Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
87c199c2 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
a844f451 NS |
25 | #include "xfs_sb.h" |
26 | #include "xfs_ag.h" | |
1da177e4 LT |
27 | #include "xfs_dir2.h" |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_mount.h" | |
30 | #include "xfs_error.h" | |
31 | #include "xfs_bmap_btree.h" | |
a844f451 NS |
32 | #include "xfs_alloc_btree.h" |
33 | #include "xfs_ialloc_btree.h" | |
1da177e4 | 34 | #include "xfs_dir2_sf.h" |
a844f451 | 35 | #include "xfs_attr_sf.h" |
1da177e4 | 36 | #include "xfs_dinode.h" |
1da177e4 | 37 | #include "xfs_inode.h" |
a844f451 | 38 | #include "xfs_inode_item.h" |
a844f451 | 39 | #include "xfs_alloc.h" |
1da177e4 LT |
40 | #include "xfs_ialloc.h" |
41 | #include "xfs_log_priv.h" | |
42 | #include "xfs_buf_item.h" | |
1da177e4 LT |
43 | #include "xfs_log_recover.h" |
44 | #include "xfs_extfree_item.h" | |
45 | #include "xfs_trans_priv.h" | |
1da177e4 LT |
46 | #include "xfs_quota.h" |
47 | #include "xfs_rw.h" | |
43355099 | 48 | #include "xfs_utils.h" |
0b1b213f | 49 | #include "xfs_trace.h" |
1da177e4 LT |
50 | |
51 | STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *); | |
52 | STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t); | |
1da177e4 LT |
53 | #if defined(DEBUG) |
54 | STATIC void xlog_recover_check_summary(xlog_t *); | |
1da177e4 LT |
55 | #else |
56 | #define xlog_recover_check_summary(log) | |
1da177e4 LT |
57 | #endif |
58 | ||
1da177e4 LT |
59 | /* |
60 | * Sector aligned buffer routines for buffer create/read/write/access | |
61 | */ | |
62 | ||
6881a229 AE |
63 | /* Number of basic blocks in a log sector */ |
64 | #define xlog_sectbb(log) (1 << (log)->l_sectbb_log) | |
65 | ||
ff30a622 AE |
66 | /* |
67 | * Verify the given count of basic blocks is valid number of blocks | |
68 | * to specify for an operation involving the given XFS log buffer. | |
69 | * Returns nonzero if the count is valid, 0 otherwise. | |
70 | */ | |
71 | ||
72 | static inline int | |
73 | xlog_buf_bbcount_valid( | |
74 | xlog_t *log, | |
75 | int bbcount) | |
76 | { | |
77 | return bbcount > 0 && bbcount <= log->l_logBBsize; | |
78 | } | |
79 | ||
36adecff AE |
80 | /* |
81 | * Allocate a buffer to hold log data. The buffer needs to be able | |
82 | * to map to a range of nbblks basic blocks at any valid (basic | |
83 | * block) offset within the log. | |
84 | */ | |
5d77c0dc | 85 | STATIC xfs_buf_t * |
1da177e4 LT |
86 | xlog_get_bp( |
87 | xlog_t *log, | |
3228149c | 88 | int nbblks) |
1da177e4 | 89 | { |
ff30a622 AE |
90 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
91 | xlog_warn("XFS: Invalid block length (0x%x) given for buffer", | |
92 | nbblks); | |
93 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
94 | return NULL; |
95 | } | |
1da177e4 | 96 | |
36adecff AE |
97 | /* |
98 | * We do log I/O in units of log sectors (a power-of-2 | |
99 | * multiple of the basic block size), so we round up the | |
100 | * requested size to acommodate the basic blocks required | |
101 | * for complete log sectors. | |
102 | * | |
103 | * In addition, the buffer may be used for a non-sector- | |
104 | * aligned block offset, in which case an I/O of the | |
105 | * requested size could extend beyond the end of the | |
106 | * buffer. If the requested size is only 1 basic block it | |
107 | * will never straddle a sector boundary, so this won't be | |
108 | * an issue. Nor will this be a problem if the log I/O is | |
109 | * done in basic blocks (sector size 1). But otherwise we | |
110 | * extend the buffer by one extra log sector to ensure | |
111 | * there's space to accomodate this possiblility. | |
112 | */ | |
113 | if (nbblks > 1 && log->l_sectbb_log) | |
114 | nbblks += xlog_sectbb(log); | |
115 | nbblks = round_up(nbblks, xlog_sectbb(log)); | |
116 | ||
3228149c | 117 | return xfs_buf_get_noaddr(BBTOB(nbblks), log->l_mp->m_logdev_targp); |
1da177e4 LT |
118 | } |
119 | ||
5d77c0dc | 120 | STATIC void |
1da177e4 LT |
121 | xlog_put_bp( |
122 | xfs_buf_t *bp) | |
123 | { | |
124 | xfs_buf_free(bp); | |
125 | } | |
126 | ||
076e6acb CH |
127 | STATIC xfs_caddr_t |
128 | xlog_align( | |
129 | xlog_t *log, | |
130 | xfs_daddr_t blk_no, | |
131 | int nbblks, | |
132 | xfs_buf_t *bp) | |
133 | { | |
134 | xfs_caddr_t ptr; | |
135 | ||
136 | if (!log->l_sectbb_log) | |
137 | return XFS_BUF_PTR(bp); | |
138 | ||
139 | ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask); | |
140 | ASSERT(XFS_BUF_SIZE(bp) >= | |
141 | BBTOB(nbblks + (blk_no & log->l_sectbb_mask))); | |
142 | return ptr; | |
143 | } | |
144 | ||
1da177e4 LT |
145 | |
146 | /* | |
147 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. | |
148 | */ | |
076e6acb CH |
149 | STATIC int |
150 | xlog_bread_noalign( | |
1da177e4 LT |
151 | xlog_t *log, |
152 | xfs_daddr_t blk_no, | |
153 | int nbblks, | |
154 | xfs_buf_t *bp) | |
155 | { | |
156 | int error; | |
157 | ||
ff30a622 AE |
158 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
159 | xlog_warn("XFS: Invalid block length (0x%x) given for buffer", | |
160 | nbblks); | |
161 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
162 | return EFSCORRUPTED; |
163 | } | |
164 | ||
36adecff AE |
165 | blk_no = round_down(blk_no, xlog_sectbb(log)); |
166 | nbblks = round_up(nbblks, xlog_sectbb(log)); | |
1da177e4 LT |
167 | |
168 | ASSERT(nbblks > 0); | |
169 | ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp)); | |
170 | ASSERT(bp); | |
171 | ||
172 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
173 | XFS_BUF_READ(bp); | |
174 | XFS_BUF_BUSY(bp); | |
175 | XFS_BUF_SET_COUNT(bp, BBTOB(nbblks)); | |
176 | XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp); | |
177 | ||
178 | xfsbdstrat(log->l_mp, bp); | |
d64e31a2 DC |
179 | error = xfs_iowait(bp); |
180 | if (error) | |
1da177e4 LT |
181 | xfs_ioerror_alert("xlog_bread", log->l_mp, |
182 | bp, XFS_BUF_ADDR(bp)); | |
183 | return error; | |
184 | } | |
185 | ||
076e6acb CH |
186 | STATIC int |
187 | xlog_bread( | |
188 | xlog_t *log, | |
189 | xfs_daddr_t blk_no, | |
190 | int nbblks, | |
191 | xfs_buf_t *bp, | |
192 | xfs_caddr_t *offset) | |
193 | { | |
194 | int error; | |
195 | ||
196 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
197 | if (error) | |
198 | return error; | |
199 | ||
200 | *offset = xlog_align(log, blk_no, nbblks, bp); | |
201 | return 0; | |
202 | } | |
203 | ||
1da177e4 LT |
204 | /* |
205 | * Write out the buffer at the given block for the given number of blocks. | |
206 | * The buffer is kept locked across the write and is returned locked. | |
207 | * This can only be used for synchronous log writes. | |
208 | */ | |
ba0f32d4 | 209 | STATIC int |
1da177e4 LT |
210 | xlog_bwrite( |
211 | xlog_t *log, | |
212 | xfs_daddr_t blk_no, | |
213 | int nbblks, | |
214 | xfs_buf_t *bp) | |
215 | { | |
216 | int error; | |
217 | ||
ff30a622 AE |
218 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
219 | xlog_warn("XFS: Invalid block length (0x%x) given for buffer", | |
220 | nbblks); | |
221 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
222 | return EFSCORRUPTED; |
223 | } | |
224 | ||
36adecff AE |
225 | blk_no = round_down(blk_no, xlog_sectbb(log)); |
226 | nbblks = round_up(nbblks, xlog_sectbb(log)); | |
1da177e4 LT |
227 | |
228 | ASSERT(nbblks > 0); | |
229 | ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp)); | |
230 | ||
231 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
232 | XFS_BUF_ZEROFLAGS(bp); | |
233 | XFS_BUF_BUSY(bp); | |
234 | XFS_BUF_HOLD(bp); | |
235 | XFS_BUF_PSEMA(bp, PRIBIO); | |
236 | XFS_BUF_SET_COUNT(bp, BBTOB(nbblks)); | |
237 | XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp); | |
238 | ||
239 | if ((error = xfs_bwrite(log->l_mp, bp))) | |
240 | xfs_ioerror_alert("xlog_bwrite", log->l_mp, | |
241 | bp, XFS_BUF_ADDR(bp)); | |
242 | return error; | |
243 | } | |
244 | ||
1da177e4 LT |
245 | #ifdef DEBUG |
246 | /* | |
247 | * dump debug superblock and log record information | |
248 | */ | |
249 | STATIC void | |
250 | xlog_header_check_dump( | |
251 | xfs_mount_t *mp, | |
252 | xlog_rec_header_t *head) | |
253 | { | |
03daa57c JP |
254 | cmn_err(CE_DEBUG, "%s: SB : uuid = %pU, fmt = %d\n", |
255 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); | |
256 | cmn_err(CE_DEBUG, " log : uuid = %pU, fmt = %d\n", | |
257 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); | |
1da177e4 LT |
258 | } |
259 | #else | |
260 | #define xlog_header_check_dump(mp, head) | |
261 | #endif | |
262 | ||
263 | /* | |
264 | * check log record header for recovery | |
265 | */ | |
266 | STATIC int | |
267 | xlog_header_check_recover( | |
268 | xfs_mount_t *mp, | |
269 | xlog_rec_header_t *head) | |
270 | { | |
b53e675d | 271 | ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM); |
1da177e4 LT |
272 | |
273 | /* | |
274 | * IRIX doesn't write the h_fmt field and leaves it zeroed | |
275 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover | |
276 | * a dirty log created in IRIX. | |
277 | */ | |
b53e675d | 278 | if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) { |
1da177e4 LT |
279 | xlog_warn( |
280 | "XFS: dirty log written in incompatible format - can't recover"); | |
281 | xlog_header_check_dump(mp, head); | |
282 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", | |
283 | XFS_ERRLEVEL_HIGH, mp); | |
284 | return XFS_ERROR(EFSCORRUPTED); | |
285 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
286 | xlog_warn( | |
287 | "XFS: dirty log entry has mismatched uuid - can't recover"); | |
288 | xlog_header_check_dump(mp, head); | |
289 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", | |
290 | XFS_ERRLEVEL_HIGH, mp); | |
291 | return XFS_ERROR(EFSCORRUPTED); | |
292 | } | |
293 | return 0; | |
294 | } | |
295 | ||
296 | /* | |
297 | * read the head block of the log and check the header | |
298 | */ | |
299 | STATIC int | |
300 | xlog_header_check_mount( | |
301 | xfs_mount_t *mp, | |
302 | xlog_rec_header_t *head) | |
303 | { | |
b53e675d | 304 | ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM); |
1da177e4 LT |
305 | |
306 | if (uuid_is_nil(&head->h_fs_uuid)) { | |
307 | /* | |
308 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If | |
309 | * h_fs_uuid is nil, we assume this log was last mounted | |
310 | * by IRIX and continue. | |
311 | */ | |
312 | xlog_warn("XFS: nil uuid in log - IRIX style log"); | |
313 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
314 | xlog_warn("XFS: log has mismatched uuid - can't recover"); | |
315 | xlog_header_check_dump(mp, head); | |
316 | XFS_ERROR_REPORT("xlog_header_check_mount", | |
317 | XFS_ERRLEVEL_HIGH, mp); | |
318 | return XFS_ERROR(EFSCORRUPTED); | |
319 | } | |
320 | return 0; | |
321 | } | |
322 | ||
323 | STATIC void | |
324 | xlog_recover_iodone( | |
325 | struct xfs_buf *bp) | |
326 | { | |
1da177e4 LT |
327 | if (XFS_BUF_GETERROR(bp)) { |
328 | /* | |
329 | * We're not going to bother about retrying | |
330 | * this during recovery. One strike! | |
331 | */ | |
1da177e4 | 332 | xfs_ioerror_alert("xlog_recover_iodone", |
15ac08a8 CH |
333 | bp->b_mount, bp, XFS_BUF_ADDR(bp)); |
334 | xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR); | |
1da177e4 | 335 | } |
15ac08a8 | 336 | bp->b_mount = NULL; |
1da177e4 LT |
337 | XFS_BUF_CLR_IODONE_FUNC(bp); |
338 | xfs_biodone(bp); | |
339 | } | |
340 | ||
341 | /* | |
342 | * This routine finds (to an approximation) the first block in the physical | |
343 | * log which contains the given cycle. It uses a binary search algorithm. | |
344 | * Note that the algorithm can not be perfect because the disk will not | |
345 | * necessarily be perfect. | |
346 | */ | |
a8272ce0 | 347 | STATIC int |
1da177e4 LT |
348 | xlog_find_cycle_start( |
349 | xlog_t *log, | |
350 | xfs_buf_t *bp, | |
351 | xfs_daddr_t first_blk, | |
352 | xfs_daddr_t *last_blk, | |
353 | uint cycle) | |
354 | { | |
355 | xfs_caddr_t offset; | |
356 | xfs_daddr_t mid_blk; | |
357 | uint mid_cycle; | |
358 | int error; | |
359 | ||
360 | mid_blk = BLK_AVG(first_blk, *last_blk); | |
361 | while (mid_blk != first_blk && mid_blk != *last_blk) { | |
076e6acb CH |
362 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
363 | if (error) | |
1da177e4 | 364 | return error; |
03bea6fe | 365 | mid_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
366 | if (mid_cycle == cycle) { |
367 | *last_blk = mid_blk; | |
368 | /* last_half_cycle == mid_cycle */ | |
369 | } else { | |
370 | first_blk = mid_blk; | |
371 | /* first_half_cycle == mid_cycle */ | |
372 | } | |
373 | mid_blk = BLK_AVG(first_blk, *last_blk); | |
374 | } | |
375 | ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) || | |
376 | (mid_blk == *last_blk && mid_blk-1 == first_blk)); | |
377 | ||
378 | return 0; | |
379 | } | |
380 | ||
381 | /* | |
382 | * Check that the range of blocks does not contain the cycle number | |
383 | * given. The scan needs to occur from front to back and the ptr into the | |
384 | * region must be updated since a later routine will need to perform another | |
385 | * test. If the region is completely good, we end up returning the same | |
386 | * last block number. | |
387 | * | |
388 | * Set blkno to -1 if we encounter no errors. This is an invalid block number | |
389 | * since we don't ever expect logs to get this large. | |
390 | */ | |
391 | STATIC int | |
392 | xlog_find_verify_cycle( | |
393 | xlog_t *log, | |
394 | xfs_daddr_t start_blk, | |
395 | int nbblks, | |
396 | uint stop_on_cycle_no, | |
397 | xfs_daddr_t *new_blk) | |
398 | { | |
399 | xfs_daddr_t i, j; | |
400 | uint cycle; | |
401 | xfs_buf_t *bp; | |
402 | xfs_daddr_t bufblks; | |
403 | xfs_caddr_t buf = NULL; | |
404 | int error = 0; | |
405 | ||
6881a229 AE |
406 | /* |
407 | * Greedily allocate a buffer big enough to handle the full | |
408 | * range of basic blocks we'll be examining. If that fails, | |
409 | * try a smaller size. We need to be able to read at least | |
410 | * a log sector, or we're out of luck. | |
411 | */ | |
1da177e4 | 412 | bufblks = 1 << ffs(nbblks); |
1da177e4 | 413 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1da177e4 | 414 | bufblks >>= 1; |
6881a229 | 415 | if (bufblks < xlog_sectbb(log)) |
1da177e4 LT |
416 | return ENOMEM; |
417 | } | |
418 | ||
419 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { | |
420 | int bcount; | |
421 | ||
422 | bcount = min(bufblks, (start_blk + nbblks - i)); | |
423 | ||
076e6acb CH |
424 | error = xlog_bread(log, i, bcount, bp, &buf); |
425 | if (error) | |
1da177e4 LT |
426 | goto out; |
427 | ||
1da177e4 | 428 | for (j = 0; j < bcount; j++) { |
03bea6fe | 429 | cycle = xlog_get_cycle(buf); |
1da177e4 LT |
430 | if (cycle == stop_on_cycle_no) { |
431 | *new_blk = i+j; | |
432 | goto out; | |
433 | } | |
434 | ||
435 | buf += BBSIZE; | |
436 | } | |
437 | } | |
438 | ||
439 | *new_blk = -1; | |
440 | ||
441 | out: | |
442 | xlog_put_bp(bp); | |
443 | return error; | |
444 | } | |
445 | ||
446 | /* | |
447 | * Potentially backup over partial log record write. | |
448 | * | |
449 | * In the typical case, last_blk is the number of the block directly after | |
450 | * a good log record. Therefore, we subtract one to get the block number | |
451 | * of the last block in the given buffer. extra_bblks contains the number | |
452 | * of blocks we would have read on a previous read. This happens when the | |
453 | * last log record is split over the end of the physical log. | |
454 | * | |
455 | * extra_bblks is the number of blocks potentially verified on a previous | |
456 | * call to this routine. | |
457 | */ | |
458 | STATIC int | |
459 | xlog_find_verify_log_record( | |
460 | xlog_t *log, | |
461 | xfs_daddr_t start_blk, | |
462 | xfs_daddr_t *last_blk, | |
463 | int extra_bblks) | |
464 | { | |
465 | xfs_daddr_t i; | |
466 | xfs_buf_t *bp; | |
467 | xfs_caddr_t offset = NULL; | |
468 | xlog_rec_header_t *head = NULL; | |
469 | int error = 0; | |
470 | int smallmem = 0; | |
471 | int num_blks = *last_blk - start_blk; | |
472 | int xhdrs; | |
473 | ||
474 | ASSERT(start_blk != 0 || *last_blk != start_blk); | |
475 | ||
476 | if (!(bp = xlog_get_bp(log, num_blks))) { | |
477 | if (!(bp = xlog_get_bp(log, 1))) | |
478 | return ENOMEM; | |
479 | smallmem = 1; | |
480 | } else { | |
076e6acb CH |
481 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
482 | if (error) | |
1da177e4 | 483 | goto out; |
1da177e4 LT |
484 | offset += ((num_blks - 1) << BBSHIFT); |
485 | } | |
486 | ||
487 | for (i = (*last_blk) - 1; i >= 0; i--) { | |
488 | if (i < start_blk) { | |
489 | /* valid log record not found */ | |
490 | xlog_warn( | |
491 | "XFS: Log inconsistent (didn't find previous header)"); | |
492 | ASSERT(0); | |
493 | error = XFS_ERROR(EIO); | |
494 | goto out; | |
495 | } | |
496 | ||
497 | if (smallmem) { | |
076e6acb CH |
498 | error = xlog_bread(log, i, 1, bp, &offset); |
499 | if (error) | |
1da177e4 | 500 | goto out; |
1da177e4 LT |
501 | } |
502 | ||
503 | head = (xlog_rec_header_t *)offset; | |
504 | ||
b53e675d | 505 | if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno)) |
1da177e4 LT |
506 | break; |
507 | ||
508 | if (!smallmem) | |
509 | offset -= BBSIZE; | |
510 | } | |
511 | ||
512 | /* | |
513 | * We hit the beginning of the physical log & still no header. Return | |
514 | * to caller. If caller can handle a return of -1, then this routine | |
