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 delayed 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 = {
185 !! 195 .sync_mode = WB_SYNC_ALL,
186 trace_jbd2_submit_inode_data(jinode->i !! 196 .nr_to_write = mapping->nrpages * 2,
187 return journal->j_submit_inode_data_bu !! 197 .range_start = dirty_start,
>> 198 .range_end = dirty_end,
>> 199 };
188 200
>> 201 ret = generic_writepages(mapping, &wbc);
>> 202 return ret;
189 } 203 }
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 } <<
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;
457 454
458 J_ASSERT (atomic_read(&commit_transact 455 J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
459 journal->j_max_transac 456 journal->j_max_transaction_buffers);
460 457
461 /* 458 /*
462 * First thing we are allowed to do is 459 * First thing we are allowed to do is to discard any remaining
463 * BJ_Reserved buffers. Note, it is _ 460 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
464 * that there are no such buffers: if 461 * that there are no such buffers: if a large filesystem
465 * operation like a truncate needs to 462 * operation like a truncate needs to split itself over multiple
466 * transactions, then it may try to do 463 * transactions, then it may try to do a jbd2_journal_restart() while
467 * there are still BJ_Reserved buffers 464 * there are still BJ_Reserved buffers outstanding. These must
468 * be released cleanly from the curren 465 * be released cleanly from the current transaction.
469 * 466 *
470 * In this case, the filesystem must s 467 * In this case, the filesystem must still reserve write access
471 * again before modifying the buffer i 468 * again before modifying the buffer in the new transaction, but
472 * we do not require it to remember ex 469 * we do not require it to remember exactly which old buffers it
473 * has reserved. This is consistent w 470 * has reserved. This is consistent with the existing behaviour
474 * that multiple jbd2_journal_get_writ 471 * that multiple jbd2_journal_get_write_access() calls to the same
475 * buffer are perfectly permissible. 472 * buffer are perfectly permissible.
476 * We use journal->j_state_lock here t 473 * We use journal->j_state_lock here to serialize processing of
477 * t_reserved_list with eviction of bu 474 * t_reserved_list with eviction of buffers from journal_unmap_buffer().
478 */ 475 */
479 while (commit_transaction->t_reserved_ 476 while (commit_transaction->t_reserved_list) {
480 jh = commit_transaction->t_res 477 jh = commit_transaction->t_reserved_list;
481 JBUFFER_TRACE(jh, "reserved, u 478 JBUFFER_TRACE(jh, "reserved, unused: refile");
482 /* 479 /*
483 * A jbd2_journal_get_undo_acc 480 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
484 * leave undo-committed data. 481 * leave undo-committed data.
485 */ 482 */
486 if (jh->b_committed_data) { 483 if (jh->b_committed_data) {
487 struct buffer_head *bh 484 struct buffer_head *bh = jh2bh(jh);
488 485
489 spin_lock(&jh->b_state !! 486 jbd_lock_bh_state(bh);
490 jbd2_free(jh->b_commit 487 jbd2_free(jh->b_committed_data, bh->b_size);
491 jh->b_committed_data = 488 jh->b_committed_data = NULL;
492 spin_unlock(&jh->b_sta !! 489 jbd_unlock_bh_state(bh);
493 } 490 }
494 jbd2_journal_refile_buffer(jou 491 jbd2_journal_refile_buffer(journal, jh);
495 } 492 }
496 493
497 write_unlock(&journal->j_state_lock); 494 write_unlock(&journal->j_state_lock);
498 /* 495 /*
499 * Now try to drop any written-back bu 496 * Now try to drop any written-back buffers from the journal's
500 * checkpoint lists. We do this *befo 497 * checkpoint lists. We do this *before* commit because it potentially
501 * frees some memory 498 * frees some memory
502 */ 499 */
503 spin_lock(&journal->j_list_lock); 500 spin_lock(&journal->j_list_lock);
504 __jbd2_journal_clean_checkpoint_list(j !! 501 __jbd2_journal_clean_checkpoint_list(journal, false);
505 spin_unlock(&journal->j_list_lock); 502 spin_unlock(&journal->j_list_lock);
506 503
507 jbd2_debug(3, "JBD2: commit phase 1\n" !! 504 jbd_debug(3, "JBD2: commit phase 1\n");
508 505
509 /* 506 /*
510 * Clear revoked flag to reflect there 507 * Clear revoked flag to reflect there is no revoked buffers
511 * in the next transaction which is go 508 * in the next transaction which is going to be started.
512 */ 509 */
513 jbd2_clear_buffer_revoked_flags(journa 510 jbd2_clear_buffer_revoked_flags(journal);
514 511
515 /* 512 /*
516 * Switch to a new revoke table. 513 * Switch to a new revoke table.
517 */ 514 */
518 jbd2_journal_switch_revoke_table(journ 515 jbd2_journal_switch_revoke_table(journal);
519 516
520 write_lock(&journal->j_state_lock); 517 write_lock(&journal->j_state_lock);
521 /* 518 /*
522 * Reserved credits cannot be claimed 519 * Reserved credits cannot be claimed anymore, free them
523 */ 520 */
524 atomic_sub(atomic_read(&journal->j_res 521 atomic_sub(atomic_read(&journal->j_reserved_credits),
525 &commit_transaction->t_outs 522 &commit_transaction->t_outstanding_credits);
526 523
527 trace_jbd2_commit_flushing(journal, co 524 trace_jbd2_commit_flushing(journal, commit_transaction);
528 stats.run.rs_flushing = jiffies; 525 stats.run.rs_flushing = jiffies;
529 stats.run.rs_locked = jbd2_time_diff(s 526 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
530 s 527 stats.run.rs_flushing);
531 528
532 commit_transaction->t_state = T_FLUSH; 529 commit_transaction->t_state = T_FLUSH;
533 journal->j_committing_transaction = co 530 journal->j_committing_transaction = commit_transaction;
534 journal->j_running_transaction = NULL; 531 journal->j_running_transaction = NULL;
535 start_time = ktime_get(); 532 start_time = ktime_get();
536 commit_transaction->t_log_start = jour 533 commit_transaction->t_log_start = journal->j_head;
537 wake_up_all(&journal->j_wait_transacti 534 wake_up_all(&journal->j_wait_transaction_locked);
538 write_unlock(&journal->j_state_lock); 535 write_unlock(&journal->j_state_lock);
539 536
540 jbd2_debug(3, "JBD2: commit phase 2a\n !! 537 jbd_debug(3, "JBD2: commit phase 2a\n");
541 538
542 /* 539 /*
543 * Now start flushing things to disk, 540 * Now start flushing things to disk, in the order they appear
544 * on the transaction lists. Data blo 541 * on the transaction lists. Data blocks go first.
