Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
87c199c2 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
a844f451 NS |
25 | #include "xfs_sb.h" |
26 | #include "xfs_ag.h" | |
1da177e4 LT |
27 | #include "xfs_dir2.h" |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_mount.h" | |
30 | #include "xfs_error.h" | |
31 | #include "xfs_bmap_btree.h" | |
a844f451 NS |
32 | #include "xfs_alloc_btree.h" |
33 | #include "xfs_ialloc_btree.h" | |
1da177e4 | 34 | #include "xfs_dir2_sf.h" |
a844f451 | 35 | #include "xfs_attr_sf.h" |
1da177e4 | 36 | #include "xfs_dinode.h" |
1da177e4 | 37 | #include "xfs_inode.h" |
a844f451 | 38 | #include "xfs_inode_item.h" |
a844f451 | 39 | #include "xfs_alloc.h" |
1da177e4 LT |
40 | #include "xfs_ialloc.h" |
41 | #include "xfs_log_priv.h" | |
42 | #include "xfs_buf_item.h" | |
1da177e4 LT |
43 | #include "xfs_log_recover.h" |
44 | #include "xfs_extfree_item.h" | |
45 | #include "xfs_trans_priv.h" | |
1da177e4 LT |
46 | #include "xfs_quota.h" |
47 | #include "xfs_rw.h" | |
43355099 | 48 | #include "xfs_utils.h" |
0b1b213f | 49 | #include "xfs_trace.h" |
1da177e4 LT |
50 | |
51 | STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *); | |
52 | STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t); | |
1da177e4 LT |
53 | #if defined(DEBUG) |
54 | STATIC void xlog_recover_check_summary(xlog_t *); | |
1da177e4 LT |
55 | #else |
56 | #define xlog_recover_check_summary(log) | |
1da177e4 LT |
57 | #endif |
58 | ||
59 | ||
60 | /* | |
61 | * Sector aligned buffer routines for buffer create/read/write/access | |
62 | */ | |
63 | ||
64 | #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \ | |
65 | ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \ | |
66 | ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) ) | |
67 | #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask) | |
68 | ||
5d77c0dc | 69 | STATIC xfs_buf_t * |
1da177e4 LT |
70 | xlog_get_bp( |
71 | xlog_t *log, | |
3228149c | 72 | int nbblks) |
1da177e4 | 73 | { |
3228149c DC |
74 | if (nbblks <= 0 || nbblks > log->l_logBBsize) { |
75 | xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks); | |
76 | XFS_ERROR_REPORT("xlog_get_bp(1)", | |
77 | XFS_ERRLEVEL_HIGH, log->l_mp); | |
78 | return NULL; | |
79 | } | |
1da177e4 LT |
80 | |
81 | if (log->l_sectbb_log) { | |
3228149c DC |
82 | if (nbblks > 1) |
83 | nbblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1); | |
84 | nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks); | |
1da177e4 | 85 | } |
3228149c | 86 | return xfs_buf_get_noaddr(BBTOB(nbblks), log->l_mp->m_logdev_targp); |
1da177e4 LT |
87 | } |
88 | ||
5d77c0dc | 89 | STATIC void |
1da177e4 LT |
90 | xlog_put_bp( |
91 | xfs_buf_t *bp) | |
92 | { | |
93 | xfs_buf_free(bp); | |
94 | } | |
95 | ||
076e6acb CH |
96 | STATIC xfs_caddr_t |
97 | xlog_align( | |
98 | xlog_t *log, | |
99 | xfs_daddr_t blk_no, | |
100 | int nbblks, | |
101 | xfs_buf_t *bp) | |
102 | { | |
103 | xfs_caddr_t ptr; | |
104 | ||
105 | if (!log->l_sectbb_log) | |
106 | return XFS_BUF_PTR(bp); | |
107 | ||
108 | ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask); | |
109 | ASSERT(XFS_BUF_SIZE(bp) >= | |
110 | BBTOB(nbblks + (blk_no & log->l_sectbb_mask))); | |
111 | return ptr; | |
112 | } | |
113 | ||
1da177e4 LT |
114 | |
115 | /* | |
116 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. | |
117 | */ | |
076e6acb CH |
118 | STATIC int |
119 | xlog_bread_noalign( | |
1da177e4 LT |
120 | xlog_t *log, |
121 | xfs_daddr_t blk_no, | |
122 | int nbblks, | |
123 | xfs_buf_t *bp) | |
124 | { | |
125 | int error; | |
126 | ||
3228149c DC |
127 | if (nbblks <= 0 || nbblks > log->l_logBBsize) { |
128 | xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks); | |
129 | XFS_ERROR_REPORT("xlog_bread(1)", | |
130 | XFS_ERRLEVEL_HIGH, log->l_mp); | |
131 | return EFSCORRUPTED; | |
132 | } | |
133 | ||
1da177e4 LT |
134 | if (log->l_sectbb_log) { |
135 | blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no); | |
136 | nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks); | |
137 | } | |
138 | ||
139 | ASSERT(nbblks > 0); | |
140 | ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp)); | |
141 | ASSERT(bp); | |
142 | ||
143 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
144 | XFS_BUF_READ(bp); | |
145 | XFS_BUF_BUSY(bp); | |
146 | XFS_BUF_SET_COUNT(bp, BBTOB(nbblks)); | |
147 | XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp); | |
148 | ||
149 | xfsbdstrat(log->l_mp, bp); | |
d64e31a2 DC |
150 | error = xfs_iowait(bp); |
151 | if (error) | |
1da177e4 LT |
152 | xfs_ioerror_alert("xlog_bread", log->l_mp, |
153 | bp, XFS_BUF_ADDR(bp)); | |
154 | return error; | |
155 | } | |
156 | ||
076e6acb CH |
157 | STATIC int |
158 | xlog_bread( | |
159 | xlog_t *log, | |
160 | xfs_daddr_t blk_no, | |
161 | int nbblks, | |
162 | xfs_buf_t *bp, | |
163 | xfs_caddr_t *offset) | |
164 | { | |
165 | int error; | |
166 | ||
167 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
168 | if (error) | |
169 | return error; | |
170 | ||
171 | *offset = xlog_align(log, blk_no, nbblks, bp); | |
172 | return 0; | |
173 | } | |
174 | ||
1da177e4 LT |
175 | /* |
176 | * Write out the buffer at the given block for the given number of blocks. | |
177 | * The buffer is kept locked across the write and is returned locked. | |
178 | * This can only be used for synchronous log writes. | |
179 | */ | |
ba0f32d4 | 180 | STATIC int |
1da177e4 LT |
181 | xlog_bwrite( |
182 | xlog_t *log, | |
183 | xfs_daddr_t blk_no, | |
184 | int nbblks, | |
185 | xfs_buf_t *bp) | |
186 | { | |
187 | int error; | |
188 | ||
3228149c DC |
189 | if (nbblks <= 0 || nbblks > log->l_logBBsize) { |
190 | xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks); | |
191 | XFS_ERROR_REPORT("xlog_bwrite(1)", | |
192 | XFS_ERRLEVEL_HIGH, log->l_mp); | |
193 | return EFSCORRUPTED; | |
194 | } | |
195 | ||
1da177e4 LT |
196 | if (log->l_sectbb_log) { |
197 | blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no); | |
198 | nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks); | |
199 | } | |
200 | ||
201 | ASSERT(nbblks > 0); | |
202 | ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp)); | |
203 | ||
204 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
205 | XFS_BUF_ZEROFLAGS(bp); | |
206 | XFS_BUF_BUSY(bp); | |
207 | XFS_BUF_HOLD(bp); | |
208 | XFS_BUF_PSEMA(bp, PRIBIO); | |
209 | XFS_BUF_SET_COUNT(bp, BBTOB(nbblks)); | |
210 | XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp); | |
211 | ||
212 | if ((error = xfs_bwrite(log->l_mp, bp))) | |
213 | xfs_ioerror_alert("xlog_bwrite", log->l_mp, | |
214 | bp, XFS_BUF_ADDR(bp)); | |
215 | return error; | |
216 | } | |
217 | ||
1da177e4 LT |
218 | #ifdef DEBUG |
219 | /* | |
220 | * dump debug superblock and log record information | |
221 | */ | |
222 | STATIC void | |
223 | xlog_header_check_dump( | |
224 | xfs_mount_t *mp, | |
225 | xlog_rec_header_t *head) | |
226 | { | |
03daa57c JP |
227 | cmn_err(CE_DEBUG, "%s: SB : uuid = %pU, fmt = %d\n", |
228 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); | |
229 | cmn_err(CE_DEBUG, " log : uuid = %pU, fmt = %d\n", | |
230 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); | |
1da177e4 LT |
231 | } |
232 | #else | |
233 | #define xlog_header_check_dump(mp, head) | |
234 | #endif | |
235 | ||
236 | /* | |
237 | * check log record header for recovery | |
238 | */ | |
239 | STATIC int | |
240 | xlog_header_check_recover( | |
241 | xfs_mount_t *mp, | |
242 | xlog_rec_header_t *head) | |
243 | { | |
b53e675d | 244 | ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM); |
1da177e4 LT |
245 | |
246 | /* | |
247 | * IRIX doesn't write the h_fmt field and leaves it zeroed | |
248 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover | |
249 | * a dirty log created in IRIX. | |
250 | */ | |
b53e675d | 251 | if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) { |
1da177e4 LT |
252 | xlog_warn( |
253 | "XFS: dirty log written in incompatible format - can't recover"); | |
254 | xlog_header_check_dump(mp, head); | |
255 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", | |
256 | XFS_ERRLEVEL_HIGH, mp); | |
257 | return XFS_ERROR(EFSCORRUPTED); | |
258 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
259 | xlog_warn( | |
260 | "XFS: dirty log entry has mismatched uuid - can't recover"); | |
261 | xlog_header_check_dump(mp, head); | |
262 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", | |
263 | XFS_ERRLEVEL_HIGH, mp); | |
264 | return XFS_ERROR(EFSCORRUPTED); | |
265 | } | |
266 | return 0; | |
267 | } | |
268 | ||
269 | /* | |
270 | * read the head block of the log and check the header | |
271 | */ | |
272 | STATIC int | |
273 | xlog_header_check_mount( | |
274 | xfs_mount_t *mp, | |
275 | xlog_rec_header_t *head) | |
276 | { | |
b53e675d | 277 | ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM); |
1da177e4 LT |
278 | |
279 | if (uuid_is_nil(&head->h_fs_uuid)) { | |
280 | /* | |
281 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If | |
282 | * h_fs_uuid is nil, we assume this log was last mounted | |
283 | * by IRIX and continue. | |
284 | */ | |
285 | xlog_warn("XFS: nil uuid in log - IRIX style log"); | |
286 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
287 | xlog_warn("XFS: log has mismatched uuid - can't recover"); | |
288 | xlog_header_check_dump(mp, head); | |
289 | XFS_ERROR_REPORT("xlog_header_check_mount", | |
290 | XFS_ERRLEVEL_HIGH, mp); | |
291 | return XFS_ERROR(EFSCORRUPTED); | |
292 | } | |
293 | return 0; | |
294 | } | |
295 | ||
296 | STATIC void | |
297 | xlog_recover_iodone( | |
298 | struct xfs_buf *bp) | |
299 | { | |
1da177e4 LT |
300 | if (XFS_BUF_GETERROR(bp)) { |
301 | /* | |
302 | * We're not going to bother about retrying | |
303 | * this during recovery. One strike! | |
304 | */ | |
1da177e4 | 305 | xfs_ioerror_alert("xlog_recover_iodone", |
15ac08a8 CH |
306 | bp->b_mount, bp, XFS_BUF_ADDR(bp)); |
307 | xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR); | |
1da177e4 | 308 | } |
15ac08a8 | 309 | bp->b_mount = NULL; |
1da177e4 LT |
310 | XFS_BUF_CLR_IODONE_FUNC(bp); |
311 | xfs_biodone(bp); | |
312 | } | |
313 | ||
314 | /* | |
315 | * This routine finds (to an approximation) the first block in the physical | |
316 | * log which contains the given cycle. It uses a binary search algorithm. | |
317 | * Note that the algorithm can not be perfect because the disk will not | |
318 | * necessarily be perfect. | |
319 | */ | |
a8272ce0 | 320 | STATIC int |
1da177e4 LT |
321 | xlog_find_cycle_start( |
322 | xlog_t *log, | |
323 | xfs_buf_t *bp, | |
324 | xfs_daddr_t first_blk, | |
325 | xfs_daddr_t *last_blk, | |
326 | uint cycle) | |
327 | { | |
328 | xfs_caddr_t offset; | |
329 | xfs_daddr_t mid_blk; | |
330 | uint mid_cycle; | |
331 | int error; | |
332 | ||
333 | mid_blk = BLK_AVG(first_blk, *last_blk); | |
334 | while (mid_blk != first_blk && mid_blk != *last_blk) { | |
076e6acb CH |
335 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
336 | if (error) | |
1da177e4 | 337 | return error; |
03bea6fe | 338 | mid_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
339 | if (mid_cycle == cycle) { |
340 | *last_blk = mid_blk; | |
341 | /* last_half_cycle == mid_cycle */ | |
342 | } else { | |
343 | first_blk = mid_blk; | |
344 | /* first_half_cycle == mid_cycle */ | |
345 | } | |
346 | mid_blk = BLK_AVG(first_blk, *last_blk); | |
347 | } | |
348 | ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) || | |
349 | (mid_blk == *last_blk && mid_blk-1 == first_blk)); | |
350 | ||
351 | return 0; | |
352 | } | |
353 | ||
354 | /* | |
355 | * Check that the range of blocks does not contain the cycle number | |
356 | * given. The scan needs to occur from front to back and the ptr into the | |
357 | * region must be updated since a later routine will need to perform another | |
358 | * test. If the region is completely good, we end up returning the same | |
359 | * last block number. | |
360 | * | |
361 | * Set blkno to -1 if we encounter no errors. This is an invalid block number | |
362 | * since we don't ever expect logs to get this large. | |
363 | */ | |
364 | STATIC int | |
365 | xlog_find_verify_cycle( | |
366 | xlog_t *log, | |
367 | xfs_daddr_t start_blk, | |
368 | int nbblks, | |
369 | uint stop_on_cycle_no, | |
370 | xfs_daddr_t *new_blk) | |
371 | { | |
372 | xfs_daddr_t i, j; | |
373 | uint cycle; | |
374 | xfs_buf_t *bp; | |
375 | xfs_daddr_t bufblks; | |
376 | xfs_caddr_t buf = NULL; | |
377 | int error = 0; | |
378 | ||
379 | bufblks = 1 << ffs(nbblks); | |
380 | ||
381 | while (!(bp = xlog_get_bp(log, bufblks))) { | |
382 | /* can't get enough memory to do everything in one big buffer */ | |
383 | bufblks >>= 1; | |
384 | if (bufblks <= log->l_sectbb_log) | |
385 | return ENOMEM; | |
386 | } | |
387 | ||
388 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { | |
389 | int bcount; | |
390 | ||
391 | bcount = min(bufblks, (start_blk + nbblks - i)); | |
392 | ||
076e6acb CH |
393 | error = xlog_bread(log, i, bcount, bp, &buf); |
394 | if (error) | |
1da177e4 LT |
395 | goto out; |
396 | ||
1da177e4 | 397 | for (j = 0; j < bcount; j++) { |
03bea6fe | 398 | cycle = xlog_get_cycle(buf); |
1da177e4 LT |
399 | if (cycle == stop_on_cycle_no) { |
400 | *new_blk = i+j; | |
401 | goto out; | |
402 | } | |
403 | ||
404 | buf += BBSIZE; | |
405 | } | |
406 | } | |
407 | ||
408 | *new_blk = -1; | |
409 | ||
410 | out: | |
411 | xlog_put_bp(bp); | |
412 | return error; | |
413 | } | |
414 | ||
415 | /* | |
416 | * Potentially backup over partial log record write. | |
417 | * | |
418 | * In the typical case, last_blk is the number of the block directly after | |
419 | * a good log record. Therefore, we subtract one to get the block number | |
420 | * of the last block in the given buffer. extra_bblks contains the number | |
421 | * of blocks we would have read on a previous read. This happens when the | |
422 | * last log record is split over the end of the physical log. | |
423 | * | |
424 | * extra_bblks is the number of blocks potentially verified on a previous | |
425 | * call to this routine. | |
426 | */ | |
427 | STATIC int | |
428 | xlog_find_verify_log_record( | |
429 | xlog_t *log, | |
430 | xfs_daddr_t start_blk, | |
431 | xfs_daddr_t *last_blk, | |
432 | int extra_bblks) | |
433 | { | |
434 | xfs_daddr_t i; | |
435 | xfs_buf_t *bp; | |
436 | xfs_caddr_t offset = NULL; | |
437 | xlog_rec_header_t *head = NULL; | |
438 | int error = 0; | |
439 | int smallmem = 0; | |
440 | int num_blks = *last_blk - start_blk; | |
441 | int xhdrs; | |
442 | ||
443 | ASSERT(start_blk != 0 || *last_blk != start_blk); | |
444 | ||
445 | if (!(bp = xlog_get_bp(log, num_blks))) { | |
446 | if (!(bp = xlog_get_bp(log, 1))) | |
447 | return ENOMEM; | |
448 | smallmem = 1; | |
449 | } else { | |
076e6acb CH |
450 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
451 | if (error) | |
1da177e4 | 452 | goto out; |
1da177e4 LT |
453 | offset += ((num_blks - 1) << BBSHIFT); |
454 | } | |
455 | ||
456 | for (i = (*last_blk) - 1; i >= 0; i--) { | |
457 | if (i < start_blk) { | |
458 | /* valid log record not found */ | |
459 | xlog_warn( | |
460 | "XFS: Log inconsistent (didn't find previous header)"); | |
461 | ASSERT(0); | |
462 | error = XFS_ERROR(EIO); | |
463 | goto out; | |
464 | } | |
465 | ||
466 | if (smallmem) { | |
076e6acb CH |
467 | error = xlog_bread(log, i, 1, bp, &offset); |
468 | if (error) | |
1da177e4 | 469 | goto out; |
1da177e4 LT |
470 | } |
471 | ||
472 | head = (xlog_rec_header_t *)offset; | |
473 | ||
b53e675d | 474 | if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno)) |
1da177e4 LT |
475 | break; |
476 | ||
477 | if (!smallmem) | |
478 | offset -= BBSIZE; | |
479 | } | |
480 | ||
481 | /* | |
482 | * We hit the beginning of the physical log & still no header. Return | |
483 | * to caller. If caller can handle a return of -1, then this routine | |
484 | * will be called again for the end of the physical log. | |
485 | */ | |
486 | if (i == -1) { | |
487 | error = -1; | |
488 | goto out; | |
489 | } | |
490 | ||
491 | /* | |
492 | * We have the final block of the good log (the first block | |
493 | * of the log record _before_ the head. So we check the uuid. | |
494 | */ | |
495 | if ((error = xlog_header_check_mount(log->l_mp, head))) | |
496 | goto out; | |
497 | ||
498 | /* | |
499 | * We may have found a log record header before we expected one. | |
500 | * last_blk will be the 1st block # with a given cycle #. We may end | |
501 | * up reading an entire log record. In this case, we don't want to | |
502 | * reset last_blk. Only when last_blk points in the middle of a log | |
503 | * record do we update last_blk. | |
504 | */ | |
62118709 | 505 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d | 506 | uint h_size = be32_to_cpu(head->h_size); |
1da177e4 LT |
507 | |
508 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; | |
509 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
510 | xhdrs++; | |
511 | } else { | |
512 | xhdrs = 1; | |
513 | } | |
514 | ||
b53e675d CH |
515 | if (*last_blk - i + extra_bblks != |
516 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) | |
1da177e4 LT |
517 | *last_blk = i; |
518 | ||
519 | out: | |
520 | xlog_put_bp(bp); | |
521 | return error; | |
522 | } | |
523 | ||
524 | /* | |
525 | * Head is defined to be the point of the log where the next log write | |
526 | * write could go. This means that incomplete LR writes at the end are | |
527 | * eliminated when calculating the head. We aren't guaranteed that previous | |
528 | * LR have complete transactions. We only know that a cycle number of | |
529 | * current cycle number -1 won't be present in the log if we start writing | |
530 | * from our current block number. | |
531 | * | |
532 | * last_blk contains the block number of the first block with a given | |
533 | * cycle number. | |
534 | * | |
535 | * Return: zero if normal, non-zero if error. | |
536 | */ | |
ba0f32d4 | 537 | STATIC int |
1da177e4 LT |
538 | xlog_find_head( |
539 | xlog_t *log, | |
540 | xfs_daddr_t *return_head_blk) | |
541 | { | |
542 | xfs_buf_t *bp; | |
543 | xfs_caddr_t offset; | |
544 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; | |
545 | int num_scan_bblks; | |
546 | uint first_half_cycle, last_half_cycle; | |
547 | uint stop_on_cycle; | |
548 | int error, log_bbnum = log->l_logBBsize; | |
549 | ||
550 | /* Is the end of the log device zeroed? */ | |
551 | if ((error = xlog_find_zeroed(log, &first_blk)) == -1) { | |
552 | *return_head_blk = first_blk; | |
553 | ||
554 | /* Is the whole lot zeroed? */ | |
555 | if (!first_blk) { | |
556 | /* Linux XFS shouldn't generate totally zeroed logs - | |
557 | * mkfs etc write a dummy unmount record to a fresh | |
558 | * log so we can store the uuid in there | |
559 | */ | |
560 | xlog_warn("XFS: totally zeroed log"); | |
561 | } | |
562 | ||
563 | return 0; | |
564 | } else if (error) { | |
565 | xlog_warn("XFS: empty log check failed"); | |
566 | return error; | |
567 | } | |
