh->b_frozen_triggers 957 jh->b_frozen_triggers = NULL;
968 } 958 }
969 959
970 spin_lock(&journal->j_list_loc 960 spin_lock(&journal->j_list_lock);
971 cp_transaction = jh->b_cp_tran 961 cp_transaction = jh->b_cp_transaction;
972 if (cp_transaction) { 962 if (cp_transaction) {
973 JBUFFER_TRACE(jh, "rem 963 JBUFFER_TRACE(jh, "remove from old cp transaction");
974 cp_transaction->t_chp_ 964 cp_transaction->t_chp_stats.cs_dropped++;
975 __jbd2_journal_remove_ 965 __jbd2_journal_remove_checkpoint(jh);
976 } 966 }
977 967
978 /* Only re-checkpoint the buff 968 /* Only re-checkpoint the buffer_head if it is marked
979 * dirty. If the buffer was a 969 * dirty. If the buffer was added to the BJ_Forget list
980 * by jbd2_journal_forget, it 970 * by jbd2_journal_forget, it may no longer be dirty and
981 * there's no point in keeping 971 * there's no point in keeping a checkpoint record for
982 * it. */ 972 * it. */
983 973
984 /* 974 /*
985 * A buffer which has been fre !! 975 * A buffer which has been freed while still being journaled by
986 * by a previous transaction, !! 976 * a previous transaction.
987 * the running transaction. If !! 977 */
988 * contains "add to orphan" op !! 978 if (buffer_freed(bh)) {
989 * invalidate the buffer now. <<
990 * since the buffer may be sti <<
991 * pagesize and it is attached <<
992 */ <<
993 if (buffer_freed(bh) && !jh->b <<
994 struct address_space * <<
995 <<
996 clear_buffer_freed(bh) <<
997 clear_buffer_jbddirty( <<
998 <<
999 /* 979 /*
1000 * Block device buffe !! 980 * If the running transaction is the one containing
1001 * time, so it is eno !! 981 * "add to orphan" operation (b_next_transaction !=
1002 * buffer_freed bits. !! 982 * NULL), we have to wait for that transaction to
1003 * journalled data) w !! 983 * commit before we can really get rid of the buffer.
1004 * more bits. We also !! 984 * So just clear b_modified to not confuse transaction
1005 * because the data p !! 985 * credit accounting and refile the buffer to
1006 * our hands. Note th !! 986 * BJ_Forget of the running transaction. If the just
1007 * need to make buffe !! 987 * committed transaction contains "add to orphan"
1008 * already detached f !! 988 * operation, we can completely invalidate the buffer
1009 * get reused. !! 989 * now. We are rather through in that since the
>> 990 * buffer may be still accessible when blocksize <
>> 991 * pagesize and it is attached to the last partial
>> 992 * page.
1010 */ 993 */
1011 mapping = READ_ONCE(b !! 994 jh->b_modified = 0;
1012 if (mapping && !sb_is !! 995 if (!jh->b_next_transaction) {
>> 996 clear_buffer_freed(bh);
>> 997 clear_buffer_jbddirty(bh);
1013 clear_buffer_ 998 clear_buffer_mapped(bh);
1014 clear_buffer_ 999 clear_buffer_new(bh);
1015 clear_buffer_ 1000 clear_buffer_req(bh);
1016 bh->b_bdev = 1001 bh->b_bdev = NULL;
1017 } 1002 }
1018 } 1003 }
1019 1004
1020 if (buffer_jbddirty(bh)) { 1005 if (buffer_jbddirty(bh)) {
1021 JBUFFER_TRACE(jh, "ad 1006 JBUFFER_TRACE(jh, "add to new checkpointing trans");
1022 __jbd2_journal_insert 1007 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
1023 if (is_journal_aborte 1008 if (is_journal_aborted(journal))
1024 clear_buffer_ 1009 clear_buffer_jbddirty(bh);
1025 } else { 1010 } else {
1026 J_ASSERT_BH(bh, !buff 1011 J_ASSERT_BH(bh, !buffer_dirty(bh));
1027 /* 1012 /*
1028 * The buffer on BJ_F 1013 * The buffer on BJ_Forget list and not jbddirty means
1029 * it has been freed 1014 * it has been freed by this transaction and hence it
1030 * could not have bee 1015 * could not have been reallocated until this
1031 * transaction has co 1016 * transaction has committed. *BUT* it could be
1032 * reallocated once w 1017 * reallocated once we have written all the data to
1033 * disk and before we 1018 * disk and before we process the buffer on BJ_Forget
1034 * list. 1019 * list.