545 */ 542 */
546 err = journal_submit_data_buffers(jour 543 err = journal_submit_data_buffers(journal, commit_transaction);
547 if (err) 544 if (err)
548 jbd2_journal_abort(journal, er 545 jbd2_journal_abort(journal, err);
549 546
550 blk_start_plug(&plug); 547 blk_start_plug(&plug);
551 jbd2_journal_write_revoke_records(comm 548 jbd2_journal_write_revoke_records(commit_transaction, &log_bufs);
552 549
553 jbd2_debug(3, "JBD2: commit phase 2b\n !! 550 jbd_debug(3, "JBD2: commit phase 2b\n");
554 551
555 /* 552 /*
556 * Way to go: we have now written out 553 * Way to go: we have now written out all of the data for a
557 * transaction! Now comes the tricky 554 * transaction! Now comes the tricky part: we need to write out
558 * metadata. Loop over the transactio 555 * metadata. Loop over the transaction's entire buffer list:
559 */ 556 */
560 write_lock(&journal->j_state_lock); 557 write_lock(&journal->j_state_lock);
561 commit_transaction->t_state = T_COMMIT 558 commit_transaction->t_state = T_COMMIT;
562 write_unlock(&journal->j_state_lock); 559 write_unlock(&journal->j_state_lock);
563 560
564 trace_jbd2_commit_logging(journal, com 561 trace_jbd2_commit_logging(journal, commit_transaction);
565 stats.run.rs_logging = jiffies; 562 stats.run.rs_logging = jiffies;
566 stats.run.rs_flushing = jbd2_time_diff 563 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
567 564 stats.run.rs_logging);
568 stats.run.rs_blocks = commit_transacti !! 565 stats.run.rs_blocks =
>> 566 atomic_read(&commit_transaction->t_outstanding_credits);
569 stats.run.rs_blocks_logged = 0; 567 stats.run.rs_blocks_logged = 0;
570 568
571 J_ASSERT(commit_transaction->t_nr_buff 569 J_ASSERT(commit_transaction->t_nr_buffers <=
572 atomic_read(&commit_transacti 570 atomic_read(&commit_transaction->t_outstanding_credits));
573 571
>> 572 err = 0;
574 bufs = 0; 573 bufs = 0;
575 descriptor = NULL; 574 descriptor = NULL;
576 while (commit_transaction->t_buffers) 575 while (commit_transaction->t_buffers) {
577 576
578 /* Find the next buffer to be 577 /* Find the next buffer to be journaled... */
579 578
580 jh = commit_transaction->t_buf 579 jh = commit_transaction->t_buffers;
581 580
582 /* If we're in abort mode, we 581 /* If we're in abort mode, we just un-journal the buffer and
583 release it. */ 582 release it. */
584 583
585 if (is_journal_aborted(journal 584 if (is_journal_aborted(journal)) {
586 clear_buffer_jbddirty( 585 clear_buffer_jbddirty(jh2bh(jh));
587 JBUFFER_TRACE(jh, "jou 586 JBUFFER_TRACE(jh, "journal is aborting: refile");
588 jbd2_buffer_abort_trig 587 jbd2_buffer_abort_trigger(jh,
589 588 jh->b_frozen_data ?
590 589 jh->b_frozen_triggers :
591 590 jh->b_triggers);
592 jbd2_journal_refile_bu 591 jbd2_journal_refile_buffer(journal, jh);
593 /* If that was the las 592 /* If that was the last one, we need to clean up
594 * any descriptor buff 593 * any descriptor buffers which may have been
595 * already allocated, 594 * already allocated, even if we are now
596 * aborting. */ 595 * aborting. */
597 if (!commit_transactio 596 if (!commit_transaction->t_buffers)
598 goto start_jou 597 goto start_journal_io;
599 continue; 598 continue;
600 } 599 }
601 600
602 /* Make sure we have a descrip 601 /* Make sure we have a descriptor block in which to
603 record the metadata buffer. 602 record the metadata buffer. */
604 603
605 if (!descriptor) { 604 if (!descriptor) {
606 J_ASSERT (bufs == 0); 605 J_ASSERT (bufs == 0);
607 606
608 jbd2_debug(4, "JBD2: g !! 607 jbd_debug(4, "JBD2: get descriptor\n");
609 608
610 descriptor = jbd2_jour 609 descriptor = jbd2_journal_get_descriptor_buffer(
611 610 commit_transaction,
612 611 JBD2_DESCRIPTOR_BLOCK);
613 if (!descriptor) { 612 if (!descriptor) {
614 jbd2_journal_a 613 jbd2_journal_abort(journal, -EIO);
615 continue; 614 continue;
616 } 615 }
617 616
618 jbd2_debug(4, "JBD2: g !! 617 jbd_debug(4, "JBD2: got buffer %llu (%p)\n",
619 (unsigned long 618 (unsigned long long)descriptor->b_blocknr,
620 descriptor->b_ 619 descriptor->b_data);
621 tagp = &descriptor->b_ 620 tagp = &descriptor->b_data[sizeof(journal_header_t)];
622 space_left = descripto 621 space_left = descriptor->b_size -
623 622 sizeof(journal_header_t);
624 first_tag = 1; 623 first_tag = 1;
625 set_buffer_jwrite(desc 624 set_buffer_jwrite(descriptor);
626 set_buffer_dirty(descr 625 set_buffer_dirty(descriptor);
627 wbuf[bufs++] = descrip 626 wbuf[bufs++] = descriptor;
628 627
629 /* Record it so that w 628 /* Record it so that we can wait for IO
630 completion later */ 629 completion later */
631 BUFFER_TRACE(descripto 630 BUFFER_TRACE(descriptor, "ph3: file as descriptor");
632 jbd2_file_log_bh(&log_ 631 jbd2_file_log_bh(&log_bufs, descriptor);
633 } 632 }
634 633
635 /* Where is the buffer to be w 634 /* Where is the buffer to be written? */
636 635
637 err = jbd2_journal_next_log_bl 636 err = jbd2_journal_next_log_block(journal, &blocknr);
638 /* If the block mapping failed 637 /* If the block mapping failed, just abandon the buffer
639 and repeat this loop: we'll 638 and repeat this loop: we'll fall into the
640 refile-on-abort condition a 639 refile-on-abort condition above. */
641 if (err) { 640 if (err) {
642 jbd2_journal_abort(jou 641 jbd2_journal_abort(journal, err);
643 continue; 642 continue;
644 } 643 }
645 644
646 /* 645 /*
647 * start_this_handle() uses t_ 646 * start_this_handle() uses t_outstanding_credits to determine
648 * the free space in the log. !! 647 * the free space in the log, but this counter is changed
>> 648 * by jbd2_journal_next_log_block() also.