568 | ||
569 | first_blk = 0; /* get cycle # of 1st block */ | |
570 | bp = xlog_get_bp(log, 1); | |
571 | if (!bp) | |
572 | return ENOMEM; | |
076e6acb CH |
573 | |
574 | error = xlog_bread(log, 0, 1, bp, &offset); | |
575 | if (error) | |
1da177e4 | 576 | goto bp_err; |
076e6acb | 577 | |
03bea6fe | 578 | first_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
579 | |
580 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ | |
076e6acb CH |
581 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
582 | if (error) | |
1da177e4 | 583 | goto bp_err; |
076e6acb | 584 | |
03bea6fe | 585 | last_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
586 | ASSERT(last_half_cycle != 0); |
587 | ||
588 | /* | |
589 | * If the 1st half cycle number is equal to the last half cycle number, | |
590 | * then the entire log is stamped with the same cycle number. In this | |
591 | * case, head_blk can't be set to zero (which makes sense). The below | |
592 | * math doesn't work out properly with head_blk equal to zero. Instead, | |
593 | * we set it to log_bbnum which is an invalid block number, but this | |
594 | * value makes the math correct. If head_blk doesn't changed through | |
595 | * all the tests below, *head_blk is set to zero at the very end rather | |
596 | * than log_bbnum. In a sense, log_bbnum and zero are the same block | |
597 | * in a circular file. | |
598 | */ | |
599 | if (first_half_cycle == last_half_cycle) { | |
600 | /* | |
601 | * In this case we believe that the entire log should have | |
602 | * cycle number last_half_cycle. We need to scan backwards | |
603 | * from the end verifying that there are no holes still | |
604 | * containing last_half_cycle - 1. If we find such a hole, | |
605 | * then the start of that hole will be the new head. The | |
606 | * simple case looks like | |
607 | * x | x ... | x - 1 | x | |
608 | * Another case that fits this picture would be | |
609 | * x | x + 1 | x ... | x | |
c41564b5 | 610 | * In this case the head really is somewhere at the end of the |
1da177e4 LT |
611 | * log, as one of the latest writes at the beginning was |
612 | * incomplete. | |
613 | * One more case is | |
614 | * x | x + 1 | x ... | x - 1 | x | |
615 | * This is really the combination of the above two cases, and | |
616 | * the head has to end up at the start of the x-1 hole at the | |
617 | * end of the log. | |
618 | * | |
619 | * In the 256k log case, we will read from the beginning to the | |
620 | * end of the log and search for cycle numbers equal to x-1. | |
621 | * We don't worry about the x+1 blocks that we encounter, | |
622 | * because we know that they cannot be the head since the log | |
623 | * started with x. | |
624 | */ | |
625 | head_blk = log_bbnum; | |
626 | stop_on_cycle = last_half_cycle - 1; | |
627 | } else { | |
628 | /* | |
629 | * In this case we want to find the first block with cycle | |
630 | * number matching last_half_cycle. We expect the log to be | |
631 | * some variation on | |
632 | * x + 1 ... | x ... | |
633 | * The first block with cycle number x (last_half_cycle) will | |
634 | * be where the new head belongs. First we do a binary search | |
635 | * for the first occurrence of last_half_cycle. The binary | |
636 | * search may not be totally accurate, so then we scan back | |
637 | * from there looking for occurrences of last_half_cycle before | |
638 | * us. If that backwards scan wraps around the beginning of | |
639 | * the log, then we look for occurrences of last_half_cycle - 1 | |
640 | * at the end of the log. The cases we're looking for look | |
641 | * like | |
642 | * x + 1 ... | x | x + 1 | x ... | |
643 | * ^ binary search stopped here | |
644 | * or | |
645 | * x + 1 ... | x ... | x - 1 | x | |
646 | * <---------> less than scan distance | |
647 | */ | |
648 | stop_on_cycle = last_half_cycle; | |
649 | if ((error = xlog_find_cycle_start(log, bp, first_blk, | |
650 | &head_blk, last_half_cycle))) | |
651 | goto bp_err; | |
652 | } | |
653 | ||
654 | /* | |
655 | * Now validate the answer. Scan back some number of maximum possible | |
656 | * blocks and make sure each one has the expected cycle number. The | |
657 | * maximum is determined by the total possible amount of buffering | |
658 | * in the in-core log. The following number can be made tighter if | |
659 | * we actually look at the block size of the filesystem. | |
660 | */ | |
661 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
662 | if (head_blk >= num_scan_bblks) { | |
663 | /* | |
664 | * We are guaranteed that the entire check can be performed | |
665 | * in one buffer. | |
666 | */ | |
667 | start_blk = head_blk - num_scan_bblks; | |
668 | if ((error = xlog_find_verify_cycle(log, | |
669 | start_blk, num_scan_bblks, | |
670 | stop_on_cycle, &new_blk))) | |
671 | goto bp_err; | |
672 | if (new_blk != -1) | |
673 | head_blk = new_blk; | |
674 | } else { /* need to read 2 parts of log */ | |
675 | /* | |
676 | * We are going to scan backwards in the log in two parts. | |
677 | * First we scan the physical end of the log. In this part | |
678 | * of the log, we are looking for blocks with cycle number | |
679 | * last_half_cycle - 1. | |
680 | * If we find one, then we know that the log starts there, as | |
681 | * we've found a hole that didn't get written in going around | |
682 | * the end of the physical log. The simple case for this is | |
683 | * x + 1 ... | x ... | x - 1 | x | |
684 | * <---------> less than scan distance | |
685 | * If all of the blocks at the end of the log have cycle number | |
686 | * last_half_cycle, then we check the blocks at the start of | |
687 | * the log looking for occurrences of last_half_cycle. If we | |
688 | * find one, then our current estimate for the location of the | |
689 | * first occurrence of last_half_cycle is wrong and we move | |
690 | * back to the hole we've found. This case looks like | |
691 | * x + 1 ... | x | x + 1 | x ... | |
692 | * ^ binary search stopped here | |
693 | * Another case we need to handle that only occurs in 256k | |
694 | * logs is | |
695 | * x + 1 ... | x ... | x+1 | x ... | |
696 | * ^ binary search stops here | |
697 | * In a 256k log, the scan at the end of the log will see the | |
698 | * x + 1 blocks. We need to skip past those since that is | |
699 | * certainly not the head of the log. By searching for | |
700 | * last_half_cycle-1 we accomplish that. | |
701 | */ | |
702 | start_blk = log_bbnum - num_scan_bblks + head_blk; | |
703 | ASSERT(head_blk <= INT_MAX && | |
704 | (xfs_daddr_t) num_scan_bblks - head_blk >= 0); | |
705 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
706 | num_scan_bblks - (int)head_blk, | |
707 | (stop_on_cycle - 1), &new_blk))) | |
708 | goto bp_err; | |
709 | if (new_blk != -1) { | |
710 | head_blk = new_blk; | |
711 | goto bad_blk; | |
712 | } | |
713 | ||
714 | /* | |
715 | * Scan beginning of log now. The last part of the physical | |
716 | * log is good. This scan needs to verify that it doesn't find | |
717 | * the last_half_cycle. | |
718 | */ | |
719 | start_blk = 0; | |
720 | ASSERT(head_blk <= INT_MAX); | |
721 | if ((error = xlog_find_verify_cycle(log, | |
722 | start_blk, (int)head_blk, | |
723 | stop_on_cycle, &new_blk))) | |
724 | goto bp_err; | |
725 | if (new_blk != -1) | |
726 | head_blk = new_blk; | |
727 | } | |
728 | ||
729 | bad_blk: | |
730 | /* | |
731 | * Now we need to make sure head_blk is not pointing to a block in | |
732 | * the middle of a log record. | |
733 | */ | |
734 | num_scan_bblks = XLOG_REC_SHIFT(log); | |
735 | if (head_blk >= num_scan_bblks) { | |
736 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ | |
737 | ||
738 | /* start ptr at last block ptr before head_blk */ | |
739 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
740 | &head_blk, 0)) == -1) { | |
741 | error = XFS_ERROR(EIO); | |
742 | goto bp_err; | |
743 | } else if (error) | |
744 | goto bp_err; | |
745 | } else { | |
746 | start_blk = 0; | |
747 | ASSERT(head_blk <= INT_MAX); | |
748 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
749 | &head_blk, 0)) == -1) { | |
750 | /* We hit the beginning of the log during our search */ | |
751 | start_blk = log_bbnum - num_scan_bblks + head_blk; | |
752 | new_blk = log_bbnum; | |
753 | ASSERT(start_blk <= INT_MAX && | |
754 | (xfs_daddr_t) log_bbnum-start_blk >= 0); | |
755 | ASSERT(head_blk <= INT_MAX); | |
756 | if ((error = xlog_find_verify_log_record(log, | |
757 | start_blk, &new_blk, | |
758 | (int)head_blk)) == -1) { | |
759 | error = XFS_ERROR(EIO); | |
760 | goto bp_err; | |
761 | } else if (error) | |
762 | goto bp_err; | |
763 | if (new_blk != log_bbnum) | |
764 | head_blk = new_blk; | |
765 | } else if (error) | |
766 | goto bp_err; | |
767 | } | |
768 | ||
769 | xlog_put_bp(bp); | |
770 | if (head_blk == log_bbnum) | |
771 | *return_head_blk = 0; | |
772 | else | |
773 | *return_head_blk = head_blk; | |
774 | /* | |
775 | * When returning here, we have a good block number. Bad block | |
776 | * means that during a previous crash, we didn't have a clean break | |
777 | * from cycle number N to cycle number N-1. In this case, we need | |
778 | * to find the first block with cycle number N-1. | |
779 | */ | |
780 | return 0; | |
781 | ||
782 | bp_err: | |
783 | xlog_put_bp(bp); | |
784 | ||
785 | if (error) | |
786 | xlog_warn("XFS: failed to find log head"); | |
787 | return error; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Find the sync block number or the tail of the log. | |
792 | * | |
793 | * This will be the block number of the last record to have its | |
794 | * associated buffers synced to disk. Every log record header has | |
795 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy | |
796 | * to get a sync block number. The only concern is to figure out which | |
797 | * log record header to believe. | |
798 | * | |
799 | * The following algorithm uses the log record header with the largest | |
800 | * lsn. The entire log record does not need to be valid. We only care | |
801 | * that the header is valid. | |
802 | * | |
803 | * We could speed up search by using current head_blk buffer, but it is not | |
804 | * available. | |
805 | */ | |
5d77c0dc | 806 | STATIC int |
1da177e4 LT |
807 | xlog_find_tail( |
808 | xlog_t *log, | |
809 | xfs_daddr_t *head_blk, | |
65be6054 | 810 | xfs_daddr_t *tail_blk) |
1da177e4 LT |
811 | { |
812 | xlog_rec_header_t *rhead; | |
813 | xlog_op_header_t *op_head; | |
814 | xfs_caddr_t offset = NULL; | |
815 | xfs_buf_t *bp; | |
816 | int error, i, found; | |
817 | xfs_daddr_t umount_data_blk; | |
818 | xfs_daddr_t after_umount_blk; | |
819 | xfs_lsn_t tail_lsn; | |
820 | int hblks; | |
821 | ||
822 | found = 0; | |
823 | ||
824 | /* | |
825 | * Find previous log record | |
826 | */ | |
827 | if ((error = xlog_find_head(log, head_blk))) | |
828 | return error; | |
829 | ||
830 | bp = xlog_get_bp(log, 1); | |
831 | if (!bp) | |
832 | return ENOMEM; | |
833 | if (*head_blk == 0) { /* special case */ | |
076e6acb CH |
834 | error = xlog_bread(log, 0, 1, bp, &offset); |
835 | if (error) | |
1da177e4 | 836 | goto bread_err; |
076e6acb | 837 | |
03bea6fe | 838 | if (xlog_get_cycle(offset) == 0) { |
1da177e4 LT |
839 | *tail_blk = 0; |
840 | /* leave all other log inited values alone */ | |
841 | goto exit; | |
842 | } | |
843 | } | |
844 | ||
845 | /* | |
846 | * Search backwards looking for log record header block | |
847 | */ | |
848 | ASSERT(*head_blk < INT_MAX); | |
849 | for (i = (int)(*head_blk) - 1; i >= 0; i--) { | |
076e6acb CH |
850 | error = xlog_bread(log, i, 1, bp, &offset); |
851 | if (error) | |
1da177e4 | 852 | goto bread_err; |
076e6acb | 853 | |
b53e675d | 854 | if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) { |
1da177e4 LT |
855 | found = 1; |
856 | break; | |
857 | } | |
858 | } | |
859 | /* | |
860 | * If we haven't found the log record header block, start looking | |
861 | * again from the end of the physical log. XXXmiken: There should be | |
862 | * a check here to make sure we didn't search more than N blocks in | |
863 | * the previous code. | |
864 | */ | |
865 | if (!found) { | |
866 | for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) { | |
076e6acb CH |
867 | error = xlog_bread(log, i, 1, bp, &offset); |
868 | if (error) | |
1da177e4 | 869 | goto bread_err; |
076e6acb | 870 | |
1da177e4 | 871 | if (XLOG_HEADER_MAGIC_NUM == |
b53e675d | 872 | be32_to_cpu(*(__be32 *)offset)) { |
1da177e4 LT |
873 | found = 2; |
874 | break; | |
875 | } | |
876 | } | |
877 | } | |
878 | if (!found) { | |
879 | xlog_warn("XFS: xlog_find_tail: couldn't find sync record"); | |
880 | ASSERT(0); | |
881 | return XFS_ERROR(EIO); | |
882 | } | |
883 | ||
884 | /* find blk_no of tail of log */ | |
885 | rhead = (xlog_rec_header_t *)offset; | |
b53e675d | 886 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); |
1da177e4 LT |
887 | |
888 | /* | |
889 | * Reset log values according to the state of the log when we | |
890 | * crashed. In the case where head_blk == 0, we bump curr_cycle | |
891 | * one because the next write starts a new cycle rather than | |
892 | * continuing the cycle of the last good log record. At this | |
893 | * point we have guaranteed that all partial log records have been | |
894 | * accounted for. Therefore, we know that the last good log record | |
895 | * written was complete and ended exactly on the end boundary | |
896 | * of the physical log. | |
897 | */ | |
898 | log->l_prev_block = i; | |
899 | log->l_curr_block = (int)*head_blk; | |
b53e675d | 900 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); |
1da177e4 LT |
901 | if (found == 2) |
902 | log->l_curr_cycle++; | |
b53e675d CH |
903 | log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn); |
904 | log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn); | |
1da177e4 LT |
905 | log->l_grant_reserve_cycle = log->l_curr_cycle; |
906 | log->l_grant_reserve_bytes = BBTOB(log->l_curr_block); | |
907 | log->l_grant_write_cycle = log->l_curr_cycle; | |
908 | log->l_grant_write_bytes = BBTOB(log->l_curr_block); | |
909 | ||
910 | /* | |
911 | * Look for unmount record. If we find it, then we know there | |
912 | * was a clean unmount. Since 'i' could be the last block in | |
913 | * the physical log, we convert to a log block before comparing | |
914 | * to the head_blk. | |
915 | * | |
916 | * Save the current tail lsn to use to pass to | |
917 | * xlog_clear_stale_blocks() below. We won't want to clear the | |
918 | * unmount record if there is one, so we pass the lsn of the | |
919 | * unmount record rather than the block after it. | |
920 | */ | |
62118709 | 921 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d CH |
922 | int h_size = be32_to_cpu(rhead->h_size); |
923 | int h_version = be32_to_cpu(rhead->h_version); | |
1da177e4 LT |
924 | |
925 | if ((h_version & XLOG_VERSION_2) && | |
926 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
927 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
928 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
929 | hblks++; | |
930 | } else { | |
931 | hblks = 1; | |
932 | } | |
933 | } else { | |
934 | hblks = 1; | |
935 | } | |
936 | after_umount_blk = (i + hblks + (int) | |
b53e675d | 937 | BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize; |
1da177e4 LT |
938 | tail_lsn = log->l_tail_lsn; |
939 | if (*head_blk == after_umount_blk && | |
b53e675d | 940 | be32_to_cpu(rhead->h_num_logops) == 1) { |
1da177e4 | 941 | umount_data_blk = (i + hblks) % log->l_logBBsize; |
076e6acb CH |
942 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); |
943 | if (error) | |
1da177e4 | 944 | goto bread_err; |
076e6acb | 945 | |
1da177e4 LT |
946 | op_head = (xlog_op_header_t *)offset; |
947 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { | |
948 | /* | |
949 | * Set tail and last sync so that newly written | |
950 | * log records will point recovery to after the | |
951 | * current unmount record. | |
952 | */ | |
03bea6fe CH |
953 | log->l_tail_lsn = |
954 | xlog_assign_lsn(log->l_curr_cycle, | |
955 | after_umount_blk); | |
956 | log->l_last_sync_lsn = | |
957 | xlog_assign_lsn(log->l_curr_cycle, | |
958 | after_umount_blk); | |
1da177e4 | 959 | *tail_blk = after_umount_blk; |
92821e2b DC |
960 | |
961 | /* | |
962 | * Note that the unmount was clean. If the unmount | |
963 | * was not clean, we need to know this to rebuild the | |
964 | * superblock counters from the perag headers if we | |
965 | * have a filesystem using non-persistent counters. | |
966 | */ | |
967 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; | |
1da177e4 LT |
968 | } |
969 | } | |
970 | ||
971 | /* | |
972 | * Make sure that there are no blocks in front of the head | |
973 | * with the same cycle number as the head. This can happen | |
974 | * because we allow multiple outstanding log writes concurrently, | |
975 | * and the later writes might make it out before earlier ones. | |
976 | * | |
977 | * We use the lsn from before modifying it so that we'll never | |
978 | * overwrite the unmount record after a clean unmount. | |
979 | * | |
980 | * Do this only if we are going to recover the filesystem | |
981 | * | |
982 | * NOTE: This used to say "if (!readonly)" | |
983 | * However on Linux, we can & do recover a read-only filesystem. | |
984 | * We only skip recovery if NORECOVERY is specified on mount, | |
985 | * in which case we would not be here. | |
986 | * | |
987 | * But... if the -device- itself is readonly, just skip this. | |
988 | * We can't recover this device anyway, so it won't matter. | |
989 | */ | |
990 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) { | |
991 | error = xlog_clear_stale_blocks(log, tail_lsn); | |
992 | } | |
993 | ||
994 | bread_err: | |
995 | exit: | |
996 | xlog_put_bp(bp); | |
997 | ||
998 | if (error) | |
999 | xlog_warn("XFS: failed to locate log tail"); | |
1000 | return error; | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Is the log zeroed at all? | |
1005 | * | |
1006 | * The last binary search should be changed to perform an X block read | |
1007 | * once X becomes small enough. You can then search linearly through | |
1008 | * the X blocks. This will cut down on the number of reads we need to do. | |
1009 | * | |
1010 | * If the log is partially zeroed, this routine will pass back the blkno | |
1011 | * of the first block with cycle number 0. It won't have a complete LR | |
1012 | * preceding it. | |
1013 | * | |
1014 | * Return: | |
1015 | * 0 => the log is completely written to | |
1016 | * -1 => use *blk_no as the first block of the log | |
1017 | * >0 => error has occurred | |
1018 | */ | |
a8272ce0 | 1019 | STATIC int |
1da177e4 LT |
1020 | xlog_find_zeroed( |
1021 | xlog_t *log, | |
1022 | xfs_daddr_t *blk_no) | |
1023 | { | |
1024 | xfs_buf_t *bp; | |
1025 | xfs_caddr_t offset; | |
1026 | uint first_cycle, last_cycle; | |
1027 | xfs_daddr_t new_blk, last_blk, start_blk; | |
1028 | xfs_daddr_t num_scan_bblks; | |
1029 | int error, log_bbnum = log->l_logBBsize; | |
1030 | ||
