1 // SPDX-License-Identifier: GPL-2.0+ <<
2 /* 1 /*
3 * linux/fs/jbd2/commit.c 2 * linux/fs/jbd2/commit.c
4 * 3 *
5 * Written by Stephen C. Tweedie <sct@redhat.c 4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * 5 *
7 * Copyright 1998 Red Hat corp --- All Rights 6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * 7 *
>> 8 * This file is part of the Linux kernel and is made available under
>> 9 * the terms of the GNU General Public License, version 2, or at your
>> 10 * option, any later version, incorporated herein by reference.
>> 11 *
9 * Journal commit routines for the generic fil 12 * Journal commit routines for the generic filesystem journaling code;
10 * part of the ext2fs journaling system. 13 * part of the ext2fs journaling system.
11 */ 14 */
12 15
13 #include <linux/time.h> 16 #include <linux/time.h>
14 #include <linux/fs.h> 17 #include <linux/fs.h>
15 #include <linux/jbd2.h> 18 #include <linux/jbd2.h>
16 #include <linux/errno.h> 19 #include <linux/errno.h>
17 #include <linux/slab.h> 20 #include <linux/slab.h>
18 #include <linux/mm.h> 21 #include <linux/mm.h>
19 #include <linux/pagemap.h> 22 #include <linux/pagemap.h>
20 #include <linux/jiffies.h> 23 #include <linux/jiffies.h>
21 #include <linux/crc32.h> 24 #include <linux/crc32.h>
22 #include <linux/writeback.h> 25 #include <linux/writeback.h>
23 #include <linux/backing-dev.h> 26 #include <linux/backing-dev.h>
24 #include <linux/bio.h> 27 #include <linux/bio.h>
25 #include <linux/blkdev.h> 28 #include <linux/blkdev.h>
26 #include <linux/bitops.h> 29 #include <linux/bitops.h>
27 #include <trace/events/jbd2.h> 30 #include <trace/events/jbd2.h>
28 31
29 /* 32 /*
30 * IO end handler for temporary buffer_heads h 33 * IO end handler for temporary buffer_heads handling writes to the journal.
31 */ 34 */
32 static void journal_end_buffer_io_sync(struct 35 static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
33 { 36 {
34 struct buffer_head *orig_bh = bh->b_pr 37 struct buffer_head *orig_bh = bh->b_private;
35 38
36 BUFFER_TRACE(bh, ""); 39 BUFFER_TRACE(bh, "");
37 if (uptodate) 40 if (uptodate)
38 set_buffer_uptodate(bh); 41 set_buffer_uptodate(bh);
39 else 42 else
40 clear_buffer_uptodate(bh); 43 clear_buffer_uptodate(bh);
41 if (orig_bh) { 44 if (orig_bh) {
42 clear_bit_unlock(BH_Shadow, &o 45 clear_bit_unlock(BH_Shadow, &orig_bh->b_state);
43 smp_mb__after_atomic(); 46 smp_mb__after_atomic();
44 wake_up_bit(&orig_bh->b_state, 47 wake_up_bit(&orig_bh->b_state, BH_Shadow);
45 } 48 }
46 unlock_buffer(bh); 49 unlock_buffer(bh);
47 } 50 }
48 51
49 /* 52 /*
50 * When an ext4 file is truncated, it is possi 53 * When an ext4 file is truncated, it is possible that some pages are not
51 * successfully freed, because they are attach 54 * successfully freed, because they are attached to a committing transaction.
52 * After the transaction commits, these pages 55 * After the transaction commits, these pages are left on the LRU, with no
53 * ->mapping, and with attached buffers. Thes 56 * ->mapping, and with attached buffers. These pages are trivially reclaimable
54 * by the VM, but their apparent absence upset 57 * by the VM, but their apparent absence upsets the VM accounting, and it makes
55 * the numbers in /proc/meminfo look odd. 58 * the numbers in /proc/meminfo look odd.
56 * 59 *
57 * So here, we have a buffer which has just co 60 * 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 61 * see if we can strip all buffers from the backing page.
59 * 62 *
60 * Called under lock_journal(), and possibly u 63 * Called under lock_journal(), and possibly under journal_datalist_lock. The
61 * caller provided us with a ref against the b 64 * caller provided us with a ref against the buffer, and we drop that here.
62 */ 65 */
63 static void release_buffer_page(struct buffer_ 66 static void release_buffer_page(struct buffer_head *bh)
64 { 67 {
65 struct folio *folio; !! 68 struct page *page;
66 69
67 if (buffer_dirty(bh)) 70 if (buffer_dirty(bh))
68 goto nope; 71 goto nope;
69 if (atomic_read(&bh->b_count) != 1) 72 if (atomic_read(&bh->b_count) != 1)
70 goto nope; 73 goto nope;
71 folio = bh->b_folio; !! 74 page = bh->b_page;
72 if (folio->mapping) !! 75 if (!page)
>> 76 goto nope;
>> 77 if (page->mapping)
73 goto nope; 78 goto nope;
74 79
75 /* OK, it's a truncated page */ 80 /* OK, it's a truncated page */
76 if (!folio_trylock(folio)) !! 81 if (!trylock_page(page))
77 goto nope; 82 goto nope;
78 83
79 folio_get(folio); !! 84 get_page(page);
80 __brelse(bh); 85 __brelse(bh);
81 try_to_free_buffers(folio); !! 86 try_to_free_buffers(page);
82 folio_unlock(folio); !! 87 unlock_page(page);
83 folio_put(folio); !! 88 put_page(page);
84 return; 89 return;
85 90
86 nope: 91 nope:
87 __brelse(bh); 92 __brelse(bh);
88 } 93 }
89 94
90 static void jbd2_commit_block_csum_set(journal 95 static void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
91 { 96 {
92 struct commit_header *h; 97 struct commit_header *h;
93 __u32 csum; 98 __u32 csum;
94 99
95 if (!jbd2_journal_has_csum_v2or3(j)) 100 if (!jbd2_journal_has_csum_v2or3(j))
96 return; 101 return;
97 102
98 h = (struct commit_header *)(bh->b_dat 103 h = (struct commit_header *)(bh->b_data);
99 h->h_chksum_type = 0; 104 h->h_chksum_type = 0;
100 h->h_chksum_size = 0; 105 h->h_chksum_size = 0;
101 h->h_chksum[0] = 0; 106 h->h_chksum[0] = 0;
102 csum = jbd2_chksum(j, j->j_csum_seed, 107 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
103 h->h_chksum[0] = cpu_to_be32(csum); 108 h->h_chksum[0] = cpu_to_be32(csum);
104 } 109 }
105 110
106 /* 111 /*
107 * Done it all: now submit the commit record. 112 * Done it all: now submit the commit record. We should have
108 * cleaned up our previous buffers by now, so 113 * 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 114 * mode we can now just skip the rest of the journal write
110 * entirely. 115 * entirely.
111 * 116 *
112 * Returns 1 if the journal needs to be aborte 117 * Returns 1 if the journal needs to be aborted or 0 on success
113 */ 118 */
114 static int journal_submit_commit_record(journa 119 static int journal_submit_commit_record(journal_t *journal,
115 transa 120 transaction_t *commit_transaction,
116 struct 121 struct buffer_head **cbh,
117 __u32 122 __u32 crc32_sum)
118 { 123 {
119 struct commit_header *tmp; 124 struct commit_header *tmp;
120 struct buffer_head *bh; 125 struct buffer_head *bh;
121 struct timespec64 now; !! 126 int ret;
122 blk_opf_t write_flags = REQ_OP_WRITE | !! 127 struct timespec64 now = current_kernel_time64();
123 128
124 *cbh = NULL; 129 *cbh = NULL;
125 130
126 if (is_journal_aborted(journal)) 131 if (is_journal_aborted(journal))
127 return 0; 132 return 0;
128 133
129 bh = jbd2_journal_get_descriptor_buffe 134 bh = jbd2_journal_get_descriptor_buffer(commit_transaction,
130 135 JBD2_COMMIT_BLOCK);
131 if (!bh) 136 if (!bh)
132 return 1; 137 return 1;
133 138
134 tmp = (struct commit_header *)bh->b_da 139 tmp = (struct commit_header *)bh->b_data;
135 ktime_get_coarse_real_ts64(&now); <<
136 tmp->h_commit_sec = cpu_to_be64(now.tv 140 tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
137 tmp->h_commit_nsec = cpu_to_be32(now.t 141 tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
138 142
139 if (jbd2_has_feature_checksum(journal) 143 if (jbd2_has_feature_checksum(journal)) {
140 tmp->h_chksum_type = JBD2 144 tmp->h_chksum_type = JBD2_CRC32_CHKSUM;
141 tmp->h_chksum_size = JBD2 145 tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE;
142 tmp->h_chksum[0] = cpu_ 146 tmp->h_chksum[0] = cpu_to_be32(crc32_sum);
143 } 147 }
144 jbd2_commit_block_csum_set(journal, bh 148 jbd2_commit_block_csum_set(journal, bh);
145 149
146 BUFFER_TRACE(bh, "submit commit block" 150 BUFFER_TRACE(bh, "submit commit block");
147 lock_buffer(bh); 151 lock_buffer(bh);
148 clear_buffer_dirty(bh); 152 clear_buffer_dirty(bh);
149 set_buffer_uptodate(bh); 153 set_buffer_uptodate(bh);
150 bh->b_end_io = journal_end_buffer_io_s 154 bh->b_end_io = journal_end_buffer_io_sync;
151 155
152 if (journal->j_flags & JBD2_BARRIER && 156 if (journal->j_flags & JBD2_BARRIER &&
153 !jbd2_has_feature_async_commit(jou 157 !jbd2_has_feature_async_commit(journal))
154 write_flags |= REQ_PREFLUSH | !! 158 ret = submit_bh(REQ_OP_WRITE,
>> 159 REQ_SYNC | REQ_PREFLUSH | REQ_FUA, bh);
>> 160 else
>> 161 ret = submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
155 162
156 submit_bh(write_flags, bh); <<
157 *cbh = bh; 163 *cbh = bh;
158 return 0; !! 164 return ret;
159 } 165 }
160 166
161 /* 167 /*
162 * This function along with journal_submit_com 168 * This function along with journal_submit_commit_record
163 * allows to write the commit record asynchron 169 * allows to write the commit record asynchronously.