649 */ 649 */
650 atomic_dec(&commit_transaction 650 atomic_dec(&commit_transaction->t_outstanding_credits);
651 651
652 /* Bump b_count to prevent tru 652 /* Bump b_count to prevent truncate from stumbling over
653 the shadowed buffer! @@@ T 653 the shadowed buffer! @@@ This can go if we ever get
654 rid of the shadow pairing o 654 rid of the shadow pairing of buffers. */
655 atomic_inc(&jh2bh(jh)->b_count 655 atomic_inc(&jh2bh(jh)->b_count);
656 656
657 /* 657 /*
658 * Make a temporary IO buffer 658 * Make a temporary IO buffer with which to write it out
659 * (this will requeue the meta 659 * (this will requeue the metadata buffer to BJ_Shadow).
660 */ 660 */
661 set_bit(BH_JWrite, &jh2bh(jh)- 661 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
662 JBUFFER_TRACE(jh, "ph3: write 662 JBUFFER_TRACE(jh, "ph3: write metadata");
663 escape = jbd2_journal_write_me !! 663 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
664 664 jh, &wbuf[bufs], blocknr);
665 if (escape < 0) { !! 665 if (flags < 0) {
666 jbd2_journal_abort(jou !! 666 jbd2_journal_abort(journal, flags);
667 continue; 667 continue;
668 } 668 }
669 jbd2_file_log_bh(&io_bufs, wbu 669 jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
670 670
671 /* Record the new block's tag 671 /* Record the new block's tag in the current descriptor
672 buffer */ 672 buffer */
673 673
674 tag_flag = 0; 674 tag_flag = 0;
675 if (escape) !! 675 if (flags & 1)
676 tag_flag |= JBD2_FLAG_ 676 tag_flag |= JBD2_FLAG_ESCAPE;
677 if (!first_tag) 677 if (!first_tag)
678 tag_flag |= JBD2_FLAG_ 678 tag_flag |= JBD2_FLAG_SAME_UUID;
679 679
680 tag = (journal_block_tag_t *) 680 tag = (journal_block_tag_t *) tagp;
681 write_tag_block(journal, tag, 681 write_tag_block(journal, tag, jh2bh(jh)->b_blocknr);
682 tag->t_flags = cpu_to_be16(tag 682 tag->t_flags = cpu_to_be16(tag_flag);
683 jbd2_block_tag_csum_set(journa 683 jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
684 commit 684 commit_transaction->t_tid);
685 tagp += tag_bytes; 685 tagp += tag_bytes;
686 space_left -= tag_bytes; 686 space_left -= tag_bytes;
687 bufs++; 687 bufs++;
688 688
689 if (first_tag) { 689 if (first_tag) {
690 memcpy (tagp, journal- 690 memcpy (tagp, journal->j_uuid, 16);
691 tagp += 16; 691 tagp += 16;
692 space_left -= 16; 692 space_left -= 16;
693 first_tag = 0; 693 first_tag = 0;
694 } 694 }
695 695
696 /* If there's no more to do, o 696 /* If there's no more to do, or if the descriptor is full,
697 let the IO rip! */ 697 let the IO rip! */
698 698
699 if (bufs == journal->j_wbufsiz 699 if (bufs == journal->j_wbufsize ||
700 commit_transaction->t_buff 700 commit_transaction->t_buffers == NULL ||
701 space_left < tag_bytes + 1 701 space_left < tag_bytes + 16 + csum_size) {
702 702
703 jbd2_debug(4, "JBD2: S !! 703 jbd_debug(4, "JBD2: Submit %d IOs\n", bufs);
704 704
705 /* Write an end-of-des 705 /* Write an end-of-descriptor marker before
706 submitting the IOs. 706 submitting the IOs. "tag" still points to
707 the last tag we set 707 the last tag we set up. */
708 708
709 tag->t_flags |= cpu_to 709 tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
710 start_journal_io: 710 start_journal_io:
711 if (descriptor) 711 if (descriptor)
712 jbd2_descripto 712 jbd2_descriptor_block_csum_set(journal,
713 713 descriptor);
714 714
715 for (i = 0; i < bufs; 715 for (i = 0; i < bufs; i++) {
716 struct buffer_ 716 struct buffer_head *bh = wbuf[i];
717 <<
718 /* 717 /*
719 * Compute che 718 * Compute checksum.