6fdf8ccc NS |
1031 | *blk_no = 0; |
1032 | ||
1da177e4 LT |
1033 | /* check totally zeroed log */ |
1034 | bp = xlog_get_bp(log, 1); | |
1035 | if (!bp) | |
1036 | return ENOMEM; | |
076e6acb CH |
1037 | error = xlog_bread(log, 0, 1, bp, &offset); |
1038 | if (error) | |
1da177e4 | 1039 | goto bp_err; |
076e6acb | 1040 | |
03bea6fe | 1041 | first_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1042 | if (first_cycle == 0) { /* completely zeroed log */ |
1043 | *blk_no = 0; | |
1044 | xlog_put_bp(bp); | |
1045 | return -1; | |
1046 | } | |
1047 | ||
1048 | /* check partially zeroed log */ | |
076e6acb CH |
1049 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
1050 | if (error) | |
1da177e4 | 1051 | goto bp_err; |
076e6acb | 1052 | |
03bea6fe | 1053 | last_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1054 | if (last_cycle != 0) { /* log completely written to */ |
1055 | xlog_put_bp(bp); | |
1056 | return 0; | |
1057 | } else if (first_cycle != 1) { | |
1058 | /* | |
1059 | * If the cycle of the last block is zero, the cycle of | |
1060 | * the first block must be 1. If it's not, maybe we're | |
1061 | * not looking at a log... Bail out. | |
1062 | */ | |
1063 | xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)"); | |
1064 | return XFS_ERROR(EINVAL); | |
1065 | } | |
1066 | ||
1067 | /* we have a partially zeroed log */ | |
1068 | last_blk = log_bbnum-1; | |
1069 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) | |
1070 | goto bp_err; | |
1071 | ||
1072 | /* | |
1073 | * Validate the answer. Because there is no way to guarantee that | |
1074 | * the entire log is made up of log records which are the same size, | |
1075 | * we scan over the defined maximum blocks. At this point, the maximum | |
1076 | * is not chosen to mean anything special. XXXmiken | |
1077 | */ | |
1078 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
1079 | ASSERT(num_scan_bblks <= INT_MAX); | |
1080 | ||
1081 | if (last_blk < num_scan_bblks) | |
1082 | num_scan_bblks = last_blk; | |
1083 | start_blk = last_blk - num_scan_bblks; | |
1084 | ||
1085 | /* | |
1086 | * We search for any instances of cycle number 0 that occur before | |
1087 | * our current estimate of the head. What we're trying to detect is | |
1088 | * 1 ... | 0 | 1 | 0... | |
1089 | * ^ binary search ends here | |
1090 | */ | |
1091 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
1092 | (int)num_scan_bblks, 0, &new_blk))) | |
1093 | goto bp_err; | |
1094 | if (new_blk != -1) | |
1095 | last_blk = new_blk; | |
1096 | ||
1097 | /* | |
1098 | * Potentially backup over partial log record write. We don't need | |
1099 | * to search the end of the log because we know it is zero. | |
1100 | */ | |
1101 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
1102 | &last_blk, 0)) == -1) { | |
1103 | error = XFS_ERROR(EIO); | |
1104 | goto bp_err; | |
1105 | } else if (error) | |
1106 | goto bp_err; | |
1107 | ||
1108 | *blk_no = last_blk; | |
1109 | bp_err: | |
1110 | xlog_put_bp(bp); | |
1111 | if (error) | |
1112 | return error; | |
1113 | return -1; | |
1114 | } | |
1115 | ||
1116 | /* | |
1117 | * These are simple subroutines used by xlog_clear_stale_blocks() below | |
1118 | * to initialize a buffer full of empty log record headers and write | |
1119 | * them into the log. | |
1120 | */ | |
1121 | STATIC void | |
1122 | xlog_add_record( | |
1123 | xlog_t *log, | |
1124 | xfs_caddr_t buf, | |
1125 | int cycle, | |
1126 | int block, | |
1127 | int tail_cycle, | |
1128 | int tail_block) | |
1129 | { | |
1130 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; | |
1131 | ||
1132 | memset(buf, 0, BBSIZE); | |
b53e675d CH |
1133 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1134 | recp->h_cycle = cpu_to_be32(cycle); | |
1135 | recp->h_version = cpu_to_be32( | |
62118709 | 1136 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d CH |
1137 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
1138 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); | |
1139 | recp->h_fmt = cpu_to_be32(XLOG_FMT); | |
1da177e4 LT |
1140 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1141 | } | |
1142 | ||
1143 | STATIC int | |
1144 | xlog_write_log_records( | |
1145 | xlog_t *log, | |
1146 | int cycle, | |
1147 | int start_block, | |
1148 | int blocks, | |
1149 | int tail_cycle, | |
1150 | int tail_block) | |
1151 | { | |
1152 | xfs_caddr_t offset; | |
1153 | xfs_buf_t *bp; | |
1154 | int balign, ealign; | |
1155 | int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1); | |
1156 | int end_block = start_block + blocks; | |
1157 | int bufblks; | |
1158 | int error = 0; | |
1159 | int i, j = 0; | |
1160 | ||
1161 | bufblks = 1 << ffs(blocks); | |
1162 | while (!(bp = xlog_get_bp(log, bufblks))) { | |
1163 | bufblks >>= 1; | |
1164 | if (bufblks <= log->l_sectbb_log) | |
1165 | return ENOMEM; | |
1166 | } | |
1167 | ||
1168 | /* We may need to do a read at the start to fill in part of | |
1169 | * the buffer in the starting sector not covered by the first | |
1170 | * write below. | |
1171 | */ | |
1172 | balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block); | |
1173 | if (balign != start_block) { | |
076e6acb CH |
1174 | error = xlog_bread_noalign(log, start_block, 1, bp); |
1175 | if (error) | |
1176 | goto out_put_bp; | |
1177 | ||
1da177e4 LT |
1178 | j = start_block - balign; |
1179 | } | |
1180 | ||
1181 | for (i = start_block; i < end_block; i += bufblks) { | |
1182 | int bcount, endcount; | |
1183 | ||
1184 | bcount = min(bufblks, end_block - start_block); | |
1185 | endcount = bcount - j; | |
1186 | ||
1187 | /* We may need to do a read at the end to fill in part of | |
1188 | * the buffer in the final sector not covered by the write. | |
1189 | * If this is the same sector as the above read, skip it. | |
1190 | */ | |
1191 | ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block); | |
1192 | if (j == 0 && (start_block + endcount > ealign)) { | |
1193 | offset = XFS_BUF_PTR(bp); | |
1194 | balign = BBTOB(ealign - start_block); | |
234f56ac DC |
1195 | error = XFS_BUF_SET_PTR(bp, offset + balign, |
1196 | BBTOB(sectbb)); | |
076e6acb CH |
1197 | if (error) |
1198 | break; | |
1199 | ||
1200 | error = xlog_bread_noalign(log, ealign, sectbb, bp); | |
1201 | if (error) | |
1202 | break; | |
1203 | ||
1204 | error = XFS_BUF_SET_PTR(bp, offset, bufblks); | |
234f56ac | 1205 | if (error) |
1da177e4 | 1206 | break; |
1da177e4 LT |
1207 | } |
1208 | ||
1209 | offset = xlog_align(log, start_block, endcount, bp); | |
1210 | for (; j < endcount; j++) { | |
1211 | xlog_add_record(log, offset, cycle, i+j, | |
1212 | tail_cycle, tail_block); | |
1213 | offset += BBSIZE; | |
1214 | } | |
1215 | error = xlog_bwrite(log, start_block, endcount, bp); | |
1216 | if (error) | |
1217 | break; | |
1218 | start_block += endcount; | |
1219 | j = 0; | |
1220 | } | |
076e6acb CH |
1221 | |
1222 | out_put_bp: | |
1da177e4 LT |
1223 | xlog_put_bp(bp); |
1224 | return error; | |
1225 | } | |
1226 | ||
1227 | /* | |
1228 | * This routine is called to blow away any incomplete log writes out | |
1229 | * in front of the log head. We do this so that we won't become confused | |
1230 | * if we come up, write only a little bit more, and then crash again. | |
1231 | * If we leave the partial log records out there, this situation could | |
1232 | * cause us to think those partial writes are valid blocks since they | |
1233 | * have the current cycle number. We get rid of them by overwriting them | |
1234 | * with empty log records with the old cycle number rather than the | |
1235 | * current one. | |
1236 | * | |
1237 | * The tail lsn is passed in rather than taken from | |
1238 | * the log so that we will not write over the unmount record after a | |
1239 | * clean unmount in a 512 block log. Doing so would leave the log without | |
1240 | * any valid log records in it until a new one was written. If we crashed | |
1241 | * during that time we would not be able to recover. | |
1242 | */ | |
1243 | STATIC int | |
1244 | xlog_clear_stale_blocks( | |
1245 | xlog_t *log, | |
1246 | xfs_lsn_t tail_lsn) | |
1247 | { | |
1248 | int tail_cycle, head_cycle; | |
1249 | int tail_block, head_block; | |
1250 | int tail_distance, max_distance; | |
1251 | int distance; | |
1252 | int error; | |
1253 | ||
1254 | tail_cycle = CYCLE_LSN(tail_lsn); | |
1255 | tail_block = BLOCK_LSN(tail_lsn); | |
1256 | head_cycle = log->l_curr_cycle; | |
1257 | head_block = log->l_curr_block; | |
1258 | ||
1259 | /* | |
1260 | * Figure out the distance between the new head of the log | |
1261 | * and the tail. We want to write over any blocks beyond the | |
1262 | * head that we may have written just before the crash, but | |
1263 | * we don't want to overwrite the tail of the log. | |
1264 | */ | |
1265 | if (head_cycle == tail_cycle) { | |
1266 | /* | |
1267 | * The tail is behind the head in the physical log, | |
1268 | * so the distance from the head to the tail is the | |
1269 | * distance from the head to the end of the log plus | |
1270 | * the distance from the beginning of the log to the | |
1271 | * tail. | |
1272 | */ | |
1273 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { | |
1274 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", | |
1275 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1276 | return XFS_ERROR(EFSCORRUPTED); | |
1277 | } | |
1278 | tail_distance = tail_block + (log->l_logBBsize - head_block); | |
1279 | } else { | |
1280 | /* | |
1281 | * The head is behind the tail in the physical log, | |
1282 | * so the distance from the head to the tail is just | |
1283 | * the tail block minus the head block. | |
1284 | */ | |
1285 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ | |
1286 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", | |
1287 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1288 | return XFS_ERROR(EFSCORRUPTED); | |
1289 | } | |
1290 | tail_distance = tail_block - head_block; | |
1291 | } | |
1292 | ||
1293 | /* | |
1294 | * If the head is right up against the tail, we can't clear | |
1295 | * anything. | |
1296 | */ | |
1297 | if (tail_distance <= 0) { | |
1298 | ASSERT(tail_distance == 0); | |
1299 | return 0; | |
1300 | } | |
1301 | ||
1302 | max_distance = XLOG_TOTAL_REC_SHIFT(log); | |
1303 | /* | |
1304 | * Take the smaller of the maximum amount of outstanding I/O | |
1305 | * we could have and the distance to the tail to clear out. | |
1306 | * We take the smaller so that we don't overwrite the tail and | |
1307 | * we don't waste all day writing from the head to the tail | |
1308 | * for no reason. | |
1309 | */ | |
1310 | max_distance = MIN(max_distance, tail_distance); | |
1311 | ||
1312 | if ((head_block + max_distance) <= log->l_logBBsize) { | |
1313 | /* | |
1314 | * We can stomp all the blocks we need to without | |
1315 | * wrapping around the end of the log. Just do it | |
1316 | * in a single write. Use the cycle number of the | |
1317 | * current cycle minus one so that the log will look like: | |
1318 | * n ... | n - 1 ... | |
1319 | */ | |
1320 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1321 | head_block, max_distance, tail_cycle, | |
1322 | tail_block); | |
1323 | if (error) | |
1324 | return error; | |
1325 | } else { | |
1326 | /* | |
1327 | * We need to wrap around the end of the physical log in | |
1328 | * order to clear all the blocks. Do it in two separate | |
1329 | * I/Os. The first write should be from the head to the | |
1330 | * end of the physical log, and it should use the current | |
1331 | * cycle number minus one just like above. | |
1332 | */ | |
1333 | distance = log->l_logBBsize - head_block; | |
1334 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1335 | head_block, distance, tail_cycle, | |
1336 | tail_block); | |
1337 | ||
1338 | if (error) | |
1339 | return error; | |
1340 | ||
1341 | /* | |
1342 | * Now write the blocks at the start of the physical log. | |
1343 | * This writes the remainder of the blocks we want to clear. | |
1344 | * It uses the current cycle number since we're now on the | |
1345 | * same cycle as the head so that we get: | |
1346 | * n ... n ... | n - 1 ... | |
1347 | * ^^^^^ blocks we're writing | |
1348 | */ | |
1349 | distance = max_distance - (log->l_logBBsize - head_block); | |
1350 | error = xlog_write_log_records(log, head_cycle, 0, distance, | |
1351 | tail_cycle, tail_block); | |
1352 | if (error) | |
1353 | return error; | |
1354 | } | |
1355 | ||
1356 | return 0; | |
1357 | } | |
1358 | ||
1359 | /****************************************************************************** | |
1360 | * | |
1361 | * Log recover routines | |
1362 | * | |
1363 | ****************************************************************************** | |
1364 | */ | |
1365 | ||
1366 | STATIC xlog_recover_t * | |
1367 | xlog_recover_find_tid( | |
f0a76953 | 1368 | struct hlist_head *head, |
1da177e4 LT |
1369 | xlog_tid_t tid) |
1370 | { | |
f0a76953 DC |
1371 | xlog_recover_t *trans; |
1372 | struct hlist_node *n; | |
1da177e4 | 1373 | |
f0a76953 DC |
1374 | hlist_for_each_entry(trans, n, head, r_list) { |
1375 | if (trans->r_log_tid == tid) | |
1376 | return trans; | |
1da177e4 | 1377 | } |
f0a76953 | 1378 | return NULL; |
1da177e4 LT |
1379 | } |
1380 | ||
1381 | STATIC void | |
f0a76953 DC |
1382 | xlog_recover_new_tid( |
1383 | struct hlist_head *head, | |
1384 | xlog_tid_t tid, | |
1385 | xfs_lsn_t lsn) | |
1da177e4 | 1386 | { |
f0a76953 DC |
1387 | xlog_recover_t *trans; |
1388 | ||
1389 | trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP); | |
1390 | trans->r_log_tid = tid; | |
1391 | trans->r_lsn = lsn; | |
1392 | INIT_LIST_HEAD(&trans->r_itemq); | |
1393 | ||
1394 | INIT_HLIST_NODE(&trans->r_list); | |
1395 | hlist_add_head(&trans->r_list, head); | |
1da177e4 LT |
1396 | } |
1397 | ||
1398 | STATIC void | |
1399 | xlog_recover_add_item( | |
f0a76953 | 1400 | struct list_head *head) |
1da177e4 LT |
1401 | { |
1402 | xlog_recover_item_t *item; | |
1403 | ||
1404 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); | |
f0a76953 DC |
1405 | INIT_LIST_HEAD(&item->ri_list); |
1406 | list_add_tail(&item->ri_list, head); | |
1da177e4 LT |
1407 | } |
1408 | ||
1409 | STATIC int | |
1410 | xlog_recover_add_to_cont_trans( | |
9abbc539 | 1411 | struct log *log, |
1da177e4 LT |
1412 | xlog_recover_t *trans, |
1413 | xfs_caddr_t dp, | |
1414 | int len) | |
1415 | { | |
1416 | xlog_recover_item_t *item; | |
1417 | xfs_caddr_t ptr, old_ptr; | |
1418 | int old_len; | |
1419 | ||
f0a76953 | 1420 | if (list_empty(&trans->r_itemq)) { |
1da177e4 LT |
1421 | /* finish copying rest of trans header */ |
1422 | xlog_recover_add_item(&trans->r_itemq); | |
1423 | ptr = (xfs_caddr_t) &trans->r_theader + | |
1424 | sizeof(xfs_trans_header_t) - len; | |
1425 | memcpy(ptr, dp, len); /* d, s, l */ | |
1426 | return 0; | |
1427 | } | |
f0a76953 DC |
1428 | /* take the tail entry */ |
1429 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
1da177e4 LT |
1430 | |
1431 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; | |
1432 | old_len = item->ri_buf[item->ri_cnt-1].i_len; | |
1433 | ||
760dea67 | 1434 | ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u); |
1da177e4 LT |
1435 | memcpy(&ptr[old_len], dp, len); /* d, s, l */ |
1436 | item->ri_buf[item->ri_cnt-1].i_len += len; | |
1437 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; | |
9abbc539 | 1438 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); |
1da177e4 LT |
1439 | return 0; |
1440 | } | |
1441 | ||
1442 | /* | |
1443 | * The next region to add is the start of a new region. It could be | |
1444 | * a whole region or it could be the first part of a new region. Because | |
1445 | * of this, the assumption here is that the type and size fields of all | |
1446 | * format structures fit into the first 32 bits of the structure. | |
1447 | * | |
1448 | * This works because all regions must be 32 bit aligned. Therefore, we | |
1449 | * either have both fields or we have neither field. In the case we have | |
1450 | * neither field, the data part of the region is zero length. We only have | |
1451 | * a log_op_header and can throw away the header since a new one will appear | |
1452 | * later. If we have at least 4 bytes, then we can determine how many regions | |
1453 | * will appear in the current log item. | |
1454 | */ | |
1455 | STATIC int | |
1456 | xlog_recover_add_to_trans( | |
9abbc539 | 1457 | struct log *log, |
1da177e4 LT |
1458 | xlog_recover_t *trans, |
1459 | xfs_caddr_t dp, | |
1460 | int len) | |
1461 | { | |
1462 | xfs_inode_log_format_t *in_f; /* any will do */ | |
1463 | xlog_recover_item_t *item; | |
1464 | xfs_caddr_t ptr; | |
1465 | ||
1466 | if (!len) | |
1467 | return 0; | |
f0a76953 | 1468 | if (list_empty(&trans->r_itemq)) { |
5a792c45 DC |
1469 | /* we need to catch log corruptions here */ |
1470 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { | |
1471 | xlog_warn("XFS: xlog_recover_add_to_trans: " | |
1472 | "bad header magic number"); | |
1473 | ASSERT(0); | |
1474 | return XFS_ERROR(EIO); | |
1475 | } | |
1da177e4 LT |
1476 | if (len == sizeof(xfs_trans_header_t)) |
1477 | xlog_recover_add_item(&trans->r_itemq); | |
1478 | memcpy(&trans->r_theader, dp, len); /* d, s, l */ | |
1479 | return 0; | |
1480 | } | |
1481 | ||
1482 | ptr = kmem_alloc(len, KM_SLEEP); | |
1483 | memcpy(ptr, dp, len); | |
1484 | in_f = (xfs_inode_log_format_t *)ptr; | |
1485 | ||
f0a76953 DC |
1486 | /* take the tail entry */ |
1487 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
1488 | if (item->ri_total != 0 && | |
1489 | item->ri_total == item->ri_cnt) { | |
1490 | /* tail item is in use, get a new one */ | |
1da177e4 | 1491 | xlog_recover_add_item(&trans->r_itemq); |
f0a76953 DC |
1492 | item = list_entry(trans->r_itemq.prev, |
1493 | xlog_recover_item_t, ri_list); | |
1da177e4 | 1494 | } |
1da177e4 LT |
1495 | |
1496 | if (item->ri_total == 0) { /* first region to be added */ | |
e8fa6b48 CH |
1497 | if (in_f->ilf_size == 0 || |
1498 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { | |
1499 | xlog_warn( | |
1500 | "XFS: bad number of regions (%d) in inode log format", | |
1501 | in_f->ilf_size); | |
1502 | ASSERT(0); | |
1503 | return XFS_ERROR(EIO); | |
1504 | } | |
1505 | ||
1506 | item->ri_total = in_f->ilf_size; | |
1507 | item->ri_buf = | |
1508 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), | |
1509 | KM_SLEEP); | |
1da177e4 LT |
1510 | } |
1511 | ASSERT(item->ri_total > item->ri_cnt); | |
1512 | /* Description region is ri_buf[0] */ | |
1513 | item->ri_buf[item->ri_cnt].i_addr = ptr; | |
1514 | item->ri_buf[item->ri_cnt].i_len = len; | |
1515 | item->ri_cnt++; | |
9abbc539 | 1516 | trace_xfs_log_recover_item_add(log, trans, item, 0); |
1da177e4 LT |
1517 | return 0; |
1518 | } | |
1519 | ||
f0a76953 DC |
1520 | /* |
1521 | * Sort the log items in the transaction. Cancelled buffers need | |
1522 | * to be put first so they are processed before any items that might | |
1523 | * modify the buffers. If they are cancelled, then the modifications | |
