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;
121 struct timespec64 now; !! 123 int ret;
122 blk_opf_t write_flags = REQ_OP_WRITE | !! 124 struct timespec64 now = current_kernel_time64();
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); <<
136 tmp->h_commit_sec = cpu_to_be64(now.tv 137 tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
137 tmp->h_commit_nsec = cpu_to_be32(now.t 138 tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
138 139
139 if (jbd2_has_feature_checksum(journal) 140 if (jbd2_has_feature_checksum(journal)) {
140 tmp->h_chksum_type = JBD2 141 tmp->h_chksum_type = JBD2_CRC32_CHKSUM;
141 tmp->h_chksum_size = JBD2 142 tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE;
142 tmp->h_chksum[0] = cpu_ 143 tmp->h_chksum[0] = cpu_to_be32(crc32_sum);
143 } 144 }
144 jbd2_commit_block_csum_set(journal, bh 145 jbd2_commit_block_csum_set(journal, bh);
145 146
146 BUFFER_TRACE(bh, "submit commit block" 147 BUFFER_TRACE(bh, "submit commit block");
147 lock_buffer(bh); 148 lock_buffer(bh);
148 clear_buffer_dirty(bh); 149 clear_buffer_dirty(bh);
149 set_buffer_uptodate(bh); 150 set_buffer_uptodate(bh);
150 bh->b_end_io = journal_end_buffer_io_s 151 bh->b_end_io = journal_end_buffer_io_sync;
151 152
152 if (journal->j_flags & JBD2_BARRIER && 153 if (journal->j_flags & JBD2_BARRIER &&
153 !jbd2_has_feature_async_commit(jou 154 !jbd2_has_feature_async_commit(journal))
154 write_flags |= REQ_PREFLUSH | !! 155 ret = submit_bh(REQ_OP_WRITE,
>> 156 REQ_SYNC | REQ_PREFLUSH | REQ_FUA, bh);
>> 157 else
>> 158 ret = submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
155 159
156 submit_bh(write_flags, bh); <<
157 *cbh = bh; 160 *cbh = bh;
158 return 0; !! 161 return ret;
159 } 162 }
160 163
161 /* 164 /*
162 * This function along with journal_submit_com 165 * This function along with journal_submit_commit_record
163 * allows to write the commit record asynchron 166 * allows to write the commit record asynchronously.
164 */ 167 */
165 static int journal_wait_on_commit_record(journ 168 static int journal_wait_on_commit_record(journal_t *journal,
166 struc 169 struct buffer_head *bh)
167 { 170 {
168 int ret = 0; 171 int ret = 0;
169 172
170 clear_buffer_dirty(bh); 173 clear_buffer_dirty(bh);
171 wait_on_buffer(bh); 174 wait_on_buffer(bh);
172 175
173 if (unlikely(!buffer_uptodate(bh))) 176 if (unlikely(!buffer_uptodate(bh)))
174 ret = -EIO; 177 ret = -EIO;
175 put_bh(bh); /* One for getb 178 put_bh(bh); /* One for getblk() */
176 179
177 return ret; 180 return ret;
178 } 181 }
179 182
180 /* Send all the data buffers related to an ino !! 183 /*
181 int jbd2_submit_inode_data(journal_t *journal, !! 184 * write the filemap data using writepage() address_space_operations.
>> 185 * We don't do block allocation here even for delalloc. We don't
>> 186 * use writepages() because with dealyed allocation we may be doing
>> 187 * block allocation in writepages().
>> 188 */
>> 189 static int journal_submit_inode_data_buffers(struct address_space *mapping)
182 { 190 {
183 if (!jinode || !(jinode->i_flags & JI_ !! 191 int ret;
184 return 0; !! 192 struct writeback_control wbc = {
>> 193 .sync_mode = WB_SYNC_ALL,
>> 194 .nr_to_write = mapping->nrpages * 2,
>> 195 .range_start = 0,
>> 196 .range_end = i_size_read(mapping->host),
>> 197 };
185 198
186 trace_jbd2_submit_inode_data(jinode->i !! 199 ret = generic_writepages(mapping, &wbc);
187 return journal->j_submit_inode_data_bu !! 200 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 } 201 }
201 EXPORT_SYMBOL(jbd2_wait_inode_data); <<
202 202
203 /* 203 /*
204 * Submit all the data buffers of inode associ 204 * Submit all the data buffers of inode associated with the transaction to
205 * disk. 205 * disk.
206 * 206 *
207 * We are in a committing transaction. Therefo 207 * We are in a committing transaction. Therefore no new inode can be added to
208 * our inode list. We use JI_COMMIT_RUNNING fl 208 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
209 * operate on from being released while we wri 209 * operate on from being released while we write out pages.
210 */ 210 */
211 static int journal_submit_data_buffers(journal 211 static int journal_submit_data_buffers(journal_t *journal,
212 transaction_t *commit_transact 212 transaction_t *commit_transaction)
213 { 213 {
214 struct jbd2_inode *jinode; 214 struct jbd2_inode *jinode;
215 int err, ret = 0; 215 int err, ret = 0;
>> 216 struct address_space *mapping;
216 217
217 spin_lock(&journal->j_list_lock); 218 spin_lock(&journal->j_list_lock);
218 list_for_each_entry(jinode, &commit_tr 219 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
219 if (!(jinode->i_flags & JI_WRI 220 if (!(jinode->i_flags & JI_WRITE_DATA))
220 continue; 221 continue;
>> 222 mapping = jinode->i_vfs_inode->i_mapping;
221 jinode->i_flags |= JI_COMMIT_R 223 jinode->i_flags |= JI_COMMIT_RUNNING;
222 spin_unlock(&journal->j_list_l 224 spin_unlock(&journal->j_list_lock);
223 /* submit the inode data buffe !! 225 /*
>> 226 * submit the inode data buffers. We use writepage
>> 227 * instead of writepages. Because writepages can do
>> 228 * block allocation with delalloc. We need to write
>> 229 * only allocated blocks here.
>> 230 */
224 trace_jbd2_submit_inode_data(j 231 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
225 if (journal->j_submit_inode_da !! 232 err = journal_submit_inode_data_buffers(mapping);
226 err = journal->j_submi !! 233 if (!ret)
227 if (!ret) !! 234 ret = err;
228 ret = err; <<
229 } <<
230 spin_lock(&journal->j_list_loc 235 spin_lock(&journal->j_list_lock);
231 J_ASSERT(jinode->i_transaction 236 J_ASSERT(jinode->i_transaction == commit_transaction);
232 jinode->i_flags &= ~JI_COMMIT_ 237 jinode->i_flags &= ~JI_COMMIT_RUNNING;
233 smp_mb(); 238 smp_mb();
234 wake_up_bit(&jinode->i_flags, 239 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
235 } 240 }
236 spin_unlock(&journal->j_list_lock); 241 spin_unlock(&journal->j_list_lock);
237 return ret; 242 return ret;
238 } 243 }
239 244
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 /* 245 /*
250 * Wait for data submitted for writeout, refil 246 * Wait for data submitted for writeout, refile inodes to proper
251 * transaction if needed. 247 * transaction if needed.
