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 NS |
38 | #include "xfs_inode_item.h" |
39 | #include "xfs_imap.h" | |
40 | #include "xfs_alloc.h" | |
1da177e4 LT |
41 | #include "xfs_ialloc.h" |
42 | #include "xfs_log_priv.h" | |
43 | #include "xfs_buf_item.h" | |
1da177e4 LT |
44 | #include "xfs_log_recover.h" |
45 | #include "xfs_extfree_item.h" | |
46 | #include "xfs_trans_priv.h" | |
1da177e4 LT |
47 | #include "xfs_quota.h" |
48 | #include "xfs_rw.h" | |
49 | ||
50 | STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *); | |
51 | STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t); | |
52 | STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q, | |
53 | xlog_recover_item_t *item); | |
54 | #if defined(DEBUG) | |
55 | STATIC void xlog_recover_check_summary(xlog_t *); | |
56 | STATIC void xlog_recover_check_ail(xfs_mount_t *, xfs_log_item_t *, int); | |
57 | #else | |
58 | #define xlog_recover_check_summary(log) | |
59 | #define xlog_recover_check_ail(mp, lip, gen) | |
60 | #endif | |
61 | ||
62 | ||
63 | /* | |
64 | * Sector aligned buffer routines for buffer create/read/write/access | |
65 | */ | |
66 | ||
67 | #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \ | |
68 | ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \ | |
69 | ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) ) | |
70 | #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask) | |
71 | ||
72 | xfs_buf_t * | |
73 | xlog_get_bp( | |
74 | xlog_t *log, | |
75 | int num_bblks) | |
76 | { | |
77 | ASSERT(num_bblks > 0); | |
78 | ||
79 | if (log->l_sectbb_log) { | |
80 | if (num_bblks > 1) | |
81 | num_bblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1); | |
82 | num_bblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, num_bblks); | |
83 | } | |
84 | return xfs_buf_get_noaddr(BBTOB(num_bblks), log->l_mp->m_logdev_targp); | |
85 | } | |
86 | ||
87 | void | |
88 | xlog_put_bp( | |
89 | xfs_buf_t *bp) | |
90 | { | |
91 | xfs_buf_free(bp); | |
92 | } | |
93 | ||
94 | ||
95 | /* | |
96 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. | |
97 | */ | |
98 | int | |
99 | xlog_bread( | |
100 | xlog_t *log, | |
101 | xfs_daddr_t blk_no, | |
102 | int nbblks, | |
103 | xfs_buf_t *bp) | |
104 | { | |
105 | int error; | |
106 | ||
107 | if (log->l_sectbb_log) { | |
108 | blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no); | |
109 | nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks); | |
110 | } | |
111 | ||
112 | ASSERT(nbblks > 0); | |
113 | ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp)); | |
114 | ASSERT(bp); | |
115 | ||
116 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
117 | XFS_BUF_READ(bp); | |
118 | XFS_BUF_BUSY(bp); | |
119 | XFS_BUF_SET_COUNT(bp, BBTOB(nbblks)); | |
120 | XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp); | |
121 | ||
122 | xfsbdstrat(log->l_mp, bp); | |
123 | if ((error = xfs_iowait(bp))) | |
124 | xfs_ioerror_alert("xlog_bread", log->l_mp, | |
125 | bp, XFS_BUF_ADDR(bp)); | |
126 | return error; | |
127 | } | |
128 | ||
129 | /* | |
130 | * Write out the buffer at the given block for the given number of blocks. | |
131 | * The buffer is kept locked across the write and is returned locked. | |
132 | * This can only be used for synchronous log writes. | |
133 | */ | |
ba0f32d4 | 134 | STATIC int |
1da177e4 LT |
135 | xlog_bwrite( |
136 | xlog_t *log, | |
137 | xfs_daddr_t blk_no, | |
138 | int nbblks, | |
139 | xfs_buf_t *bp) | |
140 | { | |
141 | int error; | |
142 | ||
143 | if (log->l_sectbb_log) { | |
144 | blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no); | |
145 | nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks); | |
146 | } | |
147 | ||
148 | ASSERT(nbblks > 0); | |
149 | ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp)); | |
150 | ||
151 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
152 | XFS_BUF_ZEROFLAGS(bp); | |
153 | XFS_BUF_BUSY(bp); | |
154 | XFS_BUF_HOLD(bp); | |
155 | XFS_BUF_PSEMA(bp, PRIBIO); | |
156 | XFS_BUF_SET_COUNT(bp, BBTOB(nbblks)); | |
157 | XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp); | |
158 | ||
159 | if ((error = xfs_bwrite(log->l_mp, bp))) | |
160 | xfs_ioerror_alert("xlog_bwrite", log->l_mp, | |
161 | bp, XFS_BUF_ADDR(bp)); | |
162 | return error; | |
163 | } | |
164 | ||
ba0f32d4 | 165 | STATIC xfs_caddr_t |
1da177e4 LT |
166 | xlog_align( |
167 | xlog_t *log, | |
168 | xfs_daddr_t blk_no, | |
169 | int nbblks, | |
170 | xfs_buf_t *bp) | |
171 | { | |
172 | xfs_caddr_t ptr; | |
173 | ||
174 | if (!log->l_sectbb_log) | |
175 | return XFS_BUF_PTR(bp); | |
176 | ||
177 | ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask); | |
178 | ASSERT(XFS_BUF_SIZE(bp) >= | |
179 | BBTOB(nbblks + (blk_no & log->l_sectbb_mask))); | |
180 | return ptr; | |
181 | } | |
182 | ||
183 | #ifdef DEBUG | |
184 | /* | |
185 | * dump debug superblock and log record information | |
186 | */ | |
187 | STATIC void | |
188 | xlog_header_check_dump( | |
189 | xfs_mount_t *mp, | |
190 | xlog_rec_header_t *head) | |
191 | { | |
192 | int b; | |
193 | ||
b6574520 | 194 | cmn_err(CE_DEBUG, "%s: SB : uuid = ", __FUNCTION__); |
1da177e4 | 195 | for (b = 0; b < 16; b++) |
b6574520 NS |
196 | cmn_err(CE_DEBUG, "%02x", ((uchar_t *)&mp->m_sb.sb_uuid)[b]); |
197 | cmn_err(CE_DEBUG, ", fmt = %d\n", XLOG_FMT); | |
198 | cmn_err(CE_DEBUG, " log : uuid = "); | |
1da177e4 | 199 | for (b = 0; b < 16; b++) |
b6574520 NS |
200 | cmn_err(CE_DEBUG, "%02x",((uchar_t *)&head->h_fs_uuid)[b]); |
201 | cmn_err(CE_DEBUG, ", fmt = %d\n", INT_GET(head->h_fmt, ARCH_CONVERT)); | |
1da177e4 LT |
202 | } |
203 | #else | |
204 | #define xlog_header_check_dump(mp, head) | |
205 | #endif | |
206 | ||
207 | /* | |
208 | * check log record header for recovery | |
209 | */ | |
210 | STATIC int | |
211 | xlog_header_check_recover( | |
212 | xfs_mount_t *mp, | |
213 | xlog_rec_header_t *head) | |
214 | { | |
215 | ASSERT(INT_GET(head->h_magicno, ARCH_CONVERT) == XLOG_HEADER_MAGIC_NUM); | |
216 | ||
217 | /* | |
218 | * IRIX doesn't write the h_fmt field and leaves it zeroed | |
219 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover | |
220 | * a dirty log created in IRIX. | |
221 | */ | |
222 | if (unlikely(INT_GET(head->h_fmt, ARCH_CONVERT) != XLOG_FMT)) { | |
223 | xlog_warn( | |
224 | "XFS: dirty log written in incompatible format - can't recover"); | |
225 | xlog_header_check_dump(mp, head); | |
226 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", | |
227 | XFS_ERRLEVEL_HIGH, mp); | |
228 | return XFS_ERROR(EFSCORRUPTED); | |
229 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
230 | xlog_warn( | |
231 | "XFS: dirty log entry has mismatched uuid - can't recover"); | |
232 | xlog_header_check_dump(mp, head); | |
233 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", | |
234 | XFS_ERRLEVEL_HIGH, mp); | |
235 | return XFS_ERROR(EFSCORRUPTED); | |
236 | } | |
237 | return 0; | |
238 | } | |
239 | ||
240 | /* | |
241 | * read the head block of the log and check the header | |
242 | */ | |
243 | STATIC int | |
244 | xlog_header_check_mount( | |
245 | xfs_mount_t *mp, | |
246 | xlog_rec_header_t *head) | |
247 | { | |
248 | ASSERT(INT_GET(head->h_magicno, ARCH_CONVERT) == XLOG_HEADER_MAGIC_NUM); | |
249 | ||
250 | if (uuid_is_nil(&head->h_fs_uuid)) { | |
251 | /* | |
252 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If | |
253 | * h_fs_uuid is nil, we assume this log was last mounted | |
254 | * by IRIX and continue. | |
255 | */ | |
256 | xlog_warn("XFS: nil uuid in log - IRIX style log"); | |
257 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
258 | xlog_warn("XFS: log has mismatched uuid - can't recover"); | |
259 | xlog_header_check_dump(mp, head); | |
260 | XFS_ERROR_REPORT("xlog_header_check_mount", | |
261 | XFS_ERRLEVEL_HIGH, mp); | |
262 | return XFS_ERROR(EFSCORRUPTED); | |
263 | } | |
264 | return 0; | |
265 | } | |
266 | ||
267 | STATIC void | |
268 | xlog_recover_iodone( | |
269 | struct xfs_buf *bp) | |
270 | { | |
271 | xfs_mount_t *mp; | |
272 | ||
273 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *)); | |
274 | ||
275 | if (XFS_BUF_GETERROR(bp)) { | |
276 | /* | |
277 | * We're not going to bother about retrying | |
278 | * this during recovery. One strike! | |
279 | */ | |
280 | mp = XFS_BUF_FSPRIVATE(bp, xfs_mount_t *); | |
281 | xfs_ioerror_alert("xlog_recover_iodone", | |
282 | mp, bp, XFS_BUF_ADDR(bp)); | |
7d04a335 | 283 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
284 | } |
285 | XFS_BUF_SET_FSPRIVATE(bp, NULL); | |
286 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
287 | xfs_biodone(bp); | |
288 | } | |
289 | ||
290 | /* | |
291 | * This routine finds (to an approximation) the first block in the physical | |
292 | * log which contains the given cycle. It uses a binary search algorithm. | |
293 | * Note that the algorithm can not be perfect because the disk will not | |
294 | * necessarily be perfect. | |
295 | */ | |
296 | int | |
297 | xlog_find_cycle_start( | |
298 | xlog_t *log, | |
299 | xfs_buf_t *bp, | |
300 | xfs_daddr_t first_blk, | |
301 | xfs_daddr_t *last_blk, | |
302 | uint cycle) | |
303 | { | |
304 | xfs_caddr_t offset; | |
305 | xfs_daddr_t mid_blk; | |
306 | uint mid_cycle; | |
307 | int error; | |
308 | ||
309 | mid_blk = BLK_AVG(first_blk, *last_blk); | |
310 | while (mid_blk != first_blk && mid_blk != *last_blk) { | |
311 | if ((error = xlog_bread(log, mid_blk, 1, bp))) | |
312 | return error; | |
313 | offset = xlog_align(log, mid_blk, 1, bp); | |
314 | mid_cycle = GET_CYCLE(offset, ARCH_CONVERT); | |
315 | if (mid_cycle == cycle) { | |
316 | *last_blk = mid_blk; | |
317 | /* last_half_cycle == mid_cycle */ | |
318 | } else { | |
319 | first_blk = mid_blk; | |
320 | /* first_half_cycle == mid_cycle */ | |
321 | } | |
322 | mid_blk = BLK_AVG(first_blk, *last_blk); | |
323 | } | |
324 | ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) || | |
325 | (mid_blk == *last_blk && mid_blk-1 == first_blk)); | |
326 | ||
327 | return 0; | |
328 | } | |
329 | ||
330 | /* | |
331 | * Check that the range of blocks does not contain the cycle number | |
332 | * given. The scan needs to occur from front to back and the ptr into the | |
333 | * region must be updated since a later routine will need to perform another | |
334 | * test. If the region is completely good, we end up returning the same | |
335 | * last block number. | |
336 | * | |
337 | * Set blkno to -1 if we encounter no errors. This is an invalid block number | |
338 | * since we don't ever expect logs to get this large. | |
339 | */ | |
340 | STATIC int | |
341 | xlog_find_verify_cycle( | |
342 | xlog_t *log, | |
343 | xfs_daddr_t start_blk, | |
344 | int nbblks, | |
345 | uint stop_on_cycle_no, | |
346 | xfs_daddr_t *new_blk) | |
347 | { | |
348 | xfs_daddr_t i, j; | |
349 | uint cycle; | |
350 | xfs_buf_t *bp; | |
351 | xfs_daddr_t bufblks; | |
352 | xfs_caddr_t buf = NULL; | |
353 | int error = 0; | |
354 | ||
355 | bufblks = 1 << ffs(nbblks); | |
356 | ||
357 | while (!(bp = xlog_get_bp(log, bufblks))) { | |
358 | /* can't get enough memory to do everything in one big buffer */ | |
359 | bufblks >>= 1; | |
360 | if (bufblks <= log->l_sectbb_log) | |
361 | return ENOMEM; | |
362 | } | |
363 | ||
364 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { | |
365 | int bcount; | |
366 | ||
367 | bcount = min(bufblks, (start_blk + nbblks - i)); | |
368 | ||
369 | if ((error = xlog_bread(log, i, bcount, bp))) | |
370 | goto out; | |
371 | ||
372 | buf = xlog_align(log, i, bcount, bp); | |
373 | for (j = 0; j < bcount; j++) { | |
374 | cycle = GET_CYCLE(buf, ARCH_CONVERT); | |
375 | if (cycle == stop_on_cycle_no) { | |
376 | *new_blk = i+j; | |
377 | goto out; | |
378 | } | |
379 | ||
380 | buf += BBSIZE; | |
381 | } | |
382 | } | |
383 | ||
384 | *new_blk = -1; | |
385 | ||
386 | out: | |
387 | xlog_put_bp(bp); | |
388 | return error; | |
389 | } | |
390 | ||
391 | /* | |
392 | * Potentially backup over partial log record write. | |
393 | * | |
394 | * In the typical case, last_blk is the number of the block directly after | |
395 | * a good log record. Therefore, we subtract one to get the block number | |
396 | * of the last block in the given buffer. extra_bblks contains the number | |
397 | * of blocks we would have read on a previous read. This happens when the | |
398 | * last log record is split over the end of the physical log. | |
399 | * | |
400 | * extra_bblks is the number of blocks potentially verified on a previous | |
401 | * call to this routine. | |
402 | */ | |
403 | STATIC int | |
404 | xlog_find_verify_log_record( | |
405 | xlog_t *log, | |
406 | xfs_daddr_t start_blk, | |
407 | xfs_daddr_t *last_blk, | |
408 | int extra_bblks) | |
409 | { | |
410 | xfs_daddr_t i; | |
411 | xfs_buf_t *bp; | |
412 | xfs_caddr_t offset = NULL; | |
413 | xlog_rec_header_t *head = NULL; | |
414 | int error = 0; | |
415 | int smallmem = 0; | |
416 | int num_blks = *last_blk - start_blk; | |
417 | int xhdrs; | |
418 | ||
419 | ASSERT(start_blk != 0 || *last_blk != start_blk); | |
420 | ||
421 | if (!(bp = xlog_get_bp(log, num_blks))) { | |
422 | if (!(bp = xlog_get_bp(log, 1))) | |
423 | return ENOMEM; | |
424 | smallmem = 1; | |
425 | } else { | |
426 | if ((error = xlog_bread(log, start_blk, num_blks, bp))) | |
427 | goto out; | |
428 | offset = xlog_align(log, start_blk, num_blks, bp); | |
429 | offset += ((num_blks - 1) << BBSHIFT); | |
430 | } | |
431 | ||
432 | for (i = (*last_blk) - 1; i >= 0; i--) { | |
433 | if (i < start_blk) { | |
434 | /* valid log record not found */ | |
435 | xlog_warn( | |
436 | "XFS: Log inconsistent (didn't find previous header)"); | |
437 | ASSERT(0); | |
438 | error = XFS_ERROR(EIO); | |
439 | goto out; | |
440 | } | |
441 | ||
442 | if (smallmem) { | |
443 | if ((error = xlog_bread(log, i, 1, bp))) | |
444 | goto out; | |
445 | offset = xlog_align(log, i, 1, bp); | |
446 | } | |
447 | ||
448 | head = (xlog_rec_header_t *)offset; | |
449 | ||
450 | if (XLOG_HEADER_MAGIC_NUM == | |
451 | INT_GET(head->h_magicno, ARCH_CONVERT)) | |
452 | break; | |
453 | ||
454 | if (!smallmem) | |
455 | offset -= BBSIZE; | |
456 | } | |
457 | ||
458 | /* | |
459 | * We hit the beginning of the physical log & still no header. Return | |
460 | * to caller. If caller can handle a return of -1, then this routine | |
461 | * will be called again for the end of the physical log. | |
462 | */ | |
463 | if (i == -1) { | |
464 | error = -1; | |
465 | goto out; | |
466 | } | |
467 | ||
468 | /* | |
469 | * We have the final block of the good log (the first block | |
470 | * of the log record _before_ the head. So we check the uuid. | |
471 | */ | |
472 | if ((error = xlog_header_check_mount(log->l_mp, head))) | |
473 | goto out; | |
474 | ||
475 | /* | |
476 | * We may have found a log record header before we expected one. | |
477 | * last_blk will be the 1st block # with a given cycle #. We may end | |
478 | * up reading an entire log record. In this case, we don't want to | |
479 | * reset last_blk. Only when last_blk points in the middle of a log | |
480 | * record do we update last_blk. | |
481 | */ | |
482 | if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) { | |
483 | uint h_size = INT_GET(head->h_size, ARCH_CONVERT); | |
484 | ||
485 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; | |
486 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
487 | xhdrs++; | |
488 | } else { | |
489 | xhdrs = 1; | |
490 | } | |
491 | ||
492 | if (*last_blk - i + extra_bblks | |
493 | != BTOBB(INT_GET(head->h_len, ARCH_CONVERT)) + xhdrs) | |
494 | *last_blk = i; | |
495 | ||
496 | out: | |
497 | xlog_put_bp(bp); | |
498 | return error; | |
499 | } | |
500 | ||
501 | /* | |
502 | * Head is defined to be the point of the log where the next log write | |
503 | * write could go. This means that incomplete LR writes at the end are | |
504 | * eliminated when calculating the head. We aren't guaranteed that previous | |
505 | * LR have complete transactions. We only know that a cycle number of | |
506 | * current cycle number -1 won't be present in the log if we start writing | |
507 | * from our current block number. | |
508 | * | |
509 | * last_blk contains the block number of the first block with a given | |
510 | * cycle number. | |
511 | * | |
512 | * Return: zero if normal, non-zero if error. | |
513 | */ | |
ba0f32d4 | 514 | STATIC int |
1da177e4 LT |
515 | xlog_find_head( |
516 | xlog_t *log, | |
517 | xfs_daddr_t *return_head_blk) | |
518 | { | |
519 | xfs_buf_t *bp; | |
520 | xfs_caddr_t offset; | |
521 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; | |
522 | int num_scan_bblks; | |
523 | uint first_half_cycle, last_half_cycle; | |
524 | uint stop_on_cycle; | |
525 | int error, log_bbnum = log->l_logBBsize; | |
526 | ||
527 | /* Is the end of the log device zeroed? */ | |
528 | if ((error = xlog_find_zeroed(log, &first_blk)) == -1) { | |
529 | *return_head_blk = first_blk; | |
530 | ||
531 | /* Is the whole lot zeroed? */ | |
532 | if (!first_blk) { | |
533 | /* Linux XFS shouldn't generate totally zeroed logs - | |
534 | * mkfs etc write a dummy unmount record to a fresh | |
535 | * log so we can store the uuid in there | |
536 | */ | |
537 | xlog_warn("XFS: totally zeroed log"); | |
538 | } | |
539 | ||
540 | return 0; | |
541 | } else if (error) { | |
542 | xlog_warn("XFS: empty log check failed"); | |
543 | return error; | |
544 | } | |
545 | ||
546 | first_blk = 0; /* get cycle # of 1st block */ | |
547 | bp = xlog_get_bp(log, 1); | |
548 | if (!bp) | |
549 | return ENOMEM; | |
550 | if ((error = xlog_bread(log, 0, 1, bp))) | |
551 | goto bp_err; | |
552 | offset = xlog_align(log, 0, 1, bp); | |
553 | first_half_cycle = GET_CYCLE(offset, ARCH_CONVERT); | |
554 | ||
555 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ | |
556 | if ((error = xlog_bread(log, last_blk, 1, bp))) | |
557 | goto bp_err; | |
558 | offset = xlog_align(log, last_blk, 1, bp); | |
559 | last_half_cycle = GET_CYCLE(offset, ARCH_CONVERT); | |
560 | ASSERT(last_half_cycle != 0); | |
561 | ||
562 | /* | |
563 | * If the 1st half cycle number is equal to the last half cycle number, | |
564 | * then the entire log is stamped with the same cycle number. In this | |
565 | * case, head_blk can't be set to zero (which makes sense). The below | |
566 | * math doesn't work out properly with head_blk equal to zero. Instead, | |
567 | * we set it to log_bbnum which is an invalid block number, but this | |
568 | * value makes the math correct. If head_blk doesn't changed through | |
569 | * all the tests below, *head_blk is set to zero at the very end rather | |