1524 | * don't need to be replayed. | |
1525 | */ | |
1da177e4 LT |
1526 | STATIC int |
1527 | xlog_recover_reorder_trans( | |
9abbc539 DC |
1528 | struct log *log, |
1529 | xlog_recover_t *trans, | |
1530 | int pass) | |
1da177e4 | 1531 | { |
f0a76953 DC |
1532 | xlog_recover_item_t *item, *n; |
1533 | LIST_HEAD(sort_list); | |
1534 | ||
1535 | list_splice_init(&trans->r_itemq, &sort_list); | |
1536 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { | |
1537 | xfs_buf_log_format_t *buf_f; | |
1da177e4 | 1538 | |
f0a76953 | 1539 | buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr; |
1da177e4 | 1540 | |
f0a76953 | 1541 | switch (ITEM_TYPE(item)) { |
1da177e4 | 1542 | case XFS_LI_BUF: |
f0a76953 | 1543 | if (!(buf_f->blf_flags & XFS_BLI_CANCEL)) { |
9abbc539 DC |
1544 | trace_xfs_log_recover_item_reorder_head(log, |
1545 | trans, item, pass); | |
f0a76953 | 1546 | list_move(&item->ri_list, &trans->r_itemq); |
1da177e4 LT |
1547 | break; |
1548 | } | |
1549 | case XFS_LI_INODE: | |
1da177e4 LT |
1550 | case XFS_LI_DQUOT: |
1551 | case XFS_LI_QUOTAOFF: | |
1552 | case XFS_LI_EFD: | |
1553 | case XFS_LI_EFI: | |
9abbc539 DC |
1554 | trace_xfs_log_recover_item_reorder_tail(log, |
1555 | trans, item, pass); | |
f0a76953 | 1556 | list_move_tail(&item->ri_list, &trans->r_itemq); |
1da177e4 LT |
1557 | break; |
1558 | default: | |
1559 | xlog_warn( | |
1560 | "XFS: xlog_recover_reorder_trans: unrecognized type of log operation"); | |
1561 | ASSERT(0); | |
1562 | return XFS_ERROR(EIO); | |
1563 | } | |
f0a76953 DC |
1564 | } |
1565 | ASSERT(list_empty(&sort_list)); | |
1da177e4 LT |
1566 | return 0; |
1567 | } | |
1568 | ||
1569 | /* | |
1570 | * Build up the table of buf cancel records so that we don't replay | |
1571 | * cancelled data in the second pass. For buffer records that are | |
1572 | * not cancel records, there is nothing to do here so we just return. | |
1573 | * | |
1574 | * If we get a cancel record which is already in the table, this indicates | |
1575 | * that the buffer was cancelled multiple times. In order to ensure | |
1576 | * that during pass 2 we keep the record in the table until we reach its | |
1577 | * last occurrence in the log, we keep a reference count in the cancel | |
1578 | * record in the table to tell us how many times we expect to see this | |
1579 | * record during the second pass. | |
1580 | */ | |
1581 | STATIC void | |
1582 | xlog_recover_do_buffer_pass1( | |
1583 | xlog_t *log, | |
1584 | xfs_buf_log_format_t *buf_f) | |
1585 | { | |
1586 | xfs_buf_cancel_t *bcp; | |
1587 | xfs_buf_cancel_t *nextp; | |
1588 | xfs_buf_cancel_t *prevp; | |
1589 | xfs_buf_cancel_t **bucket; | |
1da177e4 LT |
1590 | xfs_daddr_t blkno = 0; |
1591 | uint len = 0; | |
1592 | ushort flags = 0; | |
1593 | ||
1594 | switch (buf_f->blf_type) { | |
1595 | case XFS_LI_BUF: | |
1596 | blkno = buf_f->blf_blkno; | |
1597 | len = buf_f->blf_len; | |
1598 | flags = buf_f->blf_flags; | |
1599 | break; | |
1da177e4 LT |
1600 | } |
1601 | ||
1602 | /* | |
1603 | * If this isn't a cancel buffer item, then just return. | |
1604 | */ | |
9abbc539 DC |
1605 | if (!(flags & XFS_BLI_CANCEL)) { |
1606 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); | |
1da177e4 | 1607 | return; |
9abbc539 | 1608 | } |
1da177e4 LT |
1609 | |
1610 | /* | |
1611 | * Insert an xfs_buf_cancel record into the hash table of | |
1612 | * them. If there is already an identical record, bump | |
1613 | * its reference count. | |
1614 | */ | |
1615 | bucket = &log->l_buf_cancel_table[(__uint64_t)blkno % | |
1616 | XLOG_BC_TABLE_SIZE]; | |
1617 | /* | |
1618 | * If the hash bucket is empty then just insert a new record into | |
1619 | * the bucket. | |
1620 | */ | |
1621 | if (*bucket == NULL) { | |
1622 | bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t), | |
1623 | KM_SLEEP); | |
1624 | bcp->bc_blkno = blkno; | |
1625 | bcp->bc_len = len; | |
1626 | bcp->bc_refcount = 1; | |
1627 | bcp->bc_next = NULL; | |
1628 | *bucket = bcp; | |
1629 | return; | |
1630 | } | |
1631 | ||
1632 | /* | |
1633 | * The hash bucket is not empty, so search for duplicates of our | |
1634 | * record. If we find one them just bump its refcount. If not | |
1635 | * then add us at the end of the list. | |
1636 | */ | |
1637 | prevp = NULL; | |
1638 | nextp = *bucket; | |
1639 | while (nextp != NULL) { | |
1640 | if (nextp->bc_blkno == blkno && nextp->bc_len == len) { | |
1641 | nextp->bc_refcount++; | |
9abbc539 | 1642 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
1da177e4 LT |
1643 | return; |
1644 | } | |
1645 | prevp = nextp; | |
1646 | nextp = nextp->bc_next; | |
1647 | } | |
1648 | ASSERT(prevp != NULL); | |
1649 | bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t), | |
1650 | KM_SLEEP); | |
1651 | bcp->bc_blkno = blkno; | |
1652 | bcp->bc_len = len; | |
1653 | bcp->bc_refcount = 1; | |
1654 | bcp->bc_next = NULL; | |
1655 | prevp->bc_next = bcp; | |
9abbc539 | 1656 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
1da177e4 LT |
1657 | } |
1658 | ||
1659 | /* | |
1660 | * Check to see whether the buffer being recovered has a corresponding | |
1661 | * entry in the buffer cancel record table. If it does then return 1 | |
1662 | * so that it will be cancelled, otherwise return 0. If the buffer is | |
1663 | * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement | |
1664 | * the refcount on the entry in the table and remove it from the table | |
1665 | * if this is the last reference. | |
1666 | * | |
1667 | * We remove the cancel record from the table when we encounter its | |
1668 | * last occurrence in the log so that if the same buffer is re-used | |
1669 | * again after its last cancellation we actually replay the changes | |
1670 | * made at that point. | |
1671 | */ | |
1672 | STATIC int | |
1673 | xlog_check_buffer_cancelled( | |
1674 | xlog_t *log, | |
1675 | xfs_daddr_t blkno, | |
1676 | uint len, | |
1677 | ushort flags) | |
1678 | { | |
1679 | xfs_buf_cancel_t *bcp; | |
1680 | xfs_buf_cancel_t *prevp; | |
1681 | xfs_buf_cancel_t **bucket; | |
1682 | ||
1683 | if (log->l_buf_cancel_table == NULL) { | |
1684 | /* | |
1685 | * There is nothing in the table built in pass one, | |
1686 | * so this buffer must not be cancelled. | |
1687 | */ | |
1688 | ASSERT(!(flags & XFS_BLI_CANCEL)); | |
1689 | return 0; | |
1690 | } | |
1691 | ||
1692 | bucket = &log->l_buf_cancel_table[(__uint64_t)blkno % | |
1693 | XLOG_BC_TABLE_SIZE]; | |
1694 | bcp = *bucket; | |
1695 | if (bcp == NULL) { | |
1696 | /* | |
1697 | * There is no corresponding entry in the table built | |
1698 | * in pass one, so this buffer has not been cancelled. | |
1699 | */ | |
1700 | ASSERT(!(flags & XFS_BLI_CANCEL)); | |
1701 | return 0; | |
1702 | } | |
1703 | ||
1704 | /* | |
1705 | * Search for an entry in the buffer cancel table that | |
1706 | * matches our buffer. | |
1707 | */ | |
1708 | prevp = NULL; | |
1709 | while (bcp != NULL) { | |
1710 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) { | |
1711 | /* | |
1712 | * We've go a match, so return 1 so that the | |
1713 | * recovery of this buffer is cancelled. | |
1714 | * If this buffer is actually a buffer cancel | |
1715 | * log item, then decrement the refcount on the | |
1716 | * one in the table and remove it if this is the | |
1717 | * last reference. | |
1718 | */ | |
1719 | if (flags & XFS_BLI_CANCEL) { | |
1720 | bcp->bc_refcount--; | |
1721 | if (bcp->bc_refcount == 0) { | |
1722 | if (prevp == NULL) { | |
1723 | *bucket = bcp->bc_next; | |
1724 | } else { | |
1725 | prevp->bc_next = bcp->bc_next; | |
1726 | } | |
f0e2d93c | 1727 | kmem_free(bcp); |
1da177e4 LT |
1728 | } |
1729 | } | |
1730 | return 1; | |
1731 | } | |
1732 | prevp = bcp; | |
1733 | bcp = bcp->bc_next; | |
1734 | } | |
1735 | /* | |
1736 | * We didn't find a corresponding entry in the table, so | |
1737 | * return 0 so that the buffer is NOT cancelled. | |
1738 | */ | |
1739 | ASSERT(!(flags & XFS_BLI_CANCEL)); | |
1740 | return 0; | |
1741 | } | |
1742 | ||
1743 | STATIC int | |
1744 | xlog_recover_do_buffer_pass2( | |
1745 | xlog_t *log, | |
1746 | xfs_buf_log_format_t *buf_f) | |
1747 | { | |
1da177e4 LT |
1748 | xfs_daddr_t blkno = 0; |
1749 | ushort flags = 0; | |
1750 | uint len = 0; | |
1751 | ||
1752 | switch (buf_f->blf_type) { | |
1753 | case XFS_LI_BUF: | |
1754 | blkno = buf_f->blf_blkno; | |
1755 | flags = buf_f->blf_flags; | |
1756 | len = buf_f->blf_len; | |
1757 | break; | |
1da177e4 LT |
1758 | } |
1759 | ||
1760 | return xlog_check_buffer_cancelled(log, blkno, len, flags); | |
1761 | } | |
1762 | ||
1763 | /* | |
1764 | * Perform recovery for a buffer full of inodes. In these buffers, | |
1765 | * the only data which should be recovered is that which corresponds | |
1766 | * to the di_next_unlinked pointers in the on disk inode structures. | |
1767 | * The rest of the data for the inodes is always logged through the | |
1768 | * inodes themselves rather than the inode buffer and is recovered | |
1769 | * in xlog_recover_do_inode_trans(). | |
1770 | * | |
1771 | * The only time when buffers full of inodes are fully recovered is | |
1772 | * when the buffer is full of newly allocated inodes. In this case | |
1773 | * the buffer will not be marked as an inode buffer and so will be | |
1774 | * sent to xlog_recover_do_reg_buffer() below during recovery. | |
1775 | */ | |
1776 | STATIC int | |
1777 | xlog_recover_do_inode_buffer( | |
1778 | xfs_mount_t *mp, | |
1779 | xlog_recover_item_t *item, | |
1780 | xfs_buf_t *bp, | |
1781 | xfs_buf_log_format_t *buf_f) | |
1782 | { | |
1783 | int i; | |
1784 | int item_index; | |
1785 | int bit; | |
1786 | int nbits; | |
1787 | int reg_buf_offset; | |
1788 | int reg_buf_bytes; | |
1789 | int next_unlinked_offset; | |
1790 | int inodes_per_buf; | |
1791 | xfs_agino_t *logged_nextp; | |
1792 | xfs_agino_t *buffer_nextp; | |
1da177e4 LT |
1793 | unsigned int *data_map = NULL; |
1794 | unsigned int map_size = 0; | |
1795 | ||
9abbc539 DC |
1796 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
1797 | ||
1da177e4 LT |
1798 | switch (buf_f->blf_type) { |
1799 | case XFS_LI_BUF: | |
1800 | data_map = buf_f->blf_data_map; | |
1801 | map_size = buf_f->blf_map_size; | |
1802 | break; | |
1da177e4 LT |
1803 | } |
1804 | /* | |
1805 | * Set the variables corresponding to the current region to | |
1806 | * 0 so that we'll initialize them on the first pass through | |
1807 | * the loop. | |
1808 | */ | |
1809 | reg_buf_offset = 0; | |
1810 | reg_buf_bytes = 0; | |
1811 | bit = 0; | |
1812 | nbits = 0; | |
1813 | item_index = 0; | |
1814 | inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog; | |
1815 | for (i = 0; i < inodes_per_buf; i++) { | |
1816 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + | |
1817 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1818 | ||
1819 | while (next_unlinked_offset >= | |
1820 | (reg_buf_offset + reg_buf_bytes)) { | |
1821 | /* | |
1822 | * The next di_next_unlinked field is beyond | |
1823 | * the current logged region. Find the next | |
1824 | * logged region that contains or is beyond | |
1825 | * the current di_next_unlinked field. | |
1826 | */ | |
1827 | bit += nbits; | |
1828 | bit = xfs_next_bit(data_map, map_size, bit); | |
1829 | ||
1830 | /* | |
1831 | * If there are no more logged regions in the | |
1832 | * buffer, then we're done. | |
1833 | */ | |
1834 | if (bit == -1) { | |
1835 | return 0; | |
1836 | } | |
1837 | ||
1838 | nbits = xfs_contig_bits(data_map, map_size, | |
1839 | bit); | |
1840 | ASSERT(nbits > 0); | |
1841 | reg_buf_offset = bit << XFS_BLI_SHIFT; | |
1842 | reg_buf_bytes = nbits << XFS_BLI_SHIFT; | |
1843 | item_index++; | |
1844 | } | |
1845 | ||
1846 | /* | |
1847 | * If the current logged region starts after the current | |
1848 | * di_next_unlinked field, then move on to the next | |
1849 | * di_next_unlinked field. | |
1850 | */ | |
1851 | if (next_unlinked_offset < reg_buf_offset) { | |
1852 | continue; | |
1853 | } | |
1854 | ||
1855 | ASSERT(item->ri_buf[item_index].i_addr != NULL); | |
1856 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0); | |
1857 | ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp)); | |
1858 | ||
1859 | /* | |
1860 | * The current logged region contains a copy of the | |
1861 | * current di_next_unlinked field. Extract its value | |
1862 | * and copy it to the buffer copy. | |
1863 | */ | |
1864 | logged_nextp = (xfs_agino_t *) | |
1865 | ((char *)(item->ri_buf[item_index].i_addr) + | |
1866 | (next_unlinked_offset - reg_buf_offset)); | |
1867 | if (unlikely(*logged_nextp == 0)) { | |
1868 | xfs_fs_cmn_err(CE_ALERT, mp, | |
1869 | "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field", | |
1870 | item, bp); | |
1871 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", | |
1872 | XFS_ERRLEVEL_LOW, mp); | |
1873 | return XFS_ERROR(EFSCORRUPTED); | |
1874 | } | |
1875 | ||
1876 | buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp, | |
1877 | next_unlinked_offset); | |
87c199c2 | 1878 | *buffer_nextp = *logged_nextp; |
1da177e4 LT |
1879 | } |
1880 | ||
1881 | return 0; | |
1882 | } | |
1883 | ||
1884 | /* | |
1885 | * Perform a 'normal' buffer recovery. Each logged region of the | |
1886 | * buffer should be copied over the corresponding region in the | |
1887 | * given buffer. The bitmap in the buf log format structure indicates | |
1888 | * where to place the logged data. | |
1889 | */ | |
1890 | /*ARGSUSED*/ | |
1891 | STATIC void | |
1892 | xlog_recover_do_reg_buffer( | |
9abbc539 | 1893 | struct xfs_mount *mp, |
1da177e4 LT |
1894 | xlog_recover_item_t *item, |
1895 | xfs_buf_t *bp, | |
1896 | xfs_buf_log_format_t *buf_f) | |
1897 | { | |
1898 | int i; | |
1899 | int bit; | |
1900 | int nbits; | |
1da177e4 LT |
1901 | unsigned int *data_map = NULL; |
1902 | unsigned int map_size = 0; | |
1903 | int error; | |
1904 | ||
9abbc539 DC |
1905 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); |
1906 | ||
1da177e4 LT |
1907 | switch (buf_f->blf_type) { |
1908 | case XFS_LI_BUF: | |
1909 | data_map = buf_f->blf_data_map; | |
1910 | map_size = buf_f->blf_map_size; | |
1911 | break; | |
1da177e4 LT |
1912 | } |
1913 | bit = 0; | |
1914 | i = 1; /* 0 is the buf format structure */ | |
1915 | while (1) { | |
1916 | bit = xfs_next_bit(data_map, map_size, bit); | |
1917 | if (bit == -1) | |
1918 | break; | |
1919 | nbits = xfs_contig_bits(data_map, map_size, bit); | |
1920 | ASSERT(nbits > 0); | |
4b80916b | 1921 | ASSERT(item->ri_buf[i].i_addr != NULL); |
1da177e4 LT |
1922 | ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0); |
1923 | ASSERT(XFS_BUF_COUNT(bp) >= | |
1924 | ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT)); | |
1925 | ||
1926 | /* | |
1927 | * Do a sanity check if this is a dquot buffer. Just checking | |
1928 | * the first dquot in the buffer should do. XXXThis is | |
1929 | * probably a good thing to do for other buf types also. | |
1930 | */ | |
1931 | error = 0; | |
c8ad20ff NS |
1932 | if (buf_f->blf_flags & |
1933 | (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) { | |
0c5e1ce8 CH |
1934 | if (item->ri_buf[i].i_addr == NULL) { |
1935 | cmn_err(CE_ALERT, | |
1936 | "XFS: NULL dquot in %s.", __func__); | |
1937 | goto next; | |
1938 | } | |
8ec6dba2 | 1939 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { |
0c5e1ce8 CH |
1940 | cmn_err(CE_ALERT, |
1941 | "XFS: dquot too small (%d) in %s.", | |
1942 | item->ri_buf[i].i_len, __func__); | |
1943 | goto next; | |
1944 | } | |
1da177e4 LT |
1945 | error = xfs_qm_dqcheck((xfs_disk_dquot_t *) |
1946 | item->ri_buf[i].i_addr, | |
1947 | -1, 0, XFS_QMOPT_DOWARN, | |
1948 | "dquot_buf_recover"); | |
0c5e1ce8 CH |
1949 | if (error) |
1950 | goto next; | |
1da177e4 | 1951 | } |
0c5e1ce8 CH |
1952 | |
1953 | memcpy(xfs_buf_offset(bp, | |
1954 | (uint)bit << XFS_BLI_SHIFT), /* dest */ | |
1955 | item->ri_buf[i].i_addr, /* source */ | |
1956 | nbits<<XFS_BLI_SHIFT); /* length */ | |
1957 | next: | |
1da177e4 LT |
1958 | i++; |
1959 | bit += nbits; | |
1960 | } | |
1961 | ||
1962 | /* Shouldn't be any more regions */ | |
1963 | ASSERT(i == item->ri_total); | |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * Do some primitive error checking on ondisk dquot data structures. | |
1968 | */ | |
1969 | int | |
1970 | xfs_qm_dqcheck( | |
1971 | xfs_disk_dquot_t *ddq, | |
1972 | xfs_dqid_t id, | |
1973 | uint type, /* used only when IO_dorepair is true */ | |
1974 | uint flags, | |
1975 | char *str) | |
1976 | { | |
1977 | xfs_dqblk_t *d = (xfs_dqblk_t *)ddq; | |
1978 | int errs = 0; | |
1979 | ||
1980 | /* | |
1981 | * We can encounter an uninitialized dquot buffer for 2 reasons: | |
1982 | * 1. If we crash while deleting the quotainode(s), and those blks got | |
1983 | * used for user data. This is because we take the path of regular | |
1984 | * file deletion; however, the size field of quotainodes is never | |
1985 | * updated, so all the tricks that we play in itruncate_finish | |
1986 | * don't quite matter. | |
1987 | * | |
1988 | * 2. We don't play the quota buffers when there's a quotaoff logitem. | |
1989 | * But the allocation will be replayed so we'll end up with an | |
1990 | * uninitialized quota block. | |
1991 | * | |
1992 | * This is all fine; things are still consistent, and we haven't lost | |
1993 | * any quota information. Just don't complain about bad dquot blks. | |
1994 | */ | |
1149d96a | 1995 | if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) { |
1da177e4 LT |
1996 | if (flags & XFS_QMOPT_DOWARN) |
1997 | cmn_err(CE_ALERT, | |
1998 | "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x", | |
1149d96a | 1999 | str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC); |
1da177e4 LT |
2000 | errs++; |
2001 | } | |
1149d96a | 2002 | if (ddq->d_version != XFS_DQUOT_VERSION) { |
1da177e4 LT |
2003 | if (flags & XFS_QMOPT_DOWARN) |
2004 | cmn_err(CE_ALERT, | |
2005 | "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x", | |
1149d96a | 2006 | str, id, ddq->d_version, XFS_DQUOT_VERSION); |
1da177e4 LT |
2007 | errs++; |
2008 | } | |
2009 | ||
1149d96a CH |
2010 | if (ddq->d_flags != XFS_DQ_USER && |
2011 | ddq->d_flags != XFS_DQ_PROJ && | |
2012 | ddq->d_flags != XFS_DQ_GROUP) { | |
1da177e4 LT |
2013 | if (flags & XFS_QMOPT_DOWARN) |
2014 | cmn_err(CE_ALERT, | |
2015 | "%s : XFS dquot ID 0x%x, unknown flags 0x%x", | |
1149d96a | 2016 | str, id, ddq->d_flags); |
1da177e4 LT |
2017 | errs++; |
2018 | } | |
2019 | ||
1149d96a | 2020 | if (id != -1 && id != be32_to_cpu(ddq->d_id)) { |
1da177e4 LT |
2021 | if (flags & XFS_QMOPT_DOWARN) |
2022 | cmn_err(CE_ALERT, | |
2023 | "%s : ondisk-dquot 0x%p, ID mismatch: " | |
2024 | "0x%x expected, found id 0x%x", | |
1149d96a | 2025 | str, ddq, id, be32_to_cpu(ddq->d_id)); |
1da177e4 LT |
2026 | errs++; |
2027 | } | |
2028 | ||
2029 | if (!errs && ddq->d_id) { | |
1149d96a CH |
2030 | if (ddq->d_blk_softlimit && |
2031 | be64_to_cpu(ddq->d_bcount) >= | |
2032 | be64_to_cpu(ddq->d_blk_softlimit)) { | |
1da177e4 LT |
2033 | if (!ddq->d_btimer) { |