1035 */ 1020 */
1036 if (!jh->b_next_trans 1021 if (!jh->b_next_transaction)
1037 try_to_free = 1022 try_to_free = 1;
1038 } 1023 }
1039 JBUFFER_TRACE(jh, "refile or 1024 JBUFFER_TRACE(jh, "refile or unfile buffer");
1040 drop_ref = __jbd2_journal_ref !! 1025 __jbd2_journal_refile_buffer(jh);
1041 spin_unlock(&jh->b_state_lock !! 1026 jbd_unlock_bh_state(bh);
1042 if (drop_ref) <<
1043 jbd2_journal_put_jour <<
1044 if (try_to_free) 1027 if (try_to_free)
1045 release_buffer_page(b 1028 release_buffer_page(bh); /* Drops bh reference */
1046 else 1029 else
1047 __brelse(bh); 1030 __brelse(bh);
1048 cond_resched_lock(&journal->j 1031 cond_resched_lock(&journal->j_list_lock);
1049 } 1032 }
1050 spin_unlock(&journal->j_list_lock); 1033 spin_unlock(&journal->j_list_lock);
1051 /* 1034 /*
1052 * This is a bit sleazy. We use j_li 1035 * This is a bit sleazy. We use j_list_lock to protect transition
1053 * of a transaction into T_FINISHED s 1036 * of a transaction into T_FINISHED state and calling
1054 * __jbd2_journal_drop_transaction(). 1037 * __jbd2_journal_drop_transaction(). Otherwise we could race with
1055 * other checkpointing code processin 1038 * other checkpointing code processing the transaction...
1056 */ 1039 */
1057 write_lock(&journal->j_state_lock); 1040 write_lock(&journal->j_state_lock);
1058 spin_lock(&journal->j_list_lock); 1041 spin_lock(&journal->j_list_lock);
1059 /* 1042 /*
1060 * Now recheck if some buffers did no 1043 * Now recheck if some buffers did not get attached to the transaction
1061 * while the lock was dropped... 1044 * while the lock was dropped...
1062 */ 1045 */
1063 if (commit_transaction->t_forget) { 1046 if (commit_transaction->t_forget) {
1064 spin_unlock(&journal->j_list_ 1047 spin_unlock(&journal->j_list_lock);
1065 write_unlock(&journal->j_stat 1048 write_unlock(&journal->j_state_lock);
1066 goto restart_loop; 1049 goto restart_loop;
1067 } 1050 }
1068 1051
1069 /* Add the transaction to the checkpo 1052 /* Add the transaction to the checkpoint list
1070 * __journal_remove_checkpoint() can 1053 * __journal_remove_checkpoint() can not destroy transaction
1071 * under us because it is not marked 1054 * under us because it is not marked as T_FINISHED yet */
1072 if (journal->j_checkpoint_transaction 1055 if (journal->j_checkpoint_transactions == NULL) {
1073 journal->j_checkpoint_transac 1056 journal->j_checkpoint_transactions = commit_transaction;
1074 commit_transaction->t_cpnext 1057 commit_transaction->t_cpnext = commit_transaction;
1075 commit_transaction->t_cpprev 1058 commit_transaction->t_cpprev = commit_transaction;
1076 } else { 1059 } else {
1077 commit_transaction->t_cpnext 1060 commit_transaction->t_cpnext =
1078 journal->j_checkpoint 1061 journal->j_checkpoint_transactions;
1079 commit_transaction->t_cpprev 1062 commit_transaction->t_cpprev =
1080 commit_transaction->t 1063 commit_transaction->t_cpnext->t_cpprev;
1081 commit_transaction->t_cpnext- 1064 commit_transaction->t_cpnext->t_cpprev =
1082 commit_transaction; 1065 commit_transaction;
1083 commit_transaction->t_cpprev- 1066 commit_transaction->t_cpprev->t_cpnext =
1084 commit_transa 1067 commit_transaction;
1085 } 1068 }
1086 spin_unlock(&journal->j_list_lock); 1069 spin_unlock(&journal->j_list_lock);
1087 1070
1088 /* Done with this transaction! */ 1071 /* Done with this transaction! */
1089 1072
1090 jbd2_debug(3, "JBD2: commit phase 7\n !! 1073 jbd_debug(3, "JBD2: commit phase 7\n");
1091 1074
1092 J_ASSERT(commit_transaction->t_state 1075 J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
1093 1076
1094 commit_transaction->t_start = jiffies 1077 commit_transaction->t_start = jiffies;
1095 stats.run.rs_logging = jbd2_time_diff 1078 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
1096 1079 commit_transaction->t_start);
1097 1080
1098 /* 1081 /*
1099 * File the transaction statistics 1082 * File the transaction statistics
1100 */ 1083 */
1101 stats.ts_tid = commit_transaction->t_ 1084 stats.ts_tid = commit_transaction->t_tid;
1102 stats.run.rs_handle_count = 1085 stats.run.rs_handle_count =
1103 atomic_read(&commit_transacti 1086 atomic_read(&commit_transaction->t_handle_count);
1104 trace_jbd2_run_stats(journal->j_fs_de 1087 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
1105 commit_transacti 1088 commit_transaction->t_tid, &stats.run);
1106 stats.ts_requested = (commit_transact 1089 stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0;