252 * 248 *
253 */ 249 */
254 static int journal_finish_inode_data_buffers(j 250 static int journal_finish_inode_data_buffers(journal_t *journal,
255 transaction_t *commit_transact 251 transaction_t *commit_transaction)
256 { 252 {
257 struct jbd2_inode *jinode, *next_i; 253 struct jbd2_inode *jinode, *next_i;
258 int err, ret = 0; 254 int err, ret = 0;
259 255
260 /* For locking, see the comment in jou 256 /* For locking, see the comment in journal_submit_data_buffers() */
261 spin_lock(&journal->j_list_lock); 257 spin_lock(&journal->j_list_lock);
262 list_for_each_entry(jinode, &commit_tr 258 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
263 if (!(jinode->i_flags & JI_WAI 259 if (!(jinode->i_flags & JI_WAIT_DATA))
264 continue; 260 continue;
265 jinode->i_flags |= JI_COMMIT_R 261 jinode->i_flags |= JI_COMMIT_RUNNING;
266 spin_unlock(&journal->j_list_l 262 spin_unlock(&journal->j_list_lock);
267 /* wait for the inode data buf !! 263 err = filemap_fdatawait_keep_errors(
268 if (journal->j_finish_inode_da !! 264 jinode->i_vfs_inode->i_mapping);
269 err = journal->j_finis !! 265 if (!ret)
270 if (!ret) !! 266 ret = err;
271 ret = err; <<
272 } <<
273 cond_resched(); <<
274 spin_lock(&journal->j_list_loc 267 spin_lock(&journal->j_list_lock);
275 jinode->i_flags &= ~JI_COMMIT_ 268 jinode->i_flags &= ~JI_COMMIT_RUNNING;
276 smp_mb(); 269 smp_mb();
277 wake_up_bit(&jinode->i_flags, 270 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
278 } 271 }
279 272
280 /* Now refile inode to proper lists */ 273 /* Now refile inode to proper lists */
281 list_for_each_entry_safe(jinode, next_ 274 list_for_each_entry_safe(jinode, next_i,
282 &commit_trans 275 &commit_transaction->t_inode_list, i_list) {
283 list_del(&jinode->i_list); 276 list_del(&jinode->i_list);
284 if (jinode->i_next_transaction 277 if (jinode->i_next_transaction) {
285 jinode->i_transaction 278 jinode->i_transaction = jinode->i_next_transaction;
286 jinode->i_next_transac 279 jinode->i_next_transaction = NULL;
287 list_add(&jinode->i_li 280 list_add(&jinode->i_list,
288 &jinode->i_tra 281 &jinode->i_transaction->t_inode_list);
289 } else { 282 } else {
290 jinode->i_transaction 283 jinode->i_transaction = NULL;
291 jinode->i_dirty_start <<
292 jinode->i_dirty_end = <<
293 } 284 }
294 } 285 }
295 spin_unlock(&journal->j_list_lock); 286 spin_unlock(&journal->j_list_lock);
296 287
297 return ret; 288 return ret;
298 } 289 }
299 290
300 static __u32 jbd2_checksum_data(__u32 crc32_su 291 static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
301 { 292 {
>> 293 struct page *page = bh->b_page;
302 char *addr; 294 char *addr;
303 __u32 checksum; 295 __u32 checksum;
304 296
305 addr = kmap_local_folio(bh->b_folio, b !! 297 addr = kmap_atomic(page);
306 checksum = crc32_be(crc32_sum, addr, b !! 298 checksum = crc32_be(crc32_sum,
307 kunmap_local(addr); !! 299 (void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
>> 300 kunmap_atomic(addr);
308 301
309 return checksum; 302 return checksum;
310 } 303 }
311 304
312 static void write_tag_block(journal_t *j, jour 305 static void write_tag_block(journal_t *j, journal_block_tag_t *tag,
313 unsigned lo 306 unsigned long long block)
314 { 307 {
315 tag->t_blocknr = cpu_to_be32(block & ( 308 tag->t_blocknr = cpu_to_be32(block & (u32)~0);
316 if (jbd2_has_feature_64bit(j)) 309 if (jbd2_has_feature_64bit(j))
317 tag->t_blocknr_high = cpu_to_b 310 tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
318 } 311 }
319 312
320 static void jbd2_block_tag_csum_set(journal_t 313 static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
321 struct buf 314 struct buffer_head *bh, __u32 sequence)
322 { 315 {
323 journal_block_tag3_t *tag3 = (journal_ 316 journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
>> 317 struct page *page = bh->b_page;
324 __u8 *addr; 318 __u8 *addr;
325 __u32 csum32; 319 __u32 csum32;
326 __be32 seq; 320 __be32 seq;
327 321
328 if (!jbd2_journal_has_csum_v2or3(j)) 322 if (!jbd2_journal_has_csum_v2or3(j))
329 return; 323 return;
330 324
331 seq = cpu_to_be32(sequence); 325 seq = cpu_to_be32(sequence);
332 addr = kmap_local_folio(bh->b_folio, b !! 326 addr = kmap_atomic(page);
333 csum32 = jbd2_chksum(j, j->j_csum_seed 327 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
334 csum32 = jbd2_chksum(j, csum32, addr, !! 328 csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data),
335 kunmap_local(addr); !! 329 bh->b_size);
>> 330 kunmap_atomic(addr);
336 331
337 if (jbd2_has_feature_csum3(j)) 332 if (jbd2_has_feature_csum3(j))
338 tag3->t_checksum = cpu_to_be32 333 tag3->t_checksum = cpu_to_be32(csum32);
339 else 334 else
340 tag->t_checksum = cpu_to_be16( 335 tag->t_checksum = cpu_to_be16(csum32);
341 } 336 }
342 /* 337 /*
343 * jbd2_journal_commit_transaction 338 * jbd2_journal_commit_transaction
344 * 339 *
345 * The primary function for committing a trans 340 * The primary function for committing a transaction to the log. This
346 * function is called by the journal thread to 341 * function is called by the journal thread to begin a complete commit.
347 */ 342 */
348 void jbd2_journal_commit_transaction(journal_t 343 void jbd2_journal_commit_transaction(journal_t *journal)
349 { 344 {
350 struct transaction_stats_s stats; 345 struct transaction_stats_s stats;
351 transaction_t *commit_transaction; 346 transaction_t *commit_transaction;
352 struct journal_head *jh; 347 struct journal_head *jh;
353 struct buffer_head *descriptor; 348 struct buffer_head *descriptor;
354 struct buffer_head **wbuf = journal->j 349 struct buffer_head **wbuf = journal->j_wbuf;
355 int bufs; 350 int bufs;
356 int escape; !! 351 int flags;
357 int err; 352 int err;
358 unsigned long long blocknr; 353 unsigned long long blocknr;
359 ktime_t start_time; 354 ktime_t start_time;
360 u64 commit_time; 355 u64 commit_time;
361 char *tagp = NULL; 356 char *tagp = NULL;
362 journal_block_tag_t *tag = NULL; 357 journal_block_tag_t *tag = NULL;
363 int space_left = 0; 358 int space_left = 0;
364 int first_tag = 0; 359 int first_tag = 0;
365 int tag_flag; 360 int tag_flag;
366 int i; 361 int i;
367 int tag_bytes = journal_tag_bytes(jour 362 int tag_bytes = journal_tag_bytes(journal);
368 struct buffer_head *cbh = NULL; /* For 363 struct buffer_head *cbh = NULL; /* For transactional checksums */
369 __u32 crc32_sum = ~0; 364 __u32 crc32_sum = ~0;
370 struct blk_plug plug; 365 struct blk_plug plug;
371 /* Tail of the journal */ 366 /* Tail of the journal */
372 unsigned long first_block; 367 unsigned long first_block;
373 tid_t first_tid; 368 tid_t first_tid;
374 int update_tail; 369 int update_tail;
375 int csum_size = 0; 370 int csum_size = 0;
376 LIST_HEAD(io_bufs); 371 LIST_HEAD(io_bufs);
377 LIST_HEAD(log_bufs); 372 LIST_HEAD(log_bufs);
378 373
379 if (jbd2_journal_has_csum_v2or3(journa 374 if (jbd2_journal_has_csum_v2or3(journal))
380 csum_size = sizeof(struct jbd2 375 csum_size = sizeof(struct jbd2_journal_block_tail);
381 376
382 /* 377 /*
383 * First job: lock down the current tr 378 * First job: lock down the current transaction and wait for
384 * all outstanding updates to complete 379 * all outstanding updates to complete.
385 */ 380 */
386 381
387 /* Do we need to erase the effects of 382 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
388 if (journal->j_flags & JBD2_FLUSHED) { 383 if (journal->j_flags & JBD2_FLUSHED) {
389 jbd2_debug(3, "super block upd !! 384 jbd_debug(3, "super block updated\n");
390 mutex_lock_io(&journal->j_chec 385 mutex_lock_io(&journal->j_checkpoint_mutex);
391 /* 386 /*
392 * We hold j_checkpoint_mutex 387 * We hold j_checkpoint_mutex so tail cannot change under us.
393 * We don't need any special d 388 * We don't need any special data guarantees for writing sb
394 * since journal is empty and 389 * since journal is empty and it is ok for write to be
395 * flushed only with transacti 390 * flushed only with transaction commit.