164 */ 170 */
165 static int journal_wait_on_commit_record(journ 171 static int journal_wait_on_commit_record(journal_t *journal,
166 struc 172 struct buffer_head *bh)
167 { 173 {
168 int ret = 0; 174 int ret = 0;
169 175
170 clear_buffer_dirty(bh); 176 clear_buffer_dirty(bh);
171 wait_on_buffer(bh); 177 wait_on_buffer(bh);
172 178
173 if (unlikely(!buffer_uptodate(bh))) 179 if (unlikely(!buffer_uptodate(bh)))
174 ret = -EIO; 180 ret = -EIO;
175 put_bh(bh); /* One for getb 181 put_bh(bh); /* One for getblk() */
176 182
177 return ret; 183 return ret;
178 } 184 }
179 185
180 /* Send all the data buffers related to an ino !! 186 /*
181 int jbd2_submit_inode_data(journal_t *journal, !! 187 * write the filemap data using writepage() address_space_operations.
>> 188 * We don't do block allocation here even for delalloc. We don't
>> 189 * use writepages() because with dealyed allocation we may be doing
>> 190 * block allocation in writepages().
>> 191 */
>> 192 static int journal_submit_inode_data_buffers(struct address_space *mapping)
182 { 193 {
183 if (!jinode || !(jinode->i_flags & JI_ !! 194 int ret;
184 return 0; !! 195 struct writeback_control wbc = {
>> 196 .sync_mode = WB_SYNC_ALL,
>> 197 .nr_to_write = mapping->nrpages * 2,
>> 198 .range_start = 0,
>> 199 .range_end = i_size_read(mapping->host),
>> 200 };
185 201
186 trace_jbd2_submit_inode_data(jinode->i !! 202 ret = generic_writepages(mapping, &wbc);
187 return journal->j_submit_inode_data_bu !! 203 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 } 204 }
201 EXPORT_SYMBOL(jbd2_wait_inode_data); <<
202 205
203 /* 206 /*
204 * Submit all the data buffers of inode associ 207 * Submit all the data buffers of inode associated with the transaction to
205 * disk. 208 * disk.
206 * 209 *
207 * We are in a committing transaction. Therefo 210 * We are in a committing transaction. Therefore no new inode can be added to
208 * our inode list. We use JI_COMMIT_RUNNING fl 211 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
209 * operate on from being released while we wri 212 * operate on from being released while we write out pages.
210 */ 213 */
211 static int journal_submit_data_buffers(journal 214 static int journal_submit_data_buffers(journal_t *journal,
212 transaction_t *commit_transact 215 transaction_t *commit_transaction)
213 { 216 {
214 struct jbd2_inode *jinode; 217 struct jbd2_inode *jinode;
215 int err, ret = 0; 218 int err, ret = 0;
>> 219 struct address_space *mapping;
216 220
217 spin_lock(&journal->j_list_lock); 221 spin_lock(&journal->j_list_lock);
218 list_for_each_entry(jinode, &commit_tr 222 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
219 if (!(jinode->i_flags & JI_WRI 223 if (!(jinode->i_flags & JI_WRITE_DATA))
220 continue; 224 continue;
>> 225 mapping = jinode->i_vfs_inode->i_mapping;
221 jinode->i_flags |= JI_COMMIT_R 226 jinode->i_flags |= JI_COMMIT_RUNNING;
222 spin_unlock(&journal->j_list_l 227 spin_unlock(&journal->j_list_lock);
223 /* submit the inode data buffe !! 228 /*
>> 229 * submit the inode data buffers. We use writepage
>> 230 * instead of writepages. Because writepages can do
>> 231 * block allocation with delalloc. We need to write
>> 232 * only allocated blocks here.
>> 233 */
224 trace_jbd2_submit_inode_data(j 234 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
225 if (journal->j_submit_inode_da !! 235 err = journal_submit_inode_data_buffers(mapping);
226 err = journal->j_submi !! 236 if (!ret)
227 if (!ret) !! 237 ret = err;
228 ret = err; <<
229 } <<
230 spin_lock(&journal->j_list_loc 238 spin_lock(&journal->j_list_lock);
231 J_ASSERT(jinode->i_transaction 239 J_ASSERT(jinode->i_transaction == commit_transaction);
232 jinode->i_flags &= ~JI_COMMIT_ 240 jinode->i_flags &= ~JI_COMMIT_RUNNING;
233 smp_mb(); 241 smp_mb();
234 wake_up_bit(&jinode->i_flags, 242 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
235 } 243 }
236 spin_unlock(&journal->j_list_lock); 244 spin_unlock(&journal->j_list_lock);
237 return ret; 245 return ret;
238 } 246 }
239 247
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 /* 248 /*
250 * Wait for data submitted for writeout, refil 249 * Wait for data submitted for writeout, refile inodes to proper
251 * transaction if needed. 250 * transaction if needed.
252 * 251 *
253 */ 252 */
254 static int journal_finish_inode_data_buffers(j 253 static int journal_finish_inode_data_buffers(journal_t *journal,
255 transaction_t *commit_transact 254 transaction_t *commit_transaction)
256 { 255 {
257 struct jbd2_inode *jinode, *next_i; 256 struct jbd2_inode *jinode, *next_i;
258 int err, ret = 0; 257 int err, ret = 0;
259 258
260 /* For locking, see the comment in jou 259 /* For locking, see the comment in journal_submit_data_buffers() */
261 spin_lock(&journal->j_list_lock); 260 spin_lock(&journal->j_list_lock);
262 list_for_each_entry(jinode, &commit_tr 261 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
263 if (!(jinode->i_flags & JI_WAI 262 if (!(jinode->i_flags & JI_WAIT_DATA))
264 continue; 263 continue;
265 jinode->i_flags |= JI_COMMIT_R 264 jinode->i_flags |= JI_COMMIT_RUNNING;
266 spin_unlock(&journal->j_list_l 265 spin_unlock(&journal->j_list_lock);
267 /* wait for the inode data buf !! 266 err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping);
268 if (journal->j_finish_inode_da !! 267 if (err) {
269 err = journal->j_finis !! 268 /*
>> 269 * Because AS_EIO is cleared by
>> 270 * filemap_fdatawait_range(), set it again so
>> 271 * that user process can get -EIO from fsync().