570 | * than log_bbnum. In a sense, log_bbnum and zero are the same block | |
571 | * in a circular file. | |
572 | */ | |
573 | if (first_half_cycle == last_half_cycle) { | |
574 | /* | |
575 | * In this case we believe that the entire log should have | |
576 | * cycle number last_half_cycle. We need to scan backwards | |
577 | * from the end verifying that there are no holes still | |
578 | * containing last_half_cycle - 1. If we find such a hole, | |
579 | * then the start of that hole will be the new head. The | |
580 | * simple case looks like | |
581 | * x | x ... | x - 1 | x | |
582 | * Another case that fits this picture would be | |
583 | * x | x + 1 | x ... | x | |
c41564b5 | 584 | * In this case the head really is somewhere at the end of the |
1da177e4 LT |
585 | * log, as one of the latest writes at the beginning was |
586 | * incomplete. | |
587 | * One more case is | |
588 | * x | x + 1 | x ... | x - 1 | x | |
589 | * This is really the combination of the above two cases, and | |
590 | * the head has to end up at the start of the x-1 hole at the | |
591 | * end of the log. | |
592 | * | |
593 | * In the 256k log case, we will read from the beginning to the | |
594 | * end of the log and search for cycle numbers equal to x-1. | |
595 | * We don't worry about the x+1 blocks that we encounter, | |
596 | * because we know that they cannot be the head since the log | |
597 | * started with x. | |
598 | */ | |
599 | head_blk = log_bbnum; | |
600 | stop_on_cycle = last_half_cycle - 1; | |
601 | } else { | |
602 | /* | |
603 | * In this case we want to find the first block with cycle | |
604 | * number matching last_half_cycle. We expect the log to be | |
605 | * some variation on | |
606 | * x + 1 ... | x ... | |
607 | * The first block with cycle number x (last_half_cycle) will | |
608 | * be where the new head belongs. First we do a binary search | |
609 | * for the first occurrence of last_half_cycle. The binary | |
610 | * search may not be totally accurate, so then we scan back | |
611 | * from there looking for occurrences of last_half_cycle before | |
612 | * us. If that backwards scan wraps around the beginning of | |
613 | * the log, then we look for occurrences of last_half_cycle - 1 | |
614 | * at the end of the log. The cases we're looking for look | |
615 | * like | |
616 | * x + 1 ... | x | x + 1 | x ... | |
617 | * ^ binary search stopped here | |
618 | * or | |
619 | * x + 1 ... | x ... | x - 1 | x | |
620 | * <---------> less than scan distance | |
621 | */ | |
622 | stop_on_cycle = last_half_cycle; | |
623 | if ((error = xlog_find_cycle_start(log, bp, first_blk, | |
624 | &head_blk, last_half_cycle))) | |
625 | goto bp_err; | |
626 | } | |
627 | ||
628 | /* | |
629 | * Now validate the answer. Scan back some number of maximum possible | |
630 | * blocks and make sure each one has the expected cycle number. The | |
631 | * maximum is determined by the total possible amount of buffering | |
632 | * in the in-core log. The following number can be made tighter if | |
633 | * we actually look at the block size of the filesystem. | |
634 | */ | |
635 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
636 | if (head_blk >= num_scan_bblks) { | |
637 | /* | |
638 | * We are guaranteed that the entire check can be performed | |
639 | * in one buffer. | |
640 | */ | |
641 | start_blk = head_blk - num_scan_bblks; | |
642 | if ((error = xlog_find_verify_cycle(log, | |
643 | start_blk, num_scan_bblks, | |
644 | stop_on_cycle, &new_blk))) | |
645 | goto bp_err; | |
646 | if (new_blk != -1) | |
647 | head_blk = new_blk; | |
648 | } else { /* need to read 2 parts of log */ | |
649 | /* | |
650 | * We are going to scan backwards in the log in two parts. | |
651 | * First we scan the physical end of the log. In this part | |
652 | * of the log, we are looking for blocks with cycle number | |
653 | * last_half_cycle - 1. | |
654 | * If we find one, then we know that the log starts there, as | |
655 | * we've found a hole that didn't get written in going around | |
656 | * the end of the physical log. The simple case for this is | |
657 | * x + 1 ... | x ... | x - 1 | x | |
658 | * <---------> less than scan distance | |
659 | * If all of the blocks at the end of the log have cycle number | |
660 | * last_half_cycle, then we check the blocks at the start of | |
661 | * the log looking for occurrences of last_half_cycle. If we | |
662 | * find one, then our current estimate for the location of the | |
663 | * first occurrence of last_half_cycle is wrong and we move | |
664 | * back to the hole we've found. This case looks like | |
665 | * x + 1 ... | x | x + 1 | x ... | |
666 | * ^ binary search stopped here | |
667 | * Another case we need to handle that only occurs in 256k | |
668 | * logs is | |
669 | * x + 1 ... | x ... | x+1 | x ... | |
670 | * ^ binary search stops here | |
671 | * In a 256k log, the scan at the end of the log will see the | |
672 | * x + 1 blocks. We need to skip past those since that is | |
673 | * certainly not the head of the log. By searching for | |
674 | * last_half_cycle-1 we accomplish that. | |
675 | */ | |
676 | start_blk = log_bbnum - num_scan_bblks + head_blk; | |
677 | ASSERT(head_blk <= INT_MAX && | |
678 | (xfs_daddr_t) num_scan_bblks - head_blk >= 0); | |
679 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
680 | num_scan_bblks - (int)head_blk, | |
681 | (stop_on_cycle - 1), &new_blk))) | |
682 | goto bp_err; | |
683 | if (new_blk != -1) { | |
684 | head_blk = new_blk; | |
685 | goto bad_blk; | |
686 | } | |
687 | ||
688 | /* | |
689 | * Scan beginning of log now. The last part of the physical | |
690 | * log is good. This scan needs to verify that it doesn't find | |
691 | * the last_half_cycle. | |
692 | */ | |
693 | start_blk = 0; | |
694 | ASSERT(head_blk <= INT_MAX); | |
695 | if ((error = xlog_find_verify_cycle(log, | |
696 | start_blk, (int)head_blk, | |
697 | stop_on_cycle, &new_blk))) | |
698 | goto bp_err; | |
699 | if (new_blk != -1) | |
700 | head_blk = new_blk; | |
701 | } | |
702 | ||
703 | bad_blk: | |
704 | /* | |
705 | * Now we need to make sure head_blk is not pointing to a block in | |
706 | * the middle of a log record. | |
707 | */ | |
708 | num_scan_bblks = XLOG_REC_SHIFT(log); | |
709 | if (head_blk >= num_scan_bblks) { | |
710 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ | |
711 | ||
712 | /* start ptr at last block ptr before head_blk */ | |
713 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
714 | &head_blk, 0)) == -1) { | |
715 | error = XFS_ERROR(EIO); | |
716 | goto bp_err; | |
717 | } else if (error) | |
718 | goto bp_err; | |
719 | } else { | |
720 | start_blk = 0; | |
721 | ASSERT(head_blk <= INT_MAX); | |
722 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
723 | &head_blk, 0)) == -1) { | |
724 | /* We hit the beginning of the log during our search */ | |
725 | start_blk = log_bbnum - num_scan_bblks + head_blk; | |
726 | new_blk = log_bbnum; | |
727 | ASSERT(start_blk <= INT_MAX && | |
728 | (xfs_daddr_t) log_bbnum-start_blk >= 0); | |
729 | ASSERT(head_blk <= INT_MAX); | |
730 | if ((error = xlog_find_verify_log_record(log, | |
731 | start_blk, &new_blk, | |
732 | (int)head_blk)) == -1) { | |
733 | error = XFS_ERROR(EIO); | |
734 | goto bp_err; | |
735 | } else if (error) | |
736 | goto bp_err; | |
737 | if (new_blk != log_bbnum) | |
738 | head_blk = new_blk; | |
739 | } else if (error) | |
740 | goto bp_err; | |
741 | } | |
742 | ||
743 | xlog_put_bp(bp); | |
744 | if (head_blk == log_bbnum) | |
745 | *return_head_blk = 0; | |
746 | else | |
747 | *return_head_blk = head_blk; | |
748 | /* | |
749 | * When returning here, we have a good block number. Bad block | |
750 | * means that during a previous crash, we didn't have a clean break | |
751 | * from cycle number N to cycle number N-1. In this case, we need | |
752 | * to find the first block with cycle number N-1. | |
753 | */ | |
754 | return 0; | |
755 | ||
756 | bp_err: | |
757 | xlog_put_bp(bp); | |
758 | ||
759 | if (error) | |
760 | xlog_warn("XFS: failed to find log head"); | |
761 | return error; | |
762 | } | |
763 | ||
764 | /* | |
765 | * Find the sync block number or the tail of the log. | |
766 | * | |
767 | * This will be the block number of the last record to have its | |
768 | * associated buffers synced to disk. Every log record header has | |
769 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy | |
770 | * to get a sync block number. The only concern is to figure out which | |
771 | * log record header to believe. | |
772 | * | |
773 | * The following algorithm uses the log record header with the largest | |
774 | * lsn. The entire log record does not need to be valid. We only care | |
775 | * that the header is valid. | |
776 | * | |
777 | * We could speed up search by using current head_blk buffer, but it is not | |
778 | * available. | |
779 | */ | |
780 | int | |
781 | xlog_find_tail( | |
782 | xlog_t *log, | |
783 | xfs_daddr_t *head_blk, | |
65be6054 | 784 | xfs_daddr_t *tail_blk) |
1da177e4 LT |
785 | { |
786 | xlog_rec_header_t *rhead; | |
787 | xlog_op_header_t *op_head; | |
788 | xfs_caddr_t offset = NULL; | |
789 | xfs_buf_t *bp; | |
790 | int error, i, found; | |
791 | xfs_daddr_t umount_data_blk; | |
792 | xfs_daddr_t after_umount_blk; | |
793 | xfs_lsn_t tail_lsn; | |
794 | int hblks; | |
795 | ||
796 | found = 0; | |
797 | ||
798 | /* | |
799 | * Find previous log record | |
800 | */ | |
801 | if ((error = xlog_find_head(log, head_blk))) | |
802 | return error; | |
803 | ||
804 | bp = xlog_get_bp(log, 1); | |
805 | if (!bp) | |
806 | return ENOMEM; | |
807 | if (*head_blk == 0) { /* special case */ | |
808 | if ((error = xlog_bread(log, 0, 1, bp))) | |
809 | goto bread_err; | |
810 | offset = xlog_align(log, 0, 1, bp); | |
811 | if (GET_CYCLE(offset, ARCH_CONVERT) == 0) { | |
812 | *tail_blk = 0; | |
813 | /* leave all other log inited values alone */ | |
814 | goto exit; | |
815 | } | |
816 | } | |
817 | ||
818 | /* | |
819 | * Search backwards looking for log record header block | |
820 | */ | |
821 | ASSERT(*head_blk < INT_MAX); | |
822 | for (i = (int)(*head_blk) - 1; i >= 0; i--) { | |
823 | if ((error = xlog_bread(log, i, 1, bp))) | |
824 | goto bread_err; | |
825 | offset = xlog_align(log, i, 1, bp); | |
826 | if (XLOG_HEADER_MAGIC_NUM == | |
827 | INT_GET(*(uint *)offset, ARCH_CONVERT)) { | |
828 | found = 1; | |
829 | break; | |
830 | } | |
831 | } | |
832 | /* | |
833 | * If we haven't found the log record header block, start looking | |
834 | * again from the end of the physical log. XXXmiken: There should be | |
835 | * a check here to make sure we didn't search more than N blocks in | |
836 | * the previous code. | |
837 | */ | |
838 | if (!found) { | |
839 | for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) { | |
840 | if ((error = xlog_bread(log, i, 1, bp))) | |
841 | goto bread_err; | |
842 | offset = xlog_align(log, i, 1, bp); | |
843 | if (XLOG_HEADER_MAGIC_NUM == | |
844 | INT_GET(*(uint*)offset, ARCH_CONVERT)) { | |
845 | found = 2; | |
846 | break; | |
847 | } | |
848 | } | |
849 | } | |
850 | if (!found) { | |
851 | xlog_warn("XFS: xlog_find_tail: couldn't find sync record"); | |
852 | ASSERT(0); | |
853 | return XFS_ERROR(EIO); | |
854 | } | |
855 | ||
856 | /* find blk_no of tail of log */ | |
857 | rhead = (xlog_rec_header_t *)offset; | |
858 | *tail_blk = BLOCK_LSN(INT_GET(rhead->h_tail_lsn, ARCH_CONVERT)); | |
859 | ||
860 | /* | |
861 | * Reset log values according to the state of the log when we | |
862 | * crashed. In the case where head_blk == 0, we bump curr_cycle | |
863 | * one because the next write starts a new cycle rather than | |
864 | * continuing the cycle of the last good log record. At this | |
865 | * point we have guaranteed that all partial log records have been | |
866 | * accounted for. Therefore, we know that the last good log record | |
867 | * written was complete and ended exactly on the end boundary | |
868 | * of the physical log. | |
869 | */ | |
870 | log->l_prev_block = i; | |
871 | log->l_curr_block = (int)*head_blk; | |
872 | log->l_curr_cycle = INT_GET(rhead->h_cycle, ARCH_CONVERT); | |
873 | if (found == 2) | |
874 | log->l_curr_cycle++; | |
875 | log->l_tail_lsn = INT_GET(rhead->h_tail_lsn, ARCH_CONVERT); | |
876 | log->l_last_sync_lsn = INT_GET(rhead->h_lsn, ARCH_CONVERT); | |
877 | log->l_grant_reserve_cycle = log->l_curr_cycle; | |
878 | log->l_grant_reserve_bytes = BBTOB(log->l_curr_block); | |
879 | log->l_grant_write_cycle = log->l_curr_cycle; | |
880 | log->l_grant_write_bytes = BBTOB(log->l_curr_block); | |
881 | ||
882 | /* | |
883 | * Look for unmount record. If we find it, then we know there | |
884 | * was a clean unmount. Since 'i' could be the last block in | |
885 | * the physical log, we convert to a log block before comparing | |
886 | * to the head_blk. | |
887 | * | |
888 | * Save the current tail lsn to use to pass to | |
889 | * xlog_clear_stale_blocks() below. We won't want to clear the | |
890 | * unmount record if there is one, so we pass the lsn of the | |
891 | * unmount record rather than the block after it. | |
892 | */ | |
893 | if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) { | |
894 | int h_size = INT_GET(rhead->h_size, ARCH_CONVERT); | |
895 | int h_version = INT_GET(rhead->h_version, ARCH_CONVERT); | |
896 | ||
897 | if ((h_version & XLOG_VERSION_2) && | |
898 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
899 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
900 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
901 | hblks++; | |
902 | } else { | |
903 | hblks = 1; | |
904 | } | |
905 | } else { | |
906 | hblks = 1; | |
907 | } | |
908 | after_umount_blk = (i + hblks + (int) | |
909 | BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT))) % log->l_logBBsize; | |
910 | tail_lsn = log->l_tail_lsn; | |
911 | if (*head_blk == after_umount_blk && | |
912 | INT_GET(rhead->h_num_logops, ARCH_CONVERT) == 1) { | |
913 | umount_data_blk = (i + hblks) % log->l_logBBsize; | |
914 | if ((error = xlog_bread(log, umount_data_blk, 1, bp))) { | |
915 | goto bread_err; | |
916 | } | |
917 | offset = xlog_align(log, umount_data_blk, 1, bp); | |
918 | op_head = (xlog_op_header_t *)offset; | |
919 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { | |
920 | /* | |
921 | * Set tail and last sync so that newly written | |
922 | * log records will point recovery to after the | |
923 | * current unmount record. | |
924 | */ | |
925 | ASSIGN_ANY_LSN_HOST(log->l_tail_lsn, log->l_curr_cycle, | |
926 | after_umount_blk); | |
927 | ASSIGN_ANY_LSN_HOST(log->l_last_sync_lsn, log->l_curr_cycle, | |
928 | after_umount_blk); | |
929 | *tail_blk = after_umount_blk; | |
92821e2b DC |
930 | |
931 | /* | |
932 | * Note that the unmount was clean. If the unmount | |
933 | * was not clean, we need to know this to rebuild the | |
934 | * superblock counters from the perag headers if we | |
935 | * have a filesystem using non-persistent counters. | |
936 | */ | |
937 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; | |
1da177e4 LT |
938 | } |
939 | } | |
940 | ||
941 | /* | |
942 | * Make sure that there are no blocks in front of the head | |
943 | * with the same cycle number as the head. This can happen | |
944 | * because we allow multiple outstanding log writes concurrently, | |
945 | * and the later writes might make it out before earlier ones. | |
946 | * | |
947 | * We use the lsn from before modifying it so that we'll never | |
948 | * overwrite the unmount record after a clean unmount. | |
949 | * | |
950 | * Do this only if we are going to recover the filesystem | |
951 | * | |
952 | * NOTE: This used to say "if (!readonly)" | |
953 | * However on Linux, we can & do recover a read-only filesystem. | |
954 | * We only skip recovery if NORECOVERY is specified on mount, | |
955 | * in which case we would not be here. | |
956 | * | |
957 | * But... if the -device- itself is readonly, just skip this. | |
958 | * We can't recover this device anyway, so it won't matter. | |
959 | */ | |
960 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) { | |
961 | error = xlog_clear_stale_blocks(log, tail_lsn); | |
962 | } | |
963 | ||
964 | bread_err: | |
965 | exit: | |
966 | xlog_put_bp(bp); | |
967 | ||
968 | if (error) | |
969 | xlog_warn("XFS: failed to locate log tail"); | |
970 | return error; | |
971 | } | |
972 | ||
973 | /* | |
974 | * Is the log zeroed at all? | |
975 | * | |
976 | * The last binary search should be changed to perform an X block read | |
977 | * once X becomes small enough. You can then search linearly through | |
978 | * the X blocks. This will cut down on the number of reads we need to do. | |
979 | * | |
980 | * If the log is partially zeroed, this routine will pass back the blkno | |
981 | * of the first block with cycle number 0. It won't have a complete LR | |
982 | * preceding it. | |
983 | * | |
984 | * Return: | |
985 | * 0 => the log is completely written to | |
986 | * -1 => use *blk_no as the first block of the log | |
987 | * >0 => error has occurred | |
988 | */ | |
989 | int | |
990 | xlog_find_zeroed( | |
991 | xlog_t *log, | |
992 | xfs_daddr_t *blk_no) | |
993 | { | |
994 | xfs_buf_t *bp; | |
995 | xfs_caddr_t offset; | |
996 | uint first_cycle, last_cycle; | |
997 | xfs_daddr_t new_blk, last_blk, start_blk; | |
998 | xfs_daddr_t num_scan_bblks; | |
999 | int error, log_bbnum = log->l_logBBsize; | |
1000 | ||
6fdf8ccc NS |
1001 | *blk_no = 0; |
1002 | ||
1da177e4 LT |
1003 | /* check totally zeroed log */ |
1004 | bp = xlog_get_bp(log, 1); | |
1005 | if (!bp) | |
1006 | return ENOMEM; | |
1007 | if ((error = xlog_bread(log, 0, 1, bp))) | |
1008 | goto bp_err; | |
1009 | offset = xlog_align(log, 0, 1, bp); | |
1010 | first_cycle = GET_CYCLE(offset, ARCH_CONVERT); | |
1011 | if (first_cycle == 0) { /* completely zeroed log */ | |
1012 | *blk_no = 0; | |
1013 | xlog_put_bp(bp); | |
1014 | return -1; | |
1015 | } | |
1016 | ||
1017 | /* check partially zeroed log */ | |
1018 | if ((error = xlog_bread(log, log_bbnum-1, 1, bp))) | |
1019 | goto bp_err; | |
1020 | offset = xlog_align(log, log_bbnum-1, 1, bp); | |
1021 | last_cycle = GET_CYCLE(offset, ARCH_CONVERT); | |
1022 | if (last_cycle != 0) { /* log completely written to */ | |
1023 | xlog_put_bp(bp); | |
1024 | return 0; | |
1025 | } else if (first_cycle != 1) { | |
1026 | /* | |
1027 | * If the cycle of the last block is zero, the cycle of | |
1028 | * the first block must be 1. If it's not, maybe we're | |
1029 | * not looking at a log... Bail out. | |