720 */ 719 */
721 if (jbd2_has_f 720 if (jbd2_has_feature_checksum(journal)) {
722 crc32_ 721 crc32_sum =
723 jb 722 jbd2_checksum_data(crc32_sum, bh);
724 } 723 }
725 724
726 lock_buffer(bh 725 lock_buffer(bh);
727 clear_buffer_d 726 clear_buffer_dirty(bh);
728 set_buffer_upt 727 set_buffer_uptodate(bh);
729 bh->b_end_io = 728 bh->b_end_io = journal_end_buffer_io_sync;
730 submit_bh(REQ_ !! 729 submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
731 bh); <<
732 } 730 }
733 cond_resched(); 731 cond_resched();
734 732
735 /* Force a new descrip 733 /* Force a new descriptor to be generated next
736 time round the loop 734 time round the loop. */
737 descriptor = NULL; 735 descriptor = NULL;
738 bufs = 0; 736 bufs = 0;
739 } 737 }
740 } 738 }
741 739
742 err = journal_finish_inode_data_buffer 740 err = journal_finish_inode_data_buffers(journal, commit_transaction);
743 if (err) { 741 if (err) {
744 printk(KERN_WARNING 742 printk(KERN_WARNING
745 "JBD2: Detected IO err 743 "JBD2: Detected IO errors while flushing file data "
746 "on %s\n", journal->j_d 744 "on %s\n", journal->j_devname);
747 if (journal->j_flags & JBD2_AB 745 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
748 jbd2_journal_abort(jou 746 jbd2_journal_abort(journal, err);
749 err = 0; 747 err = 0;
750 } 748 }
751 749
752 /* 750 /*
753 * Get current oldest transaction in t 751 * Get current oldest transaction in the log before we issue flush
754 * to the filesystem device. After the 752 * to the filesystem device. After the flush we can be sure that
755 * blocks of all older transactions ar 753 * blocks of all older transactions are checkpointed to persistent
756 * storage and we will be safe to upda 754 * storage and we will be safe to update journal start in the
757 * superblock with the numbers we get 755 * superblock with the numbers we get here.
758 */ 756 */
759 update_tail = 757 update_tail =
760 jbd2_journal_get_log_tail(jour 758 jbd2_journal_get_log_tail(journal, &first_tid, &first_block);
761 759
762 write_lock(&journal->j_state_lock); 760 write_lock(&journal->j_state_lock);
763 if (update_tail) { 761 if (update_tail) {
764 long freed = first_block - jou 762 long freed = first_block - journal->j_tail;
765 763
766 if (first_block < journal->j_t 764 if (first_block < journal->j_tail)
767 freed += journal->j_la 765 freed += journal->j_last - journal->j_first;
768 /* Update tail only if we free 766 /* Update tail only if we free significant amount of space */
769 if (freed < journal->j_max_tra !! 767 if (freed < journal->j_maxlen / 4)
770 update_tail = 0; 768 update_tail = 0;
771 } 769 }
772 J_ASSERT(commit_transaction->t_state = 770 J_ASSERT(commit_transaction->t_state == T_COMMIT);
773 commit_transaction->t_state = T_COMMIT 771 commit_transaction->t_state = T_COMMIT_DFLUSH;
774 write_unlock(&journal->j_state_lock); 772 write_unlock(&journal->j_state_lock);
775 773
776 /* !! 774 /*
777 * If the journal is not located on th 775 * If the journal is not located on the file system device,
778 * then we must flush the file system 776 * then we must flush the file system device before we issue
779 * the commit record 777 * the commit record
780 */ 778 */
781 if (commit_transaction->t_need_data_fl 779 if (commit_transaction->t_need_data_flush &&
782 (journal->j_fs_dev != journal->j_d 780 (journal->j_fs_dev != journal->j_dev) &&
783 (journal->j_flags & JBD2_BARRIER)) 781 (journal->j_flags & JBD2_BARRIER))
784 blkdev_issue_flush(journal->j_ !! 782 blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL);
785 783
786 /* Done it all: now write the commit r 784 /* Done it all: now write the commit record asynchronously. */
787 if (jbd2_has_feature_async_commit(jour 785 if (jbd2_has_feature_async_commit(journal)) {
788 err = journal_submit_commit_re 786 err = journal_submit_commit_record(journal, commit_transaction,
789 787 &cbh, crc32_sum);
790 if (err) 788 if (err)
791 jbd2_journal_abort(jou 789 jbd2_journal_abort(journal, err);
792 } 790 }
793 791
794 blk_finish_plug(&plug); 792 blk_finish_plug(&plug);
795 793
796 /* Lo and behold: we have just managed 794 /* Lo and behold: we have just managed to send a transaction to
797 the log. Before we can commit it, 795 the log. Before we can commit it, wait for the IO so far to
798 complete. Control buffers being wr 796 complete. Control buffers being written are on the
799 transaction's t_log_list queue, and 797 transaction's t_log_list queue, and metadata buffers are on
800 the io_bufs list. 798 the io_bufs list.
801 799
802 Wait for the buffers in reverse ord 800 Wait for the buffers in reverse order. That way we are
803 less likely to be woken up until al 801 less likely to be woken up until all IOs have completed, and
804 so we incur less scheduling load. 802 so we incur less scheduling load.
805 */ 803 */
806 804
807 jbd2_debug(3, "JBD2: commit phase 3\n" !! 805 jbd_debug(3, "JBD2: commit phase 3\n");
808 806
809 while (!list_empty(&io_bufs)) { 807 while (!list_empty(&io_bufs)) {
810 struct buffer_head *bh = list_ 808 struct buffer_head *bh = list_entry(io_bufs.prev,
811 809 struct buffer_head,
812 810 b_assoc_buffers);
813 811
814 wait_on_buffer(bh); 812 wait_on_buffer(bh);
815 cond_resched(); 813 cond_resched();
816 814
817 if (unlikely(!buffer_uptodate( 815 if (unlikely(!buffer_uptodate(bh)))
818 err = -EIO; 816 err = -EIO;
819 jbd2_unfile_log_bh(bh); 817 jbd2_unfile_log_bh(bh);
820 stats.run.rs_blocks_logged++; 818 stats.run.rs_blocks_logged++;
821 819
822 /* 820 /*
823 * The list contains temporary 821 * The list contains temporary buffer heads created by
824 * jbd2_journal_write_metadata 822 * jbd2_journal_write_metadata_buffer().
