1 // SPDX-License-Identifier: GPL-2.0+ 1 // SPDX-License-Identifier: GPL-2.0+
2 /* 2 /*
3 * linux/fs/jbd2/recovery.c 3 * linux/fs/jbd2/recovery.c
4 * 4 *
5 * Written by Stephen C. Tweedie <sct@redhat.c 5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
6 * 6 *
7 * Copyright 1999-2000 Red Hat Software --- Al 7 * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
8 * 8 *
9 * Journal recovery routines for the generic f 9 * Journal recovery routines for the generic filesystem journaling code;
10 * part of the ext2fs journaling system. 10 * part of the ext2fs journaling system.
11 */ 11 */
12 12
13 #ifndef __KERNEL__ 13 #ifndef __KERNEL__
14 #include "jfs_user.h" 14 #include "jfs_user.h"
15 #else 15 #else
16 #include <linux/time.h> 16 #include <linux/time.h>
17 #include <linux/fs.h> 17 #include <linux/fs.h>
18 #include <linux/jbd2.h> 18 #include <linux/jbd2.h>
19 #include <linux/errno.h> 19 #include <linux/errno.h>
20 #include <linux/crc32.h> 20 #include <linux/crc32.h>
21 #include <linux/blkdev.h> 21 #include <linux/blkdev.h>
22 #include <linux/string_choices.h> <<
23 #endif 22 #endif
24 23
25 /* 24 /*
26 * Maintain information about the progress of 25 * Maintain information about the progress of the recovery job, so that
27 * the different passes can carry information 26 * the different passes can carry information between them.
28 */ 27 */
29 struct recovery_info 28 struct recovery_info
30 { 29 {
31 tid_t start_transaction; 30 tid_t start_transaction;
32 tid_t end_transaction; 31 tid_t end_transaction;
33 unsigned long head_block; <<
34 32
35 int nr_replays; 33 int nr_replays;
36 int nr_revokes; 34 int nr_revokes;
37 int nr_revoke_hits; 35 int nr_revoke_hits;
38 }; 36 };
39 37
40 static int do_one_pass(journal_t *journal, 38 static int do_one_pass(journal_t *journal,
41 struct recover 39 struct recovery_info *info, enum passtype pass);
42 static int scan_revoke_records(journal_t *, st 40 static int scan_revoke_records(journal_t *, struct buffer_head *,
43 tid_t, struct 41 tid_t, struct recovery_info *);
44 42
45 #ifdef __KERNEL__ 43 #ifdef __KERNEL__
46 44
47 /* Release readahead buffers after use */ 45 /* Release readahead buffers after use */
48 static void journal_brelse_array(struct buffer 46 static void journal_brelse_array(struct buffer_head *b[], int n)
49 { 47 {
50 while (--n >= 0) 48 while (--n >= 0)
51 brelse (b[n]); 49 brelse (b[n]);
52 } 50 }
53 51
54 52
55 /* 53 /*
56 * When reading from the journal, we are going 54 * When reading from the journal, we are going through the block device
57 * layer directly and so there is no readahead 55 * layer directly and so there is no readahead being done for us. We
58 * need to implement any readahead ourselves i 56 * need to implement any readahead ourselves if we want it to happen at
59 * all. Recovery is basically one long sequen 57 * all. Recovery is basically one long sequential read, so make sure we
60 * do the IO in reasonably large chunks. 58 * do the IO in reasonably large chunks.
61 * 59 *
62 * This is not so critical that we need to be 60 * This is not so critical that we need to be enormously clever about
63 * the readahead size, though. 128K is a pure 61 * the readahead size, though. 128K is a purely arbitrary, good-enough
64 * fixed value. 62 * fixed value.
65 */ 63 */
66 64
67 #define MAXBUF 8 65 #define MAXBUF 8
68 static int do_readahead(journal_t *journal, un 66 static int do_readahead(journal_t *journal, unsigned int start)
69 { 67 {
70 int err; 68 int err;
71 unsigned int max, nbufs, next; 69 unsigned int max, nbufs, next;
72 unsigned long long blocknr; 70 unsigned long long blocknr;
73 struct buffer_head *bh; 71 struct buffer_head *bh;
74 72
75 struct buffer_head * bufs[MAXBUF]; 73 struct buffer_head * bufs[MAXBUF];
76 74
77 /* Do up to 128K of readahead */ 75 /* Do up to 128K of readahead */
78 max = start + (128 * 1024 / journal->j 76 max = start + (128 * 1024 / journal->j_blocksize);
79 if (max > journal->j_total_len) 77 if (max > journal->j_total_len)
80 max = journal->j_total_len; 78 max = journal->j_total_len;
81 79
82 /* Do the readahead itself. We'll sub 80 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
83 * a time to the block device IO layer 81 * a time to the block device IO layer. */
84 82
85 nbufs = 0; 83 nbufs = 0;
86 84
87 for (next = start; next < max; next++) 85 for (next = start; next < max; next++) {
88 err = jbd2_journal_bmap(journa 86 err = jbd2_journal_bmap(journal, next, &blocknr);
89 87
90 if (err) { 88 if (err) {
91 printk(KERN_ERR "JBD2: 89 printk(KERN_ERR "JBD2: bad block at offset %u\n",
92 next); 90 next);
93 goto failed; 91 goto failed;
94 } 92 }
95 93
96 bh = __getblk(journal->j_dev, 94 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
97 if (!bh) { 95 if (!bh) {
98 err = -ENOMEM; 96 err = -ENOMEM;
99 goto failed; 97 goto failed;
100 } 98 }
101 99
102 if (!buffer_uptodate(bh) && !b 100 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
103 bufs[nbufs++] = bh; 101 bufs[nbufs++] = bh;
104 if (nbufs == MAXBUF) { 102 if (nbufs == MAXBUF) {
105 bh_readahead_b !! 103 ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
106 journal_brelse 104 journal_brelse_array(bufs, nbufs);
107 nbufs = 0; 105 nbufs = 0;
108 } 106 }
109 } else 107 } else
110 brelse(bh); 108 brelse(bh);
111 } 109 }
112 110
113 if (nbufs) 111 if (nbufs)
114 bh_readahead_batch(nbufs, bufs !! 112 ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
115 err = 0; 113 err = 0;
116 114
117 failed: 115 failed:
118 if (nbufs) 116 if (nbufs)
119 journal_brelse_array(bufs, nbu 117 journal_brelse_array(bufs, nbufs);
120 return err; 118 return err;
121 } 119 }
122 120
123 #endif /* __KERNEL__ */ 121 #endif /* __KERNEL__ */
124 122
125 123
126 /* 124 /*
127 * Read a block from the journal 125 * Read a block from the journal
128 */ 126 */
129 127
130 static int jread(struct buffer_head **bhp, jou 128 static int jread(struct buffer_head **bhp, journal_t *journal,
131 unsigned int offset) 129 unsigned int offset)
132 { 130 {
133 int err; 131 int err;
134 unsigned long long blocknr; 132 unsigned long long blocknr;
135 struct buffer_head *bh; 133 struct buffer_head *bh;
136 134
137 *bhp = NULL; 135 *bhp = NULL;
138 136
139 if (offset >= journal->j_total_len) { 137 if (offset >= journal->j_total_len) {
140 printk(KERN_ERR "JBD2: corrupt 138 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
141 return -EFSCORRUPTED; 139 return -EFSCORRUPTED;
142 } 140 }
143 141
144 err = jbd2_journal_bmap(journal, offse 142 err = jbd2_journal_bmap(journal, offset, &blocknr);
145 143
146 if (err) { 144 if (err) {
147 printk(KERN_ERR "JBD2: bad blo 145 printk(KERN_ERR "JBD2: bad block at offset %u\n",
148 offset); 146 offset);
149 return err; 147 return err;
150 } 148 }
151 149
152 bh = __getblk(journal->j_dev, blocknr, 150 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
153 if (!bh) 151 if (!bh)
154 return -ENOMEM; 152 return -ENOMEM;
155 153
156 if (!buffer_uptodate(bh)) { 154 if (!buffer_uptodate(bh)) {
157 /* !! 155 /* If this is a brand new buffer, start readahead.