515 | * will be called again for the end of the physical log. | |
516 | */ | |
517 | if (i == -1) { | |
518 | error = -1; | |
519 | goto out; | |
520 | } | |
521 | ||
522 | /* | |
523 | * We have the final block of the good log (the first block | |
524 | * of the log record _before_ the head. So we check the uuid. | |
525 | */ | |
526 | if ((error = xlog_header_check_mount(log->l_mp, head))) | |
527 | goto out; | |
528 | ||
529 | /* | |
530 | * We may have found a log record header before we expected one. | |
531 | * last_blk will be the 1st block # with a given cycle #. We may end | |
532 | * up reading an entire log record. In this case, we don't want to | |
533 | * reset last_blk. Only when last_blk points in the middle of a log | |
534 | * record do we update last_blk. | |
535 | */ | |
62118709 | 536 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d | 537 | uint h_size = be32_to_cpu(head->h_size); |
1da177e4 LT |
538 | |
539 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; | |
540 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
541 | xhdrs++; | |
542 | } else { | |
543 | xhdrs = 1; | |
544 | } | |
545 | ||
b53e675d CH |
546 | if (*last_blk - i + extra_bblks != |
547 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) | |
1da177e4 LT |
548 | *last_blk = i; |
549 | ||
550 | out: | |
551 | xlog_put_bp(bp); | |
552 | return error; | |
553 | } | |
554 | ||
555 | /* | |
556 | * Head is defined to be the point of the log where the next log write | |
557 | * write could go. This means that incomplete LR writes at the end are | |
558 | * eliminated when calculating the head. We aren't guaranteed that previous | |
559 | * LR have complete transactions. We only know that a cycle number of | |
560 | * current cycle number -1 won't be present in the log if we start writing | |
561 | * from our current block number. | |
562 | * | |
563 | * last_blk contains the block number of the first block with a given | |
564 | * cycle number. | |
565 | * | |
566 | * Return: zero if normal, non-zero if error. | |
567 | */ | |
ba0f32d4 | 568 | STATIC int |
1da177e4 LT |
569 | xlog_find_head( |
570 | xlog_t *log, | |
571 | xfs_daddr_t *return_head_blk) | |
572 | { | |
573 | xfs_buf_t *bp; | |
574 | xfs_caddr_t offset; | |
575 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; | |
576 | int num_scan_bblks; | |
577 | uint first_half_cycle, last_half_cycle; | |
578 | uint stop_on_cycle; | |
579 | int error, log_bbnum = log->l_logBBsize; | |
580 | ||
581 | /* Is the end of the log device zeroed? */ | |
582 | if ((error = xlog_find_zeroed(log, &first_blk)) == -1) { | |
583 | *return_head_blk = first_blk; | |
584 | ||
585 | /* Is the whole lot zeroed? */ | |
586 | if (!first_blk) { | |
587 | /* Linux XFS shouldn't generate totally zeroed logs - | |
588 | * mkfs etc write a dummy unmount record to a fresh | |
589 | * log so we can store the uuid in there | |
590 | */ | |
591 | xlog_warn("XFS: totally zeroed log"); | |
592 | } | |
593 | ||
594 | return 0; | |
595 | } else if (error) { | |
596 | xlog_warn("XFS: empty log check failed"); | |
597 | return error; | |
598 | } | |
599 | ||
600 | first_blk = 0; /* get cycle # of 1st block */ | |
601 | bp = xlog_get_bp(log, 1); | |
602 | if (!bp) | |
603 | return ENOMEM; | |
076e6acb CH |
604 | |
605 | error = xlog_bread(log, 0, 1, bp, &offset); | |
606 | if (error) | |
1da177e4 | 607 | goto bp_err; |
076e6acb | 608 | |
03bea6fe | 609 | first_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
610 | |
611 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ | |
076e6acb CH |
612 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
613 | if (error) | |
1da177e4 | 614 | goto bp_err; |
076e6acb | 615 | |
03bea6fe | 616 | last_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
617 | ASSERT(last_half_cycle != 0); |
618 | ||
619 | /* | |
620 | * If the 1st half cycle number is equal to the last half cycle number, | |
621 | * then the entire log is stamped with the same cycle number. In this | |
622 | * case, head_blk can't be set to zero (which makes sense). The below | |
623 | * math doesn't work out properly with head_blk equal to zero. Instead, | |
624 | * we set it to log_bbnum which is an invalid block number, but this | |
625 | * value makes the math correct. If head_blk doesn't changed through | |
626 | * all the tests below, *head_blk is set to zero at the very end rather | |
627 | * than log_bbnum. In a sense, log_bbnum and zero are the same block | |
628 | * in a circular file. | |
629 | */ | |
630 | if (first_half_cycle == last_half_cycle) { | |
631 | /* | |
632 | * In this case we believe that the entire log should have | |
633 | * cycle number last_half_cycle. We need to scan backwards | |
634 | * from the end verifying that there are no holes still | |
635 | * containing last_half_cycle - 1. If we find such a hole, | |
636 | * then the start of that hole will be the new head. The | |
637 | * simple case looks like | |
638 | * x | x ... | x - 1 | x | |
639 | * Another case that fits this picture would be | |
640 | * x | x + 1 | x ... | x | |
c41564b5 | 641 | * In this case the head really is somewhere at the end of the |
1da177e4 LT |
642 | * log, as one of the latest writes at the beginning was |
643 | * incomplete. | |
644 | * One more case is | |
645 | * x | x + 1 | x ... | x - 1 | x | |
646 | * This is really the combination of the above two cases, and | |
647 | * the head has to end up at the start of the x-1 hole at the | |
648 | * end of the log. | |
649 | * | |
650 | * In the 256k log case, we will read from the beginning to the | |
651 | * end of the log and search for cycle numbers equal to x-1. | |
652 | * We don't worry about the x+1 blocks that we encounter, | |
653 | * because we know that they cannot be the head since the log | |
654 | * started with x. | |
655 | */ | |
656 | head_blk = log_bbnum; | |
657 | stop_on_cycle = last_half_cycle - 1; | |
658 | } else { | |
659 | /* | |
660 | * In this case we want to find the first block with cycle | |
661 | * number matching last_half_cycle. We expect the log to be | |
662 | * some variation on | |
663 | * x + 1 ... | x ... | |
664 | * The first block with cycle number x (last_half_cycle) will | |
665 | * be where the new head belongs. First we do a binary search | |
666 | * for the first occurrence of last_half_cycle. The binary | |
667 | * search may not be totally accurate, so then we scan back | |
668 | * from there looking for occurrences of last_half_cycle before | |
669 | * us. If that backwards scan wraps around the beginning of | |
670 | * the log, then we look for occurrences of last_half_cycle - 1 | |
671 | * at the end of the log. The cases we're looking for look | |
672 | * like | |
673 | * x + 1 ... | x | x + 1 | x ... | |
674 | * ^ binary search stopped here | |
675 | * or | |
676 | * x + 1 ... | x ... | x - 1 | x | |
677 | * <---------> less than scan distance | |
678 | */ | |
679 | stop_on_cycle = last_half_cycle; | |
680 | if ((error = xlog_find_cycle_start(log, bp, first_blk, | |
681 | &head_blk, last_half_cycle))) | |
682 | goto bp_err; | |
683 | } | |
684 | ||
685 | /* | |
686 | * Now validate the answer. Scan back some number of maximum possible | |
687 | * blocks and make sure each one has the expected cycle number. The | |
688 | * maximum is determined by the total possible amount of buffering | |
689 | * in the in-core log. The following number can be made tighter if | |
690 | * we actually look at the block size of the filesystem. | |
691 | */ | |
692 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
693 | if (head_blk >= num_scan_bblks) { | |
694 | /* | |
695 | * We are guaranteed that the entire check can be performed | |
696 | * in one buffer. | |
697 | */ | |
698 | start_blk = head_blk - num_scan_bblks; | |
699 | if ((error = xlog_find_verify_cycle(log, | |
700 | start_blk, num_scan_bblks, | |
701 | stop_on_cycle, &new_blk))) | |
702 | goto bp_err; | |
703 | if (new_blk != -1) | |
704 | head_blk = new_blk; | |
705 | } else { /* need to read 2 parts of log */ | |
706 | /* | |
707 | * We are going to scan backwards in the log in two parts. | |
708 | * First we scan the physical end of the log. In this part | |
709 | * of the log, we are looking for blocks with cycle number | |
710 | * last_half_cycle - 1. | |
711 | * If we find one, then we know that the log starts there, as | |
712 | * we've found a hole that didn't get written in going around | |
713 | * the end of the physical log. The simple case for this is | |
714 | * x + 1 ... | x ... | x - 1 | x | |
715 | * <---------> less than scan distance | |
716 | * If all of the blocks at the end of the log have cycle number | |
717 | * last_half_cycle, then we check the blocks at the start of | |
718 | * the log looking for occurrences of last_half_cycle. If we | |
719 | * find one, then our current estimate for the location of the | |
720 | * first occurrence of last_half_cycle is wrong and we move | |
721 | * back to the hole we've found. This case looks like | |
722 | * x + 1 ... | x | x + 1 | x ... | |
723 | * ^ binary search stopped here | |
724 | * Another case we need to handle that only occurs in 256k | |
725 | * logs is | |
726 | * x + 1 ... | x ... | x+1 | x ... | |
727 | * ^ binary search stops here | |
728 | * In a 256k log, the scan at the end of the log will see the | |
729 | * x + 1 blocks. We need to skip past those since that is | |
730 | * certainly not the head of the log. By searching for | |
731 | * last_half_cycle-1 we accomplish that. | |
732 | */ | |
733 | start_blk = log_bbnum - num_scan_bblks + head_blk; | |
734 | ASSERT(head_blk <= INT_MAX && | |
735 | (xfs_daddr_t) num_scan_bblks - head_blk >= 0); | |
736 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
737 | num_scan_bblks - (int)head_blk, | |
738 | (stop_on_cycle - 1), &new_blk))) | |
739 | goto bp_err; | |
740 | if (new_blk != -1) { | |
741 | head_blk = new_blk; | |
9db127ed | 742 | goto validate_head; |
1da177e4 LT |
743 | } |
744 | ||
745 | /* | |
746 | * Scan beginning of log now. The last part of the physical | |
747 | * log is good. This scan needs to verify that it doesn't find | |
748 | * the last_half_cycle. | |
749 | */ | |
750 | start_blk = 0; | |
751 | ASSERT(head_blk <= INT_MAX); | |
752 | if ((error = xlog_find_verify_cycle(log, | |
753 | start_blk, (int)head_blk, | |
754 | stop_on_cycle, &new_blk))) | |
755 | goto bp_err; | |
756 | if (new_blk != -1) | |
757 | head_blk = new_blk; | |
758 | } | |
759 | ||
9db127ed | 760 | validate_head: |
1da177e4 LT |
761 | /* |
762 | * Now we need to make sure head_blk is not pointing to a block in | |
763 | * the middle of a log record. | |
764 | */ | |
765 | num_scan_bblks = XLOG_REC_SHIFT(log); | |
766 | if (head_blk >= num_scan_bblks) { | |
767 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ | |
768 | ||
769 | /* start ptr at last block ptr before head_blk */ | |
770 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
771 | &head_blk, 0)) == -1) { | |
772 | error = XFS_ERROR(EIO); | |
773 | goto bp_err; | |
774 | } else if (error) | |
775 | goto bp_err; | |
776 | } else { | |
777 | start_blk = 0; | |
778 | ASSERT(head_blk <= INT_MAX); | |
779 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
780 | &head_blk, 0)) == -1) { | |
781 | /* We hit the beginning of the log during our search */ | |
782 | start_blk = log_bbnum - num_scan_bblks + head_blk; | |
783 | new_blk = log_bbnum; | |
784 | ASSERT(start_blk <= INT_MAX && | |
785 | (xfs_daddr_t) log_bbnum-start_blk >= 0); | |
786 | ASSERT(head_blk <= INT_MAX); | |
787 | if ((error = xlog_find_verify_log_record(log, | |
788 | start_blk, &new_blk, | |
789 | (int)head_blk)) == -1) { | |
790 | error = XFS_ERROR(EIO); | |
791 | goto bp_err; | |
792 | } else if (error) | |
793 | goto bp_err; | |
794 | if (new_blk != log_bbnum) | |
795 | head_blk = new_blk; | |
796 | } else if (error) | |
797 | goto bp_err; | |
798 | } | |
799 | ||
800 | xlog_put_bp(bp); | |
801 | if (head_blk == log_bbnum) | |
802 | *return_head_blk = 0; | |
803 | else | |
804 | *return_head_blk = head_blk; | |
805 | /* | |
806 | * When returning here, we have a good block number. Bad block | |
807 | * means that during a previous crash, we didn't have a clean break | |
808 | * from cycle number N to cycle number N-1. In this case, we need | |
809 | * to find the first block with cycle number N-1. | |
810 | */ | |
811 | return 0; | |
812 | ||
813 | bp_err: | |
814 | xlog_put_bp(bp); | |
815 | ||
816 | if (error) | |
817 | xlog_warn("XFS: failed to find log head"); | |
818 | return error; | |
819 | } | |
820 | ||
821 | /* | |
822 | * Find the sync block number or the tail of the log. | |
823 | * | |
824 | * This will be the block number of the last record to have its | |
825 | * associated buffers synced to disk. Every log record header has | |
826 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy | |
827 | * to get a sync block number. The only concern is to figure out which | |
828 | * log record header to believe. | |
829 | * | |
830 | * The following algorithm uses the log record header with the largest | |
831 | * lsn. The entire log record does not need to be valid. We only care | |
832 | * that the header is valid. | |
833 | * | |
834 | * We could speed up search by using current head_blk buffer, but it is not | |
835 | * available. | |
836 | */ | |
5d77c0dc | 837 | STATIC int |
1da177e4 LT |
838 | xlog_find_tail( |
839 | xlog_t *log, | |
840 | xfs_daddr_t *head_blk, | |
65be6054 | 841 | xfs_daddr_t *tail_blk) |
1da177e4 LT |
842 | { |
843 | xlog_rec_header_t *rhead; | |
844 | xlog_op_header_t *op_head; | |
845 | xfs_caddr_t offset = NULL; | |
846 | xfs_buf_t *bp; | |
847 | int error, i, found; | |
848 | xfs_daddr_t umount_data_blk; | |
849 | xfs_daddr_t after_umount_blk; | |
850 | xfs_lsn_t tail_lsn; | |
851 | int hblks; | |
852 | ||
853 | found = 0; | |
854 | ||
855 | /* | |
856 | * Find previous log record | |
857 | */ | |
858 | if ((error = xlog_find_head(log, head_blk))) | |
859 | return error; | |
860 | ||
861 | bp = xlog_get_bp(log, 1); | |
862 | if (!bp) | |
863 | return ENOMEM; | |
864 | if (*head_blk == 0) { /* special case */ | |
076e6acb CH |
865 | error = xlog_bread(log, 0, 1, bp, &offset); |
866 | if (error) | |
9db127ed | 867 | goto done; |
076e6acb | 868 | |
03bea6fe | 869 | if (xlog_get_cycle(offset) == 0) { |
1da177e4 LT |
870 | *tail_blk = 0; |
871 | /* leave all other log inited values alone */ | |
9db127ed | 872 | goto done; |
1da177e4 LT |
873 | } |
874 | } | |
875 | ||
876 | /* | |
877 | * Search backwards looking for log record header block | |
878 | */ | |
879 | ASSERT(*head_blk < INT_MAX); | |
880 | for (i = (int)(*head_blk) - 1; i >= 0; i--) { | |
076e6acb CH |
881 | error = xlog_bread(log, i, 1, bp, &offset); |
882 | if (error) | |
9db127ed | 883 | goto done; |
076e6acb | 884 | |
b53e675d | 885 | if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) { |
1da177e4 LT |
886 | found = 1; |
887 | break; | |
888 | } | |
889 | } | |
890 | /* | |
891 | * If we haven't found the log record header block, start looking | |
892 | * again from the end of the physical log. XXXmiken: There should be | |
893 | * a check here to make sure we didn't search more than N blocks in | |
894 | * the previous code. | |
895 | */ | |
896 | if (!found) { | |
897 | for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) { | |
076e6acb CH |
898 | error = xlog_bread(log, i, 1, bp, &offset); |
899 | if (error) | |
9db127ed | 900 | goto done; |
076e6acb | 901 | |
1da177e4 | 902 | if (XLOG_HEADER_MAGIC_NUM == |
b53e675d | 903 | be32_to_cpu(*(__be32 *)offset)) { |
1da177e4 LT |
904 | found = 2; |
905 | break; | |
906 | } | |
907 | } | |
908 | } | |
909 | if (!found) { | |
910 | xlog_warn("XFS: xlog_find_tail: couldn't find sync record"); | |
911 | ASSERT(0); | |
912 | return XFS_ERROR(EIO); | |
913 | } | |
914 | ||
915 | /* find blk_no of tail of log */ | |
916 | rhead = (xlog_rec_header_t *)offset; | |
b53e675d | 917 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); |
1da177e4 LT |
918 | |
919 | /* | |
920 | * Reset log values according to the state of the log when we | |
921 | * crashed. In the case where head_blk == 0, we bump curr_cycle | |
922 | * one because the next write starts a new cycle rather than | |
923 | * continuing the cycle of the last good log record. At this | |
924 | * point we have guaranteed that all partial log records have been | |
925 | * accounted for. Therefore, we know that the last good log record | |
926 | * written was complete and ended exactly on the end boundary | |
927 | * of the physical log. | |
928 | */ | |
929 | log->l_prev_block = i; | |
930 | log->l_curr_block = (int)*head_blk; | |
b53e675d | 931 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); |
1da177e4 LT |
932 | if (found == 2) |
933 | log->l_curr_cycle++; | |
b53e675d CH |
934 | log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn); |
935 | log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn); | |
1da177e4 LT |
936 | log->l_grant_reserve_cycle = log->l_curr_cycle; |
937 | log->l_grant_reserve_bytes = BBTOB(log->l_curr_block); | |
938 | log->l_grant_write_cycle = log->l_curr_cycle; | |
939 | log->l_grant_write_bytes = BBTOB(log->l_curr_block); | |
940 | ||
941 | /* | |
942 | * Look for unmount record. If we find it, then we know there | |
943 | * was a clean unmount. Since 'i' could be the last block in | |