825 */ 823 */
826 BUFFER_TRACE(bh, "dumping temp 824 BUFFER_TRACE(bh, "dumping temporary bh");
827 __brelse(bh); 825 __brelse(bh);
828 J_ASSERT_BH(bh, atomic_read(&b 826 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
829 free_buffer_head(bh); 827 free_buffer_head(bh);
830 828
831 /* We also have to refile the 829 /* We also have to refile the corresponding shadowed buffer */
832 jh = commit_transaction->t_sha 830 jh = commit_transaction->t_shadow_list->b_tprev;
833 bh = jh2bh(jh); 831 bh = jh2bh(jh);
834 clear_buffer_jwrite(bh); 832 clear_buffer_jwrite(bh);
835 J_ASSERT_BH(bh, buffer_jbddirt 833 J_ASSERT_BH(bh, buffer_jbddirty(bh));
836 J_ASSERT_BH(bh, !buffer_shadow 834 J_ASSERT_BH(bh, !buffer_shadow(bh));
837 835
838 /* The metadata is now release 836 /* The metadata is now released for reuse, but we need
839 to remember it against this 837 to remember it against this transaction so that when
840 we finally commit, we can d 838 we finally commit, we can do any checkpointing
841 required. */ 839 required. */
842 JBUFFER_TRACE(jh, "file as BJ_ 840 JBUFFER_TRACE(jh, "file as BJ_Forget");
843 jbd2_journal_file_buffer(jh, c 841 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
844 JBUFFER_TRACE(jh, "brelse shad 842 JBUFFER_TRACE(jh, "brelse shadowed buffer");
845 __brelse(bh); 843 __brelse(bh);
846 } 844 }
847 845
848 J_ASSERT (commit_transaction->t_shadow 846 J_ASSERT (commit_transaction->t_shadow_list == NULL);
849 847
850 jbd2_debug(3, "JBD2: commit phase 4\n" !! 848 jbd_debug(3, "JBD2: commit phase 4\n");
851 849
852 /* Here we wait for the revoke record 850 /* Here we wait for the revoke record and descriptor record buffers */
853 while (!list_empty(&log_bufs)) { 851 while (!list_empty(&log_bufs)) {
854 struct buffer_head *bh; 852 struct buffer_head *bh;
855 853
856 bh = list_entry(log_bufs.prev, 854 bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
857 wait_on_buffer(bh); 855 wait_on_buffer(bh);
858 cond_resched(); 856 cond_resched();
859 857
860 if (unlikely(!buffer_uptodate( 858 if (unlikely(!buffer_uptodate(bh)))
861 err = -EIO; 859 err = -EIO;
862 860
863 BUFFER_TRACE(bh, "ph5: control 861 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
864 clear_buffer_jwrite(bh); 862 clear_buffer_jwrite(bh);
865 jbd2_unfile_log_bh(bh); 863 jbd2_unfile_log_bh(bh);
866 stats.run.rs_blocks_logged++; 864 stats.run.rs_blocks_logged++;
867 __brelse(bh); /* One 865 __brelse(bh); /* One for getblk */
868 /* AKPM: bforget here */ 866 /* AKPM: bforget here */
869 } 867 }
870 868
871 if (err) 869 if (err)
872 jbd2_journal_abort(journal, er 870 jbd2_journal_abort(journal, err);
873 871
874 jbd2_debug(3, "JBD2: commit phase 5\n" !! 872 jbd_debug(3, "JBD2: commit phase 5\n");
875 write_lock(&journal->j_state_lock); 873 write_lock(&journal->j_state_lock);
876 J_ASSERT(commit_transaction->t_state = 874 J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
877 commit_transaction->t_state = T_COMMIT 875 commit_transaction->t_state = T_COMMIT_JFLUSH;
878 write_unlock(&journal->j_state_lock); 876 write_unlock(&journal->j_state_lock);
879 877
880 if (!jbd2_has_feature_async_commit(jou 878 if (!jbd2_has_feature_async_commit(journal)) {
881 err = journal_submit_commit_re 879 err = journal_submit_commit_record(journal, commit_transaction,
882 880 &cbh, crc32_sum);
883 if (err) 881 if (err)
884 jbd2_journal_abort(jou 882 jbd2_journal_abort(journal, err);
885 } 883 }
886 if (cbh) 884 if (cbh)
887 err = journal_wait_on_commit_r 885 err = journal_wait_on_commit_record(journal, cbh);
888 stats.run.rs_blocks_logged++; 886 stats.run.rs_blocks_logged++;
889 if (jbd2_has_feature_async_commit(jour 887 if (jbd2_has_feature_async_commit(journal) &&
890 journal->j_flags & JBD2_BARRIER) { 888 journal->j_flags & JBD2_BARRIER) {
891 blkdev_issue_flush(journal->j_ !! 889 blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL);
892 } 890 }
893 891
894 if (err) 892 if (err)
895 jbd2_journal_abort(journal, er 893 jbd2_journal_abort(journal, err);
896 894
897 WARN_ON_ONCE( <<
898 atomic_read(&commit_transactio <<
899 <<
900 /* 895 /*
901 * Now disk caches for filesystem devi 896 * Now disk caches for filesystem device are flushed so we are safe to
902 * erase checkpointed transactions fro 897 * erase checkpointed transactions from the log by updating journal
903 * superblock. 898 * superblock.
904 */ 899 */
905 if (update_tail) 900 if (update_tail)
906 jbd2_update_log_tail(journal, 901 jbd2_update_log_tail(journal, first_tid, first_block);
907 902
908 /* End of a transaction! Finally, we 903 /* End of a transaction! Finally, we can do checkpoint
909 processing: any buffers committed a 904 processing: any buffers committed as a result of this
910 transaction can be removed from any 905 transaction can be removed from any checkpoint list it was on
911 before. */ 906 before. */
912 907
913 jbd2_debug(3, "JBD2: commit phase 6\n" !! 908 jbd_debug(3, "JBD2: commit phase 6\n");
914 909
915 J_ASSERT(list_empty(&commit_transactio 910 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
916 J_ASSERT(commit_transaction->t_buffers 911 J_ASSERT(commit_transaction->t_buffers == NULL);
917 J_ASSERT(commit_transaction->t_checkpo 912 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
918 J_ASSERT(commit_transaction->t_shadow_ 913 J_ASSERT(commit_transaction->t_shadow_list == NULL);
919 914
920 restart_loop: 915 restart_loop:
921 /* 916 /*
922 * As there are other places (journal_ 917 * As there are other places (journal_unmap_buffer()) adding buffers
923 * to this list we have to be careful 918 * to this list we have to be careful and hold the j_list_lock.