396 */ 391 */
397 jbd2_journal_update_sb_log_tai 392 jbd2_journal_update_sb_log_tail(journal,
398 393 journal->j_tail_sequence,
399 !! 394 journal->j_tail,
>> 395 REQ_SYNC);
400 mutex_unlock(&journal->j_check 396 mutex_unlock(&journal->j_checkpoint_mutex);
401 } else { 397 } else {
402 jbd2_debug(3, "superblock not !! 398 jbd_debug(3, "superblock not updated\n");
403 } 399 }
404 400
405 J_ASSERT(journal->j_running_transactio 401 J_ASSERT(journal->j_running_transaction != NULL);
406 J_ASSERT(journal->j_committing_transac 402 J_ASSERT(journal->j_committing_transaction == NULL);
407 403
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 404 commit_transaction = journal->j_running_transaction;
432 405
433 trace_jbd2_start_commit(journal, commi 406 trace_jbd2_start_commit(journal, commit_transaction);
434 jbd2_debug(1, "JBD2: starting commit o !! 407 jbd_debug(1, "JBD2: starting commit of transaction %d\n",
435 commit_transaction->t_ 408 commit_transaction->t_tid);
436 409
437 write_lock(&journal->j_state_lock); 410 write_lock(&journal->j_state_lock);
438 journal->j_fc_off = 0; <<
439 J_ASSERT(commit_transaction->t_state = 411 J_ASSERT(commit_transaction->t_state == T_RUNNING);
440 commit_transaction->t_state = T_LOCKED 412 commit_transaction->t_state = T_LOCKED;
441 413
442 trace_jbd2_commit_locking(journal, com 414 trace_jbd2_commit_locking(journal, commit_transaction);
443 stats.run.rs_wait = commit_transaction 415 stats.run.rs_wait = commit_transaction->t_max_wait;
444 stats.run.rs_request_delay = 0; 416 stats.run.rs_request_delay = 0;
445 stats.run.rs_locked = jiffies; 417 stats.run.rs_locked = jiffies;
446 if (commit_transaction->t_requested) 418 if (commit_transaction->t_requested)
447 stats.run.rs_request_delay = 419 stats.run.rs_request_delay =
448 jbd2_time_diff(commit_ 420 jbd2_time_diff(commit_transaction->t_requested,
449 stats.r 421 stats.run.rs_locked);
450 stats.run.rs_running = jbd2_time_diff( 422 stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
451 423 stats.run.rs_locked);
452 424
453 // waits for any t_updates to finish !! 425 spin_lock(&commit_transaction->t_handle_lock);
454 jbd2_journal_wait_updates(journal); !! 426 while (atomic_read(&commit_transaction->t_updates)) {
>> 427 DEFINE_WAIT(wait);
455 428
456 commit_transaction->t_state = T_SWITCH !! 429 prepare_to_wait(&journal->j_wait_updates, &wait,
>> 430 TASK_UNINTERRUPTIBLE);
>> 431 if (atomic_read(&commit_transaction->t_updates)) {
>> 432 spin_unlock(&commit_transaction->t_handle_lock);
>> 433 write_unlock(&journal->j_state_lock);
>> 434 schedule();
>> 435 write_lock(&journal->j_state_lock);
>> 436 spin_lock(&commit_transaction->t_handle_lock);
>> 437 }
>> 438 finish_wait(&journal->j_wait_updates, &wait);
>> 439 }
>> 440 spin_unlock(&commit_transaction->t_handle_lock);
457 441
458 J_ASSERT (atomic_read(&commit_transact 442 J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
459 journal->j_max_transac 443 journal->j_max_transaction_buffers);
460 444
461 /* 445 /*
462 * First thing we are allowed to do is 446 * First thing we are allowed to do is to discard any remaining
463 * BJ_Reserved buffers. Note, it is _ 447 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
464 * that there are no such buffers: if 448 * that there are no such buffers: if a large filesystem
465 * operation like a truncate needs to 449 * operation like a truncate needs to split itself over multiple
466 * transactions, then it may try to do 450 * transactions, then it may try to do a jbd2_journal_restart() while
467 * there are still BJ_Reserved buffers 451 * there are still BJ_Reserved buffers outstanding. These must
468 * be released cleanly from the curren 452 * be released cleanly from the current transaction.
469 * 453 *
470 * In this case, the filesystem must s 454 * In this case, the filesystem must still reserve write access
471 * again before modifying the buffer i 455 * again before modifying the buffer in the new transaction, but
472 * we do not require it to remember ex 456 * we do not require it to remember exactly which old buffers it
473 * has reserved. This is consistent w 457 * has reserved. This is consistent with the existing behaviour
474 * that multiple jbd2_journal_get_writ 458 * that multiple jbd2_journal_get_write_access() calls to the same
475 * buffer are perfectly permissible. 459 * buffer are perfectly permissible.
476 * We use journal->j_state_lock here t <<
477 * t_reserved_list with eviction of bu <<
478 */ 460 */
479 while (commit_transaction->t_reserved_ 461 while (commit_transaction->t_reserved_list) {
480 jh = commit_transaction->t_res 462 jh = commit_transaction->t_reserved_list;
481 JBUFFER_TRACE(jh, "reserved, u 463 JBUFFER_TRACE(jh, "reserved, unused: refile");
482 /* 464 /*
483 * A jbd2_journal_get_undo_acc 465 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
484 * leave undo-committed data. 466 * leave undo-committed data.
485 */ 467 */
486 if (jh->b_committed_data) { 468 if (jh->b_committed_data) {
487 struct buffer_head *bh 469 struct buffer_head *bh = jh2bh(jh);
488 470
489 spin_lock(&jh->b_state !! 471 jbd_lock_bh_state(bh);
490 jbd2_free(jh->b_commit 472 jbd2_free(jh->b_committed_data, bh->b_size);
491 jh->b_committed_data = 473 jh->b_committed_data = NULL;
492 spin_unlock(&jh->b_sta !! 474 jbd_unlock_bh_state(bh);
493 } 475 }
494 jbd2_journal_refile_buffer(jou 476 jbd2_journal_refile_buffer(journal, jh);
495 } 477 }
496 478
497 write_unlock(&journal->j_state_lock); <<
498 /* 479 /*
499 * Now try to drop any written-back bu 480 * Now try to drop any written-back buffers from the journal's
500 * checkpoint lists. We do this *befo 481 * checkpoint lists. We do this *before* commit because it potentially
501 * frees some memory 482 * frees some memory
502 */ 483 */
503 spin_lock(&journal->j_list_lock); 484 spin_lock(&journal->j_list_lock);
504 __jbd2_journal_clean_checkpoint_list(j !! 485 __jbd2_journal_clean_checkpoint_list(journal, false);
505 spin_unlock(&journal->j_list_lock); 486 spin_unlock(&journal->j_list_lock);
506 487
507 jbd2_debug(3, "JBD2: commit phase 1\n" !! 488 jbd_debug(3, "JBD2: commit phase 1\n");
508 489
509 /* 490 /*
510 * Clear revoked flag to reflect there 491 * Clear revoked flag to reflect there is no revoked buffers
511 * in the next transaction which is go 492 * in the next transaction which is going to be started.
512 */ 493 */
513 jbd2_clear_buffer_revoked_flags(journa 494 jbd2_clear_buffer_revoked_flags(journal);
514 495
515 /* 496 /*
516 * Switch to a new revoke table. 497 * Switch to a new revoke table.
517 */ 498 */
518 jbd2_journal_switch_revoke_table(journ 499 jbd2_journal_switch_revoke_table(journal);
519 500
520 write_lock(&journal->j_state_lock); <<
521 /* 501 /*
522 * Reserved credits cannot be claimed 502 * Reserved credits cannot be claimed anymore, free them
523 */ 503 */
524 atomic_sub(atomic_read(&journal->j_res 504 atomic_sub(atomic_read(&journal->j_reserved_credits),
525 &commit_transaction->t_outs 505 &commit_transaction->t_outstanding_credits);
526 506
527 trace_jbd2_commit_flushing(journal, co 507 trace_jbd2_commit_flushing(journal, commit_transaction);
528 stats.run.rs_flushing = jiffies; 508 stats.run.rs_flushing = jiffies;
529 stats.run.rs_locked = jbd2_time_diff(s 509 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
530 s 510 stats.run.rs_flushing);
531 511
532 commit_transaction->t_state = T_FLUSH; 512 commit_transaction->t_state = T_FLUSH;
533 journal->j_committing_transaction = co 513 journal->j_committing_transaction = commit_transaction;
534 journal->j_running_transaction = NULL; 514 journal->j_running_transaction = NULL;
535 start_time = ktime_get(); 515 start_time = ktime_get();
536 commit_transaction->t_log_start = jour 516 commit_transaction->t_log_start = journal->j_head;
537 wake_up_all(&journal->j_wait_transacti !! 517 wake_up(&journal->j_wait_transaction_locked);
538 write_unlock(&journal->j_state_lock); 518 write_unlock(&journal->j_state_lock);
539 519
540 jbd2_debug(3, "JBD2: commit phase 2a\n !! 520 jbd_debug(3, "JBD2: commit phase 2a\n");
541 521
542 /* 522 /*
543 * Now start flushing things to disk, 523 * Now start flushing things to disk, in the order they appear
544 * on the transaction lists. Data blo 524 * on the transaction lists. Data blocks go first.