1030 | */ | |
1031 | xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)"); | |
1032 | return XFS_ERROR(EINVAL); | |
1033 | } | |
1034 | ||
1035 | /* we have a partially zeroed log */ | |
1036 | last_blk = log_bbnum-1; | |
1037 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) | |
1038 | goto bp_err; | |
1039 | ||
1040 | /* | |
1041 | * Validate the answer. Because there is no way to guarantee that | |
1042 | * the entire log is made up of log records which are the same size, | |
1043 | * we scan over the defined maximum blocks. At this point, the maximum | |
1044 | * is not chosen to mean anything special. XXXmiken | |
1045 | */ | |
1046 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
1047 | ASSERT(num_scan_bblks <= INT_MAX); | |
1048 | ||
1049 | if (last_blk < num_scan_bblks) | |
1050 | num_scan_bblks = last_blk; | |
1051 | start_blk = last_blk - num_scan_bblks; | |
1052 | ||
1053 | /* | |
1054 | * We search for any instances of cycle number 0 that occur before | |
1055 | * our current estimate of the head. What we're trying to detect is | |
1056 | * 1 ... | 0 | 1 | 0... | |
1057 | * ^ binary search ends here | |
1058 | */ | |
1059 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
1060 | (int)num_scan_bblks, 0, &new_blk))) | |
1061 | goto bp_err; | |
1062 | if (new_blk != -1) | |
1063 | last_blk = new_blk; | |
1064 | ||
1065 | /* | |
1066 | * Potentially backup over partial log record write. We don't need | |
1067 | * to search the end of the log because we know it is zero. | |
1068 | */ | |
1069 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
1070 | &last_blk, 0)) == -1) { | |
1071 | error = XFS_ERROR(EIO); | |
1072 | goto bp_err; | |
1073 | } else if (error) | |
1074 | goto bp_err; | |
1075 | ||
1076 | *blk_no = last_blk; | |
1077 | bp_err: | |
1078 | xlog_put_bp(bp); | |
1079 | if (error) | |
1080 | return error; | |
1081 | return -1; | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * These are simple subroutines used by xlog_clear_stale_blocks() below | |
1086 | * to initialize a buffer full of empty log record headers and write | |
1087 | * them into the log. | |
1088 | */ | |
1089 | STATIC void | |
1090 | xlog_add_record( | |
1091 | xlog_t *log, | |
1092 | xfs_caddr_t buf, | |
1093 | int cycle, | |
1094 | int block, | |
1095 | int tail_cycle, | |
1096 | int tail_block) | |
1097 | { | |
1098 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; | |
1099 | ||
1100 | memset(buf, 0, BBSIZE); | |
1101 | INT_SET(recp->h_magicno, ARCH_CONVERT, XLOG_HEADER_MAGIC_NUM); | |
1102 | INT_SET(recp->h_cycle, ARCH_CONVERT, cycle); | |
1103 | INT_SET(recp->h_version, ARCH_CONVERT, | |
1104 | XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? 2 : 1); | |
1105 | ASSIGN_ANY_LSN_DISK(recp->h_lsn, cycle, block); | |
1106 | ASSIGN_ANY_LSN_DISK(recp->h_tail_lsn, tail_cycle, tail_block); | |
1107 | INT_SET(recp->h_fmt, ARCH_CONVERT, XLOG_FMT); | |
1108 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); | |
1109 | } | |
1110 | ||
1111 | STATIC int | |
1112 | xlog_write_log_records( | |
1113 | xlog_t *log, | |
1114 | int cycle, | |
1115 | int start_block, | |
1116 | int blocks, | |
1117 | int tail_cycle, | |
1118 | int tail_block) | |
1119 | { | |
1120 | xfs_caddr_t offset; | |
1121 | xfs_buf_t *bp; | |
1122 | int balign, ealign; | |
1123 | int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1); | |
1124 | int end_block = start_block + blocks; | |
1125 | int bufblks; | |
1126 | int error = 0; | |
1127 | int i, j = 0; | |
1128 | ||
1129 | bufblks = 1 << ffs(blocks); | |
1130 | while (!(bp = xlog_get_bp(log, bufblks))) { | |
1131 | bufblks >>= 1; | |
1132 | if (bufblks <= log->l_sectbb_log) | |
1133 | return ENOMEM; | |
1134 | } | |
1135 | ||
1136 | /* We may need to do a read at the start to fill in part of | |
1137 | * the buffer in the starting sector not covered by the first | |
1138 | * write below. | |
1139 | */ | |
1140 | balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block); | |
1141 | if (balign != start_block) { | |
1142 | if ((error = xlog_bread(log, start_block, 1, bp))) { | |
1143 | xlog_put_bp(bp); | |
1144 | return error; | |
1145 | } | |
1146 | j = start_block - balign; | |
1147 | } | |
1148 | ||
1149 | for (i = start_block; i < end_block; i += bufblks) { | |
1150 | int bcount, endcount; | |
1151 | ||
1152 | bcount = min(bufblks, end_block - start_block); | |
1153 | endcount = bcount - j; | |
1154 | ||
1155 | /* We may need to do a read at the end to fill in part of | |
1156 | * the buffer in the final sector not covered by the write. | |
1157 | * If this is the same sector as the above read, skip it. | |
1158 | */ | |
1159 | ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block); | |
1160 | if (j == 0 && (start_block + endcount > ealign)) { | |
1161 | offset = XFS_BUF_PTR(bp); | |
1162 | balign = BBTOB(ealign - start_block); | |
1163 | XFS_BUF_SET_PTR(bp, offset + balign, BBTOB(sectbb)); | |
1164 | if ((error = xlog_bread(log, ealign, sectbb, bp))) | |
1165 | break; | |
1166 | XFS_BUF_SET_PTR(bp, offset, bufblks); | |
1167 | } | |
1168 | ||
1169 | offset = xlog_align(log, start_block, endcount, bp); | |
1170 | for (; j < endcount; j++) { | |
1171 | xlog_add_record(log, offset, cycle, i+j, | |
1172 | tail_cycle, tail_block); | |
1173 | offset += BBSIZE; | |
1174 | } | |
1175 | error = xlog_bwrite(log, start_block, endcount, bp); | |
1176 | if (error) | |
1177 | break; | |
1178 | start_block += endcount; | |
1179 | j = 0; | |
1180 | } | |
1181 | xlog_put_bp(bp); | |
1182 | return error; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * This routine is called to blow away any incomplete log writes out | |
1187 | * in front of the log head. We do this so that we won't become confused | |
1188 | * if we come up, write only a little bit more, and then crash again. | |
1189 | * If we leave the partial log records out there, this situation could | |
1190 | * cause us to think those partial writes are valid blocks since they | |
1191 | * have the current cycle number. We get rid of them by overwriting them | |
1192 | * with empty log records with the old cycle number rather than the | |
1193 | * current one. | |
1194 | * | |
1195 | * The tail lsn is passed in rather than taken from | |
1196 | * the log so that we will not write over the unmount record after a | |
1197 | * clean unmount in a 512 block log. Doing so would leave the log without | |
1198 | * any valid log records in it until a new one was written. If we crashed | |
1199 | * during that time we would not be able to recover. | |
1200 | */ | |
1201 | STATIC int | |
1202 | xlog_clear_stale_blocks( | |
1203 | xlog_t *log, | |
1204 | xfs_lsn_t tail_lsn) | |
1205 | { | |
1206 | int tail_cycle, head_cycle; | |
1207 | int tail_block, head_block; | |
1208 | int tail_distance, max_distance; | |
1209 | int distance; | |
1210 | int error; | |
1211 | ||
1212 | tail_cycle = CYCLE_LSN(tail_lsn); | |
1213 | tail_block = BLOCK_LSN(tail_lsn); | |
1214 | head_cycle = log->l_curr_cycle; | |
1215 | head_block = log->l_curr_block; | |
1216 | ||
1217 | /* | |
1218 | * Figure out the distance between the new head of the log | |
1219 | * and the tail. We want to write over any blocks beyond the | |
1220 | * head that we may have written just before the crash, but | |
1221 | * we don't want to overwrite the tail of the log. | |
1222 | */ | |
1223 | if (head_cycle == tail_cycle) { | |
1224 | /* | |
1225 | * The tail is behind the head in the physical log, | |
1226 | * so the distance from the head to the tail is the | |
1227 | * distance from the head to the end of the log plus | |
1228 | * the distance from the beginning of the log to the | |
1229 | * tail. | |
1230 | */ | |
1231 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { | |
1232 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", | |
1233 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1234 | return XFS_ERROR(EFSCORRUPTED); | |
1235 | } | |
1236 | tail_distance = tail_block + (log->l_logBBsize - head_block); | |
1237 | } else { | |
1238 | /* | |
1239 | * The head is behind the tail in the physical log, | |
1240 | * so the distance from the head to the tail is just | |
1241 | * the tail block minus the head block. | |
1242 | */ | |
1243 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ | |
1244 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", | |
1245 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1246 | return XFS_ERROR(EFSCORRUPTED); | |
1247 | } | |
1248 | tail_distance = tail_block - head_block; | |
1249 | } | |
1250 | ||
1251 | /* | |
1252 | * If the head is right up against the tail, we can't clear | |
1253 | * anything. | |
1254 | */ | |
1255 | if (tail_distance <= 0) { | |
1256 | ASSERT(tail_distance == 0); | |
1257 | return 0; | |
1258 | } | |
1259 | ||
1260 | max_distance = XLOG_TOTAL_REC_SHIFT(log); | |
1261 | /* | |
1262 | * Take the smaller of the maximum amount of outstanding I/O | |
1263 | * we could have and the distance to the tail to clear out. | |
1264 | * We take the smaller so that we don't overwrite the tail and | |
1265 | * we don't waste all day writing from the head to the tail | |
1266 | * for no reason. | |
1267 | */ | |
1268 | max_distance = MIN(max_distance, tail_distance); | |
1269 | ||
1270 | if ((head_block + max_distance) <= log->l_logBBsize) { | |
1271 | /* | |
1272 | * We can stomp all the blocks we need to without | |
1273 | * wrapping around the end of the log. Just do it | |
1274 | * in a single write. Use the cycle number of the | |
1275 | * current cycle minus one so that the log will look like: | |
1276 | * n ... | n - 1 ... | |
1277 | */ | |
1278 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1279 | head_block, max_distance, tail_cycle, | |
1280 | tail_block); | |
1281 | if (error) | |
1282 | return error; | |
1283 | } else { | |
1284 | /* | |
1285 | * We need to wrap around the end of the physical log in | |
1286 | * order to clear all the blocks. Do it in two separate | |
1287 | * I/Os. The first write should be from the head to the | |
1288 | * end of the physical log, and it should use the current | |
1289 | * cycle number minus one just like above. | |
1290 | */ | |
1291 | distance = log->l_logBBsize - head_block; | |
1292 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1293 | head_block, distance, tail_cycle, | |
1294 | tail_block); | |
1295 | ||
1296 | if (error) | |
1297 | return error; | |
1298 | ||
1299 | /* | |
1300 | * Now write the blocks at the start of the physical log. | |
1301 | * This writes the remainder of the blocks we want to clear. | |
1302 | * It uses the current cycle number since we're now on the | |
1303 | * same cycle as the head so that we get: | |
1304 | * n ... n ... | n - 1 ... | |
1305 | * ^^^^^ blocks we're writing | |
1306 | */ | |
1307 | distance = max_distance - (log->l_logBBsize - head_block); | |
1308 | error = xlog_write_log_records(log, head_cycle, 0, distance, | |
1309 | tail_cycle, tail_block); | |
1310 | if (error) | |
1311 | return error; | |
1312 | } | |
1313 | ||
1314 | return 0; | |
1315 | } | |
1316 | ||
1317 | /****************************************************************************** | |
1318 | * | |
1319 | * Log recover routines | |
1320 | * | |
1321 | ****************************************************************************** | |
1322 | */ | |
1323 | ||
1324 | STATIC xlog_recover_t * | |
1325 | xlog_recover_find_tid( | |
1326 | xlog_recover_t *q, | |
1327 | xlog_tid_t tid) | |
1328 | { | |
1329 | xlog_recover_t *p = q; | |
1330 | ||
1331 | while (p != NULL) { | |
1332 | if (p->r_log_tid == tid) | |
1333 | break; | |
1334 | p = p->r_next; | |
1335 | } | |
1336 | return p; | |
1337 | } | |
1338 | ||
1339 | STATIC void | |
1340 | xlog_recover_put_hashq( | |
1341 | xlog_recover_t **q, | |
1342 | xlog_recover_t *trans) | |
1343 | { | |
1344 | trans->r_next = *q; | |
1345 | *q = trans; | |
1346 | } | |
1347 | ||
1348 | STATIC void | |
1349 | xlog_recover_add_item( | |
1350 | xlog_recover_item_t **itemq) | |
1351 | { | |
1352 | xlog_recover_item_t *item; | |
1353 | ||
1354 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); | |
1355 | xlog_recover_insert_item_backq(itemq, item); | |
1356 | } | |
1357 | ||
1358 | STATIC int | |
1359 | xlog_recover_add_to_cont_trans( | |
1360 | xlog_recover_t *trans, | |
1361 | xfs_caddr_t dp, | |
1362 | int len) | |
1363 | { | |
1364 | xlog_recover_item_t *item; | |
1365 | xfs_caddr_t ptr, old_ptr; | |
1366 | int old_len; | |
1367 | ||
1368 | item = trans->r_itemq; | |
1369 | if (item == 0) { | |
1370 | /* finish copying rest of trans header */ | |
1371 | xlog_recover_add_item(&trans->r_itemq); | |
1372 | ptr = (xfs_caddr_t) &trans->r_theader + | |
1373 | sizeof(xfs_trans_header_t) - len; | |
1374 | memcpy(ptr, dp, len); /* d, s, l */ | |
1375 | return 0; | |
1376 | } | |
1377 | item = item->ri_prev; | |
1378 | ||
1379 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; | |
1380 | old_len = item->ri_buf[item->ri_cnt-1].i_len; | |
1381 | ||
760dea67 | 1382 | ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u); |
1da177e4 LT |
1383 | memcpy(&ptr[old_len], dp, len); /* d, s, l */ |
1384 | item->ri_buf[item->ri_cnt-1].i_len += len; | |
1385 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; | |
1386 | return 0; | |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * The next region to add is the start of a new region. It could be | |
1391 | * a whole region or it could be the first part of a new region. Because | |
1392 | * of this, the assumption here is that the type and size fields of all | |
1393 | * format structures fit into the first 32 bits of the structure. | |
1394 | * | |
1395 | * This works because all regions must be 32 bit aligned. Therefore, we | |
1396 | * either have both fields or we have neither field. In the case we have | |
1397 | * neither field, the data part of the region is zero length. We only have | |
1398 | * a log_op_header and can throw away the header since a new one will appear | |
1399 | * later. If we have at least 4 bytes, then we can determine how many regions | |
1400 | * will appear in the current log item. | |
1401 | */ | |
1402 | STATIC int | |
1403 | xlog_recover_add_to_trans( | |
1404 | xlog_recover_t *trans, | |
1405 | xfs_caddr_t dp, | |
1406 | int len) | |
1407 | { | |
1408 | xfs_inode_log_format_t *in_f; /* any will do */ | |
1409 | xlog_recover_item_t *item; | |
1410 | xfs_caddr_t ptr; | |
1411 | ||
1412 | if (!len) | |
1413 | return 0; | |
1414 | item = trans->r_itemq; | |
1415 | if (item == 0) { | |
1416 | ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC); | |
1417 | if (len == sizeof(xfs_trans_header_t)) | |
1418 | xlog_recover_add_item(&trans->r_itemq); | |
1419 | memcpy(&trans->r_theader, dp, len); /* d, s, l */ | |
1420 | return 0; | |
1421 | } | |
1422 | ||
1423 | ptr = kmem_alloc(len, KM_SLEEP); | |
1424 | memcpy(ptr, dp, len); | |
1425 | in_f = (xfs_inode_log_format_t *)ptr; | |
1426 | ||
1427 | if (item->ri_prev->ri_total != 0 && | |
1428 | item->ri_prev->ri_total == item->ri_prev->ri_cnt) { | |
1429 | xlog_recover_add_item(&trans->r_itemq); | |
1430 | } | |
1431 | item = trans->r_itemq; | |
1432 | item = item->ri_prev; | |
1433 | ||
1434 | if (item->ri_total == 0) { /* first region to be added */ | |
1435 | item->ri_total = in_f->ilf_size; | |
1436 | ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM); | |
1437 | item->ri_buf = kmem_zalloc((item->ri_total * | |
1438 | sizeof(xfs_log_iovec_t)), KM_SLEEP); | |
1439 | } | |
1440 | ASSERT(item->ri_total > item->ri_cnt); | |
1441 | /* Description region is ri_buf[0] */ | |
1442 | item->ri_buf[item->ri_cnt].i_addr = ptr; | |
1443 | item->ri_buf[item->ri_cnt].i_len = len; | |
1444 | item->ri_cnt++; | |
1445 | return 0; | |
1446 | } | |
1447 | ||
1448 | STATIC void | |
1449 | xlog_recover_new_tid( | |
1450 | xlog_recover_t **q, | |
1451 | xlog_tid_t tid, | |
1452 | xfs_lsn_t lsn) | |
1453 | { | |
1454 | xlog_recover_t *trans; | |
1455 | ||
1456 | trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP); | |
1457 | trans->r_log_tid = tid; | |
1458 | trans->r_lsn = lsn; | |
1459 | xlog_recover_put_hashq(q, trans); | |
1460 | } | |
1461 | ||
1462 | STATIC int | |
1463 | xlog_recover_unlink_tid( | |
1464 | xlog_recover_t **q, | |
1465 | xlog_recover_t *trans) | |
1466 | { | |
1467 | xlog_recover_t *tp; | |
1468 | int found = 0; | |
1469 | ||
1470 | ASSERT(trans != 0); | |
1471 | if (trans == *q) { | |
1472 | *q = (*q)->r_next; | |
1473 | } else { | |
1474 | tp = *q; | |
1475 | while (tp != 0) { | |
1476 | if (tp->r_next == trans) { | |
1477 | found = 1; | |
1478 | break; | |
1479 | } | |
1480 | tp = tp->r_next; | |
1481 | } | |
1482 | if (!found) { | |
1483 | xlog_warn( | |
1484 | "XFS: xlog_recover_unlink_tid: trans not found"); | |
1485 | ASSERT(0); | |
1486 | return XFS_ERROR(EIO); | |
1487 | } | |
1488 | tp->r_next = tp->r_next->r_next; | |
1489 | } | |
1490 | return 0; | |
1491 | } | |
1492 | ||
1493 | STATIC void | |
1494 | xlog_recover_insert_item_backq( | |
1495 | xlog_recover_item_t **q, | |
1496 | xlog_recover_item_t *item) | |
1497 | { | |
1498 | if (*q == 0) { | |
1499 | item->ri_prev = item->ri_next = item; | |
1500 | *q = item; | |
1501 | } else { | |
1502 | item->ri_next = *q; | |
1503 | item->ri_prev = (*q)->ri_prev; | |
1504 | (*q)->ri_prev = item; | |
1505 | item->ri_prev->ri_next = item; | |
1506 | } | |
1507 | } | |
1508 | ||
1509 | STATIC void | |
1510 | xlog_recover_insert_item_frontq( | |
1511 | xlog_recover_item_t **q, | |
1512 | xlog_recover_item_t *item) | |
1513 | { | |
1514 | xlog_recover_insert_item_backq(q, item); | |
1515 | *q = item; | |
1516 | } | |
1517 | ||
1518 | STATIC int | |
1519 | xlog_recover_reorder_trans( | |
1da177e4 LT |
1520 | xlog_recover_t *trans) |
1521 | { | |
1522 | xlog_recover_item_t *first_item, *itemq, *itemq_next; | |
1523 | xfs_buf_log_format_t *buf_f; | |
1da177e4 LT |
1524 | ushort flags = 0; |
1525 | ||
1526 | first_item = itemq = trans->r_itemq; | |
1527 | trans->r_itemq = NULL; | |
1528 | do { | |
1529 | itemq_next = itemq->ri_next; | |
1530 | buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr; | |
1da177e4 LT |
1531 | |
1532 | switch (ITEM_TYPE(itemq)) { | |
1533 | case XFS_LI_BUF: | |
804195b6 | 1534 | flags = buf_f->blf_flags; |
1da177e4 LT |
1535 | if (!(flags & XFS_BLI_CANCEL)) { |
1536 | xlog_recover_insert_item_frontq(&trans->r_itemq, | |
1537 | itemq); | |
1538 | break; | |
1539 | } | |
1540 | case XFS_LI_INODE: | |
1da177e4 LT |
1541 | case XFS_LI_DQUOT: |
1542 | case XFS_LI_QUOTAOFF: | |
1543 | case XFS_LI_EFD: | |
1544 | case XFS_LI_EFI: | |