924 */ 919 */
925 spin_lock(&journal->j_list_lock); 920 spin_lock(&journal->j_list_lock);
926 while (commit_transaction->t_forget) { 921 while (commit_transaction->t_forget) {
927 transaction_t *cp_transaction; 922 transaction_t *cp_transaction;
928 struct buffer_head *bh; 923 struct buffer_head *bh;
929 int try_to_free = 0; 924 int try_to_free = 0;
930 bool drop_ref; <<
931 925
932 jh = commit_transaction->t_for 926 jh = commit_transaction->t_forget;
933 spin_unlock(&journal->j_list_l 927 spin_unlock(&journal->j_list_lock);
934 bh = jh2bh(jh); 928 bh = jh2bh(jh);
935 /* 929 /*
936 * Get a reference so that bh 930 * Get a reference so that bh cannot be freed before we are
937 * done with it. 931 * done with it.
938 */ 932 */
939 get_bh(bh); 933 get_bh(bh);
940 spin_lock(&jh->b_state_lock); !! 934 jbd_lock_bh_state(bh);
941 J_ASSERT_JH(jh, jh->b_transact 935 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
942 936
943 /* 937 /*
944 * If there is undo-protected 938 * If there is undo-protected committed data against
945 * this buffer, then we can re 939 * this buffer, then we can remove it now. If it is a
946 * buffer needing such protect 940 * buffer needing such protection, the old frozen_data
947 * field now points to a commi 941 * field now points to a committed version of the
948 * buffer, so rotate that fiel 942 * buffer, so rotate that field to the new committed
949 * data. 943 * data.
950 * 944 *
951 * Otherwise, we can just thro 945 * Otherwise, we can just throw away the frozen data now.
952 * 946 *
953 * We also know that the froze 947 * We also know that the frozen data has already fired
954 * its triggers if they exist, 948 * its triggers if they exist, so we can clear that too.
955 */ 949 */
956 if (jh->b_committed_data) { 950 if (jh->b_committed_data) {
957 jbd2_free(jh->b_commit 951 jbd2_free(jh->b_committed_data, bh->b_size);
958 jh->b_committed_data = 952 jh->b_committed_data = NULL;
959 if (jh->b_frozen_data) 953 if (jh->b_frozen_data) {
960 jh->b_committe 954 jh->b_committed_data = jh->b_frozen_data;
961 jh->b_frozen_d 955 jh->b_frozen_data = NULL;
962 jh->b_frozen_t 956 jh->b_frozen_triggers = NULL;
963 } 957 }
964 } else if (jh->b_frozen_data) 958 } else if (jh->b_frozen_data) {
965 jbd2_free(jh->b_frozen 959 jbd2_free(jh->b_frozen_data, bh->b_size);
966 jh->b_frozen_data = NU 960 jh->b_frozen_data = NULL;
967 jh->b_frozen_triggers 961 jh->b_frozen_triggers = NULL;
968 } 962 }
969 963
970 spin_lock(&journal->j_list_loc 964 spin_lock(&journal->j_list_lock);
971 cp_transaction = jh->b_cp_tran 965 cp_transaction = jh->b_cp_transaction;
972 if (cp_transaction) { 966 if (cp_transaction) {
973 JBUFFER_TRACE(jh, "rem 967 JBUFFER_TRACE(jh, "remove from old cp transaction");
974 cp_transaction->t_chp_ 968 cp_transaction->t_chp_stats.cs_dropped++;
975 __jbd2_journal_remove_ 969 __jbd2_journal_remove_checkpoint(jh);
976 } 970 }
977 971
978 /* Only re-checkpoint the buff 972 /* Only re-checkpoint the buffer_head if it is marked
979 * dirty. If the buffer was a 973 * dirty. If the buffer was added to the BJ_Forget list
980 * by jbd2_journal_forget, it 974 * by jbd2_journal_forget, it may no longer be dirty and
981 * there's no point in keeping 975 * there's no point in keeping a checkpoint record for
982 * it. */ 976 * it. */
983 977
984 /* 978 /*
985 * A buffer which has been fre 979 * A buffer which has been freed while still being journaled
986 * by a previous transaction, 980 * by a previous transaction, refile the buffer to BJ_Forget of
987 * the running transaction. If 981 * the running transaction. If the just committed transaction
988 * contains "add to orphan" op 982 * contains "add to orphan" operation, we can completely
989 * invalidate the buffer now. 983 * invalidate the buffer now. We are rather through in that
990 * since the buffer may be sti 984 * since the buffer may be still accessible when blocksize <
991 * pagesize and it is attached 985 * pagesize and it is attached to the last partial page.
992 */ 986 */
993 if (buffer_freed(bh) && !jh->b 987 if (buffer_freed(bh) && !jh->b_next_transaction) {
994 struct address_space * 988 struct address_space *mapping;
995 989
996 clear_buffer_freed(bh) 990 clear_buffer_freed(bh);
997 clear_buffer_jbddirty( 991 clear_buffer_jbddirty(bh);
998 992
999 /* 993 /*
1000 * Block device buffe 994 * Block device buffers need to stay mapped all the
1001 * time, so it is eno 995 * time, so it is enough to clear buffer_jbddirty and
1002 * buffer_freed bits. 996 * buffer_freed bits. For the file mapping buffers (i.e.