158 * If this is a brand new buff !! 156 Otherwise, we assume we are already reading it. */
159 * Otherwise, we assume we are !! 157 if (!buffer_req(bh))
160 */ <<
161 bool need_readahead = !buffer_ <<
162 <<
163 bh_read_nowait(bh, 0); <<
164 if (need_readahead) <<
165 do_readahead(journal, 158 do_readahead(journal, offset);
166 wait_on_buffer(bh); 159 wait_on_buffer(bh);
167 } 160 }
168 161
169 if (!buffer_uptodate(bh)) { 162 if (!buffer_uptodate(bh)) {
170 printk(KERN_ERR "JBD2: Failed 163 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
171 offset); 164 offset);
172 brelse(bh); 165 brelse(bh);
173 return -EIO; 166 return -EIO;
174 } 167 }
175 168
176 *bhp = bh; 169 *bhp = bh;
177 return 0; 170 return 0;
178 } 171 }
179 172
180 static int jbd2_descriptor_block_csum_verify(j 173 static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
181 { 174 {
182 struct jbd2_journal_block_tail *tail; 175 struct jbd2_journal_block_tail *tail;
183 __be32 provided; 176 __be32 provided;
184 __u32 calculated; 177 __u32 calculated;
185 178
186 if (!jbd2_journal_has_csum_v2or3(j)) 179 if (!jbd2_journal_has_csum_v2or3(j))
187 return 1; 180 return 1;
188 181
189 tail = (struct jbd2_journal_block_tail !! 182 tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize -
190 j->j_blocksize - sizeof(struct !! 183 sizeof(struct jbd2_journal_block_tail));
191 provided = tail->t_checksum; 184 provided = tail->t_checksum;
192 tail->t_checksum = 0; 185 tail->t_checksum = 0;
193 calculated = jbd2_chksum(j, j->j_csum_ 186 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
194 tail->t_checksum = provided; 187 tail->t_checksum = provided;
195 188
196 return provided == cpu_to_be32(calcula 189 return provided == cpu_to_be32(calculated);
197 } 190 }
198 191
199 /* 192 /*
200 * Count the number of in-use tags in a journa 193 * Count the number of in-use tags in a journal descriptor block.
201 */ 194 */
202 195
203 static int count_tags(journal_t *journal, stru 196 static int count_tags(journal_t *journal, struct buffer_head *bh)
204 { 197 {
205 char * tagp; 198 char * tagp;
206 journal_block_tag_t tag; !! 199 journal_block_tag_t * tag;
207 int nr = 0, size = 200 int nr = 0, size = journal->j_blocksize;
208 int tag_bytes = jo 201 int tag_bytes = journal_tag_bytes(journal);
209 202
210 if (jbd2_journal_has_csum_v2or3(journa 203 if (jbd2_journal_has_csum_v2or3(journal))
211 size -= sizeof(struct jbd2_jou 204 size -= sizeof(struct jbd2_journal_block_tail);
212 205
213 tagp = &bh->b_data[sizeof(journal_head 206 tagp = &bh->b_data[sizeof(journal_header_t)];
214 207
215 while ((tagp - bh->b_data + tag_bytes) 208 while ((tagp - bh->b_data + tag_bytes) <= size) {
216 memcpy(&tag, tagp, sizeof(tag) !! 209 tag = (journal_block_tag_t *) tagp;
217 210
218 nr++; 211 nr++;
219 tagp += tag_bytes; 212 tagp += tag_bytes;
220 if (!(tag.t_flags & cpu_to_be1 !! 213 if (!(tag->t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
221 tagp += 16; 214 tagp += 16;
222 215
223 if (tag.t_flags & cpu_to_be16( !! 216 if (tag->t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
224 break; 217 break;
225 } 218 }
226 219
227 return nr; 220 return nr;
228 } 221 }
229 222
230 223
231 /* Make sure we wrap around the log correctly! 224 /* Make sure we wrap around the log correctly! */
232 #define wrap(journal, var) 225 #define wrap(journal, var) \
233 do { 226 do { \
234 if (var >= (journal)->j_last) !! 227 unsigned long _wrap_last = \
235 var -= ((journal)->j_last - (j !! 228 jbd2_has_feature_fast_commit(journal) ? \
>> 229 (journal)->j_fc_last : (journal)->j_last; \
>> 230 \
>> 231 if (var >= _wrap_last) \
>> 232 var -= (_wrap_last - (journal)->j_first); \
236 } while (0) 233 } while (0)
237 234
238 static int fc_do_one_pass(journal_t *journal, 235 static int fc_do_one_pass(journal_t *journal,
239 struct recovery_info 236 struct recovery_info *info, enum passtype pass)
240 { 237 {
241 unsigned int expected_commit_id = info 238 unsigned int expected_commit_id = info->end_transaction;
242 unsigned long next_fc_block; 239 unsigned long next_fc_block;
243 struct buffer_head *bh; 240 struct buffer_head *bh;
244 int err = 0; 241 int err = 0;
245 242
246 next_fc_block = journal->j_fc_first; 243 next_fc_block = journal->j_fc_first;
247 if (!journal->j_fc_replay_callback) 244 if (!journal->j_fc_replay_callback)
248 return 0; 245 return 0;
249 246
250 while (next_fc_block <= journal->j_fc_ 247 while (next_fc_block <= journal->j_fc_last) {
251 jbd2_debug(3, "Fast commit rep !! 248 jbd_debug(3, "Fast commit replay: next block %ld\n",
252 next_fc_block); 249 next_fc_block);
253 err = jread(&bh, journal, next 250 err = jread(&bh, journal, next_fc_block);
254 if (err) { 251 if (err) {
255 jbd2_debug(3, "Fast co !! 252 jbd_debug(3, "Fast commit replay: read error\n");
256 break; 253 break;
257 } 254 }
258 255
259 err = journal->j_fc_replay_cal 256 err = journal->j_fc_replay_callback(journal, bh, pass,