1545 | xlog_recover_insert_item_backq(&trans->r_itemq, itemq); | |
1546 | break; | |
1547 | default: | |
1548 | xlog_warn( | |
1549 | "XFS: xlog_recover_reorder_trans: unrecognized type of log operation"); | |
1550 | ASSERT(0); | |
1551 | return XFS_ERROR(EIO); | |
1552 | } | |
1553 | itemq = itemq_next; | |
1554 | } while (first_item != itemq); | |
1555 | return 0; | |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * Build up the table of buf cancel records so that we don't replay | |
1560 | * cancelled data in the second pass. For buffer records that are | |
1561 | * not cancel records, there is nothing to do here so we just return. | |
1562 | * | |
1563 | * If we get a cancel record which is already in the table, this indicates | |
1564 | * that the buffer was cancelled multiple times. In order to ensure | |
1565 | * that during pass 2 we keep the record in the table until we reach its | |
1566 | * last occurrence in the log, we keep a reference count in the cancel | |
1567 | * record in the table to tell us how many times we expect to see this | |
1568 | * record during the second pass. | |
1569 | */ | |
1570 | STATIC void | |
1571 | xlog_recover_do_buffer_pass1( | |
1572 | xlog_t *log, | |
1573 | xfs_buf_log_format_t *buf_f) | |
1574 | { | |
1575 | xfs_buf_cancel_t *bcp; | |
1576 | xfs_buf_cancel_t *nextp; | |
1577 | xfs_buf_cancel_t *prevp; | |
1578 | xfs_buf_cancel_t **bucket; | |
1da177e4 LT |
1579 | xfs_daddr_t blkno = 0; |
1580 | uint len = 0; | |
1581 | ushort flags = 0; | |
1582 | ||
1583 | switch (buf_f->blf_type) { | |
1584 | case XFS_LI_BUF: | |
1585 | blkno = buf_f->blf_blkno; | |
1586 | len = buf_f->blf_len; | |
1587 | flags = buf_f->blf_flags; | |
1588 | break; | |
1da177e4 LT |
1589 | } |
1590 | ||
1591 | /* | |
1592 | * If this isn't a cancel buffer item, then just return. | |
1593 | */ | |
1594 | if (!(flags & XFS_BLI_CANCEL)) | |
1595 | return; | |
1596 | ||
1597 | /* | |
1598 | * Insert an xfs_buf_cancel record into the hash table of | |
1599 | * them. If there is already an identical record, bump | |
1600 | * its reference count. | |
1601 | */ | |
1602 | bucket = &log->l_buf_cancel_table[(__uint64_t)blkno % | |
1603 | XLOG_BC_TABLE_SIZE]; | |
1604 | /* | |
1605 | * If the hash bucket is empty then just insert a new record into | |
1606 | * the bucket. | |
1607 | */ | |
1608 | if (*bucket == NULL) { | |
1609 | bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t), | |
1610 | KM_SLEEP); | |
1611 | bcp->bc_blkno = blkno; | |
1612 | bcp->bc_len = len; | |
1613 | bcp->bc_refcount = 1; | |
1614 | bcp->bc_next = NULL; | |
1615 | *bucket = bcp; | |
1616 | return; | |
1617 | } | |
1618 | ||
1619 | /* | |
1620 | * The hash bucket is not empty, so search for duplicates of our | |
1621 | * record. If we find one them just bump its refcount. If not | |
1622 | * then add us at the end of the list. | |
1623 | */ | |
1624 | prevp = NULL; | |
1625 | nextp = *bucket; | |
1626 | while (nextp != NULL) { | |
1627 | if (nextp->bc_blkno == blkno && nextp->bc_len == len) { | |
1628 | nextp->bc_refcount++; | |
1629 | return; | |
1630 | } | |
1631 | prevp = nextp; | |
1632 | nextp = nextp->bc_next; | |
1633 | } | |
1634 | ASSERT(prevp != NULL); | |
1635 | bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t), | |
1636 | KM_SLEEP); | |
1637 | bcp->bc_blkno = blkno; | |
1638 | bcp->bc_len = len; | |
1639 | bcp->bc_refcount = 1; | |
1640 | bcp->bc_next = NULL; | |
1641 | prevp->bc_next = bcp; | |
1642 | } | |
1643 | ||
1644 | /* | |
1645 | * Check to see whether the buffer being recovered has a corresponding | |
1646 | * entry in the buffer cancel record table. If it does then return 1 | |
1647 | * so that it will be cancelled, otherwise return 0. If the buffer is | |
1648 | * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement | |
1649 | * the refcount on the entry in the table and remove it from the table | |
1650 | * if this is the last reference. | |
1651 | * | |
1652 | * We remove the cancel record from the table when we encounter its | |
1653 | * last occurrence in the log so that if the same buffer is re-used | |
1654 | * again after its last cancellation we actually replay the changes | |
1655 | * made at that point. | |
1656 | */ | |
1657 | STATIC int | |
1658 | xlog_check_buffer_cancelled( | |
1659 | xlog_t *log, | |
1660 | xfs_daddr_t blkno, | |
1661 | uint len, | |
1662 | ushort flags) | |
1663 | { | |
1664 | xfs_buf_cancel_t *bcp; | |
1665 | xfs_buf_cancel_t *prevp; | |
1666 | xfs_buf_cancel_t **bucket; | |
1667 | ||
1668 | if (log->l_buf_cancel_table == NULL) { | |
1669 | /* | |
1670 | * There is nothing in the table built in pass one, | |
1671 | * so this buffer must not be cancelled. | |
1672 | */ | |
1673 | ASSERT(!(flags & XFS_BLI_CANCEL)); | |
1674 | return 0; | |
1675 | } | |
1676 | ||
1677 | bucket = &log->l_buf_cancel_table[(__uint64_t)blkno % | |
1678 | XLOG_BC_TABLE_SIZE]; | |
1679 | bcp = *bucket; | |
1680 | if (bcp == NULL) { | |
1681 | /* | |
1682 | * There is no corresponding entry in the table built | |
1683 | * in pass one, so this buffer has not been cancelled. | |
1684 | */ | |
1685 | ASSERT(!(flags & XFS_BLI_CANCEL)); | |
1686 | return 0; | |
1687 | } | |
1688 | ||
1689 | /* | |
1690 | * Search for an entry in the buffer cancel table that | |
1691 | * matches our buffer. | |
1692 | */ | |
1693 | prevp = NULL; | |
1694 | while (bcp != NULL) { | |
1695 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) { | |
1696 | /* | |
1697 | * We've go a match, so return 1 so that the | |
1698 | * recovery of this buffer is cancelled. | |
1699 | * If this buffer is actually a buffer cancel | |
1700 | * log item, then decrement the refcount on the | |
1701 | * one in the table and remove it if this is the | |
1702 | * last reference. | |
1703 | */ | |
1704 | if (flags & XFS_BLI_CANCEL) { | |
1705 | bcp->bc_refcount--; | |
1706 | if (bcp->bc_refcount == 0) { | |
1707 | if (prevp == NULL) { | |
1708 | *bucket = bcp->bc_next; | |
1709 | } else { | |
1710 | prevp->bc_next = bcp->bc_next; | |
1711 | } | |
1712 | kmem_free(bcp, | |
1713 | sizeof(xfs_buf_cancel_t)); | |
1714 | } | |
1715 | } | |
1716 | return 1; | |
1717 | } | |
1718 | prevp = bcp; | |
1719 | bcp = bcp->bc_next; | |
1720 | } | |
1721 | /* | |
1722 | * We didn't find a corresponding entry in the table, so | |
1723 | * return 0 so that the buffer is NOT cancelled. | |
1724 | */ | |
1725 | ASSERT(!(flags & XFS_BLI_CANCEL)); | |
1726 | return 0; | |
1727 | } | |
1728 | ||
1729 | STATIC int | |
1730 | xlog_recover_do_buffer_pass2( | |
1731 | xlog_t *log, | |
1732 | xfs_buf_log_format_t *buf_f) | |
1733 | { | |
1da177e4 LT |
1734 | xfs_daddr_t blkno = 0; |
1735 | ushort flags = 0; | |
1736 | uint len = 0; | |
1737 | ||
1738 | switch (buf_f->blf_type) { | |
1739 | case XFS_LI_BUF: | |
1740 | blkno = buf_f->blf_blkno; | |
1741 | flags = buf_f->blf_flags; | |
1742 | len = buf_f->blf_len; | |
1743 | break; | |
1da177e4 LT |
1744 | } |
1745 | ||
1746 | return xlog_check_buffer_cancelled(log, blkno, len, flags); | |
1747 | } | |
1748 | ||
1749 | /* | |
1750 | * Perform recovery for a buffer full of inodes. In these buffers, | |
1751 | * the only data which should be recovered is that which corresponds | |
1752 | * to the di_next_unlinked pointers in the on disk inode structures. | |
1753 | * The rest of the data for the inodes is always logged through the | |
1754 | * inodes themselves rather than the inode buffer and is recovered | |
1755 | * in xlog_recover_do_inode_trans(). | |
1756 | * | |
1757 | * The only time when buffers full of inodes are fully recovered is | |
1758 | * when the buffer is full of newly allocated inodes. In this case | |
1759 | * the buffer will not be marked as an inode buffer and so will be | |
1760 | * sent to xlog_recover_do_reg_buffer() below during recovery. | |
1761 | */ | |
1762 | STATIC int | |
1763 | xlog_recover_do_inode_buffer( | |
1764 | xfs_mount_t *mp, | |
1765 | xlog_recover_item_t *item, | |
1766 | xfs_buf_t *bp, | |
1767 | xfs_buf_log_format_t *buf_f) | |
1768 | { | |
1769 | int i; | |
1770 | int item_index; | |
1771 | int bit; | |
1772 | int nbits; | |
1773 | int reg_buf_offset; | |
1774 | int reg_buf_bytes; | |
1775 | int next_unlinked_offset; | |
1776 | int inodes_per_buf; | |
1777 | xfs_agino_t *logged_nextp; | |
1778 | xfs_agino_t *buffer_nextp; | |
1da177e4 LT |
1779 | unsigned int *data_map = NULL; |
1780 | unsigned int map_size = 0; | |
1781 | ||
1782 | switch (buf_f->blf_type) { | |
1783 | case XFS_LI_BUF: | |
1784 | data_map = buf_f->blf_data_map; | |
1785 | map_size = buf_f->blf_map_size; | |
1786 | break; | |
1da177e4 LT |
1787 | } |
1788 | /* | |
1789 | * Set the variables corresponding to the current region to | |
1790 | * 0 so that we'll initialize them on the first pass through | |
1791 | * the loop. | |
1792 | */ | |
1793 | reg_buf_offset = 0; | |
1794 | reg_buf_bytes = 0; | |
1795 | bit = 0; | |
1796 | nbits = 0; | |
1797 | item_index = 0; | |
1798 | inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog; | |
1799 | for (i = 0; i < inodes_per_buf; i++) { | |
1800 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + | |
1801 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1802 | ||
1803 | while (next_unlinked_offset >= | |
1804 | (reg_buf_offset + reg_buf_bytes)) { | |
1805 | /* | |
1806 | * The next di_next_unlinked field is beyond | |
1807 | * the current logged region. Find the next | |
1808 | * logged region that contains or is beyond | |
1809 | * the current di_next_unlinked field. | |
1810 | */ | |
1811 | bit += nbits; | |
1812 | bit = xfs_next_bit(data_map, map_size, bit); | |
1813 | ||
1814 | /* | |
1815 | * If there are no more logged regions in the | |
1816 | * buffer, then we're done. | |
1817 | */ | |
1818 | if (bit == -1) { | |
1819 | return 0; | |
1820 | } | |
1821 | ||
1822 | nbits = xfs_contig_bits(data_map, map_size, | |
1823 | bit); | |
1824 | ASSERT(nbits > 0); | |
1825 | reg_buf_offset = bit << XFS_BLI_SHIFT; | |
1826 | reg_buf_bytes = nbits << XFS_BLI_SHIFT; | |
1827 | item_index++; | |
1828 | } | |
1829 | ||
1830 | /* | |
1831 | * If the current logged region starts after the current | |
1832 | * di_next_unlinked field, then move on to the next | |
1833 | * di_next_unlinked field. | |
1834 | */ | |
1835 | if (next_unlinked_offset < reg_buf_offset) { | |
1836 | continue; | |
1837 | } | |
1838 | ||
1839 | ASSERT(item->ri_buf[item_index].i_addr != NULL); | |
1840 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0); | |
1841 | ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp)); | |
1842 | ||
1843 | /* | |
1844 | * The current logged region contains a copy of the | |
1845 | * current di_next_unlinked field. Extract its value | |
1846 | * and copy it to the buffer copy. | |
1847 | */ | |
1848 | logged_nextp = (xfs_agino_t *) | |
1849 | ((char *)(item->ri_buf[item_index].i_addr) + | |
1850 | (next_unlinked_offset - reg_buf_offset)); | |
1851 | if (unlikely(*logged_nextp == 0)) { | |
1852 | xfs_fs_cmn_err(CE_ALERT, mp, | |
1853 | "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field", | |
1854 | item, bp); | |
1855 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", | |
1856 | XFS_ERRLEVEL_LOW, mp); | |
1857 | return XFS_ERROR(EFSCORRUPTED); | |
1858 | } | |
1859 | ||
1860 | buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp, | |
1861 | next_unlinked_offset); | |
87c199c2 | 1862 | *buffer_nextp = *logged_nextp; |
1da177e4 LT |
1863 | } |
1864 | ||
1865 | return 0; | |
1866 | } | |
1867 | ||
1868 | /* | |
1869 | * Perform a 'normal' buffer recovery. Each logged region of the | |
1870 | * buffer should be copied over the corresponding region in the | |
1871 | * given buffer. The bitmap in the buf log format structure indicates | |
1872 | * where to place the logged data. | |
1873 | */ | |
1874 | /*ARGSUSED*/ | |
1875 | STATIC void | |
1876 | xlog_recover_do_reg_buffer( | |
1da177e4 LT |
1877 | xlog_recover_item_t *item, |
1878 | xfs_buf_t *bp, | |
1879 | xfs_buf_log_format_t *buf_f) | |
1880 | { | |
1881 | int i; | |
1882 | int bit; | |
1883 | int nbits; | |
1da177e4 LT |
1884 | unsigned int *data_map = NULL; |
1885 | unsigned int map_size = 0; | |
1886 | int error; | |
1887 | ||
1888 | switch (buf_f->blf_type) { | |
1889 | case XFS_LI_BUF: | |
1890 | data_map = buf_f->blf_data_map; | |
1891 | map_size = buf_f->blf_map_size; | |
1892 | break; | |
1da177e4 LT |
1893 | } |
1894 | bit = 0; | |
1895 | i = 1; /* 0 is the buf format structure */ | |
1896 | while (1) { | |
1897 | bit = xfs_next_bit(data_map, map_size, bit); | |
1898 | if (bit == -1) | |
1899 | break; | |
1900 | nbits = xfs_contig_bits(data_map, map_size, bit); | |
1901 | ASSERT(nbits > 0); | |
1902 | ASSERT(item->ri_buf[i].i_addr != 0); | |
1903 | ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0); | |
1904 | ASSERT(XFS_BUF_COUNT(bp) >= | |
1905 | ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT)); | |
1906 | ||
1907 | /* | |
1908 | * Do a sanity check if this is a dquot buffer. Just checking | |
1909 | * the first dquot in the buffer should do. XXXThis is | |
1910 | * probably a good thing to do for other buf types also. | |
1911 | */ | |
1912 | error = 0; | |
c8ad20ff NS |
1913 | if (buf_f->blf_flags & |
1914 | (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) { | |
1da177e4 LT |
1915 | error = xfs_qm_dqcheck((xfs_disk_dquot_t *) |
1916 | item->ri_buf[i].i_addr, | |
1917 | -1, 0, XFS_QMOPT_DOWARN, | |
1918 | "dquot_buf_recover"); | |
1919 | } | |
1920 | if (!error) | |
1921 | memcpy(xfs_buf_offset(bp, | |
1922 | (uint)bit << XFS_BLI_SHIFT), /* dest */ | |
1923 | item->ri_buf[i].i_addr, /* source */ | |
1924 | nbits<<XFS_BLI_SHIFT); /* length */ | |
1925 | i++; | |
1926 | bit += nbits; | |
1927 | } | |
1928 | ||
1929 | /* Shouldn't be any more regions */ | |
1930 | ASSERT(i == item->ri_total); | |
1931 | } | |
1932 | ||
1933 | /* | |
1934 | * Do some primitive error checking on ondisk dquot data structures. | |
1935 | */ | |
1936 | int | |
1937 | xfs_qm_dqcheck( | |
1938 | xfs_disk_dquot_t *ddq, | |
1939 | xfs_dqid_t id, | |
1940 | uint type, /* used only when IO_dorepair is true */ | |
1941 | uint flags, | |
1942 | char *str) | |
1943 | { | |
1944 | xfs_dqblk_t *d = (xfs_dqblk_t *)ddq; | |
1945 | int errs = 0; | |
1946 | ||
1947 | /* | |
1948 | * We can encounter an uninitialized dquot buffer for 2 reasons: | |
1949 | * 1. If we crash while deleting the quotainode(s), and those blks got | |
1950 | * used for user data. This is because we take the path of regular | |
1951 | * file deletion; however, the size field of quotainodes is never | |
1952 | * updated, so all the tricks that we play in itruncate_finish | |
1953 | * don't quite matter. | |
1954 | * | |
1955 | * 2. We don't play the quota buffers when there's a quotaoff logitem. | |
1956 | * But the allocation will be replayed so we'll end up with an | |
1957 | * uninitialized quota block. | |
1958 | * | |
1959 | * This is all fine; things are still consistent, and we haven't lost | |
1960 | * any quota information. Just don't complain about bad dquot blks. | |
1961 | */ | |
1149d96a | 1962 | if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) { |
1da177e4 LT |
1963 | if (flags & XFS_QMOPT_DOWARN) |
1964 | cmn_err(CE_ALERT, | |
1965 | "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x", | |
1149d96a | 1966 | str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC); |
1da177e4 LT |
1967 | errs++; |
1968 | } | |
1149d96a | 1969 | if (ddq->d_version != XFS_DQUOT_VERSION) { |
1da177e4 LT |
1970 | if (flags & XFS_QMOPT_DOWARN) |
1971 | cmn_err(CE_ALERT, | |
1972 | "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x", | |
1149d96a | 1973 | str, id, ddq->d_version, XFS_DQUOT_VERSION); |
1da177e4 LT |
1974 | errs++; |
1975 | } | |
1976 | ||
1149d96a CH |
1977 | if (ddq->d_flags != XFS_DQ_USER && |
1978 | ddq->d_flags != XFS_DQ_PROJ && | |
1979 | ddq->d_flags != XFS_DQ_GROUP) { | |
1da177e4 LT |
1980 | if (flags & XFS_QMOPT_DOWARN) |
1981 | cmn_err(CE_ALERT, | |
1982 | "%s : XFS dquot ID 0x%x, unknown flags 0x%x", | |
1149d96a | 1983 | str, id, ddq->d_flags); |
1da177e4 LT |
1984 | errs++; |
1985 | } | |
1986 | ||
1149d96a | 1987 | if (id != -1 && id != be32_to_cpu(ddq->d_id)) { |
1da177e4 LT |
1988 | if (flags & XFS_QMOPT_DOWARN) |
1989 | cmn_err(CE_ALERT, | |
1990 | "%s : ondisk-dquot 0x%p, ID mismatch: " | |
1991 | "0x%x expected, found id 0x%x", | |
1149d96a | 1992 | str, ddq, id, be32_to_cpu(ddq->d_id)); |
1da177e4 LT |
1993 | errs++; |
1994 | } | |
1995 | ||
1996 | if (!errs && ddq->d_id) { | |
1149d96a CH |
1997 | if (ddq->d_blk_softlimit && |
1998 | be64_to_cpu(ddq->d_bcount) >= | |
1999 | be64_to_cpu(ddq->d_blk_softlimit)) { | |
1da177e4 LT |
2000 | if (!ddq->d_btimer) { |
2001 | if (flags & XFS_QMOPT_DOWARN) | |
2002 | cmn_err(CE_ALERT, | |
2003 | "%s : Dquot ID 0x%x (0x%p) " | |
2004 | "BLK TIMER NOT STARTED", | |
1149d96a | 2005 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2006 | errs++; |
2007 | } | |
2008 | } | |
1149d96a CH |
2009 | if (ddq->d_ino_softlimit && |
2010 | be64_to_cpu(ddq->d_icount) >= | |
2011 | be64_to_cpu(ddq->d_ino_softlimit)) { | |
1da177e4 LT |
2012 | if (!ddq->d_itimer) { |
2013 | if (flags & XFS_QMOPT_DOWARN) | |
2014 | cmn_err(CE_ALERT, | |
2015 | "%s : Dquot ID 0x%x (0x%p) " | |
2016 | "INODE TIMER NOT STARTED", | |
1149d96a | 2017 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2018 | errs++; |
2019 | } | |
2020 | } | |
1149d96a CH |
2021 | if (ddq->d_rtb_softlimit && |
2022 | be64_to_cpu(ddq->d_rtbcount) >= | |
2023 | be64_to_cpu(ddq->d_rtb_softlimit)) { | |
1da177e4 LT |
2024 | if (!ddq->d_rtbtimer) { |
2025 | if (flags & XFS_QMOPT_DOWARN) | |
2026 | cmn_err(CE_ALERT, | |
2027 | "%s : Dquot ID 0x%x (0x%p) " | |
2028 | "RTBLK TIMER NOT STARTED", | |
1149d96a | 2029 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2030 | errs++; |
2031 | } | |
2032 | } | |
2033 | } | |
2034 | ||
2035 | if (!errs || !(flags & XFS_QMOPT_DQREPAIR)) | |
2036 | return errs; | |
2037 | ||
2038 | if (flags & XFS_QMOPT_DOWARN) | |
2039 | cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id); | |
2040 | ||
2041 | /* | |
2042 | * Typically, a repair is only requested by quotacheck. | |
2043 | */ | |
2044 | ASSERT(id != -1); | |
2045 | ASSERT(flags & XFS_QMOPT_DQREPAIR); | |
2046 | memset(d, 0, sizeof(xfs_dqblk_t)); | |
1149d96a CH |
2047 | |
2048 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
2049 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
2050 | d->dd_diskdq.d_flags = type; | |
2051 | d->dd_diskdq.d_id = cpu_to_be32(id); | |
1da177e4 LT |
2052 | |
2053 | return errs; | |