>> 272 */
>> 273 mapping_set_error(jinode->i_vfs_inode->i_mapping, -EIO);
>> 274
270 if (!ret) 275 if (!ret)
271 ret = err; 276 ret = err;
272 } 277 }
273 cond_resched(); <<
274 spin_lock(&journal->j_list_loc 278 spin_lock(&journal->j_list_lock);
275 jinode->i_flags &= ~JI_COMMIT_ 279 jinode->i_flags &= ~JI_COMMIT_RUNNING;
276 smp_mb(); 280 smp_mb();
277 wake_up_bit(&jinode->i_flags, 281 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
278 } 282 }
279 283
280 /* Now refile inode to proper lists */ 284 /* Now refile inode to proper lists */
281 list_for_each_entry_safe(jinode, next_ 285 list_for_each_entry_safe(jinode, next_i,
282 &commit_trans 286 &commit_transaction->t_inode_list, i_list) {
283 list_del(&jinode->i_list); 287 list_del(&jinode->i_list);
284 if (jinode->i_next_transaction 288 if (jinode->i_next_transaction) {
285 jinode->i_transaction 289 jinode->i_transaction = jinode->i_next_transaction;
286 jinode->i_next_transac 290 jinode->i_next_transaction = NULL;
287 list_add(&jinode->i_li 291 list_add(&jinode->i_list,
288 &jinode->i_tra 292 &jinode->i_transaction->t_inode_list);
289 } else { 293 } else {
290 jinode->i_transaction 294 jinode->i_transaction = NULL;
291 jinode->i_dirty_start <<
292 jinode->i_dirty_end = <<
293 } 295 }
294 } 296 }
295 spin_unlock(&journal->j_list_lock); 297 spin_unlock(&journal->j_list_lock);
296 298
297 return ret; 299 return ret;
298 } 300 }
299 301
300 static __u32 jbd2_checksum_data(__u32 crc32_su 302 static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
301 { 303 {
>> 304 struct page *page = bh->b_page;
302 char *addr; 305 char *addr;
303 __u32 checksum; 306 __u32 checksum;
304 307
305 addr = kmap_local_folio(bh->b_folio, b !! 308 addr = kmap_atomic(page);
306 checksum = crc32_be(crc32_sum, addr, b !! 309 checksum = crc32_be(crc32_sum,
307 kunmap_local(addr); !! 310 (void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
>> 311 kunmap_atomic(addr);
308 312
309 return checksum; 313 return checksum;
310 } 314 }
311 315
312 static void write_tag_block(journal_t *j, jour 316 static void write_tag_block(journal_t *j, journal_block_tag_t *tag,
313 unsigned lo 317 unsigned long long block)
314 { 318 {
315 tag->t_blocknr = cpu_to_be32(block & ( 319 tag->t_blocknr = cpu_to_be32(block & (u32)~0);
316 if (jbd2_has_feature_64bit(j)) 320 if (jbd2_has_feature_64bit(j))
317 tag->t_blocknr_high = cpu_to_b 321 tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
318 } 322 }
319 323
320 static void jbd2_block_tag_csum_set(journal_t 324 static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
321 struct buf 325 struct buffer_head *bh, __u32 sequence)
322 { 326 {
323 journal_block_tag3_t *tag3 = (journal_ 327 journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
>> 328 struct page *page = bh->b_page;
324 __u8 *addr; 329 __u8 *addr;
325 __u32 csum32; 330 __u32 csum32;
326 __be32 seq; 331 __be32 seq;
327 332
328 if (!jbd2_journal_has_csum_v2or3(j)) 333 if (!jbd2_journal_has_csum_v2or3(j))
329 return; 334 return;
330 335
331 seq = cpu_to_be32(sequence); 336 seq = cpu_to_be32(sequence);
332 addr = kmap_local_folio(bh->b_folio, b !! 337 addr = kmap_atomic(page);
333 csum32 = jbd2_chksum(j, j->j_csum_seed 338 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
334 csum32 = jbd2_chksum(j, csum32, addr, !! 339 csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data),
335 kunmap_local(addr); !! 340 bh->b_size);
>> 341 kunmap_atomic(addr);
336 342
337 if (jbd2_has_feature_csum3(j)) 343 if (jbd2_has_feature_csum3(j))
338 tag3->t_checksum = cpu_to_be32 344 tag3->t_checksum = cpu_to_be32(csum32);
339 else 345 else
340 tag->t_checksum = cpu_to_be16( 346 tag->t_checksum = cpu_to_be16(csum32);
341 } 347 }
342 /* 348 /*
343 * jbd2_journal_commit_transaction 349 * jbd2_journal_commit_transaction
344 * 350 *
345 * The primary function for committing a trans 351 * The primary function for committing a transaction to the log. This
346 * function is called by the journal thread to 352 * function is called by the journal thread to begin a complete commit.
347 */ 353 */
348 void jbd2_journal_commit_transaction(journal_t 354 void jbd2_journal_commit_transaction(journal_t *journal)
349 { 355 {
350 struct transaction_stats_s stats; 356 struct transaction_stats_s stats;
351 transaction_t *commit_transaction; 357 transaction_t *commit_transaction;
352 struct journal_head *jh; 358 struct journal_head *jh;
353 struct buffer_head *descriptor; 359 struct buffer_head *descriptor;
354 struct buffer_head **wbuf = journal->j 360 struct buffer_head **wbuf = journal->j_wbuf;
355 int bufs; 361 int bufs;
356 int escape; !! 362 int flags;
357 int err; 363 int err;
358 unsigned long long blocknr; 364 unsigned long long blocknr;
359 ktime_t start_time; 365 ktime_t start_time;
360 u64 commit_time; 366 u64 commit_time;
361 char *tagp = NULL; 367 char *tagp = NULL;
362 journal_block_tag_t *tag = NULL; 368 journal_block_tag_t *tag = NULL;
363 int space_left = 0; 369 int space_left = 0;
364 int first_tag = 0; 370 int first_tag = 0;
365 int tag_flag; 371 int tag_flag;
366 int i; 372 int i;
367 int tag_bytes = journal_tag_bytes(jour 373 int tag_bytes = journal_tag_bytes(journal);
368 struct buffer_head *cbh = NULL; /* For 374 struct buffer_head *cbh = NULL; /* For transactional checksums */
369 __u32 crc32_sum = ~0; 375 __u32 crc32_sum = ~0;
370 struct blk_plug plug; 376 struct blk_plug plug;
371 /* Tail of the journal */ 377 /* Tail of the journal */
372 unsigned long first_block; 378 unsigned long first_block;
373 tid_t first_tid; 379 tid_t first_tid;
374 int update_tail; 380 int update_tail;
375 int csum_size = 0; 381 int csum_size = 0;
376 LIST_HEAD(io_bufs); 382 LIST_HEAD(io_bufs);
377 LIST_HEAD(log_bufs); 383 LIST_HEAD(log_bufs);
378 384
379 if (jbd2_journal_has_csum_v2or3(journa 385 if (jbd2_journal_has_csum_v2or3(journal))
380 csum_size = sizeof(struct jbd2 386 csum_size = sizeof(struct jbd2_journal_block_tail);
381 387
382 /* 388 /*
383 * First job: lock down the current tr 389 * First job: lock down the current transaction and wait for
384 * all outstanding updates to complete 390 * all outstanding updates to complete.
385 */ 391 */
386 392
387 /* Do we need to erase the effects of 393 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
388 if (journal->j_flags & JBD2_FLUSHED) { 394 if (journal->j_flags & JBD2_FLUSHED) {
389 jbd2_debug(3, "super block upd !! 395 jbd_debug(3, "super block updated\n");
390 mutex_lock_io(&journal->j_chec !! 396 mutex_lock(&journal->j_checkpoint_mutex);
391 /* 397 /*
392 * We hold j_checkpoint_mutex 398 * We hold j_checkpoint_mutex so tail cannot change under us.
393 * We don't need any special d 399 * We don't need any special data guarantees for writing sb
394 * since journal is empty and 400 * since journal is empty and it is ok for write to be
395 * flushed only with transacti 401 * flushed only with transaction commit.
396 */ 402 */
397 jbd2_journal_update_sb_log_tai 403 jbd2_journal_update_sb_log_tail(journal,
398 404 journal->j_tail_sequence,
399 !! 405 journal->j_tail,
>> 406 REQ_SYNC);
400 mutex_unlock(&journal->j_check 407 mutex_unlock(&journal->j_checkpoint_mutex);
401 } else { 408 } else {
402 jbd2_debug(3, "superblock not !! 409 jbd_debug(3, "superblock not updated\n");
403 } 410 }
404 411
405 J_ASSERT(journal->j_running_transactio 412 J_ASSERT(journal->j_running_transaction != NULL);
406 J_ASSERT(journal->j_committing_transac 413 J_ASSERT(journal->j_committing_transaction == NULL);
407 414
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 415 commit_transaction = journal->j_running_transaction;
432 416
433 trace_jbd2_start_commit(journal, commi 417 trace_jbd2_start_commit(journal, commit_transaction);
434 jbd2_debug(1, "JBD2: starting commit o !! 418 jbd_debug(1, "JBD2: starting commit of transaction %d\n",
435 commit_transaction->t_ 419 commit_transaction->t_tid);
436 420
437 write_lock(&journal->j_state_lock); 421 write_lock(&journal->j_state_lock);
438 journal->j_fc_off = 0; <<
439 J_ASSERT(commit_transaction->t_state = 422 J_ASSERT(commit_transaction->t_state == T_RUNNING);
440 commit_transaction->t_state = T_LOCKED 423 commit_transaction->t_state = T_LOCKED;
441 424
442 trace_jbd2_commit_locking(journal, com 425 trace_jbd2_commit_locking(journal, commit_transaction);
443 stats.run.rs_wait = commit_transaction 426 stats.run.rs_wait = commit_transaction->t_max_wait;
444 stats.run.rs_request_delay = 0; 427 stats.run.rs_request_delay = 0;
445 stats.run.rs_locked = jiffies; 428 stats.run.rs_locked = jiffies;
446 if (commit_transaction->t_requested) 429 if (commit_transaction->t_requested)
447 stats.run.rs_request_delay = 430 stats.run.rs_request_delay =
448 jbd2_time_diff(commit_ 431 jbd2_time_diff(commit_transaction->t_requested,
449 stats.r 432 stats.run.rs_locked);
450 stats.run.rs_running = jbd2_time_diff( 433 stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
451 434 stats.run.rs_locked);
452 435
453 // waits for any t_updates to finish !! 436 spin_lock(&commit_transaction->t_handle_lock);
454 jbd2_journal_wait_updates(journal); !! 437 while (atomic_read(&commit_transaction->t_updates)) {
>> 438 DEFINE_WAIT(wait);
455 439
456 commit_transaction->t_state = T_SWITCH !! 440 prepare_to_wait(&journal->j_wait_updates, &wait,
>> 441 TASK_UNINTERRUPTIBLE);
>> 442 if (atomic_read(&commit_transaction->t_updates)) {
>> 443 spin_unlock(&commit_transaction->t_handle_lock);
>> 444 write_unlock(&journal->j_state_lock);
>> 445 schedule();
>> 446 write_lock(&journal->j_state_lock);
>> 447 spin_lock(&commit_transaction->t_handle_lock);
>> 448 }
>> 449 finish_wait(&journal->j_wait_updates, &wait);
>> 450 }
>> 451 spin_unlock(&commit_transaction->t_handle_lock);
457 452
458 J_ASSERT (atomic_read(&commit_transact 453 J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
459 journal->j_max_transac 454 journal->j_max_transaction_buffers);
460 455
461 /* 456 /*
462 * First thing we are allowed to do is 457 * First thing we are allowed to do is to discard any remaining
463 * BJ_Reserved buffers. Note, it is _ 458 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
464 * that there are no such buffers: if 459 * that there are no such buffers: if a large filesystem
465 * operation like a truncate needs to 460 * operation like a truncate needs to split itself over multiple
466 * transactions, then it may try to do 461 * transactions, then it may try to do a jbd2_journal_restart() while
467 * there are still BJ_Reserved buffers 462 * there are still BJ_Reserved buffers outstanding. These must
468 * be released cleanly from the curren 463 * be released cleanly from the current transaction.