2034 | if (flags & XFS_QMOPT_DOWARN) | |
2035 | cmn_err(CE_ALERT, | |
2036 | "%s : Dquot ID 0x%x (0x%p) " | |
2037 | "BLK TIMER NOT STARTED", | |
1149d96a | 2038 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2039 | errs++; |
2040 | } | |
2041 | } | |
1149d96a CH |
2042 | if (ddq->d_ino_softlimit && |
2043 | be64_to_cpu(ddq->d_icount) >= | |
2044 | be64_to_cpu(ddq->d_ino_softlimit)) { | |
1da177e4 LT |
2045 | if (!ddq->d_itimer) { |
2046 | if (flags & XFS_QMOPT_DOWARN) | |
2047 | cmn_err(CE_ALERT, | |
2048 | "%s : Dquot ID 0x%x (0x%p) " | |
2049 | "INODE TIMER NOT STARTED", | |
1149d96a | 2050 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2051 | errs++; |
2052 | } | |
2053 | } | |
1149d96a CH |
2054 | if (ddq->d_rtb_softlimit && |
2055 | be64_to_cpu(ddq->d_rtbcount) >= | |
2056 | be64_to_cpu(ddq->d_rtb_softlimit)) { | |
1da177e4 LT |
2057 | if (!ddq->d_rtbtimer) { |
2058 | if (flags & XFS_QMOPT_DOWARN) | |
2059 | cmn_err(CE_ALERT, | |
2060 | "%s : Dquot ID 0x%x (0x%p) " | |
2061 | "RTBLK TIMER NOT STARTED", | |
1149d96a | 2062 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2063 | errs++; |
2064 | } | |
2065 | } | |
2066 | } | |
2067 | ||
2068 | if (!errs || !(flags & XFS_QMOPT_DQREPAIR)) | |
2069 | return errs; | |
2070 | ||
2071 | if (flags & XFS_QMOPT_DOWARN) | |
2072 | cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id); | |
2073 | ||
2074 | /* | |
2075 | * Typically, a repair is only requested by quotacheck. | |
2076 | */ | |
2077 | ASSERT(id != -1); | |
2078 | ASSERT(flags & XFS_QMOPT_DQREPAIR); | |
2079 | memset(d, 0, sizeof(xfs_dqblk_t)); | |
1149d96a CH |
2080 | |
2081 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
2082 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
2083 | d->dd_diskdq.d_flags = type; | |
2084 | d->dd_diskdq.d_id = cpu_to_be32(id); | |
1da177e4 LT |
2085 | |
2086 | return errs; | |
2087 | } | |
2088 | ||
2089 | /* | |
2090 | * Perform a dquot buffer recovery. | |
2091 | * Simple algorithm: if we have found a QUOTAOFF logitem of the same type | |
2092 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. | |
2093 | * Else, treat it as a regular buffer and do recovery. | |
2094 | */ | |
2095 | STATIC void | |
2096 | xlog_recover_do_dquot_buffer( | |
2097 | xfs_mount_t *mp, | |
2098 | xlog_t *log, | |
2099 | xlog_recover_item_t *item, | |
2100 | xfs_buf_t *bp, | |
2101 | xfs_buf_log_format_t *buf_f) | |
2102 | { | |
2103 | uint type; | |
2104 | ||
9abbc539 DC |
2105 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
2106 | ||
1da177e4 LT |
2107 | /* |
2108 | * Filesystems are required to send in quota flags at mount time. | |
2109 | */ | |
2110 | if (mp->m_qflags == 0) { | |
2111 | return; | |
2112 | } | |
2113 | ||
2114 | type = 0; | |
2115 | if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF) | |
2116 | type |= XFS_DQ_USER; | |
c8ad20ff NS |
2117 | if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF) |
2118 | type |= XFS_DQ_PROJ; | |
1da177e4 LT |
2119 | if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF) |
2120 | type |= XFS_DQ_GROUP; | |
2121 | /* | |
2122 | * This type of quotas was turned off, so ignore this buffer | |
2123 | */ | |
2124 | if (log->l_quotaoffs_flag & type) | |
2125 | return; | |
2126 | ||
9abbc539 | 2127 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 LT |
2128 | } |
2129 | ||
2130 | /* | |
2131 | * This routine replays a modification made to a buffer at runtime. | |
2132 | * There are actually two types of buffer, regular and inode, which | |
2133 | * are handled differently. Inode buffers are handled differently | |
2134 | * in that we only recover a specific set of data from them, namely | |
2135 | * the inode di_next_unlinked fields. This is because all other inode | |
2136 | * data is actually logged via inode records and any data we replay | |
2137 | * here which overlaps that may be stale. | |
2138 | * | |
2139 | * When meta-data buffers are freed at run time we log a buffer item | |
2140 | * with the XFS_BLI_CANCEL bit set to indicate that previous copies | |
2141 | * of the buffer in the log should not be replayed at recovery time. | |
2142 | * This is so that if the blocks covered by the buffer are reused for | |
2143 | * file data before we crash we don't end up replaying old, freed | |
2144 | * meta-data into a user's file. | |
2145 | * | |
2146 | * To handle the cancellation of buffer log items, we make two passes | |
2147 | * over the log during recovery. During the first we build a table of | |
2148 | * those buffers which have been cancelled, and during the second we | |
2149 | * only replay those buffers which do not have corresponding cancel | |
2150 | * records in the table. See xlog_recover_do_buffer_pass[1,2] above | |
2151 | * for more details on the implementation of the table of cancel records. | |
2152 | */ | |
2153 | STATIC int | |
2154 | xlog_recover_do_buffer_trans( | |
2155 | xlog_t *log, | |
2156 | xlog_recover_item_t *item, | |
2157 | int pass) | |
2158 | { | |
2159 | xfs_buf_log_format_t *buf_f; | |
1da177e4 LT |
2160 | xfs_mount_t *mp; |
2161 | xfs_buf_t *bp; | |
2162 | int error; | |
2163 | int cancel; | |
2164 | xfs_daddr_t blkno; | |
2165 | int len; | |
2166 | ushort flags; | |
6ad112bf | 2167 | uint buf_flags; |
1da177e4 LT |
2168 | |
2169 | buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr; | |
2170 | ||
2171 | if (pass == XLOG_RECOVER_PASS1) { | |
2172 | /* | |
2173 | * In this pass we're only looking for buf items | |
2174 | * with the XFS_BLI_CANCEL bit set. | |
2175 | */ | |
2176 | xlog_recover_do_buffer_pass1(log, buf_f); | |
2177 | return 0; | |
2178 | } else { | |
2179 | /* | |
2180 | * In this pass we want to recover all the buffers | |
2181 | * which have not been cancelled and are not | |
2182 | * cancellation buffers themselves. The routine | |
2183 | * we call here will tell us whether or not to | |
2184 | * continue with the replay of this buffer. | |
2185 | */ | |
2186 | cancel = xlog_recover_do_buffer_pass2(log, buf_f); | |
2187 | if (cancel) { | |
9abbc539 | 2188 | trace_xfs_log_recover_buf_cancel(log, buf_f); |
1da177e4 LT |
2189 | return 0; |
2190 | } | |
2191 | } | |
9abbc539 | 2192 | trace_xfs_log_recover_buf_recover(log, buf_f); |
1da177e4 LT |
2193 | switch (buf_f->blf_type) { |
2194 | case XFS_LI_BUF: | |
2195 | blkno = buf_f->blf_blkno; | |
2196 | len = buf_f->blf_len; | |
2197 | flags = buf_f->blf_flags; | |
2198 | break; | |
1da177e4 LT |
2199 | default: |
2200 | xfs_fs_cmn_err(CE_ALERT, log->l_mp, | |
fc1f8c1c NS |
2201 | "xfs_log_recover: unknown buffer type 0x%x, logdev %s", |
2202 | buf_f->blf_type, log->l_mp->m_logname ? | |
2203 | log->l_mp->m_logname : "internal"); | |
1da177e4 LT |
2204 | XFS_ERROR_REPORT("xlog_recover_do_buffer_trans", |
2205 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2206 | return XFS_ERROR(EFSCORRUPTED); | |
2207 | } | |
2208 | ||
2209 | mp = log->l_mp; | |
0cadda1c | 2210 | buf_flags = XBF_LOCK; |
6ad112bf | 2211 | if (!(flags & XFS_BLI_INODE_BUF)) |
0cadda1c | 2212 | buf_flags |= XBF_MAPPED; |
6ad112bf CH |
2213 | |
2214 | bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, buf_flags); | |
1da177e4 LT |
2215 | if (XFS_BUF_ISERROR(bp)) { |
2216 | xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp, | |
2217 | bp, blkno); | |
2218 | error = XFS_BUF_GETERROR(bp); | |
2219 | xfs_buf_relse(bp); | |
2220 | return error; | |
2221 | } | |
2222 | ||
2223 | error = 0; | |
2224 | if (flags & XFS_BLI_INODE_BUF) { | |
2225 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); | |
c8ad20ff NS |
2226 | } else if (flags & |
2227 | (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) { | |
1da177e4 LT |
2228 | xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); |
2229 | } else { | |
9abbc539 | 2230 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 LT |
2231 | } |
2232 | if (error) | |
2233 | return XFS_ERROR(error); | |
2234 | ||
2235 | /* | |
2236 | * Perform delayed write on the buffer. Asynchronous writes will be | |
2237 | * slower when taking into account all the buffers to be flushed. | |
2238 | * | |
2239 | * Also make sure that only inode buffers with good sizes stay in | |
2240 | * the buffer cache. The kernel moves inodes in buffers of 1 block | |
2241 | * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode | |
2242 | * buffers in the log can be a different size if the log was generated | |
2243 | * by an older kernel using unclustered inode buffers or a newer kernel | |
2244 | * running with a different inode cluster size. Regardless, if the | |
2245 | * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE) | |
2246 | * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep | |
2247 | * the buffer out of the buffer cache so that the buffer won't | |
2248 | * overlap with future reads of those inodes. | |
2249 | */ | |
2250 | if (XFS_DINODE_MAGIC == | |
b53e675d | 2251 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
1da177e4 LT |
2252 | (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize, |
2253 | (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) { | |
2254 | XFS_BUF_STALE(bp); | |
2255 | error = xfs_bwrite(mp, bp); | |
2256 | } else { | |
15ac08a8 CH |
2257 | ASSERT(bp->b_mount == NULL || bp->b_mount == mp); |
2258 | bp->b_mount = mp; | |
1da177e4 LT |
2259 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone); |
2260 | xfs_bdwrite(mp, bp); | |
2261 | } | |
2262 | ||
2263 | return (error); | |
2264 | } | |
2265 | ||
2266 | STATIC int | |
2267 | xlog_recover_do_inode_trans( | |
2268 | xlog_t *log, | |
2269 | xlog_recover_item_t *item, | |
2270 | int pass) | |
2271 | { | |
2272 | xfs_inode_log_format_t *in_f; | |
2273 | xfs_mount_t *mp; | |
2274 | xfs_buf_t *bp; | |
1da177e4 LT |
2275 | xfs_dinode_t *dip; |
2276 | xfs_ino_t ino; | |
2277 | int len; | |
2278 | xfs_caddr_t src; | |
2279 | xfs_caddr_t dest; | |
2280 | int error; | |
2281 | int attr_index; | |
2282 | uint fields; | |
347d1c01 | 2283 | xfs_icdinode_t *dicp; |
6d192a9b | 2284 | int need_free = 0; |
1da177e4 LT |
2285 | |
2286 | if (pass == XLOG_RECOVER_PASS1) { | |
2287 | return 0; | |
2288 | } | |
2289 | ||
6d192a9b TS |
2290 | if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) { |
2291 | in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr; | |
2292 | } else { | |
2293 | in_f = (xfs_inode_log_format_t *)kmem_alloc( | |
2294 | sizeof(xfs_inode_log_format_t), KM_SLEEP); | |
2295 | need_free = 1; | |
2296 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); | |
2297 | if (error) | |
2298 | goto error; | |
2299 | } | |
1da177e4 LT |
2300 | ino = in_f->ilf_ino; |
2301 | mp = log->l_mp; | |
1da177e4 LT |
2302 | |
2303 | /* | |
2304 | * Inode buffers can be freed, look out for it, | |
2305 | * and do not replay the inode. | |
2306 | */ | |
a1941895 CH |
2307 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
2308 | in_f->ilf_len, 0)) { | |
6d192a9b | 2309 | error = 0; |
9abbc539 | 2310 | trace_xfs_log_recover_inode_cancel(log, in_f); |
6d192a9b TS |
2311 | goto error; |
2312 | } | |
9abbc539 | 2313 | trace_xfs_log_recover_inode_recover(log, in_f); |
1da177e4 | 2314 | |
6ad112bf | 2315 | bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, |
0cadda1c | 2316 | XBF_LOCK); |
1da177e4 LT |
2317 | if (XFS_BUF_ISERROR(bp)) { |
2318 | xfs_ioerror_alert("xlog_recover_do..(read#2)", mp, | |
a1941895 | 2319 | bp, in_f->ilf_blkno); |
1da177e4 LT |
2320 | error = XFS_BUF_GETERROR(bp); |
2321 | xfs_buf_relse(bp); | |
6d192a9b | 2322 | goto error; |
1da177e4 LT |
2323 | } |
2324 | error = 0; | |
2325 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); | |
a1941895 | 2326 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset); |
1da177e4 LT |
2327 | |
2328 | /* | |
2329 | * Make sure the place we're flushing out to really looks | |
2330 | * like an inode! | |
2331 | */ | |
81591fe2 | 2332 | if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) { |
1da177e4 LT |
2333 | xfs_buf_relse(bp); |
2334 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2335 | "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld", | |
2336 | dip, bp, ino); | |
2337 | XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)", | |
2338 | XFS_ERRLEVEL_LOW, mp); | |
6d192a9b TS |
2339 | error = EFSCORRUPTED; |
2340 | goto error; | |
1da177e4 | 2341 | } |
347d1c01 | 2342 | dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr); |
1da177e4 LT |
2343 | if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) { |
2344 | xfs_buf_relse(bp); | |
2345 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2346 | "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld", | |
2347 | item, ino); | |
2348 | XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)", | |
2349 | XFS_ERRLEVEL_LOW, mp); | |
6d192a9b TS |
2350 | error = EFSCORRUPTED; |
2351 | goto error; | |
1da177e4 LT |
2352 | } |
2353 | ||
2354 | /* Skip replay when the on disk inode is newer than the log one */ | |
81591fe2 | 2355 | if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) { |
1da177e4 LT |
2356 | /* |
2357 | * Deal with the wrap case, DI_MAX_FLUSH is less | |
2358 | * than smaller numbers | |
2359 | */ | |
81591fe2 | 2360 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
347d1c01 | 2361 | dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
1da177e4 LT |
2362 | /* do nothing */ |
2363 | } else { | |
2364 | xfs_buf_relse(bp); | |
9abbc539 | 2365 | trace_xfs_log_recover_inode_skip(log, in_f); |
6d192a9b TS |
2366 | error = 0; |
2367 | goto error; | |
1da177e4 LT |
2368 | } |
2369 | } | |
2370 | /* Take the opportunity to reset the flush iteration count */ | |
2371 | dicp->di_flushiter = 0; | |
2372 | ||
2373 | if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) { | |
2374 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && | |
2375 | (dicp->di_format != XFS_DINODE_FMT_BTREE)) { | |
2376 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)", | |
2377 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2378 | xfs_buf_relse(bp); | |
2379 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2380 | "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2381 | item, dip, bp, ino); | |
6d192a9b TS |
2382 | error = EFSCORRUPTED; |
2383 | goto error; | |
1da177e4 LT |
2384 | } |
2385 | } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) { | |
2386 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && | |
2387 | (dicp->di_format != XFS_DINODE_FMT_BTREE) && | |
2388 | (dicp->di_format != XFS_DINODE_FMT_LOCAL)) { | |
2389 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)", | |
2390 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2391 | xfs_buf_relse(bp); | |
2392 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2393 | "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2394 | item, dip, bp, ino); | |
6d192a9b TS |
2395 | error = EFSCORRUPTED; |
2396 | goto error; | |
1da177e4 LT |
2397 | } |
2398 | } | |
2399 | if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){ | |
2400 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)", | |
2401 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2402 | xfs_buf_relse(bp); | |
2403 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2404 | "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld", | |
2405 | item, dip, bp, ino, | |
2406 | dicp->di_nextents + dicp->di_anextents, | |
2407 | dicp->di_nblocks); | |
6d192a9b TS |
2408 | error = EFSCORRUPTED; |
2409 | goto error; | |
1da177e4 LT |
2410 | } |
2411 | if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) { | |
2412 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)", | |
2413 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2414 | xfs_buf_relse(bp); | |
2415 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2416 | "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x", | |
2417 | item, dip, bp, ino, dicp->di_forkoff); | |
6d192a9b TS |
2418 | error = EFSCORRUPTED; |
2419 | goto error; | |
1da177e4 | 2420 | } |
81591fe2 | 2421 | if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) { |
1da177e4 LT |
2422 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)", |
2423 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2424 | xfs_buf_relse(bp); | |
2425 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2426 | "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p", | |
2427 | item->ri_buf[1].i_len, item); | |
6d192a9b TS |
2428 | error = EFSCORRUPTED; |
2429 | goto error; | |
1da177e4 LT |
2430 | } |
2431 | ||
2432 | /* The core is in in-core format */ | |
81591fe2 | 2433 | xfs_dinode_to_disk(dip, (xfs_icdinode_t *)item->ri_buf[1].i_addr); |
1da177e4 LT |
2434 | |
2435 | /* the rest is in on-disk format */ | |
81591fe2 CH |
2436 | if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) { |
2437 | memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode), | |
2438 | item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode), | |
2439 | item->ri_buf[1].i_len - sizeof(struct xfs_icdinode)); | |
1da177e4 LT |
2440 | } |
2441 | ||
2442 | fields = in_f->ilf_fields; | |
2443 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { | |
2444 | case XFS_ILOG_DEV: | |
81591fe2 | 2445 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
1da177e4 LT |
2446 | break; |
2447 | case XFS_ILOG_UUID: | |
81591fe2 CH |
2448 | memcpy(XFS_DFORK_DPTR(dip), |
2449 | &in_f->ilf_u.ilfu_uuid, | |
2450 | sizeof(uuid_t)); | |
1da177e4 LT |
2451 | break; |
2452 | } | |
2453 | ||
2454 | if (in_f->ilf_size == 2) | |
2455 | goto write_inode_buffer; | |
2456 | len = item->ri_buf[2].i_len; | |
2457 | src = item->ri_buf[2].i_addr; | |
2458 | ASSERT(in_f->ilf_size <= 4); | |
2459 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); | |
2460 | ASSERT(!(fields & XFS_ILOG_DFORK) || | |
2461 | (len == in_f->ilf_dsize)); | |
2462 | ||
2463 | switch (fields & XFS_ILOG_DFORK) { | |
2464 | case XFS_ILOG_DDATA: | |
2465 | case XFS_ILOG_DEXT: | |
81591fe2 | 2466 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
1da177e4 LT |
2467 | break; |
2468 | ||
2469 | case XFS_ILOG_DBROOT: | |
7cc95a82 | 2470 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
81591fe2 | 2471 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
1da177e4 LT |
2472 | XFS_DFORK_DSIZE(dip, mp)); |
2473 | break; | |
2474 | ||
2475 | default: | |
2476 | /* | |
2477 | * There are no data fork flags set. | |
2478 | */ | |
2479 | ASSERT((fields & XFS_ILOG_DFORK) == 0); | |
2480 | break; | |
2481 | } | |
2482 | ||
2483 | /* | |
2484 | * If we logged any attribute data, recover it. There may or | |
2485 | * may not have been any other non-core data logged in this | |
2486 | * transaction. | |
2487 | */ | |
2488 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2489 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { | |
2490 | attr_index = 3; | |
2491 | } else { | |
2492 | attr_index = 2; | |
2493 | } | |
2494 | len = item->ri_buf[attr_index].i_len; | |
2495 | src = item->ri_buf[attr_index].i_addr; | |
2496 | ASSERT(len == in_f->ilf_asize); | |
2497 | ||
2498 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2499 | case XFS_ILOG_ADATA: | |