545 */ 525 */
546 err = journal_submit_data_buffers(jour 526 err = journal_submit_data_buffers(journal, commit_transaction);
547 if (err) 527 if (err)
548 jbd2_journal_abort(journal, er 528 jbd2_journal_abort(journal, err);
549 529
550 blk_start_plug(&plug); 530 blk_start_plug(&plug);
551 jbd2_journal_write_revoke_records(comm 531 jbd2_journal_write_revoke_records(commit_transaction, &log_bufs);
552 532
553 jbd2_debug(3, "JBD2: commit phase 2b\n !! 533 jbd_debug(3, "JBD2: commit phase 2b\n");
554 534
555 /* 535 /*
556 * Way to go: we have now written out 536 * Way to go: we have now written out all of the data for a
557 * transaction! Now comes the tricky 537 * transaction! Now comes the tricky part: we need to write out
558 * metadata. Loop over the transactio 538 * metadata. Loop over the transaction's entire buffer list:
559 */ 539 */
560 write_lock(&journal->j_state_lock); 540 write_lock(&journal->j_state_lock);
561 commit_transaction->t_state = T_COMMIT 541 commit_transaction->t_state = T_COMMIT;
562 write_unlock(&journal->j_state_lock); 542 write_unlock(&journal->j_state_lock);
563 543
564 trace_jbd2_commit_logging(journal, com 544 trace_jbd2_commit_logging(journal, commit_transaction);
565 stats.run.rs_logging = jiffies; 545 stats.run.rs_logging = jiffies;
566 stats.run.rs_flushing = jbd2_time_diff 546 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
567 547 stats.run.rs_logging);
568 stats.run.rs_blocks = commit_transacti !! 548 stats.run.rs_blocks =
>> 549 atomic_read(&commit_transaction->t_outstanding_credits);
569 stats.run.rs_blocks_logged = 0; 550 stats.run.rs_blocks_logged = 0;
570 551
571 J_ASSERT(commit_transaction->t_nr_buff 552 J_ASSERT(commit_transaction->t_nr_buffers <=
572 atomic_read(&commit_transacti 553 atomic_read(&commit_transaction->t_outstanding_credits));
573 554
>> 555 err = 0;
574 bufs = 0; 556 bufs = 0;
575 descriptor = NULL; 557 descriptor = NULL;
576 while (commit_transaction->t_buffers) 558 while (commit_transaction->t_buffers) {
577 559
578 /* Find the next buffer to be 560 /* Find the next buffer to be journaled... */
579 561
580 jh = commit_transaction->t_buf 562 jh = commit_transaction->t_buffers;
581 563
582 /* If we're in abort mode, we 564 /* If we're in abort mode, we just un-journal the buffer and
583 release it. */ 565 release it. */
584 566
585 if (is_journal_aborted(journal 567 if (is_journal_aborted(journal)) {
586 clear_buffer_jbddirty( 568 clear_buffer_jbddirty(jh2bh(jh));
587 JBUFFER_TRACE(jh, "jou 569 JBUFFER_TRACE(jh, "journal is aborting: refile");
588 jbd2_buffer_abort_trig 570 jbd2_buffer_abort_trigger(jh,
589 571 jh->b_frozen_data ?
590 572 jh->b_frozen_triggers :
591 573 jh->b_triggers);
592 jbd2_journal_refile_bu 574 jbd2_journal_refile_buffer(journal, jh);
593 /* If that was the las 575 /* If that was the last one, we need to clean up
594 * any descriptor buff 576 * any descriptor buffers which may have been
595 * already allocated, 577 * already allocated, even if we are now
596 * aborting. */ 578 * aborting. */
597 if (!commit_transactio 579 if (!commit_transaction->t_buffers)
598 goto start_jou 580 goto start_journal_io;
599 continue; 581 continue;
600 } 582 }
601 583
602 /* Make sure we have a descrip 584 /* Make sure we have a descriptor block in which to
603 record the metadata buffer. 585 record the metadata buffer. */
604 586
605 if (!descriptor) { 587 if (!descriptor) {
606 J_ASSERT (bufs == 0); 588 J_ASSERT (bufs == 0);
607 589
608 jbd2_debug(4, "JBD2: g !! 590 jbd_debug(4, "JBD2: get descriptor\n");
609 591
610 descriptor = jbd2_jour 592 descriptor = jbd2_journal_get_descriptor_buffer(
611 593 commit_transaction,
612 594 JBD2_DESCRIPTOR_BLOCK);
613 if (!descriptor) { 595 if (!descriptor) {
614 jbd2_journal_a 596 jbd2_journal_abort(journal, -EIO);
615 continue; 597 continue;
616 } 598 }
617 599
618 jbd2_debug(4, "JBD2: g !! 600 jbd_debug(4, "JBD2: got buffer %llu (%p)\n",
619 (unsigned long 601 (unsigned long long)descriptor->b_blocknr,
620 descriptor->b_ 602 descriptor->b_data);
621 tagp = &descriptor->b_ 603 tagp = &descriptor->b_data[sizeof(journal_header_t)];
622 space_left = descripto 604 space_left = descriptor->b_size -
623 605 sizeof(journal_header_t);
624 first_tag = 1; 606 first_tag = 1;
625 set_buffer_jwrite(desc 607 set_buffer_jwrite(descriptor);
626 set_buffer_dirty(descr 608 set_buffer_dirty(descriptor);
627 wbuf[bufs++] = descrip 609 wbuf[bufs++] = descriptor;
628 610
629 /* Record it so that w 611 /* Record it so that we can wait for IO
630 completion later */ 612 completion later */
631 BUFFER_TRACE(descripto 613 BUFFER_TRACE(descriptor, "ph3: file as descriptor");
632 jbd2_file_log_bh(&log_ 614 jbd2_file_log_bh(&log_bufs, descriptor);
633 } 615 }
634 616
635 /* Where is the buffer to be w 617 /* Where is the buffer to be written? */
636 618
637 err = jbd2_journal_next_log_bl 619 err = jbd2_journal_next_log_block(journal, &blocknr);
638 /* If the block mapping failed 620 /* If the block mapping failed, just abandon the buffer
639 and repeat this loop: we'll 621 and repeat this loop: we'll fall into the
640 refile-on-abort condition a 622 refile-on-abort condition above. */
641 if (err) { 623 if (err) {
642 jbd2_journal_abort(jou 624 jbd2_journal_abort(journal, err);
643 continue; 625 continue;
644 } 626 }
645 627
646 /* 628 /*
647 * start_this_handle() uses t_ 629 * start_this_handle() uses t_outstanding_credits to determine
648 * the free space in the log. !! 630 * the free space in the log, but this counter is changed
>> 631 * by jbd2_journal_next_log_block() also.
649 */ 632 */
650 atomic_dec(&commit_transaction 633 atomic_dec(&commit_transaction->t_outstanding_credits);
651 634
652 /* Bump b_count to prevent tru 635 /* Bump b_count to prevent truncate from stumbling over
653 the shadowed buffer! @@@ T 636 the shadowed buffer! @@@ This can go if we ever get
654 rid of the shadow pairing o 637 rid of the shadow pairing of buffers. */
655 atomic_inc(&jh2bh(jh)->b_count 638 atomic_inc(&jh2bh(jh)->b_count);
656 639
657 /* 640 /*
658 * Make a temporary IO buffer 641 * Make a temporary IO buffer with which to write it out
659 * (this will requeue the meta 642 * (this will requeue the metadata buffer to BJ_Shadow).