469 * 464 *
470 * In this case, the filesystem must s 465 * In this case, the filesystem must still reserve write access
471 * again before modifying the buffer i 466 * again before modifying the buffer in the new transaction, but
472 * we do not require it to remember ex 467 * we do not require it to remember exactly which old buffers it
473 * has reserved. This is consistent w 468 * has reserved. This is consistent with the existing behaviour
474 * that multiple jbd2_journal_get_writ 469 * that multiple jbd2_journal_get_write_access() calls to the same
475 * buffer are perfectly permissible. 470 * buffer are perfectly permissible.
476 * We use journal->j_state_lock here t <<
477 * t_reserved_list with eviction of bu <<
478 */ 471 */
479 while (commit_transaction->t_reserved_ 472 while (commit_transaction->t_reserved_list) {
480 jh = commit_transaction->t_res 473 jh = commit_transaction->t_reserved_list;
481 JBUFFER_TRACE(jh, "reserved, u 474 JBUFFER_TRACE(jh, "reserved, unused: refile");
482 /* 475 /*
483 * A jbd2_journal_get_undo_acc 476 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
484 * leave undo-committed data. 477 * leave undo-committed data.
485 */ 478 */
486 if (jh->b_committed_data) { 479 if (jh->b_committed_data) {
487 struct buffer_head *bh 480 struct buffer_head *bh = jh2bh(jh);
488 481
489 spin_lock(&jh->b_state !! 482 jbd_lock_bh_state(bh);
490 jbd2_free(jh->b_commit 483 jbd2_free(jh->b_committed_data, bh->b_size);
491 jh->b_committed_data = 484 jh->b_committed_data = NULL;
492 spin_unlock(&jh->b_sta !! 485 jbd_unlock_bh_state(bh);
493 } 486 }
494 jbd2_journal_refile_buffer(jou 487 jbd2_journal_refile_buffer(journal, jh);
495 } 488 }
496 489
497 write_unlock(&journal->j_state_lock); <<
498 /* 490 /*
499 * Now try to drop any written-back bu 491 * Now try to drop any written-back buffers from the journal's
500 * checkpoint lists. We do this *befo 492 * checkpoint lists. We do this *before* commit because it potentially
501 * frees some memory 493 * frees some memory
502 */ 494 */
503 spin_lock(&journal->j_list_lock); 495 spin_lock(&journal->j_list_lock);
504 __jbd2_journal_clean_checkpoint_list(j !! 496 __jbd2_journal_clean_checkpoint_list(journal, false);
505 spin_unlock(&journal->j_list_lock); 497 spin_unlock(&journal->j_list_lock);
506 498
507 jbd2_debug(3, "JBD2: commit phase 1\n" !! 499 jbd_debug(3, "JBD2: commit phase 1\n");
508 500
509 /* 501 /*
510 * Clear revoked flag to reflect there 502 * Clear revoked flag to reflect there is no revoked buffers
511 * in the next transaction which is go 503 * in the next transaction which is going to be started.
512 */ 504 */
513 jbd2_clear_buffer_revoked_flags(journa 505 jbd2_clear_buffer_revoked_flags(journal);
514 506
515 /* 507 /*
516 * Switch to a new revoke table. 508 * Switch to a new revoke table.
517 */ 509 */
518 jbd2_journal_switch_revoke_table(journ 510 jbd2_journal_switch_revoke_table(journal);
519 511
520 write_lock(&journal->j_state_lock); <<
521 /* 512 /*
522 * Reserved credits cannot be claimed 513 * Reserved credits cannot be claimed anymore, free them
523 */ 514 */
524 atomic_sub(atomic_read(&journal->j_res 515 atomic_sub(atomic_read(&journal->j_reserved_credits),
525 &commit_transaction->t_outs 516 &commit_transaction->t_outstanding_credits);
526 517
527 trace_jbd2_commit_flushing(journal, co 518 trace_jbd2_commit_flushing(journal, commit_transaction);
528 stats.run.rs_flushing = jiffies; 519 stats.run.rs_flushing = jiffies;
529 stats.run.rs_locked = jbd2_time_diff(s 520 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
530 s 521 stats.run.rs_flushing);
531 522
532 commit_transaction->t_state = T_FLUSH; 523 commit_transaction->t_state = T_FLUSH;
533 journal->j_committing_transaction = co 524 journal->j_committing_transaction = commit_transaction;
534 journal->j_running_transaction = NULL; 525 journal->j_running_transaction = NULL;
535 start_time = ktime_get(); 526 start_time = ktime_get();
536 commit_transaction->t_log_start = jour 527 commit_transaction->t_log_start = journal->j_head;
537 wake_up_all(&journal->j_wait_transacti !! 528 wake_up(&journal->j_wait_transaction_locked);
538 write_unlock(&journal->j_state_lock); 529 write_unlock(&journal->j_state_lock);
539 530
540 jbd2_debug(3, "JBD2: commit phase 2a\n !! 531 jbd_debug(3, "JBD2: commit phase 2a\n");
541 532
542 /* 533 /*
543 * Now start flushing things to disk, 534 * Now start flushing things to disk, in the order they appear
544 * on the transaction lists. Data blo 535 * on the transaction lists. Data blocks go first.
545 */ 536 */
546 err = journal_submit_data_buffers(jour 537 err = journal_submit_data_buffers(journal, commit_transaction);
547 if (err) 538 if (err)
548 jbd2_journal_abort(journal, er 539 jbd2_journal_abort(journal, err);
549 540
550 blk_start_plug(&plug); 541 blk_start_plug(&plug);
551 jbd2_journal_write_revoke_records(comm 542 jbd2_journal_write_revoke_records(commit_transaction, &log_bufs);
552 543
553 jbd2_debug(3, "JBD2: commit phase 2b\n !! 544 jbd_debug(3, "JBD2: commit phase 2b\n");
554 545
555 /* 546 /*
556 * Way to go: we have now written out 547 * Way to go: we have now written out all of the data for a
557 * transaction! Now comes the tricky 548 * transaction! Now comes the tricky part: we need to write out
558 * metadata. Loop over the transactio 549 * metadata. Loop over the transaction's entire buffer list:
559 */ 550 */
560 write_lock(&journal->j_state_lock); 551 write_lock(&journal->j_state_lock);
561 commit_transaction->t_state = T_COMMIT 552 commit_transaction->t_state = T_COMMIT;
562 write_unlock(&journal->j_state_lock); 553 write_unlock(&journal->j_state_lock);
563 554
564 trace_jbd2_commit_logging(journal, com 555 trace_jbd2_commit_logging(journal, commit_transaction);
565 stats.run.rs_logging = jiffies; 556 stats.run.rs_logging = jiffies;
566 stats.run.rs_flushing = jbd2_time_diff 557 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
567 558 stats.run.rs_logging);
568 stats.run.rs_blocks = commit_transacti !! 559 stats.run.rs_blocks =
>> 560 atomic_read(&commit_transaction->t_outstanding_credits);
569 stats.run.rs_blocks_logged = 0; 561 stats.run.rs_blocks_logged = 0;
570 562
571 J_ASSERT(commit_transaction->t_nr_buff 563 J_ASSERT(commit_transaction->t_nr_buffers <=
572 atomic_read(&commit_transacti 564 atomic_read(&commit_transaction->t_outstanding_credits));
573 565
>> 566 err = 0;
574 bufs = 0; 567 bufs = 0;
575 descriptor = NULL; 568 descriptor = NULL;
576 while (commit_transaction->t_buffers) 569 while (commit_transaction->t_buffers) {
577 570
578 /* Find the next buffer to be 571 /* Find the next buffer to be journaled... */
579 572
580 jh = commit_transaction->t_buf 573 jh = commit_transaction->t_buffers;
581 574
582 /* If we're in abort mode, we 575 /* If we're in abort mode, we just un-journal the buffer and
583 release it. */ 576 release it. */
584 577
585 if (is_journal_aborted(journal 578 if (is_journal_aborted(journal)) {
586 clear_buffer_jbddirty( 579 clear_buffer_jbddirty(jh2bh(jh));
587 JBUFFER_TRACE(jh, "jou 580 JBUFFER_TRACE(jh, "journal is aborting: refile");
588 jbd2_buffer_abort_trig 581 jbd2_buffer_abort_trigger(jh,
589 582 jh->b_frozen_data ?