260 next_f 257 next_fc_block - journal->j_fc_first,
261 expect 258 expected_commit_id);
262 brelse(bh); <<
263 next_fc_block++; 259 next_fc_block++;
264 if (err < 0 || err == JBD2_FC_ 260 if (err < 0 || err == JBD2_FC_REPLAY_STOP)
265 break; 261 break;
266 err = 0; 262 err = 0;
267 } 263 }
268 264
269 if (err) 265 if (err)
270 jbd2_debug(3, "Fast commit rep !! 266 jbd_debug(3, "Fast commit replay failed, err = %d\n", err);
271 267
272 return err; 268 return err;
273 } 269 }
274 270
275 /** 271 /**
276 * jbd2_journal_recover - recovers a on-disk j 272 * jbd2_journal_recover - recovers a on-disk journal
277 * @journal: the journal to recover 273 * @journal: the journal to recover
278 * 274 *
279 * The primary function for recovering the log 275 * The primary function for recovering the log contents when mounting a
280 * journaled device. 276 * journaled device.
281 * 277 *
282 * Recovery is done in three passes. In the f 278 * Recovery is done in three passes. In the first pass, we look for the
283 * end of the log. In the second, we assemble 279 * end of the log. In the second, we assemble the list of revoke
284 * blocks. In the third and final pass, we re 280 * blocks. In the third and final pass, we replay any un-revoked blocks
285 * in the log. 281 * in the log.
286 */ 282 */
287 int jbd2_journal_recover(journal_t *journal) 283 int jbd2_journal_recover(journal_t *journal)
288 { 284 {
289 int err, err2; 285 int err, err2;
290 journal_superblock_t * sb; 286 journal_superblock_t * sb;
291 287
292 struct recovery_info info; 288 struct recovery_info info;
293 289
294 memset(&info, 0, sizeof(info)); 290 memset(&info, 0, sizeof(info));
295 sb = journal->j_superblock; 291 sb = journal->j_superblock;
296 292
297 /* 293 /*
298 * The journal superblock's s_start fi 294 * The journal superblock's s_start field (the current log head)
299 * is always zero if, and only if, the 295 * is always zero if, and only if, the journal was cleanly
300 * unmounted. 296 * unmounted.
301 */ 297 */
>> 298
302 if (!sb->s_start) { 299 if (!sb->s_start) {
303 jbd2_debug(1, "No recovery req !! 300 jbd_debug(1, "No recovery required, last transaction %d\n",
304 be32_to_cpu(sb->s_se !! 301 be32_to_cpu(sb->s_sequence));
305 journal->j_transaction_sequenc 302 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
306 journal->j_head = be32_to_cpu( <<
307 return 0; 303 return 0;
308 } 304 }
309 305
310 err = do_one_pass(journal, &info, PASS 306 err = do_one_pass(journal, &info, PASS_SCAN);
311 if (!err) 307 if (!err)
312 err = do_one_pass(journal, &in 308 err = do_one_pass(journal, &info, PASS_REVOKE);
313 if (!err) 309 if (!err)
314 err = do_one_pass(journal, &in 310 err = do_one_pass(journal, &info, PASS_REPLAY);
315 311
316 jbd2_debug(1, "JBD2: recovery, exit st !! 312 jbd_debug(1, "JBD2: recovery, exit status %d, "
317 "recovered transactions %u t 313 "recovered transactions %u to %u\n",
318 err, info.start_transaction, 314 err, info.start_transaction, info.end_transaction);
319 jbd2_debug(1, "JBD2: Replayed %d and r !! 315 jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
320 info.nr_replays, info.nr_rev 316 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
321 317
322 /* Restart the log at the next transac 318 /* Restart the log at the next transaction ID, thus invalidating
323 * any existing commit records in the 319 * any existing commit records in the log. */
324 journal->j_transaction_sequence = ++in 320 journal->j_transaction_sequence = ++info.end_transaction;
325 journal->j_head = info.head_block; <<
326 jbd2_debug(1, "JBD2: last transaction <<
327 journal->j_transaction_seque <<
328 321
329 jbd2_journal_clear_revoke(journal); 322 jbd2_journal_clear_revoke(journal);
330 err2 = sync_blockdev(journal->j_fs_dev 323 err2 = sync_blockdev(journal->j_fs_dev);
331 if (!err) 324 if (!err)
332 err = err2; 325 err = err2;
333 err2 = jbd2_check_fs_dev_write_error(j <<
334 if (!err) <<
335 err = err2; <<
336 /* Make sure all replayed data is on p 326 /* Make sure all replayed data is on permanent storage */
337 if (journal->j_flags & JBD2_BARRIER) { 327 if (journal->j_flags & JBD2_BARRIER) {
338 err2 = blkdev_issue_flush(jour 328 err2 = blkdev_issue_flush(journal->j_fs_dev);
339 if (!err) 329 if (!err)
340 err = err2; 330 err = err2;
341 } 331 }
342 return err; 332 return err;
343 } 333 }
344 334
345 /** 335 /**
346 * jbd2_journal_skip_recovery - Start journal 336 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
347 * @journal: journal to startup 337 * @journal: journal to startup
348 * 338 *
349 * Locate any valid recovery information from 339 * Locate any valid recovery information from the journal and set up the
350 * journal structures in memory to ignore it ( 340 * journal structures in memory to ignore it (presumably because the
351 * caller has evidence that it is out of date) 341 * caller has evidence that it is out of date).