1107 1090
1108 commit_transaction->t_state = T_COMMI 1091 commit_transaction->t_state = T_COMMIT_CALLBACK;
1109 J_ASSERT(commit_transaction == journa 1092 J_ASSERT(commit_transaction == journal->j_committing_transaction);
1110 WRITE_ONCE(journal->j_commit_sequence !! 1093 journal->j_commit_sequence = commit_transaction->t_tid;
1111 journal->j_committing_transaction = N 1094 journal->j_committing_transaction = NULL;
1112 commit_time = ktime_to_ns(ktime_sub(k 1095 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1113 1096
1114 /* 1097 /*
1115 * weight the commit time higher than 1098 * weight the commit time higher than the average time so we don't
1116 * react too strongly to vast changes 1099 * react too strongly to vast changes in the commit time
1117 */ 1100 */
1118 if (likely(journal->j_average_commit_ 1101 if (likely(journal->j_average_commit_time))
1119 journal->j_average_commit_tim 1102 journal->j_average_commit_time = (commit_time +
1120 journal->j_av 1103 journal->j_average_commit_time*3) / 4;
1121 else 1104 else
1122 journal->j_average_commit_tim 1105 journal->j_average_commit_time = commit_time;
1123 1106
1124 write_unlock(&journal->j_state_lock); 1107 write_unlock(&journal->j_state_lock);
1125 1108
1126 if (journal->j_commit_callback) 1109 if (journal->j_commit_callback)
1127 journal->j_commit_callback(jo 1110 journal->j_commit_callback(journal, commit_transaction);
1128 if (journal->j_fc_cleanup_callback) <<
1129 journal->j_fc_cleanup_callbac <<
1130 1111
1131 trace_jbd2_end_commit(journal, commit 1112 trace_jbd2_end_commit(journal, commit_transaction);
1132 jbd2_debug(1, "JBD2: commit %d comple !! 1113 jbd_debug(1, "JBD2: commit %d complete, head %d\n",
1133 journal->j_commit_sequence, 1114 journal->j_commit_sequence, journal->j_tail_sequence);
1134 1115
1135 write_lock(&journal->j_state_lock); 1116 write_lock(&journal->j_state_lock);
1136 journal->j_flags &= ~JBD2_FULL_COMMIT <<
1137 journal->j_flags &= ~JBD2_FAST_COMMIT <<
1138 spin_lock(&journal->j_list_lock); 1117 spin_lock(&journal->j_list_lock);
1139 commit_transaction->t_state = T_FINIS 1118 commit_transaction->t_state = T_FINISHED;
1140 /* Check if the transaction can be dr 1119 /* Check if the transaction can be dropped now that we are finished */
1141 if (commit_transaction->t_checkpoint_ !! 1120 if (commit_transaction->t_checkpoint_list == NULL &&
>> 1121 commit_transaction->t_checkpoint_io_list == NULL) {
1142 __jbd2_journal_drop_transacti 1122 __jbd2_journal_drop_transaction(journal, commit_transaction);
1143 jbd2_journal_free_transaction 1123 jbd2_journal_free_transaction(commit_transaction);
1144 } 1124 }
1145 spin_unlock(&journal->j_list_lock); 1125 spin_unlock(&journal->j_list_lock);
1146 write_unlock(&journal->j_state_lock); 1126 write_unlock(&journal->j_state_lock);
1147 wake_up(&journal->j_wait_done_commit) 1127 wake_up(&journal->j_wait_done_commit);
1148 wake_up(&journal->j_fc_wait); <<
1149 1128
1150 /* 1129 /*
1151 * Calculate overall stats 1130 * Calculate overall stats
1152 */ 1131 */
1153 spin_lock(&journal->j_history_lock); 1132 spin_lock(&journal->j_history_lock);
1154 journal->j_stats.ts_tid++; 1133 journal->j_stats.ts_tid++;
1155 journal->j_stats.ts_requested += stat 1134 journal->j_stats.ts_requested += stats.ts_requested;
1156 journal->j_stats.run.rs_wait += stats 1135 journal->j_stats.run.rs_wait += stats.run.rs_wait;
1157 journal->j_stats.run.rs_request_delay 1136 journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
1158 journal->j_stats.run.rs_running += st 1137 journal->j_stats.run.rs_running += stats.run.rs_running;
1159 journal->j_stats.run.rs_locked += sta 1138 journal->j_stats.run.rs_locked += stats.run.rs_locked;
1160 journal->j_stats.run.rs_flushing += s 1139 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
1161 journal->j_stats.run.rs_logging += st 1140 journal->j_stats.run.rs_logging += stats.run.rs_logging;
1162 journal->j_stats.run.rs_handle_count 1141 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
1163 journal->j_stats.run.rs_blocks += sta 1142 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
1164 journal->j_stats.run.rs_blocks_logged 1143 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
1165 spin_unlock(&journal->j_history_lock) 1144 spin_unlock(&journal->j_history_lock);
1166 } 1145 }
1167 1146
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.