2500 | case XFS_ILOG_AEXT: | |
2501 | dest = XFS_DFORK_APTR(dip); | |
2502 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); | |
2503 | memcpy(dest, src, len); | |
2504 | break; | |
2505 | ||
2506 | case XFS_ILOG_ABROOT: | |
2507 | dest = XFS_DFORK_APTR(dip); | |
7cc95a82 CH |
2508 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
2509 | len, (xfs_bmdr_block_t*)dest, | |
1da177e4 LT |
2510 | XFS_DFORK_ASIZE(dip, mp)); |
2511 | break; | |
2512 | ||
2513 | default: | |
2514 | xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag"); | |
2515 | ASSERT(0); | |
2516 | xfs_buf_relse(bp); | |
6d192a9b TS |
2517 | error = EIO; |
2518 | goto error; | |
1da177e4 LT |
2519 | } |
2520 | } | |
2521 | ||
2522 | write_inode_buffer: | |
dd0bbad8 CH |
2523 | ASSERT(bp->b_mount == NULL || bp->b_mount == mp); |
2524 | bp->b_mount = mp; | |
2525 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone); | |
2526 | xfs_bdwrite(mp, bp); | |
6d192a9b TS |
2527 | error: |
2528 | if (need_free) | |
f0e2d93c | 2529 | kmem_free(in_f); |
6d192a9b | 2530 | return XFS_ERROR(error); |
1da177e4 LT |
2531 | } |
2532 | ||
2533 | /* | |
2534 | * Recover QUOTAOFF records. We simply make a note of it in the xlog_t | |
2535 | * structure, so that we know not to do any dquot item or dquot buffer recovery, | |
2536 | * of that type. | |
2537 | */ | |
2538 | STATIC int | |
2539 | xlog_recover_do_quotaoff_trans( | |
2540 | xlog_t *log, | |
2541 | xlog_recover_item_t *item, | |
2542 | int pass) | |
2543 | { | |
2544 | xfs_qoff_logformat_t *qoff_f; | |
2545 | ||
2546 | if (pass == XLOG_RECOVER_PASS2) { | |
2547 | return (0); | |
2548 | } | |
2549 | ||
2550 | qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr; | |
2551 | ASSERT(qoff_f); | |
2552 | ||
2553 | /* | |
2554 | * The logitem format's flag tells us if this was user quotaoff, | |
77a7cce4 | 2555 | * group/project quotaoff or both. |
1da177e4 LT |
2556 | */ |
2557 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) | |
2558 | log->l_quotaoffs_flag |= XFS_DQ_USER; | |
77a7cce4 NS |
2559 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
2560 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; | |
1da177e4 LT |
2561 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
2562 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; | |
2563 | ||
2564 | return (0); | |
2565 | } | |
2566 | ||
2567 | /* | |
2568 | * Recover a dquot record | |
2569 | */ | |
2570 | STATIC int | |
2571 | xlog_recover_do_dquot_trans( | |
2572 | xlog_t *log, | |
2573 | xlog_recover_item_t *item, | |
2574 | int pass) | |
2575 | { | |
2576 | xfs_mount_t *mp; | |
2577 | xfs_buf_t *bp; | |
2578 | struct xfs_disk_dquot *ddq, *recddq; | |
2579 | int error; | |
2580 | xfs_dq_logformat_t *dq_f; | |
2581 | uint type; | |
2582 | ||
2583 | if (pass == XLOG_RECOVER_PASS1) { | |
2584 | return 0; | |
2585 | } | |
2586 | mp = log->l_mp; | |
2587 | ||
2588 | /* | |
2589 | * Filesystems are required to send in quota flags at mount time. | |
2590 | */ | |
2591 | if (mp->m_qflags == 0) | |
2592 | return (0); | |
2593 | ||
2594 | recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr; | |
0c5e1ce8 CH |
2595 | |
2596 | if (item->ri_buf[1].i_addr == NULL) { | |
2597 | cmn_err(CE_ALERT, | |
2598 | "XFS: NULL dquot in %s.", __func__); | |
2599 | return XFS_ERROR(EIO); | |
2600 | } | |
8ec6dba2 | 2601 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
0c5e1ce8 CH |
2602 | cmn_err(CE_ALERT, |
2603 | "XFS: dquot too small (%d) in %s.", | |
2604 | item->ri_buf[1].i_len, __func__); | |
2605 | return XFS_ERROR(EIO); | |
2606 | } | |
2607 | ||
1da177e4 LT |
2608 | /* |
2609 | * This type of quotas was turned off, so ignore this record. | |
2610 | */ | |
b53e675d | 2611 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
1da177e4 LT |
2612 | ASSERT(type); |
2613 | if (log->l_quotaoffs_flag & type) | |
2614 | return (0); | |
2615 | ||
2616 | /* | |
2617 | * At this point we know that quota was _not_ turned off. | |
2618 | * Since the mount flags are not indicating to us otherwise, this | |
2619 | * must mean that quota is on, and the dquot needs to be replayed. | |
2620 | * Remember that we may not have fully recovered the superblock yet, | |
2621 | * so we can't do the usual trick of looking at the SB quota bits. | |
2622 | * | |
2623 | * The other possibility, of course, is that the quota subsystem was | |
2624 | * removed since the last mount - ENOSYS. | |
2625 | */ | |
2626 | dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr; | |
2627 | ASSERT(dq_f); | |
2628 | if ((error = xfs_qm_dqcheck(recddq, | |
2629 | dq_f->qlf_id, | |
2630 | 0, XFS_QMOPT_DOWARN, | |
2631 | "xlog_recover_do_dquot_trans (log copy)"))) { | |
2632 | return XFS_ERROR(EIO); | |
2633 | } | |
2634 | ASSERT(dq_f->qlf_len == 1); | |
2635 | ||
2636 | error = xfs_read_buf(mp, mp->m_ddev_targp, | |
2637 | dq_f->qlf_blkno, | |
2638 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), | |
2639 | 0, &bp); | |
2640 | if (error) { | |
2641 | xfs_ioerror_alert("xlog_recover_do..(read#3)", mp, | |
2642 | bp, dq_f->qlf_blkno); | |
2643 | return error; | |
2644 | } | |
2645 | ASSERT(bp); | |
2646 | ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset); | |
2647 | ||
2648 | /* | |
2649 | * At least the magic num portion should be on disk because this | |
2650 | * was among a chunk of dquots created earlier, and we did some | |
2651 | * minimal initialization then. | |
2652 | */ | |
2653 | if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, | |
2654 | "xlog_recover_do_dquot_trans")) { | |
2655 | xfs_buf_relse(bp); | |
2656 | return XFS_ERROR(EIO); | |
2657 | } | |
2658 | ||
2659 | memcpy(ddq, recddq, item->ri_buf[1].i_len); | |
2660 | ||
2661 | ASSERT(dq_f->qlf_size == 2); | |
15ac08a8 CH |
2662 | ASSERT(bp->b_mount == NULL || bp->b_mount == mp); |
2663 | bp->b_mount = mp; | |
1da177e4 LT |
2664 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone); |
2665 | xfs_bdwrite(mp, bp); | |
2666 | ||
2667 | return (0); | |
2668 | } | |
2669 | ||
2670 | /* | |
2671 | * This routine is called to create an in-core extent free intent | |
2672 | * item from the efi format structure which was logged on disk. | |
2673 | * It allocates an in-core efi, copies the extents from the format | |
2674 | * structure into it, and adds the efi to the AIL with the given | |
2675 | * LSN. | |
2676 | */ | |
6d192a9b | 2677 | STATIC int |
1da177e4 LT |
2678 | xlog_recover_do_efi_trans( |
2679 | xlog_t *log, | |
2680 | xlog_recover_item_t *item, | |
2681 | xfs_lsn_t lsn, | |
2682 | int pass) | |
2683 | { | |
6d192a9b | 2684 | int error; |
1da177e4 LT |
2685 | xfs_mount_t *mp; |
2686 | xfs_efi_log_item_t *efip; | |
2687 | xfs_efi_log_format_t *efi_formatp; | |
1da177e4 LT |
2688 | |
2689 | if (pass == XLOG_RECOVER_PASS1) { | |
6d192a9b | 2690 | return 0; |
1da177e4 LT |
2691 | } |
2692 | ||
2693 | efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr; | |
1da177e4 LT |
2694 | |
2695 | mp = log->l_mp; | |
2696 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); | |
6d192a9b TS |
2697 | if ((error = xfs_efi_copy_format(&(item->ri_buf[0]), |
2698 | &(efip->efi_format)))) { | |
2699 | xfs_efi_item_free(efip); | |
2700 | return error; | |
2701 | } | |
1da177e4 LT |
2702 | efip->efi_next_extent = efi_formatp->efi_nextents; |
2703 | efip->efi_flags |= XFS_EFI_COMMITTED; | |
2704 | ||
a9c21c1b | 2705 | spin_lock(&log->l_ailp->xa_lock); |
1da177e4 | 2706 | /* |
783a2f65 | 2707 | * xfs_trans_ail_update() drops the AIL lock. |
1da177e4 | 2708 | */ |
783a2f65 | 2709 | xfs_trans_ail_update(log->l_ailp, (xfs_log_item_t *)efip, lsn); |
6d192a9b | 2710 | return 0; |
1da177e4 LT |
2711 | } |
2712 | ||
2713 | ||
2714 | /* | |
2715 | * This routine is called when an efd format structure is found in | |
2716 | * a committed transaction in the log. It's purpose is to cancel | |
2717 | * the corresponding efi if it was still in the log. To do this | |
2718 | * it searches the AIL for the efi with an id equal to that in the | |
2719 | * efd format structure. If we find it, we remove the efi from the | |
2720 | * AIL and free it. | |
2721 | */ | |
2722 | STATIC void | |
2723 | xlog_recover_do_efd_trans( | |
2724 | xlog_t *log, | |
2725 | xlog_recover_item_t *item, | |
2726 | int pass) | |
2727 | { | |
1da177e4 LT |
2728 | xfs_efd_log_format_t *efd_formatp; |
2729 | xfs_efi_log_item_t *efip = NULL; | |
2730 | xfs_log_item_t *lip; | |
1da177e4 | 2731 | __uint64_t efi_id; |
27d8d5fe | 2732 | struct xfs_ail_cursor cur; |
783a2f65 | 2733 | struct xfs_ail *ailp = log->l_ailp; |
1da177e4 LT |
2734 | |
2735 | if (pass == XLOG_RECOVER_PASS1) { | |
2736 | return; | |
2737 | } | |
2738 | ||
2739 | efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr; | |
6d192a9b TS |
2740 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
2741 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || | |
2742 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + | |
2743 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); | |
1da177e4 LT |
2744 | efi_id = efd_formatp->efd_efi_id; |
2745 | ||
2746 | /* | |
2747 | * Search for the efi with the id in the efd format structure | |
2748 | * in the AIL. | |
2749 | */ | |
a9c21c1b DC |
2750 | spin_lock(&ailp->xa_lock); |
2751 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
2752 | while (lip != NULL) { |
2753 | if (lip->li_type == XFS_LI_EFI) { | |
2754 | efip = (xfs_efi_log_item_t *)lip; | |
2755 | if (efip->efi_format.efi_id == efi_id) { | |
2756 | /* | |
783a2f65 | 2757 | * xfs_trans_ail_delete() drops the |
1da177e4 LT |
2758 | * AIL lock. |
2759 | */ | |
783a2f65 | 2760 | xfs_trans_ail_delete(ailp, lip); |
8ae2c0f6 | 2761 | xfs_efi_item_free(efip); |
a9c21c1b | 2762 | spin_lock(&ailp->xa_lock); |
27d8d5fe | 2763 | break; |
1da177e4 LT |
2764 | } |
2765 | } | |
a9c21c1b | 2766 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 2767 | } |
a9c21c1b DC |
2768 | xfs_trans_ail_cursor_done(ailp, &cur); |
2769 | spin_unlock(&ailp->xa_lock); | |
1da177e4 LT |
2770 | } |
2771 | ||
2772 | /* | |
2773 | * Perform the transaction | |
2774 | * | |
2775 | * If the transaction modifies a buffer or inode, do it now. Otherwise, | |
2776 | * EFIs and EFDs get queued up by adding entries into the AIL for them. | |
2777 | */ | |
2778 | STATIC int | |
2779 | xlog_recover_do_trans( | |
2780 | xlog_t *log, | |
2781 | xlog_recover_t *trans, | |
2782 | int pass) | |
2783 | { | |
2784 | int error = 0; | |
f0a76953 | 2785 | xlog_recover_item_t *item; |
1da177e4 | 2786 | |
9abbc539 | 2787 | error = xlog_recover_reorder_trans(log, trans, pass); |
ff0205e0 | 2788 | if (error) |
1da177e4 | 2789 | return error; |
ff0205e0 | 2790 | |
f0a76953 | 2791 | list_for_each_entry(item, &trans->r_itemq, ri_list) { |
9abbc539 | 2792 | trace_xfs_log_recover_item_recover(log, trans, item, pass); |
ff0205e0 CH |
2793 | switch (ITEM_TYPE(item)) { |
2794 | case XFS_LI_BUF: | |
2795 | error = xlog_recover_do_buffer_trans(log, item, pass); | |
2796 | break; | |
2797 | case XFS_LI_INODE: | |
2798 | error = xlog_recover_do_inode_trans(log, item, pass); | |
2799 | break; | |
2800 | case XFS_LI_EFI: | |
2801 | error = xlog_recover_do_efi_trans(log, item, | |
2802 | trans->r_lsn, pass); | |
2803 | break; | |
2804 | case XFS_LI_EFD: | |
1da177e4 | 2805 | xlog_recover_do_efd_trans(log, item, pass); |
ff0205e0 CH |
2806 | error = 0; |
2807 | break; | |
2808 | case XFS_LI_DQUOT: | |
2809 | error = xlog_recover_do_dquot_trans(log, item, pass); | |
2810 | break; | |
2811 | case XFS_LI_QUOTAOFF: | |
2812 | error = xlog_recover_do_quotaoff_trans(log, item, | |
2813 | pass); | |
2814 | break; | |
2815 | default: | |
2816 | xlog_warn( | |
2817 | "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item)); | |
1da177e4 LT |
2818 | ASSERT(0); |
2819 | error = XFS_ERROR(EIO); | |
2820 | break; | |
2821 | } | |
ff0205e0 CH |
2822 | |
2823 | if (error) | |
2824 | return error; | |
f0a76953 | 2825 | } |
1da177e4 | 2826 | |
ff0205e0 | 2827 | return 0; |
1da177e4 LT |
2828 | } |
2829 | ||
2830 | /* | |
2831 | * Free up any resources allocated by the transaction | |
2832 | * | |
2833 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. | |
2834 | */ | |
2835 | STATIC void | |
2836 | xlog_recover_free_trans( | |
2837 | xlog_recover_t *trans) | |
2838 | { | |
f0a76953 | 2839 | xlog_recover_item_t *item, *n; |
1da177e4 LT |
2840 | int i; |
2841 | ||
f0a76953 DC |
2842 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { |
2843 | /* Free the regions in the item. */ | |
2844 | list_del(&item->ri_list); | |
2845 | for (i = 0; i < item->ri_cnt; i++) | |
2846 | kmem_free(item->ri_buf[i].i_addr); | |
1da177e4 | 2847 | /* Free the item itself */ |
f0a76953 DC |
2848 | kmem_free(item->ri_buf); |
2849 | kmem_free(item); | |
2850 | } | |
1da177e4 | 2851 | /* Free the transaction recover structure */ |
f0e2d93c | 2852 | kmem_free(trans); |
1da177e4 LT |
2853 | } |
2854 | ||
2855 | STATIC int | |
2856 | xlog_recover_commit_trans( | |
2857 | xlog_t *log, | |
1da177e4 LT |
2858 | xlog_recover_t *trans, |
2859 | int pass) | |
2860 | { | |
2861 | int error; | |
2862 | ||
f0a76953 | 2863 | hlist_del(&trans->r_list); |
1da177e4 LT |
2864 | if ((error = xlog_recover_do_trans(log, trans, pass))) |
2865 | return error; | |
2866 | xlog_recover_free_trans(trans); /* no error */ | |
2867 | return 0; | |
2868 | } | |
2869 | ||
2870 | STATIC int | |
2871 | xlog_recover_unmount_trans( | |
2872 | xlog_recover_t *trans) | |
2873 | { | |
2874 | /* Do nothing now */ | |
2875 | xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR"); | |
2876 | return 0; | |
2877 | } | |
2878 | ||
2879 | /* | |
2880 | * There are two valid states of the r_state field. 0 indicates that the | |
2881 | * transaction structure is in a normal state. We have either seen the | |
2882 | * start of the transaction or the last operation we added was not a partial | |
2883 | * operation. If the last operation we added to the transaction was a | |
2884 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. | |
2885 | * | |
2886 | * NOTE: skip LRs with 0 data length. | |
2887 | */ | |
2888 | STATIC int | |
2889 | xlog_recover_process_data( | |
2890 | xlog_t *log, | |
f0a76953 | 2891 | struct hlist_head rhash[], |
1da177e4 LT |
2892 | xlog_rec_header_t *rhead, |
2893 | xfs_caddr_t dp, | |
2894 | int pass) | |
2895 | { | |
2896 | xfs_caddr_t lp; | |
2897 | int num_logops; | |
2898 | xlog_op_header_t *ohead; | |
2899 | xlog_recover_t *trans; | |
2900 | xlog_tid_t tid; | |
2901 | int error; | |
2902 | unsigned long hash; | |
2903 | uint flags; | |
2904 | ||
b53e675d CH |
2905 | lp = dp + be32_to_cpu(rhead->h_len); |
2906 | num_logops = be32_to_cpu(rhead->h_num_logops); | |
1da177e4 LT |
2907 | |
2908 | /* check the log format matches our own - else we can't recover */ | |
2909 | if (xlog_header_check_recover(log->l_mp, rhead)) | |
2910 | return (XFS_ERROR(EIO)); | |
2911 | ||
2912 | while ((dp < lp) && num_logops) { | |
2913 | ASSERT(dp + sizeof(xlog_op_header_t) <= lp); | |
2914 | ohead = (xlog_op_header_t *)dp; | |
2915 | dp += sizeof(xlog_op_header_t); | |
2916 | if (ohead->oh_clientid != XFS_TRANSACTION && | |
2917 | ohead->oh_clientid != XFS_LOG) { | |
2918 | xlog_warn( | |
2919 | "XFS: xlog_recover_process_data: bad clientid"); | |
2920 | ASSERT(0); | |
2921 | return (XFS_ERROR(EIO)); | |
2922 | } | |
67fcb7bf | 2923 | tid = be32_to_cpu(ohead->oh_tid); |
1da177e4 | 2924 | hash = XLOG_RHASH(tid); |
f0a76953 | 2925 | trans = xlog_recover_find_tid(&rhash[hash], tid); |
1da177e4 LT |
2926 | if (trans == NULL) { /* not found; add new tid */ |
2927 | if (ohead->oh_flags & XLOG_START_TRANS) | |
2928 | xlog_recover_new_tid(&rhash[hash], tid, | |
b53e675d | 2929 | be64_to_cpu(rhead->h_lsn)); |
1da177e4 | 2930 | } else { |
9742bb93 LM |
2931 | if (dp + be32_to_cpu(ohead->oh_len) > lp) { |
2932 | xlog_warn( | |
2933 | "XFS: xlog_recover_process_data: bad length"); | |
2934 | WARN_ON(1); | |
2935 | return (XFS_ERROR(EIO)); | |
2936 | } | |
1da177e4 LT |
2937 | flags = ohead->oh_flags & ~XLOG_END_TRANS; |
2938 | if (flags & XLOG_WAS_CONT_TRANS) | |
2939 | flags &= ~XLOG_CONTINUE_TRANS; | |
2940 | switch (flags) { | |
2941 | case XLOG_COMMIT_TRANS: | |
2942 | error = xlog_recover_commit_trans(log, | |
f0a76953 | 2943 | trans, pass); |
1da177e4 LT |
2944 | break; |
2945 | case XLOG_UNMOUNT_TRANS: | |
2946 | error = xlog_recover_unmount_trans(trans); | |
2947 | break; | |
2948 | case XLOG_WAS_CONT_TRANS: | |
9abbc539 DC |
2949 | error = xlog_recover_add_to_cont_trans(log, |
2950 | trans, dp, | |
2951 | be32_to_cpu(ohead->oh_len)); | |
1da177e4 LT |
2952 | break; |
2953 | case XLOG_START_TRANS: | |
2954 | xlog_warn( | |
2955 | "XFS: xlog_recover_process_data: bad transaction"); | |
2956 | ASSERT(0); | |
2957 | error = XFS_ERROR(EIO); | |
2958 | break; | |
2959 | case 0: | |
2960 | case XLOG_CONTINUE_TRANS: | |
9abbc539 | 2961 | error = xlog_recover_add_to_trans(log, trans, |
67fcb7bf | 2962 | dp, be32_to_cpu(ohead->oh_len)); |
1da177e4 LT |
2963 | break; |
2964 | default: | |
2965 | xlog_warn( | |
2966 | "XFS: xlog_recover_process_data: bad flag"); | |
2967 | ASSERT(0); | |
2968 | error = XFS_ERROR(EIO); | |
2969 | break; | |
2970 | } | |
2971 | if (error) | |
2972 | return error; | |
2973 | } | |
67fcb7bf | 2974 | dp += be32_to_cpu(ohead->oh_len); |
1da177e4 LT |
2975 | num_logops--; |
2976 | } | |
2977 | return 0; | |
2978 | } | |
2979 | ||
2980 | /* | |
2981 | * Process an extent free intent item that was recovered from | |
2982 | * the log. We need to free the extents that it describes. | |
2983 | */ | |
3c1e2bbe | 2984 | STATIC int |
1da177e4 LT |
2985 | xlog_recover_process_efi( |
2986 | xfs_mount_t *mp, | |
2987 | xfs_efi_log_item_t *efip) | |
2988 | { | |
2989 | xfs_efd_log_item_t *efdp; | |
2990 | xfs_trans_t *tp; | |
2991 | int i; | |
3c1e2bbe | 2992 | int error = 0; |
1da177e4 LT |
2993 | xfs_extent_t *extp; |
2994 | xfs_fsblock_t startblock_fsb; | |
2995 | ||
2996 | ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED)); | |
2997 | ||
2998 | /* | |
2999 | * First check the validity of the extents described by the | |
3000 | * EFI. If any are bad, then assume that all are bad and | |
3001 | * just toss the EFI. | |
3002 | */ | |
3003 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3004 | extp = &(efip->efi_format.efi_extents[i]); | |
3005 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
3006 | XFS_FSB_TO_DADDR(mp, extp->ext_start)); | |
3007 | if ((startblock_fsb == 0) || | |
3008 | (extp->ext_len == 0) || | |
3009 | (startblock_fsb >= mp->m_sb.sb_dblocks) || | |
3010 | (extp->ext_len >= mp->m_sb.sb_agblocks)) { | |
3011 | /* | |
3012 | * This will pull the EFI from the AIL and | |
3013 | * free the memory associated with it. | |
3014 | */ | |
3015 | xfs_efi_release(efip, efip->efi_format.efi_nextents); | |
3c1e2bbe | 3016 | return XFS_ERROR(EIO); |
1da177e4 LT |
3017 | } |
3018 | } | |
3019 | ||
3020 | tp = xfs_trans_alloc(mp, 0); | |
3c1e2bbe | 3021 | error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0); |
