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> <<
27 #include <trace/events/jbd2.h> 29 #include <trace/events/jbd2.h>
28 30
29 /* 31 /*
30 * IO end handler for temporary buffer_heads h !! 32 * Default IO end handler for temporary BJ_IO buffer_heads.
31 */ 33 */
32 static void journal_end_buffer_io_sync(struct 34 static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
33 { 35 {
34 struct buffer_head *orig_bh = bh->b_pr <<
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) { <<
42 clear_bit_unlock(BH_Shadow, &o <<
43 smp_mb__after_atomic(); <<
44 wake_up_bit(&orig_bh->b_state, <<
45 } <<
46 unlock_buffer(bh); 41 unlock_buffer(bh);
47 } 42 }
48 43
49 /* 44 /*
50 * When an ext4 file is truncated, it is possi 45 * When an ext4 file is truncated, it is possible that some pages are not
51 * successfully freed, because they are attach 46 * successfully freed, because they are attached to a committing transaction.
52 * After the transaction commits, these pages 47 * After the transaction commits, these pages are left on the LRU, with no
53 * ->mapping, and with attached buffers. Thes 48 * ->mapping, and with attached buffers. These pages are trivially reclaimable
54 * by the VM, but their apparent absence upset 49 * by the VM, but their apparent absence upsets the VM accounting, and it makes
55 * the numbers in /proc/meminfo look odd. 50 * the numbers in /proc/meminfo look odd.
56 * 51 *
57 * So here, we have a buffer which has just co 52 * 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 53 * see if we can strip all buffers from the backing page.
59 * 54 *
60 * Called under lock_journal(), and possibly u 55 * Called under lock_journal(), and possibly under journal_datalist_lock. The
61 * caller provided us with a ref against the b 56 * caller provided us with a ref against the buffer, and we drop that here.
62 */ 57 */
63 static void release_buffer_page(struct buffer_ 58 static void release_buffer_page(struct buffer_head *bh)
64 { 59 {
65 struct folio *folio; !! 60 struct page *page;
66 61
67 if (buffer_dirty(bh)) 62 if (buffer_dirty(bh))
68 goto nope; 63 goto nope;
69 if (atomic_read(&bh->b_count) != 1) 64 if (atomic_read(&bh->b_count) != 1)
70 goto nope; 65 goto nope;
71 folio = bh->b_folio; !! 66 page = bh->b_page;
72 if (folio->mapping) !! 67 if (!page)
>> 68 goto nope;
>> 69 if (page->mapping)
73 goto nope; 70 goto nope;
74 71
75 /* OK, it's a truncated page */ 72 /* OK, it's a truncated page */
76 if (!folio_trylock(folio)) !! 73 if (!trylock_page(page))
77 goto nope; 74 goto nope;
78 75
79 folio_get(folio); !! 76 page_cache_get(page);
80 __brelse(bh); 77 __brelse(bh);
81 try_to_free_buffers(folio); !! 78 try_to_free_buffers(page);
82 folio_unlock(folio); !! 79 unlock_page(page);
83 folio_put(folio); !! 80 page_cache_release(page);
84 return; 81 return;
85 82
86 nope: 83 nope:
87 __brelse(bh); 84 __brelse(bh);
88 } 85 }
89 86
90 static void jbd2_commit_block_csum_set(journal <<
91 { <<
92 struct commit_header *h; <<
93 __u32 csum; <<
94 <<
95 if (!jbd2_journal_has_csum_v2or3(j)) <<
96 return; <<
97 <<
98 h = (struct commit_header *)(bh->b_dat <<
99 h->h_chksum_type = 0; <<
100 h->h_chksum_size = 0; <<
101 h->h_chksum[0] = 0; <<
102 csum = jbd2_chksum(j, j->j_csum_seed, <<
103 h->h_chksum[0] = cpu_to_be32(csum); <<
104 } <<
105 <<
106 /* 87 /*
107 * Done it all: now submit the commit record. 88 * Done it all: now submit the commit record. We should have
108 * cleaned up our previous buffers by now, so 89 * 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 90 * mode we can now just skip the rest of the journal write
110 * entirely. 91 * entirely.
111 * 92 *
112 * Returns 1 if the journal needs to be aborte 93 * Returns 1 if the journal needs to be aborted or 0 on success
113 */ 94 */
114 static int journal_submit_commit_record(journa 95 static int journal_submit_commit_record(journal_t *journal,
115 transa 96 transaction_t *commit_transaction,
116 struct 97 struct buffer_head **cbh,
117 __u32 98 __u32 crc32_sum)
118 { 99 {
>> 100 struct journal_head *descriptor;
119 struct commit_header *tmp; 101 struct commit_header *tmp;
120 struct buffer_head *bh; 102 struct buffer_head *bh;
121 struct timespec64 now; !! 103 int ret;
122 blk_opf_t write_flags = REQ_OP_WRITE | !! 104 int barrier_done = 0;
123 !! 105 struct timespec now = current_kernel_time();
124 *cbh = NULL; <<
125 106
126 if (is_journal_aborted(journal)) 107 if (is_journal_aborted(journal))
127 return 0; 108 return 0;
128 109
129 bh = jbd2_journal_get_descriptor_buffe !! 110 descriptor = jbd2_journal_get_descriptor_buffer(journal);
130 !! 111 if (!descriptor)
131 if (!bh) <<
132 return 1; 112 return 1;
133 113
>> 114 bh = jh2bh(descriptor);
>> 115
134 tmp = (struct commit_header *)bh->b_da 116 tmp = (struct commit_header *)bh->b_data;
135 ktime_get_coarse_real_ts64(&now); !! 117 tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
>> 118 tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK);
>> 119 tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
136 tmp->h_commit_sec = cpu_to_be64(now.tv 120 tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
137 tmp->h_commit_nsec = cpu_to_be32(now.t 121 tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
138 122
139 if (jbd2_has_feature_checksum(journal) !! 123 if (JBD2_HAS_COMPAT_FEATURE(journal,
>> 124 JBD2_FEATURE_COMPAT_CHECKSUM)) {
140 tmp->h_chksum_type = JBD2 125 tmp->h_chksum_type = JBD2_CRC32_CHKSUM;
141 tmp->h_chksum_size = JBD2 126 tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE;
142 tmp->h_chksum[0] = cpu_ 127 tmp->h_chksum[0] = cpu_to_be32(crc32_sum);
143 } 128 }
144 jbd2_commit_block_csum_set(journal, bh <<
145 129
146 BUFFER_TRACE(bh, "submit commit block" !! 130 JBUFFER_TRACE(descriptor, "submit commit block");
147 lock_buffer(bh); 131 lock_buffer(bh);
148 clear_buffer_dirty(bh); 132 clear_buffer_dirty(bh);
149 set_buffer_uptodate(bh); 133 set_buffer_uptodate(bh);
150 bh->b_end_io = journal_end_buffer_io_s 134 bh->b_end_io = journal_end_buffer_io_sync;
151 135
152 if (journal->j_flags & JBD2_BARRIER && 136 if (journal->j_flags & JBD2_BARRIER &&
153 !jbd2_has_feature_async_commit(jou !! 137 !JBD2_HAS_INCOMPAT_FEATURE(journal,
154 write_flags |= REQ_PREFLUSH | !! 138 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
>> 139 set_buffer_ordered(bh);
>> 140 barrier_done = 1;
>> 141 }
>> 142 ret = submit_bh(WRITE_SYNC_PLUG, bh);
>> 143 if (barrier_done)
>> 144 clear_buffer_ordered(bh);
>> 145
>> 146 /* is it possible for another commit to fail at roughly
>> 147 * the same time as this one? If so, we don't want to
>> 148 * trust the barrier flag in the super, but instead want
>> 149 * to remember if we sent a barrier request
>> 150 */
>> 151 if (ret == -EOPNOTSUPP && barrier_done) {
>> 152 printk(KERN_WARNING
>> 153 "JBD: barrier-based sync failed on %s - "
>> 154 "disabling barriers\n", journal->j_devname);
>> 155 spin_lock(&journal->j_state_lock);
>> 156 journal->j_flags &= ~JBD2_BARRIER;
>> 157 spin_unlock(&journal->j_state_lock);
155 158
156 submit_bh(write_flags, bh); !! 159 /* And try again, without the barrier */
>> 160 lock_buffer(bh);
>> 161 set_buffer_uptodate(bh);
>> 162 clear_buffer_dirty(bh);
>> 163 ret = submit_bh(WRITE_SYNC_PLUG, bh);
>> 164 }
157 *cbh = bh; 165 *cbh = bh;
158 return 0; !! 166 return ret;
159 } 167 }
160 168
161 /* 169 /*
162 * This function along with journal_submit_com 170 * This function along with journal_submit_commit_record
163 * allows to write the commit record asynchron 171 * allows to write the commit record asynchronously.