1003 * journalled data) w 997 * journalled data) we need to unmap buffer and clear
1004 * more bits. We also 998 * more bits. We also need to be careful about the check
1005 * because the data p 999 * because the data page mapping can get cleared under
1006 * our hands. Note th 1000 * our hands. Note that if mapping == NULL, we don't
1007 * need to make buffe 1001 * need to make buffer unmapped because the page is
1008 * already detached f 1002 * already detached from the mapping and buffers cannot
1009 * get reused. 1003 * get reused.
1010 */ 1004 */
1011 mapping = READ_ONCE(b !! 1005 mapping = READ_ONCE(bh->b_page->mapping);
1012 if (mapping && !sb_is 1006 if (mapping && !sb_is_blkdev_sb(mapping->host->i_sb)) {
1013 clear_buffer_ 1007 clear_buffer_mapped(bh);
1014 clear_buffer_ 1008 clear_buffer_new(bh);
1015 clear_buffer_ 1009 clear_buffer_req(bh);
1016 bh->b_bdev = 1010 bh->b_bdev = NULL;
1017 } 1011 }
1018 } 1012 }
1019 1013
1020 if (buffer_jbddirty(bh)) { 1014 if (buffer_jbddirty(bh)) {
1021 JBUFFER_TRACE(jh, "ad 1015 JBUFFER_TRACE(jh, "add to new checkpointing trans");
1022 __jbd2_journal_insert 1016 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
1023 if (is_journal_aborte 1017 if (is_journal_aborted(journal))
1024 clear_buffer_ 1018 clear_buffer_jbddirty(bh);
1025 } else { 1019 } else {
1026 J_ASSERT_BH(bh, !buff 1020 J_ASSERT_BH(bh, !buffer_dirty(bh));
1027 /* 1021 /*
1028 * The buffer on BJ_F 1022 * The buffer on BJ_Forget list and not jbddirty means
1029 * it has been freed 1023 * it has been freed by this transaction and hence it
1030 * could not have bee 1024 * could not have been reallocated until this
1031 * transaction has co 1025 * transaction has committed. *BUT* it could be
1032 * reallocated once w 1026 * reallocated once we have written all the data to
1033 * disk and before we 1027 * disk and before we process the buffer on BJ_Forget
1034 * list. 1028 * list.
1035 */ 1029 */
1036 if (!jh->b_next_trans 1030 if (!jh->b_next_transaction)
1037 try_to_free = 1031 try_to_free = 1;
1038 } 1032 }
1039 JBUFFER_TRACE(jh, "refile or 1033 JBUFFER_TRACE(jh, "refile or unfile buffer");
1040 drop_ref = __jbd2_journal_ref !! 1034 __jbd2_journal_refile_buffer(jh);
1041 spin_unlock(&jh->b_state_lock !! 1035 jbd_unlock_bh_state(bh);
1042 if (drop_ref) <<
1043 jbd2_journal_put_jour <<
1044 if (try_to_free) 1036 if (try_to_free)
1045 release_buffer_page(b 1037 release_buffer_page(bh); /* Drops bh reference */
1046 else 1038 else
1047 __brelse(bh); 1039 __brelse(bh);
1048 cond_resched_lock(&journal->j 1040 cond_resched_lock(&journal->j_list_lock);
1049 } 1041 }
1050 spin_unlock(&journal->j_list_lock); 1042 spin_unlock(&journal->j_list_lock);
1051 /* 1043 /*
1052 * This is a bit sleazy. We use j_li 1044 * This is a bit sleazy. We use j_list_lock to protect transition
1053 * of a transaction into T_FINISHED s 1045 * of a transaction into T_FINISHED state and calling
1054 * __jbd2_journal_drop_transaction(). 1046 * __jbd2_journal_drop_transaction(). Otherwise we could race with
1055 * other checkpointing code processin 1047 * other checkpointing code processing the transaction...
1056 */ 1048 */
1057 write_lock(&journal->j_state_lock); 1049 write_lock(&journal->j_state_lock);
1058 spin_lock(&journal->j_list_lock); 1050 spin_lock(&journal->j_list_lock);
1059 /* 1051 /*
1060 * Now recheck if some buffers did no 1052 * Now recheck if some buffers did not get attached to the transaction
1061 * while the lock was dropped... 1053 * while the lock was dropped...
1062 */ 1054 */
1063 if (commit_transaction->t_forget) { 1055 if (commit_transaction->t_forget) {
1064 spin_unlock(&journal->j_list_ 1056 spin_unlock(&journal->j_list_lock);
1065 write_unlock(&journal->j_stat 1057 write_unlock(&journal->j_state_lock);
1066 goto restart_loop; 1058 goto restart_loop;
1067 } 1059 }
1068 1060
1069 /* Add the transaction to the checkpo 1061 /* Add the transaction to the checkpoint list
1070 * __journal_remove_checkpoint() can 1062 * __journal_remove_checkpoint() can not destroy transaction
1071 * under us because it is not marked 1063 * under us because it is not marked as T_FINISHED yet */
1072 if (journal->j_checkpoint_transaction 1064 if (journal->j_checkpoint_transactions == NULL) {
1073 journal->j_checkpoint_transac 1065 journal->j_checkpoint_transactions = commit_transaction;
1074 commit_transaction->t_cpnext 1066 commit_transaction->t_cpnext = commit_transaction;
1075 commit_transaction->t_cpprev 1067 commit_transaction->t_cpprev = commit_transaction;
1076 } else { 1068 } else {
1077 commit_transaction->t_cpnext 1069 commit_transaction->t_cpnext =
1078 journal->j_checkpoint 1070 journal->j_checkpoint_transactions;
1079 commit_transaction->t_cpprev 1071 commit_transaction->t_cpprev =
1080 commit_transaction->t 1072 commit_transaction->t_cpnext->t_cpprev;
1081 commit_transaction->t_cpnext- 1073 commit_transaction->t_cpnext->t_cpprev =
1082 commit_transaction; 1074 commit_transaction;
1083 commit_transaction->t_cpprev- 1075 commit_transaction->t_cpprev->t_cpnext =
1084 commit_transa 1076 commit_transaction;
1085 } 1077 }
1086 spin_unlock(&journal->j_list_lock); 1078 spin_unlock(&journal->j_list_lock);
1087 1079
1088 /* Done with this transaction! */ 1080 /* Done with this transaction! */
1089 1081