660 */ 643 */
661 set_bit(BH_JWrite, &jh2bh(jh)- 644 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
662 JBUFFER_TRACE(jh, "ph3: write 645 JBUFFER_TRACE(jh, "ph3: write metadata");
663 escape = jbd2_journal_write_me !! 646 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
664 647 jh, &wbuf[bufs], blocknr);
665 if (escape < 0) { !! 648 if (flags < 0) {
666 jbd2_journal_abort(jou !! 649 jbd2_journal_abort(journal, flags);
667 continue; 650 continue;
668 } 651 }
669 jbd2_file_log_bh(&io_bufs, wbu 652 jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
670 653
671 /* Record the new block's tag 654 /* Record the new block's tag in the current descriptor
672 buffer */ 655 buffer */
673 656
674 tag_flag = 0; 657 tag_flag = 0;
675 if (escape) !! 658 if (flags & 1)
676 tag_flag |= JBD2_FLAG_ 659 tag_flag |= JBD2_FLAG_ESCAPE;
677 if (!first_tag) 660 if (!first_tag)
678 tag_flag |= JBD2_FLAG_ 661 tag_flag |= JBD2_FLAG_SAME_UUID;
679 662
680 tag = (journal_block_tag_t *) 663 tag = (journal_block_tag_t *) tagp;
681 write_tag_block(journal, tag, 664 write_tag_block(journal, tag, jh2bh(jh)->b_blocknr);
682 tag->t_flags = cpu_to_be16(tag 665 tag->t_flags = cpu_to_be16(tag_flag);
683 jbd2_block_tag_csum_set(journa 666 jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
684 commit 667 commit_transaction->t_tid);
685 tagp += tag_bytes; 668 tagp += tag_bytes;
686 space_left -= tag_bytes; 669 space_left -= tag_bytes;
687 bufs++; 670 bufs++;
688 671
689 if (first_tag) { 672 if (first_tag) {
690 memcpy (tagp, journal- 673 memcpy (tagp, journal->j_uuid, 16);
691 tagp += 16; 674 tagp += 16;
692 space_left -= 16; 675 space_left -= 16;
693 first_tag = 0; 676 first_tag = 0;
694 } 677 }
695 678
696 /* If there's no more to do, o 679 /* If there's no more to do, or if the descriptor is full,
697 let the IO rip! */ 680 let the IO rip! */
698 681
699 if (bufs == journal->j_wbufsiz 682 if (bufs == journal->j_wbufsize ||
700 commit_transaction->t_buff 683 commit_transaction->t_buffers == NULL ||
701 space_left < tag_bytes + 1 684 space_left < tag_bytes + 16 + csum_size) {
702 685
703 jbd2_debug(4, "JBD2: S !! 686 jbd_debug(4, "JBD2: Submit %d IOs\n", bufs);
704 687
705 /* Write an end-of-des 688 /* Write an end-of-descriptor marker before
706 submitting the IOs. 689 submitting the IOs. "tag" still points to
707 the last tag we set 690 the last tag we set up. */
708 691
709 tag->t_flags |= cpu_to 692 tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
710 start_journal_io: <<
711 if (descriptor) <<
712 jbd2_descripto <<
713 <<
714 693
>> 694 jbd2_descriptor_block_csum_set(journal, descriptor);
>> 695 start_journal_io:
715 for (i = 0; i < bufs; 696 for (i = 0; i < bufs; i++) {
716 struct buffer_ 697 struct buffer_head *bh = wbuf[i];
717 <<
718 /* 698 /*
719 * Compute che 699 * Compute checksum.
720 */ 700 */
721 if (jbd2_has_f 701 if (jbd2_has_feature_checksum(journal)) {
722 crc32_ 702 crc32_sum =
723 jb 703 jbd2_checksum_data(crc32_sum, bh);
724 } 704 }
725 705
726 lock_buffer(bh 706 lock_buffer(bh);
727 clear_buffer_d 707 clear_buffer_dirty(bh);
728 set_buffer_upt 708 set_buffer_uptodate(bh);
729 bh->b_end_io = 709 bh->b_end_io = journal_end_buffer_io_sync;
730 submit_bh(REQ_ !! 710 submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
731 bh); <<
732 } 711 }
733 cond_resched(); 712 cond_resched();
>> 713 stats.run.rs_blocks_logged += bufs;
734 714
735 /* Force a new descrip 715 /* Force a new descriptor to be generated next
736 time round the loop 716 time round the loop. */
737 descriptor = NULL; 717 descriptor = NULL;
738 bufs = 0; 718 bufs = 0;
739 } 719 }
740 } 720 }
741 721
742 err = journal_finish_inode_data_buffer 722 err = journal_finish_inode_data_buffers(journal, commit_transaction);
743 if (err) { 723 if (err) {
744 printk(KERN_WARNING 724 printk(KERN_WARNING
745 "JBD2: Detected IO err 725 "JBD2: Detected IO errors while flushing file data "
746 "on %s\n", journal->j_d 726 "on %s\n", journal->j_devname);
747 if (journal->j_flags & JBD2_AB 727 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
748 jbd2_journal_abort(jou 728 jbd2_journal_abort(journal, err);
749 err = 0; 729 err = 0;
750 } 730 }
751 731
752 /* 732 /*
753 * Get current oldest transaction in t 733 * Get current oldest transaction in the log before we issue flush
754 * to the filesystem device. After the 734 * to the filesystem device. After the flush we can be sure that
755 * blocks of all older transactions ar 735 * blocks of all older transactions are checkpointed to persistent
756 * storage and we will be safe to upda 736 * storage and we will be safe to update journal start in the
757 * superblock with the numbers we get 737 * superblock with the numbers we get here.
758 */ 738 */
759 update_tail = 739 update_tail =
760 jbd2_journal_get_log_tail(jour 740 jbd2_journal_get_log_tail(journal, &first_tid, &first_block);
761 741
762 write_lock(&journal->j_state_lock); 742 write_lock(&journal->j_state_lock);
763 if (update_tail) { 743 if (update_tail) {
764 long freed = first_block - jou 744 long freed = first_block - journal->j_tail;
765 745
766 if (first_block < journal->j_t 746 if (first_block < journal->j_tail)
767 freed += journal->j_la 747 freed += journal->j_last - journal->j_first;
768 /* Update tail only if we free 748 /* Update tail only if we free significant amount of space */
769 if (freed < journal->j_max_tra !! 749 if (freed < journal->j_maxlen / 4)
770 update_tail = 0; 750 update_tail = 0;
771 } 751 }
772 J_ASSERT(commit_transaction->t_state = 752 J_ASSERT(commit_transaction->t_state == T_COMMIT);
773 commit_transaction->t_state = T_COMMIT 753 commit_transaction->t_state = T_COMMIT_DFLUSH;
774 write_unlock(&journal->j_state_lock); 754 write_unlock(&journal->j_state_lock);
775 755
776 /* !! 756 /*
777 * If the journal is not located on th 757 * If the journal is not located on the file system device,
778 * then we must flush the file system 758 * then we must flush the file system device before we issue
779 * the commit record 759 * the commit record
780 */ 760 */
781 if (commit_transaction->t_need_data_fl 761 if (commit_transaction->t_need_data_flush &&
782 (journal->j_fs_dev != journal->j_d 762 (journal->j_fs_dev != journal->j_dev) &&
783 (journal->j_flags & JBD2_BARRIER)) 763 (journal->j_flags & JBD2_BARRIER))
784 blkdev_issue_flush(journal->j_ !! 764 blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL);
785 765
786 /* Done it all: now write the commit r 766 /* Done it all: now write the commit record asynchronously. */
787 if (jbd2_has_feature_async_commit(jour 767 if (jbd2_has_feature_async_commit(journal)) {
788 err = journal_submit_commit_re 768 err = journal_submit_commit_record(journal, commit_transaction,
789 769 &cbh, crc32_sum);
790 if (err) 770 if (err)
791 jbd2_journal_abort(jou !! 771 __jbd2_journal_abort_hard(journal);
792 } 772 }
793 773
794 blk_finish_plug(&plug); 774 blk_finish_plug(&plug);
795 775
796 /* Lo and behold: we have just managed 776 /* Lo and behold: we have just managed to send a transaction to
797 the log. Before we can commit it, 777 the log. Before we can commit it, wait for the IO so far to
798 complete. Control buffers being wr 778 complete. Control buffers being written are on the
799 transaction's t_log_list queue, and 779 transaction's t_log_list queue, and metadata buffers are on
800 the io_bufs list. 780 the io_bufs list.