944 | * the physical log, we convert to a log block before comparing | |
945 | * to the head_blk. | |
946 | * | |
947 | * Save the current tail lsn to use to pass to | |
948 | * xlog_clear_stale_blocks() below. We won't want to clear the | |
949 | * unmount record if there is one, so we pass the lsn of the | |
950 | * unmount record rather than the block after it. | |
951 | */ | |
62118709 | 952 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d CH |
953 | int h_size = be32_to_cpu(rhead->h_size); |
954 | int h_version = be32_to_cpu(rhead->h_version); | |
1da177e4 LT |
955 | |
956 | if ((h_version & XLOG_VERSION_2) && | |
957 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
958 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
959 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
960 | hblks++; | |
961 | } else { | |
962 | hblks = 1; | |
963 | } | |
964 | } else { | |
965 | hblks = 1; | |
966 | } | |
967 | after_umount_blk = (i + hblks + (int) | |
b53e675d | 968 | BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize; |
1da177e4 LT |
969 | tail_lsn = log->l_tail_lsn; |
970 | if (*head_blk == after_umount_blk && | |
b53e675d | 971 | be32_to_cpu(rhead->h_num_logops) == 1) { |
1da177e4 | 972 | umount_data_blk = (i + hblks) % log->l_logBBsize; |
076e6acb CH |
973 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); |
974 | if (error) | |
9db127ed | 975 | goto done; |
076e6acb | 976 | |
1da177e4 LT |
977 | op_head = (xlog_op_header_t *)offset; |
978 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { | |
979 | /* | |
980 | * Set tail and last sync so that newly written | |
981 | * log records will point recovery to after the | |
982 | * current unmount record. | |
983 | */ | |
03bea6fe CH |
984 | log->l_tail_lsn = |
985 | xlog_assign_lsn(log->l_curr_cycle, | |
986 | after_umount_blk); | |
987 | log->l_last_sync_lsn = | |
988 | xlog_assign_lsn(log->l_curr_cycle, | |
989 | after_umount_blk); | |
1da177e4 | 990 | *tail_blk = after_umount_blk; |
92821e2b DC |
991 | |
992 | /* | |
993 | * Note that the unmount was clean. If the unmount | |
994 | * was not clean, we need to know this to rebuild the | |
995 | * superblock counters from the perag headers if we | |
996 | * have a filesystem using non-persistent counters. | |
997 | */ | |
998 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; | |
1da177e4 LT |
999 | } |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * Make sure that there are no blocks in front of the head | |
1004 | * with the same cycle number as the head. This can happen | |
1005 | * because we allow multiple outstanding log writes concurrently, | |
1006 | * and the later writes might make it out before earlier ones. | |
1007 | * | |
1008 | * We use the lsn from before modifying it so that we'll never | |
1009 | * overwrite the unmount record after a clean unmount. | |
1010 | * | |
1011 | * Do this only if we are going to recover the filesystem | |
1012 | * | |
1013 | * NOTE: This used to say "if (!readonly)" | |
1014 | * However on Linux, we can & do recover a read-only filesystem. | |
1015 | * We only skip recovery if NORECOVERY is specified on mount, | |
1016 | * in which case we would not be here. | |
1017 | * | |
1018 | * But... if the -device- itself is readonly, just skip this. | |
1019 | * We can't recover this device anyway, so it won't matter. | |
1020 | */ | |
9db127ed | 1021 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) |
1da177e4 | 1022 | error = xlog_clear_stale_blocks(log, tail_lsn); |
1da177e4 | 1023 | |
9db127ed | 1024 | done: |
1da177e4 LT |
1025 | xlog_put_bp(bp); |
1026 | ||
1027 | if (error) | |
1028 | xlog_warn("XFS: failed to locate log tail"); | |
1029 | return error; | |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * Is the log zeroed at all? | |
1034 | * | |
1035 | * The last binary search should be changed to perform an X block read | |
1036 | * once X becomes small enough. You can then search linearly through | |
1037 | * the X blocks. This will cut down on the number of reads we need to do. | |
1038 | * | |
1039 | * If the log is partially zeroed, this routine will pass back the blkno | |
1040 | * of the first block with cycle number 0. It won't have a complete LR | |
1041 | * preceding it. | |
1042 | * | |
1043 | * Return: | |
1044 | * 0 => the log is completely written to | |
1045 | * -1 => use *blk_no as the first block of the log | |
1046 | * >0 => error has occurred | |
1047 | */ | |
a8272ce0 | 1048 | STATIC int |
1da177e4 LT |
1049 | xlog_find_zeroed( |
1050 | xlog_t *log, | |
1051 | xfs_daddr_t *blk_no) | |
1052 | { | |
1053 | xfs_buf_t *bp; | |
1054 | xfs_caddr_t offset; | |
1055 | uint first_cycle, last_cycle; | |
1056 | xfs_daddr_t new_blk, last_blk, start_blk; | |
1057 | xfs_daddr_t num_scan_bblks; | |
1058 | int error, log_bbnum = log->l_logBBsize; | |
1059 | ||
6fdf8ccc NS |
1060 | *blk_no = 0; |
1061 | ||
1da177e4 LT |
1062 | /* check totally zeroed log */ |
1063 | bp = xlog_get_bp(log, 1); | |
1064 | if (!bp) | |
1065 | return ENOMEM; | |
076e6acb CH |
1066 | error = xlog_bread(log, 0, 1, bp, &offset); |
1067 | if (error) | |
1da177e4 | 1068 | goto bp_err; |
076e6acb | 1069 | |
03bea6fe | 1070 | first_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1071 | if (first_cycle == 0) { /* completely zeroed log */ |
1072 | *blk_no = 0; | |
1073 | xlog_put_bp(bp); | |
1074 | return -1; | |
1075 | } | |
1076 | ||
1077 | /* check partially zeroed log */ | |
076e6acb CH |
1078 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
1079 | if (error) | |
1da177e4 | 1080 | goto bp_err; |
076e6acb | 1081 | |
03bea6fe | 1082 | last_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1083 | if (last_cycle != 0) { /* log completely written to */ |
1084 | xlog_put_bp(bp); | |
1085 | return 0; | |
1086 | } else if (first_cycle != 1) { | |
1087 | /* | |
1088 | * If the cycle of the last block is zero, the cycle of | |
1089 | * the first block must be 1. If it's not, maybe we're | |
1090 | * not looking at a log... Bail out. | |
1091 | */ | |
1092 | xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)"); | |
1093 | return XFS_ERROR(EINVAL); | |
1094 | } | |
1095 | ||
1096 | /* we have a partially zeroed log */ | |
1097 | last_blk = log_bbnum-1; | |
1098 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) | |
1099 | goto bp_err; | |
1100 | ||
1101 | /* | |
1102 | * Validate the answer. Because there is no way to guarantee that | |
1103 | * the entire log is made up of log records which are the same size, | |
1104 | * we scan over the defined maximum blocks. At this point, the maximum | |
1105 | * is not chosen to mean anything special. XXXmiken | |
1106 | */ | |
1107 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
1108 | ASSERT(num_scan_bblks <= INT_MAX); | |
1109 | ||
1110 | if (last_blk < num_scan_bblks) | |
1111 | num_scan_bblks = last_blk; | |
1112 | start_blk = last_blk - num_scan_bblks; | |
1113 | ||
1114 | /* | |
1115 | * We search for any instances of cycle number 0 that occur before | |
1116 | * our current estimate of the head. What we're trying to detect is | |
1117 | * 1 ... | 0 | 1 | 0... | |
1118 | * ^ binary search ends here | |
1119 | */ | |
1120 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
1121 | (int)num_scan_bblks, 0, &new_blk))) | |
1122 | goto bp_err; | |
1123 | if (new_blk != -1) | |
1124 | last_blk = new_blk; | |
1125 | ||
1126 | /* | |
1127 | * Potentially backup over partial log record write. We don't need | |
1128 | * to search the end of the log because we know it is zero. | |
1129 | */ | |
1130 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
1131 | &last_blk, 0)) == -1) { | |
1132 | error = XFS_ERROR(EIO); | |
1133 | goto bp_err; | |
1134 | } else if (error) | |
1135 | goto bp_err; | |
1136 | ||
1137 | *blk_no = last_blk; | |
1138 | bp_err: | |
1139 | xlog_put_bp(bp); | |
1140 | if (error) | |
1141 | return error; | |
1142 | return -1; | |
1143 | } | |
1144 | ||
1145 | /* | |
1146 | * These are simple subroutines used by xlog_clear_stale_blocks() below | |
1147 | * to initialize a buffer full of empty log record headers and write | |
1148 | * them into the log. | |
1149 | */ | |
1150 | STATIC void | |
1151 | xlog_add_record( | |
1152 | xlog_t *log, | |
1153 | xfs_caddr_t buf, | |
1154 | int cycle, | |
1155 | int block, | |
1156 | int tail_cycle, | |
1157 | int tail_block) | |
1158 | { | |
1159 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; | |
1160 | ||
1161 | memset(buf, 0, BBSIZE); | |
b53e675d CH |
1162 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1163 | recp->h_cycle = cpu_to_be32(cycle); | |
1164 | recp->h_version = cpu_to_be32( | |
62118709 | 1165 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d CH |
1166 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
1167 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); | |
1168 | recp->h_fmt = cpu_to_be32(XLOG_FMT); | |
1da177e4 LT |
1169 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1170 | } | |
1171 | ||
1172 | STATIC int | |
1173 | xlog_write_log_records( | |
1174 | xlog_t *log, | |
1175 | int cycle, | |
1176 | int start_block, | |
1177 | int blocks, | |
1178 | int tail_cycle, | |
1179 | int tail_block) | |
1180 | { | |
1181 | xfs_caddr_t offset; | |
1182 | xfs_buf_t *bp; | |
1183 | int balign, ealign; | |
5c17f533 | 1184 | int sectbb = xlog_sectbb(log); |
1da177e4 LT |
1185 | int end_block = start_block + blocks; |
1186 | int bufblks; | |
1187 | int error = 0; | |
1188 | int i, j = 0; | |
1189 | ||
6881a229 AE |
1190 | /* |
1191 | * Greedily allocate a buffer big enough to handle the full | |
1192 | * range of basic blocks to be written. If that fails, try | |
1193 | * a smaller size. We need to be able to write at least a | |
1194 | * log sector, or we're out of luck. | |
1195 | */ | |
1da177e4 LT |
1196 | bufblks = 1 << ffs(blocks); |
1197 | while (!(bp = xlog_get_bp(log, bufblks))) { | |
1198 | bufblks >>= 1; | |
6881a229 | 1199 | if (bufblks < xlog_sectbb(log)) |
1da177e4 LT |
1200 | return ENOMEM; |
1201 | } | |
1202 | ||
1203 | /* We may need to do a read at the start to fill in part of | |
1204 | * the buffer in the starting sector not covered by the first | |
1205 | * write below. | |
1206 | */ | |
5c17f533 | 1207 | balign = round_down(start_block, sectbb); |
1da177e4 | 1208 | if (balign != start_block) { |
076e6acb CH |
1209 | error = xlog_bread_noalign(log, start_block, 1, bp); |
1210 | if (error) | |
1211 | goto out_put_bp; | |
1212 | ||
1da177e4 LT |
1213 | j = start_block - balign; |
1214 | } | |
1215 | ||
1216 | for (i = start_block; i < end_block; i += bufblks) { | |
1217 | int bcount, endcount; | |
1218 | ||
1219 | bcount = min(bufblks, end_block - start_block); | |
1220 | endcount = bcount - j; | |
1221 | ||
1222 | /* We may need to do a read at the end to fill in part of | |
1223 | * the buffer in the final sector not covered by the write. | |
1224 | * If this is the same sector as the above read, skip it. | |
1225 | */ | |
5c17f533 | 1226 | ealign = round_down(end_block, sectbb); |
1da177e4 LT |
1227 | if (j == 0 && (start_block + endcount > ealign)) { |
1228 | offset = XFS_BUF_PTR(bp); | |
1229 | balign = BBTOB(ealign - start_block); | |
234f56ac DC |
1230 | error = XFS_BUF_SET_PTR(bp, offset + balign, |
1231 | BBTOB(sectbb)); | |
076e6acb CH |
1232 | if (error) |
1233 | break; | |
1234 | ||
1235 | error = xlog_bread_noalign(log, ealign, sectbb, bp); | |
1236 | if (error) | |
1237 | break; | |
1238 | ||
1239 | error = XFS_BUF_SET_PTR(bp, offset, bufblks); | |
234f56ac | 1240 | if (error) |
1da177e4 | 1241 | break; |
1da177e4 LT |
1242 | } |
1243 | ||
1244 | offset = xlog_align(log, start_block, endcount, bp); | |
1245 | for (; j < endcount; j++) { | |
1246 | xlog_add_record(log, offset, cycle, i+j, | |
1247 | tail_cycle, tail_block); | |
1248 | offset += BBSIZE; | |
1249 | } | |
1250 | error = xlog_bwrite(log, start_block, endcount, bp); | |
1251 | if (error) | |
1252 | break; | |
1253 | start_block += endcount; | |
1254 | j = 0; | |
1255 | } | |
076e6acb CH |
1256 | |
1257 | out_put_bp: | |
1da177e4 LT |
1258 | xlog_put_bp(bp); |
1259 | return error; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * This routine is called to blow away any incomplete log writes out | |
1264 | * in front of the log head. We do this so that we won't become confused | |
1265 | * if we come up, write only a little bit more, and then crash again. | |
1266 | * If we leave the partial log records out there, this situation could | |
1267 | * cause us to think those partial writes are valid blocks since they | |
1268 | * have the current cycle number. We get rid of them by overwriting them | |
1269 | * with empty log records with the old cycle number rather than the | |
1270 | * current one. | |
1271 | * | |
1272 | * The tail lsn is passed in rather than taken from | |
1273 | * the log so that we will not write over the unmount record after a | |
1274 | * clean unmount in a 512 block log. Doing so would leave the log without | |
1275 | * any valid log records in it until a new one was written. If we crashed | |
1276 | * during that time we would not be able to recover. | |
1277 | */ | |
1278 | STATIC int | |
1279 | xlog_clear_stale_blocks( | |
1280 | xlog_t *log, | |
1281 | xfs_lsn_t tail_lsn) | |
1282 | { | |
1283 | int tail_cycle, head_cycle; | |
1284 | int tail_block, head_block; | |
1285 | int tail_distance, max_distance; | |
1286 | int distance; | |
1287 | int error; | |
1288 | ||
1289 | tail_cycle = CYCLE_LSN(tail_lsn); | |
1290 | tail_block = BLOCK_LSN(tail_lsn); | |
1291 | head_cycle = log->l_curr_cycle; | |
1292 | head_block = log->l_curr_block; | |
1293 | ||
1294 | /* | |
1295 | * Figure out the distance between the new head of the log | |
1296 | * and the tail. We want to write over any blocks beyond the | |
1297 | * head that we may have written just before the crash, but | |
1298 | * we don't want to overwrite the tail of the log. | |
1299 | */ | |
1300 | if (head_cycle == tail_cycle) { | |
1301 | /* | |
1302 | * The tail is behind the head in the physical log, | |
1303 | * so the distance from the head to the tail is the | |
1304 | * distance from the head to the end of the log plus | |
1305 | * the distance from the beginning of the log to the | |
1306 | * tail. | |
1307 | */ | |
1308 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { | |
1309 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", | |
1310 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1311 | return XFS_ERROR(EFSCORRUPTED); | |
1312 | } | |
1313 | tail_distance = tail_block + (log->l_logBBsize - head_block); | |
1314 | } else { | |
1315 | /* | |
1316 | * The head is behind the tail in the physical log, | |
1317 | * so the distance from the head to the tail is just | |
1318 | * the tail block minus the head block. | |
1319 | */ | |
1320 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ | |
1321 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", | |
1322 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1323 | return XFS_ERROR(EFSCORRUPTED); | |
1324 | } | |
1325 | tail_distance = tail_block - head_block; | |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * If the head is right up against the tail, we can't clear | |
1330 | * anything. | |
1331 | */ | |
1332 | if (tail_distance <= 0) { | |
1333 | ASSERT(tail_distance == 0); | |
1334 | return 0; | |
1335 | } | |
1336 | ||
1337 | max_distance = XLOG_TOTAL_REC_SHIFT(log); | |
1338 | /* | |
1339 | * Take the smaller of the maximum amount of outstanding I/O | |
1340 | * we could have and the distance to the tail to clear out. | |
1341 | * We take the smaller so that we don't overwrite the tail and | |
1342 | * we don't waste all day writing from the head to the tail | |
1343 | * for no reason. | |
1344 | */ | |
1345 | max_distance = MIN(max_distance, tail_distance); | |
1346 | ||
1347 | if ((head_block + max_distance) <= log->l_logBBsize) { | |
1348 | /* | |
1349 | * We can stomp all the blocks we need to without | |
1350 | * wrapping around the end of the log. Just do it | |
1351 | * in a single write. Use the cycle number of the | |
1352 | * current cycle minus one so that the log will look like: | |
1353 | * n ... | n - 1 ... | |
1354 | */ | |
1355 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1356 | head_block, max_distance, tail_cycle, | |
1357 | tail_block); | |
1358 | if (error) | |
1359 | return error; | |
1360 | } else { | |
1361 | /* | |
1362 | * We need to wrap around the end of the physical log in | |
1363 | * order to clear all the blocks. Do it in two separate | |
1364 | * I/Os. The first write should be from the head to the | |
1365 | * end of the physical log, and it should use the current | |
1366 | * cycle number minus one just like above. | |
1367 | */ | |
1368 | distance = log->l_logBBsize - head_block; | |
1369 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1370 | head_block, distance, tail_cycle, | |
1371 | tail_block); | |
1372 | ||
1373 | if (error) | |
1374 | return error; | |
1375 | ||
1376 | /* | |
1377 | * Now write the blocks at the start of the physical log. | |
1378 | * This writes the remainder of the blocks we want to clear. | |
1379 | * It uses the current cycle number since we're now on the | |
1380 | * same cycle as the head so that we get: | |
1381 | * n ... n ... | n - 1 ... | |