2054 | } | |
2055 | ||
2056 | /* | |
2057 | * Perform a dquot buffer recovery. | |
2058 | * Simple algorithm: if we have found a QUOTAOFF logitem of the same type | |
2059 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. | |
2060 | * Else, treat it as a regular buffer and do recovery. | |
2061 | */ | |
2062 | STATIC void | |
2063 | xlog_recover_do_dquot_buffer( | |
2064 | xfs_mount_t *mp, | |
2065 | xlog_t *log, | |
2066 | xlog_recover_item_t *item, | |
2067 | xfs_buf_t *bp, | |
2068 | xfs_buf_log_format_t *buf_f) | |
2069 | { | |
2070 | uint type; | |
2071 | ||
2072 | /* | |
2073 | * Filesystems are required to send in quota flags at mount time. | |
2074 | */ | |
2075 | if (mp->m_qflags == 0) { | |
2076 | return; | |
2077 | } | |
2078 | ||
2079 | type = 0; | |
2080 | if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF) | |
2081 | type |= XFS_DQ_USER; | |
c8ad20ff NS |
2082 | if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF) |
2083 | type |= XFS_DQ_PROJ; | |
1da177e4 LT |
2084 | if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF) |
2085 | type |= XFS_DQ_GROUP; | |
2086 | /* | |
2087 | * This type of quotas was turned off, so ignore this buffer | |
2088 | */ | |
2089 | if (log->l_quotaoffs_flag & type) | |
2090 | return; | |
2091 | ||
e9ed9d22 | 2092 | xlog_recover_do_reg_buffer(item, bp, buf_f); |
1da177e4 LT |
2093 | } |
2094 | ||
2095 | /* | |
2096 | * This routine replays a modification made to a buffer at runtime. | |
2097 | * There are actually two types of buffer, regular and inode, which | |
2098 | * are handled differently. Inode buffers are handled differently | |
2099 | * in that we only recover a specific set of data from them, namely | |
2100 | * the inode di_next_unlinked fields. This is because all other inode | |
2101 | * data is actually logged via inode records and any data we replay | |
2102 | * here which overlaps that may be stale. | |
2103 | * | |
2104 | * When meta-data buffers are freed at run time we log a buffer item | |
2105 | * with the XFS_BLI_CANCEL bit set to indicate that previous copies | |
2106 | * of the buffer in the log should not be replayed at recovery time. | |
2107 | * This is so that if the blocks covered by the buffer are reused for | |
2108 | * file data before we crash we don't end up replaying old, freed | |
2109 | * meta-data into a user's file. | |
2110 | * | |
2111 | * To handle the cancellation of buffer log items, we make two passes | |
2112 | * over the log during recovery. During the first we build a table of | |
2113 | * those buffers which have been cancelled, and during the second we | |
2114 | * only replay those buffers which do not have corresponding cancel | |
2115 | * records in the table. See xlog_recover_do_buffer_pass[1,2] above | |
2116 | * for more details on the implementation of the table of cancel records. | |
2117 | */ | |
2118 | STATIC int | |
2119 | xlog_recover_do_buffer_trans( | |
2120 | xlog_t *log, | |
2121 | xlog_recover_item_t *item, | |
2122 | int pass) | |
2123 | { | |
2124 | xfs_buf_log_format_t *buf_f; | |
1da177e4 LT |
2125 | xfs_mount_t *mp; |
2126 | xfs_buf_t *bp; | |
2127 | int error; | |
2128 | int cancel; | |
2129 | xfs_daddr_t blkno; | |
2130 | int len; | |
2131 | ushort flags; | |
2132 | ||
2133 | buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr; | |
2134 | ||
2135 | if (pass == XLOG_RECOVER_PASS1) { | |
2136 | /* | |
2137 | * In this pass we're only looking for buf items | |
2138 | * with the XFS_BLI_CANCEL bit set. | |
2139 | */ | |
2140 | xlog_recover_do_buffer_pass1(log, buf_f); | |
2141 | return 0; | |
2142 | } else { | |
2143 | /* | |
2144 | * In this pass we want to recover all the buffers | |
2145 | * which have not been cancelled and are not | |
2146 | * cancellation buffers themselves. The routine | |
2147 | * we call here will tell us whether or not to | |
2148 | * continue with the replay of this buffer. | |
2149 | */ | |
2150 | cancel = xlog_recover_do_buffer_pass2(log, buf_f); | |
2151 | if (cancel) { | |
2152 | return 0; | |
2153 | } | |
2154 | } | |
2155 | switch (buf_f->blf_type) { | |
2156 | case XFS_LI_BUF: | |
2157 | blkno = buf_f->blf_blkno; | |
2158 | len = buf_f->blf_len; | |
2159 | flags = buf_f->blf_flags; | |
2160 | break; | |
1da177e4 LT |
2161 | default: |
2162 | xfs_fs_cmn_err(CE_ALERT, log->l_mp, | |
fc1f8c1c NS |
2163 | "xfs_log_recover: unknown buffer type 0x%x, logdev %s", |
2164 | buf_f->blf_type, log->l_mp->m_logname ? | |
2165 | log->l_mp->m_logname : "internal"); | |
1da177e4 LT |
2166 | XFS_ERROR_REPORT("xlog_recover_do_buffer_trans", |
2167 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2168 | return XFS_ERROR(EFSCORRUPTED); | |
2169 | } | |
2170 | ||
2171 | mp = log->l_mp; | |
2172 | if (flags & XFS_BLI_INODE_BUF) { | |
2173 | bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len, | |
2174 | XFS_BUF_LOCK); | |
2175 | } else { | |
2176 | bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0); | |
2177 | } | |
2178 | if (XFS_BUF_ISERROR(bp)) { | |
2179 | xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp, | |
2180 | bp, blkno); | |
2181 | error = XFS_BUF_GETERROR(bp); | |
2182 | xfs_buf_relse(bp); | |
2183 | return error; | |
2184 | } | |
2185 | ||
2186 | error = 0; | |
2187 | if (flags & XFS_BLI_INODE_BUF) { | |
2188 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); | |
c8ad20ff NS |
2189 | } else if (flags & |
2190 | (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) { | |
1da177e4 LT |
2191 | xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); |
2192 | } else { | |
e9ed9d22 | 2193 | xlog_recover_do_reg_buffer(item, bp, buf_f); |
1da177e4 LT |
2194 | } |
2195 | if (error) | |
2196 | return XFS_ERROR(error); | |
2197 | ||
2198 | /* | |
2199 | * Perform delayed write on the buffer. Asynchronous writes will be | |
2200 | * slower when taking into account all the buffers to be flushed. | |
2201 | * | |
2202 | * Also make sure that only inode buffers with good sizes stay in | |
2203 | * the buffer cache. The kernel moves inodes in buffers of 1 block | |
2204 | * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode | |
2205 | * buffers in the log can be a different size if the log was generated | |
2206 | * by an older kernel using unclustered inode buffers or a newer kernel | |
2207 | * running with a different inode cluster size. Regardless, if the | |
2208 | * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE) | |
2209 | * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep | |
2210 | * the buffer out of the buffer cache so that the buffer won't | |
2211 | * overlap with future reads of those inodes. | |
2212 | */ | |
2213 | if (XFS_DINODE_MAGIC == | |
2214 | INT_GET(*((__uint16_t *)(xfs_buf_offset(bp, 0))), ARCH_CONVERT) && | |
2215 | (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize, | |
2216 | (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) { | |
2217 | XFS_BUF_STALE(bp); | |
2218 | error = xfs_bwrite(mp, bp); | |
2219 | } else { | |
2220 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL || | |
2221 | XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp); | |
2222 | XFS_BUF_SET_FSPRIVATE(bp, mp); | |
2223 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone); | |
2224 | xfs_bdwrite(mp, bp); | |
2225 | } | |
2226 | ||
2227 | return (error); | |
2228 | } | |
2229 | ||
2230 | STATIC int | |
2231 | xlog_recover_do_inode_trans( | |
2232 | xlog_t *log, | |
2233 | xlog_recover_item_t *item, | |
2234 | int pass) | |
2235 | { | |
2236 | xfs_inode_log_format_t *in_f; | |
2237 | xfs_mount_t *mp; | |
2238 | xfs_buf_t *bp; | |
2239 | xfs_imap_t imap; | |
2240 | xfs_dinode_t *dip; | |
2241 | xfs_ino_t ino; | |
2242 | int len; | |
2243 | xfs_caddr_t src; | |
2244 | xfs_caddr_t dest; | |
2245 | int error; | |
2246 | int attr_index; | |
2247 | uint fields; | |
2248 | xfs_dinode_core_t *dicp; | |
6d192a9b | 2249 | int need_free = 0; |
1da177e4 LT |
2250 | |
2251 | if (pass == XLOG_RECOVER_PASS1) { | |
2252 | return 0; | |
2253 | } | |
2254 | ||
6d192a9b TS |
2255 | if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) { |
2256 | in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr; | |
2257 | } else { | |
2258 | in_f = (xfs_inode_log_format_t *)kmem_alloc( | |
2259 | sizeof(xfs_inode_log_format_t), KM_SLEEP); | |
2260 | need_free = 1; | |
2261 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); | |
2262 | if (error) | |
2263 | goto error; | |
2264 | } | |
1da177e4 LT |
2265 | ino = in_f->ilf_ino; |
2266 | mp = log->l_mp; | |
2267 | if (ITEM_TYPE(item) == XFS_LI_INODE) { | |
2268 | imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno; | |
2269 | imap.im_len = in_f->ilf_len; | |
2270 | imap.im_boffset = in_f->ilf_boffset; | |
2271 | } else { | |
2272 | /* | |
2273 | * It's an old inode format record. We don't know where | |
2274 | * its cluster is located on disk, and we can't allow | |
2275 | * xfs_imap() to figure it out because the inode btrees | |
2276 | * are not ready to be used. Therefore do not pass the | |
2277 | * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give | |
2278 | * us only the single block in which the inode lives | |
2279 | * rather than its cluster, so we must make sure to | |
2280 | * invalidate the buffer when we write it out below. | |
2281 | */ | |
2282 | imap.im_blkno = 0; | |
2283 | xfs_imap(log->l_mp, NULL, ino, &imap, 0); | |
2284 | } | |
2285 | ||
2286 | /* | |
2287 | * Inode buffers can be freed, look out for it, | |
2288 | * and do not replay the inode. | |
2289 | */ | |
6d192a9b TS |
2290 | if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) { |
2291 | error = 0; | |
2292 | goto error; | |
2293 | } | |
1da177e4 LT |
2294 | |
2295 | bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len, | |
2296 | XFS_BUF_LOCK); | |
2297 | if (XFS_BUF_ISERROR(bp)) { | |
2298 | xfs_ioerror_alert("xlog_recover_do..(read#2)", mp, | |
2299 | bp, imap.im_blkno); | |
2300 | error = XFS_BUF_GETERROR(bp); | |
2301 | xfs_buf_relse(bp); | |
6d192a9b | 2302 | goto error; |
1da177e4 LT |
2303 | } |
2304 | error = 0; | |
2305 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); | |
2306 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); | |
2307 | ||
2308 | /* | |
2309 | * Make sure the place we're flushing out to really looks | |
2310 | * like an inode! | |
2311 | */ | |
2312 | if (unlikely(INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC)) { | |
2313 | xfs_buf_relse(bp); | |
2314 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2315 | "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld", | |
2316 | dip, bp, ino); | |
2317 | XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)", | |
2318 | XFS_ERRLEVEL_LOW, mp); | |
6d192a9b TS |
2319 | error = EFSCORRUPTED; |
2320 | goto error; | |
1da177e4 LT |
2321 | } |
2322 | dicp = (xfs_dinode_core_t*)(item->ri_buf[1].i_addr); | |
2323 | if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) { | |
2324 | xfs_buf_relse(bp); | |
2325 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2326 | "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld", | |
2327 | item, ino); | |
2328 | XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)", | |
2329 | XFS_ERRLEVEL_LOW, mp); | |
6d192a9b TS |
2330 | error = EFSCORRUPTED; |
2331 | goto error; | |
1da177e4 LT |
2332 | } |
2333 | ||
2334 | /* Skip replay when the on disk inode is newer than the log one */ | |
2335 | if (dicp->di_flushiter < | |
2336 | INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT)) { | |
2337 | /* | |
2338 | * Deal with the wrap case, DI_MAX_FLUSH is less | |
2339 | * than smaller numbers | |
2340 | */ | |
2341 | if ((INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT) | |
2342 | == DI_MAX_FLUSH) && | |
2343 | (dicp->di_flushiter < (DI_MAX_FLUSH>>1))) { | |
2344 | /* do nothing */ | |
2345 | } else { | |
2346 | xfs_buf_relse(bp); | |
6d192a9b TS |
2347 | error = 0; |
2348 | goto error; | |
1da177e4 LT |
2349 | } |
2350 | } | |
2351 | /* Take the opportunity to reset the flush iteration count */ | |
2352 | dicp->di_flushiter = 0; | |
2353 | ||
2354 | if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) { | |
2355 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && | |
2356 | (dicp->di_format != XFS_DINODE_FMT_BTREE)) { | |
2357 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)", | |
2358 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2359 | xfs_buf_relse(bp); | |
2360 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2361 | "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2362 | item, dip, bp, ino); | |
6d192a9b TS |
2363 | error = EFSCORRUPTED; |
2364 | goto error; | |
1da177e4 LT |
2365 | } |
2366 | } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) { | |
2367 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && | |
2368 | (dicp->di_format != XFS_DINODE_FMT_BTREE) && | |
2369 | (dicp->di_format != XFS_DINODE_FMT_LOCAL)) { | |
2370 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)", | |
2371 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2372 | xfs_buf_relse(bp); | |
2373 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2374 | "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2375 | item, dip, bp, ino); | |
6d192a9b TS |
2376 | error = EFSCORRUPTED; |
2377 | goto error; | |
1da177e4 LT |
2378 | } |
2379 | } | |
2380 | if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){ | |
2381 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)", | |
2382 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2383 | xfs_buf_relse(bp); | |
2384 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2385 | "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", | |
2386 | item, dip, bp, ino, | |
2387 | dicp->di_nextents + dicp->di_anextents, | |
2388 | dicp->di_nblocks); | |
6d192a9b TS |
2389 | error = EFSCORRUPTED; |
2390 | goto error; | |
1da177e4 LT |
2391 | } |
2392 | if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) { | |
2393 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)", | |
2394 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2395 | xfs_buf_relse(bp); | |
2396 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2397 | "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x", | |
2398 | item, dip, bp, ino, dicp->di_forkoff); | |
6d192a9b TS |
2399 | error = EFSCORRUPTED; |
2400 | goto error; | |
1da177e4 LT |
2401 | } |
2402 | if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) { | |
2403 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)", | |
2404 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2405 | xfs_buf_relse(bp); | |
2406 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2407 | "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p", | |
2408 | item->ri_buf[1].i_len, item); | |
6d192a9b TS |
2409 | error = EFSCORRUPTED; |
2410 | goto error; | |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | /* The core is in in-core format */ | |
2414 | xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core, | |
2415 | (xfs_dinode_core_t*)item->ri_buf[1].i_addr, -1); | |
2416 | ||
2417 | /* the rest is in on-disk format */ | |
2418 | if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) { | |
2419 | memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t), | |
2420 | item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t), | |
2421 | item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t)); | |
2422 | } | |
2423 | ||
2424 | fields = in_f->ilf_fields; | |
2425 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { | |
2426 | case XFS_ILOG_DEV: | |
2427 | INT_SET(dip->di_u.di_dev, ARCH_CONVERT, in_f->ilf_u.ilfu_rdev); | |
2428 | ||
2429 | break; | |
2430 | case XFS_ILOG_UUID: | |
2431 | dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid; | |
2432 | break; | |
2433 | } | |
2434 | ||
2435 | if (in_f->ilf_size == 2) | |
2436 | goto write_inode_buffer; | |
2437 | len = item->ri_buf[2].i_len; | |
2438 | src = item->ri_buf[2].i_addr; | |
2439 | ASSERT(in_f->ilf_size <= 4); | |
2440 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); | |
2441 | ASSERT(!(fields & XFS_ILOG_DFORK) || | |
2442 | (len == in_f->ilf_dsize)); | |
2443 | ||
2444 | switch (fields & XFS_ILOG_DFORK) { | |
2445 | case XFS_ILOG_DDATA: | |
2446 | case XFS_ILOG_DEXT: | |
2447 | memcpy(&dip->di_u, src, len); | |
2448 | break; | |
2449 | ||
2450 | case XFS_ILOG_DBROOT: | |
2451 | xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len, | |
2452 | &(dip->di_u.di_bmbt), | |
2453 | XFS_DFORK_DSIZE(dip, mp)); | |
2454 | break; | |
2455 | ||
2456 | default: | |
2457 | /* | |
2458 | * There are no data fork flags set. | |
2459 | */ | |
2460 | ASSERT((fields & XFS_ILOG_DFORK) == 0); | |
2461 | break; | |
2462 | } | |
2463 | ||
2464 | /* | |
2465 | * If we logged any attribute data, recover it. There may or | |
2466 | * may not have been any other non-core data logged in this | |
2467 | * transaction. | |
2468 | */ | |
2469 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2470 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { | |
2471 | attr_index = 3; | |
2472 | } else { | |
2473 | attr_index = 2; | |
2474 | } | |
2475 | len = item->ri_buf[attr_index].i_len; | |
2476 | src = item->ri_buf[attr_index].i_addr; | |
2477 | ASSERT(len == in_f->ilf_asize); | |
2478 | ||
2479 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2480 | case XFS_ILOG_ADATA: | |
2481 | case XFS_ILOG_AEXT: | |
2482 | dest = XFS_DFORK_APTR(dip); | |
2483 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); | |
2484 | memcpy(dest, src, len); | |
2485 | break; | |
2486 | ||
2487 | case XFS_ILOG_ABROOT: | |
2488 | dest = XFS_DFORK_APTR(dip); | |
2489 | xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len, | |
2490 | (xfs_bmdr_block_t*)dest, | |
2491 | XFS_DFORK_ASIZE(dip, mp)); | |
2492 | break; | |
2493 | ||
2494 | default: | |
2495 | xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag"); | |
2496 | ASSERT(0); | |
2497 | xfs_buf_relse(bp); | |
6d192a9b TS |
2498 | error = EIO; |
2499 | goto error; | |
1da177e4 LT |
2500 | } |
2501 | } | |
2502 | ||
2503 | write_inode_buffer: | |
2504 | if (ITEM_TYPE(item) == XFS_LI_INODE) { | |
2505 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL || | |
2506 | XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp); | |
2507 | XFS_BUF_SET_FSPRIVATE(bp, mp); | |
2508 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone); | |
2509 | xfs_bdwrite(mp, bp); | |
2510 | } else { | |
2511 | XFS_BUF_STALE(bp); | |
2512 | error = xfs_bwrite(mp, bp); | |
2513 | } | |
2514 | ||
6d192a9b TS |
2515 | error: |
2516 | if (need_free) | |
2517 | kmem_free(in_f, sizeof(*in_f)); | |