164 */ 172 */
165 static int journal_wait_on_commit_record(journ 173 static int journal_wait_on_commit_record(journal_t *journal,
166 struc 174 struct buffer_head *bh)
167 { 175 {
168 int ret = 0; 176 int ret = 0;
169 177
>> 178 retry:
170 clear_buffer_dirty(bh); 179 clear_buffer_dirty(bh);
171 wait_on_buffer(bh); 180 wait_on_buffer(bh);
>> 181 if (buffer_eopnotsupp(bh) && (journal->j_flags & JBD2_BARRIER)) {
>> 182 printk(KERN_WARNING
>> 183 "JBD2: wait_on_commit_record: sync failed on %s - "
>> 184 "disabling barriers\n", journal->j_devname);
>> 185 spin_lock(&journal->j_state_lock);
>> 186 journal->j_flags &= ~JBD2_BARRIER;
>> 187 spin_unlock(&journal->j_state_lock);
>> 188
>> 189 lock_buffer(bh);
>> 190 clear_buffer_dirty(bh);
>> 191 set_buffer_uptodate(bh);
>> 192 bh->b_end_io = journal_end_buffer_io_sync;
>> 193
>> 194 ret = submit_bh(WRITE_SYNC_PLUG, bh);
>> 195 if (ret) {
>> 196 unlock_buffer(bh);
>> 197 return ret;
>> 198 }
>> 199 goto retry;
>> 200 }
172 201
173 if (unlikely(!buffer_uptodate(bh))) 202 if (unlikely(!buffer_uptodate(bh)))
174 ret = -EIO; 203 ret = -EIO;
175 put_bh(bh); /* One for getb 204 put_bh(bh); /* One for getblk() */
>> 205 jbd2_journal_put_journal_head(bh2jh(bh));
176 206
177 return ret; 207 return ret;
178 } 208 }
179 209
180 /* Send all the data buffers related to an ino !! 210 /*
181 int jbd2_submit_inode_data(journal_t *journal, !! 211 * write the filemap data using writepage() address_space_operations.
>> 212 * We don't do block allocation here even for delalloc. We don't
>> 213 * use writepages() because with dealyed allocation we may be doing
>> 214 * block allocation in writepages().
>> 215 */
>> 216 static int journal_submit_inode_data_buffers(struct address_space *mapping)
182 { 217 {
183 if (!jinode || !(jinode->i_flags & JI_ !! 218 int ret;
184 return 0; !! 219 struct writeback_control wbc = {
185 !! 220 .sync_mode = WB_SYNC_ALL,
186 trace_jbd2_submit_inode_data(jinode->i !! 221 .nr_to_write = mapping->nrpages * 2,
187 return journal->j_submit_inode_data_bu !! 222 .range_start = 0,
>> 223 .range_end = i_size_read(mapping->host),
>> 224 };
188 225
>> 226 ret = generic_writepages(mapping, &wbc);
>> 227 return ret;
189 } 228 }
190 EXPORT_SYMBOL(jbd2_submit_inode_data); <<
191 <<
192 int jbd2_wait_inode_data(journal_t *journal, s <<
193 { <<
194 if (!jinode || !(jinode->i_flags & JI_ <<
195 !jinode->i_vfs_inode || !jinod <<
196 return 0; <<
197 return filemap_fdatawait_range_keep_er <<
198 jinode->i_vfs_inode->i_mapping <<
199 jinode->i_dirty_end); <<
200 } <<
201 EXPORT_SYMBOL(jbd2_wait_inode_data); <<
202 229
203 /* 230 /*
204 * Submit all the data buffers of inode associ 231 * Submit all the data buffers of inode associated with the transaction to
205 * disk. 232 * disk.
206 * 233 *
207 * We are in a committing transaction. Therefo 234 * We are in a committing transaction. Therefore no new inode can be added to
208 * our inode list. We use JI_COMMIT_RUNNING fl 235 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
209 * operate on from being released while we wri 236 * operate on from being released while we write out pages.
210 */ 237 */
211 static int journal_submit_data_buffers(journal 238 static int journal_submit_data_buffers(journal_t *journal,
212 transaction_t *commit_transact 239 transaction_t *commit_transaction)
213 { 240 {
214 struct jbd2_inode *jinode; 241 struct jbd2_inode *jinode;
215 int err, ret = 0; 242 int err, ret = 0;
>> 243 struct address_space *mapping;
216 244
217 spin_lock(&journal->j_list_lock); 245 spin_lock(&journal->j_list_lock);
218 list_for_each_entry(jinode, &commit_tr 246 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
219 if (!(jinode->i_flags & JI_WRI !! 247 mapping = jinode->i_vfs_inode->i_mapping;
220 continue; <<
221 jinode->i_flags |= JI_COMMIT_R 248 jinode->i_flags |= JI_COMMIT_RUNNING;
222 spin_unlock(&journal->j_list_l 249 spin_unlock(&journal->j_list_lock);
223 /* submit the inode data buffe !! 250 /*
>> 251 * submit the inode data buffers. We use writepage
>> 252 * instead of writepages. Because writepages can do
>> 253 * block allocation with delalloc. We need to write
>> 254 * only allocated blocks here.
>> 255 */
224 trace_jbd2_submit_inode_data(j 256 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
225 if (journal->j_submit_inode_da !! 257 err = journal_submit_inode_data_buffers(mapping);
226 err = journal->j_submi !! 258 if (!ret)
227 if (!ret) !! 259 ret = err;
228 ret = err; <<
229 } <<
230 spin_lock(&journal->j_list_loc 260 spin_lock(&journal->j_list_lock);
231 J_ASSERT(jinode->i_transaction 261 J_ASSERT(jinode->i_transaction == commit_transaction);
>> 262 commit_transaction->t_flushed_data_blocks = 1;
232 jinode->i_flags &= ~JI_COMMIT_ 263 jinode->i_flags &= ~JI_COMMIT_RUNNING;
233 smp_mb(); <<
234 wake_up_bit(&jinode->i_flags, 264 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
235 } 265 }
236 spin_unlock(&journal->j_list_lock); 266 spin_unlock(&journal->j_list_lock);
237 return ret; 267 return ret;
238 } 268 }
239 269
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 /* 270 /*
250 * Wait for data submitted for writeout, refil 271 * Wait for data submitted for writeout, refile inodes to proper
251 * transaction if needed. 272 * transaction if needed.
252 * 273 *
253 */ 274 */
254 static int journal_finish_inode_data_buffers(j 275 static int journal_finish_inode_data_buffers(journal_t *journal,
255 transaction_t *commit_transact 276 transaction_t *commit_transaction)
256 { 277 {
257 struct jbd2_inode *jinode, *next_i; 278 struct jbd2_inode *jinode, *next_i;
258 int err, ret = 0; 279 int err, ret = 0;
259 280
260 /* For locking, see the comment in jou 281 /* For locking, see the comment in journal_submit_data_buffers() */
261 spin_lock(&journal->j_list_lock); 282 spin_lock(&journal->j_list_lock);
262 list_for_each_entry(jinode, &commit_tr 283 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
263 if (!(jinode->i_flags & JI_WAI <<
264 continue; <<
265 jinode->i_flags |= JI_COMMIT_R 284 jinode->i_flags |= JI_COMMIT_RUNNING;
266 spin_unlock(&journal->j_list_l 285 spin_unlock(&journal->j_list_lock);
267 /* wait for the inode data buf !! 286 err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping);
268 if (journal->j_finish_inode_da !! 287 if (err) {
269 err = journal->j_finis !! 288 /*
>> 289 * Because AS_EIO is cleared by
>> 290 * wait_on_page_writeback_range(), set it again so
>> 291 * that user process can get -EIO from fsync().
>> 292 */
>> 293 set_bit(AS_EIO,
>> 294 &jinode->i_vfs_inode->i_mapping->flags);
>> 295
270 if (!ret) 296 if (!ret)
271 ret = err; 297 ret = err;
272 } 298 }
273 cond_resched(); <<
274 spin_lock(&journal->j_list_loc 299 spin_lock(&journal->j_list_lock);
275 jinode->i_flags &= ~JI_COMMIT_ 300 jinode->i_flags &= ~JI_COMMIT_RUNNING;
276 smp_mb(); <<
277 wake_up_bit(&jinode->i_flags, 301 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
278 } 302 }
279 303
280 /* Now refile inode to proper lists */ 304 /* Now refile inode to proper lists */
281 list_for_each_entry_safe(jinode, next_ 305 list_for_each_entry_safe(jinode, next_i,
282 &commit_trans 306 &commit_transaction->t_inode_list, i_list) {
283 list_del(&jinode->i_list); 307 list_del(&jinode->i_list);
284 if (jinode->i_next_transaction 308 if (jinode->i_next_transaction) {
285 jinode->i_transaction 309 jinode->i_transaction = jinode->i_next_transaction;
286 jinode->i_next_transac 310 jinode->i_next_transaction = NULL;
287 list_add(&jinode->i_li 311 list_add(&jinode->i_list,
288 &jinode->i_tra 312 &jinode->i_transaction->t_inode_list);
289 } else { 313 } else {
290 jinode->i_transaction 314 jinode->i_transaction = NULL;
291 jinode->i_dirty_start <<
292 jinode->i_dirty_end = <<
293 } 315 }
294 } 316 }
295 spin_unlock(&journal->j_list_lock); 317 spin_unlock(&journal->j_list_lock);
296 318
297 return ret; 319 return ret;
298 } 320 }
299 321
300 static __u32 jbd2_checksum_data(__u32 crc32_su 322 static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
301 { 323 {
>> 324 struct page *page = bh->b_page;
302 char *addr; 325 char *addr;
303 __u32 checksum; 326 __u32 checksum;
304 327
305 addr = kmap_local_folio(bh->b_folio, b !! 328 addr = kmap_atomic(page, KM_USER0);
306 checksum = crc32_be(crc32_sum, addr, b !! 329 checksum = crc32_be(crc32_sum,
307 kunmap_local(addr); !! 330 (void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
>> 331 kunmap_atomic(addr, KM_USER0);
308 332
309 return checksum; 333 return checksum;
310 } 334 }
311 335
312 static void write_tag_block(journal_t *j, jour !! 336 static void write_tag_block(int tag_bytes, journal_block_tag_t *tag,
313 unsigned lo 337 unsigned long long block)
314 { 338 {
315 tag->t_blocknr = cpu_to_be32(block & ( 339 tag->t_blocknr = cpu_to_be32(block & (u32)~0);
316 if (jbd2_has_feature_64bit(j)) !! 340 if (tag_bytes > JBD2_TAG_SIZE32)
317 tag->t_blocknr_high = cpu_to_b 341 tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
318 } 342 }
319 343
320 static void jbd2_block_tag_csum_set(journal_t <<
321 struct buf <<
322 { <<
323 journal_block_tag3_t *tag3 = (journal_ <<
324 __u8 *addr; <<
325 __u32 csum32; <<
326 __be32 seq; <<
327 <<
328 if (!jbd2_journal_has_csum_v2or3(j)) <<
329 return; <<
330 <<
331 seq = cpu_to_be32(sequence); <<
332 addr = kmap_local_folio(bh->b_folio, b <<
333 csum32 = jbd2_chksum(j, j->j_csum_seed <<
334 csum32 = jbd2_chksum(j, csum32, addr, <<
335 kunmap_local(addr); <<
336 <<
337 if (jbd2_has_feature_csum3(j)) <<
338 tag3->t_checksum = cpu_to_be32 <<
339 else <<
340 tag->t_checksum = cpu_to_be16( <<
341 } <<
342 /* 344 /*
343 * jbd2_journal_commit_transaction 345 * jbd2_journal_commit_transaction
344 * 346 *
345 * The primary function for committing a trans 347 * The primary function for committing a transaction to the log. This
346 * function is called by the journal thread to 348 * function is called by the journal thread to begin a complete commit.