590 583 jh->b_frozen_triggers :
591 584 jh->b_triggers);
592 jbd2_journal_refile_bu 585 jbd2_journal_refile_buffer(journal, jh);
593 /* If that was the las 586 /* If that was the last one, we need to clean up
594 * any descriptor buff 587 * any descriptor buffers which may have been
595 * already allocated, 588 * already allocated, even if we are now
596 * aborting. */ 589 * aborting. */
597 if (!commit_transactio 590 if (!commit_transaction->t_buffers)
598 goto start_jou 591 goto start_journal_io;
599 continue; 592 continue;
600 } 593 }
601 594
602 /* Make sure we have a descrip 595 /* Make sure we have a descriptor block in which to
603 record the metadata buffer. 596 record the metadata buffer. */
604 597
605 if (!descriptor) { 598 if (!descriptor) {
606 J_ASSERT (bufs == 0); 599 J_ASSERT (bufs == 0);
607 600
608 jbd2_debug(4, "JBD2: g !! 601 jbd_debug(4, "JBD2: get descriptor\n");
609 602
610 descriptor = jbd2_jour 603 descriptor = jbd2_journal_get_descriptor_buffer(
611 604 commit_transaction,
612 605 JBD2_DESCRIPTOR_BLOCK);
613 if (!descriptor) { 606 if (!descriptor) {
614 jbd2_journal_a 607 jbd2_journal_abort(journal, -EIO);
615 continue; 608 continue;
616 } 609 }
617 610
618 jbd2_debug(4, "JBD2: g !! 611 jbd_debug(4, "JBD2: got buffer %llu (%p)\n",
619 (unsigned long 612 (unsigned long long)descriptor->b_blocknr,
620 descriptor->b_ 613 descriptor->b_data);
621 tagp = &descriptor->b_ 614 tagp = &descriptor->b_data[sizeof(journal_header_t)];
622 space_left = descripto 615 space_left = descriptor->b_size -
623 616 sizeof(journal_header_t);
624 first_tag = 1; 617 first_tag = 1;
625 set_buffer_jwrite(desc 618 set_buffer_jwrite(descriptor);
626 set_buffer_dirty(descr 619 set_buffer_dirty(descriptor);
627 wbuf[bufs++] = descrip 620 wbuf[bufs++] = descriptor;
628 621
629 /* Record it so that w 622 /* Record it so that we can wait for IO
630 completion later */ 623 completion later */
631 BUFFER_TRACE(descripto 624 BUFFER_TRACE(descriptor, "ph3: file as descriptor");
632 jbd2_file_log_bh(&log_ 625 jbd2_file_log_bh(&log_bufs, descriptor);
633 } 626 }
634 627
635 /* Where is the buffer to be w 628 /* Where is the buffer to be written? */
636 629
637 err = jbd2_journal_next_log_bl 630 err = jbd2_journal_next_log_block(journal, &blocknr);
638 /* If the block mapping failed 631 /* If the block mapping failed, just abandon the buffer
639 and repeat this loop: we'll 632 and repeat this loop: we'll fall into the
640 refile-on-abort condition a 633 refile-on-abort condition above. */
641 if (err) { 634 if (err) {
642 jbd2_journal_abort(jou 635 jbd2_journal_abort(journal, err);
643 continue; 636 continue;
644 } 637 }
645 638
646 /* 639 /*
647 * start_this_handle() uses t_ 640 * start_this_handle() uses t_outstanding_credits to determine
648 * the free space in the log. !! 641 * the free space in the log, but this counter is changed
>> 642 * by jbd2_journal_next_log_block() also.
649 */ 643 */
650 atomic_dec(&commit_transaction 644 atomic_dec(&commit_transaction->t_outstanding_credits);
651 645
652 /* Bump b_count to prevent tru 646 /* Bump b_count to prevent truncate from stumbling over
653 the shadowed buffer! @@@ T 647 the shadowed buffer! @@@ This can go if we ever get
654 rid of the shadow pairing o 648 rid of the shadow pairing of buffers. */
655 atomic_inc(&jh2bh(jh)->b_count 649 atomic_inc(&jh2bh(jh)->b_count);
656 650
657 /* 651 /*
658 * Make a temporary IO buffer 652 * Make a temporary IO buffer with which to write it out
659 * (this will requeue the meta 653 * (this will requeue the metadata buffer to BJ_Shadow).
660 */ 654 */
661 set_bit(BH_JWrite, &jh2bh(jh)- 655 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
662 JBUFFER_TRACE(jh, "ph3: write 656 JBUFFER_TRACE(jh, "ph3: write metadata");
663 escape = jbd2_journal_write_me !! 657 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
664 658 jh, &wbuf[bufs], blocknr);
665 if (escape < 0) { !! 659 if (flags < 0) {
666 jbd2_journal_abort(jou !! 660 jbd2_journal_abort(journal, flags);
667 continue; 661 continue;
668 } 662 }
669 jbd2_file_log_bh(&io_bufs, wbu 663 jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
670 664
671 /* Record the new block's tag 665 /* Record the new block's tag in the current descriptor
672 buffer */ 666 buffer */
673 667
674 tag_flag = 0; 668 tag_flag = 0;
675 if (escape) !! 669 if (flags & 1)
676 tag_flag |= JBD2_FLAG_ 670 tag_flag |= JBD2_FLAG_ESCAPE;
677 if (!first_tag) 671 if (!first_tag)
678 tag_flag |= JBD2_FLAG_ 672 tag_flag |= JBD2_FLAG_SAME_UUID;
679 673
680 tag = (journal_block_tag_t *) 674 tag = (journal_block_tag_t *) tagp;
681 write_tag_block(journal, tag, 675 write_tag_block(journal, tag, jh2bh(jh)->b_blocknr);
682 tag->t_flags = cpu_to_be16(tag 676 tag->t_flags = cpu_to_be16(tag_flag);
683 jbd2_block_tag_csum_set(journa 677 jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
684 commit 678 commit_transaction->t_tid);
685 tagp += tag_bytes; 679 tagp += tag_bytes;
686 space_left -= tag_bytes; 680 space_left -= tag_bytes;
687 bufs++; 681 bufs++;
688 682
689 if (first_tag) { 683 if (first_tag) {
690 memcpy (tagp, journal- 684 memcpy (tagp, journal->j_uuid, 16);
691 tagp += 16; 685 tagp += 16;
692 space_left -= 16; 686 space_left -= 16;
693 first_tag = 0; 687 first_tag = 0;
694 } 688 }
695 689
696 /* If there's no more to do, o 690 /* If there's no more to do, or if the descriptor is full,
697 let the IO rip! */ 691 let the IO rip! */
698 692
699 if (bufs == journal->j_wbufsiz 693 if (bufs == journal->j_wbufsize ||
700 commit_transaction->t_buff 694 commit_transaction->t_buffers == NULL ||
701 space_left < tag_bytes + 1 695 space_left < tag_bytes + 16 + csum_size) {
702 696
703 jbd2_debug(4, "JBD2: S !! 697 jbd_debug(4, "JBD2: Submit %d IOs\n", bufs);
704 698
705 /* Write an end-of-des 699 /* Write an end-of-descriptor marker before
706 submitting the IOs. 700 submitting the IOs. "tag" still points to
707 the last tag we set 701 the last tag we set up. */
708 702
709 tag->t_flags |= cpu_to 703 tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
710 start_journal_io: <<
711 if (descriptor) <<
712 jbd2_descripto <<
713 <<
714 704
>> 705 jbd2_descriptor_block_csum_set(journal, descriptor);
>> 706 start_journal_io:
715 for (i = 0; i < bufs; 707 for (i = 0; i < bufs; i++) {
716 struct buffer_ 708 struct buffer_head *bh = wbuf[i];
717 <<
718 /* 709 /*
719 * Compute che 710 * Compute checksum.
720 */ 711 */
721 if (jbd2_has_f 712 if (jbd2_has_feature_checksum(journal)) {
722 crc32_ 713 crc32_sum =
723 jb 714 jbd2_checksum_data(crc32_sum, bh);
724 } 715 }
725 716
726 lock_buffer(bh 717 lock_buffer(bh);
727 clear_buffer_d 718 clear_buffer_dirty(bh);
728 set_buffer_upt 719 set_buffer_uptodate(bh);
729 bh->b_end_io = 720 bh->b_end_io = journal_end_buffer_io_sync;
730 submit_bh(REQ_ !! 721 submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
731 bh); <<
732 } 722 }
733 cond_resched(); 723 cond_resched();
>> 724 stats.run.rs_blocks_logged += bufs;
734 725
735 /* Force a new descrip 726 /* Force a new descriptor to be generated next
736 time round the loop 727 time round the loop. */
737 descriptor = NULL; 728 descriptor = NULL;
738 bufs = 0; 729 bufs = 0;
739 } 730 }
740 } 731 }
741 732
742 err = journal_finish_inode_data_buffer 733 err = journal_finish_inode_data_buffers(journal, commit_transaction);
743 if (err) { 734 if (err) {
744 printk(KERN_WARNING 735 printk(KERN_WARNING
745 "JBD2: Detected IO err 736 "JBD2: Detected IO errors while flushing file data "
746 "on %s\n", journal->j_d 737 "on %s\n", journal->j_devname);
747 if (journal->j_flags & JBD2_AB 738 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
748 jbd2_journal_abort(jou 739 jbd2_journal_abort(journal, err);
749 err = 0; 740 err = 0;
750 } 741 }
751 742
752 /* 743 /*
753 * Get current oldest transaction in t 744 * Get current oldest transaction in the log before we issue flush
754 * to the filesystem device. After the 745 * to the filesystem device. After the flush we can be sure that
755 * blocks of all older transactions ar 746 * blocks of all older transactions are checkpointed to persistent
756 * storage and we will be safe to upda 747 * storage and we will be safe to update journal start in the
757 * superblock with the numbers we get 748 * superblock with the numbers we get here.