801 781
802 Wait for the buffers in reverse ord 782 Wait for the buffers in reverse order. That way we are
803 less likely to be woken up until al 783 less likely to be woken up until all IOs have completed, and
804 so we incur less scheduling load. 784 so we incur less scheduling load.
805 */ 785 */
806 786
807 jbd2_debug(3, "JBD2: commit phase 3\n" !! 787 jbd_debug(3, "JBD2: commit phase 3\n");
808 788
809 while (!list_empty(&io_bufs)) { 789 while (!list_empty(&io_bufs)) {
810 struct buffer_head *bh = list_ 790 struct buffer_head *bh = list_entry(io_bufs.prev,
811 791 struct buffer_head,
812 792 b_assoc_buffers);
813 793
814 wait_on_buffer(bh); 794 wait_on_buffer(bh);
815 cond_resched(); 795 cond_resched();
816 796
817 if (unlikely(!buffer_uptodate( 797 if (unlikely(!buffer_uptodate(bh)))
818 err = -EIO; 798 err = -EIO;
819 jbd2_unfile_log_bh(bh); 799 jbd2_unfile_log_bh(bh);
820 stats.run.rs_blocks_logged++; <<
821 800
822 /* 801 /*
823 * The list contains temporary 802 * The list contains temporary buffer heads created by
824 * jbd2_journal_write_metadata 803 * jbd2_journal_write_metadata_buffer().
825 */ 804 */
826 BUFFER_TRACE(bh, "dumping temp 805 BUFFER_TRACE(bh, "dumping temporary bh");
827 __brelse(bh); 806 __brelse(bh);
828 J_ASSERT_BH(bh, atomic_read(&b 807 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
829 free_buffer_head(bh); 808 free_buffer_head(bh);
830 809
831 /* We also have to refile the 810 /* We also have to refile the corresponding shadowed buffer */
832 jh = commit_transaction->t_sha 811 jh = commit_transaction->t_shadow_list->b_tprev;
833 bh = jh2bh(jh); 812 bh = jh2bh(jh);
834 clear_buffer_jwrite(bh); 813 clear_buffer_jwrite(bh);
835 J_ASSERT_BH(bh, buffer_jbddirt 814 J_ASSERT_BH(bh, buffer_jbddirty(bh));
836 J_ASSERT_BH(bh, !buffer_shadow 815 J_ASSERT_BH(bh, !buffer_shadow(bh));
837 816
838 /* The metadata is now release 817 /* The metadata is now released for reuse, but we need
839 to remember it against this 818 to remember it against this transaction so that when
840 we finally commit, we can d 819 we finally commit, we can do any checkpointing
841 required. */ 820 required. */
842 JBUFFER_TRACE(jh, "file as BJ_ 821 JBUFFER_TRACE(jh, "file as BJ_Forget");
843 jbd2_journal_file_buffer(jh, c 822 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
844 JBUFFER_TRACE(jh, "brelse shad 823 JBUFFER_TRACE(jh, "brelse shadowed buffer");
845 __brelse(bh); 824 __brelse(bh);
846 } 825 }
847 826
848 J_ASSERT (commit_transaction->t_shadow 827 J_ASSERT (commit_transaction->t_shadow_list == NULL);
849 828
850 jbd2_debug(3, "JBD2: commit phase 4\n" !! 829 jbd_debug(3, "JBD2: commit phase 4\n");
851 830
852 /* Here we wait for the revoke record 831 /* Here we wait for the revoke record and descriptor record buffers */
853 while (!list_empty(&log_bufs)) { 832 while (!list_empty(&log_bufs)) {
854 struct buffer_head *bh; 833 struct buffer_head *bh;
855 834
856 bh = list_entry(log_bufs.prev, 835 bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
857 wait_on_buffer(bh); 836 wait_on_buffer(bh);
858 cond_resched(); 837 cond_resched();
859 838
860 if (unlikely(!buffer_uptodate( 839 if (unlikely(!buffer_uptodate(bh)))
861 err = -EIO; 840 err = -EIO;
862 841
863 BUFFER_TRACE(bh, "ph5: control 842 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
864 clear_buffer_jwrite(bh); 843 clear_buffer_jwrite(bh);
865 jbd2_unfile_log_bh(bh); 844 jbd2_unfile_log_bh(bh);
866 stats.run.rs_blocks_logged++; <<
867 __brelse(bh); /* One 845 __brelse(bh); /* One for getblk */
868 /* AKPM: bforget here */ 846 /* AKPM: bforget here */
869 } 847 }
870 848
871 if (err) 849 if (err)
872 jbd2_journal_abort(journal, er 850 jbd2_journal_abort(journal, err);
873 851
874 jbd2_debug(3, "JBD2: commit phase 5\n" !! 852 jbd_debug(3, "JBD2: commit phase 5\n");
875 write_lock(&journal->j_state_lock); 853 write_lock(&journal->j_state_lock);
876 J_ASSERT(commit_transaction->t_state = 854 J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
877 commit_transaction->t_state = T_COMMIT 855 commit_transaction->t_state = T_COMMIT_JFLUSH;
878 write_unlock(&journal->j_state_lock); 856 write_unlock(&journal->j_state_lock);
879 857
880 if (!jbd2_has_feature_async_commit(jou 858 if (!jbd2_has_feature_async_commit(journal)) {
881 err = journal_submit_commit_re 859 err = journal_submit_commit_record(journal, commit_transaction,
882 860 &cbh, crc32_sum);
883 if (err) 861 if (err)
884 jbd2_journal_abort(jou !! 862 __jbd2_journal_abort_hard(journal);
885 } 863 }
886 if (cbh) 864 if (cbh)
887 err = journal_wait_on_commit_r 865 err = journal_wait_on_commit_record(journal, cbh);
888 stats.run.rs_blocks_logged++; <<
889 if (jbd2_has_feature_async_commit(jour 866 if (jbd2_has_feature_async_commit(journal) &&
890 journal->j_flags & JBD2_BARRIER) { 867 journal->j_flags & JBD2_BARRIER) {
891 blkdev_issue_flush(journal->j_ !! 868 blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL);
892 } 869 }
893 870
894 if (err) 871 if (err)
895 jbd2_journal_abort(journal, er 872 jbd2_journal_abort(journal, err);
896 873
897 WARN_ON_ONCE( <<
898 atomic_read(&commit_transactio <<
899 <<
900 /* 874 /*
901 * Now disk caches for filesystem devi 875 * Now disk caches for filesystem device are flushed so we are safe to
902 * erase checkpointed transactions fro 876 * erase checkpointed transactions from the log by updating journal
903 * superblock. 877 * superblock.
904 */ 878 */
905 if (update_tail) 879 if (update_tail)
906 jbd2_update_log_tail(journal, 880 jbd2_update_log_tail(journal, first_tid, first_block);
907 881
908 /* End of a transaction! Finally, we 882 /* End of a transaction! Finally, we can do checkpoint
909 processing: any buffers committed a 883 processing: any buffers committed as a result of this
910 transaction can be removed from any 884 transaction can be removed from any checkpoint list it was on
911 before. */ 885 before. */
912 886
913 jbd2_debug(3, "JBD2: commit phase 6\n" !! 887 jbd_debug(3, "JBD2: commit phase 6\n");
914 888
915 J_ASSERT(list_empty(&commit_transactio 889 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
916 J_ASSERT(commit_transaction->t_buffers 890 J_ASSERT(commit_transaction->t_buffers == NULL);
917 J_ASSERT(commit_transaction->t_checkpo 891 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
918 J_ASSERT(commit_transaction->t_shadow_ 892 J_ASSERT(commit_transaction->t_shadow_list == NULL);
919 893
920 restart_loop: 894 restart_loop:
921 /* 895 /*
922 * As there are other places (journal_ 896 * As there are other places (journal_unmap_buffer()) adding buffers
923 * to this list we have to be careful 897 * to this list we have to be careful and hold the j_list_lock.