352 * This function doesn't appear to be exported 342 * This function doesn't appear to be exported..
353 * 343 *
354 * We perform one pass over the journal to all 344 * We perform one pass over the journal to allow us to tell the user how
355 * much recovery information is being erased, 345 * much recovery information is being erased, and to let us initialise
356 * the journal transaction sequence numbers to 346 * the journal transaction sequence numbers to the next unused ID.
357 */ 347 */
358 int jbd2_journal_skip_recovery(journal_t *jour 348 int jbd2_journal_skip_recovery(journal_t *journal)
359 { 349 {
360 int err; 350 int err;
361 351
362 struct recovery_info info; 352 struct recovery_info info;
363 353
364 memset (&info, 0, sizeof(info)); 354 memset (&info, 0, sizeof(info));
365 355
366 err = do_one_pass(journal, &info, PASS 356 err = do_one_pass(journal, &info, PASS_SCAN);
367 357
368 if (err) { 358 if (err) {
369 printk(KERN_ERR "JBD2: error % 359 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
370 ++journal->j_transaction_seque 360 ++journal->j_transaction_sequence;
371 journal->j_head = journal->j_f <<
372 } else { 361 } else {
373 #ifdef CONFIG_JBD2_DEBUG 362 #ifdef CONFIG_JBD2_DEBUG
374 int dropped = info.end_transac 363 int dropped = info.end_transaction -
375 be32_to_cpu(journal->j 364 be32_to_cpu(journal->j_superblock->s_sequence);
376 jbd2_debug(1, !! 365 jbd_debug(1,
377 "JBD2: ignoring %d t 366 "JBD2: ignoring %d transaction%s from the journal.\n",
378 dropped, str_plural( !! 367 dropped, (dropped == 1) ? "" : "s");
379 #endif 368 #endif
380 journal->j_transaction_sequenc 369 journal->j_transaction_sequence = ++info.end_transaction;
381 journal->j_head = info.head_bl <<
382 } 370 }
383 371
384 journal->j_tail = 0; 372 journal->j_tail = 0;
385 return err; 373 return err;
386 } 374 }
387 375
388 static inline unsigned long long read_tag_bloc 376 static inline unsigned long long read_tag_block(journal_t *journal,
389 377 journal_block_tag_t *tag)
390 { 378 {
391 unsigned long long block = be32_to_cpu 379 unsigned long long block = be32_to_cpu(tag->t_blocknr);
392 if (jbd2_has_feature_64bit(journal)) 380 if (jbd2_has_feature_64bit(journal))
393 block |= (u64)be32_to_cpu(tag- 381 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
394 return block; 382 return block;
395 } 383 }
396 384
397 /* 385 /*
398 * calc_chksums calculates the checksums for t 386 * calc_chksums calculates the checksums for the blocks described in the
399 * descriptor block. 387 * descriptor block.
400 */ 388 */
401 static int calc_chksums(journal_t *journal, st 389 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
402 unsigned long *next_lo 390 unsigned long *next_log_block, __u32 *crc32_sum)
403 { 391 {
404 int i, num_blks, err; 392 int i, num_blks, err;
405 unsigned long io_block; 393 unsigned long io_block;
406 struct buffer_head *obh; 394 struct buffer_head *obh;
407 395
408 num_blks = count_tags(journal, bh); 396 num_blks = count_tags(journal, bh);
409 /* Calculate checksum of the descripto 397 /* Calculate checksum of the descriptor block. */
410 *crc32_sum = crc32_be(*crc32_sum, (voi 398 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
411 399
412 for (i = 0; i < num_blks; i++) { 400 for (i = 0; i < num_blks; i++) {
413 io_block = (*next_log_block)++ 401 io_block = (*next_log_block)++;
414 wrap(journal, *next_log_block) 402 wrap(journal, *next_log_block);
415 err = jread(&obh, journal, io_ 403 err = jread(&obh, journal, io_block);
416 if (err) { 404 if (err) {
417 printk(KERN_ERR "JBD2: 405 printk(KERN_ERR "JBD2: IO error %d recovering block "
418 "%lu in log\n" 406 "%lu in log\n", err, io_block);
419 return 1; 407 return 1;
420 } else { 408 } else {
421 *crc32_sum = crc32_be( 409 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
422 obh->b_si 410 obh->b_size);
423 } 411 }
424 put_bh(obh); 412 put_bh(obh);
425 } 413 }
426 return 0; 414 return 0;
427 } 415 }
428 416
429 static int jbd2_commit_block_csum_verify(journ 417 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
430 { 418 {
431 struct commit_header *h; 419 struct commit_header *h;
432 __be32 provided; 420 __be32 provided;
433 __u32 calculated; 421 __u32 calculated;
434 422
435 if (!jbd2_journal_has_csum_v2or3(j)) 423 if (!jbd2_journal_has_csum_v2or3(j))
436 return 1; 424 return 1;
437 425
438 h = buf; 426 h = buf;
439 provided = h->h_chksum[0]; 427 provided = h->h_chksum[0];
440 h->h_chksum[0] = 0; 428 h->h_chksum[0] = 0;
441 calculated = jbd2_chksum(j, j->j_csum_ 429 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
442 h->h_chksum[0] = provided; 430 h->h_chksum[0] = provided;
443 431
444 return provided == cpu_to_be32(calcula 432 return provided == cpu_to_be32(calculated);
445 } 433 }
446 434
447 static bool jbd2_commit_block_csum_verify_part <<
448 { <<
449 struct commit_header *h; <<
450 __be32 provided; <<
451 __u32 calculated; <<
452 void *tmpbuf; <<
453 <<
454 tmpbuf = kzalloc(j->j_blocksize, GFP_K <<
455 if (!tmpbuf) <<
456 return false; <<
457 <<
458 memcpy(tmpbuf, buf, sizeof(struct comm <<
459 h = tmpbuf; <<
460 provided = h->h_chksum[0]; <<
461 h->h_chksum[0] = 0; <<
462 calculated = jbd2_chksum(j, j->j_csum_ <<
463 kfree(tmpbuf); <<
464 <<
465 return provided == cpu_to_be32(calcula <<
466 } <<
467 <<
468 static int jbd2_block_tag_csum_verify(journal_ 435 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
469 journal_ <<
470 void *bu 436 void *buf, __u32 sequence)
471 { 437 {
>> 438 journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
472 __u32 csum32; 439 __u32 csum32;
473 __be32 seq; 440 __be32 seq;
474 441
475 if (!jbd2_journal_has_csum_v2or3(j)) 442 if (!jbd2_journal_has_csum_v2or3(j))
476 return 1; 443 return 1;
477 444
478 seq = cpu_to_be32(sequence); 445 seq = cpu_to_be32(sequence);
479 csum32 = jbd2_chksum(j, j->j_csum_seed 446 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
480 csum32 = jbd2_chksum(j, csum32, buf, j 447 csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
481 448
482 if (jbd2_has_feature_csum3(j)) 449 if (jbd2_has_feature_csum3(j))