758 */ 749 */
759 update_tail = 750 update_tail =
760 jbd2_journal_get_log_tail(jour 751 jbd2_journal_get_log_tail(journal, &first_tid, &first_block);
761 752
762 write_lock(&journal->j_state_lock); 753 write_lock(&journal->j_state_lock);
763 if (update_tail) { 754 if (update_tail) {
764 long freed = first_block - jou 755 long freed = first_block - journal->j_tail;
765 756
766 if (first_block < journal->j_t 757 if (first_block < journal->j_tail)
767 freed += journal->j_la 758 freed += journal->j_last - journal->j_first;
768 /* Update tail only if we free 759 /* Update tail only if we free significant amount of space */
769 if (freed < journal->j_max_tra !! 760 if (freed < journal->j_maxlen / 4)
770 update_tail = 0; 761 update_tail = 0;
771 } 762 }
772 J_ASSERT(commit_transaction->t_state = 763 J_ASSERT(commit_transaction->t_state == T_COMMIT);
773 commit_transaction->t_state = T_COMMIT 764 commit_transaction->t_state = T_COMMIT_DFLUSH;
774 write_unlock(&journal->j_state_lock); 765 write_unlock(&journal->j_state_lock);
775 766
776 /* !! 767 /*
777 * If the journal is not located on th 768 * If the journal is not located on the file system device,
778 * then we must flush the file system 769 * then we must flush the file system device before we issue
779 * the commit record 770 * the commit record
780 */ 771 */
781 if (commit_transaction->t_need_data_fl 772 if (commit_transaction->t_need_data_flush &&
782 (journal->j_fs_dev != journal->j_d 773 (journal->j_fs_dev != journal->j_dev) &&
783 (journal->j_flags & JBD2_BARRIER)) 774 (journal->j_flags & JBD2_BARRIER))
784 blkdev_issue_flush(journal->j_ !! 775 blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL);
785 776
786 /* Done it all: now write the commit r 777 /* Done it all: now write the commit record asynchronously. */
787 if (jbd2_has_feature_async_commit(jour 778 if (jbd2_has_feature_async_commit(journal)) {
788 err = journal_submit_commit_re 779 err = journal_submit_commit_record(journal, commit_transaction,
789 780 &cbh, crc32_sum);
790 if (err) 781 if (err)
791 jbd2_journal_abort(jou !! 782 __jbd2_journal_abort_hard(journal);
792 } 783 }
793 784
794 blk_finish_plug(&plug); 785 blk_finish_plug(&plug);
795 786
796 /* Lo and behold: we have just managed 787 /* Lo and behold: we have just managed to send a transaction to
797 the log. Before we can commit it, 788 the log. Before we can commit it, wait for the IO so far to
798 complete. Control buffers being wr 789 complete. Control buffers being written are on the
799 transaction's t_log_list queue, and 790 transaction's t_log_list queue, and metadata buffers are on
800 the io_bufs list. 791 the io_bufs list.
801 792
802 Wait for the buffers in reverse ord 793 Wait for the buffers in reverse order. That way we are
803 less likely to be woken up until al 794 less likely to be woken up until all IOs have completed, and
804 so we incur less scheduling load. 795 so we incur less scheduling load.
805 */ 796 */
806 797
807 jbd2_debug(3, "JBD2: commit phase 3\n" !! 798 jbd_debug(3, "JBD2: commit phase 3\n");
808 799
809 while (!list_empty(&io_bufs)) { 800 while (!list_empty(&io_bufs)) {
810 struct buffer_head *bh = list_ 801 struct buffer_head *bh = list_entry(io_bufs.prev,
811 802 struct buffer_head,
812 803 b_assoc_buffers);
813 804
814 wait_on_buffer(bh); 805 wait_on_buffer(bh);
815 cond_resched(); 806 cond_resched();
816 807
817 if (unlikely(!buffer_uptodate( 808 if (unlikely(!buffer_uptodate(bh)))
818 err = -EIO; 809 err = -EIO;
819 jbd2_unfile_log_bh(bh); 810 jbd2_unfile_log_bh(bh);
820 stats.run.rs_blocks_logged++; <<
821 811
822 /* 812 /*
823 * The list contains temporary 813 * The list contains temporary buffer heads created by
824 * jbd2_journal_write_metadata 814 * jbd2_journal_write_metadata_buffer().
825 */ 815 */
826 BUFFER_TRACE(bh, "dumping temp 816 BUFFER_TRACE(bh, "dumping temporary bh");
827 __brelse(bh); 817 __brelse(bh);
828 J_ASSERT_BH(bh, atomic_read(&b 818 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
829 free_buffer_head(bh); 819 free_buffer_head(bh);
830 820
831 /* We also have to refile the 821 /* We also have to refile the corresponding shadowed buffer */
832 jh = commit_transaction->t_sha 822 jh = commit_transaction->t_shadow_list->b_tprev;
833 bh = jh2bh(jh); 823 bh = jh2bh(jh);
834 clear_buffer_jwrite(bh); 824 clear_buffer_jwrite(bh);
835 J_ASSERT_BH(bh, buffer_jbddirt 825 J_ASSERT_BH(bh, buffer_jbddirty(bh));
836 J_ASSERT_BH(bh, !buffer_shadow 826 J_ASSERT_BH(bh, !buffer_shadow(bh));
837 827
838 /* The metadata is now release 828 /* The metadata is now released for reuse, but we need
839 to remember it against this 829 to remember it against this transaction so that when
840 we finally commit, we can d 830 we finally commit, we can do any checkpointing
841 required. */ 831 required. */
842 JBUFFER_TRACE(jh, "file as BJ_ 832 JBUFFER_TRACE(jh, "file as BJ_Forget");
843 jbd2_journal_file_buffer(jh, c 833 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
844 JBUFFER_TRACE(jh, "brelse shad 834 JBUFFER_TRACE(jh, "brelse shadowed buffer");
845 __brelse(bh); 835 __brelse(bh);
846 } 836 }
847 837
848 J_ASSERT (commit_transaction->t_shadow 838 J_ASSERT (commit_transaction->t_shadow_list == NULL);
849 839
850 jbd2_debug(3, "JBD2: commit phase 4\n" !! 840 jbd_debug(3, "JBD2: commit phase 4\n");
851 841
852 /* Here we wait for the revoke record 842 /* Here we wait for the revoke record and descriptor record buffers */
853 while (!list_empty(&log_bufs)) { 843 while (!list_empty(&log_bufs)) {
854 struct buffer_head *bh; 844 struct buffer_head *bh;
855 845
856 bh = list_entry(log_bufs.prev, 846 bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
857 wait_on_buffer(bh); 847 wait_on_buffer(bh);
858 cond_resched(); 848 cond_resched();
859 849
860 if (unlikely(!buffer_uptodate( 850 if (unlikely(!buffer_uptodate(bh)))
861 err = -EIO; 851 err = -EIO;
862 852
863 BUFFER_TRACE(bh, "ph5: control 853 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
864 clear_buffer_jwrite(bh); 854 clear_buffer_jwrite(bh);
865 jbd2_unfile_log_bh(bh); 855 jbd2_unfile_log_bh(bh);
866 stats.run.rs_blocks_logged++; <<
867 __brelse(bh); /* One 856 __brelse(bh); /* One for getblk */
868 /* AKPM: bforget here */ 857 /* AKPM: bforget here */
869 } 858 }
870 859
871 if (err) 860 if (err)
872 jbd2_journal_abort(journal, er 861 jbd2_journal_abort(journal, err);
873 862
874 jbd2_debug(3, "JBD2: commit phase 5\n" !! 863 jbd_debug(3, "JBD2: commit phase 5\n");
875 write_lock(&journal->j_state_lock); 864 write_lock(&journal->j_state_lock);
876 J_ASSERT(commit_transaction->t_state = 865 J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
877 commit_transaction->t_state = T_COMMIT 866 commit_transaction->t_state = T_COMMIT_JFLUSH;
878 write_unlock(&journal->j_state_lock); 867 write_unlock(&journal->j_state_lock);
879 868
880 if (!jbd2_has_feature_async_commit(jou 869 if (!jbd2_has_feature_async_commit(journal)) {
881 err = journal_submit_commit_re 870 err = journal_submit_commit_record(journal, commit_transaction,
882 871 &cbh, crc32_sum);
883 if (err) 872 if (err)
884 jbd2_journal_abort(jou !! 873 __jbd2_journal_abort_hard(journal);
885 } 874 }
886 if (cbh) 875 if (cbh)
887 err = journal_wait_on_commit_r 876 err = journal_wait_on_commit_record(journal, cbh);
888 stats.run.rs_blocks_logged++; <<
889 if (jbd2_has_feature_async_commit(jour 877 if (jbd2_has_feature_async_commit(journal) &&
890 journal->j_flags & JBD2_BARRIER) { 878 journal->j_flags & JBD2_BARRIER) {
891 blkdev_issue_flush(journal->j_ !! 879 blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL);
892 } 880 }
893 881
894 if (err) 882 if (err)
895 jbd2_journal_abort(journal, er 883 jbd2_journal_abort(journal, err);
896 884
897 WARN_ON_ONCE( <<
898 atomic_read(&commit_transactio <<
899 <<
900 /* 885 /*
901 * Now disk caches for filesystem devi 886 * Now disk caches for filesystem device are flushed so we are safe to
902 * erase checkpointed transactions fro 887 * erase checkpointed transactions from the log by updating journal
903 * superblock. 888 * superblock.