1382 | * ^^^^^ blocks we're writing | |
1383 | */ | |
1384 | distance = max_distance - (log->l_logBBsize - head_block); | |
1385 | error = xlog_write_log_records(log, head_cycle, 0, distance, | |
1386 | tail_cycle, tail_block); | |
1387 | if (error) | |
1388 | return error; | |
1389 | } | |
1390 | ||
1391 | return 0; | |
1392 | } | |
1393 | ||
1394 | /****************************************************************************** | |
1395 | * | |
1396 | * Log recover routines | |
1397 | * | |
1398 | ****************************************************************************** | |
1399 | */ | |
1400 | ||
1401 | STATIC xlog_recover_t * | |
1402 | xlog_recover_find_tid( | |
f0a76953 | 1403 | struct hlist_head *head, |
1da177e4 LT |
1404 | xlog_tid_t tid) |
1405 | { | |
f0a76953 DC |
1406 | xlog_recover_t *trans; |
1407 | struct hlist_node *n; | |
1da177e4 | 1408 | |
f0a76953 DC |
1409 | hlist_for_each_entry(trans, n, head, r_list) { |
1410 | if (trans->r_log_tid == tid) | |
1411 | return trans; | |
1da177e4 | 1412 | } |
f0a76953 | 1413 | return NULL; |
1da177e4 LT |
1414 | } |
1415 | ||
1416 | STATIC void | |
f0a76953 DC |
1417 | xlog_recover_new_tid( |
1418 | struct hlist_head *head, | |
1419 | xlog_tid_t tid, | |
1420 | xfs_lsn_t lsn) | |
1da177e4 | 1421 | { |
f0a76953 DC |
1422 | xlog_recover_t *trans; |
1423 | ||
1424 | trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP); | |
1425 | trans->r_log_tid = tid; | |
1426 | trans->r_lsn = lsn; | |
1427 | INIT_LIST_HEAD(&trans->r_itemq); | |
1428 | ||
1429 | INIT_HLIST_NODE(&trans->r_list); | |
1430 | hlist_add_head(&trans->r_list, head); | |
1da177e4 LT |
1431 | } |
1432 | ||
1433 | STATIC void | |
1434 | xlog_recover_add_item( | |
f0a76953 | 1435 | struct list_head *head) |
1da177e4 LT |
1436 | { |
1437 | xlog_recover_item_t *item; | |
1438 | ||
1439 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); | |
f0a76953 DC |
1440 | INIT_LIST_HEAD(&item->ri_list); |
1441 | list_add_tail(&item->ri_list, head); | |
1da177e4 LT |
1442 | } |
1443 | ||
1444 | STATIC int | |
1445 | xlog_recover_add_to_cont_trans( | |
9abbc539 | 1446 | struct log *log, |
1da177e4 LT |
1447 | xlog_recover_t *trans, |
1448 | xfs_caddr_t dp, | |
1449 | int len) | |
1450 | { | |
1451 | xlog_recover_item_t *item; | |
1452 | xfs_caddr_t ptr, old_ptr; | |
1453 | int old_len; | |
1454 | ||
f0a76953 | 1455 | if (list_empty(&trans->r_itemq)) { |
1da177e4 LT |
1456 | /* finish copying rest of trans header */ |
1457 | xlog_recover_add_item(&trans->r_itemq); | |
1458 | ptr = (xfs_caddr_t) &trans->r_theader + | |
1459 | sizeof(xfs_trans_header_t) - len; | |
1460 | memcpy(ptr, dp, len); /* d, s, l */ | |
1461 | return 0; | |
1462 | } | |
f0a76953 DC |
1463 | /* take the tail entry */ |
1464 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
1da177e4 LT |
1465 | |
1466 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; | |
1467 | old_len = item->ri_buf[item->ri_cnt-1].i_len; | |
1468 | ||
760dea67 | 1469 | ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u); |
1da177e4 LT |
1470 | memcpy(&ptr[old_len], dp, len); /* d, s, l */ |
1471 | item->ri_buf[item->ri_cnt-1].i_len += len; | |
1472 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; | |
9abbc539 | 1473 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); |
1da177e4 LT |
1474 | return 0; |
1475 | } | |
1476 | ||
1477 | /* | |
1478 | * The next region to add is the start of a new region. It could be | |
1479 | * a whole region or it could be the first part of a new region. Because | |
1480 | * of this, the assumption here is that the type and size fields of all | |
1481 | * format structures fit into the first 32 bits of the structure. | |
1482 | * | |
1483 | * This works because all regions must be 32 bit aligned. Therefore, we | |
1484 | * either have both fields or we have neither field. In the case we have | |
1485 | * neither field, the data part of the region is zero length. We only have | |
1486 | * a log_op_header and can throw away the header since a new one will appear | |
1487 | * later. If we have at least 4 bytes, then we can determine how many regions | |
1488 | * will appear in the current log item. | |
1489 | */ | |
1490 | STATIC int | |
1491 | xlog_recover_add_to_trans( | |
9abbc539 | 1492 | struct log *log, |
1da177e4 LT |
1493 | xlog_recover_t *trans, |
1494 | xfs_caddr_t dp, | |
1495 | int len) | |
1496 | { | |
1497 | xfs_inode_log_format_t *in_f; /* any will do */ | |
1498 | xlog_recover_item_t *item; | |
1499 | xfs_caddr_t ptr; | |
1500 | ||
1501 | if (!len) | |
1502 | return 0; | |
f0a76953 | 1503 | if (list_empty(&trans->r_itemq)) { |
5a792c45 DC |
1504 | /* we need to catch log corruptions here */ |
1505 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { | |
1506 | xlog_warn("XFS: xlog_recover_add_to_trans: " | |
1507 | "bad header magic number"); | |
1508 | ASSERT(0); | |
1509 | return XFS_ERROR(EIO); | |
1510 | } | |
1da177e4 LT |
1511 | if (len == sizeof(xfs_trans_header_t)) |
1512 | xlog_recover_add_item(&trans->r_itemq); | |
1513 | memcpy(&trans->r_theader, dp, len); /* d, s, l */ | |
1514 | return 0; | |
1515 | } | |
1516 | ||
1517 | ptr = kmem_alloc(len, KM_SLEEP); | |
1518 | memcpy(ptr, dp, len); | |
1519 | in_f = (xfs_inode_log_format_t *)ptr; | |
1520 | ||
f0a76953 DC |
1521 | /* take the tail entry */ |
1522 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
1523 | if (item->ri_total != 0 && | |
1524 | item->ri_total == item->ri_cnt) { | |
1525 | /* tail item is in use, get a new one */ | |
1da177e4 | 1526 | xlog_recover_add_item(&trans->r_itemq); |
f0a76953 DC |
1527 | item = list_entry(trans->r_itemq.prev, |
1528 | xlog_recover_item_t, ri_list); | |
1da177e4 | 1529 | } |
1da177e4 LT |
1530 | |
1531 | if (item->ri_total == 0) { /* first region to be added */ | |
e8fa6b48 CH |
1532 | if (in_f->ilf_size == 0 || |
1533 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { | |
1534 | xlog_warn( | |
1535 | "XFS: bad number of regions (%d) in inode log format", | |
1536 | in_f->ilf_size); | |
1537 | ASSERT(0); | |
1538 | return XFS_ERROR(EIO); | |
1539 | } | |
1540 | ||
1541 | item->ri_total = in_f->ilf_size; | |
1542 | item->ri_buf = | |
1543 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), | |
1544 | KM_SLEEP); | |
1da177e4 LT |
1545 | } |
1546 | ASSERT(item->ri_total > item->ri_cnt); | |
1547 | /* Description region is ri_buf[0] */ | |
1548 | item->ri_buf[item->ri_cnt].i_addr = ptr; | |
1549 | item->ri_buf[item->ri_cnt].i_len = len; | |
1550 | item->ri_cnt++; | |
9abbc539 | 1551 | trace_xfs_log_recover_item_add(log, trans, item, 0); |
1da177e4 LT |
1552 | return 0; |
1553 | } | |
1554 | ||
f0a76953 DC |
1555 | /* |
1556 | * Sort the log items in the transaction. Cancelled buffers need | |
1557 | * to be put first so they are processed before any items that might | |
1558 | * modify the buffers. If they are cancelled, then the modifications | |
1559 | * don't need to be replayed. | |
1560 | */ | |
1da177e4 LT |
1561 | STATIC int |
1562 | xlog_recover_reorder_trans( | |
9abbc539 DC |
1563 | struct log *log, |
1564 | xlog_recover_t *trans, | |
1565 | int pass) | |
1da177e4 | 1566 | { |
f0a76953 DC |
1567 | xlog_recover_item_t *item, *n; |
1568 | LIST_HEAD(sort_list); | |
1569 | ||
1570 | list_splice_init(&trans->r_itemq, &sort_list); | |
1571 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { | |
1572 | xfs_buf_log_format_t *buf_f; | |
1da177e4 | 1573 | |
f0a76953 | 1574 | buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr; |
1da177e4 | 1575 | |
f0a76953 | 1576 | switch (ITEM_TYPE(item)) { |
1da177e4 | 1577 | case XFS_LI_BUF: |
f0a76953 | 1578 | if (!(buf_f->blf_flags & XFS_BLI_CANCEL)) { |
9abbc539 DC |
1579 | trace_xfs_log_recover_item_reorder_head(log, |
1580 | trans, item, pass); | |
f0a76953 | 1581 | list_move(&item->ri_list, &trans->r_itemq); |
1da177e4 LT |
1582 | break; |
1583 | } | |
1584 | case XFS_LI_INODE: | |
1da177e4 LT |
1585 | case XFS_LI_DQUOT: |
1586 | case XFS_LI_QUOTAOFF: | |
1587 | case XFS_LI_EFD: | |
1588 | case XFS_LI_EFI: | |
9abbc539 DC |
1589 | trace_xfs_log_recover_item_reorder_tail(log, |
1590 | trans, item, pass); | |
f0a76953 | 1591 | list_move_tail(&item->ri_list, &trans->r_itemq); |
1da177e4 LT |
1592 | break; |
1593 | default: | |
1594 | xlog_warn( | |
1595 | "XFS: xlog_recover_reorder_trans: unrecognized type of log operation"); | |
1596 | ASSERT(0); | |
1597 | return XFS_ERROR(EIO); | |
1598 | } | |
f0a76953 DC |
1599 | } |
1600 | ASSERT(list_empty(&sort_list)); | |
1da177e4 LT |
1601 | return 0; |
1602 | } | |
1603 | ||
1604 | /* | |
1605 | * Build up the table of buf cancel records so that we don't replay | |
1606 | * cancelled data in the second pass. For buffer records that are | |
1607 | * not cancel records, there is nothing to do here so we just return. | |
1608 | * | |
1609 | * If we get a cancel record which is already in the table, this indicates | |
1610 | * that the buffer was cancelled multiple times. In order to ensure | |
1611 | * that during pass 2 we keep the record in the table until we reach its | |
1612 | * last occurrence in the log, we keep a reference count in the cancel | |
1613 | * record in the table to tell us how many times we expect to see this | |
1614 | * record during the second pass. | |
1615 | */ | |
1616 | STATIC void | |
1617 | xlog_recover_do_buffer_pass1( | |
1618 | xlog_t *log, | |
1619 | xfs_buf_log_format_t *buf_f) | |
1620 | { | |
1621 | xfs_buf_cancel_t *bcp; | |
1622 | xfs_buf_cancel_t *nextp; | |
1623 | xfs_buf_cancel_t *prevp; | |
1624 | xfs_buf_cancel_t **bucket; | |
1da177e4 LT |
1625 | xfs_daddr_t blkno = 0; |
1626 | uint len = 0; | |
1627 | ushort flags = 0; | |
1628 | ||
1629 | switch (buf_f->blf_type) { | |
1630 | case XFS_LI_BUF: | |
1631 | blkno = buf_f->blf_blkno; | |
1632 | len = buf_f->blf_len; | |
1633 | flags = buf_f->blf_flags; | |
1634 | break; | |
1da177e4 LT |
1635 | } |
1636 | ||
1637 | /* | |
1638 | * If this isn't a cancel buffer item, then just return. | |
1639 | */ | |
9abbc539 DC |
1640 | if (!(flags & XFS_BLI_CANCEL)) { |
1641 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); | |
1da177e4 | 1642 | return; |
9abbc539 | 1643 | } |
1da177e4 LT |
1644 | |
1645 | /* | |
1646 | * Insert an xfs_buf_cancel record into the hash table of | |
1647 | * them. If there is already an identical record, bump | |
1648 | * its reference count. | |
1649 | */ | |
1650 | bucket = &log->l_buf_cancel_table[(__uint64_t)blkno % | |
1651 | XLOG_BC_TABLE_SIZE]; | |
1652 | /* | |
1653 | * If the hash bucket is empty then just insert a new record into | |
1654 | * the bucket. | |
1655 | */ | |
1656 | if (*bucket == NULL) { | |
1657 | bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t), | |
1658 | KM_SLEEP); | |
1659 | bcp->bc_blkno = blkno; | |
1660 | bcp->bc_len = len; | |
1661 | bcp->bc_refcount = 1; | |
1662 | bcp->bc_next = NULL; | |
1663 | *bucket = bcp; | |
1664 | return; | |
1665 | } | |
1666 | ||
1667 | /* | |
1668 | * The hash bucket is not empty, so search for duplicates of our | |
1669 | * record. If we find one them just bump its refcount. If not | |
1670 | * then add us at the end of the list. | |
1671 | */ | |
1672 | prevp = NULL; | |
1673 | nextp = *bucket; | |
1674 | while (nextp != NULL) { | |
1675 | if (nextp->bc_blkno == blkno && nextp->bc_len == len) { | |
1676 | nextp->bc_refcount++; | |
9abbc539 | 1677 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
1da177e4 LT |
1678 | return; |
1679 | } | |
1680 | prevp = nextp; | |
1681 | nextp = nextp->bc_next; | |
1682 | } | |
1683 | ASSERT(prevp != NULL); | |
1684 | bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t), | |
1685 | KM_SLEEP); | |
1686 | bcp->bc_blkno = blkno; | |
1687 | bcp->bc_len = len; | |
1688 | bcp->bc_refcount = 1; | |
1689 | bcp->bc_next = NULL; | |
1690 | prevp->bc_next = bcp; | |
9abbc539 | 1691 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
1da177e4 LT |
1692 | } |
1693 | ||
1694 | /* | |
1695 | * Check to see whether the buffer being recovered has a corresponding | |
1696 | * entry in the buffer cancel record table. If it does then return 1 | |
1697 | * so that it will be cancelled, otherwise return 0. If the buffer is | |
1698 | * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement | |
1699 | * the refcount on the entry in the table and remove it from the table | |
1700 | * if this is the last reference. | |
1701 | * | |
1702 | * We remove the cancel record from the table when we encounter its | |
1703 | * last occurrence in the log so that if the same buffer is re-used | |
1704 | * again after its last cancellation we actually replay the changes | |
1705 | * made at that point. | |
1706 | */ | |
1707 | STATIC int | |
1708 | xlog_check_buffer_cancelled( | |
1709 | xlog_t *log, | |
1710 | xfs_daddr_t blkno, | |
1711 | uint len, | |
1712 | ushort flags) | |
1713 | { | |
1714 | xfs_buf_cancel_t *bcp; | |
1715 | xfs_buf_cancel_t *prevp; | |
1716 | xfs_buf_cancel_t **bucket; | |
1717 | ||
1718 | if (log->l_buf_cancel_table == NULL) { | |
1719 | /* | |
1720 | * There is nothing in the table built in pass one, | |
1721 | * so this buffer must not be cancelled. | |
1722 | */ | |
1723 | ASSERT(!(flags & XFS_BLI_CANCEL)); | |
1724 | return 0; | |
1725 | } | |
1726 | ||
1727 | bucket = &log->l_buf_cancel_table[(__uint64_t)blkno % | |
1728 | XLOG_BC_TABLE_SIZE]; | |
1729 | bcp = *bucket; | |
1730 | if (bcp == NULL) { | |
1731 | /* | |
1732 | * There is no corresponding entry in the table built | |
1733 | * in pass one, so this buffer has not been cancelled. | |
1734 | */ | |
1735 | ASSERT(!(flags & XFS_BLI_CANCEL)); | |
1736 | return 0; | |
1737 | } | |
1738 | ||
1739 | /* | |
1740 | * Search for an entry in the buffer cancel table that | |
1741 | * matches our buffer. | |
1742 | */ | |
1743 | prevp = NULL; | |
1744 | while (bcp != NULL) { | |
1745 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) { | |
1746 | /* | |
1747 | * We've go a match, so return 1 so that the | |
1748 | * recovery of this buffer is cancelled. | |
1749 | * If this buffer is actually a buffer cancel | |
1750 | * log item, then decrement the refcount on the | |
1751 | * one in the table and remove it if this is the | |
1752 | * last reference. | |
1753 | */ | |
1754 | if (flags & XFS_BLI_CANCEL) { | |
1755 | bcp->bc_refcount--; | |
1756 | if (bcp->bc_refcount == 0) { | |
1757 | if (prevp == NULL) { | |
1758 | *bucket = bcp->bc_next; | |
1759 | } else { | |
1760 | prevp->bc_next = bcp->bc_next; | |
1761 | } | |
f0e2d93c | 1762 | kmem_free(bcp); |
1da177e4 LT |
1763 | } |
1764 | } | |
1765 | return 1; | |
1766 | } | |
1767 | prevp = bcp; | |
1768 | bcp = bcp->bc_next; | |
1769 | } | |
1770 | /* | |
1771 | * We didn't find a corresponding entry in the table, so | |
1772 | * return 0 so that the buffer is NOT cancelled. | |
1773 | */ | |
1774 | ASSERT(!(flags & XFS_BLI_CANCEL)); | |
1775 | return 0; | |
1776 | } | |
1777 | ||
1778 | STATIC int | |
1779 | xlog_recover_do_buffer_pass2( | |
1780 | xlog_t *log, | |
1781 | xfs_buf_log_format_t *buf_f) | |
1782 | { | |
1da177e4 LT |
1783 | xfs_daddr_t blkno = 0; |
1784 | ushort flags = 0; | |
1785 | uint len = 0; | |
1786 | ||
1787 | switch (buf_f->blf_type) { | |
1788 | case XFS_LI_BUF: | |
1789 | blkno = buf_f->blf_blkno; | |
1790 | flags = buf_f->blf_flags; | |
1791 | len = buf_f->blf_len; | |
1792 | break; | |
1da177e4 LT |
1793 | } |
1794 | ||
1795 | return xlog_check_buffer_cancelled(log, blkno, len, flags); | |
1796 | } | |
1797 | ||
1798 | /* | |
1799 | * Perform recovery for a buffer full of inodes. In these buffers, | |
1800 | * the only data which should be recovered is that which corresponds | |
1801 | * to the di_next_unlinked pointers in the on disk inode structures. | |
1802 | * The rest of the data for the inodes is always logged through the | |
1803 | * inodes themselves rather than the inode buffer and is recovered | |
1804 | * in xlog_recover_do_inode_trans(). | |
1805 | * | |
1806 | * The only time when buffers full of inodes are fully recovered is | |
1807 | * when the buffer is full of newly allocated inodes. In this case | |
1808 | * the buffer will not be marked as an inode buffer and so will be | |
1809 | * sent to xlog_recover_do_reg_buffer() below during recovery. | |
1810 | */ | |
1811 | STATIC int | |
1812 | xlog_recover_do_inode_buffer( | |
1813 | xfs_mount_t *mp, | |
1814 | xlog_recover_item_t *item, | |
1815 | xfs_buf_t *bp, | |
1816 | xfs_buf_log_format_t *buf_f) | |
1817 | { | |
1818 | int i; | |
1819 | int item_index; | |
1820 | int bit; | |
1821 | int nbits; | |
1822 | int reg_buf_offset; | |
1823 | int reg_buf_bytes; | |
1824 | int next_unlinked_offset; | |
1825 | int inodes_per_buf; | |
1826 | xfs_agino_t *logged_nextp; | |
1827 | xfs_agino_t *buffer_nextp; | |
1da177e4 LT |
1828 | unsigned int *data_map = NULL; |
1829 | unsigned int map_size = 0; | |
1830 | ||
9abbc539 DC |
1831 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
1832 | ||
1da177e4 LT |
1833 | switch (buf_f->blf_type) { |
1834 | case XFS_LI_BUF: | |
1835 | data_map = buf_f->blf_data_map; | |
1836 | map_size = buf_f->blf_map_size; | |
1837 | break; | |
1da177e4 LT |
1838 | } |
1839 | /* | |
1840 | * Set the variables corresponding to the current region to | |
1841 | * 0 so that we'll initialize them on the first pass through | |
1842 | * the loop. | |
1843 | */ | |
1844 | reg_buf_offset = 0; | |
1845 | reg_buf_bytes = 0; | |