483 return tag3->t_checksum == cpu 450 return tag3->t_checksum == cpu_to_be32(csum32);
484 else 451 else
485 return tag->t_checksum == cpu_ 452 return tag->t_checksum == cpu_to_be16(csum32);
486 } 453 }
487 454
488 static int do_one_pass(journal_t *journal, 455 static int do_one_pass(journal_t *journal,
489 struct recovery_info * 456 struct recovery_info *info, enum passtype pass)
490 { 457 {
491 unsigned int first_commit_I 458 unsigned int first_commit_ID, next_commit_ID;
492 unsigned long next_log_block !! 459 unsigned long next_log_block;
493 int err, success = 460 int err, success = 0;
494 journal_superblock_t * sb; 461 journal_superblock_t * sb;
495 journal_header_t * tmp; 462 journal_header_t * tmp;
496 struct buffer_head * bh; 463 struct buffer_head * bh;
497 unsigned int sequence; 464 unsigned int sequence;
498 int blocktype; 465 int blocktype;
499 int tag_bytes = jo 466 int tag_bytes = journal_tag_bytes(journal);
500 __u32 crc32_sum = ~0 467 __u32 crc32_sum = ~0; /* Transactional Checksums */
501 int descr_csum_siz 468 int descr_csum_size = 0;
502 int block_error = 469 int block_error = 0;
503 bool need_check_com 470 bool need_check_commit_time = false;
504 __u64 last_trans_com 471 __u64 last_trans_commit_time = 0, commit_time;
505 472
506 /* 473 /*
507 * First thing is to establish what we 474 * First thing is to establish what we expect to find in the log
508 * (in terms of transaction IDs), and 475 * (in terms of transaction IDs), and where (in terms of log
509 * block offsets): query the superbloc 476 * block offsets): query the superblock.
510 */ 477 */
511 478
512 sb = journal->j_superblock; 479 sb = journal->j_superblock;
513 next_commit_ID = be32_to_cpu(sb->s_seq 480 next_commit_ID = be32_to_cpu(sb->s_sequence);
514 next_log_block = be32_to_cpu(sb->s_sta 481 next_log_block = be32_to_cpu(sb->s_start);
515 head_block = next_log_block; <<
516 482
517 first_commit_ID = next_commit_ID; 483 first_commit_ID = next_commit_ID;
518 if (pass == PASS_SCAN) 484 if (pass == PASS_SCAN)
519 info->start_transaction = firs 485 info->start_transaction = first_commit_ID;
520 486
521 jbd2_debug(1, "Starting recovery pass !! 487 jbd_debug(1, "Starting recovery pass %d\n", pass);
522 488
523 /* 489 /*
524 * Now we walk through the log, transa 490 * Now we walk through the log, transaction by transaction,
525 * making sure that each transaction h 491 * making sure that each transaction has a commit block in the
526 * expected place. Each complete tran 492 * expected place. Each complete transaction gets replayed back
527 * into the main filesystem. 493 * into the main filesystem.
528 */ 494 */
529 495
530 while (1) { 496 while (1) {
531 int flags; 497 int flags;
532 char * tagp; 498 char * tagp;
533 journal_block_tag_t tag; !! 499 journal_block_tag_t * tag;
534 struct buffer_head * obh; 500 struct buffer_head * obh;
535 struct buffer_head * nbh; 501 struct buffer_head * nbh;
536 502
537 cond_resched(); 503 cond_resched();
538 504
539 /* If we already know where to 505 /* If we already know where to stop the log traversal,
540 * check right now that we hav 506 * check right now that we haven't gone past the end of
541 * the log. */ 507 * the log. */
542 508
543 if (pass != PASS_SCAN) 509 if (pass != PASS_SCAN)
544 if (tid_geq(next_commi 510 if (tid_geq(next_commit_ID, info->end_transaction))
545 break; 511 break;
546 512
547 jbd2_debug(2, "Scanning for se !! 513 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
548 next_commit_ID, next !! 514 next_commit_ID, next_log_block,
>> 515 jbd2_has_feature_fast_commit(journal) ?
>> 516 journal->j_fc_last : journal->j_last);
549 517
550 /* Skip over each chunk of the 518 /* Skip over each chunk of the transaction looking
551 * either the next descriptor 519 * either the next descriptor block or the final commit
552 * record. */ 520 * record. */
553 521
554 jbd2_debug(3, "JBD2: checking !! 522 jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
555 err = jread(&bh, journal, next 523 err = jread(&bh, journal, next_log_block);
556 if (err) 524 if (err)
557 goto failed; 525 goto failed;
558 526
559 next_log_block++; 527 next_log_block++;
560 wrap(journal, next_log_block); 528 wrap(journal, next_log_block);
561 529
562 /* What kind of buffer is it? 530 /* What kind of buffer is it?
563 * 531 *
564 * If it is a descriptor block 532 * If it is a descriptor block, check that it has the
565 * expected sequence number. 533 * expected sequence number. Otherwise, we're all done
566 * here. */ 534 * here. */
567 535
568 tmp = (journal_header_t *)bh-> 536 tmp = (journal_header_t *)bh->b_data;
569 537
570 if (tmp->h_magic != cpu_to_be3 538 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
571 brelse(bh); 539 brelse(bh);
572 break; 540 break;
573 } 541 }
574 542
575 blocktype = be32_to_cpu(tmp->h 543 blocktype = be32_to_cpu(tmp->h_blocktype);
576 sequence = be32_to_cpu(tmp->h_ 544 sequence = be32_to_cpu(tmp->h_sequence);
577 jbd2_debug(3, "Found magic %d, !! 545 jbd_debug(3, "Found magic %d, sequence %d\n",
578 blocktype, sequence) 546 blocktype, sequence);
579 547
580 if (sequence != next_commit_ID 548 if (sequence != next_commit_ID) {
581 brelse(bh); 549 brelse(bh);
582 break; 550 break;
583 } 551 }
584 552
585 /* OK, we have a valid descrip 553 /* OK, we have a valid descriptor block which matches
586 * all of the sequence number 554 * all of the sequence number checks. What are we going
587 * to do with it? That depend 555 * to do with it? That depends on the pass... */
588 556
589 switch(blocktype) { 557 switch(blocktype) {
590 case JBD2_DESCRIPTOR_BLOCK: 558 case JBD2_DESCRIPTOR_BLOCK:
591 /* Verify checksum fir 559 /* Verify checksum first */
592 if (jbd2_journal_has_c 560 if (jbd2_journal_has_csum_v2or3(journal))
593 descr_csum_siz 561 descr_csum_size =
594 sizeof 562 sizeof(struct jbd2_journal_block_tail);
595 if (descr_csum_size > 563 if (descr_csum_size > 0 &&
596 !jbd2_descriptor_b 564 !jbd2_descriptor_block_csum_verify(journal,
597 565 bh->b_data)) {
598 /* 566 /*
599 * PASS_SCAN c 567 * PASS_SCAN can see stale blocks due to lazy
600 * journal ini 568 * journal init. Don't error out on those yet.