904 */ 889 */
905 if (update_tail) 890 if (update_tail)
906 jbd2_update_log_tail(journal, 891 jbd2_update_log_tail(journal, first_tid, first_block);
907 892
908 /* End of a transaction! Finally, we 893 /* End of a transaction! Finally, we can do checkpoint
909 processing: any buffers committed a 894 processing: any buffers committed as a result of this
910 transaction can be removed from any 895 transaction can be removed from any checkpoint list it was on
911 before. */ 896 before. */
912 897
913 jbd2_debug(3, "JBD2: commit phase 6\n" !! 898 jbd_debug(3, "JBD2: commit phase 6\n");
914 899
915 J_ASSERT(list_empty(&commit_transactio 900 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
916 J_ASSERT(commit_transaction->t_buffers 901 J_ASSERT(commit_transaction->t_buffers == NULL);
917 J_ASSERT(commit_transaction->t_checkpo 902 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
918 J_ASSERT(commit_transaction->t_shadow_ 903 J_ASSERT(commit_transaction->t_shadow_list == NULL);
919 904
920 restart_loop: 905 restart_loop:
921 /* 906 /*
922 * As there are other places (journal_ 907 * As there are other places (journal_unmap_buffer()) adding buffers
923 * to this list we have to be careful 908 * to this list we have to be careful and hold the j_list_lock.
924 */ 909 */
925 spin_lock(&journal->j_list_lock); 910 spin_lock(&journal->j_list_lock);
926 while (commit_transaction->t_forget) { 911 while (commit_transaction->t_forget) {
927 transaction_t *cp_transaction; 912 transaction_t *cp_transaction;
928 struct buffer_head *bh; 913 struct buffer_head *bh;
929 int try_to_free = 0; 914 int try_to_free = 0;
930 bool drop_ref; <<
931 915
932 jh = commit_transaction->t_for 916 jh = commit_transaction->t_forget;
933 spin_unlock(&journal->j_list_l 917 spin_unlock(&journal->j_list_lock);
934 bh = jh2bh(jh); 918 bh = jh2bh(jh);
935 /* 919 /*
936 * Get a reference so that bh 920 * Get a reference so that bh cannot be freed before we are
937 * done with it. 921 * done with it.
938 */ 922 */
939 get_bh(bh); 923 get_bh(bh);
940 spin_lock(&jh->b_state_lock); !! 924 jbd_lock_bh_state(bh);
941 J_ASSERT_JH(jh, jh->b_transact 925 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
942 926
943 /* 927 /*
944 * If there is undo-protected 928 * If there is undo-protected committed data against
945 * this buffer, then we can re 929 * this buffer, then we can remove it now. If it is a
946 * buffer needing such protect 930 * buffer needing such protection, the old frozen_data
947 * field now points to a commi 931 * field now points to a committed version of the
948 * buffer, so rotate that fiel 932 * buffer, so rotate that field to the new committed
949 * data. 933 * data.
950 * 934 *
951 * Otherwise, we can just thro 935 * Otherwise, we can just throw away the frozen data now.
952 * 936 *
953 * We also know that the froze 937 * We also know that the frozen data has already fired
954 * its triggers if they exist, 938 * its triggers if they exist, so we can clear that too.
955 */ 939 */
956 if (jh->b_committed_data) { 940 if (jh->b_committed_data) {
957 jbd2_free(jh->b_commit 941 jbd2_free(jh->b_committed_data, bh->b_size);
958 jh->b_committed_data = 942 jh->b_committed_data = NULL;
959 if (jh->b_frozen_data) 943 if (jh->b_frozen_data) {
960 jh->b_committe 944 jh->b_committed_data = jh->b_frozen_data;
961 jh->b_frozen_d 945 jh->b_frozen_data = NULL;
962 jh->b_frozen_t 946 jh->b_frozen_triggers = NULL;
963 } 947 }
964 } else if (jh->b_frozen_data) 948 } else if (jh->b_frozen_data) {
965 jbd2_free(jh->b_frozen 949 jbd2_free(jh->b_frozen_data, bh->b_size);
966 jh->b_frozen_data = NU 950 jh->b_frozen_data = NULL;
967 jh->b_frozen_triggers 951 jh->b_frozen_triggers = NULL;
968 } 952 }
969 953
970 spin_lock(&journal->j_list_loc 954 spin_lock(&journal->j_list_lock);
971 cp_transaction = jh->b_cp_tran 955 cp_transaction = jh->b_cp_transaction;
972 if (cp_transaction) { 956 if (cp_transaction) {
973 JBUFFER_TRACE(jh, "rem 957 JBUFFER_TRACE(jh, "remove from old cp transaction");
974 cp_transaction->t_chp_ 958 cp_transaction->t_chp_stats.cs_dropped++;
975 __jbd2_journal_remove_ 959 __jbd2_journal_remove_checkpoint(jh);
976 } 960 }
977 961
978 /* Only re-checkpoint the buff 962 /* Only re-checkpoint the buffer_head if it is marked
979 * dirty. If the buffer was a 963 * dirty. If the buffer was added to the BJ_Forget list
980 * by jbd2_journal_forget, it 964 * by jbd2_journal_forget, it may no longer be dirty and
981 * there's no point in keeping 965 * there's no point in keeping a checkpoint record for
982 * it. */ 966 * it. */
983 967
984 /* 968 /*
985 * A buffer which has been fre !! 969 * A buffer which has been freed while still being journaled by
986 * by a previous transaction, !! 970 * a previous transaction.
987 * the running transaction. If !! 971 */
988 * contains "add to orphan" op !! 972 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 /* 973 /*
1000 * Block device buffe !! 974 * If the running transaction is the one containing
1001 * time, so it is eno !! 975 * "add to orphan" operation (b_next_transaction !=
1002 * buffer_freed bits. !! 976 * NULL), we have to wait for that transaction to
1003 * journalled data) w !! 977 * commit before we can really get rid of the buffer.
1004 * more bits. We also !! 978 * So just clear b_modified to not confuse transaction
1005 * because the data p !! 979 * credit accounting and refile the buffer to
1006 * our hands. Note th !! 980 * BJ_Forget of the running transaction. If the just
1007 * need to make buffe !! 981 * committed transaction contains "add to orphan"
1008 * already detached f !! 982 * operation, we can completely invalidate the buffer
1009 * get reused. !! 983 * now. We are rather through in that since the
>> 984 * buffer may be still accessible when blocksize <
>> 985 * pagesize and it is attached to the last partial
>> 986 * page.
1010 */ 987 */
1011 mapping = READ_ONCE(b !! 988 jh->b_modified = 0;
1012 if (mapping && !sb_is !! 989 if (!jh->b_next_transaction) {
>> 990 clear_buffer_freed(bh);
>> 991 clear_buffer_jbddirty(bh);
1013 clear_buffer_ 992 clear_buffer_mapped(bh);
1014 clear_buffer_ 993 clear_buffer_new(bh);
1015 clear_buffer_ 994 clear_buffer_req(bh);
1016 bh->b_bdev = 995 bh->b_bdev = NULL;
1017 } 996 }
1018 } 997 }
1019 998
1020 if (buffer_jbddirty(bh)) { 999 if (buffer_jbddirty(bh)) {
1021 JBUFFER_TRACE(jh, "ad 1000 JBUFFER_TRACE(jh, "add to new checkpointing trans");
1022 __jbd2_journal_insert 1001 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
1023 if (is_journal_aborte 1002 if (is_journal_aborted(journal))
1024 clear_buffer_ 1003 clear_buffer_jbddirty(bh);
1025 } else { 1004 } else {
1026 J_ASSERT_BH(bh, !buff 1005 J_ASSERT_BH(bh, !buffer_dirty(bh));
1027 /* 1006 /*
1028 * The buffer on BJ_F 1007 * The buffer on BJ_Forget list and not jbddirty means
1029 * it has been freed 1008 * it has been freed by this transaction and hence it
1030 * could not have bee 1009 * could not have been reallocated until this
1031 * transaction has co 1010 * transaction has committed. *BUT* it could be
1032 * reallocated once w 1011 * reallocated once we have written all the data to
1033 * disk and before we 1012 * disk and before we process the buffer on BJ_Forget
1034 * list. 1013 * list.