2518 | return XFS_ERROR(error); | |
1da177e4 LT |
2519 | } |
2520 | ||
2521 | /* | |
2522 | * Recover QUOTAOFF records. We simply make a note of it in the xlog_t | |
2523 | * structure, so that we know not to do any dquot item or dquot buffer recovery, | |
2524 | * of that type. | |
2525 | */ | |
2526 | STATIC int | |
2527 | xlog_recover_do_quotaoff_trans( | |
2528 | xlog_t *log, | |
2529 | xlog_recover_item_t *item, | |
2530 | int pass) | |
2531 | { | |
2532 | xfs_qoff_logformat_t *qoff_f; | |
2533 | ||
2534 | if (pass == XLOG_RECOVER_PASS2) { | |
2535 | return (0); | |
2536 | } | |
2537 | ||
2538 | qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr; | |
2539 | ASSERT(qoff_f); | |
2540 | ||
2541 | /* | |
2542 | * The logitem format's flag tells us if this was user quotaoff, | |
77a7cce4 | 2543 | * group/project quotaoff or both. |
1da177e4 LT |
2544 | */ |
2545 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) | |
2546 | log->l_quotaoffs_flag |= XFS_DQ_USER; | |
77a7cce4 NS |
2547 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
2548 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; | |
1da177e4 LT |
2549 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
2550 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; | |
2551 | ||
2552 | return (0); | |
2553 | } | |
2554 | ||
2555 | /* | |
2556 | * Recover a dquot record | |
2557 | */ | |
2558 | STATIC int | |
2559 | xlog_recover_do_dquot_trans( | |
2560 | xlog_t *log, | |
2561 | xlog_recover_item_t *item, | |
2562 | int pass) | |
2563 | { | |
2564 | xfs_mount_t *mp; | |
2565 | xfs_buf_t *bp; | |
2566 | struct xfs_disk_dquot *ddq, *recddq; | |
2567 | int error; | |
2568 | xfs_dq_logformat_t *dq_f; | |
2569 | uint type; | |
2570 | ||
2571 | if (pass == XLOG_RECOVER_PASS1) { | |
2572 | return 0; | |
2573 | } | |
2574 | mp = log->l_mp; | |
2575 | ||
2576 | /* | |
2577 | * Filesystems are required to send in quota flags at mount time. | |
2578 | */ | |
2579 | if (mp->m_qflags == 0) | |
2580 | return (0); | |
2581 | ||
2582 | recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr; | |
2583 | ASSERT(recddq); | |
2584 | /* | |
2585 | * This type of quotas was turned off, so ignore this record. | |
2586 | */ | |
2587 | type = INT_GET(recddq->d_flags, ARCH_CONVERT) & | |
c8ad20ff | 2588 | (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
1da177e4 LT |
2589 | ASSERT(type); |
2590 | if (log->l_quotaoffs_flag & type) | |
2591 | return (0); | |
2592 | ||
2593 | /* | |
2594 | * At this point we know that quota was _not_ turned off. | |
2595 | * Since the mount flags are not indicating to us otherwise, this | |
2596 | * must mean that quota is on, and the dquot needs to be replayed. | |
2597 | * Remember that we may not have fully recovered the superblock yet, | |
2598 | * so we can't do the usual trick of looking at the SB quota bits. | |
2599 | * | |
2600 | * The other possibility, of course, is that the quota subsystem was | |
2601 | * removed since the last mount - ENOSYS. | |
2602 | */ | |
2603 | dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr; | |
2604 | ASSERT(dq_f); | |
2605 | if ((error = xfs_qm_dqcheck(recddq, | |
2606 | dq_f->qlf_id, | |
2607 | 0, XFS_QMOPT_DOWARN, | |
2608 | "xlog_recover_do_dquot_trans (log copy)"))) { | |
2609 | return XFS_ERROR(EIO); | |
2610 | } | |
2611 | ASSERT(dq_f->qlf_len == 1); | |
2612 | ||
2613 | error = xfs_read_buf(mp, mp->m_ddev_targp, | |
2614 | dq_f->qlf_blkno, | |
2615 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), | |
2616 | 0, &bp); | |
2617 | if (error) { | |
2618 | xfs_ioerror_alert("xlog_recover_do..(read#3)", mp, | |
2619 | bp, dq_f->qlf_blkno); | |
2620 | return error; | |
2621 | } | |
2622 | ASSERT(bp); | |
2623 | ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset); | |
2624 | ||
2625 | /* | |
2626 | * At least the magic num portion should be on disk because this | |
2627 | * was among a chunk of dquots created earlier, and we did some | |
2628 | * minimal initialization then. | |
2629 | */ | |
2630 | if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, | |
2631 | "xlog_recover_do_dquot_trans")) { | |
2632 | xfs_buf_relse(bp); | |
2633 | return XFS_ERROR(EIO); | |
2634 | } | |
2635 | ||
2636 | memcpy(ddq, recddq, item->ri_buf[1].i_len); | |
2637 | ||
2638 | ASSERT(dq_f->qlf_size == 2); | |
2639 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL || | |
2640 | XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp); | |
2641 | XFS_BUF_SET_FSPRIVATE(bp, mp); | |
2642 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone); | |
2643 | xfs_bdwrite(mp, bp); | |
2644 | ||
2645 | return (0); | |
2646 | } | |
2647 | ||
2648 | /* | |
2649 | * This routine is called to create an in-core extent free intent | |
2650 | * item from the efi format structure which was logged on disk. | |
2651 | * It allocates an in-core efi, copies the extents from the format | |
2652 | * structure into it, and adds the efi to the AIL with the given | |
2653 | * LSN. | |
2654 | */ | |
6d192a9b | 2655 | STATIC int |
1da177e4 LT |
2656 | xlog_recover_do_efi_trans( |
2657 | xlog_t *log, | |
2658 | xlog_recover_item_t *item, | |
2659 | xfs_lsn_t lsn, | |
2660 | int pass) | |
2661 | { | |
6d192a9b | 2662 | int error; |
1da177e4 LT |
2663 | xfs_mount_t *mp; |
2664 | xfs_efi_log_item_t *efip; | |
2665 | xfs_efi_log_format_t *efi_formatp; | |
2666 | SPLDECL(s); | |
2667 | ||
2668 | if (pass == XLOG_RECOVER_PASS1) { | |
6d192a9b | 2669 | return 0; |
1da177e4 LT |
2670 | } |
2671 | ||
2672 | efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr; | |
1da177e4 LT |
2673 | |
2674 | mp = log->l_mp; | |
2675 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); | |
6d192a9b TS |
2676 | if ((error = xfs_efi_copy_format(&(item->ri_buf[0]), |
2677 | &(efip->efi_format)))) { | |
2678 | xfs_efi_item_free(efip); | |
2679 | return error; | |
2680 | } | |
1da177e4 LT |
2681 | efip->efi_next_extent = efi_formatp->efi_nextents; |
2682 | efip->efi_flags |= XFS_EFI_COMMITTED; | |
2683 | ||
2684 | AIL_LOCK(mp,s); | |
2685 | /* | |
2686 | * xfs_trans_update_ail() drops the AIL lock. | |
2687 | */ | |
2688 | xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn, s); | |
6d192a9b | 2689 | return 0; |
1da177e4 LT |
2690 | } |
2691 | ||
2692 | ||
2693 | /* | |
2694 | * This routine is called when an efd format structure is found in | |
2695 | * a committed transaction in the log. It's purpose is to cancel | |
2696 | * the corresponding efi if it was still in the log. To do this | |
2697 | * it searches the AIL for the efi with an id equal to that in the | |
2698 | * efd format structure. If we find it, we remove the efi from the | |
2699 | * AIL and free it. | |
2700 | */ | |
2701 | STATIC void | |
2702 | xlog_recover_do_efd_trans( | |
2703 | xlog_t *log, | |
2704 | xlog_recover_item_t *item, | |
2705 | int pass) | |
2706 | { | |
2707 | xfs_mount_t *mp; | |
2708 | xfs_efd_log_format_t *efd_formatp; | |
2709 | xfs_efi_log_item_t *efip = NULL; | |
2710 | xfs_log_item_t *lip; | |
2711 | int gen; | |
1da177e4 LT |
2712 | __uint64_t efi_id; |
2713 | SPLDECL(s); | |
2714 | ||
2715 | if (pass == XLOG_RECOVER_PASS1) { | |
2716 | return; | |
2717 | } | |
2718 | ||
2719 | efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr; | |
6d192a9b TS |
2720 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
2721 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || | |
2722 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + | |
2723 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); | |
1da177e4 LT |
2724 | efi_id = efd_formatp->efd_efi_id; |
2725 | ||
2726 | /* | |
2727 | * Search for the efi with the id in the efd format structure | |
2728 | * in the AIL. | |
2729 | */ | |
2730 | mp = log->l_mp; | |
2731 | AIL_LOCK(mp,s); | |
2732 | lip = xfs_trans_first_ail(mp, &gen); | |
2733 | while (lip != NULL) { | |
2734 | if (lip->li_type == XFS_LI_EFI) { | |
2735 | efip = (xfs_efi_log_item_t *)lip; | |
2736 | if (efip->efi_format.efi_id == efi_id) { | |
2737 | /* | |
2738 | * xfs_trans_delete_ail() drops the | |
2739 | * AIL lock. | |
2740 | */ | |
2741 | xfs_trans_delete_ail(mp, lip, s); | |
2742 | break; | |
2743 | } | |
2744 | } | |
2745 | lip = xfs_trans_next_ail(mp, lip, &gen, NULL); | |
2746 | } | |
1da177e4 LT |
2747 | |
2748 | /* | |
2749 | * If we found it, then free it up. If it wasn't there, it | |
2750 | * must have been overwritten in the log. Oh well. | |
2751 | */ | |
2752 | if (lip != NULL) { | |
7d795ca3 CH |
2753 | xfs_efi_item_free(efip); |
2754 | } else { | |
2755 | AIL_UNLOCK(mp, s); | |
1da177e4 LT |
2756 | } |
2757 | } | |
2758 | ||
2759 | /* | |
2760 | * Perform the transaction | |
2761 | * | |
2762 | * If the transaction modifies a buffer or inode, do it now. Otherwise, | |
2763 | * EFIs and EFDs get queued up by adding entries into the AIL for them. | |
2764 | */ | |
2765 | STATIC int | |
2766 | xlog_recover_do_trans( | |
2767 | xlog_t *log, | |
2768 | xlog_recover_t *trans, | |
2769 | int pass) | |
2770 | { | |
2771 | int error = 0; | |
2772 | xlog_recover_item_t *item, *first_item; | |
2773 | ||
e9ed9d22 | 2774 | if ((error = xlog_recover_reorder_trans(trans))) |
1da177e4 LT |
2775 | return error; |
2776 | first_item = item = trans->r_itemq; | |
2777 | do { | |
2778 | /* | |
2779 | * we don't need to worry about the block number being | |
2780 | * truncated in > 1 TB buffers because in user-land, | |
2781 | * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so | |
c41564b5 | 2782 | * the blknos will get through the user-mode buffer |
1da177e4 LT |
2783 | * cache properly. The only bad case is o32 kernels |
2784 | * where xfs_daddr_t is 32-bits but mount will warn us | |
2785 | * off a > 1 TB filesystem before we get here. | |
2786 | */ | |
804195b6 | 2787 | if ((ITEM_TYPE(item) == XFS_LI_BUF)) { |
1da177e4 LT |
2788 | if ((error = xlog_recover_do_buffer_trans(log, item, |
2789 | pass))) | |
2790 | break; | |
6d192a9b | 2791 | } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) { |
1da177e4 LT |
2792 | if ((error = xlog_recover_do_inode_trans(log, item, |
2793 | pass))) | |
2794 | break; | |
2795 | } else if (ITEM_TYPE(item) == XFS_LI_EFI) { | |
6d192a9b TS |
2796 | if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn, |
2797 | pass))) | |
2798 | break; | |
1da177e4 LT |
2799 | } else if (ITEM_TYPE(item) == XFS_LI_EFD) { |
2800 | xlog_recover_do_efd_trans(log, item, pass); | |
2801 | } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) { | |
2802 | if ((error = xlog_recover_do_dquot_trans(log, item, | |
2803 | pass))) | |
2804 | break; | |
2805 | } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) { | |
2806 | if ((error = xlog_recover_do_quotaoff_trans(log, item, | |
2807 | pass))) | |
2808 | break; | |
2809 | } else { | |
2810 | xlog_warn("XFS: xlog_recover_do_trans"); | |
2811 | ASSERT(0); | |
2812 | error = XFS_ERROR(EIO); | |
2813 | break; | |
2814 | } | |
2815 | item = item->ri_next; | |
2816 | } while (first_item != item); | |
2817 | ||
2818 | return error; | |
2819 | } | |
2820 | ||
2821 | /* | |
2822 | * Free up any resources allocated by the transaction | |
2823 | * | |
2824 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. | |
2825 | */ | |
2826 | STATIC void | |
2827 | xlog_recover_free_trans( | |
2828 | xlog_recover_t *trans) | |
2829 | { | |
2830 | xlog_recover_item_t *first_item, *item, *free_item; | |
2831 | int i; | |
2832 | ||
2833 | item = first_item = trans->r_itemq; | |
2834 | do { | |
2835 | free_item = item; | |
2836 | item = item->ri_next; | |
2837 | /* Free the regions in the item. */ | |
2838 | for (i = 0; i < free_item->ri_cnt; i++) { | |
2839 | kmem_free(free_item->ri_buf[i].i_addr, | |
2840 | free_item->ri_buf[i].i_len); | |
2841 | } | |
2842 | /* Free the item itself */ | |
2843 | kmem_free(free_item->ri_buf, | |
2844 | (free_item->ri_total * sizeof(xfs_log_iovec_t))); | |
2845 | kmem_free(free_item, sizeof(xlog_recover_item_t)); | |
2846 | } while (first_item != item); | |
2847 | /* Free the transaction recover structure */ | |
2848 | kmem_free(trans, sizeof(xlog_recover_t)); | |
2849 | } | |
2850 | ||
2851 | STATIC int | |
2852 | xlog_recover_commit_trans( | |
2853 | xlog_t *log, | |
2854 | xlog_recover_t **q, | |
2855 | xlog_recover_t *trans, | |
2856 | int pass) | |
2857 | { | |
2858 | int error; | |
2859 | ||
2860 | if ((error = xlog_recover_unlink_tid(q, trans))) | |
2861 | return error; | |
2862 | if ((error = xlog_recover_do_trans(log, trans, pass))) | |
2863 | return error; | |
2864 | xlog_recover_free_trans(trans); /* no error */ | |
2865 | return 0; | |
2866 | } | |
2867 | ||
2868 | STATIC int | |
2869 | xlog_recover_unmount_trans( | |
2870 | xlog_recover_t *trans) | |
2871 | { | |
2872 | /* Do nothing now */ | |
2873 | xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR"); | |
2874 | return 0; | |
2875 | } | |
2876 | ||
2877 | /* | |
2878 | * There are two valid states of the r_state field. 0 indicates that the | |
2879 | * transaction structure is in a normal state. We have either seen the | |
2880 | * start of the transaction or the last operation we added was not a partial | |
2881 | * operation. If the last operation we added to the transaction was a | |
2882 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. | |
2883 | * | |
2884 | * NOTE: skip LRs with 0 data length. | |
2885 | */ | |
2886 | STATIC int | |
2887 | xlog_recover_process_data( | |
2888 | xlog_t *log, | |
2889 | xlog_recover_t *rhash[], | |
2890 | xlog_rec_header_t *rhead, | |
2891 | xfs_caddr_t dp, | |
2892 | int pass) | |
2893 | { | |
2894 | xfs_caddr_t lp; | |
2895 | int num_logops; | |
2896 | xlog_op_header_t *ohead; | |
2897 | xlog_recover_t *trans; | |
2898 | xlog_tid_t tid; | |
2899 | int error; | |
2900 | unsigned long hash; | |
2901 | uint flags; | |
2902 | ||
2903 | lp = dp + INT_GET(rhead->h_len, ARCH_CONVERT); | |
2904 | num_logops = INT_GET(rhead->h_num_logops, ARCH_CONVERT); | |
2905 | ||
2906 | /* check the log format matches our own - else we can't recover */ | |
2907 | if (xlog_header_check_recover(log->l_mp, rhead)) | |
2908 | return (XFS_ERROR(EIO)); | |
2909 | ||
2910 | while ((dp < lp) && num_logops) { | |
2911 | ASSERT(dp + sizeof(xlog_op_header_t) <= lp); | |
2912 | ohead = (xlog_op_header_t *)dp; | |
2913 | dp += sizeof(xlog_op_header_t); | |
2914 | if (ohead->oh_clientid != XFS_TRANSACTION && | |
2915 | ohead->oh_clientid != XFS_LOG) { | |
2916 | xlog_warn( | |
2917 | "XFS: xlog_recover_process_data: bad clientid"); | |
2918 | ASSERT(0); | |
2919 | return (XFS_ERROR(EIO)); | |
2920 | } | |
2921 | tid = INT_GET(ohead->oh_tid, ARCH_CONVERT); | |
2922 | hash = XLOG_RHASH(tid); | |
2923 | trans = xlog_recover_find_tid(rhash[hash], tid); | |
2924 | if (trans == NULL) { /* not found; add new tid */ | |
2925 | if (ohead->oh_flags & XLOG_START_TRANS) | |
2926 | xlog_recover_new_tid(&rhash[hash], tid, | |
2927 | INT_GET(rhead->h_lsn, ARCH_CONVERT)); | |
2928 | } else { | |
2929 | ASSERT(dp+INT_GET(ohead->oh_len, ARCH_CONVERT) <= lp); | |
2930 | flags = ohead->oh_flags & ~XLOG_END_TRANS; | |
2931 | if (flags & XLOG_WAS_CONT_TRANS) | |
2932 | flags &= ~XLOG_CONTINUE_TRANS; | |
2933 | switch (flags) { | |
2934 | case XLOG_COMMIT_TRANS: | |
2935 | error = xlog_recover_commit_trans(log, | |
2936 | &rhash[hash], trans, pass); | |
2937 | break; | |
2938 | case XLOG_UNMOUNT_TRANS: | |
2939 | error = xlog_recover_unmount_trans(trans); | |
2940 | break; | |
2941 | case XLOG_WAS_CONT_TRANS: | |
2942 | error = xlog_recover_add_to_cont_trans(trans, | |
2943 | dp, INT_GET(ohead->oh_len, | |
2944 | ARCH_CONVERT)); | |
2945 | break; | |
2946 | case XLOG_START_TRANS: | |
2947 | xlog_warn( | |
2948 | "XFS: xlog_recover_process_data: bad transaction"); | |
2949 | ASSERT(0); | |
2950 | error = XFS_ERROR(EIO); | |
2951 | break; | |
2952 | case 0: | |
2953 | case XLOG_CONTINUE_TRANS: | |
2954 | error = xlog_recover_add_to_trans(trans, | |
2955 | dp, INT_GET(ohead->oh_len, | |
2956 | ARCH_CONVERT)); | |
2957 | break; | |
2958 | default: | |
2959 | xlog_warn( | |
2960 | "XFS: xlog_recover_process_data: bad flag"); | |
2961 | ASSERT(0); | |
2962 | error = XFS_ERROR(EIO); | |
2963 | break; | |
2964 | } | |
2965 | if (error) | |
2966 | return error; | |
2967 | } | |
2968 | dp += INT_GET(ohead->oh_len, ARCH_CONVERT); | |
2969 | num_logops--; | |
2970 | } | |
2971 | return 0; | |
2972 | } | |
2973 | ||
2974 | /* | |
2975 | * Process an extent free intent item that was recovered from | |
2976 | * the log. We need to free the extents that it describes. | |
2977 | */ | |
2978 | STATIC void | |
2979 | xlog_recover_process_efi( | |
2980 | xfs_mount_t *mp, | |
2981 | xfs_efi_log_item_t *efip) | |
2982 | { | |
2983 | xfs_efd_log_item_t *efdp; | |
2984 | xfs_trans_t *tp; | |
2985 | int i; | |
2986 | xfs_extent_t *extp; | |
2987 | xfs_fsblock_t startblock_fsb; | |
2988 | ||
2989 | ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED)); | |
2990 | ||
2991 | /* | |
2992 | * First check the validity of the extents described by the | |
2993 | * EFI. If any are bad, then assume that all are bad and | |
2994 | * just toss the EFI. | |
2995 | */ | |
2996 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
2997 | extp = &(efip->efi_format.efi_extents[i]); | |
2998 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
2999 | XFS_FSB_TO_DADDR(mp, extp->ext_start)); | |
3000 | if ((startblock_fsb == 0) || | |
3001 | (extp->ext_len == 0) || | |
3002 | (startblock_fsb >= mp->m_sb.sb_dblocks) || | |
3003 | (extp->ext_len >= mp->m_sb.sb_agblocks)) { | |
3004 | /* | |
3005 | * This will pull the EFI from the AIL and | |
3006 | * free the memory associated with it. | |
3007 | */ | |
3008 | xfs_efi_release(efip, efip->efi_format.efi_nextents); | |
3009 | return; | |
3010 | } | |
3011 | } | |
3012 | ||
3013 | tp = xfs_trans_alloc(mp, 0); | |
3014 | xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0); | |
3015 | efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents); | |
3016 | ||
3017 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3018 | extp = &(efip->efi_format.efi_extents[i]); | |
3019 | xfs_free_extent(tp, extp->ext_start, extp->ext_len); | |
3020 | xfs_trans_log_efd_extent(tp, efdp, extp->ext_start, | |
3021 | extp->ext_len); | |
3022 | } | |
3023 | ||
3024 | efip->efi_flags |= XFS_EFI_RECOVERED; | |
1c72bf90 | 3025 | xfs_trans_commit(tp, 0); |
1da177e4 LT |
3026 | } |
3027 | ||
3028 | /* | |
3029 | * Verify that once we've encountered something other than an EFI | |
3030 | * in the AIL that there are no more EFIs in the AIL. | |
3031 | */ | |
3032 | #if defined(DEBUG) | |
3033 | STATIC void | |