601 */ 569 */
602 if (pass != PA 570 if (pass != PASS_SCAN) {
603 pr_err 571 pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
604 572 next_log_block);
605 err = 573 err = -EFSBADCRC;
606 brelse 574 brelse(bh);
607 goto f 575 goto failed;
608 } 576 }
609 need_check_com 577 need_check_commit_time = true;
610 jbd2_debug(1, !! 578 jbd_debug(1,
611 "inval 579 "invalid descriptor block found in %lu\n",
612 next_l 580 next_log_block);
613 } 581 }
614 582
615 /* If it is a valid de 583 /* If it is a valid descriptor block, replay it
616 * in pass REPLAY; if 584 * in pass REPLAY; if journal_checksums enabled, then
617 * calculate checksums 585 * calculate checksums in PASS_SCAN, otherwise,
618 * just skip over the 586 * just skip over the blocks it describes. */
619 if (pass != PASS_REPLA 587 if (pass != PASS_REPLAY) {
620 if (pass == PA 588 if (pass == PASS_SCAN &&
621 jbd2_has_f 589 jbd2_has_feature_checksum(journal) &&
622 !need_chec 590 !need_check_commit_time &&
623 !info->end 591 !info->end_transaction) {
624 if (ca 592 if (calc_chksums(journal, bh,
625 593 &next_log_block,
626 594 &crc32_sum)) {
627 595 put_bh(bh);
628 596 break;
629 } 597 }
630 put_bh 598 put_bh(bh);
631 contin 599 continue;
632 } 600 }
633 next_log_block 601 next_log_block += count_tags(journal, bh);
634 wrap(journal, 602 wrap(journal, next_log_block);
635 put_bh(bh); 603 put_bh(bh);
636 continue; 604 continue;
637 } 605 }
638 606
639 /* A descriptor block: 607 /* A descriptor block: we can now write all of
640 * the data blocks. Y 608 * the data blocks. Yay, useful work is finally
641 * getting done here! 609 * getting done here! */
642 610
643 tagp = &bh->b_data[siz 611 tagp = &bh->b_data[sizeof(journal_header_t)];
644 while ((tagp - bh->b_d 612 while ((tagp - bh->b_data + tag_bytes)
645 <= journal->j_b 613 <= journal->j_blocksize - descr_csum_size) {
646 unsigned long 614 unsigned long io_block;
647 615
648 memcpy(&tag, t !! 616 tag = (journal_block_tag_t *) tagp;
649 flags = be16_t !! 617 flags = be16_to_cpu(tag->t_flags);
650 618
651 io_block = nex 619 io_block = next_log_block++;
652 wrap(journal, 620 wrap(journal, next_log_block);
653 err = jread(&o 621 err = jread(&obh, journal, io_block);
654 if (err) { 622 if (err) {
655 /* Rec 623 /* Recover what we can, but
656 * rep 624 * report failure at the end. */
657 succes 625 success = err;
658 printk 626 printk(KERN_ERR
659 627 "JBD2: IO error %d recovering "
660 !! 628 "block %ld in log\n",
661 629 err, io_block);
662 } else { 630 } else {
663 unsign 631 unsigned long long blocknr;
664 632
665 J_ASSE 633 J_ASSERT(obh != NULL);
666 blockn 634 blocknr = read_tag_block(journal,
667 !! 635 tag);
668 636
669 /* If 637 /* If the block has been
670 * rev 638 * revoked, then we're all done
671 * her 639 * here. */
672 if (jb 640 if (jbd2_journal_test_revoke
673 (j 641 (journal, blocknr,
674 n 642 next_commit_ID)) {
675 643 brelse(obh);
676 644 ++info->nr_revoke_hits;
677 645 goto skip_write;
678 } 646 }
679 647
680 /* Loo 648 /* Look for block corruption */
681 if (!j 649 if (!jbd2_block_tag_csum_verify(
682 journal, &tag, (journa !! 650 journal, tag, obh->b_data,
683 obh->b_data, be32_to_c !! 651 be32_to_cpu(tmp->h_sequence))) {
684 652 brelse(obh);
685 653 success = -EFSBADCRC;
686 654 printk(KERN_ERR "JBD2: Invalid "
687 655 "checksum recovering "
688 656 "data block %llu in "
689 !! 657 "log\n", blocknr);
690 <<
691 658 block_error = 1;
692 659 goto skip_write;
693 } 660 }
694 661
695 /* Fin 662 /* Find a buffer for the new
696 * dat 663 * data being restored */
697 nbh = 664 nbh = __getblk(journal->j_fs_dev,
698 665 blocknr,
699 666 journal->j_blocksize);
700 if (nb 667 if (nbh == NULL) {
701 668 printk(KERN_ERR
702 669 "JBD2: Out of memory "
703 670 "during recovery.\n");
704 671 err = -ENOMEM;
705 672 brelse(bh);
706 673 brelse(obh);
707 674 goto failed;
708 } 675 }
709 676
710 lock_b 677 lock_buffer(nbh);
711 memcpy 678 memcpy(nbh->b_data, obh->b_data,
712 679 journal->j_blocksize);
713 if (fl 680 if (flags & JBD2_FLAG_ESCAPE) {
714 681 *((__be32 *)nbh->b_data) =
715 682 cpu_to_be32(JBD2_MAGIC_NUMBER);
716 } 683 }
717 684
718 BUFFER 685 BUFFER_TRACE(nbh, "marking dirty");
719 set_bu 686 set_buffer_uptodate(nbh);
720 mark_b 687 mark_buffer_dirty(nbh);
721 BUFFER 688 BUFFER_TRACE(nbh, "marking uptodate");
722 ++info 689 ++info->nr_replays;
>> 690 /* ll_rw_block(WRITE, 1, &nbh); */
723 unlock 691 unlock_buffer(nbh);
724 brelse 692 brelse(obh);
725 brelse 693 brelse(nbh);
726 } 694 }
727 695
728 skip_write: 696 skip_write:
729 tagp += tag_by 697 tagp += tag_bytes;
730 if (!(flags & 698 if (!(flags & JBD2_FLAG_SAME_UUID))
731 tagp + 699 tagp += 16;
732 700
733 if (flags & JB 701 if (flags & JBD2_FLAG_LAST_TAG)
734 break; 702 break;
735 } 703 }
736 704
737 brelse(bh); 705 brelse(bh);
738 continue; 706 continue;
739 707
740 case JBD2_COMMIT_BLOCK: 708 case JBD2_COMMIT_BLOCK:
741 /* How to differen 709 /* How to differentiate between interrupted commit
742 * and j 710 * and journal corruption ?