347 */ 349 */
348 void jbd2_journal_commit_transaction(journal_t 350 void jbd2_journal_commit_transaction(journal_t *journal)
349 { 351 {
350 struct transaction_stats_s stats; 352 struct transaction_stats_s stats;
351 transaction_t *commit_transaction; 353 transaction_t *commit_transaction;
352 struct journal_head *jh; !! 354 struct journal_head *jh, *new_jh, *descriptor;
353 struct buffer_head *descriptor; <<
354 struct buffer_head **wbuf = journal->j 355 struct buffer_head **wbuf = journal->j_wbuf;
355 int bufs; 356 int bufs;
356 int escape; !! 357 int flags;
357 int err; 358 int err;
358 unsigned long long blocknr; 359 unsigned long long blocknr;
359 ktime_t start_time; 360 ktime_t start_time;
360 u64 commit_time; 361 u64 commit_time;
361 char *tagp = NULL; 362 char *tagp = NULL;
>> 363 journal_header_t *header;
362 journal_block_tag_t *tag = NULL; 364 journal_block_tag_t *tag = NULL;
363 int space_left = 0; 365 int space_left = 0;
364 int first_tag = 0; 366 int first_tag = 0;
365 int tag_flag; 367 int tag_flag;
366 int i; !! 368 int i, to_free = 0;
367 int tag_bytes = journal_tag_bytes(jour 369 int tag_bytes = journal_tag_bytes(journal);
368 struct buffer_head *cbh = NULL; /* For 370 struct buffer_head *cbh = NULL; /* For transactional checksums */
369 __u32 crc32_sum = ~0; 371 __u32 crc32_sum = ~0;
370 struct blk_plug plug; !! 372 int write_op = WRITE;
371 /* Tail of the journal */ <<
372 unsigned long first_block; <<
373 tid_t first_tid; <<
374 int update_tail; <<
375 int csum_size = 0; <<
376 LIST_HEAD(io_bufs); <<
377 LIST_HEAD(log_bufs); <<
378 <<
379 if (jbd2_journal_has_csum_v2or3(journa <<
380 csum_size = sizeof(struct jbd2 <<
381 373
382 /* 374 /*
383 * First job: lock down the current tr 375 * First job: lock down the current transaction and wait for
384 * all outstanding updates to complete 376 * all outstanding updates to complete.
385 */ 377 */
386 378
>> 379 #ifdef COMMIT_STATS
>> 380 spin_lock(&journal->j_list_lock);
>> 381 summarise_journal_usage(journal);
>> 382 spin_unlock(&journal->j_list_lock);
>> 383 #endif
>> 384
387 /* Do we need to erase the effects of 385 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
388 if (journal->j_flags & JBD2_FLUSHED) { 386 if (journal->j_flags & JBD2_FLUSHED) {
389 jbd2_debug(3, "super block upd !! 387 jbd_debug(3, "super block updated\n");
390 mutex_lock_io(&journal->j_chec !! 388 jbd2_journal_update_superblock(journal, 1);
391 /* <<
392 * We hold j_checkpoint_mutex <<
393 * We don't need any special d <<
394 * since journal is empty and <<
395 * flushed only with transacti <<
396 */ <<
397 jbd2_journal_update_sb_log_tai <<
398 <<
399 <<
400 mutex_unlock(&journal->j_check <<
401 } else { 389 } else {
402 jbd2_debug(3, "superblock not !! 390 jbd_debug(3, "superblock not updated\n");
403 } 391 }
404 392
405 J_ASSERT(journal->j_running_transactio 393 J_ASSERT(journal->j_running_transaction != NULL);
406 J_ASSERT(journal->j_committing_transac 394 J_ASSERT(journal->j_committing_transaction == NULL);
407 395
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 396 commit_transaction = journal->j_running_transaction;
>> 397 J_ASSERT(commit_transaction->t_state == T_RUNNING);
432 398
433 trace_jbd2_start_commit(journal, commi 399 trace_jbd2_start_commit(journal, commit_transaction);
434 jbd2_debug(1, "JBD2: starting commit o !! 400 jbd_debug(1, "JBD: starting commit of transaction %d\n",
435 commit_transaction->t_ 401 commit_transaction->t_tid);
436 402
437 write_lock(&journal->j_state_lock); !! 403 spin_lock(&journal->j_state_lock);
438 journal->j_fc_off = 0; <<
439 J_ASSERT(commit_transaction->t_state = <<
440 commit_transaction->t_state = T_LOCKED 404 commit_transaction->t_state = T_LOCKED;
441 405
>> 406 /*
>> 407 * Use plugged writes here, since we want to submit several before
>> 408 * we unplug the device. We don't do explicit unplugging in here,
>> 409 * instead we rely on sync_buffer() doing the unplug for us.
>> 410 */
>> 411 if (commit_transaction->t_synchronous_commit)
>> 412 write_op = WRITE_SYNC_PLUG;
442 trace_jbd2_commit_locking(journal, com 413 trace_jbd2_commit_locking(journal, commit_transaction);
443 stats.run.rs_wait = commit_transaction 414 stats.run.rs_wait = commit_transaction->t_max_wait;
444 stats.run.rs_request_delay = 0; <<
445 stats.run.rs_locked = jiffies; 415 stats.run.rs_locked = jiffies;
446 if (commit_transaction->t_requested) <<
447 stats.run.rs_request_delay = <<
448 jbd2_time_diff(commit_ <<
449 stats.r <<
450 stats.run.rs_running = jbd2_time_diff( 416 stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
451 417 stats.run.rs_locked);
452 418
453 // waits for any t_updates to finish !! 419 spin_lock(&commit_transaction->t_handle_lock);
454 jbd2_journal_wait_updates(journal); !! 420 while (commit_transaction->t_updates) {
>> 421 DEFINE_WAIT(wait);
455 422
456 commit_transaction->t_state = T_SWITCH !! 423 prepare_to_wait(&journal->j_wait_updates, &wait,
>> 424 TASK_UNINTERRUPTIBLE);
>> 425 if (commit_transaction->t_updates) {
>> 426 spin_unlock(&commit_transaction->t_handle_lock);
>> 427 spin_unlock(&journal->j_state_lock);
>> 428 schedule();
>> 429 spin_lock(&journal->j_state_lock);
>> 430 spin_lock(&commit_transaction->t_handle_lock);
>> 431 }
>> 432 finish_wait(&journal->j_wait_updates, &wait);
>> 433 }
>> 434 spin_unlock(&commit_transaction->t_handle_lock);
457 435
458 J_ASSERT (atomic_read(&commit_transact !! 436 J_ASSERT (commit_transaction->t_outstanding_credits <=
459 journal->j_max_transac 437 journal->j_max_transaction_buffers);
460 438
461 /* 439 /*
462 * First thing we are allowed to do is 440 * First thing we are allowed to do is to discard any remaining
463 * BJ_Reserved buffers. Note, it is _ 441 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
464 * that there are no such buffers: if 442 * that there are no such buffers: if a large filesystem
465 * operation like a truncate needs to 443 * operation like a truncate needs to split itself over multiple
466 * transactions, then it may try to do 444 * transactions, then it may try to do a jbd2_journal_restart() while
467 * there are still BJ_Reserved buffers 445 * there are still BJ_Reserved buffers outstanding. These must
468 * be released cleanly from the curren 446 * be released cleanly from the current transaction.