3034 | xlog_recover_check_ail( | |
3035 | xfs_mount_t *mp, | |
3036 | xfs_log_item_t *lip, | |
3037 | int gen) | |
3038 | { | |
3039 | int orig_gen = gen; | |
3040 | ||
3041 | do { | |
3042 | ASSERT(lip->li_type != XFS_LI_EFI); | |
3043 | lip = xfs_trans_next_ail(mp, lip, &gen, NULL); | |
3044 | /* | |
3045 | * The check will be bogus if we restart from the | |
3046 | * beginning of the AIL, so ASSERT that we don't. | |
3047 | * We never should since we're holding the AIL lock | |
3048 | * the entire time. | |
3049 | */ | |
3050 | ASSERT(gen == orig_gen); | |
3051 | } while (lip != NULL); | |
3052 | } | |
3053 | #endif /* DEBUG */ | |
3054 | ||
3055 | /* | |
3056 | * When this is called, all of the EFIs which did not have | |
3057 | * corresponding EFDs should be in the AIL. What we do now | |
3058 | * is free the extents associated with each one. | |
3059 | * | |
3060 | * Since we process the EFIs in normal transactions, they | |
3061 | * will be removed at some point after the commit. This prevents | |
3062 | * us from just walking down the list processing each one. | |
3063 | * We'll use a flag in the EFI to skip those that we've already | |
3064 | * processed and use the AIL iteration mechanism's generation | |
3065 | * count to try to speed this up at least a bit. | |
3066 | * | |
3067 | * When we start, we know that the EFIs are the only things in | |
3068 | * the AIL. As we process them, however, other items are added | |
3069 | * to the AIL. Since everything added to the AIL must come after | |
3070 | * everything already in the AIL, we stop processing as soon as | |
3071 | * we see something other than an EFI in the AIL. | |
3072 | */ | |
3073 | STATIC void | |
3074 | xlog_recover_process_efis( | |
3075 | xlog_t *log) | |
3076 | { | |
3077 | xfs_log_item_t *lip; | |
3078 | xfs_efi_log_item_t *efip; | |
3079 | int gen; | |
3080 | xfs_mount_t *mp; | |
3081 | SPLDECL(s); | |
3082 | ||
3083 | mp = log->l_mp; | |
3084 | AIL_LOCK(mp,s); | |
3085 | ||
3086 | lip = xfs_trans_first_ail(mp, &gen); | |
3087 | while (lip != NULL) { | |
3088 | /* | |
3089 | * We're done when we see something other than an EFI. | |
3090 | */ | |
3091 | if (lip->li_type != XFS_LI_EFI) { | |
3092 | xlog_recover_check_ail(mp, lip, gen); | |
3093 | break; | |
3094 | } | |
3095 | ||
3096 | /* | |
3097 | * Skip EFIs that we've already processed. | |
3098 | */ | |
3099 | efip = (xfs_efi_log_item_t *)lip; | |
3100 | if (efip->efi_flags & XFS_EFI_RECOVERED) { | |
3101 | lip = xfs_trans_next_ail(mp, lip, &gen, NULL); | |
3102 | continue; | |
3103 | } | |
3104 | ||
3105 | AIL_UNLOCK(mp, s); | |
3106 | xlog_recover_process_efi(mp, efip); | |
3107 | AIL_LOCK(mp,s); | |
3108 | lip = xfs_trans_next_ail(mp, lip, &gen, NULL); | |
3109 | } | |
3110 | AIL_UNLOCK(mp, s); | |
3111 | } | |
3112 | ||
3113 | /* | |
3114 | * This routine performs a transaction to null out a bad inode pointer | |
3115 | * in an agi unlinked inode hash bucket. | |
3116 | */ | |
3117 | STATIC void | |
3118 | xlog_recover_clear_agi_bucket( | |
3119 | xfs_mount_t *mp, | |
3120 | xfs_agnumber_t agno, | |
3121 | int bucket) | |
3122 | { | |
3123 | xfs_trans_t *tp; | |
3124 | xfs_agi_t *agi; | |
3125 | xfs_buf_t *agibp; | |
3126 | int offset; | |
3127 | int error; | |
3128 | ||
3129 | tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET); | |
3130 | xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0); | |
3131 | ||
3132 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, | |
3133 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), | |
3134 | XFS_FSS_TO_BB(mp, 1), 0, &agibp); | |
3135 | if (error) { | |
3136 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3137 | return; | |
3138 | } | |
3139 | ||
3140 | agi = XFS_BUF_TO_AGI(agibp); | |
16259e7d | 3141 | if (be32_to_cpu(agi->agi_magicnum) != XFS_AGI_MAGIC) { |
1da177e4 LT |
3142 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); |
3143 | return; | |
3144 | } | |
1da177e4 | 3145 | |
16259e7d | 3146 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
3147 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
3148 | (sizeof(xfs_agino_t) * bucket); | |
3149 | xfs_trans_log_buf(tp, agibp, offset, | |
3150 | (offset + sizeof(xfs_agino_t) - 1)); | |
3151 | ||
1c72bf90 | 3152 | (void) xfs_trans_commit(tp, 0); |
1da177e4 LT |
3153 | } |
3154 | ||
3155 | /* | |
3156 | * xlog_iunlink_recover | |
3157 | * | |
3158 | * This is called during recovery to process any inodes which | |
3159 | * we unlinked but not freed when the system crashed. These | |
3160 | * inodes will be on the lists in the AGI blocks. What we do | |
3161 | * here is scan all the AGIs and fully truncate and free any | |
3162 | * inodes found on the lists. Each inode is removed from the | |
3163 | * lists when it has been fully truncated and is freed. The | |
3164 | * freeing of the inode and its removal from the list must be | |
3165 | * atomic. | |
3166 | */ | |
3167 | void | |
3168 | xlog_recover_process_iunlinks( | |
3169 | xlog_t *log) | |
3170 | { | |
3171 | xfs_mount_t *mp; | |
3172 | xfs_agnumber_t agno; | |
3173 | xfs_agi_t *agi; | |
3174 | xfs_buf_t *agibp; | |
3175 | xfs_buf_t *ibp; | |
3176 | xfs_dinode_t *dip; | |
3177 | xfs_inode_t *ip; | |
3178 | xfs_agino_t agino; | |
3179 | xfs_ino_t ino; | |
3180 | int bucket; | |
3181 | int error; | |
3182 | uint mp_dmevmask; | |
3183 | ||
3184 | mp = log->l_mp; | |
3185 | ||
3186 | /* | |
3187 | * Prevent any DMAPI event from being sent while in this function. | |
3188 | */ | |
3189 | mp_dmevmask = mp->m_dmevmask; | |
3190 | mp->m_dmevmask = 0; | |
3191 | ||
3192 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
3193 | /* | |
3194 | * Find the agi for this ag. | |
3195 | */ | |
3196 | agibp = xfs_buf_read(mp->m_ddev_targp, | |
3197 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), | |
3198 | XFS_FSS_TO_BB(mp, 1), 0); | |
3199 | if (XFS_BUF_ISERROR(agibp)) { | |
3200 | xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)", | |
3201 | log->l_mp, agibp, | |
3202 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp))); | |
3203 | } | |
3204 | agi = XFS_BUF_TO_AGI(agibp); | |
16259e7d | 3205 | ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agi->agi_magicnum)); |
1da177e4 LT |
3206 | |
3207 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { | |
3208 | ||
16259e7d | 3209 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
1da177e4 LT |
3210 | while (agino != NULLAGINO) { |
3211 | ||
3212 | /* | |
3213 | * Release the agi buffer so that it can | |
3214 | * be acquired in the normal course of the | |
3215 | * transaction to truncate and free the inode. | |
3216 | */ | |
3217 | xfs_buf_relse(agibp); | |
3218 | ||
3219 | ino = XFS_AGINO_TO_INO(mp, agno, agino); | |
3220 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0); | |
3221 | ASSERT(error || (ip != NULL)); | |
3222 | ||
3223 | if (!error) { | |
3224 | /* | |
3225 | * Get the on disk inode to find the | |
3226 | * next inode in the bucket. | |
3227 | */ | |
3228 | error = xfs_itobp(mp, NULL, ip, &dip, | |
b12dd342 | 3229 | &ibp, 0, 0); |
1da177e4 LT |
3230 | ASSERT(error || (dip != NULL)); |
3231 | } | |
3232 | ||
3233 | if (!error) { | |
3234 | ASSERT(ip->i_d.di_nlink == 0); | |
3235 | ||
3236 | /* setup for the next pass */ | |
3237 | agino = INT_GET(dip->di_next_unlinked, | |
3238 | ARCH_CONVERT); | |
3239 | xfs_buf_relse(ibp); | |
3240 | /* | |
3241 | * Prevent any DMAPI event from | |
3242 | * being sent when the | |
3243 | * reference on the inode is | |
3244 | * dropped. | |
3245 | */ | |
3246 | ip->i_d.di_dmevmask = 0; | |
3247 | ||
3248 | /* | |
3249 | * If this is a new inode, handle | |
3250 | * it specially. Otherwise, | |
3251 | * just drop our reference to the | |
3252 | * inode. If there are no | |
3253 | * other references, this will | |
3254 | * send the inode to | |
3255 | * xfs_inactive() which will | |
3256 | * truncate the file and free | |
3257 | * the inode. | |
3258 | */ | |
3259 | if (ip->i_d.di_mode == 0) | |
3260 | xfs_iput_new(ip, 0); | |
3261 | else | |
3262 | VN_RELE(XFS_ITOV(ip)); | |
3263 | } else { | |
3264 | /* | |
3265 | * We can't read in the inode | |
3266 | * this bucket points to, or | |
3267 | * this inode is messed up. Just | |
3268 | * ditch this bucket of inodes. We | |
3269 | * will lose some inodes and space, | |
3270 | * but at least we won't hang. Call | |
3271 | * xlog_recover_clear_agi_bucket() | |
3272 | * to perform a transaction to clear | |
3273 | * the inode pointer in the bucket. | |
3274 | */ | |
3275 | xlog_recover_clear_agi_bucket(mp, agno, | |
3276 | bucket); | |
3277 | ||
3278 | agino = NULLAGINO; | |
3279 | } | |
3280 | ||
3281 | /* | |
3282 | * Reacquire the agibuffer and continue around | |
3283 | * the loop. | |
3284 | */ | |
3285 | agibp = xfs_buf_read(mp->m_ddev_targp, | |
3286 | XFS_AG_DADDR(mp, agno, | |
3287 | XFS_AGI_DADDR(mp)), | |
3288 | XFS_FSS_TO_BB(mp, 1), 0); | |
3289 | if (XFS_BUF_ISERROR(agibp)) { | |
3290 | xfs_ioerror_alert( | |
3291 | "xlog_recover_process_iunlinks(#2)", | |
3292 | log->l_mp, agibp, | |
3293 | XFS_AG_DADDR(mp, agno, | |
3294 | XFS_AGI_DADDR(mp))); | |
3295 | } | |
3296 | agi = XFS_BUF_TO_AGI(agibp); | |
16259e7d CH |
3297 | ASSERT(XFS_AGI_MAGIC == be32_to_cpu( |
3298 | agi->agi_magicnum)); | |
1da177e4 LT |
3299 | } |
3300 | } | |
3301 | ||
3302 | /* | |
3303 | * Release the buffer for the current agi so we can | |
3304 | * go on to the next one. | |
3305 | */ | |
3306 | xfs_buf_relse(agibp); | |
3307 | } | |
3308 | ||
3309 | mp->m_dmevmask = mp_dmevmask; | |
3310 | } | |
3311 | ||
3312 | ||
3313 | #ifdef DEBUG | |
3314 | STATIC void | |
3315 | xlog_pack_data_checksum( | |
3316 | xlog_t *log, | |
3317 | xlog_in_core_t *iclog, | |
3318 | int size) | |
3319 | { | |
3320 | int i; | |
3321 | uint *up; | |
3322 | uint chksum = 0; | |
3323 | ||
3324 | up = (uint *)iclog->ic_datap; | |
3325 | /* divide length by 4 to get # words */ | |
3326 | for (i = 0; i < (size >> 2); i++) { | |
3327 | chksum ^= INT_GET(*up, ARCH_CONVERT); | |
3328 | up++; | |
3329 | } | |
3330 | INT_SET(iclog->ic_header.h_chksum, ARCH_CONVERT, chksum); | |
3331 | } | |
3332 | #else | |
3333 | #define xlog_pack_data_checksum(log, iclog, size) | |
3334 | #endif | |
3335 | ||
3336 | /* | |
3337 | * Stamp cycle number in every block | |
3338 | */ | |
3339 | void | |
3340 | xlog_pack_data( | |
3341 | xlog_t *log, | |
3342 | xlog_in_core_t *iclog, | |
3343 | int roundoff) | |
3344 | { | |
3345 | int i, j, k; | |
3346 | int size = iclog->ic_offset + roundoff; | |
3347 | uint cycle_lsn; | |
3348 | xfs_caddr_t dp; | |
3349 | xlog_in_core_2_t *xhdr; | |
3350 | ||
3351 | xlog_pack_data_checksum(log, iclog, size); | |
3352 | ||
3353 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
3354 | ||
3355 | dp = iclog->ic_datap; | |
3356 | for (i = 0; i < BTOBB(size) && | |
3357 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { | |
3358 | iclog->ic_header.h_cycle_data[i] = *(uint *)dp; | |
3359 | *(uint *)dp = cycle_lsn; | |
3360 | dp += BBSIZE; | |
3361 | } | |
3362 | ||
3363 | if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) { | |
3364 | xhdr = (xlog_in_core_2_t *)&iclog->ic_header; | |
3365 | for ( ; i < BTOBB(size); i++) { | |
3366 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3367 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3368 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(uint *)dp; | |
3369 | *(uint *)dp = cycle_lsn; | |
3370 | dp += BBSIZE; | |
3371 | } | |
3372 | ||
3373 | for (i = 1; i < log->l_iclog_heads; i++) { | |
3374 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
3375 | } | |
3376 | } | |
3377 | } | |
3378 | ||
3379 | #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY) | |
3380 | STATIC void | |
3381 | xlog_unpack_data_checksum( | |
3382 | xlog_rec_header_t *rhead, | |
3383 | xfs_caddr_t dp, | |
3384 | xlog_t *log) | |
3385 | { | |
3386 | uint *up = (uint *)dp; | |
3387 | uint chksum = 0; | |
3388 | int i; | |
3389 | ||
3390 | /* divide length by 4 to get # words */ | |
3391 | for (i=0; i < INT_GET(rhead->h_len, ARCH_CONVERT) >> 2; i++) { | |
3392 | chksum ^= INT_GET(*up, ARCH_CONVERT); | |
3393 | up++; | |
3394 | } | |
3395 | if (chksum != INT_GET(rhead->h_chksum, ARCH_CONVERT)) { | |
3396 | if (rhead->h_chksum || | |
3397 | ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) { | |
3398 | cmn_err(CE_DEBUG, | |
b6574520 | 3399 | "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n", |
1da177e4 LT |
3400 | INT_GET(rhead->h_chksum, ARCH_CONVERT), chksum); |
3401 | cmn_err(CE_DEBUG, | |
3402 | "XFS: Disregard message if filesystem was created with non-DEBUG kernel"); | |
3403 | if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) { | |
3404 | cmn_err(CE_DEBUG, | |
b6574520 | 3405 | "XFS: LogR this is a LogV2 filesystem\n"); |
1da177e4 LT |
3406 | } |
3407 | log->l_flags |= XLOG_CHKSUM_MISMATCH; | |
3408 | } | |
3409 | } | |
3410 | } | |
3411 | #else | |
3412 | #define xlog_unpack_data_checksum(rhead, dp, log) | |
3413 | #endif | |
3414 | ||
3415 | STATIC void | |
3416 | xlog_unpack_data( | |
3417 | xlog_rec_header_t *rhead, | |
3418 | xfs_caddr_t dp, | |
3419 | xlog_t *log) | |
3420 | { | |
3421 | int i, j, k; | |
3422 | xlog_in_core_2_t *xhdr; | |
3423 | ||
3424 | for (i = 0; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)) && | |
3425 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { | |
3426 | *(uint *)dp = *(uint *)&rhead->h_cycle_data[i]; | |
3427 | dp += BBSIZE; | |
3428 | } | |
3429 | ||
3430 | if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) { | |
3431 | xhdr = (xlog_in_core_2_t *)rhead; | |
3432 | for ( ; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)); i++) { | |
3433 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3434 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3435 | *(uint *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; | |
3436 | dp += BBSIZE; | |
3437 | } | |
3438 | } | |
3439 | ||
3440 | xlog_unpack_data_checksum(rhead, dp, log); | |
3441 | } | |
3442 | ||
3443 | STATIC int | |
3444 | xlog_valid_rec_header( | |
3445 | xlog_t *log, | |
3446 | xlog_rec_header_t *rhead, | |
3447 | xfs_daddr_t blkno) | |
3448 | { | |
3449 | int hlen; | |
3450 | ||
3451 | if (unlikely( | |
3452 | (INT_GET(rhead->h_magicno, ARCH_CONVERT) != | |
3453 | XLOG_HEADER_MAGIC_NUM))) { | |
3454 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", | |
3455 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3456 | return XFS_ERROR(EFSCORRUPTED); | |
3457 | } | |
3458 | if (unlikely( | |
3459 | (!rhead->h_version || | |
3460 | (INT_GET(rhead->h_version, ARCH_CONVERT) & | |
3461 | (~XLOG_VERSION_OKBITS)) != 0))) { | |
3462 | xlog_warn("XFS: %s: unrecognised log version (%d).", | |
3463 | __FUNCTION__, INT_GET(rhead->h_version, ARCH_CONVERT)); | |
3464 | return XFS_ERROR(EIO); | |
3465 | } | |
3466 | ||
3467 | /* LR body must have data or it wouldn't have been written */ | |
3468 | hlen = INT_GET(rhead->h_len, ARCH_CONVERT); | |
3469 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { | |
3470 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", | |
3471 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3472 | return XFS_ERROR(EFSCORRUPTED); | |
3473 | } | |
3474 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { | |
3475 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", | |
3476 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3477 | return XFS_ERROR(EFSCORRUPTED); | |
3478 | } | |
3479 | return 0; | |
3480 | } | |
3481 | ||
3482 | /* | |
3483 | * Read the log from tail to head and process the log records found. | |
3484 | * Handle the two cases where the tail and head are in the same cycle | |
3485 | * and where the active portion of the log wraps around the end of | |
3486 | * the physical log separately. The pass parameter is passed through | |
3487 | * to the routines called to process the data and is not looked at | |
3488 | * here. | |
3489 | */ | |
3490 | STATIC int | |
3491 | xlog_do_recovery_pass( | |
3492 | xlog_t *log, | |
3493 | xfs_daddr_t head_blk, | |
3494 | xfs_daddr_t tail_blk, | |
3495 | int pass) | |
3496 | { | |
3497 | xlog_rec_header_t *rhead; | |
3498 | xfs_daddr_t blk_no; | |
3499 | xfs_caddr_t bufaddr, offset; | |
3500 | xfs_buf_t *hbp, *dbp; | |
3501 | int error = 0, h_size; | |
3502 | int bblks, split_bblks; | |
3503 | int hblks, split_hblks, wrapped_hblks; | |
3504 | xlog_recover_t *rhash[XLOG_RHASH_SIZE]; | |
3505 | ||
3506 | ASSERT(head_blk != tail_blk); | |
3507 | ||
3508 | /* | |
3509 | * Read the header of the tail block and get the iclog buffer size from | |
3510 | * h_size. Use this to tell how many sectors make up the log header. | |
3511 | */ | |
3512 | if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) { | |
3513 | /* | |
3514 | * When using variable length iclogs, read first sector of | |
3515 | * iclog header and extract the header size from it. Get a | |
3516 | * new hbp that is the correct size. | |
3517 | */ | |
3518 | hbp = xlog_get_bp(log, 1); | |
3519 | if (!hbp) | |
3520 | return ENOMEM; | |
3521 | if ((error = xlog_bread(log, tail_blk, 1, hbp))) | |
3522 | goto bread_err1; | |
3523 | offset = xlog_align(log, tail_blk, 1, hbp); | |
3524 | rhead = (xlog_rec_header_t *)offset; | |
3525 | error = xlog_valid_rec_header(log, rhead, tail_blk); | |
3526 | if (error) | |
3527 | goto bread_err1; | |
3528 | h_size = INT_GET(rhead->h_size, ARCH_CONVERT); | |
3529 | if ((INT_GET(rhead->h_version, ARCH_CONVERT) | |
3530 | & XLOG_VERSION_2) && | |
3531 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
3532 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
3533 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
3534 | hblks++; | |
3535 | xlog_put_bp(hbp); | |
3536 | hbp = xlog_get_bp(log, hblks); | |
3537 | } else { | |
3538 | hblks = 1; | |
3539 | } | |
3540 | } else { | |
3541 | ASSERT(log->l_sectbb_log == 0); | |
3542 | hblks = 1; | |
3543 | hbp = xlog_get_bp(log, 1); | |
3544 | h_size = XLOG_BIG_RECORD_BSIZE; | |
3545 | } | |
3546 | ||
3547 | if (!hbp) | |
3548 | return ENOMEM; | |
3549 | dbp = xlog_get_bp(log, BTOBB(h_size)); | |
3550 | if (!dbp) { | |
3551 | xlog_put_bp(hbp); | |
3552 | return ENOMEM; | |
3553 | } | |
3554 | ||
3555 | memset(rhash, 0, sizeof(rhash)); | |
3556 | if (tail_blk <= head_blk) { | |
3557 | for (blk_no = tail_blk; blk_no < head_blk; ) { | |
3558 | if ((error = xlog_bread(log, blk_no, hblks, hbp))) | |
3559 | goto bread_err2; | |
3560 | offset = xlog_align(log, blk_no, hblks, hbp); | |
3561 | rhead = (xlog_rec_header_t *)offset; | |
3562 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3563 | if (error) | |