743 * 711 *
744 * {nth transaction} 712 * {nth transaction}
745 * Checksum Ver 713 * Checksum Verification Failed
746 * 714 * |
747 * _____ 715 * ____________________
748 * | 716 * | |
749 * async_commit 717 * async_commit sync_commit
750 * | 718 * | |
751 * | GO T 719 * | GO TO NEXT "Journal Corruption"
752 * | TRAN 720 * | TRANSACTION
753 * | 721 * |
754 * {(n+1)th transancti 722 * {(n+1)th transanction}
755 * | 723 * |
756 * _______|_____ 724 * _______|______________
757 * | 725 * | |
758 * Commit block found 726 * Commit block found Commit block not found
759 * | 727 * | |
760 * "Journal Corruption 728 * "Journal Corruption" |
761 * _____ 729 * _____________|_________
762 * | 730 * | |
763 * nth trans corr 731 * nth trans corrupt OR nth trans
764 * and (n+1)th in 732 * and (n+1)th interrupted interrupted
765 * before commit 733 * before commit block
766 * could reach th 734 * could reach the disk.
767 * (Cannot find t 735 * (Cannot find the difference in above
768 * mentioned con 736 * mentioned conditions. Hence assume
769 * "Interrupted 737 * "Interrupted Commit".)
770 */ 738 */
771 commit_time = be64_to_ 739 commit_time = be64_to_cpu(
772 ((struct commi 740 ((struct commit_header *)bh->b_data)->h_commit_sec);
773 /* 741 /*
774 * If need_check_commi 742 * If need_check_commit_time is set, it means we are in
775 * PASS_SCAN and csum 743 * PASS_SCAN and csum verify failed before. If
776 * commit_time is incr 744 * commit_time is increasing, it's the same journal,
777 * otherwise it is sta 745 * otherwise it is stale journal block, just end this
778 * recovery. 746 * recovery.
779 */ 747 */
780 if (need_check_commit_ 748 if (need_check_commit_time) {
781 if (commit_tim 749 if (commit_time >= last_trans_commit_time) {
782 pr_err 750 pr_err("JBD2: Invalid checksum found in transaction %u\n",
783 751 next_commit_ID);
784 err = 752 err = -EFSBADCRC;
785 brelse 753 brelse(bh);
786 goto f 754 goto failed;
787 } 755 }
788 ignore_crc_mismatch: 756 ignore_crc_mismatch:
789 /* 757 /*
790 * It likely d 758 * It likely does not belong to same journal,
791 * just end th 759 * just end this recovery with success.
792 */ 760 */
793 jbd2_debug(1, !! 761 jbd_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
794 next 762 next_commit_ID);
>> 763 err = 0;
795 brelse(bh); 764 brelse(bh);
796 goto done; 765 goto done;
797 } 766 }
798 767
799 /* 768 /*
800 * Found an expected c 769 * Found an expected commit block: if checksums
801 * are present, verify 770 * are present, verify them in PASS_SCAN; else not
802 * much to do other th 771 * much to do other than move on to the next sequence
803 * number. 772 * number.
804 */ 773 */
805 if (pass == PASS_SCAN 774 if (pass == PASS_SCAN &&
806 jbd2_has_feature_c 775 jbd2_has_feature_checksum(journal)) {
807 struct commit_ 776 struct commit_header *cbh =
808 (struc 777 (struct commit_header *)bh->b_data;
809 unsigned found 778 unsigned found_chksum =
810 be32_t 779 be32_to_cpu(cbh->h_chksum[0]);
811 780
812 if (info->end_ 781 if (info->end_transaction) {
813 journa 782 journal->j_failed_commit =
814 783 info->end_transaction;
815 brelse 784 brelse(bh);
816 break; 785 break;
817 } 786 }
818 787
819 /* Neither che 788 /* Neither checksum match nor unused? */
820 if (!((crc32_s 789 if (!((crc32_sum == found_chksum &&
821 cbh->h_ 790 cbh->h_chksum_type ==
822 791 JBD2_CRC32_CHKSUM &&
823 cbh->h_ 792 cbh->h_chksum_size ==
824 793 JBD2_CRC32_CHKSUM_SIZE) ||
825 (cbh->h_ 794 (cbh->h_chksum_type == 0 &&
826 cbh->h_ 795 cbh->h_chksum_size == 0 &&
827 found_c 796 found_chksum == 0)))
828 goto c 797 goto chksum_error;
829 798
830 crc32_sum = ~0 799 crc32_sum = ~0;
831 } 800 }
832 if (pass == PASS_SCAN 801 if (pass == PASS_SCAN &&
833 !jbd2_commit_block 802 !jbd2_commit_block_csum_verify(journal,
834 803 bh->b_data)) {
835 if (jbd2_commi <<
836 <<
837 <<
838 pr_not <<
839 <<
840 goto c <<
841 } <<
842 chksum_error: 804 chksum_error:
843 if (commit_tim 805 if (commit_time < last_trans_commit_time)
844 goto i 806 goto ignore_crc_mismatch;
845 info->end_tran 807 info->end_transaction = next_commit_ID;
846 info->head_blo <<
847 808
848 if (!jbd2_has_ 809 if (!jbd2_has_feature_async_commit(journal)) {
849 journa 810 journal->j_failed_commit =
850 811 next_commit_ID;
851 brelse 812 brelse(bh);
852 break; 813 break;
853 } 814 }
854 } 815 }
855 if (pass == PASS_SCAN) !! 816 if (pass == PASS_SCAN)
856 chksum_ok: <<
857 last_trans_com 817 last_trans_commit_time = commit_time;
858 head_block = n <<
859 } <<
860 brelse(bh); 818 brelse(bh);
861 next_commit_ID++; 819 next_commit_ID++;
862 continue; 820 continue;
863 821
864 case JBD2_REVOKE_BLOCK: 822 case JBD2_REVOKE_BLOCK:
865 /* 823 /*
866 * Check revoke block 824 * Check revoke block crc in pass_scan, if csum verify
867 * failed, check commi 825 * failed, check commit block time later.