469 * 447 *
470 * In this case, the filesystem must s 448 * In this case, the filesystem must still reserve write access
471 * again before modifying the buffer i 449 * again before modifying the buffer in the new transaction, but
472 * we do not require it to remember ex 450 * we do not require it to remember exactly which old buffers it
473 * has reserved. This is consistent w 451 * has reserved. This is consistent with the existing behaviour
474 * that multiple jbd2_journal_get_writ 452 * that multiple jbd2_journal_get_write_access() calls to the same
475 * buffer are perfectly permissible. !! 453 * buffer are perfectly permissable.
476 * We use journal->j_state_lock here t <<
477 * t_reserved_list with eviction of bu <<
478 */ 454 */
479 while (commit_transaction->t_reserved_ 455 while (commit_transaction->t_reserved_list) {
480 jh = commit_transaction->t_res 456 jh = commit_transaction->t_reserved_list;
481 JBUFFER_TRACE(jh, "reserved, u 457 JBUFFER_TRACE(jh, "reserved, unused: refile");
482 /* 458 /*
483 * A jbd2_journal_get_undo_acc 459 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
484 * leave undo-committed data. 460 * leave undo-committed data.
485 */ 461 */
486 if (jh->b_committed_data) { 462 if (jh->b_committed_data) {
487 struct buffer_head *bh 463 struct buffer_head *bh = jh2bh(jh);
488 464
489 spin_lock(&jh->b_state !! 465 jbd_lock_bh_state(bh);
490 jbd2_free(jh->b_commit 466 jbd2_free(jh->b_committed_data, bh->b_size);
491 jh->b_committed_data = 467 jh->b_committed_data = NULL;
492 spin_unlock(&jh->b_sta !! 468 jbd_unlock_bh_state(bh);
493 } 469 }
494 jbd2_journal_refile_buffer(jou 470 jbd2_journal_refile_buffer(journal, jh);
495 } 471 }
496 472
497 write_unlock(&journal->j_state_lock); <<
498 /* 473 /*
499 * Now try to drop any written-back bu 474 * Now try to drop any written-back buffers from the journal's
500 * checkpoint lists. We do this *befo 475 * checkpoint lists. We do this *before* commit because it potentially
501 * frees some memory 476 * frees some memory
502 */ 477 */
503 spin_lock(&journal->j_list_lock); 478 spin_lock(&journal->j_list_lock);
504 __jbd2_journal_clean_checkpoint_list(j !! 479 __jbd2_journal_clean_checkpoint_list(journal);
505 spin_unlock(&journal->j_list_lock); 480 spin_unlock(&journal->j_list_lock);
506 481
507 jbd2_debug(3, "JBD2: commit phase 1\n" !! 482 jbd_debug (3, "JBD: commit phase 1\n");
508 <<
509 /* <<
510 * Clear revoked flag to reflect there <<
511 * in the next transaction which is go <<
512 */ <<
513 jbd2_clear_buffer_revoked_flags(journa <<
514 483
515 /* 484 /*
516 * Switch to a new revoke table. 485 * Switch to a new revoke table.
517 */ 486 */
518 jbd2_journal_switch_revoke_table(journ 487 jbd2_journal_switch_revoke_table(journal);
519 488
520 write_lock(&journal->j_state_lock); <<
521 /* <<
522 * Reserved credits cannot be claimed <<
523 */ <<
524 atomic_sub(atomic_read(&journal->j_res <<
525 &commit_transaction->t_outs <<
526 <<
527 trace_jbd2_commit_flushing(journal, co 489 trace_jbd2_commit_flushing(journal, commit_transaction);
528 stats.run.rs_flushing = jiffies; 490 stats.run.rs_flushing = jiffies;
529 stats.run.rs_locked = jbd2_time_diff(s 491 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
530 s 492 stats.run.rs_flushing);
531 493
532 commit_transaction->t_state = T_FLUSH; 494 commit_transaction->t_state = T_FLUSH;
533 journal->j_committing_transaction = co 495 journal->j_committing_transaction = commit_transaction;
534 journal->j_running_transaction = NULL; 496 journal->j_running_transaction = NULL;
535 start_time = ktime_get(); 497 start_time = ktime_get();
536 commit_transaction->t_log_start = jour 498 commit_transaction->t_log_start = journal->j_head;
537 wake_up_all(&journal->j_wait_transacti !! 499 wake_up(&journal->j_wait_transaction_locked);
538 write_unlock(&journal->j_state_lock); !! 500 spin_unlock(&journal->j_state_lock);
539 501
540 jbd2_debug(3, "JBD2: commit phase 2a\n !! 502 jbd_debug (3, "JBD: commit phase 2\n");
541 503
542 /* 504 /*
543 * Now start flushing things to disk, 505 * Now start flushing things to disk, in the order they appear
544 * on the transaction lists. Data blo 506 * on the transaction lists. Data blocks go first.
545 */ 507 */
546 err = journal_submit_data_buffers(jour 508 err = journal_submit_data_buffers(journal, commit_transaction);
547 if (err) 509 if (err)
548 jbd2_journal_abort(journal, er 510 jbd2_journal_abort(journal, err);
549 511
550 blk_start_plug(&plug); !! 512 jbd2_journal_write_revoke_records(journal, commit_transaction,
551 jbd2_journal_write_revoke_records(comm !! 513 write_op);
552 514
553 jbd2_debug(3, "JBD2: commit phase 2b\n !! 515 jbd_debug(3, "JBD: commit phase 2\n");
554 516
555 /* 517 /*
556 * Way to go: we have now written out 518 * Way to go: we have now written out all of the data for a
557 * transaction! Now comes the tricky 519 * transaction! Now comes the tricky part: we need to write out
558 * metadata. Loop over the transactio 520 * metadata. Loop over the transaction's entire buffer list:
559 */ 521 */
560 write_lock(&journal->j_state_lock); !! 522 spin_lock(&journal->j_state_lock);
561 commit_transaction->t_state = T_COMMIT 523 commit_transaction->t_state = T_COMMIT;
562 write_unlock(&journal->j_state_lock); !! 524 spin_unlock(&journal->j_state_lock);
563 525
564 trace_jbd2_commit_logging(journal, com 526 trace_jbd2_commit_logging(journal, commit_transaction);
565 stats.run.rs_logging = jiffies; 527 stats.run.rs_logging = jiffies;
566 stats.run.rs_flushing = jbd2_time_diff 528 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
567 529 stats.run.rs_logging);
568 stats.run.rs_blocks = commit_transacti !! 530 stats.run.rs_blocks = commit_transaction->t_outstanding_credits;
569 stats.run.rs_blocks_logged = 0; 531 stats.run.rs_blocks_logged = 0;
570 532
571 J_ASSERT(commit_transaction->t_nr_buff 533 J_ASSERT(commit_transaction->t_nr_buffers <=
572 atomic_read(&commit_transacti !! 534 commit_transaction->t_outstanding_credits);
573 535
574 bufs = 0; !! 536 err = 0;
575 descriptor = NULL; 537 descriptor = NULL;
>> 538 bufs = 0;
576 while (commit_transaction->t_buffers) 539 while (commit_transaction->t_buffers) {
577 540
578 /* Find the next buffer to be 541 /* Find the next buffer to be journaled... */
579 542
580 jh = commit_transaction->t_buf 543 jh = commit_transaction->t_buffers;
581 544
582 /* If we're in abort mode, we 545 /* If we're in abort mode, we just un-journal the buffer and
583 release it. */ 546 release it. */
584 547
585 if (is_journal_aborted(journal 548 if (is_journal_aborted(journal)) {
586 clear_buffer_jbddirty( 549 clear_buffer_jbddirty(jh2bh(jh));
587 JBUFFER_TRACE(jh, "jou 550 JBUFFER_TRACE(jh, "journal is aborting: refile");
588 jbd2_buffer_abort_trig 551 jbd2_buffer_abort_trigger(jh,
589 552 jh->b_frozen_data ?