1846 | bit = 0; | |
1847 | nbits = 0; | |
1848 | item_index = 0; | |
1849 | inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog; | |
1850 | for (i = 0; i < inodes_per_buf; i++) { | |
1851 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + | |
1852 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1853 | ||
1854 | while (next_unlinked_offset >= | |
1855 | (reg_buf_offset + reg_buf_bytes)) { | |
1856 | /* | |
1857 | * The next di_next_unlinked field is beyond | |
1858 | * the current logged region. Find the next | |
1859 | * logged region that contains or is beyond | |
1860 | * the current di_next_unlinked field. | |
1861 | */ | |
1862 | bit += nbits; | |
1863 | bit = xfs_next_bit(data_map, map_size, bit); | |
1864 | ||
1865 | /* | |
1866 | * If there are no more logged regions in the | |
1867 | * buffer, then we're done. | |
1868 | */ | |
1869 | if (bit == -1) { | |
1870 | return 0; | |
1871 | } | |
1872 | ||
1873 | nbits = xfs_contig_bits(data_map, map_size, | |
1874 | bit); | |
1875 | ASSERT(nbits > 0); | |
1876 | reg_buf_offset = bit << XFS_BLI_SHIFT; | |
1877 | reg_buf_bytes = nbits << XFS_BLI_SHIFT; | |
1878 | item_index++; | |
1879 | } | |
1880 | ||
1881 | /* | |
1882 | * If the current logged region starts after the current | |
1883 | * di_next_unlinked field, then move on to the next | |
1884 | * di_next_unlinked field. | |
1885 | */ | |
1886 | if (next_unlinked_offset < reg_buf_offset) { | |
1887 | continue; | |
1888 | } | |
1889 | ||
1890 | ASSERT(item->ri_buf[item_index].i_addr != NULL); | |
1891 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0); | |
1892 | ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp)); | |
1893 | ||
1894 | /* | |
1895 | * The current logged region contains a copy of the | |
1896 | * current di_next_unlinked field. Extract its value | |
1897 | * and copy it to the buffer copy. | |
1898 | */ | |
1899 | logged_nextp = (xfs_agino_t *) | |
1900 | ((char *)(item->ri_buf[item_index].i_addr) + | |
1901 | (next_unlinked_offset - reg_buf_offset)); | |
1902 | if (unlikely(*logged_nextp == 0)) { | |
1903 | xfs_fs_cmn_err(CE_ALERT, mp, | |
1904 | "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field", | |
1905 | item, bp); | |
1906 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", | |
1907 | XFS_ERRLEVEL_LOW, mp); | |
1908 | return XFS_ERROR(EFSCORRUPTED); | |
1909 | } | |
1910 | ||
1911 | buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp, | |
1912 | next_unlinked_offset); | |
87c199c2 | 1913 | *buffer_nextp = *logged_nextp; |
1da177e4 LT |
1914 | } |
1915 | ||
1916 | return 0; | |
1917 | } | |
1918 | ||
1919 | /* | |
1920 | * Perform a 'normal' buffer recovery. Each logged region of the | |
1921 | * buffer should be copied over the corresponding region in the | |
1922 | * given buffer. The bitmap in the buf log format structure indicates | |
1923 | * where to place the logged data. | |
1924 | */ | |
1925 | /*ARGSUSED*/ | |
1926 | STATIC void | |
1927 | xlog_recover_do_reg_buffer( | |
9abbc539 | 1928 | struct xfs_mount *mp, |
1da177e4 LT |
1929 | xlog_recover_item_t *item, |
1930 | xfs_buf_t *bp, | |
1931 | xfs_buf_log_format_t *buf_f) | |
1932 | { | |
1933 | int i; | |
1934 | int bit; | |
1935 | int nbits; | |
1da177e4 LT |
1936 | unsigned int *data_map = NULL; |
1937 | unsigned int map_size = 0; | |
1938 | int error; | |
1939 | ||
9abbc539 DC |
1940 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); |
1941 | ||
1da177e4 LT |
1942 | switch (buf_f->blf_type) { |
1943 | case XFS_LI_BUF: | |
1944 | data_map = buf_f->blf_data_map; | |
1945 | map_size = buf_f->blf_map_size; | |
1946 | break; | |
1da177e4 LT |
1947 | } |
1948 | bit = 0; | |
1949 | i = 1; /* 0 is the buf format structure */ | |
1950 | while (1) { | |
1951 | bit = xfs_next_bit(data_map, map_size, bit); | |
1952 | if (bit == -1) | |
1953 | break; | |
1954 | nbits = xfs_contig_bits(data_map, map_size, bit); | |
1955 | ASSERT(nbits > 0); | |
4b80916b | 1956 | ASSERT(item->ri_buf[i].i_addr != NULL); |
1da177e4 LT |
1957 | ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0); |
1958 | ASSERT(XFS_BUF_COUNT(bp) >= | |
1959 | ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT)); | |
1960 | ||
1961 | /* | |
1962 | * Do a sanity check if this is a dquot buffer. Just checking | |
1963 | * the first dquot in the buffer should do. XXXThis is | |
1964 | * probably a good thing to do for other buf types also. | |
1965 | */ | |
1966 | error = 0; | |
c8ad20ff NS |
1967 | if (buf_f->blf_flags & |
1968 | (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) { | |
0c5e1ce8 CH |
1969 | if (item->ri_buf[i].i_addr == NULL) { |
1970 | cmn_err(CE_ALERT, | |
1971 | "XFS: NULL dquot in %s.", __func__); | |
1972 | goto next; | |
1973 | } | |
8ec6dba2 | 1974 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { |
0c5e1ce8 CH |
1975 | cmn_err(CE_ALERT, |
1976 | "XFS: dquot too small (%d) in %s.", | |
1977 | item->ri_buf[i].i_len, __func__); | |
1978 | goto next; | |
1979 | } | |
1da177e4 LT |
1980 | error = xfs_qm_dqcheck((xfs_disk_dquot_t *) |
1981 | item->ri_buf[i].i_addr, | |
1982 | -1, 0, XFS_QMOPT_DOWARN, | |
1983 | "dquot_buf_recover"); | |
0c5e1ce8 CH |
1984 | if (error) |
1985 | goto next; | |
1da177e4 | 1986 | } |
0c5e1ce8 CH |
1987 | |
1988 | memcpy(xfs_buf_offset(bp, | |
1989 | (uint)bit << XFS_BLI_SHIFT), /* dest */ | |
1990 | item->ri_buf[i].i_addr, /* source */ | |
1991 | nbits<<XFS_BLI_SHIFT); /* length */ | |
1992 | next: | |
1da177e4 LT |
1993 | i++; |
1994 | bit += nbits; | |
1995 | } | |
1996 | ||
1997 | /* Shouldn't be any more regions */ | |
1998 | ASSERT(i == item->ri_total); | |
1999 | } | |
2000 | ||
2001 | /* | |
2002 | * Do some primitive error checking on ondisk dquot data structures. | |
2003 | */ | |
2004 | int | |
2005 | xfs_qm_dqcheck( | |
2006 | xfs_disk_dquot_t *ddq, | |
2007 | xfs_dqid_t id, | |
2008 | uint type, /* used only when IO_dorepair is true */ | |
2009 | uint flags, | |
2010 | char *str) | |
2011 | { | |
2012 | xfs_dqblk_t *d = (xfs_dqblk_t *)ddq; | |
2013 | int errs = 0; | |
2014 | ||
2015 | /* | |
2016 | * We can encounter an uninitialized dquot buffer for 2 reasons: | |
2017 | * 1. If we crash while deleting the quotainode(s), and those blks got | |
2018 | * used for user data. This is because we take the path of regular | |
2019 | * file deletion; however, the size field of quotainodes is never | |
2020 | * updated, so all the tricks that we play in itruncate_finish | |
2021 | * don't quite matter. | |
2022 | * | |
2023 | * 2. We don't play the quota buffers when there's a quotaoff logitem. | |
2024 | * But the allocation will be replayed so we'll end up with an | |
2025 | * uninitialized quota block. | |
2026 | * | |
2027 | * This is all fine; things are still consistent, and we haven't lost | |
2028 | * any quota information. Just don't complain about bad dquot blks. | |
2029 | */ | |
1149d96a | 2030 | if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) { |
1da177e4 LT |
2031 | if (flags & XFS_QMOPT_DOWARN) |
2032 | cmn_err(CE_ALERT, | |
2033 | "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x", | |
1149d96a | 2034 | str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC); |
1da177e4 LT |
2035 | errs++; |
2036 | } | |
1149d96a | 2037 | if (ddq->d_version != XFS_DQUOT_VERSION) { |
1da177e4 LT |
2038 | if (flags & XFS_QMOPT_DOWARN) |
2039 | cmn_err(CE_ALERT, | |
2040 | "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x", | |
1149d96a | 2041 | str, id, ddq->d_version, XFS_DQUOT_VERSION); |
1da177e4 LT |
2042 | errs++; |
2043 | } | |
2044 | ||
1149d96a CH |
2045 | if (ddq->d_flags != XFS_DQ_USER && |
2046 | ddq->d_flags != XFS_DQ_PROJ && | |
2047 | ddq->d_flags != XFS_DQ_GROUP) { | |
1da177e4 LT |
2048 | if (flags & XFS_QMOPT_DOWARN) |
2049 | cmn_err(CE_ALERT, | |
2050 | "%s : XFS dquot ID 0x%x, unknown flags 0x%x", | |
1149d96a | 2051 | str, id, ddq->d_flags); |
1da177e4 LT |
2052 | errs++; |
2053 | } | |
2054 | ||
1149d96a | 2055 | if (id != -1 && id != be32_to_cpu(ddq->d_id)) { |
1da177e4 LT |
2056 | if (flags & XFS_QMOPT_DOWARN) |
2057 | cmn_err(CE_ALERT, | |
2058 | "%s : ondisk-dquot 0x%p, ID mismatch: " | |
2059 | "0x%x expected, found id 0x%x", | |
1149d96a | 2060 | str, ddq, id, be32_to_cpu(ddq->d_id)); |
1da177e4 LT |
2061 | errs++; |
2062 | } | |
2063 | ||
2064 | if (!errs && ddq->d_id) { | |
1149d96a CH |
2065 | if (ddq->d_blk_softlimit && |
2066 | be64_to_cpu(ddq->d_bcount) >= | |
2067 | be64_to_cpu(ddq->d_blk_softlimit)) { | |
1da177e4 LT |
2068 | if (!ddq->d_btimer) { |
2069 | if (flags & XFS_QMOPT_DOWARN) | |
2070 | cmn_err(CE_ALERT, | |
2071 | "%s : Dquot ID 0x%x (0x%p) " | |
2072 | "BLK TIMER NOT STARTED", | |
1149d96a | 2073 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2074 | errs++; |
2075 | } | |
2076 | } | |
1149d96a CH |
2077 | if (ddq->d_ino_softlimit && |
2078 | be64_to_cpu(ddq->d_icount) >= | |
2079 | be64_to_cpu(ddq->d_ino_softlimit)) { | |
1da177e4 LT |
2080 | if (!ddq->d_itimer) { |
2081 | if (flags & XFS_QMOPT_DOWARN) | |
2082 | cmn_err(CE_ALERT, | |
2083 | "%s : Dquot ID 0x%x (0x%p) " | |
2084 | "INODE TIMER NOT STARTED", | |
1149d96a | 2085 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2086 | errs++; |
2087 | } | |
2088 | } | |
1149d96a CH |
2089 | if (ddq->d_rtb_softlimit && |
2090 | be64_to_cpu(ddq->d_rtbcount) >= | |
2091 | be64_to_cpu(ddq->d_rtb_softlimit)) { | |
1da177e4 LT |
2092 | if (!ddq->d_rtbtimer) { |
2093 | if (flags & XFS_QMOPT_DOWARN) | |
2094 | cmn_err(CE_ALERT, | |
2095 | "%s : Dquot ID 0x%x (0x%p) " | |
2096 | "RTBLK TIMER NOT STARTED", | |
1149d96a | 2097 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2098 | errs++; |
2099 | } | |
2100 | } | |
2101 | } | |
2102 | ||
2103 | if (!errs || !(flags & XFS_QMOPT_DQREPAIR)) | |
2104 | return errs; | |
2105 | ||
2106 | if (flags & XFS_QMOPT_DOWARN) | |
2107 | cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id); | |
2108 | ||
2109 | /* | |
2110 | * Typically, a repair is only requested by quotacheck. | |
2111 | */ | |
2112 | ASSERT(id != -1); | |
2113 | ASSERT(flags & XFS_QMOPT_DQREPAIR); | |
2114 | memset(d, 0, sizeof(xfs_dqblk_t)); | |
1149d96a CH |
2115 | |
2116 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
2117 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
2118 | d->dd_diskdq.d_flags = type; | |
2119 | d->dd_diskdq.d_id = cpu_to_be32(id); | |
1da177e4 LT |
2120 | |
2121 | return errs; | |
2122 | } | |
2123 | ||
2124 | /* | |
2125 | * Perform a dquot buffer recovery. | |
2126 | * Simple algorithm: if we have found a QUOTAOFF logitem of the same type | |
2127 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. | |
2128 | * Else, treat it as a regular buffer and do recovery. | |
2129 | */ | |
2130 | STATIC void | |
2131 | xlog_recover_do_dquot_buffer( | |
2132 | xfs_mount_t *mp, | |
2133 | xlog_t *log, | |
2134 | xlog_recover_item_t *item, | |
2135 | xfs_buf_t *bp, | |
2136 | xfs_buf_log_format_t *buf_f) | |
2137 | { | |
2138 | uint type; | |
2139 | ||
9abbc539 DC |
2140 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
2141 | ||
1da177e4 LT |
2142 | /* |
2143 | * Filesystems are required to send in quota flags at mount time. | |
2144 | */ | |
2145 | if (mp->m_qflags == 0) { | |
2146 | return; | |
2147 | } | |
2148 | ||
2149 | type = 0; | |
2150 | if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF) | |
2151 | type |= XFS_DQ_USER; | |
c8ad20ff NS |
2152 | if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF) |
2153 | type |= XFS_DQ_PROJ; | |
1da177e4 LT |
2154 | if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF) |
2155 | type |= XFS_DQ_GROUP; | |
2156 | /* | |
2157 | * This type of quotas was turned off, so ignore this buffer | |
2158 | */ | |
2159 | if (log->l_quotaoffs_flag & type) | |
2160 | return; | |
2161 | ||
9abbc539 | 2162 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 LT |
2163 | } |
2164 | ||
2165 | /* | |
2166 | * This routine replays a modification made to a buffer at runtime. | |
2167 | * There are actually two types of buffer, regular and inode, which | |
2168 | * are handled differently. Inode buffers are handled differently | |
2169 | * in that we only recover a specific set of data from them, namely | |
2170 | * the inode di_next_unlinked fields. This is because all other inode | |
2171 | * data is actually logged via inode records and any data we replay | |
2172 | * here which overlaps that may be stale. | |
2173 | * | |
2174 | * When meta-data buffers are freed at run time we log a buffer item | |
2175 | * with the XFS_BLI_CANCEL bit set to indicate that previous copies | |
2176 | * of the buffer in the log should not be replayed at recovery time. | |
2177 | * This is so that if the blocks covered by the buffer are reused for | |
2178 | * file data before we crash we don't end up replaying old, freed | |
2179 | * meta-data into a user's file. | |
2180 | * | |
2181 | * To handle the cancellation of buffer log items, we make two passes | |
2182 | * over the log during recovery. During the first we build a table of | |
2183 | * those buffers which have been cancelled, and during the second we | |
2184 | * only replay those buffers which do not have corresponding cancel | |
2185 | * records in the table. See xlog_recover_do_buffer_pass[1,2] above | |
2186 | * for more details on the implementation of the table of cancel records. | |
2187 | */ | |
2188 | STATIC int | |
2189 | xlog_recover_do_buffer_trans( | |
2190 | xlog_t *log, | |
2191 | xlog_recover_item_t *item, | |
2192 | int pass) | |
2193 | { | |
2194 | xfs_buf_log_format_t *buf_f; | |
1da177e4 LT |
2195 | xfs_mount_t *mp; |
2196 | xfs_buf_t *bp; | |
2197 | int error; | |
2198 | int cancel; | |
2199 | xfs_daddr_t blkno; | |
2200 | int len; | |
2201 | ushort flags; | |
6ad112bf | 2202 | uint buf_flags; |
1da177e4 LT |
2203 | |
2204 | buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr; | |
2205 | ||
2206 | if (pass == XLOG_RECOVER_PASS1) { | |
2207 | /* | |
2208 | * In this pass we're only looking for buf items | |
2209 | * with the XFS_BLI_CANCEL bit set. | |
2210 | */ | |
2211 | xlog_recover_do_buffer_pass1(log, buf_f); | |
2212 | return 0; | |
2213 | } else { | |
2214 | /* | |
2215 | * In this pass we want to recover all the buffers | |
2216 | * which have not been cancelled and are not | |
2217 | * cancellation buffers themselves. The routine | |
2218 | * we call here will tell us whether or not to | |
2219 | * continue with the replay of this buffer. | |
2220 | */ | |
2221 | cancel = xlog_recover_do_buffer_pass2(log, buf_f); | |
2222 | if (cancel) { | |
9abbc539 | 2223 | trace_xfs_log_recover_buf_cancel(log, buf_f); |
1da177e4 LT |
2224 | return 0; |
2225 | } | |
2226 | } | |
9abbc539 | 2227 | trace_xfs_log_recover_buf_recover(log, buf_f); |
1da177e4 LT |
2228 | switch (buf_f->blf_type) { |
2229 | case XFS_LI_BUF: | |
2230 | blkno = buf_f->blf_blkno; | |
2231 | len = buf_f->blf_len; | |
2232 | flags = buf_f->blf_flags; | |
2233 | break; | |
1da177e4 LT |
2234 | default: |
2235 | xfs_fs_cmn_err(CE_ALERT, log->l_mp, | |
fc1f8c1c NS |
2236 | "xfs_log_recover: unknown buffer type 0x%x, logdev %s", |
2237 | buf_f->blf_type, log->l_mp->m_logname ? | |
2238 | log->l_mp->m_logname : "internal"); | |
1da177e4 LT |
2239 | XFS_ERROR_REPORT("xlog_recover_do_buffer_trans", |
2240 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2241 | return XFS_ERROR(EFSCORRUPTED); | |
2242 | } | |
2243 | ||
2244 | mp = log->l_mp; | |
0cadda1c | 2245 | buf_flags = XBF_LOCK; |
6ad112bf | 2246 | if (!(flags & XFS_BLI_INODE_BUF)) |
0cadda1c | 2247 | buf_flags |= XBF_MAPPED; |
6ad112bf CH |
2248 | |
2249 | bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, buf_flags); | |
1da177e4 LT |
2250 | if (XFS_BUF_ISERROR(bp)) { |
2251 | xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp, | |
2252 | bp, blkno); | |
2253 | error = XFS_BUF_GETERROR(bp); | |
2254 | xfs_buf_relse(bp); | |
2255 | return error; | |
2256 | } | |
2257 | ||
2258 | error = 0; | |
2259 | if (flags & XFS_BLI_INODE_BUF) { | |
2260 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); | |
c8ad20ff NS |
2261 | } else if (flags & |
2262 | (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) { | |
1da177e4 LT |
2263 | xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); |
2264 | } else { | |
9abbc539 | 2265 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 LT |
2266 | } |
2267 | if (error) | |
2268 | return XFS_ERROR(error); | |
2269 | ||
2270 | /* | |
2271 | * Perform delayed write on the buffer. Asynchronous writes will be | |
2272 | * slower when taking into account all the buffers to be flushed. | |
2273 | * | |
2274 | * Also make sure that only inode buffers with good sizes stay in | |
2275 | * the buffer cache. The kernel moves inodes in buffers of 1 block | |
2276 | * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode | |
2277 | * buffers in the log can be a different size if the log was generated | |
2278 | * by an older kernel using unclustered inode buffers or a newer kernel | |
2279 | * running with a different inode cluster size. Regardless, if the | |
2280 | * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE) | |
2281 | * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep | |
2282 | * the buffer out of the buffer cache so that the buffer won't | |
2283 | * overlap with future reads of those inodes. | |
2284 | */ | |
2285 | if (XFS_DINODE_MAGIC == | |