924 */ 898 */
925 spin_lock(&journal->j_list_lock); 899 spin_lock(&journal->j_list_lock);
926 while (commit_transaction->t_forget) { 900 while (commit_transaction->t_forget) {
927 transaction_t *cp_transaction; 901 transaction_t *cp_transaction;
928 struct buffer_head *bh; 902 struct buffer_head *bh;
929 int try_to_free = 0; 903 int try_to_free = 0;
930 bool drop_ref; <<
931 904
932 jh = commit_transaction->t_for 905 jh = commit_transaction->t_forget;
933 spin_unlock(&journal->j_list_l 906 spin_unlock(&journal->j_list_lock);
934 bh = jh2bh(jh); 907 bh = jh2bh(jh);
935 /* 908 /*
936 * Get a reference so that bh 909 * Get a reference so that bh cannot be freed before we are
937 * done with it. 910 * done with it.
938 */ 911 */
939 get_bh(bh); 912 get_bh(bh);
940 spin_lock(&jh->b_state_lock); !! 913 jbd_lock_bh_state(bh);
941 J_ASSERT_JH(jh, jh->b_transact 914 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
942 915
943 /* 916 /*
944 * If there is undo-protected 917 * If there is undo-protected committed data against
945 * this buffer, then we can re 918 * this buffer, then we can remove it now. If it is a
946 * buffer needing such protect 919 * buffer needing such protection, the old frozen_data
947 * field now points to a commi 920 * field now points to a committed version of the
948 * buffer, so rotate that fiel 921 * buffer, so rotate that field to the new committed
949 * data. 922 * data.
950 * 923 *
951 * Otherwise, we can just thro 924 * Otherwise, we can just throw away the frozen data now.
952 * 925 *
953 * We also know that the froze 926 * We also know that the frozen data has already fired
954 * its triggers if they exist, 927 * its triggers if they exist, so we can clear that too.
955 */ 928 */
956 if (jh->b_committed_data) { 929 if (jh->b_committed_data) {
957 jbd2_free(jh->b_commit 930 jbd2_free(jh->b_committed_data, bh->b_size);
958 jh->b_committed_data = 931 jh->b_committed_data = NULL;
959 if (jh->b_frozen_data) 932 if (jh->b_frozen_data) {
960 jh->b_committe 933 jh->b_committed_data = jh->b_frozen_data;
961 jh->b_frozen_d 934 jh->b_frozen_data = NULL;
962 jh->b_frozen_t 935 jh->b_frozen_triggers = NULL;
963 } 936 }
964 } else if (jh->b_frozen_data) 937 } else if (jh->b_frozen_data) {
965 jbd2_free(jh->b_frozen 938 jbd2_free(jh->b_frozen_data, bh->b_size);
966 jh->b_frozen_data = NU 939 jh->b_frozen_data = NULL;
967 jh->b_frozen_triggers 940 jh->b_frozen_triggers = NULL;
968 } 941 }
969 942
970 spin_lock(&journal->j_list_loc 943 spin_lock(&journal->j_list_lock);
971 cp_transaction = jh->b_cp_tran 944 cp_transaction = jh->b_cp_transaction;
972 if (cp_transaction) { 945 if (cp_transaction) {
973 JBUFFER_TRACE(jh, "rem 946 JBUFFER_TRACE(jh, "remove from old cp transaction");
974 cp_transaction->t_chp_ 947 cp_transaction->t_chp_stats.cs_dropped++;
975 __jbd2_journal_remove_ 948 __jbd2_journal_remove_checkpoint(jh);
976 } 949 }
977 950
978 /* Only re-checkpoint the buff 951 /* Only re-checkpoint the buffer_head if it is marked
979 * dirty. If the buffer was a 952 * dirty. If the buffer was added to the BJ_Forget list
980 * by jbd2_journal_forget, it 953 * by jbd2_journal_forget, it may no longer be dirty and
981 * there's no point in keeping 954 * there's no point in keeping a checkpoint record for
982 * it. */ 955 * it. */
983 956
984 /* 957 /*
985 * A buffer which has been fre !! 958 * A buffer which has been freed while still being journaled by
986 * by a previous transaction, !! 959 * a previous transaction.
987 * the running transaction. If !! 960 */
988 * contains "add to orphan" op !! 961 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 /* 962 /*
1000 * Block device buffe !! 963 * If the running transaction is the one containing
1001 * time, so it is eno !! 964 * "add to orphan" operation (b_next_transaction !=
1002 * buffer_freed bits. !! 965 * NULL), we have to wait for that transaction to
1003 * journalled data) w !! 966 * commit before we can really get rid of the buffer.
1004 * more bits. We also !! 967 * So just clear b_modified to not confuse transaction
1005 * because the data p !! 968 * credit accounting and refile the buffer to
1006 * our hands. Note th !! 969 * BJ_Forget of the running transaction. If the just
1007 * need to make buffe !! 970 * committed transaction contains "add to orphan"
1008 * already detached f !! 971 * operation, we can completely invalidate the buffer
1009 * get reused. !! 972 * now. We are rather through in that since the
>> 973 * buffer may be still accessible when blocksize <
>> 974 * pagesize and it is attached to the last partial
>> 975 * page.
1010 */ 976 */
1011 mapping = READ_ONCE(b !! 977 jh->b_modified = 0;
1012 if (mapping && !sb_is !! 978 if (!jh->b_next_transaction) {
>> 979 clear_buffer_freed(bh);
>> 980 clear_buffer_jbddirty(bh);
1013 clear_buffer_ 981 clear_buffer_mapped(bh);
1014 clear_buffer_ 982 clear_buffer_new(bh);
1015 clear_buffer_ 983 clear_buffer_req(bh);
1016 bh->b_bdev = 984 bh->b_bdev = NULL;
1017 } 985 }
1018 } 986 }
1019 987
1020 if (buffer_jbddirty(bh)) { 988 if (buffer_jbddirty(bh)) {
1021 JBUFFER_TRACE(jh, "ad 989 JBUFFER_TRACE(jh, "add to new checkpointing trans");
1022 __jbd2_journal_insert 990 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
1023 if (is_journal_aborte 991 if (is_journal_aborted(journal))
1024 clear_buffer_ 992 clear_buffer_jbddirty(bh);
1025 } else { 993 } else {
1026 J_ASSERT_BH(bh, !buff 994 J_ASSERT_BH(bh, !buffer_dirty(bh));
1027 /* 995 /*
1028 * The buffer on BJ_F 996 * The buffer on BJ_Forget list and not jbddirty means
1029 * it has been freed 997 * it has been freed by this transaction and hence it
1030 * could not have bee 998 * could not have been reallocated until this
1031 * transaction has co 999 * transaction has committed. *BUT* it could be
1032 * reallocated once w 1000 * reallocated once we have written all the data to
1033 * disk and before we 1001 * disk and before we process the buffer on BJ_Forget
1034 * list. 1002 * list.
1035 */ 1003 */
1036 if (!jh->b_next_trans 1004 if (!jh->b_next_transaction)
1037 try_to_free = 1005 try_to_free = 1;
1038 } 1006 }
1039 JBUFFER_TRACE(jh, "refile or 1007 JBUFFER_TRACE(jh, "refile or unfile buffer");
1040 drop_ref = __jbd2_journal_ref !! 1008 __jbd2_journal_refile_buffer(jh);
1041 spin_unlock(&jh->b_state_lock !! 1009 jbd_unlock_bh_state(bh);
1042 if (drop_ref) <<
1043 jbd2_journal_put_jour <<
1044 if (try_to_free) 1010 if (try_to_free)
1045 release_buffer_page(b 1011 release_buffer_page(bh); /* Drops bh reference */
1046 else 1012 else
1047 __brelse(bh); 1013 __brelse(bh);
1048 cond_resched_lock(&journal->j 1014 cond_resched_lock(&journal->j_list_lock);
1049 } 1015 }
1050 spin_unlock(&journal->j_list_lock); 1016 spin_unlock(&journal->j_list_lock);
1051 /* 1017 /*
1052 * This is a bit sleazy. We use j_li 1018 * This is a bit sleazy. We use j_list_lock to protect transition
1053 * of a transaction into T_FINISHED s 1019 * of a transaction into T_FINISHED state and calling
1054 * __jbd2_journal_drop_transaction(). 1020 * __jbd2_journal_drop_transaction(). Otherwise we could race with
1055 * other checkpointing code processin 1021 * other checkpointing code processing the transaction...