590 553 jh->b_frozen_triggers :
591 554 jh->b_triggers);
592 jbd2_journal_refile_bu 555 jbd2_journal_refile_buffer(journal, jh);
593 /* If that was the las 556 /* If that was the last one, we need to clean up
594 * any descriptor buff 557 * any descriptor buffers which may have been
595 * already allocated, 558 * already allocated, even if we are now
596 * aborting. */ 559 * aborting. */
597 if (!commit_transactio 560 if (!commit_transaction->t_buffers)
598 goto start_jou 561 goto start_journal_io;
599 continue; 562 continue;
600 } 563 }
601 564
602 /* Make sure we have a descrip 565 /* Make sure we have a descriptor block in which to
603 record the metadata buffer. 566 record the metadata buffer. */
604 567
605 if (!descriptor) { 568 if (!descriptor) {
>> 569 struct buffer_head *bh;
>> 570
606 J_ASSERT (bufs == 0); 571 J_ASSERT (bufs == 0);
607 572
608 jbd2_debug(4, "JBD2: g !! 573 jbd_debug(4, "JBD: get descriptor\n");
609 574
610 descriptor = jbd2_jour !! 575 descriptor = jbd2_journal_get_descriptor_buffer(journal);
611 <<
612 <<
613 if (!descriptor) { 576 if (!descriptor) {
614 jbd2_journal_a 577 jbd2_journal_abort(journal, -EIO);
615 continue; 578 continue;
616 } 579 }
617 580
618 jbd2_debug(4, "JBD2: g !! 581 bh = jh2bh(descriptor);
619 (unsigned long !! 582 jbd_debug(4, "JBD: got buffer %llu (%p)\n",
620 descriptor->b_ !! 583 (unsigned long long)bh->b_blocknr, bh->b_data);
621 tagp = &descriptor->b_ !! 584 header = (journal_header_t *)&bh->b_data[0];
622 space_left = descripto !! 585 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
623 !! 586 header->h_blocktype = cpu_to_be32(JBD2_DESCRIPTOR_BLOCK);
>> 587 header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
>> 588
>> 589 tagp = &bh->b_data[sizeof(journal_header_t)];
>> 590 space_left = bh->b_size - sizeof(journal_header_t);
624 first_tag = 1; 591 first_tag = 1;
625 set_buffer_jwrite(desc !! 592 set_buffer_jwrite(bh);
626 set_buffer_dirty(descr !! 593 set_buffer_dirty(bh);
627 wbuf[bufs++] = descrip !! 594 wbuf[bufs++] = bh;
628 595
629 /* Record it so that w 596 /* Record it so that we can wait for IO
630 completion later */ 597 completion later */
631 BUFFER_TRACE(descripto !! 598 BUFFER_TRACE(bh, "ph3: file as descriptor");
632 jbd2_file_log_bh(&log_ !! 599 jbd2_journal_file_buffer(descriptor, commit_transaction,
>> 600 BJ_LogCtl);
633 } 601 }
634 602
635 /* Where is the buffer to be w 603 /* Where is the buffer to be written? */
636 604
637 err = jbd2_journal_next_log_bl 605 err = jbd2_journal_next_log_block(journal, &blocknr);
638 /* If the block mapping failed 606 /* If the block mapping failed, just abandon the buffer
639 and repeat this loop: we'll 607 and repeat this loop: we'll fall into the
640 refile-on-abort condition a 608 refile-on-abort condition above. */
641 if (err) { 609 if (err) {
642 jbd2_journal_abort(jou 610 jbd2_journal_abort(journal, err);
643 continue; 611 continue;
644 } 612 }
645 613
646 /* 614 /*
647 * start_this_handle() uses t_ 615 * start_this_handle() uses t_outstanding_credits to determine
648 * the free space in the log. !! 616 * the free space in the log, but this counter is changed
>> 617 * by jbd2_journal_next_log_block() also.
649 */ 618 */
650 atomic_dec(&commit_transaction !! 619 commit_transaction->t_outstanding_credits--;
651 620
652 /* Bump b_count to prevent tru 621 /* Bump b_count to prevent truncate from stumbling over
653 the shadowed buffer! @@@ T 622 the shadowed buffer! @@@ This can go if we ever get
654 rid of the shadow pairing o !! 623 rid of the BJ_IO/BJ_Shadow pairing of buffers. */
655 atomic_inc(&jh2bh(jh)->b_count 624 atomic_inc(&jh2bh(jh)->b_count);
656 625
>> 626 /* Make a temporary IO buffer with which to write it out
>> 627 (this will requeue both the metadata buffer and the
>> 628 temporary IO buffer). new_bh goes on BJ_IO*/
>> 629
>> 630 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
657 /* 631 /*
658 * Make a temporary IO buffer !! 632 * akpm: jbd2_journal_write_metadata_buffer() sets
659 * (this will requeue the meta !! 633 * new_bh->b_transaction to commit_transaction.
>> 634 * We need to clean this up before we release new_bh
>> 635 * (which is of type BJ_IO)
660 */ 636 */
661 set_bit(BH_JWrite, &jh2bh(jh)- <<
662 JBUFFER_TRACE(jh, "ph3: write 637 JBUFFER_TRACE(jh, "ph3: write metadata");
663 escape = jbd2_journal_write_me !! 638 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
664 !! 639 jh, &new_jh, blocknr);
665 if (escape < 0) { !! 640 if (flags < 0) {
666 jbd2_journal_abort(jou !! 641 jbd2_journal_abort(journal, flags);
667 continue; 642 continue;
668 } 643 }
669 jbd2_file_log_bh(&io_bufs, wbu !! 644 set_bit(BH_JWrite, &jh2bh(new_jh)->b_state);
>> 645 wbuf[bufs++] = jh2bh(new_jh);
670 646
671 /* Record the new block's tag 647 /* Record the new block's tag in the current descriptor
672 buffer */ 648 buffer */
673 649
674 tag_flag = 0; 650 tag_flag = 0;
675 if (escape) !! 651 if (flags & 1)
676 tag_flag |= JBD2_FLAG_ 652 tag_flag |= JBD2_FLAG_ESCAPE;
677 if (!first_tag) 653 if (!first_tag)
678 tag_flag |= JBD2_FLAG_ 654 tag_flag |= JBD2_FLAG_SAME_UUID;
679 655
680 tag = (journal_block_tag_t *) 656 tag = (journal_block_tag_t *) tagp;
681 write_tag_block(journal, tag, !! 657 write_tag_block(tag_bytes, tag, jh2bh(jh)->b_blocknr);
682 tag->t_flags = cpu_to_be16(tag !! 658 tag->t_flags = cpu_to_be32(tag_flag);
683 jbd2_block_tag_csum_set(journa <<
684 commit <<
685 tagp += tag_bytes; 659 tagp += tag_bytes;
686 space_left -= tag_bytes; 660 space_left -= tag_bytes;
687 bufs++; <<
688 661
689 if (first_tag) { 662 if (first_tag) {
690 memcpy (tagp, journal- 663 memcpy (tagp, journal->j_uuid, 16);
691 tagp += 16; 664 tagp += 16;
692 space_left -= 16; 665 space_left -= 16;
693 first_tag = 0; 666 first_tag = 0;
694 } 667 }
695 668
696 /* If there's no more to do, o 669 /* If there's no more to do, or if the descriptor is full,
697 let the IO rip! */ 670 let the IO rip! */
698 671
699 if (bufs == journal->j_wbufsiz 672 if (bufs == journal->j_wbufsize ||
700 commit_transaction->t_buff 673 commit_transaction->t_buffers == NULL ||
701 space_left < tag_bytes + 1 !! 674 space_left < tag_bytes + 16) {
702 675
703 jbd2_debug(4, "JBD2: S !! 676 jbd_debug(4, "JBD: Submit %d IOs\n", bufs);
704 677
705 /* Write an end-of-des 678 /* Write an end-of-descriptor marker before
706 submitting the IOs. 679 submitting the IOs. "tag" still points to
707 the last tag we set 680 the last tag we set up. */
708 681
709 tag->t_flags |= cpu_to !! 682 tag->t_flags |= cpu_to_be32(JBD2_FLAG_LAST_TAG);
710 start_journal_io: <<
711 if (descriptor) <<
712 jbd2_descripto <<
713 <<
714 683
>> 684 start_journal_io:
715 for (i = 0; i < bufs; 685 for (i = 0; i < bufs; i++) {
716 struct buffer_ 686 struct buffer_head *bh = wbuf[i];
717 <<
718 /* 687 /*
719 * Compute che 688 * Compute checksum.