b53e675d | 2286 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
1da177e4 LT |
2287 | (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize, |
2288 | (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) { | |
2289 | XFS_BUF_STALE(bp); | |
2290 | error = xfs_bwrite(mp, bp); | |
2291 | } else { | |
15ac08a8 CH |
2292 | ASSERT(bp->b_mount == NULL || bp->b_mount == mp); |
2293 | bp->b_mount = mp; | |
1da177e4 LT |
2294 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone); |
2295 | xfs_bdwrite(mp, bp); | |
2296 | } | |
2297 | ||
2298 | return (error); | |
2299 | } | |
2300 | ||
2301 | STATIC int | |
2302 | xlog_recover_do_inode_trans( | |
2303 | xlog_t *log, | |
2304 | xlog_recover_item_t *item, | |
2305 | int pass) | |
2306 | { | |
2307 | xfs_inode_log_format_t *in_f; | |
2308 | xfs_mount_t *mp; | |
2309 | xfs_buf_t *bp; | |
1da177e4 LT |
2310 | xfs_dinode_t *dip; |
2311 | xfs_ino_t ino; | |
2312 | int len; | |
2313 | xfs_caddr_t src; | |
2314 | xfs_caddr_t dest; | |
2315 | int error; | |
2316 | int attr_index; | |
2317 | uint fields; | |
347d1c01 | 2318 | xfs_icdinode_t *dicp; |
6d192a9b | 2319 | int need_free = 0; |
1da177e4 LT |
2320 | |
2321 | if (pass == XLOG_RECOVER_PASS1) { | |
2322 | return 0; | |
2323 | } | |
2324 | ||
6d192a9b TS |
2325 | if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) { |
2326 | in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr; | |
2327 | } else { | |
2328 | in_f = (xfs_inode_log_format_t *)kmem_alloc( | |
2329 | sizeof(xfs_inode_log_format_t), KM_SLEEP); | |
2330 | need_free = 1; | |
2331 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); | |
2332 | if (error) | |
2333 | goto error; | |
2334 | } | |
1da177e4 LT |
2335 | ino = in_f->ilf_ino; |
2336 | mp = log->l_mp; | |
1da177e4 LT |
2337 | |
2338 | /* | |
2339 | * Inode buffers can be freed, look out for it, | |
2340 | * and do not replay the inode. | |
2341 | */ | |
a1941895 CH |
2342 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
2343 | in_f->ilf_len, 0)) { | |
6d192a9b | 2344 | error = 0; |
9abbc539 | 2345 | trace_xfs_log_recover_inode_cancel(log, in_f); |
6d192a9b TS |
2346 | goto error; |
2347 | } | |
9abbc539 | 2348 | trace_xfs_log_recover_inode_recover(log, in_f); |
1da177e4 | 2349 | |
6ad112bf | 2350 | bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, |
0cadda1c | 2351 | XBF_LOCK); |
1da177e4 LT |
2352 | if (XFS_BUF_ISERROR(bp)) { |
2353 | xfs_ioerror_alert("xlog_recover_do..(read#2)", mp, | |
a1941895 | 2354 | bp, in_f->ilf_blkno); |
1da177e4 LT |
2355 | error = XFS_BUF_GETERROR(bp); |
2356 | xfs_buf_relse(bp); | |
6d192a9b | 2357 | goto error; |
1da177e4 LT |
2358 | } |
2359 | error = 0; | |
2360 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); | |
a1941895 | 2361 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset); |
1da177e4 LT |
2362 | |
2363 | /* | |
2364 | * Make sure the place we're flushing out to really looks | |
2365 | * like an inode! | |
2366 | */ | |
81591fe2 | 2367 | if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) { |
1da177e4 LT |
2368 | xfs_buf_relse(bp); |
2369 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2370 | "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld", | |
2371 | dip, bp, ino); | |
2372 | XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)", | |
2373 | XFS_ERRLEVEL_LOW, mp); | |
6d192a9b TS |
2374 | error = EFSCORRUPTED; |
2375 | goto error; | |
1da177e4 | 2376 | } |
347d1c01 | 2377 | dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr); |
1da177e4 LT |
2378 | if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) { |
2379 | xfs_buf_relse(bp); | |
2380 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2381 | "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld", | |
2382 | item, ino); | |
2383 | XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)", | |
2384 | XFS_ERRLEVEL_LOW, mp); | |
6d192a9b TS |
2385 | error = EFSCORRUPTED; |
2386 | goto error; | |
1da177e4 LT |
2387 | } |
2388 | ||
2389 | /* Skip replay when the on disk inode is newer than the log one */ | |
81591fe2 | 2390 | if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) { |
1da177e4 LT |
2391 | /* |
2392 | * Deal with the wrap case, DI_MAX_FLUSH is less | |
2393 | * than smaller numbers | |
2394 | */ | |
81591fe2 | 2395 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
347d1c01 | 2396 | dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
1da177e4 LT |
2397 | /* do nothing */ |
2398 | } else { | |
2399 | xfs_buf_relse(bp); | |
9abbc539 | 2400 | trace_xfs_log_recover_inode_skip(log, in_f); |
6d192a9b TS |
2401 | error = 0; |
2402 | goto error; | |
1da177e4 LT |
2403 | } |
2404 | } | |
2405 | /* Take the opportunity to reset the flush iteration count */ | |
2406 | dicp->di_flushiter = 0; | |
2407 | ||
2408 | if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) { | |
2409 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && | |
2410 | (dicp->di_format != XFS_DINODE_FMT_BTREE)) { | |
2411 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)", | |
2412 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2413 | xfs_buf_relse(bp); | |
2414 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2415 | "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2416 | item, dip, bp, ino); | |
6d192a9b TS |
2417 | error = EFSCORRUPTED; |
2418 | goto error; | |
1da177e4 LT |
2419 | } |
2420 | } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) { | |
2421 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && | |
2422 | (dicp->di_format != XFS_DINODE_FMT_BTREE) && | |
2423 | (dicp->di_format != XFS_DINODE_FMT_LOCAL)) { | |
2424 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)", | |
2425 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2426 | xfs_buf_relse(bp); | |
2427 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2428 | "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2429 | item, dip, bp, ino); | |
6d192a9b TS |
2430 | error = EFSCORRUPTED; |
2431 | goto error; | |
1da177e4 LT |
2432 | } |
2433 | } | |
2434 | if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){ | |
2435 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)", | |
2436 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2437 | xfs_buf_relse(bp); | |
2438 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2439 | "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", | |
2440 | item, dip, bp, ino, | |
2441 | dicp->di_nextents + dicp->di_anextents, | |
2442 | dicp->di_nblocks); | |
6d192a9b TS |
2443 | error = EFSCORRUPTED; |
2444 | goto error; | |
1da177e4 LT |
2445 | } |
2446 | if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) { | |
2447 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)", | |
2448 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2449 | xfs_buf_relse(bp); | |
2450 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2451 | "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x", | |
2452 | item, dip, bp, ino, dicp->di_forkoff); | |
6d192a9b TS |
2453 | error = EFSCORRUPTED; |
2454 | goto error; | |
1da177e4 | 2455 | } |
81591fe2 | 2456 | if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) { |
1da177e4 LT |
2457 | XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)", |
2458 | XFS_ERRLEVEL_LOW, mp, dicp); | |
2459 | xfs_buf_relse(bp); | |
2460 | xfs_fs_cmn_err(CE_ALERT, mp, | |
2461 | "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p", | |
2462 | item->ri_buf[1].i_len, item); | |
6d192a9b TS |
2463 | error = EFSCORRUPTED; |
2464 | goto error; | |
1da177e4 LT |
2465 | } |
2466 | ||
2467 | /* The core is in in-core format */ | |
81591fe2 | 2468 | xfs_dinode_to_disk(dip, (xfs_icdinode_t *)item->ri_buf[1].i_addr); |
1da177e4 LT |
2469 | |
2470 | /* the rest is in on-disk format */ | |
81591fe2 CH |
2471 | if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) { |
2472 | memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode), | |
2473 | item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode), | |
2474 | item->ri_buf[1].i_len - sizeof(struct xfs_icdinode)); | |
1da177e4 LT |
2475 | } |
2476 | ||
2477 | fields = in_f->ilf_fields; | |
2478 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { | |
2479 | case XFS_ILOG_DEV: | |
81591fe2 | 2480 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
1da177e4 LT |
2481 | break; |
2482 | case XFS_ILOG_UUID: | |
81591fe2 CH |
2483 | memcpy(XFS_DFORK_DPTR(dip), |
2484 | &in_f->ilf_u.ilfu_uuid, | |
2485 | sizeof(uuid_t)); | |
1da177e4 LT |
2486 | break; |
2487 | } | |
2488 | ||
2489 | if (in_f->ilf_size == 2) | |
2490 | goto write_inode_buffer; | |
2491 | len = item->ri_buf[2].i_len; | |
2492 | src = item->ri_buf[2].i_addr; | |
2493 | ASSERT(in_f->ilf_size <= 4); | |
2494 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); | |
2495 | ASSERT(!(fields & XFS_ILOG_DFORK) || | |
2496 | (len == in_f->ilf_dsize)); | |
2497 | ||
2498 | switch (fields & XFS_ILOG_DFORK) { | |
2499 | case XFS_ILOG_DDATA: | |
2500 | case XFS_ILOG_DEXT: | |
81591fe2 | 2501 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
1da177e4 LT |
2502 | break; |
2503 | ||
2504 | case XFS_ILOG_DBROOT: | |
7cc95a82 | 2505 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
81591fe2 | 2506 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
1da177e4 LT |
2507 | XFS_DFORK_DSIZE(dip, mp)); |
2508 | break; | |
2509 | ||
2510 | default: | |
2511 | /* | |
2512 | * There are no data fork flags set. | |
2513 | */ | |
2514 | ASSERT((fields & XFS_ILOG_DFORK) == 0); | |
2515 | break; | |
2516 | } | |
2517 | ||
2518 | /* | |
2519 | * If we logged any attribute data, recover it. There may or | |
2520 | * may not have been any other non-core data logged in this | |
2521 | * transaction. | |
2522 | */ | |
2523 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2524 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { | |
2525 | attr_index = 3; | |
2526 | } else { | |
2527 | attr_index = 2; | |
2528 | } | |
2529 | len = item->ri_buf[attr_index].i_len; | |
2530 | src = item->ri_buf[attr_index].i_addr; | |
2531 | ASSERT(len == in_f->ilf_asize); | |
2532 | ||
2533 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2534 | case XFS_ILOG_ADATA: | |
2535 | case XFS_ILOG_AEXT: | |
2536 | dest = XFS_DFORK_APTR(dip); | |
2537 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); | |
2538 | memcpy(dest, src, len); | |
2539 | break; | |
2540 | ||
2541 | case XFS_ILOG_ABROOT: | |
2542 | dest = XFS_DFORK_APTR(dip); | |
7cc95a82 CH |
2543 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
2544 | len, (xfs_bmdr_block_t*)dest, | |
1da177e4 LT |
2545 | XFS_DFORK_ASIZE(dip, mp)); |
2546 | break; | |
2547 | ||
2548 | default: | |
2549 | xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag"); | |
2550 | ASSERT(0); | |
2551 | xfs_buf_relse(bp); | |
6d192a9b TS |
2552 | error = EIO; |
2553 | goto error; | |
1da177e4 LT |
2554 | } |
2555 | } | |
2556 | ||
2557 | write_inode_buffer: | |
dd0bbad8 CH |
2558 | ASSERT(bp->b_mount == NULL || bp->b_mount == mp); |
2559 | bp->b_mount = mp; | |
2560 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone); | |
2561 | xfs_bdwrite(mp, bp); | |
6d192a9b TS |
2562 | error: |
2563 | if (need_free) | |
f0e2d93c | 2564 | kmem_free(in_f); |
6d192a9b | 2565 | return XFS_ERROR(error); |
1da177e4 LT |
2566 | } |
2567 | ||
2568 | /* | |
2569 | * Recover QUOTAOFF records. We simply make a note of it in the xlog_t | |
2570 | * structure, so that we know not to do any dquot item or dquot buffer recovery, | |
2571 | * of that type. | |
2572 | */ | |
2573 | STATIC int | |
2574 | xlog_recover_do_quotaoff_trans( | |
2575 | xlog_t *log, | |
2576 | xlog_recover_item_t *item, | |
2577 | int pass) | |
2578 | { | |
2579 | xfs_qoff_logformat_t *qoff_f; | |
2580 | ||
2581 | if (pass == XLOG_RECOVER_PASS2) { | |
2582 | return (0); | |
2583 | } | |
2584 | ||
2585 | qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr; | |
2586 | ASSERT(qoff_f); | |
2587 | ||
2588 | /* | |
2589 | * The logitem format's flag tells us if this was user quotaoff, | |
77a7cce4 | 2590 | * group/project quotaoff or both. |
1da177e4 LT |
2591 | */ |
2592 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) | |
2593 | log->l_quotaoffs_flag |= XFS_DQ_USER; | |
77a7cce4 NS |
2594 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
2595 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; | |
1da177e4 LT |
2596 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
2597 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; | |
2598 | ||
2599 | return (0); | |
2600 | } | |
2601 | ||
2602 | /* | |
2603 | * Recover a dquot record | |
2604 | */ | |
2605 | STATIC int | |
2606 | xlog_recover_do_dquot_trans( | |
2607 | xlog_t *log, | |
2608 | xlog_recover_item_t *item, | |
2609 | int pass) | |
2610 | { | |
2611 | xfs_mount_t *mp; | |
2612 | xfs_buf_t *bp; | |
2613 | struct xfs_disk_dquot *ddq, *recddq; | |
2614 | int error; | |
2615 | xfs_dq_logformat_t *dq_f; | |
2616 | uint type; | |
2617 | ||
2618 | if (pass == XLOG_RECOVER_PASS1) { | |
2619 | return 0; | |
2620 | } | |
2621 | mp = log->l_mp; | |
2622 | ||
2623 | /* | |
2624 | * Filesystems are required to send in quota flags at mount time. | |
2625 | */ | |
2626 | if (mp->m_qflags == 0) | |
2627 | return (0); | |
2628 | ||
2629 | recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr; | |
0c5e1ce8 CH |
2630 | |
2631 | if (item->ri_buf[1].i_addr == NULL) { | |
2632 | cmn_err(CE_ALERT, | |
2633 | "XFS: NULL dquot in %s.", __func__); | |
2634 | return XFS_ERROR(EIO); | |
2635 | } | |
8ec6dba2 | 2636 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
0c5e1ce8 CH |
2637 | cmn_err(CE_ALERT, |
2638 | "XFS: dquot too small (%d) in %s.", | |
2639 | item->ri_buf[1].i_len, __func__); | |
2640 | return XFS_ERROR(EIO); | |
2641 | } | |
2642 | ||
1da177e4 LT |
2643 | /* |
2644 | * This type of quotas was turned off, so ignore this record. | |
2645 | */ | |
b53e675d | 2646 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
1da177e4 LT |
2647 | ASSERT(type); |
2648 | if (log->l_quotaoffs_flag & type) | |
2649 | return (0); | |
2650 | ||
2651 | /* | |
2652 | * At this point we know that quota was _not_ turned off. | |
2653 | * Since the mount flags are not indicating to us otherwise, this | |
2654 | * must mean that quota is on, and the dquot needs to be replayed. | |
2655 | * Remember that we may not have fully recovered the superblock yet, | |
2656 | * so we can't do the usual trick of looking at the SB quota bits. | |
2657 | * | |
2658 | * The other possibility, of course, is that the quota subsystem was | |
2659 | * removed since the last mount - ENOSYS. | |
2660 | */ | |
2661 | dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr; | |
2662 | ASSERT(dq_f); | |
2663 | if ((error = xfs_qm_dqcheck(recddq, | |
2664 | dq_f->qlf_id, | |
2665 | 0, XFS_QMOPT_DOWARN, | |
2666 | "xlog_recover_do_dquot_trans (log copy)"))) { | |
2667 | return XFS_ERROR(EIO); | |
2668 | } | |
2669 | ASSERT(dq_f->qlf_len == 1); | |
2670 | ||
2671 | error = xfs_read_buf(mp, mp->m_ddev_targp, | |
2672 | dq_f->qlf_blkno, | |
2673 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), | |
2674 | 0, &bp); | |
2675 | if (error) { | |
2676 | xfs_ioerror_alert("xlog_recover_do..(read#3)", mp, | |
2677 | bp, dq_f->qlf_blkno); | |
2678 | return error; | |
2679 | } | |
2680 | ASSERT(bp); | |
2681 | ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset); | |
2682 | ||
2683 | /* | |
2684 | * At least the magic num portion should be on disk because this | |
2685 | * was among a chunk of dquots created earlier, and we did some | |
2686 | * minimal initialization then. | |
2687 | */ | |
2688 | if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, | |
2689 | "xlog_recover_do_dquot_trans")) { | |
2690 | xfs_buf_relse(bp); | |
2691 | return XFS_ERROR(EIO); | |
2692 | } | |
2693 | ||
2694 | memcpy(ddq, recddq, item->ri_buf[1].i_len); | |
2695 | ||
2696 | ASSERT(dq_f->qlf_size == 2); | |
15ac08a8 CH |
2697 | ASSERT(bp->b_mount == NULL || bp->b_mount == mp); |
2698 | bp->b_mount = mp; | |
1da177e4 LT |
2699 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone); |
2700 | xfs_bdwrite(mp, bp); | |
2701 | ||
2702 | return (0); | |
2703 | } | |
2704 | ||
2705 | /* | |
2706 | * This routine is called to create an in-core extent free intent | |
2707 | * item from the efi format structure which was logged on disk. | |
2708 | * It allocates an in-core efi, copies the extents from the format | |
2709 | * structure into it, and adds the efi to the AIL with the given | |
2710 | * LSN. | |
2711 | */ | |
6d192a9b | 2712 | STATIC int |
1da177e4 LT |
2713 | xlog_recover_do_efi_trans( |
2714 | xlog_t *log, | |
2715 | xlog_recover_item_t *item, | |
2716 | xfs_lsn_t lsn, | |
2717 | int pass) | |
2718 | { | |
6d192a9b | 2719 | int error; |
1da177e4 LT |
2720 | xfs_mount_t *mp; |
2721 | xfs_efi_log_item_t *efip; | |
2722 | xfs_efi_log_format_t *efi_formatp; | |
1da177e4 LT |
2723 | |
2724 | if (pass == XLOG_RECOVER_PASS1) { | |
6d192a9b | 2725 | return 0; |
1da177e4 LT |
2726 | } |
2727 | ||
2728 | efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr; | |
1da177e4 LT |
2729 | |
2730 | mp = log->l_mp; | |
2731 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); | |
6d192a9b TS |
2732 | if ((error = xfs_efi_copy_format(&(item->ri_buf[0]), |
2733 | &(efip->efi_format)))) { | |
2734 | xfs_efi_item_free(efip); | |
2735 | return error; | |