fc6149d8 DC |
3022 | if (error) |
3023 | goto abort_error; | |
1da177e4 LT |
3024 | efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents); |
3025 | ||
3026 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3027 | extp = &(efip->efi_format.efi_extents[i]); | |
fc6149d8 DC |
3028 | error = xfs_free_extent(tp, extp->ext_start, extp->ext_len); |
3029 | if (error) | |
3030 | goto abort_error; | |
1da177e4 LT |
3031 | xfs_trans_log_efd_extent(tp, efdp, extp->ext_start, |
3032 | extp->ext_len); | |
3033 | } | |
3034 | ||
3035 | efip->efi_flags |= XFS_EFI_RECOVERED; | |
e5720eec | 3036 | error = xfs_trans_commit(tp, 0); |
3c1e2bbe | 3037 | return error; |
fc6149d8 DC |
3038 | |
3039 | abort_error: | |
3040 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3041 | return error; | |
1da177e4 LT |
3042 | } |
3043 | ||
1da177e4 LT |
3044 | /* |
3045 | * When this is called, all of the EFIs which did not have | |
3046 | * corresponding EFDs should be in the AIL. What we do now | |
3047 | * is free the extents associated with each one. | |
3048 | * | |
3049 | * Since we process the EFIs in normal transactions, they | |
3050 | * will be removed at some point after the commit. This prevents | |
3051 | * us from just walking down the list processing each one. | |
3052 | * We'll use a flag in the EFI to skip those that we've already | |
3053 | * processed and use the AIL iteration mechanism's generation | |
3054 | * count to try to speed this up at least a bit. | |
3055 | * | |
3056 | * When we start, we know that the EFIs are the only things in | |
3057 | * the AIL. As we process them, however, other items are added | |
3058 | * to the AIL. Since everything added to the AIL must come after | |
3059 | * everything already in the AIL, we stop processing as soon as | |
3060 | * we see something other than an EFI in the AIL. | |
3061 | */ | |
3c1e2bbe | 3062 | STATIC int |
1da177e4 LT |
3063 | xlog_recover_process_efis( |
3064 | xlog_t *log) | |
3065 | { | |
3066 | xfs_log_item_t *lip; | |
3067 | xfs_efi_log_item_t *efip; | |
3c1e2bbe | 3068 | int error = 0; |
27d8d5fe | 3069 | struct xfs_ail_cursor cur; |
a9c21c1b | 3070 | struct xfs_ail *ailp; |
1da177e4 | 3071 | |
a9c21c1b DC |
3072 | ailp = log->l_ailp; |
3073 | spin_lock(&ailp->xa_lock); | |
3074 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
3075 | while (lip != NULL) { |
3076 | /* | |
3077 | * We're done when we see something other than an EFI. | |
27d8d5fe | 3078 | * There should be no EFIs left in the AIL now. |
1da177e4 LT |
3079 | */ |
3080 | if (lip->li_type != XFS_LI_EFI) { | |
27d8d5fe | 3081 | #ifdef DEBUG |
a9c21c1b | 3082 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
27d8d5fe DC |
3083 | ASSERT(lip->li_type != XFS_LI_EFI); |
3084 | #endif | |
1da177e4 LT |
3085 | break; |
3086 | } | |
3087 | ||
3088 | /* | |
3089 | * Skip EFIs that we've already processed. | |
3090 | */ | |
3091 | efip = (xfs_efi_log_item_t *)lip; | |
3092 | if (efip->efi_flags & XFS_EFI_RECOVERED) { | |
a9c21c1b | 3093 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 LT |
3094 | continue; |
3095 | } | |
3096 | ||
a9c21c1b DC |
3097 | spin_unlock(&ailp->xa_lock); |
3098 | error = xlog_recover_process_efi(log->l_mp, efip); | |
3099 | spin_lock(&ailp->xa_lock); | |
27d8d5fe DC |
3100 | if (error) |
3101 | goto out; | |
a9c21c1b | 3102 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 3103 | } |
27d8d5fe | 3104 | out: |
a9c21c1b DC |
3105 | xfs_trans_ail_cursor_done(ailp, &cur); |
3106 | spin_unlock(&ailp->xa_lock); | |
3c1e2bbe | 3107 | return error; |
1da177e4 LT |
3108 | } |
3109 | ||
3110 | /* | |
3111 | * This routine performs a transaction to null out a bad inode pointer | |
3112 | * in an agi unlinked inode hash bucket. | |
3113 | */ | |
3114 | STATIC void | |
3115 | xlog_recover_clear_agi_bucket( | |
3116 | xfs_mount_t *mp, | |
3117 | xfs_agnumber_t agno, | |
3118 | int bucket) | |
3119 | { | |
3120 | xfs_trans_t *tp; | |
3121 | xfs_agi_t *agi; | |
3122 | xfs_buf_t *agibp; | |
3123 | int offset; | |
3124 | int error; | |
3125 | ||
3126 | tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET); | |
5e1be0fb CH |
3127 | error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), |
3128 | 0, 0, 0); | |
e5720eec DC |
3129 | if (error) |
3130 | goto out_abort; | |
1da177e4 | 3131 | |
5e1be0fb CH |
3132 | error = xfs_read_agi(mp, tp, agno, &agibp); |
3133 | if (error) | |
e5720eec | 3134 | goto out_abort; |
1da177e4 | 3135 | |
5e1be0fb | 3136 | agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 3137 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
3138 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
3139 | (sizeof(xfs_agino_t) * bucket); | |
3140 | xfs_trans_log_buf(tp, agibp, offset, | |
3141 | (offset + sizeof(xfs_agino_t) - 1)); | |
3142 | ||
e5720eec DC |
3143 | error = xfs_trans_commit(tp, 0); |
3144 | if (error) | |
3145 | goto out_error; | |
3146 | return; | |
3147 | ||
3148 | out_abort: | |
3149 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3150 | out_error: | |
3151 | xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: " | |
3152 | "failed to clear agi %d. Continuing.", agno); | |
3153 | return; | |
1da177e4 LT |
3154 | } |
3155 | ||
23fac50f CH |
3156 | STATIC xfs_agino_t |
3157 | xlog_recover_process_one_iunlink( | |
3158 | struct xfs_mount *mp, | |
3159 | xfs_agnumber_t agno, | |
3160 | xfs_agino_t agino, | |
3161 | int bucket) | |
3162 | { | |
3163 | struct xfs_buf *ibp; | |
3164 | struct xfs_dinode *dip; | |
3165 | struct xfs_inode *ip; | |
3166 | xfs_ino_t ino; | |
3167 | int error; | |
3168 | ||
3169 | ino = XFS_AGINO_TO_INO(mp, agno, agino); | |
3170 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0); | |
3171 | if (error) | |
3172 | goto fail; | |
3173 | ||
3174 | /* | |
3175 | * Get the on disk inode to find the next inode in the bucket. | |
3176 | */ | |
0cadda1c | 3177 | error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK); |
23fac50f | 3178 | if (error) |
0e446673 | 3179 | goto fail_iput; |
23fac50f | 3180 | |
23fac50f | 3181 | ASSERT(ip->i_d.di_nlink == 0); |
0e446673 | 3182 | ASSERT(ip->i_d.di_mode != 0); |
23fac50f CH |
3183 | |
3184 | /* setup for the next pass */ | |
3185 | agino = be32_to_cpu(dip->di_next_unlinked); | |
3186 | xfs_buf_relse(ibp); | |
3187 | ||
3188 | /* | |
3189 | * Prevent any DMAPI event from being sent when the reference on | |
3190 | * the inode is dropped. | |
3191 | */ | |
3192 | ip->i_d.di_dmevmask = 0; | |
3193 | ||
0e446673 | 3194 | IRELE(ip); |
23fac50f CH |
3195 | return agino; |
3196 | ||
0e446673 CH |
3197 | fail_iput: |
3198 | IRELE(ip); | |
23fac50f CH |
3199 | fail: |
3200 | /* | |
3201 | * We can't read in the inode this bucket points to, or this inode | |
3202 | * is messed up. Just ditch this bucket of inodes. We will lose | |
3203 | * some inodes and space, but at least we won't hang. | |
3204 | * | |
3205 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to | |
3206 | * clear the inode pointer in the bucket. | |
3207 | */ | |
3208 | xlog_recover_clear_agi_bucket(mp, agno, bucket); | |
3209 | return NULLAGINO; | |
3210 | } | |
3211 | ||
1da177e4 LT |
3212 | /* |
3213 | * xlog_iunlink_recover | |
3214 | * | |
3215 | * This is called during recovery to process any inodes which | |
3216 | * we unlinked but not freed when the system crashed. These | |
3217 | * inodes will be on the lists in the AGI blocks. What we do | |
3218 | * here is scan all the AGIs and fully truncate and free any | |
3219 | * inodes found on the lists. Each inode is removed from the | |
3220 | * lists when it has been fully truncated and is freed. The | |
3221 | * freeing of the inode and its removal from the list must be | |
3222 | * atomic. | |
3223 | */ | |
d96f8f89 | 3224 | STATIC void |
1da177e4 LT |
3225 | xlog_recover_process_iunlinks( |
3226 | xlog_t *log) | |
3227 | { | |
3228 | xfs_mount_t *mp; | |
3229 | xfs_agnumber_t agno; | |
3230 | xfs_agi_t *agi; | |
3231 | xfs_buf_t *agibp; | |
1da177e4 | 3232 | xfs_agino_t agino; |
1da177e4 LT |
3233 | int bucket; |
3234 | int error; | |
3235 | uint mp_dmevmask; | |
3236 | ||
3237 | mp = log->l_mp; | |
3238 | ||
3239 | /* | |
3240 | * Prevent any DMAPI event from being sent while in this function. | |
3241 | */ | |
3242 | mp_dmevmask = mp->m_dmevmask; | |
3243 | mp->m_dmevmask = 0; | |
3244 | ||
3245 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
3246 | /* | |
3247 | * Find the agi for this ag. | |
3248 | */ | |
5e1be0fb CH |
3249 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
3250 | if (error) { | |
3251 | /* | |
3252 | * AGI is b0rked. Don't process it. | |
3253 | * | |
3254 | * We should probably mark the filesystem as corrupt | |
3255 | * after we've recovered all the ag's we can.... | |
3256 | */ | |
3257 | continue; | |
1da177e4 LT |
3258 | } |
3259 | agi = XFS_BUF_TO_AGI(agibp); | |
1da177e4 LT |
3260 | |
3261 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { | |
16259e7d | 3262 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
1da177e4 | 3263 | while (agino != NULLAGINO) { |
1da177e4 LT |
3264 | /* |
3265 | * Release the agi buffer so that it can | |
3266 | * be acquired in the normal course of the | |
3267 | * transaction to truncate and free the inode. | |
3268 | */ | |
3269 | xfs_buf_relse(agibp); | |
3270 | ||
23fac50f CH |
3271 | agino = xlog_recover_process_one_iunlink(mp, |
3272 | agno, agino, bucket); | |
1da177e4 LT |
3273 | |
3274 | /* | |
3275 | * Reacquire the agibuffer and continue around | |
5e1be0fb CH |
3276 | * the loop. This should never fail as we know |
3277 | * the buffer was good earlier on. | |
1da177e4 | 3278 | */ |
5e1be0fb CH |
3279 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
3280 | ASSERT(error == 0); | |
1da177e4 | 3281 | agi = XFS_BUF_TO_AGI(agibp); |
1da177e4 LT |
3282 | } |
3283 | } | |
3284 | ||
3285 | /* | |
3286 | * Release the buffer for the current agi so we can | |
3287 | * go on to the next one. | |
3288 | */ | |
3289 | xfs_buf_relse(agibp); | |
3290 | } | |
3291 | ||
3292 | mp->m_dmevmask = mp_dmevmask; | |
3293 | } | |
3294 | ||
3295 | ||
3296 | #ifdef DEBUG | |
3297 | STATIC void | |
3298 | xlog_pack_data_checksum( | |
3299 | xlog_t *log, | |
3300 | xlog_in_core_t *iclog, | |
3301 | int size) | |
3302 | { | |
3303 | int i; | |
b53e675d | 3304 | __be32 *up; |
1da177e4 LT |
3305 | uint chksum = 0; |
3306 | ||
b53e675d | 3307 | up = (__be32 *)iclog->ic_datap; |
1da177e4 LT |
3308 | /* divide length by 4 to get # words */ |
3309 | for (i = 0; i < (size >> 2); i++) { | |
b53e675d | 3310 | chksum ^= be32_to_cpu(*up); |
1da177e4 LT |
3311 | up++; |
3312 | } | |
b53e675d | 3313 | iclog->ic_header.h_chksum = cpu_to_be32(chksum); |
1da177e4 LT |
3314 | } |
3315 | #else | |
3316 | #define xlog_pack_data_checksum(log, iclog, size) | |
3317 | #endif | |
3318 | ||
3319 | /* | |
3320 | * Stamp cycle number in every block | |
3321 | */ | |
3322 | void | |
3323 | xlog_pack_data( | |
3324 | xlog_t *log, | |
3325 | xlog_in_core_t *iclog, | |
3326 | int roundoff) | |
3327 | { | |
3328 | int i, j, k; | |
3329 | int size = iclog->ic_offset + roundoff; | |
b53e675d | 3330 | __be32 cycle_lsn; |
1da177e4 | 3331 | xfs_caddr_t dp; |
1da177e4 LT |
3332 | |
3333 | xlog_pack_data_checksum(log, iclog, size); | |
3334 | ||
3335 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
3336 | ||
3337 | dp = iclog->ic_datap; | |
3338 | for (i = 0; i < BTOBB(size) && | |
3339 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { | |
b53e675d CH |
3340 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; |
3341 | *(__be32 *)dp = cycle_lsn; | |
1da177e4 LT |
3342 | dp += BBSIZE; |
3343 | } | |
3344 | ||
62118709 | 3345 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b28708d6 CH |
3346 | xlog_in_core_2_t *xhdr = iclog->ic_data; |
3347 | ||
1da177e4 LT |
3348 | for ( ; i < BTOBB(size); i++) { |
3349 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3350 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d CH |
3351 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; |
3352 | *(__be32 *)dp = cycle_lsn; | |
1da177e4 LT |
3353 | dp += BBSIZE; |
3354 | } | |
3355 | ||
3356 | for (i = 1; i < log->l_iclog_heads; i++) { | |
3357 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
3358 | } | |
3359 | } | |
3360 | } | |
3361 | ||
3362 | #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY) | |
3363 | STATIC void | |
3364 | xlog_unpack_data_checksum( | |
3365 | xlog_rec_header_t *rhead, | |
3366 | xfs_caddr_t dp, | |
3367 | xlog_t *log) | |
3368 | { | |
b53e675d | 3369 | __be32 *up = (__be32 *)dp; |
1da177e4 LT |
3370 | uint chksum = 0; |
3371 | int i; | |
3372 | ||
3373 | /* divide length by 4 to get # words */ | |
b53e675d CH |
3374 | for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) { |
3375 | chksum ^= be32_to_cpu(*up); | |
1da177e4 LT |
3376 | up++; |
3377 | } | |
b53e675d | 3378 | if (chksum != be32_to_cpu(rhead->h_chksum)) { |
1da177e4 LT |
3379 | if (rhead->h_chksum || |
3380 | ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) { | |
3381 | cmn_err(CE_DEBUG, | |
b6574520 | 3382 | "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n", |
b53e675d | 3383 | be32_to_cpu(rhead->h_chksum), chksum); |
1da177e4 LT |
3384 | cmn_err(CE_DEBUG, |
3385 | "XFS: Disregard message if filesystem was created with non-DEBUG kernel"); | |
62118709 | 3386 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
1da177e4 | 3387 | cmn_err(CE_DEBUG, |
b6574520 | 3388 | "XFS: LogR this is a LogV2 filesystem\n"); |
1da177e4 LT |
3389 | } |
3390 | log->l_flags |= XLOG_CHKSUM_MISMATCH; | |
3391 | } | |
3392 | } | |
3393 | } | |
3394 | #else | |
3395 | #define xlog_unpack_data_checksum(rhead, dp, log) | |
3396 | #endif | |
3397 | ||
3398 | STATIC void | |
3399 | xlog_unpack_data( | |
3400 | xlog_rec_header_t *rhead, | |
3401 | xfs_caddr_t dp, | |
3402 | xlog_t *log) | |
3403 | { | |
3404 | int i, j, k; | |
1da177e4 | 3405 | |
b53e675d | 3406 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
1da177e4 | 3407 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
b53e675d | 3408 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
1da177e4 LT |
3409 | dp += BBSIZE; |
3410 | } | |
3411 | ||
62118709 | 3412 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b28708d6 | 3413 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
b53e675d | 3414 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
1da177e4 LT |
3415 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
3416 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3417 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
1da177e4 LT |
3418 | dp += BBSIZE; |
3419 | } | |
3420 | } | |
3421 | ||
3422 | xlog_unpack_data_checksum(rhead, dp, log); | |
3423 | } | |
3424 | ||
3425 | STATIC int | |
3426 | xlog_valid_rec_header( | |
3427 | xlog_t *log, | |
3428 | xlog_rec_header_t *rhead, | |
3429 | xfs_daddr_t blkno) | |
3430 | { | |
3431 | int hlen; | |
3432 | ||
b53e675d | 3433 | if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) { |
1da177e4 LT |
3434 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
3435 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3436 | return XFS_ERROR(EFSCORRUPTED); | |
3437 | } | |
3438 | if (unlikely( | |
3439 | (!rhead->h_version || | |
b53e675d | 3440 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
1da177e4 | 3441 | xlog_warn("XFS: %s: unrecognised log version (%d).", |
34a622b2 | 3442 | __func__, be32_to_cpu(rhead->h_version)); |
1da177e4 LT |
3443 | return XFS_ERROR(EIO); |
3444 | } | |
3445 | ||
3446 | /* LR body must have data or it wouldn't have been written */ | |
b53e675d | 3447 | hlen = be32_to_cpu(rhead->h_len); |
1da177e4 LT |
3448 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
3449 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", | |
3450 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3451 | return XFS_ERROR(EFSCORRUPTED); | |
3452 | } | |
3453 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { | |
3454 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", | |
3455 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3456 | return XFS_ERROR(EFSCORRUPTED); | |
3457 | } | |
3458 | return 0; | |
3459 | } | |
3460 | ||
3461 | /* | |
3462 | * Read the log from tail to head and process the log records found. | |
3463 | * Handle the two cases where the tail and head are in the same cycle | |
3464 | * and where the active portion of the log wraps around the end of | |
3465 | * the physical log separately. The pass parameter is passed through | |
3466 | * to the routines called to process the data and is not looked at | |
3467 | * here. | |
3468 | */ | |
3469 | STATIC int | |
3470 | xlog_do_recovery_pass( | |
3471 | xlog_t *log, | |
3472 | xfs_daddr_t head_blk, | |
3473 | xfs_daddr_t tail_blk, | |
3474 | int pass) | |
3475 | { | |
3476 | xlog_rec_header_t *rhead; | |
3477 | xfs_daddr_t blk_no; | |
fc5bc4c8 | 3478 | xfs_caddr_t offset; |
1da177e4 LT |
3479 | xfs_buf_t *hbp, *dbp; |
3480 | int error = 0, h_size; | |
3481 | int bblks, split_bblks; | |
3482 | int hblks, split_hblks, wrapped_hblks; | |
f0a76953 | 3483 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
1da177e4 LT |
3484 | |
3485 | ASSERT(head_blk != tail_blk); | |
3486 | ||
3487 | /* | |
3488 | * Read the header of the tail block and get the iclog buffer size from | |
3489 | * h_size. Use this to tell how many sectors make up the log header. | |
3490 | */ | |
62118709 | 3491 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
1da177e4 LT |
3492 | /* |
3493 | * When using variable length iclogs, read first sector of | |
3494 | * iclog header and extract the header size from it. Get a | |
3495 | * new hbp that is the correct size. | |
3496 | */ | |
3497 | hbp = xlog_get_bp(log, 1); | |
3498 | if (!hbp) | |
3499 | return ENOMEM; | |
076e6acb CH |
3500 | |
3501 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); | |
3502 | if (error) | |
1da177e4 | 3503 | goto bread_err1; |
076e6acb | 3504 | |
1da177e4 LT |
3505 | rhead = (xlog_rec_header_t *)offset; |
3506 | error = xlog_valid_rec_header(log, rhead, tail_blk); | |
3507 | if (error) | |
3508 | goto bread_err1; | |
b53e675d CH |
3509 | h_size = be32_to_cpu(rhead->h_size); |
3510 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && | |
1da177e4 LT |
3511 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
3512 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
3513 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
3514 | hblks++; | |
3515 | xlog_put_bp(hbp); | |
3516 | hbp = xlog_get_bp(log, hblks); | |
3517 | } else { | |
3518 | hblks = 1; | |
3519 | } | |
3520 | } else { | |
3521 | ASSERT(log->l_sectbb_log == 0); | |
3522 | hblks = 1; | |
3523 | hbp = xlog_get_bp(log, 1); | |
3524 | h_size = XLOG_BIG_RECORD_BSIZE; | |
3525 | } | |
3526 | ||
3527 | if (!hbp) | |
3528 | return ENOMEM; | |
3529 | dbp = xlog_get_bp(log, BTOBB(h_size)); | |
3530 | if (!dbp) { | |
3531 | xlog_put_bp(hbp); | |
3532 | return ENOMEM; | |
3533 | } | |
3534 | ||
3535 | memset(rhash, 0, sizeof(rhash)); | |