1090 jbd2_debug(3, "JBD2: commit phase 7\n !! 1082 jbd_debug(3, "JBD2: commit phase 7\n");
1091 1083
1092 J_ASSERT(commit_transaction->t_state 1084 J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
1093 1085
1094 commit_transaction->t_start = jiffies 1086 commit_transaction->t_start = jiffies;
1095 stats.run.rs_logging = jbd2_time_diff 1087 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
1096 1088 commit_transaction->t_start);
1097 1089
1098 /* 1090 /*
1099 * File the transaction statistics 1091 * File the transaction statistics
1100 */ 1092 */
1101 stats.ts_tid = commit_transaction->t_ 1093 stats.ts_tid = commit_transaction->t_tid;
1102 stats.run.rs_handle_count = 1094 stats.run.rs_handle_count =
1103 atomic_read(&commit_transacti 1095 atomic_read(&commit_transaction->t_handle_count);
1104 trace_jbd2_run_stats(journal->j_fs_de 1096 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
1105 commit_transacti 1097 commit_transaction->t_tid, &stats.run);
1106 stats.ts_requested = (commit_transact 1098 stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0;
1107 1099
1108 commit_transaction->t_state = T_COMMI 1100 commit_transaction->t_state = T_COMMIT_CALLBACK;
1109 J_ASSERT(commit_transaction == journa 1101 J_ASSERT(commit_transaction == journal->j_committing_transaction);
1110 WRITE_ONCE(journal->j_commit_sequence !! 1102 journal->j_commit_sequence = commit_transaction->t_tid;
1111 journal->j_committing_transaction = N 1103 journal->j_committing_transaction = NULL;
1112 commit_time = ktime_to_ns(ktime_sub(k 1104 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1113 1105
1114 /* 1106 /*
1115 * weight the commit time higher than 1107 * weight the commit time higher than the average time so we don't
1116 * react too strongly to vast changes 1108 * react too strongly to vast changes in the commit time
1117 */ 1109 */
1118 if (likely(journal->j_average_commit_ 1110 if (likely(journal->j_average_commit_time))
1119 journal->j_average_commit_tim 1111 journal->j_average_commit_time = (commit_time +
1120 journal->j_av 1112 journal->j_average_commit_time*3) / 4;
1121 else 1113 else
1122 journal->j_average_commit_tim 1114 journal->j_average_commit_time = commit_time;
1123 1115
1124 write_unlock(&journal->j_state_lock); 1116 write_unlock(&journal->j_state_lock);
1125 1117
1126 if (journal->j_commit_callback) 1118 if (journal->j_commit_callback)
1127 journal->j_commit_callback(jo 1119 journal->j_commit_callback(journal, commit_transaction);
1128 if (journal->j_fc_cleanup_callback) <<
1129 journal->j_fc_cleanup_callbac <<
1130 1120
1131 trace_jbd2_end_commit(journal, commit 1121 trace_jbd2_end_commit(journal, commit_transaction);
1132 jbd2_debug(1, "JBD2: commit %d comple !! 1122 jbd_debug(1, "JBD2: commit %d complete, head %d\n",
1133 journal->j_commit_sequence, 1123 journal->j_commit_sequence, journal->j_tail_sequence);
1134 1124
1135 write_lock(&journal->j_state_lock); 1125 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); 1126 spin_lock(&journal->j_list_lock);
1139 commit_transaction->t_state = T_FINIS 1127 commit_transaction->t_state = T_FINISHED;
1140 /* Check if the transaction can be dr 1128 /* Check if the transaction can be dropped now that we are finished */
1141 if (commit_transaction->t_checkpoint_ !! 1129 if (commit_transaction->t_checkpoint_list == NULL &&
>> 1130 commit_transaction->t_checkpoint_io_list == NULL) {
1142 __jbd2_journal_drop_transacti 1131 __jbd2_journal_drop_transaction(journal, commit_transaction);
1143 jbd2_journal_free_transaction 1132 jbd2_journal_free_transaction(commit_transaction);
1144 } 1133 }
1145 spin_unlock(&journal->j_list_lock); 1134 spin_unlock(&journal->j_list_lock);
1146 write_unlock(&journal->j_state_lock); 1135 write_unlock(&journal->j_state_lock);
1147 wake_up(&journal->j_wait_done_commit) 1136 wake_up(&journal->j_wait_done_commit);
1148 wake_up(&journal->j_fc_wait); <<
1149 1137
1150 /* 1138 /*
1151 * Calculate overall stats 1139 * Calculate overall stats
1152 */ 1140 */
1153 spin_lock(&journal->j_history_lock); 1141 spin_lock(&journal->j_history_lock);
1154 journal->j_stats.ts_tid++; 1142 journal->j_stats.ts_tid++;
1155 journal->j_stats.ts_requested += stat 1143 journal->j_stats.ts_requested += stats.ts_requested;
1156 journal->j_stats.run.rs_wait += stats 1144 journal->j_stats.run.rs_wait += stats.run.rs_wait;
1157 journal->j_stats.run.rs_request_delay 1145 journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
1158 journal->j_stats.run.rs_running += st 1146 journal->j_stats.run.rs_running += stats.run.rs_running;
1159 journal->j_stats.run.rs_locked += sta 1147 journal->j_stats.run.rs_locked += stats.run.rs_locked;
1160 journal->j_stats.run.rs_flushing += s 1148 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
1161 journal->j_stats.run.rs_logging += st 1149 journal->j_stats.run.rs_logging += stats.run.rs_logging;
1162 journal->j_stats.run.rs_handle_count 1150 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
1163 journal->j_stats.run.rs_blocks += sta 1151 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
1164 journal->j_stats.run.rs_blocks_logged 1152 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
1165 spin_unlock(&journal->j_history_lock) 1153 spin_unlock(&journal->j_history_lock);
1166 } 1154 }
1167 1155
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.