2736 | } | |
1da177e4 LT |
2737 | efip->efi_next_extent = efi_formatp->efi_nextents; |
2738 | efip->efi_flags |= XFS_EFI_COMMITTED; | |
2739 | ||
a9c21c1b | 2740 | spin_lock(&log->l_ailp->xa_lock); |
1da177e4 | 2741 | /* |
783a2f65 | 2742 | * xfs_trans_ail_update() drops the AIL lock. |
1da177e4 | 2743 | */ |
783a2f65 | 2744 | xfs_trans_ail_update(log->l_ailp, (xfs_log_item_t *)efip, lsn); |
6d192a9b | 2745 | return 0; |
1da177e4 LT |
2746 | } |
2747 | ||
2748 | ||
2749 | /* | |
2750 | * This routine is called when an efd format structure is found in | |
2751 | * a committed transaction in the log. It's purpose is to cancel | |
2752 | * the corresponding efi if it was still in the log. To do this | |
2753 | * it searches the AIL for the efi with an id equal to that in the | |
2754 | * efd format structure. If we find it, we remove the efi from the | |
2755 | * AIL and free it. | |
2756 | */ | |
2757 | STATIC void | |
2758 | xlog_recover_do_efd_trans( | |
2759 | xlog_t *log, | |
2760 | xlog_recover_item_t *item, | |
2761 | int pass) | |
2762 | { | |
1da177e4 LT |
2763 | xfs_efd_log_format_t *efd_formatp; |
2764 | xfs_efi_log_item_t *efip = NULL; | |
2765 | xfs_log_item_t *lip; | |
1da177e4 | 2766 | __uint64_t efi_id; |
27d8d5fe | 2767 | struct xfs_ail_cursor cur; |
783a2f65 | 2768 | struct xfs_ail *ailp = log->l_ailp; |
1da177e4 LT |
2769 | |
2770 | if (pass == XLOG_RECOVER_PASS1) { | |
2771 | return; | |
2772 | } | |
2773 | ||
2774 | efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr; | |
6d192a9b TS |
2775 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
2776 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || | |
2777 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + | |
2778 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); | |
1da177e4 LT |
2779 | efi_id = efd_formatp->efd_efi_id; |
2780 | ||
2781 | /* | |
2782 | * Search for the efi with the id in the efd format structure | |
2783 | * in the AIL. | |
2784 | */ | |
a9c21c1b DC |
2785 | spin_lock(&ailp->xa_lock); |
2786 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
2787 | while (lip != NULL) { |
2788 | if (lip->li_type == XFS_LI_EFI) { | |
2789 | efip = (xfs_efi_log_item_t *)lip; | |
2790 | if (efip->efi_format.efi_id == efi_id) { | |
2791 | /* | |
783a2f65 | 2792 | * xfs_trans_ail_delete() drops the |
1da177e4 LT |
2793 | * AIL lock. |
2794 | */ | |
783a2f65 | 2795 | xfs_trans_ail_delete(ailp, lip); |
8ae2c0f6 | 2796 | xfs_efi_item_free(efip); |
a9c21c1b | 2797 | spin_lock(&ailp->xa_lock); |
27d8d5fe | 2798 | break; |
1da177e4 LT |
2799 | } |
2800 | } | |
a9c21c1b | 2801 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 2802 | } |
a9c21c1b DC |
2803 | xfs_trans_ail_cursor_done(ailp, &cur); |
2804 | spin_unlock(&ailp->xa_lock); | |
1da177e4 LT |
2805 | } |
2806 | ||
2807 | /* | |
2808 | * Perform the transaction | |
2809 | * | |
2810 | * If the transaction modifies a buffer or inode, do it now. Otherwise, | |
2811 | * EFIs and EFDs get queued up by adding entries into the AIL for them. | |
2812 | */ | |
2813 | STATIC int | |
2814 | xlog_recover_do_trans( | |
2815 | xlog_t *log, | |
2816 | xlog_recover_t *trans, | |
2817 | int pass) | |
2818 | { | |
2819 | int error = 0; | |
f0a76953 | 2820 | xlog_recover_item_t *item; |
1da177e4 | 2821 | |
9abbc539 | 2822 | error = xlog_recover_reorder_trans(log, trans, pass); |
ff0205e0 | 2823 | if (error) |
1da177e4 | 2824 | return error; |
ff0205e0 | 2825 | |
f0a76953 | 2826 | list_for_each_entry(item, &trans->r_itemq, ri_list) { |
9abbc539 | 2827 | trace_xfs_log_recover_item_recover(log, trans, item, pass); |
ff0205e0 CH |
2828 | switch (ITEM_TYPE(item)) { |
2829 | case XFS_LI_BUF: | |
2830 | error = xlog_recover_do_buffer_trans(log, item, pass); | |
2831 | break; | |
2832 | case XFS_LI_INODE: | |
2833 | error = xlog_recover_do_inode_trans(log, item, pass); | |
2834 | break; | |
2835 | case XFS_LI_EFI: | |
2836 | error = xlog_recover_do_efi_trans(log, item, | |
2837 | trans->r_lsn, pass); | |
2838 | break; | |
2839 | case XFS_LI_EFD: | |
1da177e4 | 2840 | xlog_recover_do_efd_trans(log, item, pass); |
ff0205e0 CH |
2841 | error = 0; |
2842 | break; | |
2843 | case XFS_LI_DQUOT: | |
2844 | error = xlog_recover_do_dquot_trans(log, item, pass); | |
2845 | break; | |
2846 | case XFS_LI_QUOTAOFF: | |
2847 | error = xlog_recover_do_quotaoff_trans(log, item, | |
2848 | pass); | |
2849 | break; | |
2850 | default: | |
2851 | xlog_warn( | |
2852 | "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item)); | |
1da177e4 LT |
2853 | ASSERT(0); |
2854 | error = XFS_ERROR(EIO); | |
2855 | break; | |
2856 | } | |
ff0205e0 CH |
2857 | |
2858 | if (error) | |
2859 | return error; | |
f0a76953 | 2860 | } |
1da177e4 | 2861 | |
ff0205e0 | 2862 | return 0; |
1da177e4 LT |
2863 | } |
2864 | ||
2865 | /* | |
2866 | * Free up any resources allocated by the transaction | |
2867 | * | |
2868 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. | |
2869 | */ | |
2870 | STATIC void | |
2871 | xlog_recover_free_trans( | |
2872 | xlog_recover_t *trans) | |
2873 | { | |
f0a76953 | 2874 | xlog_recover_item_t *item, *n; |
1da177e4 LT |
2875 | int i; |
2876 | ||
f0a76953 DC |
2877 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { |
2878 | /* Free the regions in the item. */ | |
2879 | list_del(&item->ri_list); | |
2880 | for (i = 0; i < item->ri_cnt; i++) | |
2881 | kmem_free(item->ri_buf[i].i_addr); | |
1da177e4 | 2882 | /* Free the item itself */ |
f0a76953 DC |
2883 | kmem_free(item->ri_buf); |
2884 | kmem_free(item); | |
2885 | } | |
1da177e4 | 2886 | /* Free the transaction recover structure */ |
f0e2d93c | 2887 | kmem_free(trans); |
1da177e4 LT |
2888 | } |
2889 | ||
2890 | STATIC int | |
2891 | xlog_recover_commit_trans( | |
2892 | xlog_t *log, | |
1da177e4 LT |
2893 | xlog_recover_t *trans, |
2894 | int pass) | |
2895 | { | |
2896 | int error; | |
2897 | ||
f0a76953 | 2898 | hlist_del(&trans->r_list); |
1da177e4 LT |
2899 | if ((error = xlog_recover_do_trans(log, trans, pass))) |
2900 | return error; | |
2901 | xlog_recover_free_trans(trans); /* no error */ | |
2902 | return 0; | |
2903 | } | |
2904 | ||
2905 | STATIC int | |
2906 | xlog_recover_unmount_trans( | |
2907 | xlog_recover_t *trans) | |
2908 | { | |
2909 | /* Do nothing now */ | |
2910 | xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR"); | |
2911 | return 0; | |
2912 | } | |
2913 | ||
2914 | /* | |
2915 | * There are two valid states of the r_state field. 0 indicates that the | |
2916 | * transaction structure is in a normal state. We have either seen the | |
2917 | * start of the transaction or the last operation we added was not a partial | |
2918 | * operation. If the last operation we added to the transaction was a | |
2919 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. | |
2920 | * | |
2921 | * NOTE: skip LRs with 0 data length. | |
2922 | */ | |
2923 | STATIC int | |
2924 | xlog_recover_process_data( | |
2925 | xlog_t *log, | |
f0a76953 | 2926 | struct hlist_head rhash[], |
1da177e4 LT |
2927 | xlog_rec_header_t *rhead, |
2928 | xfs_caddr_t dp, | |
2929 | int pass) | |
2930 | { | |
2931 | xfs_caddr_t lp; | |
2932 | int num_logops; | |
2933 | xlog_op_header_t *ohead; | |
2934 | xlog_recover_t *trans; | |
2935 | xlog_tid_t tid; | |
2936 | int error; | |
2937 | unsigned long hash; | |
2938 | uint flags; | |
2939 | ||
b53e675d CH |
2940 | lp = dp + be32_to_cpu(rhead->h_len); |
2941 | num_logops = be32_to_cpu(rhead->h_num_logops); | |
1da177e4 LT |
2942 | |
2943 | /* check the log format matches our own - else we can't recover */ | |
2944 | if (xlog_header_check_recover(log->l_mp, rhead)) | |
2945 | return (XFS_ERROR(EIO)); | |
2946 | ||
2947 | while ((dp < lp) && num_logops) { | |
2948 | ASSERT(dp + sizeof(xlog_op_header_t) <= lp); | |
2949 | ohead = (xlog_op_header_t *)dp; | |
2950 | dp += sizeof(xlog_op_header_t); | |
2951 | if (ohead->oh_clientid != XFS_TRANSACTION && | |
2952 | ohead->oh_clientid != XFS_LOG) { | |
2953 | xlog_warn( | |
2954 | "XFS: xlog_recover_process_data: bad clientid"); | |
2955 | ASSERT(0); | |
2956 | return (XFS_ERROR(EIO)); | |
2957 | } | |
67fcb7bf | 2958 | tid = be32_to_cpu(ohead->oh_tid); |
1da177e4 | 2959 | hash = XLOG_RHASH(tid); |
f0a76953 | 2960 | trans = xlog_recover_find_tid(&rhash[hash], tid); |
1da177e4 LT |
2961 | if (trans == NULL) { /* not found; add new tid */ |
2962 | if (ohead->oh_flags & XLOG_START_TRANS) | |
2963 | xlog_recover_new_tid(&rhash[hash], tid, | |
b53e675d | 2964 | be64_to_cpu(rhead->h_lsn)); |
1da177e4 | 2965 | } else { |
9742bb93 LM |
2966 | if (dp + be32_to_cpu(ohead->oh_len) > lp) { |
2967 | xlog_warn( | |
2968 | "XFS: xlog_recover_process_data: bad length"); | |
2969 | WARN_ON(1); | |
2970 | return (XFS_ERROR(EIO)); | |
2971 | } | |
1da177e4 LT |
2972 | flags = ohead->oh_flags & ~XLOG_END_TRANS; |
2973 | if (flags & XLOG_WAS_CONT_TRANS) | |
2974 | flags &= ~XLOG_CONTINUE_TRANS; | |
2975 | switch (flags) { | |
2976 | case XLOG_COMMIT_TRANS: | |
2977 | error = xlog_recover_commit_trans(log, | |
f0a76953 | 2978 | trans, pass); |
1da177e4 LT |
2979 | break; |
2980 | case XLOG_UNMOUNT_TRANS: | |
2981 | error = xlog_recover_unmount_trans(trans); | |
2982 | break; | |
2983 | case XLOG_WAS_CONT_TRANS: | |
9abbc539 DC |
2984 | error = xlog_recover_add_to_cont_trans(log, |
2985 | trans, dp, | |
2986 | be32_to_cpu(ohead->oh_len)); | |
1da177e4 LT |
2987 | break; |
2988 | case XLOG_START_TRANS: | |
2989 | xlog_warn( | |
2990 | "XFS: xlog_recover_process_data: bad transaction"); | |
2991 | ASSERT(0); | |
2992 | error = XFS_ERROR(EIO); | |
2993 | break; | |
2994 | case 0: | |
2995 | case XLOG_CONTINUE_TRANS: | |
9abbc539 | 2996 | error = xlog_recover_add_to_trans(log, trans, |
67fcb7bf | 2997 | dp, be32_to_cpu(ohead->oh_len)); |
1da177e4 LT |
2998 | break; |
2999 | default: | |
3000 | xlog_warn( | |
3001 | "XFS: xlog_recover_process_data: bad flag"); | |
3002 | ASSERT(0); | |
3003 | error = XFS_ERROR(EIO); | |
3004 | break; | |
3005 | } | |
3006 | if (error) | |
3007 | return error; | |
3008 | } | |
67fcb7bf | 3009 | dp += be32_to_cpu(ohead->oh_len); |
1da177e4 LT |
3010 | num_logops--; |
3011 | } | |
3012 | return 0; | |
3013 | } | |
3014 | ||
3015 | /* | |
3016 | * Process an extent free intent item that was recovered from | |
3017 | * the log. We need to free the extents that it describes. | |
3018 | */ | |
3c1e2bbe | 3019 | STATIC int |
1da177e4 LT |
3020 | xlog_recover_process_efi( |
3021 | xfs_mount_t *mp, | |
3022 | xfs_efi_log_item_t *efip) | |
3023 | { | |
3024 | xfs_efd_log_item_t *efdp; | |
3025 | xfs_trans_t *tp; | |
3026 | int i; | |
3c1e2bbe | 3027 | int error = 0; |
1da177e4 LT |
3028 | xfs_extent_t *extp; |
3029 | xfs_fsblock_t startblock_fsb; | |
3030 | ||
3031 | ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED)); | |
3032 | ||
3033 | /* | |
3034 | * First check the validity of the extents described by the | |
3035 | * EFI. If any are bad, then assume that all are bad and | |
3036 | * just toss the EFI. | |
3037 | */ | |
3038 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3039 | extp = &(efip->efi_format.efi_extents[i]); | |
3040 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
3041 | XFS_FSB_TO_DADDR(mp, extp->ext_start)); | |
3042 | if ((startblock_fsb == 0) || | |
3043 | (extp->ext_len == 0) || | |
3044 | (startblock_fsb >= mp->m_sb.sb_dblocks) || | |
3045 | (extp->ext_len >= mp->m_sb.sb_agblocks)) { | |
3046 | /* | |
3047 | * This will pull the EFI from the AIL and | |
3048 | * free the memory associated with it. | |
3049 | */ | |
3050 | xfs_efi_release(efip, efip->efi_format.efi_nextents); | |
3c1e2bbe | 3051 | return XFS_ERROR(EIO); |
1da177e4 LT |
3052 | } |
3053 | } | |
3054 | ||
3055 | tp = xfs_trans_alloc(mp, 0); | |
3c1e2bbe | 3056 | error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0); |
fc6149d8 DC |
3057 | if (error) |
3058 | goto abort_error; | |
1da177e4 LT |
3059 | efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents); |
3060 | ||
3061 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3062 | extp = &(efip->efi_format.efi_extents[i]); | |
fc6149d8 DC |
3063 | error = xfs_free_extent(tp, extp->ext_start, extp->ext_len); |
3064 | if (error) | |
3065 | goto abort_error; | |
1da177e4 LT |
3066 | xfs_trans_log_efd_extent(tp, efdp, extp->ext_start, |
3067 | extp->ext_len); | |
3068 | } | |
3069 | ||
3070 | efip->efi_flags |= XFS_EFI_RECOVERED; | |
e5720eec | 3071 | error = xfs_trans_commit(tp, 0); |
3c1e2bbe | 3072 | return error; |
fc6149d8 DC |
3073 | |
3074 | abort_error: | |
3075 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3076 | return error; | |
1da177e4 LT |
3077 | } |
3078 | ||
1da177e4 LT |
3079 | /* |
3080 | * When this is called, all of the EFIs which did not have | |
3081 | * corresponding EFDs should be in the AIL. What we do now | |
3082 | * is free the extents associated with each one. | |
3083 | * | |
3084 | * Since we process the EFIs in normal transactions, they | |
3085 | * will be removed at some point after the commit. This prevents | |
3086 | * us from just walking down the list processing each one. | |
3087 | * We'll use a flag in the EFI to skip those that we've already | |
3088 | * processed and use the AIL iteration mechanism's generation | |
3089 | * count to try to speed this up at least a bit. | |
3090 | * | |
3091 | * When we start, we know that the EFIs are the only things in | |
3092 | * the AIL. As we process them, however, other items are added | |
3093 | * to the AIL. Since everything added to the AIL must come after | |
3094 | * everything already in the AIL, we stop processing as soon as | |
3095 | * we see something other than an EFI in the AIL. | |
3096 | */ | |
3c1e2bbe | 3097 | STATIC int |
1da177e4 LT |
3098 | xlog_recover_process_efis( |
3099 | xlog_t *log) | |
3100 | { | |
3101 | xfs_log_item_t *lip; | |
3102 | xfs_efi_log_item_t *efip; | |
3c1e2bbe | 3103 | int error = 0; |
27d8d5fe | 3104 | struct xfs_ail_cursor cur; |
a9c21c1b | 3105 | struct xfs_ail *ailp; |
1da177e4 | 3106 | |
a9c21c1b DC |
3107 | ailp = log->l_ailp; |
3108 | spin_lock(&ailp->xa_lock); | |
3109 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
3110 | while (lip != NULL) { |
3111 | /* | |
3112 | * We're done when we see something other than an EFI. | |
27d8d5fe | 3113 | * There should be no EFIs left in the AIL now. |
1da177e4 LT |
3114 | */ |
3115 | if (lip->li_type != XFS_LI_EFI) { | |
27d8d5fe | 3116 | #ifdef DEBUG |
a9c21c1b | 3117 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
27d8d5fe DC |
3118 | ASSERT(lip->li_type != XFS_LI_EFI); |
3119 | #endif | |
1da177e4 LT |
3120 | break; |
3121 | } | |
3122 | ||
3123 | /* | |
3124 | * Skip EFIs that we've already processed. | |
3125 | */ | |
3126 | efip = (xfs_efi_log_item_t *)lip; | |
3127 | if (efip->efi_flags & XFS_EFI_RECOVERED) { | |
a9c21c1b | 3128 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 LT |
3129 | continue; |
3130 | } | |
3131 | ||
a9c21c1b DC |
3132 | spin_unlock(&ailp->xa_lock); |
3133 | error = xlog_recover_process_efi(log->l_mp, efip); | |
3134 | spin_lock(&ailp->xa_lock); | |
27d8d5fe DC |
3135 | if (error) |
3136 | goto out; | |
a9c21c1b | 3137 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 3138 | } |
27d8d5fe | 3139 | out: |
a9c21c1b DC |
3140 | xfs_trans_ail_cursor_done(ailp, &cur); |
3141 | spin_unlock(&ailp->xa_lock); | |
3c1e2bbe | 3142 | return error; |
1da177e4 LT |
3143 | } |
3144 | ||
3145 | /* | |
3146 | * This routine performs a transaction to null out a bad inode pointer | |
3147 | * in an agi unlinked inode hash bucket. | |
3148 | */ | |
3149 | STATIC void | |
3150 | xlog_recover_clear_agi_bucket( | |
3151 | xfs_mount_t *mp, | |
3152 | xfs_agnumber_t agno, | |
3153 | int bucket) | |
3154 | { | |
3155 | xfs_trans_t *tp; | |
3156 | xfs_agi_t *agi; | |
3157 | xfs_buf_t *agibp; | |
3158 | int offset; | |
3159 | int error; | |
3160 | ||
3161 | tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET); | |
5e1be0fb CH |
3162 | error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), |
3163 | 0, 0, 0); | |
e5720eec DC |
3164 | if (error) |
3165 | goto out_abort; | |
1da177e4 | 3166 | |
5e1be0fb CH |
3167 | error = xfs_read_agi(mp, tp, agno, &agibp); |
3168 | if (error) | |
e5720eec | 3169 | goto out_abort; |
1da177e4 | 3170 | |
5e1be0fb | 3171 | agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 3172 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
3173 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
3174 | (sizeof(xfs_agino_t) * bucket); | |
3175 | xfs_trans_log_buf(tp, agibp, offset, | |
3176 | (offset + sizeof(xfs_agino_t) - 1)); | |
3177 | ||
e5720eec DC |
3178 | error = xfs_trans_commit(tp, 0); |
3179 | if (error) | |
3180 | goto out_error; | |
3181 | return; | |
3182 | ||
3183 | out_abort: | |
3184 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3185 | out_error: | |
3186 | xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: " | |
3187 | "failed to clear agi %d. Continuing.", agno); | |
3188 | return; | |
1da177e4 LT |
3189 | } |
3190 | ||
23fac50f CH |
3191 | STATIC xfs_agino_t |
3192 | xlog_recover_process_one_iunlink( | |
3193 | struct xfs_mount *mp, | |
3194 | xfs_agnumber_t agno, | |
3195 | xfs_agino_t agino, | |
3196 | int bucket) | |
3197 | { | |
3198 | struct xfs_buf *ibp; | |
3199 | struct xfs_dinode *dip; | |
3200 | struct xfs_inode *ip; | |
3201 | xfs_ino_t ino; | |
3202 | int error; | |