868 */ 826 */
869 if (pass == PASS_SCAN 827 if (pass == PASS_SCAN &&
870 !jbd2_descriptor_b 828 !jbd2_descriptor_block_csum_verify(journal,
871 829 bh->b_data)) {
872 jbd2_debug(1, !! 830 jbd_debug(1, "JBD2: invalid revoke block found in %lu\n",
873 next 831 next_log_block);
874 need_check_com 832 need_check_commit_time = true;
875 } 833 }
876 <<
877 /* If we aren't in the 834 /* If we aren't in the REVOKE pass, then we can
878 * just skip over this 835 * just skip over this block. */
879 if (pass != PASS_REVOK 836 if (pass != PASS_REVOKE) {
880 brelse(bh); 837 brelse(bh);
881 continue; 838 continue;
882 } 839 }
883 840
884 err = scan_revoke_reco 841 err = scan_revoke_records(journal, bh,
885 842 next_commit_ID, info);
886 brelse(bh); 843 brelse(bh);
887 if (err) 844 if (err)
888 goto failed; 845 goto failed;
889 continue; 846 continue;
890 847
891 default: 848 default:
892 jbd2_debug(3, "Unrecog !! 849 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
893 blocktype); 850 blocktype);
894 brelse(bh); 851 brelse(bh);
895 goto done; 852 goto done;
896 } 853 }
897 } 854 }
898 855
899 done: 856 done:
900 /* 857 /*
901 * We broke out of the log scan loop: 858 * We broke out of the log scan loop: either we came to the
902 * known end of the log or we found an 859 * known end of the log or we found an unexpected block in the
903 * log. If the latter happened, then 860 * log. If the latter happened, then we know that the "current"
904 * transaction marks the end of the va 861 * transaction marks the end of the valid log.
905 */ 862 */
906 863
907 if (pass == PASS_SCAN) { 864 if (pass == PASS_SCAN) {
908 if (!info->end_transaction) 865 if (!info->end_transaction)
909 info->end_transaction 866 info->end_transaction = next_commit_ID;
910 if (!info->head_block) <<
911 info->head_block = hea <<
912 } else { 867 } else {
913 /* It's really bad news if dif 868 /* It's really bad news if different passes end up at
914 * different places (but possi 869 * different places (but possible due to IO errors). */
915 if (info->end_transaction != n 870 if (info->end_transaction != next_commit_ID) {
916 printk(KERN_ERR "JBD2: 871 printk(KERN_ERR "JBD2: recovery pass %d ended at "
917 "transaction % 872 "transaction %u, expected %u\n",
918 pass, next_com 873 pass, next_commit_ID, info->end_transaction);
919 if (!success) 874 if (!success)
920 success = -EIO 875 success = -EIO;
921 } 876 }
922 } 877 }
923 878
924 if (jbd2_has_feature_fast_commit(journ 879 if (jbd2_has_feature_fast_commit(journal) && pass != PASS_REVOKE) {
925 err = fc_do_one_pass(journal, 880 err = fc_do_one_pass(journal, info, pass);
926 if (err) 881 if (err)
927 success = err; 882 success = err;
928 } 883 }
929 884
930 if (block_error && success == 0) 885 if (block_error && success == 0)
931 success = -EIO; 886 success = -EIO;
932 return success; 887 return success;
933 888
934 failed: 889 failed:
935 return err; 890 return err;
936 } 891 }
937 892
938 /* Scan a revoke record, marking all blocks me 893 /* Scan a revoke record, marking all blocks mentioned as revoked. */
939 894
940 static int scan_revoke_records(journal_t *jour 895 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
941 tid_t sequence, 896 tid_t sequence, struct recovery_info *info)
942 { 897 {
943 jbd2_journal_revoke_header_t *header; 898 jbd2_journal_revoke_header_t *header;
944 int offset, max; 899 int offset, max;
945 unsigned csum_size = 0; !! 900 int csum_size = 0;
946 __u32 rcount; 901 __u32 rcount;
947 int record_len = 4; 902 int record_len = 4;
948 903
949 header = (jbd2_journal_revoke_header_t 904 header = (jbd2_journal_revoke_header_t *) bh->b_data;
950 offset = sizeof(jbd2_journal_revoke_he 905 offset = sizeof(jbd2_journal_revoke_header_t);
951 rcount = be32_to_cpu(header->r_count); 906 rcount = be32_to_cpu(header->r_count);
952 907
953 if (jbd2_journal_has_csum_v2or3(journa 908 if (jbd2_journal_has_csum_v2or3(journal))
954 csum_size = sizeof(struct jbd2 909 csum_size = sizeof(struct jbd2_journal_block_tail);
955 if (rcount > journal->j_blocksize - cs 910 if (rcount > journal->j_blocksize - csum_size)
956 return -EINVAL; 911 return -EINVAL;
957 max = rcount; 912 max = rcount;
958 913
959 if (jbd2_has_feature_64bit(journal)) 914 if (jbd2_has_feature_64bit(journal))
960 record_len = 8; 915 record_len = 8;
961 916
962 while (offset + record_len <= max) { 917 while (offset + record_len <= max) {
963 unsigned long long blocknr; 918 unsigned long long blocknr;
964 int err; 919 int err;
965 920
966 if (record_len == 4) 921 if (record_len == 4)
967 blocknr = be32_to_cpu( 922 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
968 else 923 else
969 blocknr = be64_to_cpu( 924 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
970 offset += record_len; 925 offset += record_len;
971 err = jbd2_journal_set_revoke( 926 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
972 if (err) 927 if (err)
973 return err; 928 return err;
974 ++info->nr_revokes; 929 ++info->nr_revokes;
975 } 930 }
976 return 0; 931 return 0;
977 } 932 }
978 933
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.