1056 */ 1022 */
1057 write_lock(&journal->j_state_lock); 1023 write_lock(&journal->j_state_lock);
1058 spin_lock(&journal->j_list_lock); 1024 spin_lock(&journal->j_list_lock);
1059 /* 1025 /*
1060 * Now recheck if some buffers did no 1026 * Now recheck if some buffers did not get attached to the transaction
1061 * while the lock was dropped... 1027 * while the lock was dropped...
1062 */ 1028 */
1063 if (commit_transaction->t_forget) { 1029 if (commit_transaction->t_forget) {
1064 spin_unlock(&journal->j_list_ 1030 spin_unlock(&journal->j_list_lock);
1065 write_unlock(&journal->j_stat 1031 write_unlock(&journal->j_state_lock);
1066 goto restart_loop; 1032 goto restart_loop;
1067 } 1033 }
1068 1034
1069 /* Add the transaction to the checkpo 1035 /* Add the transaction to the checkpoint list
1070 * __journal_remove_checkpoint() can 1036 * __journal_remove_checkpoint() can not destroy transaction
1071 * under us because it is not marked 1037 * under us because it is not marked as T_FINISHED yet */
1072 if (journal->j_checkpoint_transaction 1038 if (journal->j_checkpoint_transactions == NULL) {
1073 journal->j_checkpoint_transac 1039 journal->j_checkpoint_transactions = commit_transaction;
1074 commit_transaction->t_cpnext 1040 commit_transaction->t_cpnext = commit_transaction;
1075 commit_transaction->t_cpprev 1041 commit_transaction->t_cpprev = commit_transaction;
1076 } else { 1042 } else {
1077 commit_transaction->t_cpnext 1043 commit_transaction->t_cpnext =
1078 journal->j_checkpoint 1044 journal->j_checkpoint_transactions;
1079 commit_transaction->t_cpprev 1045 commit_transaction->t_cpprev =
1080 commit_transaction->t 1046 commit_transaction->t_cpnext->t_cpprev;
1081 commit_transaction->t_cpnext- 1047 commit_transaction->t_cpnext->t_cpprev =
1082 commit_transaction; 1048 commit_transaction;
1083 commit_transaction->t_cpprev- 1049 commit_transaction->t_cpprev->t_cpnext =
1084 commit_transa 1050 commit_transaction;
1085 } 1051 }
1086 spin_unlock(&journal->j_list_lock); 1052 spin_unlock(&journal->j_list_lock);
1087 1053
1088 /* Done with this transaction! */ 1054 /* Done with this transaction! */
1089 1055
1090 jbd2_debug(3, "JBD2: commit phase 7\n !! 1056 jbd_debug(3, "JBD2: commit phase 7\n");
1091 1057
1092 J_ASSERT(commit_transaction->t_state 1058 J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
1093 1059
1094 commit_transaction->t_start = jiffies 1060 commit_transaction->t_start = jiffies;
1095 stats.run.rs_logging = jbd2_time_diff 1061 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
1096 1062 commit_transaction->t_start);
1097 1063
1098 /* 1064 /*
1099 * File the transaction statistics 1065 * File the transaction statistics
1100 */ 1066 */
1101 stats.ts_tid = commit_transaction->t_ 1067 stats.ts_tid = commit_transaction->t_tid;
1102 stats.run.rs_handle_count = 1068 stats.run.rs_handle_count =
1103 atomic_read(&commit_transacti 1069 atomic_read(&commit_transaction->t_handle_count);
1104 trace_jbd2_run_stats(journal->j_fs_de 1070 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
1105 commit_transacti 1071 commit_transaction->t_tid, &stats.run);
1106 stats.ts_requested = (commit_transact 1072 stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0;
1107 1073
1108 commit_transaction->t_state = T_COMMI 1074 commit_transaction->t_state = T_COMMIT_CALLBACK;
1109 J_ASSERT(commit_transaction == journa 1075 J_ASSERT(commit_transaction == journal->j_committing_transaction);
1110 WRITE_ONCE(journal->j_commit_sequence !! 1076 journal->j_commit_sequence = commit_transaction->t_tid;
1111 journal->j_committing_transaction = N 1077 journal->j_committing_transaction = NULL;
1112 commit_time = ktime_to_ns(ktime_sub(k 1078 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1113 1079
1114 /* 1080 /*
1115 * weight the commit time higher than 1081 * weight the commit time higher than the average time so we don't
1116 * react too strongly to vast changes 1082 * react too strongly to vast changes in the commit time
1117 */ 1083 */
1118 if (likely(journal->j_average_commit_ 1084 if (likely(journal->j_average_commit_time))
1119 journal->j_average_commit_tim 1085 journal->j_average_commit_time = (commit_time +
1120 journal->j_av 1086 journal->j_average_commit_time*3) / 4;
1121 else 1087 else
1122 journal->j_average_commit_tim 1088 journal->j_average_commit_time = commit_time;
1123 1089
1124 write_unlock(&journal->j_state_lock); 1090 write_unlock(&journal->j_state_lock);
1125 1091
1126 if (journal->j_commit_callback) 1092 if (journal->j_commit_callback)
1127 journal->j_commit_callback(jo 1093 journal->j_commit_callback(journal, commit_transaction);
1128 if (journal->j_fc_cleanup_callback) <<
1129 journal->j_fc_cleanup_callbac <<
1130 1094
1131 trace_jbd2_end_commit(journal, commit 1095 trace_jbd2_end_commit(journal, commit_transaction);
1132 jbd2_debug(1, "JBD2: commit %d comple !! 1096 jbd_debug(1, "JBD2: commit %d complete, head %d\n",
1133 journal->j_commit_sequence, 1097 journal->j_commit_sequence, journal->j_tail_sequence);
1134 1098
1135 write_lock(&journal->j_state_lock); 1099 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); 1100 spin_lock(&journal->j_list_lock);
1139 commit_transaction->t_state = T_FINIS 1101 commit_transaction->t_state = T_FINISHED;
1140 /* Check if the transaction can be dr 1102 /* Check if the transaction can be dropped now that we are finished */
1141 if (commit_transaction->t_checkpoint_ !! 1103 if (commit_transaction->t_checkpoint_list == NULL &&
>> 1104 commit_transaction->t_checkpoint_io_list == NULL) {
1142 __jbd2_journal_drop_transacti 1105 __jbd2_journal_drop_transaction(journal, commit_transaction);
1143 jbd2_journal_free_transaction 1106 jbd2_journal_free_transaction(commit_transaction);
1144 } 1107 }
1145 spin_unlock(&journal->j_list_lock); 1108 spin_unlock(&journal->j_list_lock);
1146 write_unlock(&journal->j_state_lock); 1109 write_unlock(&journal->j_state_lock);
1147 wake_up(&journal->j_wait_done_commit) 1110 wake_up(&journal->j_wait_done_commit);
1148 wake_up(&journal->j_fc_wait); <<
1149 1111
1150 /* 1112 /*
1151 * Calculate overall stats 1113 * Calculate overall stats
1152 */ 1114 */
1153 spin_lock(&journal->j_history_lock); 1115 spin_lock(&journal->j_history_lock);
1154 journal->j_stats.ts_tid++; 1116 journal->j_stats.ts_tid++;
1155 journal->j_stats.ts_requested += stat 1117 journal->j_stats.ts_requested += stats.ts_requested;
1156 journal->j_stats.run.rs_wait += stats 1118 journal->j_stats.run.rs_wait += stats.run.rs_wait;
1157 journal->j_stats.run.rs_request_delay 1119 journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
1158 journal->j_stats.run.rs_running += st 1120 journal->j_stats.run.rs_running += stats.run.rs_running;
1159 journal->j_stats.run.rs_locked += sta 1121 journal->j_stats.run.rs_locked += stats.run.rs_locked;
1160 journal->j_stats.run.rs_flushing += s 1122 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
1161 journal->j_stats.run.rs_logging += st 1123 journal->j_stats.run.rs_logging += stats.run.rs_logging;
1162 journal->j_stats.run.rs_handle_count 1124 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
1163 journal->j_stats.run.rs_blocks += sta 1125 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
1164 journal->j_stats.run.rs_blocks_logged 1126 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
1165 spin_unlock(&journal->j_history_lock) 1127 spin_unlock(&journal->j_history_lock);
1166 } 1128 }
1167 1129
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.