3564 | goto bread_err2; | |
3565 | ||
3566 | /* blocks in data section */ | |
3567 | bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)); | |
3568 | error = xlog_bread(log, blk_no + hblks, bblks, dbp); | |
3569 | if (error) | |
3570 | goto bread_err2; | |
3571 | offset = xlog_align(log, blk_no + hblks, bblks, dbp); | |
3572 | xlog_unpack_data(rhead, offset, log); | |
3573 | if ((error = xlog_recover_process_data(log, | |
3574 | rhash, rhead, offset, pass))) | |
3575 | goto bread_err2; | |
3576 | blk_no += bblks + hblks; | |
3577 | } | |
3578 | } else { | |
3579 | /* | |
3580 | * Perform recovery around the end of the physical log. | |
3581 | * When the head is not on the same cycle number as the tail, | |
3582 | * we can't do a sequential recovery as above. | |
3583 | */ | |
3584 | blk_no = tail_blk; | |
3585 | while (blk_no < log->l_logBBsize) { | |
3586 | /* | |
3587 | * Check for header wrapping around physical end-of-log | |
3588 | */ | |
3589 | offset = NULL; | |
3590 | split_hblks = 0; | |
3591 | wrapped_hblks = 0; | |
3592 | if (blk_no + hblks <= log->l_logBBsize) { | |
3593 | /* Read header in one read */ | |
3594 | error = xlog_bread(log, blk_no, hblks, hbp); | |
3595 | if (error) | |
3596 | goto bread_err2; | |
3597 | offset = xlog_align(log, blk_no, hblks, hbp); | |
3598 | } else { | |
3599 | /* This LR is split across physical log end */ | |
3600 | if (blk_no != log->l_logBBsize) { | |
3601 | /* some data before physical log end */ | |
3602 | ASSERT(blk_no <= INT_MAX); | |
3603 | split_hblks = log->l_logBBsize - (int)blk_no; | |
3604 | ASSERT(split_hblks > 0); | |
3605 | if ((error = xlog_bread(log, blk_no, | |
3606 | split_hblks, hbp))) | |
3607 | goto bread_err2; | |
3608 | offset = xlog_align(log, blk_no, | |
3609 | split_hblks, hbp); | |
3610 | } | |
3611 | /* | |
3612 | * Note: this black magic still works with | |
3613 | * large sector sizes (non-512) only because: | |
3614 | * - we increased the buffer size originally | |
3615 | * by 1 sector giving us enough extra space | |
3616 | * for the second read; | |
3617 | * - the log start is guaranteed to be sector | |
3618 | * aligned; | |
3619 | * - we read the log end (LR header start) | |
3620 | * _first_, then the log start (LR header end) | |
3621 | * - order is important. | |
3622 | */ | |
3623 | bufaddr = XFS_BUF_PTR(hbp); | |
3624 | XFS_BUF_SET_PTR(hbp, | |
3625 | bufaddr + BBTOB(split_hblks), | |
3626 | BBTOB(hblks - split_hblks)); | |
3627 | wrapped_hblks = hblks - split_hblks; | |
3628 | error = xlog_bread(log, 0, wrapped_hblks, hbp); | |
3629 | if (error) | |
3630 | goto bread_err2; | |
3631 | XFS_BUF_SET_PTR(hbp, bufaddr, BBTOB(hblks)); | |
3632 | if (!offset) | |
3633 | offset = xlog_align(log, 0, | |
3634 | wrapped_hblks, hbp); | |
3635 | } | |
3636 | rhead = (xlog_rec_header_t *)offset; | |
3637 | error = xlog_valid_rec_header(log, rhead, | |
3638 | split_hblks ? blk_no : 0); | |
3639 | if (error) | |
3640 | goto bread_err2; | |
3641 | ||
3642 | bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)); | |
3643 | blk_no += hblks; | |
3644 | ||
3645 | /* Read in data for log record */ | |
3646 | if (blk_no + bblks <= log->l_logBBsize) { | |
3647 | error = xlog_bread(log, blk_no, bblks, dbp); | |
3648 | if (error) | |
3649 | goto bread_err2; | |
3650 | offset = xlog_align(log, blk_no, bblks, dbp); | |
3651 | } else { | |
3652 | /* This log record is split across the | |
3653 | * physical end of log */ | |
3654 | offset = NULL; | |
3655 | split_bblks = 0; | |
3656 | if (blk_no != log->l_logBBsize) { | |
3657 | /* some data is before the physical | |
3658 | * end of log */ | |
3659 | ASSERT(!wrapped_hblks); | |
3660 | ASSERT(blk_no <= INT_MAX); | |
3661 | split_bblks = | |
3662 | log->l_logBBsize - (int)blk_no; | |
3663 | ASSERT(split_bblks > 0); | |
3664 | if ((error = xlog_bread(log, blk_no, | |
3665 | split_bblks, dbp))) | |
3666 | goto bread_err2; | |
3667 | offset = xlog_align(log, blk_no, | |
3668 | split_bblks, dbp); | |
3669 | } | |
3670 | /* | |
3671 | * Note: this black magic still works with | |
3672 | * large sector sizes (non-512) only because: | |
3673 | * - we increased the buffer size originally | |
3674 | * by 1 sector giving us enough extra space | |
3675 | * for the second read; | |
3676 | * - the log start is guaranteed to be sector | |
3677 | * aligned; | |
3678 | * - we read the log end (LR header start) | |
3679 | * _first_, then the log start (LR header end) | |
3680 | * - order is important. | |
3681 | */ | |
3682 | bufaddr = XFS_BUF_PTR(dbp); | |
3683 | XFS_BUF_SET_PTR(dbp, | |
3684 | bufaddr + BBTOB(split_bblks), | |
3685 | BBTOB(bblks - split_bblks)); | |
3686 | if ((error = xlog_bread(log, wrapped_hblks, | |
3687 | bblks - split_bblks, dbp))) | |
3688 | goto bread_err2; | |
3689 | XFS_BUF_SET_PTR(dbp, bufaddr, h_size); | |
3690 | if (!offset) | |
3691 | offset = xlog_align(log, wrapped_hblks, | |
3692 | bblks - split_bblks, dbp); | |
3693 | } | |
3694 | xlog_unpack_data(rhead, offset, log); | |
3695 | if ((error = xlog_recover_process_data(log, rhash, | |
3696 | rhead, offset, pass))) | |
3697 | goto bread_err2; | |
3698 | blk_no += bblks; | |
3699 | } | |
3700 | ||
3701 | ASSERT(blk_no >= log->l_logBBsize); | |
3702 | blk_no -= log->l_logBBsize; | |
3703 | ||
3704 | /* read first part of physical log */ | |
3705 | while (blk_no < head_blk) { | |
3706 | if ((error = xlog_bread(log, blk_no, hblks, hbp))) | |
3707 | goto bread_err2; | |
3708 | offset = xlog_align(log, blk_no, hblks, hbp); | |
3709 | rhead = (xlog_rec_header_t *)offset; | |
3710 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3711 | if (error) | |
3712 | goto bread_err2; | |
3713 | bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)); | |
3714 | if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp))) | |
3715 | goto bread_err2; | |
3716 | offset = xlog_align(log, blk_no+hblks, bblks, dbp); | |
3717 | xlog_unpack_data(rhead, offset, log); | |
3718 | if ((error = xlog_recover_process_data(log, rhash, | |
3719 | rhead, offset, pass))) | |
3720 | goto bread_err2; | |
3721 | blk_no += bblks + hblks; | |
3722 | } | |
3723 | } | |
3724 | ||
3725 | bread_err2: | |
3726 | xlog_put_bp(dbp); | |
3727 | bread_err1: | |
3728 | xlog_put_bp(hbp); | |
3729 | return error; | |
3730 | } | |
3731 | ||
3732 | /* | |
3733 | * Do the recovery of the log. We actually do this in two phases. | |
3734 | * The two passes are necessary in order to implement the function | |
3735 | * of cancelling a record written into the log. The first pass | |
3736 | * determines those things which have been cancelled, and the | |
3737 | * second pass replays log items normally except for those which | |
3738 | * have been cancelled. The handling of the replay and cancellations | |
3739 | * takes place in the log item type specific routines. | |
3740 | * | |
3741 | * The table of items which have cancel records in the log is allocated | |
3742 | * and freed at this level, since only here do we know when all of | |
3743 | * the log recovery has been completed. | |
3744 | */ | |
3745 | STATIC int | |
3746 | xlog_do_log_recovery( | |
3747 | xlog_t *log, | |
3748 | xfs_daddr_t head_blk, | |
3749 | xfs_daddr_t tail_blk) | |
3750 | { | |
3751 | int error; | |
3752 | ||
3753 | ASSERT(head_blk != tail_blk); | |
3754 | ||
3755 | /* | |
3756 | * First do a pass to find all of the cancelled buf log items. | |
3757 | * Store them in the buf_cancel_table for use in the second pass. | |
3758 | */ | |
3759 | log->l_buf_cancel_table = | |
3760 | (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE * | |
3761 | sizeof(xfs_buf_cancel_t*), | |
3762 | KM_SLEEP); | |
3763 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
3764 | XLOG_RECOVER_PASS1); | |
3765 | if (error != 0) { | |
3766 | kmem_free(log->l_buf_cancel_table, | |
3767 | XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*)); | |
3768 | log->l_buf_cancel_table = NULL; | |
3769 | return error; | |
3770 | } | |
3771 | /* | |
3772 | * Then do a second pass to actually recover the items in the log. | |
3773 | * When it is complete free the table of buf cancel items. | |
3774 | */ | |
3775 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
3776 | XLOG_RECOVER_PASS2); | |
3777 | #ifdef DEBUG | |
6d192a9b | 3778 | if (!error) { |
1da177e4 LT |
3779 | int i; |
3780 | ||
3781 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
3782 | ASSERT(log->l_buf_cancel_table[i] == NULL); | |
3783 | } | |
3784 | #endif /* DEBUG */ | |
3785 | ||
3786 | kmem_free(log->l_buf_cancel_table, | |
3787 | XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*)); | |
3788 | log->l_buf_cancel_table = NULL; | |
3789 | ||
3790 | return error; | |
3791 | } | |
3792 | ||
3793 | /* | |
3794 | * Do the actual recovery | |
3795 | */ | |
3796 | STATIC int | |
3797 | xlog_do_recover( | |
3798 | xlog_t *log, | |
3799 | xfs_daddr_t head_blk, | |
3800 | xfs_daddr_t tail_blk) | |
3801 | { | |
3802 | int error; | |
3803 | xfs_buf_t *bp; | |
3804 | xfs_sb_t *sbp; | |
3805 | ||
3806 | /* | |
3807 | * First replay the images in the log. | |
3808 | */ | |
3809 | error = xlog_do_log_recovery(log, head_blk, tail_blk); | |
3810 | if (error) { | |
3811 | return error; | |
3812 | } | |
3813 | ||
3814 | XFS_bflush(log->l_mp->m_ddev_targp); | |
3815 | ||
3816 | /* | |
3817 | * If IO errors happened during recovery, bail out. | |
3818 | */ | |
3819 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) { | |
3820 | return (EIO); | |
3821 | } | |
3822 | ||
3823 | /* | |
3824 | * We now update the tail_lsn since much of the recovery has completed | |
3825 | * and there may be space available to use. If there were no extent | |
3826 | * or iunlinks, we can free up the entire log and set the tail_lsn to | |
3827 | * be the last_sync_lsn. This was set in xlog_find_tail to be the | |
3828 | * lsn of the last known good LR on disk. If there are extent frees | |
3829 | * or iunlinks they will have some entries in the AIL; so we look at | |
3830 | * the AIL to determine how to set the tail_lsn. | |
3831 | */ | |
3832 | xlog_assign_tail_lsn(log->l_mp); | |
3833 | ||
3834 | /* | |
3835 | * Now that we've finished replaying all buffer and inode | |
3836 | * updates, re-read in the superblock. | |
3837 | */ | |
3838 | bp = xfs_getsb(log->l_mp, 0); | |
3839 | XFS_BUF_UNDONE(bp); | |
3840 | XFS_BUF_READ(bp); | |
3841 | xfsbdstrat(log->l_mp, bp); | |
3842 | if ((error = xfs_iowait(bp))) { | |
3843 | xfs_ioerror_alert("xlog_do_recover", | |
3844 | log->l_mp, bp, XFS_BUF_ADDR(bp)); | |
3845 | ASSERT(0); | |
3846 | xfs_buf_relse(bp); | |
3847 | return error; | |
3848 | } | |
3849 | ||
3850 | /* Convert superblock from on-disk format */ | |
3851 | sbp = &log->l_mp->m_sb; | |
3852 | xfs_xlatesb(XFS_BUF_TO_SBP(bp), sbp, 1, XFS_SB_ALL_BITS); | |
3853 | ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC); | |
3854 | ASSERT(XFS_SB_GOOD_VERSION(sbp)); | |
3855 | xfs_buf_relse(bp); | |
3856 | ||
5478eead LM |
3857 | /* We've re-read the superblock so re-initialize per-cpu counters */ |
3858 | xfs_icsb_reinit_counters(log->l_mp); | |
3859 | ||
1da177e4 LT |
3860 | xlog_recover_check_summary(log); |
3861 | ||
3862 | /* Normal transactions can now occur */ | |
3863 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
3864 | return 0; | |
3865 | } | |
3866 | ||
3867 | /* | |
3868 | * Perform recovery and re-initialize some log variables in xlog_find_tail. | |
3869 | * | |
3870 | * Return error or zero. | |
3871 | */ | |
3872 | int | |
3873 | xlog_recover( | |
65be6054 | 3874 | xlog_t *log) |
1da177e4 LT |
3875 | { |
3876 | xfs_daddr_t head_blk, tail_blk; | |
3877 | int error; | |
3878 | ||
3879 | /* find the tail of the log */ | |
65be6054 | 3880 | if ((error = xlog_find_tail(log, &head_blk, &tail_blk))) |
1da177e4 LT |
3881 | return error; |
3882 | ||
3883 | if (tail_blk != head_blk) { | |
3884 | /* There used to be a comment here: | |
3885 | * | |
3886 | * disallow recovery on read-only mounts. note -- mount | |
3887 | * checks for ENOSPC and turns it into an intelligent | |
3888 | * error message. | |
3889 | * ...but this is no longer true. Now, unless you specify | |
3890 | * NORECOVERY (in which case this function would never be | |
3891 | * called), we just go ahead and recover. We do this all | |
3892 | * under the vfs layer, so we can get away with it unless | |
3893 | * the device itself is read-only, in which case we fail. | |
3894 | */ | |
3a02ee18 | 3895 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
1da177e4 LT |
3896 | return error; |
3897 | } | |
3898 | ||
3899 | cmn_err(CE_NOTE, | |
fc1f8c1c NS |
3900 | "Starting XFS recovery on filesystem: %s (logdev: %s)", |
3901 | log->l_mp->m_fsname, log->l_mp->m_logname ? | |
3902 | log->l_mp->m_logname : "internal"); | |
1da177e4 LT |
3903 | |
3904 | error = xlog_do_recover(log, head_blk, tail_blk); | |
3905 | log->l_flags |= XLOG_RECOVERY_NEEDED; | |
3906 | } | |
3907 | return error; | |
3908 | } | |
3909 | ||
3910 | /* | |
3911 | * In the first part of recovery we replay inodes and buffers and build | |
3912 | * up the list of extent free items which need to be processed. Here | |
3913 | * we process the extent free items and clean up the on disk unlinked | |
3914 | * inode lists. This is separated from the first part of recovery so | |
3915 | * that the root and real-time bitmap inodes can be read in from disk in | |
3916 | * between the two stages. This is necessary so that we can free space | |
3917 | * in the real-time portion of the file system. | |
3918 | */ | |
3919 | int | |
3920 | xlog_recover_finish( | |
3921 | xlog_t *log, | |
3922 | int mfsi_flags) | |
3923 | { | |
3924 | /* | |
3925 | * Now we're ready to do the transactions needed for the | |
3926 | * rest of recovery. Start with completing all the extent | |
3927 | * free intent records and then process the unlinked inode | |
3928 | * lists. At this point, we essentially run in normal mode | |
3929 | * except that we're still performing recovery actions | |
3930 | * rather than accepting new requests. | |
3931 | */ | |
3932 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { | |
3933 | xlog_recover_process_efis(log); | |
3934 | /* | |
3935 | * Sync the log to get all the EFIs out of the AIL. | |
3936 | * This isn't absolutely necessary, but it helps in | |
3937 | * case the unlink transactions would have problems | |
3938 | * pushing the EFIs out of the way. | |
3939 | */ | |
3940 | xfs_log_force(log->l_mp, (xfs_lsn_t)0, | |
3941 | (XFS_LOG_FORCE | XFS_LOG_SYNC)); | |
3942 | ||
3943 | if ( (mfsi_flags & XFS_MFSI_NOUNLINK) == 0 ) { | |
3944 | xlog_recover_process_iunlinks(log); | |
3945 | } | |
3946 | ||
3947 | xlog_recover_check_summary(log); | |
3948 | ||
3949 | cmn_err(CE_NOTE, | |
fc1f8c1c NS |
3950 | "Ending XFS recovery on filesystem: %s (logdev: %s)", |
3951 | log->l_mp->m_fsname, log->l_mp->m_logname ? | |
3952 | log->l_mp->m_logname : "internal"); | |
1da177e4 LT |
3953 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
3954 | } else { | |
3955 | cmn_err(CE_DEBUG, | |
b6574520 | 3956 | "!Ending clean XFS mount for filesystem: %s\n", |
1da177e4 LT |
3957 | log->l_mp->m_fsname); |
3958 | } | |
3959 | return 0; | |
3960 | } | |
3961 | ||
3962 | ||
3963 | #if defined(DEBUG) | |
3964 | /* | |
3965 | * Read all of the agf and agi counters and check that they | |
3966 | * are consistent with the superblock counters. | |
3967 | */ | |
3968 | void | |
3969 | xlog_recover_check_summary( | |
3970 | xlog_t *log) | |
3971 | { | |
3972 | xfs_mount_t *mp; | |
3973 | xfs_agf_t *agfp; | |
3974 | xfs_agi_t *agip; | |
3975 | xfs_buf_t *agfbp; | |
3976 | xfs_buf_t *agibp; | |
3977 | xfs_daddr_t agfdaddr; | |
3978 | xfs_daddr_t agidaddr; | |
3979 | xfs_buf_t *sbbp; | |
3980 | #ifdef XFS_LOUD_RECOVERY | |
3981 | xfs_sb_t *sbp; | |
3982 | #endif | |
3983 | xfs_agnumber_t agno; | |
3984 | __uint64_t freeblks; | |
3985 | __uint64_t itotal; | |
3986 | __uint64_t ifree; | |
3987 | ||
3988 | mp = log->l_mp; | |
3989 | ||
3990 | freeblks = 0LL; | |
3991 | itotal = 0LL; | |
3992 | ifree = 0LL; | |
3993 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
3994 | agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)); | |
3995 | agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr, | |
3996 | XFS_FSS_TO_BB(mp, 1), 0); | |
3997 | if (XFS_BUF_ISERROR(agfbp)) { | |
3998 | xfs_ioerror_alert("xlog_recover_check_summary(agf)", | |
3999 | mp, agfbp, agfdaddr); | |
4000 | } | |
4001 | agfp = XFS_BUF_TO_AGF(agfbp); | |
16259e7d CH |
4002 | ASSERT(XFS_AGF_MAGIC == be32_to_cpu(agfp->agf_magicnum)); |
4003 | ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp->agf_versionnum))); | |
4004 | ASSERT(be32_to_cpu(agfp->agf_seqno) == agno); | |
4005 | ||
4006 | freeblks += be32_to_cpu(agfp->agf_freeblks) + | |
4007 | be32_to_cpu(agfp->agf_flcount); | |
1da177e4 LT |
4008 | xfs_buf_relse(agfbp); |
4009 | ||
4010 | agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); | |
4011 | agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr, | |
4012 | XFS_FSS_TO_BB(mp, 1), 0); | |
4013 | if (XFS_BUF_ISERROR(agibp)) { | |
4014 | xfs_ioerror_alert("xlog_recover_check_summary(agi)", | |
4015 | mp, agibp, agidaddr); | |
4016 | } | |
4017 | agip = XFS_BUF_TO_AGI(agibp); | |
16259e7d CH |
4018 | ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agip->agi_magicnum)); |
4019 | ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip->agi_versionnum))); | |
4020 | ASSERT(be32_to_cpu(agip->agi_seqno) == agno); | |
4021 | ||
4022 | itotal += be32_to_cpu(agip->agi_count); | |
4023 | ifree += be32_to_cpu(agip->agi_freecount); | |
1da177e4 LT |
4024 | xfs_buf_relse(agibp); |
4025 | } | |
4026 | ||
4027 | sbbp = xfs_getsb(mp, 0); | |
4028 | #ifdef XFS_LOUD_RECOVERY | |
4029 | sbp = &mp->m_sb; | |
4030 | xfs_xlatesb(XFS_BUF_TO_SBP(sbbp), sbp, 1, XFS_SB_ALL_BITS); | |
4031 | cmn_err(CE_NOTE, | |
4032 | "xlog_recover_check_summary: sb_icount %Lu itotal %Lu", | |
4033 | sbp->sb_icount, itotal); | |
4034 | cmn_err(CE_NOTE, | |
4035 | "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu", | |
4036 | sbp->sb_ifree, ifree); | |
4037 | cmn_err(CE_NOTE, | |
4038 | "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu", | |
4039 | sbp->sb_fdblocks, freeblks); | |
4040 | #if 0 | |
4041 | /* | |
4042 | * This is turned off until I account for the allocation | |
4043 | * btree blocks which live in free space. | |
4044 | */ | |
4045 | ASSERT(sbp->sb_icount == itotal); | |
4046 | ASSERT(sbp->sb_ifree == ifree); | |
4047 | ASSERT(sbp->sb_fdblocks == freeblks); | |
4048 | #endif | |
4049 | #endif | |
4050 | xfs_buf_relse(sbbp); | |
4051 | } | |
4052 | #endif /* DEBUG */ |