720 */ 689 */
721 if (jbd2_has_f !! 690 if (JBD2_HAS_COMPAT_FEATURE(journal,
>> 691 JBD2_FEATURE_COMPAT_CHECKSUM)) {
722 crc32_ 692 crc32_sum =
723 jb 693 jbd2_checksum_data(crc32_sum, bh);
724 } 694 }
725 695
726 lock_buffer(bh 696 lock_buffer(bh);
727 clear_buffer_d 697 clear_buffer_dirty(bh);
728 set_buffer_upt 698 set_buffer_uptodate(bh);
729 bh->b_end_io = 699 bh->b_end_io = journal_end_buffer_io_sync;
730 submit_bh(REQ_ !! 700 submit_bh(write_op, bh);
731 bh); <<
732 } 701 }
733 cond_resched(); 702 cond_resched();
>> 703 stats.run.rs_blocks_logged += bufs;
734 704
735 /* Force a new descrip 705 /* Force a new descriptor to be generated next
736 time round the loop 706 time round the loop. */
737 descriptor = NULL; 707 descriptor = NULL;
738 bufs = 0; 708 bufs = 0;
739 } 709 }
740 } 710 }
741 711
742 err = journal_finish_inode_data_buffer <<
743 if (err) { <<
744 printk(KERN_WARNING <<
745 "JBD2: Detected IO err <<
746 "on %s\n", journal->j_d <<
747 if (journal->j_flags & JBD2_AB <<
748 jbd2_journal_abort(jou <<
749 err = 0; <<
750 } <<
751 <<
752 /* <<
753 * Get current oldest transaction in t <<
754 * to the filesystem device. After the <<
755 * blocks of all older transactions ar <<
756 * storage and we will be safe to upda <<
757 * superblock with the numbers we get <<
758 */ <<
759 update_tail = <<
760 jbd2_journal_get_log_tail(jour <<
761 <<
762 write_lock(&journal->j_state_lock); <<
763 if (update_tail) { <<
764 long freed = first_block - jou <<
765 <<
766 if (first_block < journal->j_t <<
767 freed += journal->j_la <<
768 /* Update tail only if we free <<
769 if (freed < journal->j_max_tra <<
770 update_tail = 0; <<
771 } <<
772 J_ASSERT(commit_transaction->t_state = <<
773 commit_transaction->t_state = T_COMMIT <<
774 write_unlock(&journal->j_state_lock); <<
775 <<
776 /* 712 /*
777 * If the journal is not located on th 713 * If the journal is not located on the file system device,
778 * then we must flush the file system 714 * then we must flush the file system device before we issue
779 * the commit record 715 * the commit record
780 */ 716 */
781 if (commit_transaction->t_need_data_fl !! 717 if (commit_transaction->t_flushed_data_blocks &&
782 (journal->j_fs_dev != journal->j_d 718 (journal->j_fs_dev != journal->j_dev) &&
783 (journal->j_flags & JBD2_BARRIER)) 719 (journal->j_flags & JBD2_BARRIER))
784 blkdev_issue_flush(journal->j_ !! 720 blkdev_issue_flush(journal->j_fs_dev, NULL);
785 721
786 /* Done it all: now write the commit r 722 /* Done it all: now write the commit record asynchronously. */
787 if (jbd2_has_feature_async_commit(jour !! 723 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
>> 724 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
788 err = journal_submit_commit_re 725 err = journal_submit_commit_record(journal, commit_transaction,
789 726 &cbh, crc32_sum);
790 if (err) 727 if (err)
791 jbd2_journal_abort(jou !! 728 __jbd2_journal_abort_hard(journal);
>> 729 if (journal->j_flags & JBD2_BARRIER)
>> 730 blkdev_issue_flush(journal->j_dev, NULL);
792 } 731 }
793 732
794 blk_finish_plug(&plug); !! 733 err = journal_finish_inode_data_buffers(journal, commit_transaction);
>> 734 if (err) {
>> 735 printk(KERN_WARNING
>> 736 "JBD2: Detected IO errors while flushing file data "
>> 737 "on %s\n", journal->j_devname);
>> 738 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
>> 739 jbd2_journal_abort(journal, err);
>> 740 err = 0;
>> 741 }
795 742
796 /* Lo and behold: we have just managed 743 /* Lo and behold: we have just managed to send a transaction to
797 the log. Before we can commit it, 744 the log. Before we can commit it, wait for the IO so far to
798 complete. Control buffers being wr 745 complete. Control buffers being written are on the
799 transaction's t_log_list queue, and 746 transaction's t_log_list queue, and metadata buffers are on
800 the io_bufs list. !! 747 the t_iobuf_list queue.
801 748
802 Wait for the buffers in reverse ord 749 Wait for the buffers in reverse order. That way we are
803 less likely to be woken up until al 750 less likely to be woken up until all IOs have completed, and
804 so we incur less scheduling load. 751 so we incur less scheduling load.
805 */ 752 */
806 753
807 jbd2_debug(3, "JBD2: commit phase 3\n" !! 754 jbd_debug(3, "JBD: commit phase 3\n");
808 755
809 while (!list_empty(&io_bufs)) { !! 756 /*
810 struct buffer_head *bh = list_ !! 757 * akpm: these are BJ_IO, and j_list_lock is not needed.
811 !! 758 * See __journal_try_to_free_buffer.
812 !! 759 */
>> 760 wait_for_iobuf:
>> 761 while (commit_transaction->t_iobuf_list != NULL) {
>> 762 struct buffer_head *bh;
813 763
814 wait_on_buffer(bh); !! 764 jh = commit_transaction->t_iobuf_list->b_tprev;
815 cond_resched(); !! 765 bh = jh2bh(jh);
>> 766 if (buffer_locked(bh)) {
>> 767 wait_on_buffer(bh);
>> 768 goto wait_for_iobuf;
>> 769 }
>> 770 if (cond_resched())
>> 771 goto wait_for_iobuf;
816 772
817 if (unlikely(!buffer_uptodate( 773 if (unlikely(!buffer_uptodate(bh)))
818 err = -EIO; 774 err = -EIO;
819 jbd2_unfile_log_bh(bh); !! 775
820 stats.run.rs_blocks_logged++; !! 776 clear_buffer_jwrite(bh);
>> 777
>> 778 JBUFFER_TRACE(jh, "ph4: unfile after journal write");
>> 779 jbd2_journal_unfile_buffer(journal, jh);
821 780
822 /* 781 /*
823 * The list contains temporary !! 782 * ->t_iobuf_list should contain only dummy buffer_heads
824 * jbd2_journal_write_metadata !! 783 * which were created by jbd2_journal_write_metadata_buffer().
825 */ 784 */
826 BUFFER_TRACE(bh, "dumping temp 785 BUFFER_TRACE(bh, "dumping temporary bh");
>> 786 jbd2_journal_put_journal_head(jh);
827 __brelse(bh); 787 __brelse(bh);
828 J_ASSERT_BH(bh, atomic_read(&b 788 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
829 free_buffer_head(bh); 789 free_buffer_head(bh);
830 790
831 /* We also have to refile the !! 791 /* We also have to unlock and free the corresponding
>> 792 shadowed buffer */
832 jh = commit_transaction->t_sha 793 jh = commit_transaction->t_shadow_list->b_tprev;
833 bh = jh2bh(jh); 794 bh = jh2bh(jh);
834 clear_buffer_jwrite(bh); !! 795 clear_bit(BH_JWrite, &bh->b_state);
835 J_ASSERT_BH(bh, buffer_jbddirt 796 J_ASSERT_BH(bh, buffer_jbddirty(bh));
836 J_ASSERT_BH(bh, !buffer_shadow <<
837 797
838 /* The metadata is now release 798 /* The metadata is now released for reuse, but we need
839 to remember it against this 799 to remember it against this transaction so that when
840 we finally commit, we can d 800 we finally commit, we can do any checkpointing
841 required. */ 801 required. */
842 JBUFFER_TRACE(jh, "file as BJ_ 802 JBUFFER_TRACE(jh, "file as BJ_Forget");
843 jbd2_journal_file_buffer(jh, c 803 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
>> 804 /* Wake up any transactions which were waiting for this
>> 805 IO to complete */
>> 806 wake_up_bit(&bh->b_state, BH_Unshadow);
844 JBUFFER_TRACE(jh, "brelse shad 807 JBUFFER_TRACE(jh, "brelse shadowed buffer");
845 __brelse(bh); 808 __brelse(bh);
846 } 809 }
847 810
848 J_ASSERT (commit_transaction->t_shadow 811 J_ASSERT (commit_transaction->t_shadow_list == NULL);
849 812
850 jbd2_debug(3, "JBD2: commit phase 4\n" !! 813 jbd_debug(3, "JBD: commit phase 4\n");
851 814
852 /* Here we wait for the revoke record 815 /* Here we wait for the revoke record and descriptor record buffers */
853 while (!list_empty(&log_bufs)) { !! 816 wait_for_ctlbuf:
>> 817 while (commit_transaction->t_log_list != NULL) {
854 struct buffer_head *bh; 818 struct buffer_head *bh;
855 819
856 bh = list_entry(log_bufs.prev, !! 820 jh = commit_transaction->t_log_list->b_tprev;
857 wait_on_buffer(bh); !! 821 bh = jh2bh(jh);
858 cond_resched(); !! 822 if (buffer_locked(bh)) {
>> 823 wait_on_buffer(bh);
>> 824 goto wait_for_ctlbuf;
>> 825 }
>> 826 if (cond_resched())
>> 827 goto wait_for_ctlbuf;
859 828
860 if (unlikely(!buffer_uptodate( 829 if (unlikely(!buffer_uptodate(bh)))
861 err = -EIO; 830 err = -EIO;
862 831
863 BUFFER_TRACE(bh, "ph5: control 832 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
864 clear_buffer_jwrite(bh); 833 clear_buffer_jwrite(bh);
865 jbd2_unfile_log_bh(bh); !! 834 jbd2_journal_unfile_buffer(journal, jh);
866 stats.run.rs_blocks_logged++; !! 835 jbd2_journal_put_journal_head(jh);
867 __brelse(bh); /* One 836 __brelse(bh); /* One for getblk */
868 /* AKPM: bforget here */ 837 /* AKPM: bforget here */
869 } 838 }
870 839
871 if (err) 840 if (err)
872 jbd2_journal_abort(journal, er 841 jbd2_journal_abort(journal, err);
873 842
874 jbd2_debug(3, "JBD2: commit phase 5\n" !! 843 jbd_debug(3, "JBD: commit phase 5\n");
875 write_lock(&journal->j_state_lock); <<
876 J_ASSERT(commit_transaction->t_state = <<
877 commit_transaction->t_state = T_COMMIT <<
878 write_unlock(&journal->j_state_lock); <<
879 844
880 if (!jbd2_has_feature_async_commit(jou !! 845 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
>> 846 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
881 err = journal_submit_commit_re 847 err = journal_submit_commit_record(journal, commit_transaction,
882 848 &cbh, crc32_sum);
883 if (err) 849 if (err)
884 jbd2_journal_abort(jou !! 850 __jbd2_journal_abort_hard(journal);
885 } 851 }
886 if (cbh) !! 852 if (!err && !is_journal_aborted(journal))
887 err = journal_wait_on_commit_r 853 err = journal_wait_on_commit_record(journal, cbh);
888 stats.run.rs_blocks_logged++; <<
889 if (jbd2_has_feature_async_commit(jour <<
890 journal->j_flags & JBD2_BARRIER) { <<
891 blkdev_issue_flush(journal->j_ <<
892 } <<
893 854
894 if (err) 855 if (err)
895 jbd2_journal_abort(journal, er 856 jbd2_journal_abort(journal, err);
896 857
897 WARN_ON_ONCE( <<
898 atomic_read(&commit_transactio <<
899 <<
900 /* <<
901 * Now disk caches for filesystem devi <<
902 * erase checkpointed transactions fro <<
903 * superblock. <<
904 */ <<
905 if (update_tail) <<
906 jbd2_update_log_tail(journal, <<
907 <<
908 /* End of a transaction! Finally, we 858 /* End of a transaction! Finally, we can do checkpoint
909 processing: any buffers committed a 859 processing: any buffers committed as a result of this
910 transaction can be removed from any 860 transaction can be removed from any checkpoint list it was on
911 before. */ 861 before. */
912 862
913 jbd2_debug(3, "JBD2: commit phase 6\n" !! 863 jbd_debug(3, "JBD: commit phase 6\n");
914 864
915 J_ASSERT(list_empty(&commit_transactio 865 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
916 J_ASSERT(commit_transaction->t_buffers 866 J_ASSERT(commit_transaction->t_buffers == NULL);
917 J_ASSERT(commit_transaction->t_checkpo 867 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
>> 868 J_ASSERT(commit_transaction->t_iobuf_list == NULL);
918 J_ASSERT(commit_transaction->t_shadow_ 869 J_ASSERT(commit_transaction->t_shadow_list == NULL);
>> 870 J_ASSERT(commit_transaction->t_log_list == NULL);
919 871
920 restart_loop: 872 restart_loop:
921 /* 873 /*
922 * As there are other places (journal_ 874 * As there are other places (journal_unmap_buffer()) adding buffers
923 * to this list we have to be careful 875 * to this list we have to be careful and hold the j_list_lock.