3536 | if (tail_blk <= head_blk) { | |
3537 | for (blk_no = tail_blk; blk_no < head_blk; ) { | |
076e6acb CH |
3538 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
3539 | if (error) | |
1da177e4 | 3540 | goto bread_err2; |
076e6acb | 3541 | |
1da177e4 LT |
3542 | rhead = (xlog_rec_header_t *)offset; |
3543 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3544 | if (error) | |
3545 | goto bread_err2; | |
3546 | ||
3547 | /* blocks in data section */ | |
b53e675d | 3548 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
3549 | error = xlog_bread(log, blk_no + hblks, bblks, dbp, |
3550 | &offset); | |
1da177e4 LT |
3551 | if (error) |
3552 | goto bread_err2; | |
076e6acb | 3553 | |
1da177e4 LT |
3554 | xlog_unpack_data(rhead, offset, log); |
3555 | if ((error = xlog_recover_process_data(log, | |
3556 | rhash, rhead, offset, pass))) | |
3557 | goto bread_err2; | |
3558 | blk_no += bblks + hblks; | |
3559 | } | |
3560 | } else { | |
3561 | /* | |
3562 | * Perform recovery around the end of the physical log. | |
3563 | * When the head is not on the same cycle number as the tail, | |
3564 | * we can't do a sequential recovery as above. | |
3565 | */ | |
3566 | blk_no = tail_blk; | |
3567 | while (blk_no < log->l_logBBsize) { | |
3568 | /* | |
3569 | * Check for header wrapping around physical end-of-log | |
3570 | */ | |
fc5bc4c8 | 3571 | offset = XFS_BUF_PTR(hbp); |
1da177e4 LT |
3572 | split_hblks = 0; |
3573 | wrapped_hblks = 0; | |
3574 | if (blk_no + hblks <= log->l_logBBsize) { | |
3575 | /* Read header in one read */ | |
076e6acb CH |
3576 | error = xlog_bread(log, blk_no, hblks, hbp, |
3577 | &offset); | |
1da177e4 LT |
3578 | if (error) |
3579 | goto bread_err2; | |
1da177e4 LT |
3580 | } else { |
3581 | /* This LR is split across physical log end */ | |
3582 | if (blk_no != log->l_logBBsize) { | |
3583 | /* some data before physical log end */ | |
3584 | ASSERT(blk_no <= INT_MAX); | |
3585 | split_hblks = log->l_logBBsize - (int)blk_no; | |
3586 | ASSERT(split_hblks > 0); | |
076e6acb CH |
3587 | error = xlog_bread(log, blk_no, |
3588 | split_hblks, hbp, | |
3589 | &offset); | |
3590 | if (error) | |
1da177e4 | 3591 | goto bread_err2; |
1da177e4 | 3592 | } |
076e6acb | 3593 | |
1da177e4 LT |
3594 | /* |
3595 | * Note: this black magic still works with | |
3596 | * large sector sizes (non-512) only because: | |
3597 | * - we increased the buffer size originally | |
3598 | * by 1 sector giving us enough extra space | |
3599 | * for the second read; | |
3600 | * - the log start is guaranteed to be sector | |
3601 | * aligned; | |
3602 | * - we read the log end (LR header start) | |
3603 | * _first_, then the log start (LR header end) | |
3604 | * - order is important. | |
3605 | */ | |
234f56ac | 3606 | wrapped_hblks = hblks - split_hblks; |
234f56ac | 3607 | error = XFS_BUF_SET_PTR(hbp, |
fc5bc4c8 | 3608 | offset + BBTOB(split_hblks), |
1da177e4 | 3609 | BBTOB(hblks - split_hblks)); |
076e6acb CH |
3610 | if (error) |
3611 | goto bread_err2; | |
3612 | ||
3613 | error = xlog_bread_noalign(log, 0, | |
3614 | wrapped_hblks, hbp); | |
3615 | if (error) | |
3616 | goto bread_err2; | |
3617 | ||
fc5bc4c8 | 3618 | error = XFS_BUF_SET_PTR(hbp, offset, |
234f56ac | 3619 | BBTOB(hblks)); |
1da177e4 LT |
3620 | if (error) |
3621 | goto bread_err2; | |
1da177e4 LT |
3622 | } |
3623 | rhead = (xlog_rec_header_t *)offset; | |
3624 | error = xlog_valid_rec_header(log, rhead, | |
3625 | split_hblks ? blk_no : 0); | |
3626 | if (error) | |
3627 | goto bread_err2; | |
3628 | ||
b53e675d | 3629 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
1da177e4 LT |
3630 | blk_no += hblks; |
3631 | ||
3632 | /* Read in data for log record */ | |
3633 | if (blk_no + bblks <= log->l_logBBsize) { | |
076e6acb CH |
3634 | error = xlog_bread(log, blk_no, bblks, dbp, |
3635 | &offset); | |
1da177e4 LT |
3636 | if (error) |
3637 | goto bread_err2; | |
1da177e4 LT |
3638 | } else { |
3639 | /* This log record is split across the | |
3640 | * physical end of log */ | |
fc5bc4c8 | 3641 | offset = XFS_BUF_PTR(dbp); |
1da177e4 LT |
3642 | split_bblks = 0; |
3643 | if (blk_no != log->l_logBBsize) { | |
3644 | /* some data is before the physical | |
3645 | * end of log */ | |
3646 | ASSERT(!wrapped_hblks); | |
3647 | ASSERT(blk_no <= INT_MAX); | |
3648 | split_bblks = | |
3649 | log->l_logBBsize - (int)blk_no; | |
3650 | ASSERT(split_bblks > 0); | |
076e6acb CH |
3651 | error = xlog_bread(log, blk_no, |
3652 | split_bblks, dbp, | |
3653 | &offset); | |
3654 | if (error) | |
1da177e4 | 3655 | goto bread_err2; |
1da177e4 | 3656 | } |
076e6acb | 3657 | |
1da177e4 LT |
3658 | /* |
3659 | * Note: this black magic still works with | |
3660 | * large sector sizes (non-512) only because: | |
3661 | * - we increased the buffer size originally | |
3662 | * by 1 sector giving us enough extra space | |
3663 | * for the second read; | |
3664 | * - the log start is guaranteed to be sector | |
3665 | * aligned; | |
3666 | * - we read the log end (LR header start) | |
3667 | * _first_, then the log start (LR header end) | |
3668 | * - order is important. | |
3669 | */ | |
234f56ac | 3670 | error = XFS_BUF_SET_PTR(dbp, |
fc5bc4c8 | 3671 | offset + BBTOB(split_bblks), |
1da177e4 | 3672 | BBTOB(bblks - split_bblks)); |
234f56ac | 3673 | if (error) |
1da177e4 | 3674 | goto bread_err2; |
076e6acb CH |
3675 | |
3676 | error = xlog_bread_noalign(log, wrapped_hblks, | |
3677 | bblks - split_bblks, | |
3678 | dbp); | |
3679 | if (error) | |
3680 | goto bread_err2; | |
3681 | ||
fc5bc4c8 | 3682 | error = XFS_BUF_SET_PTR(dbp, offset, h_size); |
076e6acb CH |
3683 | if (error) |
3684 | goto bread_err2; | |
1da177e4 LT |
3685 | } |
3686 | xlog_unpack_data(rhead, offset, log); | |
3687 | if ((error = xlog_recover_process_data(log, rhash, | |
3688 | rhead, offset, pass))) | |
3689 | goto bread_err2; | |
3690 | blk_no += bblks; | |
3691 | } | |
3692 | ||
3693 | ASSERT(blk_no >= log->l_logBBsize); | |
3694 | blk_no -= log->l_logBBsize; | |
3695 | ||
3696 | /* read first part of physical log */ | |
3697 | while (blk_no < head_blk) { | |
076e6acb CH |
3698 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
3699 | if (error) | |
1da177e4 | 3700 | goto bread_err2; |
076e6acb | 3701 | |
1da177e4 LT |
3702 | rhead = (xlog_rec_header_t *)offset; |
3703 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3704 | if (error) | |
3705 | goto bread_err2; | |
076e6acb | 3706 | |
b53e675d | 3707 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
3708 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, |
3709 | &offset); | |
3710 | if (error) | |
1da177e4 | 3711 | goto bread_err2; |
076e6acb | 3712 | |
1da177e4 LT |
3713 | xlog_unpack_data(rhead, offset, log); |
3714 | if ((error = xlog_recover_process_data(log, rhash, | |
3715 | rhead, offset, pass))) | |
3716 | goto bread_err2; | |
3717 | blk_no += bblks + hblks; | |
3718 | } | |
3719 | } | |
3720 | ||
3721 | bread_err2: | |
3722 | xlog_put_bp(dbp); | |
3723 | bread_err1: | |
3724 | xlog_put_bp(hbp); | |
3725 | return error; | |
3726 | } | |
3727 | ||
3728 | /* | |
3729 | * Do the recovery of the log. We actually do this in two phases. | |
3730 | * The two passes are necessary in order to implement the function | |
3731 | * of cancelling a record written into the log. The first pass | |
3732 | * determines those things which have been cancelled, and the | |
3733 | * second pass replays log items normally except for those which | |
3734 | * have been cancelled. The handling of the replay and cancellations | |
3735 | * takes place in the log item type specific routines. | |
3736 | * | |
3737 | * The table of items which have cancel records in the log is allocated | |
3738 | * and freed at this level, since only here do we know when all of | |
3739 | * the log recovery has been completed. | |
3740 | */ | |
3741 | STATIC int | |
3742 | xlog_do_log_recovery( | |
3743 | xlog_t *log, | |
3744 | xfs_daddr_t head_blk, | |
3745 | xfs_daddr_t tail_blk) | |
3746 | { | |
3747 | int error; | |
3748 | ||
3749 | ASSERT(head_blk != tail_blk); | |
3750 | ||
3751 | /* | |
3752 | * First do a pass to find all of the cancelled buf log items. | |
3753 | * Store them in the buf_cancel_table for use in the second pass. | |
3754 | */ | |
3755 | log->l_buf_cancel_table = | |
3756 | (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE * | |
3757 | sizeof(xfs_buf_cancel_t*), | |
3758 | KM_SLEEP); | |
3759 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
3760 | XLOG_RECOVER_PASS1); | |
3761 | if (error != 0) { | |
f0e2d93c | 3762 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
3763 | log->l_buf_cancel_table = NULL; |
3764 | return error; | |
3765 | } | |
3766 | /* | |
3767 | * Then do a second pass to actually recover the items in the log. | |
3768 | * When it is complete free the table of buf cancel items. | |
3769 | */ | |
3770 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
3771 | XLOG_RECOVER_PASS2); | |
3772 | #ifdef DEBUG | |
6d192a9b | 3773 | if (!error) { |
1da177e4 LT |
3774 | int i; |
3775 | ||
3776 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
3777 | ASSERT(log->l_buf_cancel_table[i] == NULL); | |
3778 | } | |
3779 | #endif /* DEBUG */ | |
3780 | ||
f0e2d93c | 3781 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
3782 | log->l_buf_cancel_table = NULL; |
3783 | ||
3784 | return error; | |
3785 | } | |
3786 | ||
3787 | /* | |
3788 | * Do the actual recovery | |
3789 | */ | |
3790 | STATIC int | |
3791 | xlog_do_recover( | |
3792 | xlog_t *log, | |
3793 | xfs_daddr_t head_blk, | |
3794 | xfs_daddr_t tail_blk) | |
3795 | { | |
3796 | int error; | |
3797 | xfs_buf_t *bp; | |
3798 | xfs_sb_t *sbp; | |
3799 | ||
3800 | /* | |
3801 | * First replay the images in the log. | |
3802 | */ | |
3803 | error = xlog_do_log_recovery(log, head_blk, tail_blk); | |
3804 | if (error) { | |
3805 | return error; | |
3806 | } | |
3807 | ||
3808 | XFS_bflush(log->l_mp->m_ddev_targp); | |
3809 | ||
3810 | /* | |
3811 | * If IO errors happened during recovery, bail out. | |
3812 | */ | |
3813 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) { | |
3814 | return (EIO); | |
3815 | } | |
3816 | ||
3817 | /* | |
3818 | * We now update the tail_lsn since much of the recovery has completed | |
3819 | * and there may be space available to use. If there were no extent | |
3820 | * or iunlinks, we can free up the entire log and set the tail_lsn to | |
3821 | * be the last_sync_lsn. This was set in xlog_find_tail to be the | |
3822 | * lsn of the last known good LR on disk. If there are extent frees | |
3823 | * or iunlinks they will have some entries in the AIL; so we look at | |
3824 | * the AIL to determine how to set the tail_lsn. | |
3825 | */ | |
3826 | xlog_assign_tail_lsn(log->l_mp); | |
3827 | ||
3828 | /* | |
3829 | * Now that we've finished replaying all buffer and inode | |
3830 | * updates, re-read in the superblock. | |
3831 | */ | |
3832 | bp = xfs_getsb(log->l_mp, 0); | |
3833 | XFS_BUF_UNDONE(bp); | |
bebf963f LM |
3834 | ASSERT(!(XFS_BUF_ISWRITE(bp))); |
3835 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); | |
1da177e4 | 3836 | XFS_BUF_READ(bp); |
bebf963f | 3837 | XFS_BUF_UNASYNC(bp); |
1da177e4 | 3838 | xfsbdstrat(log->l_mp, bp); |
d64e31a2 DC |
3839 | error = xfs_iowait(bp); |
3840 | if (error) { | |
1da177e4 LT |
3841 | xfs_ioerror_alert("xlog_do_recover", |
3842 | log->l_mp, bp, XFS_BUF_ADDR(bp)); | |
3843 | ASSERT(0); | |
3844 | xfs_buf_relse(bp); | |
3845 | return error; | |
3846 | } | |
3847 | ||
3848 | /* Convert superblock from on-disk format */ | |
3849 | sbp = &log->l_mp->m_sb; | |
2bdf7cd0 | 3850 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 3851 | ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC); |
62118709 | 3852 | ASSERT(xfs_sb_good_version(sbp)); |
1da177e4 LT |
3853 | xfs_buf_relse(bp); |
3854 | ||
5478eead LM |
3855 | /* We've re-read the superblock so re-initialize per-cpu counters */ |
3856 | xfs_icsb_reinit_counters(log->l_mp); | |
3857 | ||
1da177e4 LT |
3858 | xlog_recover_check_summary(log); |
3859 | ||
3860 | /* Normal transactions can now occur */ | |
3861 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
3862 | return 0; | |
3863 | } | |
3864 | ||
3865 | /* | |
3866 | * Perform recovery and re-initialize some log variables in xlog_find_tail. | |
3867 | * | |
3868 | * Return error or zero. | |
3869 | */ | |
3870 | int | |
3871 | xlog_recover( | |
65be6054 | 3872 | xlog_t *log) |
1da177e4 LT |
3873 | { |
3874 | xfs_daddr_t head_blk, tail_blk; | |
3875 | int error; | |
3876 | ||
3877 | /* find the tail of the log */ | |
65be6054 | 3878 | if ((error = xlog_find_tail(log, &head_blk, &tail_blk))) |
1da177e4 LT |
3879 | return error; |
3880 | ||
3881 | if (tail_blk != head_blk) { | |
3882 | /* There used to be a comment here: | |
3883 | * | |
3884 | * disallow recovery on read-only mounts. note -- mount | |
3885 | * checks for ENOSPC and turns it into an intelligent | |
3886 | * error message. | |
3887 | * ...but this is no longer true. Now, unless you specify | |
3888 | * NORECOVERY (in which case this function would never be | |
3889 | * called), we just go ahead and recover. We do this all | |
3890 | * under the vfs layer, so we can get away with it unless | |
3891 | * the device itself is read-only, in which case we fail. | |
3892 | */ | |
3a02ee18 | 3893 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
1da177e4 LT |
3894 | return error; |
3895 | } | |
3896 | ||
3897 | cmn_err(CE_NOTE, | |
fc1f8c1c NS |
3898 | "Starting XFS recovery on filesystem: %s (logdev: %s)", |
3899 | log->l_mp->m_fsname, log->l_mp->m_logname ? | |
3900 | log->l_mp->m_logname : "internal"); | |
1da177e4 LT |
3901 | |
3902 | error = xlog_do_recover(log, head_blk, tail_blk); | |
3903 | log->l_flags |= XLOG_RECOVERY_NEEDED; | |
3904 | } | |
3905 | return error; | |
3906 | } | |
3907 | ||
3908 | /* | |
3909 | * In the first part of recovery we replay inodes and buffers and build | |
3910 | * up the list of extent free items which need to be processed. Here | |
3911 | * we process the extent free items and clean up the on disk unlinked | |
3912 | * inode lists. This is separated from the first part of recovery so | |
3913 | * that the root and real-time bitmap inodes can be read in from disk in | |
3914 | * between the two stages. This is necessary so that we can free space | |
3915 | * in the real-time portion of the file system. | |
3916 | */ | |
3917 | int | |
3918 | xlog_recover_finish( | |
4249023a | 3919 | xlog_t *log) |
1da177e4 LT |
3920 | { |
3921 | /* | |
3922 | * Now we're ready to do the transactions needed for the | |
3923 | * rest of recovery. Start with completing all the extent | |
3924 | * free intent records and then process the unlinked inode | |
3925 | * lists. At this point, we essentially run in normal mode | |
3926 | * except that we're still performing recovery actions | |
3927 | * rather than accepting new requests. | |
3928 | */ | |
3929 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { | |
3c1e2bbe DC |
3930 | int error; |
3931 | error = xlog_recover_process_efis(log); | |
3932 | if (error) { | |
3933 | cmn_err(CE_ALERT, | |
3934 | "Failed to recover EFIs on filesystem: %s", | |
3935 | log->l_mp->m_fsname); | |
3936 | return error; | |
3937 | } | |
1da177e4 LT |
3938 | /* |
3939 | * Sync the log to get all the EFIs out of the AIL. | |
3940 | * This isn't absolutely necessary, but it helps in | |
3941 | * case the unlink transactions would have problems | |
3942 | * pushing the EFIs out of the way. | |
3943 | */ | |
a14a348b | 3944 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
1da177e4 | 3945 | |
4249023a | 3946 | xlog_recover_process_iunlinks(log); |
1da177e4 LT |
3947 | |
3948 | xlog_recover_check_summary(log); | |
3949 | ||
3950 | cmn_err(CE_NOTE, | |
fc1f8c1c NS |
3951 | "Ending XFS recovery on filesystem: %s (logdev: %s)", |
3952 | log->l_mp->m_fsname, log->l_mp->m_logname ? | |
3953 | log->l_mp->m_logname : "internal"); | |
1da177e4 LT |
3954 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
3955 | } else { | |
3956 | cmn_err(CE_DEBUG, | |
b6574520 | 3957 | "!Ending clean XFS mount for filesystem: %s\n", |
1da177e4 LT |
3958 | log->l_mp->m_fsname); |
3959 | } | |
3960 | return 0; | |
3961 | } | |
3962 | ||
3963 | ||
3964 | #if defined(DEBUG) | |
3965 | /* | |
3966 | * Read all of the agf and agi counters and check that they | |
3967 | * are consistent with the superblock counters. | |
3968 | */ | |
3969 | void | |
3970 | xlog_recover_check_summary( | |
3971 | xlog_t *log) | |
3972 | { | |
3973 | xfs_mount_t *mp; | |
3974 | xfs_agf_t *agfp; | |
1da177e4 LT |
3975 | xfs_buf_t *agfbp; |
3976 | xfs_buf_t *agibp; | |
1da177e4 LT |
3977 | xfs_buf_t *sbbp; |
3978 | #ifdef XFS_LOUD_RECOVERY | |
3979 | xfs_sb_t *sbp; | |
3980 | #endif | |
3981 | xfs_agnumber_t agno; | |
3982 | __uint64_t freeblks; | |
3983 | __uint64_t itotal; | |
3984 | __uint64_t ifree; | |
5e1be0fb | 3985 | int error; |
1da177e4 LT |
3986 | |
3987 | mp = log->l_mp; | |
3988 | ||
3989 | freeblks = 0LL; | |
3990 | itotal = 0LL; | |
3991 | ifree = 0LL; | |
3992 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
4805621a FCH |
3993 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
3994 | if (error) { | |
3995 | xfs_fs_cmn_err(CE_ALERT, mp, | |
3996 | "xlog_recover_check_summary(agf)" | |
3997 | "agf read failed agno %d error %d", | |
3998 | agno, error); | |
3999 | } else { | |
4000 | agfp = XFS_BUF_TO_AGF(agfbp); | |
4001 | freeblks += be32_to_cpu(agfp->agf_freeblks) + | |
4002 | be32_to_cpu(agfp->agf_flcount); | |
4003 | xfs_buf_relse(agfbp); | |
1da177e4 | 4004 | } |
1da177e4 | 4005 | |
5e1be0fb CH |
4006 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
4007 | if (!error) { | |
4008 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); | |
16259e7d | 4009 | |
5e1be0fb CH |
4010 | itotal += be32_to_cpu(agi->agi_count); |
4011 | ifree += be32_to_cpu(agi->agi_freecount); | |
4012 | xfs_buf_relse(agibp); | |
4013 | } | |
1da177e4 LT |
4014 | } |
4015 | ||
4016 | sbbp = xfs_getsb(mp, 0); | |
4017 | #ifdef XFS_LOUD_RECOVERY | |
4018 | sbp = &mp->m_sb; | |
2bdf7cd0 | 4019 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp)); |
1da177e4 LT |
4020 | cmn_err(CE_NOTE, |
4021 | "xlog_recover_check_summary: sb_icount %Lu itotal %Lu", | |
4022 | sbp->sb_icount, itotal); | |
4023 | cmn_err(CE_NOTE, | |
4024 | "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu", | |
4025 | sbp->sb_ifree, ifree); | |
4026 | cmn_err(CE_NOTE, | |
4027 | "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu", | |
4028 | sbp->sb_fdblocks, freeblks); | |
4029 | #if 0 | |
4030 | /* | |
4031 | * This is turned off until I account for the allocation | |
4032 | * btree blocks which live in free space. | |
4033 | */ | |
4034 | ASSERT(sbp->sb_icount == itotal); | |
4035 | ASSERT(sbp->sb_ifree == ifree); | |
4036 | ASSERT(sbp->sb_fdblocks == freeblks); | |
4037 | #endif | |
4038 | #endif | |
4039 | xfs_buf_relse(sbbp); | |
4040 | } | |
4041 | #endif /* DEBUG */ |