1035 */ 1014 */
1036 if (!jh->b_next_trans 1015 if (!jh->b_next_transaction)
1037 try_to_free = 1016 try_to_free = 1;
1038 } 1017 }
1039 JBUFFER_TRACE(jh, "refile or 1018 JBUFFER_TRACE(jh, "refile or unfile buffer");
1040 drop_ref = __jbd2_journal_ref !! 1019 __jbd2_journal_refile_buffer(jh);
1041 spin_unlock(&jh->b_state_lock !! 1020 jbd_unlock_bh_state(bh);
1042 if (drop_ref) <<
1043 jbd2_journal_put_jour <<
1044 if (try_to_free) 1021 if (try_to_free)
1045 release_buffer_page(b 1022 release_buffer_page(bh); /* Drops bh reference */
1046 else 1023 else
1047 __brelse(bh); 1024 __brelse(bh);
1048 cond_resched_lock(&journal->j 1025 cond_resched_lock(&journal->j_list_lock);
1049 } 1026 }
1050 spin_unlock(&journal->j_list_lock); 1027 spin_unlock(&journal->j_list_lock);
1051 /* 1028 /*
1052 * This is a bit sleazy. We use j_li 1029 * This is a bit sleazy. We use j_list_lock to protect transition
1053 * of a transaction into T_FINISHED s 1030 * of a transaction into T_FINISHED state and calling
1054 * __jbd2_journal_drop_transaction(). 1031 * __jbd2_journal_drop_transaction(). Otherwise we could race with
1055 * other checkpointing code processin 1032 * other checkpointing code processing the transaction...
1056 */ 1033 */
1057 write_lock(&journal->j_state_lock); 1034 write_lock(&journal->j_state_lock);
1058 spin_lock(&journal->j_list_lock); 1035 spin_lock(&journal->j_list_lock);
1059 /* 1036 /*
1060 * Now recheck if some buffers did no 1037 * Now recheck if some buffers did not get attached to the transaction
1061 * while the lock was dropped... 1038 * while the lock was dropped...
1062 */ 1039 */
1063 if (commit_transaction->t_forget) { 1040 if (commit_transaction->t_forget) {
1064 spin_unlock(&journal->j_list_ 1041 spin_unlock(&journal->j_list_lock);
1065 write_unlock(&journal->j_stat 1042 write_unlock(&journal->j_state_lock);
1066 goto restart_loop; 1043 goto restart_loop;
1067 } 1044 }
1068 1045
1069 /* Add the transaction to the checkpo 1046 /* Add the transaction to the checkpoint list
1070 * __journal_remove_checkpoint() can 1047 * __journal_remove_checkpoint() can not destroy transaction
1071 * under us because it is not marked 1048 * under us because it is not marked as T_FINISHED yet */
1072 if (journal->j_checkpoint_transaction 1049 if (journal->j_checkpoint_transactions == NULL) {
1073 journal->j_checkpoint_transac 1050 journal->j_checkpoint_transactions = commit_transaction;
1074 commit_transaction->t_cpnext 1051 commit_transaction->t_cpnext = commit_transaction;
1075 commit_transaction->t_cpprev 1052 commit_transaction->t_cpprev = commit_transaction;
1076 } else { 1053 } else {
1077 commit_transaction->t_cpnext 1054 commit_transaction->t_cpnext =
1078 journal->j_checkpoint 1055 journal->j_checkpoint_transactions;
1079 commit_transaction->t_cpprev 1056 commit_transaction->t_cpprev =
1080 commit_transaction->t 1057 commit_transaction->t_cpnext->t_cpprev;
1081 commit_transaction->t_cpnext- 1058 commit_transaction->t_cpnext->t_cpprev =
1082 commit_transaction; 1059 commit_transaction;
1083 commit_transaction->t_cpprev- 1060 commit_transaction->t_cpprev->t_cpnext =
1084 commit_transa 1061 commit_transaction;
1085 } 1062 }
1086 spin_unlock(&journal->j_list_lock); 1063 spin_unlock(&journal->j_list_lock);
1087 1064
1088 /* Done with this transaction! */ 1065 /* Done with this transaction! */
1089 1066
1090 jbd2_debug(3, "JBD2: commit phase 7\n !! 1067 jbd_debug(3, "JBD2: commit phase 7\n");
1091 1068
1092 J_ASSERT(commit_transaction->t_state 1069 J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
1093 1070
1094 commit_transaction->t_start = jiffies 1071 commit_transaction->t_start = jiffies;
1095 stats.run.rs_logging = jbd2_time_diff 1072 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
1096 1073 commit_transaction->t_start);
1097 1074
1098 /* 1075 /*
1099 * File the transaction statistics 1076 * File the transaction statistics
1100 */ 1077 */
1101 stats.ts_tid = commit_transaction->t_ 1078 stats.ts_tid = commit_transaction->t_tid;
1102 stats.run.rs_handle_count = 1079 stats.run.rs_handle_count =
1103 atomic_read(&commit_transacti 1080 atomic_read(&commit_transaction->t_handle_count);
1104 trace_jbd2_run_stats(journal->j_fs_de 1081 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
1105 commit_transacti 1082 commit_transaction->t_tid, &stats.run);
1106 stats.ts_requested = (commit_transact 1083 stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0;
1107 1084
1108 commit_transaction->t_state = T_COMMI 1085 commit_transaction->t_state = T_COMMIT_CALLBACK;
1109 J_ASSERT(commit_transaction == journa 1086 J_ASSERT(commit_transaction == journal->j_committing_transaction);
1110 WRITE_ONCE(journal->j_commit_sequence !! 1087 journal->j_commit_sequence = commit_transaction->t_tid;
1111 journal->j_committing_transaction = N 1088 journal->j_committing_transaction = NULL;
1112 commit_time = ktime_to_ns(ktime_sub(k 1089 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1113 1090
1114 /* 1091 /*
1115 * weight the commit time higher than 1092 * weight the commit time higher than the average time so we don't
1116 * react too strongly to vast changes 1093 * react too strongly to vast changes in the commit time
1117 */ 1094 */
1118 if (likely(journal->j_average_commit_ 1095 if (likely(journal->j_average_commit_time))
1119 journal->j_average_commit_tim 1096 journal->j_average_commit_time = (commit_time +
1120 journal->j_av 1097 journal->j_average_commit_time*3) / 4;
1121 else 1098 else
1122 journal->j_average_commit_tim 1099 journal->j_average_commit_time = commit_time;
1123 1100
1124 write_unlock(&journal->j_state_lock); 1101 write_unlock(&journal->j_state_lock);
1125 1102
1126 if (journal->j_commit_callback) 1103 if (journal->j_commit_callback)
1127 journal->j_commit_callback(jo 1104 journal->j_commit_callback(journal, commit_transaction);
1128 if (journal->j_fc_cleanup_callback) <<
1129 journal->j_fc_cleanup_callbac <<
1130 1105
1131 trace_jbd2_end_commit(journal, commit 1106 trace_jbd2_end_commit(journal, commit_transaction);
1132 jbd2_debug(1, "JBD2: commit %d comple !! 1107 jbd_debug(1, "JBD2: commit %d complete, head %d\n",
1133 journal->j_commit_sequence, 1108 journal->j_commit_sequence, journal->j_tail_sequence);
1134 1109
1135 write_lock(&journal->j_state_lock); 1110 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); 1111 spin_lock(&journal->j_list_lock);
1139 commit_transaction->t_state = T_FINIS 1112 commit_transaction->t_state = T_FINISHED;
1140 /* Check if the transaction can be dr 1113 /* Check if the transaction can be dropped now that we are finished */
1141 if (commit_transaction->t_checkpoint_ !! 1114 if (commit_transaction->t_checkpoint_list == NULL &&
>> 1115 commit_transaction->t_checkpoint_io_list == NULL) {
1142 __jbd2_journal_drop_transacti 1116 __jbd2_journal_drop_transaction(journal, commit_transaction);
1143 jbd2_journal_free_transaction 1117 jbd2_journal_free_transaction(commit_transaction);
1144 } 1118 }
1145 spin_unlock(&journal->j_list_lock); 1119 spin_unlock(&journal->j_list_lock);
1146 write_unlock(&journal->j_state_lock); 1120 write_unlock(&journal->j_state_lock);
1147 wake_up(&journal->j_wait_done_commit) 1121 wake_up(&journal->j_wait_done_commit);
1148 wake_up(&journal->j_fc_wait); <<
1149 1122
1150 /* 1123 /*
1151 * Calculate overall stats 1124 * Calculate overall stats
1152 */ 1125 */
1153 spin_lock(&journal->j_history_lock); 1126 spin_lock(&journal->j_history_lock);
1154 journal->j_stats.ts_tid++; 1127 journal->j_stats.ts_tid++;
1155 journal->j_stats.ts_requested += stat 1128 journal->j_stats.ts_requested += stats.ts_requested;
1156 journal->j_stats.run.rs_wait += stats 1129 journal->j_stats.run.rs_wait += stats.run.rs_wait;
1157 journal->j_stats.run.rs_request_delay 1130 journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
1158 journal->j_stats.run.rs_running += st 1131 journal->j_stats.run.rs_running += stats.run.rs_running;
1159 journal->j_stats.run.rs_locked += sta 1132 journal->j_stats.run.rs_locked += stats.run.rs_locked;
1160 journal->j_stats.run.rs_flushing += s 1133 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
1161 journal->j_stats.run.rs_logging += st 1134 journal->j_stats.run.rs_logging += stats.run.rs_logging;
1162 journal->j_stats.run.rs_handle_count 1135 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
1163 journal->j_stats.run.rs_blocks += sta 1136 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
1164 journal->j_stats.run.rs_blocks_logged 1137 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
1165 spin_unlock(&journal->j_history_lock) 1138 spin_unlock(&journal->j_history_lock);
1166 } 1139 }
1167 1140
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.