924 */ 876 */
925 spin_lock(&journal->j_list_lock); 877 spin_lock(&journal->j_list_lock);
926 while (commit_transaction->t_forget) { 878 while (commit_transaction->t_forget) {
927 transaction_t *cp_transaction; 879 transaction_t *cp_transaction;
928 struct buffer_head *bh; 880 struct buffer_head *bh;
929 int try_to_free = 0; <<
930 bool drop_ref; <<
931 881
932 jh = commit_transaction->t_for 882 jh = commit_transaction->t_forget;
933 spin_unlock(&journal->j_list_l 883 spin_unlock(&journal->j_list_lock);
934 bh = jh2bh(jh); 884 bh = jh2bh(jh);
935 /* !! 885 jbd_lock_bh_state(bh);
936 * Get a reference so that bh !! 886 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction ||
937 * done with it. !! 887 jh->b_transaction == journal->j_running_transaction);
938 */ <<
939 get_bh(bh); <<
940 spin_lock(&jh->b_state_lock); <<
941 J_ASSERT_JH(jh, jh->b_transact <<
942 888
943 /* 889 /*
944 * If there is undo-protected 890 * If there is undo-protected committed data against
945 * this buffer, then we can re 891 * this buffer, then we can remove it now. If it is a
946 * buffer needing such protect 892 * buffer needing such protection, the old frozen_data
947 * field now points to a commi 893 * field now points to a committed version of the
948 * buffer, so rotate that fiel 894 * buffer, so rotate that field to the new committed
949 * data. 895 * data.
950 * 896 *
951 * Otherwise, we can just thro 897 * Otherwise, we can just throw away the frozen data now.
952 * 898 *
953 * We also know that the froze 899 * We also know that the frozen data has already fired
954 * its triggers if they exist, 900 * its triggers if they exist, so we can clear that too.
955 */ 901 */
956 if (jh->b_committed_data) { 902 if (jh->b_committed_data) {
957 jbd2_free(jh->b_commit 903 jbd2_free(jh->b_committed_data, bh->b_size);
958 jh->b_committed_data = 904 jh->b_committed_data = NULL;
959 if (jh->b_frozen_data) 905 if (jh->b_frozen_data) {
960 jh->b_committe 906 jh->b_committed_data = jh->b_frozen_data;
961 jh->b_frozen_d 907 jh->b_frozen_data = NULL;
962 jh->b_frozen_t 908 jh->b_frozen_triggers = NULL;
963 } 909 }
964 } else if (jh->b_frozen_data) 910 } else if (jh->b_frozen_data) {
965 jbd2_free(jh->b_frozen 911 jbd2_free(jh->b_frozen_data, bh->b_size);
966 jh->b_frozen_data = NU 912 jh->b_frozen_data = NULL;
967 jh->b_frozen_triggers 913 jh->b_frozen_triggers = NULL;
968 } 914 }
969 915
970 spin_lock(&journal->j_list_loc 916 spin_lock(&journal->j_list_lock);
971 cp_transaction = jh->b_cp_tran 917 cp_transaction = jh->b_cp_transaction;
972 if (cp_transaction) { 918 if (cp_transaction) {
973 JBUFFER_TRACE(jh, "rem 919 JBUFFER_TRACE(jh, "remove from old cp transaction");
974 cp_transaction->t_chp_ 920 cp_transaction->t_chp_stats.cs_dropped++;
975 __jbd2_journal_remove_ 921 __jbd2_journal_remove_checkpoint(jh);
976 } 922 }
977 923
978 /* Only re-checkpoint the buff 924 /* Only re-checkpoint the buffer_head if it is marked
979 * dirty. If the buffer was a 925 * dirty. If the buffer was added to the BJ_Forget list
980 * by jbd2_journal_forget, it 926 * by jbd2_journal_forget, it may no longer be dirty and
981 * there's no point in keeping 927 * there's no point in keeping a checkpoint record for
982 * it. */ 928 * it. */
983 929
984 /* !! 930 /* A buffer which has been freed while still being
985 * A buffer which has been fre !! 931 * journaled by a previous transaction may end up still
986 * by a previous transaction, !! 932 * being dirty here, but we want to avoid writing back
987 * the running transaction. If !! 933 * that buffer in the future now that the last use has
988 * contains "add to orphan" op !! 934 * been committed. That's not only a performance gain,
989 * invalidate the buffer now. !! 935 * it also stops aliasing problems if the buffer is left
990 * since the buffer may be sti !! 936 * behind for writeback and gets reallocated for another
991 * pagesize and it is attached !! 937 * use in a different page. */
992 */ !! 938 if (buffer_freed(bh)) {
993 if (buffer_freed(bh) && !jh->b <<
994 struct address_space * <<
995 <<
996 clear_buffer_freed(bh) 939 clear_buffer_freed(bh);
997 clear_buffer_jbddirty( 940 clear_buffer_jbddirty(bh);
998 <<
999 /* <<
1000 * Block device buffe <<
1001 * time, so it is eno <<
1002 * buffer_freed bits. <<
1003 * journalled data) w <<
1004 * more bits. We also <<
1005 * because the data p <<
1006 * our hands. Note th <<
1007 * need to make buffe <<
1008 * already detached f <<
1009 * get reused. <<
1010 */ <<
1011 mapping = READ_ONCE(b <<
1012 if (mapping && !sb_is <<
1013 clear_buffer_ <<
1014 clear_buffer_ <<
1015 clear_buffer_ <<
1016 bh->b_bdev = <<
1017 } <<
1018 } 941 }
1019 942
1020 if (buffer_jbddirty(bh)) { 943 if (buffer_jbddirty(bh)) {
1021 JBUFFER_TRACE(jh, "ad 944 JBUFFER_TRACE(jh, "add to new checkpointing trans");
1022 __jbd2_journal_insert 945 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
1023 if (is_journal_aborte 946 if (is_journal_aborted(journal))
1024 clear_buffer_ 947 clear_buffer_jbddirty(bh);
>> 948 JBUFFER_TRACE(jh, "refile for checkpoint writeback");
>> 949 __jbd2_journal_refile_buffer(jh);
>> 950 jbd_unlock_bh_state(bh);
1025 } else { 951 } else {
1026 J_ASSERT_BH(bh, !buff 952 J_ASSERT_BH(bh, !buffer_dirty(bh));
1027 /* !! 953 /* The buffer on BJ_Forget list and not jbddirty means
1028 * The buffer on BJ_F <<
1029 * it has been freed 954 * it has been freed by this transaction and hence it
1030 * could not have bee 955 * could not have been reallocated until this
1031 * transaction has co 956 * transaction has committed. *BUT* it could be
1032 * reallocated once w 957 * reallocated once we have written all the data to
1033 * disk and before we 958 * disk and before we process the buffer on BJ_Forget
1034 * list. !! 959 * list. */
1035 */ !! 960 JBUFFER_TRACE(jh, "refile or unfile freed buffer");
1036 if (!jh->b_next_trans !! 961 __jbd2_journal_refile_buffer(jh);
1037 try_to_free = !! 962 if (!jh->b_transaction) {
>> 963 jbd_unlock_bh_state(bh);
>> 964 /* needs a brelse */
>> 965 jbd2_journal_remove_journal_head(bh);
>> 966 release_buffer_page(bh);
>> 967 } else
>> 968 jbd_unlock_bh_state(bh);
1038 } 969 }
1039 JBUFFER_TRACE(jh, "refile or <<
1040 drop_ref = __jbd2_journal_ref <<
1041 spin_unlock(&jh->b_state_lock <<
1042 if (drop_ref) <<
1043 jbd2_journal_put_jour <<
1044 if (try_to_free) <<
1045 release_buffer_page(b <<
1046 else <<
1047 __brelse(bh); <<
1048 cond_resched_lock(&journal->j 970 cond_resched_lock(&journal->j_list_lock);
1049 } 971 }
1050 spin_unlock(&journal->j_list_lock); 972 spin_unlock(&journal->j_list_lock);
1051 /* 973 /*
1052 * This is a bit sleazy. We use j_li 974 * This is a bit sleazy. We use j_list_lock to protect transition
1053 * of a transaction into T_FINISHED s 975 * of a transaction into T_FINISHED state and calling
1054 * __jbd2_journal_drop_transaction(). 976 * __jbd2_journal_drop_transaction(). Otherwise we could race with
1055 * other checkpointing code processin 977 * other checkpointing code processing the transaction...
1056 */ 978 */
1057 write_lock(&journal->j_state_lock); !! 979 spin_lock(&journal->j_state_lock);
1058 spin_lock(&journal->j_list_lock); 980 spin_lock(&journal->j_list_lock);
1059 /* 981 /*
1060 * Now recheck if some buffers did no 982 * Now recheck if some buffers did not get attached to the transaction
1061 * while the lock was dropped... 983 * while the lock was dropped...
1062 */ 984 */
1063 if (commit_transaction->t_forget) { 985 if (commit_transaction->t_forget) {
1064 spin_unlock(&journal->j_list_ 986 spin_unlock(&journal->j_list_lock);
1065 write_unlock(&journal->j_stat !! 987 spin_unlock(&journal->j_state_lock);
1066 goto restart_loop; 988 goto restart_loop;
1067 } 989 }
1068 990
1069 /* Add the transaction to the checkpo <<
1070 * __journal_remove_checkpoint() can <<
1071 * under us because it is not marked <<
1072 if (journal->j_checkpoint_transaction <<
1073 journal->j_checkpoint_transac <<
1074 commit_transaction->t_cpnext <<
1075 commit_transaction->t_cpprev <<
1076 } else { <<
1077 commit_transaction->t_cpnext <<
1078 journal->j_checkpoint <<
1079 commit_transaction->t_cpprev <<
1080 commit_transaction->t <<
1081 commit_transaction->t_cpnext- <<
1082 commit_transaction; <<
1083 commit_transaction->t_cpprev- <<
1084 commit_transa <<
1085 } <<
1086 spin_unlock(&journal->j_list_lock); <<
1087 <<
1088 /* Done with this transaction! */ 991 /* Done with this transaction! */
1089 992
1090 jbd2_debug(3, "JBD2: commit phase 7\n !! 993 jbd_debug(3, "JBD: commit phase 7\n");
1091 994
1092 J_ASSERT(commit_transaction->t_state !! 995 J_ASSERT(commit_transaction->t_state == T_COMMIT);
1093 996
1094 commit_transaction->t_start = jiffies 997 commit_transaction->t_start = jiffies;
1095 stats.run.rs_logging = jbd2_time_diff 998 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
1096 999 commit_transaction->t_start);
1097 1000
1098 /* 1001 /*
1099 * File the transaction statistics 1002 * File the transaction statistics
1100 */ 1003 */
1101 stats.ts_tid = commit_transaction->t_ 1004 stats.ts_tid = commit_transaction->t_tid;
1102 stats.run.rs_handle_count = !! 1005 stats.run.rs_handle_count = commit_transaction->t_handle_count;
1103 atomic_read(&commit_transacti <<
1104 trace_jbd2_run_stats(journal->j_fs_de 1006 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
1105 commit_transacti 1007 commit_transaction->t_tid, &stats.run);
1106 stats.ts_requested = (commit_transact <<
1107 1008
1108 commit_transaction->t_state = T_COMMI !! 1009 /*
>> 1010 * Calculate overall stats
>> 1011 */
>> 1012 spin_lock(&journal->j_history_lock);
>> 1013 journal->j_stats.ts_tid++;
>> 1014 journal->j_stats.run.rs_wait += stats.run.rs_wait;
>> 1015 journal->j_stats.run.rs_running += stats.run.rs_running;
>> 1016 journal->j_stats.run.rs_locked += stats.run.rs_locked;
>> 1017 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
>> 1018 journal->j_stats.run.rs_logging += stats.run.rs_logging;
>> 1019 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
>> 1020 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
>> 1021 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
>> 1022 spin_unlock(&journal->j_history_lock);
>> 1023
>> 1024 commit_transaction->t_state = T_FINISHED;
1109 J_ASSERT(commit_transaction == journa 1025 J_ASSERT(commit_transaction == journal->j_committing_transaction);
1110 WRITE_ONCE(journal->j_commit_sequence !! 1026 journal->j_commit_sequence = commit_transaction->t_tid;
1111 journal->j_committing_transaction = N 1027 journal->j_committing_transaction = NULL;
1112 commit_time = ktime_to_ns(ktime_sub(k 1028 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1113 1029
1114 /* 1030 /*
1115 * weight the commit time higher than 1031 * weight the commit time higher than the average time so we don't
1116 * react too strongly to vast changes 1032 * react too strongly to vast changes in the commit time
1117 */ 1033 */
1118 if (likely(journal->j_average_commit_ 1034 if (likely(journal->j_average_commit_time))
1119 journal->j_average_commit_tim 1035 journal->j_average_commit_time = (commit_time +
1120 journal->j_av 1036 journal->j_average_commit_time*3) / 4;
1121 else 1037 else
1122 journal->j_average_commit_tim 1038 journal->j_average_commit_time = commit_time;
>> 1039 spin_unlock(&journal->j_state_lock);
1123 1040
1124 write_unlock(&journal->j_state_lock); !! 1041 if (commit_transaction->t_checkpoint_list == NULL &&
>> 1042 commit_transaction->t_checkpoint_io_list == NULL) {
>> 1043 __jbd2_journal_drop_transaction(journal, commit_transaction);
>> 1044 to_free = 1;
>> 1045 } else {
>> 1046 if (journal->j_checkpoint_transactions == NULL) {
>> 1047 journal->j_checkpoint_transactions = commit_transaction;
>> 1048 commit_transaction->t_cpnext = commit_transaction;
>> 1049 commit_transaction->t_cpprev = commit_transaction;
>> 1050 } else {
>> 1051 commit_transaction->t_cpnext =
>> 1052 journal->j_checkpoint_transactions;
>> 1053 commit_transaction->t_cpprev =
>> 1054 commit_transaction->t_cpnext->t_cpprev;
>> 1055 commit_transaction->t_cpnext->t_cpprev =
>> 1056 commit_transaction;
>> 1057 commit_transaction->t_cpprev->t_cpnext =
>> 1058 commit_transaction;
>> 1059 }
>> 1060 }
>> 1061 spin_unlock(&journal->j_list_lock);
1125 1062
1126 if (journal->j_commit_callback) 1063 if (journal->j_commit_callback)
1127 journal->j_commit_callback(jo 1064 journal->j_commit_callback(journal, commit_transaction);
1128 if (journal->j_fc_cleanup_callback) <<
1129 journal->j_fc_cleanup_callbac <<
1130 1065
1131 trace_jbd2_end_commit(journal, commit 1066 trace_jbd2_end_commit(journal, commit_transaction);
1132 jbd2_debug(1, "JBD2: commit %d comple !! 1067 jbd_debug(1, "JBD: commit %d complete, head %d\n",
1133 journal->j_commit_sequence, 1068 journal->j_commit_sequence, journal->j_tail_sequence);
>> 1069 if (to_free)
>> 1070 kfree(commit_transaction);
1134 1071
1135 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); <<
1139 commit_transaction->t_state = T_FINIS <<
1140 /* Check if the transaction can be dr <<
1141 if (commit_transaction->t_checkpoint_ <<
1142 __jbd2_journal_drop_transacti <<
1143 jbd2_journal_free_transaction <<
1144 } <<
1145 spin_unlock(&journal->j_list_lock); <<
1146 write_unlock(&journal->j_state_lock); <<
1147 wake_up(&journal->j_wait_done_commit) 1072 wake_up(&journal->j_wait_done_commit);
1148 wake_up(&journal->j_fc_wait); <<
1149 <<
1150 /* <<
1151 * Calculate overall stats <<
1152 */ <<
1153 spin_lock(&journal->j_history_lock); <<
1154 journal->j_stats.ts_tid++; <<
1155 journal->j_stats.ts_requested += stat <<
1156 journal->j_stats.run.rs_wait += stats <<
1157 journal->j_stats.run.rs_request_delay <<
1158 journal->j_stats.run.rs_running += st <<
1159 journal->j_stats.run.rs_locked += sta <<
1160 journal->j_stats.run.rs_flushing += s <<
1161 journal->j_stats.run.rs_logging += st <<
1162 journal->j_stats.run.rs_handle_count <<
1163 journal->j_stats.run.rs_blocks += sta <<
1164 journal->j_stats.run.rs_blocks_logged <<
1165 spin_unlock(&journal->j_history_lock) <<
1166 } 1073 }
1167 1074
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.