3203 | ||
3204 | ino = XFS_AGINO_TO_INO(mp, agno, agino); | |
3205 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0); | |
3206 | if (error) | |
3207 | goto fail; | |
3208 | ||
3209 | /* | |
3210 | * Get the on disk inode to find the next inode in the bucket. | |
3211 | */ | |
0cadda1c | 3212 | error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK); |
23fac50f | 3213 | if (error) |
0e446673 | 3214 | goto fail_iput; |
23fac50f | 3215 | |
23fac50f | 3216 | ASSERT(ip->i_d.di_nlink == 0); |
0e446673 | 3217 | ASSERT(ip->i_d.di_mode != 0); |
23fac50f CH |
3218 | |
3219 | /* setup for the next pass */ | |
3220 | agino = be32_to_cpu(dip->di_next_unlinked); | |
3221 | xfs_buf_relse(ibp); | |
3222 | ||
3223 | /* | |
3224 | * Prevent any DMAPI event from being sent when the reference on | |
3225 | * the inode is dropped. | |
3226 | */ | |
3227 | ip->i_d.di_dmevmask = 0; | |
3228 | ||
0e446673 | 3229 | IRELE(ip); |
23fac50f CH |
3230 | return agino; |
3231 | ||
0e446673 CH |
3232 | fail_iput: |
3233 | IRELE(ip); | |
23fac50f CH |
3234 | fail: |
3235 | /* | |
3236 | * We can't read in the inode this bucket points to, or this inode | |
3237 | * is messed up. Just ditch this bucket of inodes. We will lose | |
3238 | * some inodes and space, but at least we won't hang. | |
3239 | * | |
3240 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to | |
3241 | * clear the inode pointer in the bucket. | |
3242 | */ | |
3243 | xlog_recover_clear_agi_bucket(mp, agno, bucket); | |
3244 | return NULLAGINO; | |
3245 | } | |
3246 | ||
1da177e4 LT |
3247 | /* |
3248 | * xlog_iunlink_recover | |
3249 | * | |
3250 | * This is called during recovery to process any inodes which | |
3251 | * we unlinked but not freed when the system crashed. These | |
3252 | * inodes will be on the lists in the AGI blocks. What we do | |
3253 | * here is scan all the AGIs and fully truncate and free any | |
3254 | * inodes found on the lists. Each inode is removed from the | |
3255 | * lists when it has been fully truncated and is freed. The | |
3256 | * freeing of the inode and its removal from the list must be | |
3257 | * atomic. | |
3258 | */ | |
d96f8f89 | 3259 | STATIC void |
1da177e4 LT |
3260 | xlog_recover_process_iunlinks( |
3261 | xlog_t *log) | |
3262 | { | |
3263 | xfs_mount_t *mp; | |
3264 | xfs_agnumber_t agno; | |
3265 | xfs_agi_t *agi; | |
3266 | xfs_buf_t *agibp; | |
1da177e4 | 3267 | xfs_agino_t agino; |
1da177e4 LT |
3268 | int bucket; |
3269 | int error; | |
3270 | uint mp_dmevmask; | |
3271 | ||
3272 | mp = log->l_mp; | |
3273 | ||
3274 | /* | |
3275 | * Prevent any DMAPI event from being sent while in this function. | |
3276 | */ | |
3277 | mp_dmevmask = mp->m_dmevmask; | |
3278 | mp->m_dmevmask = 0; | |
3279 | ||
3280 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
3281 | /* | |
3282 | * Find the agi for this ag. | |
3283 | */ | |
5e1be0fb CH |
3284 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
3285 | if (error) { | |
3286 | /* | |
3287 | * AGI is b0rked. Don't process it. | |
3288 | * | |
3289 | * We should probably mark the filesystem as corrupt | |
3290 | * after we've recovered all the ag's we can.... | |
3291 | */ | |
3292 | continue; | |
1da177e4 LT |
3293 | } |
3294 | agi = XFS_BUF_TO_AGI(agibp); | |
1da177e4 LT |
3295 | |
3296 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { | |
16259e7d | 3297 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
1da177e4 | 3298 | while (agino != NULLAGINO) { |
1da177e4 LT |
3299 | /* |
3300 | * Release the agi buffer so that it can | |
3301 | * be acquired in the normal course of the | |
3302 | * transaction to truncate and free the inode. | |
3303 | */ | |
3304 | xfs_buf_relse(agibp); | |
3305 | ||
23fac50f CH |
3306 | agino = xlog_recover_process_one_iunlink(mp, |
3307 | agno, agino, bucket); | |
1da177e4 LT |
3308 | |
3309 | /* | |
3310 | * Reacquire the agibuffer and continue around | |
5e1be0fb CH |
3311 | * the loop. This should never fail as we know |
3312 | * the buffer was good earlier on. | |
1da177e4 | 3313 | */ |
5e1be0fb CH |
3314 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
3315 | ASSERT(error == 0); | |
1da177e4 | 3316 | agi = XFS_BUF_TO_AGI(agibp); |
1da177e4 LT |
3317 | } |
3318 | } | |
3319 | ||
3320 | /* | |
3321 | * Release the buffer for the current agi so we can | |
3322 | * go on to the next one. | |
3323 | */ | |
3324 | xfs_buf_relse(agibp); | |
3325 | } | |
3326 | ||
3327 | mp->m_dmevmask = mp_dmevmask; | |
3328 | } | |
3329 | ||
3330 | ||
3331 | #ifdef DEBUG | |
3332 | STATIC void | |
3333 | xlog_pack_data_checksum( | |
3334 | xlog_t *log, | |
3335 | xlog_in_core_t *iclog, | |
3336 | int size) | |
3337 | { | |
3338 | int i; | |
b53e675d | 3339 | __be32 *up; |
1da177e4 LT |
3340 | uint chksum = 0; |
3341 | ||
b53e675d | 3342 | up = (__be32 *)iclog->ic_datap; |
1da177e4 LT |
3343 | /* divide length by 4 to get # words */ |
3344 | for (i = 0; i < (size >> 2); i++) { | |
b53e675d | 3345 | chksum ^= be32_to_cpu(*up); |
1da177e4 LT |
3346 | up++; |
3347 | } | |
b53e675d | 3348 | iclog->ic_header.h_chksum = cpu_to_be32(chksum); |
1da177e4 LT |
3349 | } |
3350 | #else | |
3351 | #define xlog_pack_data_checksum(log, iclog, size) | |
3352 | #endif | |
3353 | ||
3354 | /* | |
3355 | * Stamp cycle number in every block | |
3356 | */ | |
3357 | void | |
3358 | xlog_pack_data( | |
3359 | xlog_t *log, | |
3360 | xlog_in_core_t *iclog, | |
3361 | int roundoff) | |
3362 | { | |
3363 | int i, j, k; | |
3364 | int size = iclog->ic_offset + roundoff; | |
b53e675d | 3365 | __be32 cycle_lsn; |
1da177e4 | 3366 | xfs_caddr_t dp; |
1da177e4 LT |
3367 | |
3368 | xlog_pack_data_checksum(log, iclog, size); | |
3369 | ||
3370 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
3371 | ||
3372 | dp = iclog->ic_datap; | |
3373 | for (i = 0; i < BTOBB(size) && | |
3374 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { | |
b53e675d CH |
3375 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; |
3376 | *(__be32 *)dp = cycle_lsn; | |
1da177e4 LT |
3377 | dp += BBSIZE; |
3378 | } | |
3379 | ||
62118709 | 3380 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b28708d6 CH |
3381 | xlog_in_core_2_t *xhdr = iclog->ic_data; |
3382 | ||
1da177e4 LT |
3383 | for ( ; i < BTOBB(size); i++) { |
3384 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3385 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d CH |
3386 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; |
3387 | *(__be32 *)dp = cycle_lsn; | |
1da177e4 LT |
3388 | dp += BBSIZE; |
3389 | } | |
3390 | ||
3391 | for (i = 1; i < log->l_iclog_heads; i++) { | |
3392 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
3393 | } | |
3394 | } | |
3395 | } | |
3396 | ||
3397 | #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY) | |
3398 | STATIC void | |
3399 | xlog_unpack_data_checksum( | |
3400 | xlog_rec_header_t *rhead, | |
3401 | xfs_caddr_t dp, | |
3402 | xlog_t *log) | |
3403 | { | |
b53e675d | 3404 | __be32 *up = (__be32 *)dp; |
1da177e4 LT |
3405 | uint chksum = 0; |
3406 | int i; | |
3407 | ||
3408 | /* divide length by 4 to get # words */ | |
b53e675d CH |
3409 | for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) { |
3410 | chksum ^= be32_to_cpu(*up); | |
1da177e4 LT |
3411 | up++; |
3412 | } | |
b53e675d | 3413 | if (chksum != be32_to_cpu(rhead->h_chksum)) { |
1da177e4 LT |
3414 | if (rhead->h_chksum || |
3415 | ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) { | |
3416 | cmn_err(CE_DEBUG, | |
b6574520 | 3417 | "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n", |
b53e675d | 3418 | be32_to_cpu(rhead->h_chksum), chksum); |
1da177e4 LT |
3419 | cmn_err(CE_DEBUG, |
3420 | "XFS: Disregard message if filesystem was created with non-DEBUG kernel"); | |
62118709 | 3421 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
1da177e4 | 3422 | cmn_err(CE_DEBUG, |
b6574520 | 3423 | "XFS: LogR this is a LogV2 filesystem\n"); |
1da177e4 LT |
3424 | } |
3425 | log->l_flags |= XLOG_CHKSUM_MISMATCH; | |
3426 | } | |
3427 | } | |
3428 | } | |
3429 | #else | |
3430 | #define xlog_unpack_data_checksum(rhead, dp, log) | |
3431 | #endif | |
3432 | ||
3433 | STATIC void | |
3434 | xlog_unpack_data( | |
3435 | xlog_rec_header_t *rhead, | |
3436 | xfs_caddr_t dp, | |
3437 | xlog_t *log) | |
3438 | { | |
3439 | int i, j, k; | |
1da177e4 | 3440 | |
b53e675d | 3441 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
1da177e4 | 3442 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
b53e675d | 3443 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
1da177e4 LT |
3444 | dp += BBSIZE; |
3445 | } | |
3446 | ||
62118709 | 3447 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b28708d6 | 3448 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
b53e675d | 3449 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
1da177e4 LT |
3450 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
3451 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3452 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
1da177e4 LT |
3453 | dp += BBSIZE; |
3454 | } | |
3455 | } | |
3456 | ||
3457 | xlog_unpack_data_checksum(rhead, dp, log); | |
3458 | } | |
3459 | ||
3460 | STATIC int | |
3461 | xlog_valid_rec_header( | |
3462 | xlog_t *log, | |
3463 | xlog_rec_header_t *rhead, | |
3464 | xfs_daddr_t blkno) | |
3465 | { | |
3466 | int hlen; | |
3467 | ||
b53e675d | 3468 | if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) { |
1da177e4 LT |
3469 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
3470 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3471 | return XFS_ERROR(EFSCORRUPTED); | |
3472 | } | |
3473 | if (unlikely( | |
3474 | (!rhead->h_version || | |
b53e675d | 3475 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
1da177e4 | 3476 | xlog_warn("XFS: %s: unrecognised log version (%d).", |
34a622b2 | 3477 | __func__, be32_to_cpu(rhead->h_version)); |
1da177e4 LT |
3478 | return XFS_ERROR(EIO); |
3479 | } | |
3480 | ||
3481 | /* LR body must have data or it wouldn't have been written */ | |
b53e675d | 3482 | hlen = be32_to_cpu(rhead->h_len); |
1da177e4 LT |
3483 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
3484 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", | |
3485 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3486 | return XFS_ERROR(EFSCORRUPTED); | |
3487 | } | |
3488 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { | |
3489 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", | |
3490 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3491 | return XFS_ERROR(EFSCORRUPTED); | |
3492 | } | |
3493 | return 0; | |
3494 | } | |
3495 | ||
3496 | /* | |
3497 | * Read the log from tail to head and process the log records found. | |
3498 | * Handle the two cases where the tail and head are in the same cycle | |
3499 | * and where the active portion of the log wraps around the end of | |
3500 | * the physical log separately. The pass parameter is passed through | |
3501 | * to the routines called to process the data and is not looked at | |
3502 | * here. | |
3503 | */ | |
3504 | STATIC int | |
3505 | xlog_do_recovery_pass( | |
3506 | xlog_t *log, | |
3507 | xfs_daddr_t head_blk, | |
3508 | xfs_daddr_t tail_blk, | |
3509 | int pass) | |
3510 | { | |
3511 | xlog_rec_header_t *rhead; | |
3512 | xfs_daddr_t blk_no; | |
fc5bc4c8 | 3513 | xfs_caddr_t offset; |
1da177e4 LT |
3514 | xfs_buf_t *hbp, *dbp; |
3515 | int error = 0, h_size; | |
3516 | int bblks, split_bblks; | |
3517 | int hblks, split_hblks, wrapped_hblks; | |
f0a76953 | 3518 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
1da177e4 LT |
3519 | |
3520 | ASSERT(head_blk != tail_blk); | |
3521 | ||
3522 | /* | |
3523 | * Read the header of the tail block and get the iclog buffer size from | |
3524 | * h_size. Use this to tell how many sectors make up the log header. | |
3525 | */ | |
62118709 | 3526 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
1da177e4 LT |
3527 | /* |
3528 | * When using variable length iclogs, read first sector of | |
3529 | * iclog header and extract the header size from it. Get a | |
3530 | * new hbp that is the correct size. | |
3531 | */ | |
3532 | hbp = xlog_get_bp(log, 1); | |
3533 | if (!hbp) | |
3534 | return ENOMEM; | |
076e6acb CH |
3535 | |
3536 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); | |
3537 | if (error) | |
1da177e4 | 3538 | goto bread_err1; |
076e6acb | 3539 | |
1da177e4 LT |
3540 | rhead = (xlog_rec_header_t *)offset; |
3541 | error = xlog_valid_rec_header(log, rhead, tail_blk); | |
3542 | if (error) | |
3543 | goto bread_err1; | |
b53e675d CH |
3544 | h_size = be32_to_cpu(rhead->h_size); |
3545 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && | |
1da177e4 LT |
3546 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
3547 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
3548 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
3549 | hblks++; | |
3550 | xlog_put_bp(hbp); | |
3551 | hbp = xlog_get_bp(log, hblks); | |
3552 | } else { | |
3553 | hblks = 1; | |
3554 | } | |
3555 | } else { | |
3556 | ASSERT(log->l_sectbb_log == 0); | |
3557 | hblks = 1; | |
3558 | hbp = xlog_get_bp(log, 1); | |
3559 | h_size = XLOG_BIG_RECORD_BSIZE; | |
3560 | } | |
3561 | ||
3562 | if (!hbp) | |
3563 | return ENOMEM; | |
3564 | dbp = xlog_get_bp(log, BTOBB(h_size)); | |
3565 | if (!dbp) { | |
3566 | xlog_put_bp(hbp); | |
3567 | return ENOMEM; | |
3568 | } | |
3569 | ||
3570 | memset(rhash, 0, sizeof(rhash)); | |
3571 | if (tail_blk <= head_blk) { | |
3572 | for (blk_no = tail_blk; blk_no < head_blk; ) { | |
076e6acb CH |
3573 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
3574 | if (error) | |
1da177e4 | 3575 | goto bread_err2; |
076e6acb | 3576 | |
1da177e4 LT |
3577 | rhead = (xlog_rec_header_t *)offset; |
3578 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3579 | if (error) | |
3580 | goto bread_err2; | |
3581 | ||
3582 | /* blocks in data section */ | |
b53e675d | 3583 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
3584 | error = xlog_bread(log, blk_no + hblks, bblks, dbp, |
3585 | &offset); | |
1da177e4 LT |
3586 | if (error) |
3587 | goto bread_err2; | |
076e6acb | 3588 | |
1da177e4 LT |
3589 | xlog_unpack_data(rhead, offset, log); |
3590 | if ((error = xlog_recover_process_data(log, | |
3591 | rhash, rhead, offset, pass))) | |
3592 | goto bread_err2; | |
3593 | blk_no += bblks + hblks; | |
3594 | } | |
3595 | } else { | |
3596 | /* | |
3597 | * Perform recovery around the end of the physical log. | |
3598 | * When the head is not on the same cycle number as the tail, | |
3599 | * we can't do a sequential recovery as above. | |
3600 | */ | |
3601 | blk_no = tail_blk; | |
3602 | while (blk_no < log->l_logBBsize) { | |
3603 | /* | |
3604 | * Check for header wrapping around physical end-of-log | |
3605 | */ | |
fc5bc4c8 | 3606 | offset = XFS_BUF_PTR(hbp); |
1da177e4 LT |
3607 | split_hblks = 0; |
3608 | wrapped_hblks = 0; | |
3609 | if (blk_no + hblks <= log->l_logBBsize) { | |
3610 | /* Read header in one read */ | |
076e6acb CH |
3611 | error = xlog_bread(log, blk_no, hblks, hbp, |
3612 | &offset); | |
1da177e4 LT |
3613 | if (error) |
3614 | goto bread_err2; | |
1da177e4 LT |
3615 | } else { |
3616 | /* This LR is split across physical log end */ | |
3617 | if (blk_no != log->l_logBBsize) { | |
3618 | /* some data before physical log end */ | |
3619 | ASSERT(blk_no <= INT_MAX); | |
3620 | split_hblks = log->l_logBBsize - (int)blk_no; | |
3621 | ASSERT(split_hblks > 0); | |
076e6acb CH |
3622 | error = xlog_bread(log, blk_no, |
3623 | split_hblks, hbp, | |
3624 | &offset); | |
3625 | if (error) | |
1da177e4 | 3626 | goto bread_err2; |
1da177e4 | 3627 | } |
076e6acb | 3628 | |
1da177e4 LT |
3629 | /* |
3630 | * Note: this black magic still works with | |
3631 | * large sector sizes (non-512) only because: | |
3632 | * - we increased the buffer size originally | |
3633 | * by 1 sector giving us enough extra space | |
3634 | * for the second read; | |
3635 | * - the log start is guaranteed to be sector | |
3636 | * aligned; | |
3637 | * - we read the log end (LR header start) | |
3638 | * _first_, then the log start (LR header end) | |
3639 | * - order is important. | |
3640 | */ | |
234f56ac | 3641 | wrapped_hblks = hblks - split_hblks; |
234f56ac | 3642 | error = XFS_BUF_SET_PTR(hbp, |
fc5bc4c8 | 3643 | offset + BBTOB(split_hblks), |
1da177e4 | 3644 | BBTOB(hblks - split_hblks)); |
076e6acb CH |
3645 | if (error) |
3646 | goto bread_err2; | |
3647 | ||
3648 | error = xlog_bread_noalign(log, 0, | |
3649 | wrapped_hblks, hbp); | |
3650 | if (error) | |
3651 | goto bread_err2; | |
3652 | ||
fc5bc4c8 | 3653 | error = XFS_BUF_SET_PTR(hbp, offset, |
234f56ac | 3654 | BBTOB(hblks)); |
1da177e4 LT |
3655 | if (error) |
3656 | goto bread_err2; | |
1da177e4 LT |
3657 | } |
3658 | rhead = (xlog_rec_header_t *)offset; | |
3659 | error = xlog_valid_rec_header(log, rhead, | |
3660 | split_hblks ? blk_no : 0); | |
3661 | if (error) | |
3662 | goto bread_err2; | |
3663 | ||
b53e675d | 3664 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
1da177e4 LT |
3665 | blk_no += hblks; |
3666 | ||
3667 | /* Read in data for log record */ | |
3668 | if (blk_no + bblks <= log->l_logBBsize) { | |
076e6acb CH |
3669 | error = xlog_bread(log, blk_no, bblks, dbp, |
3670 | &offset); | |
1da177e4 LT |
3671 | if (error) |
3672 | goto bread_err2; | |
1da177e4 LT |
3673 | } else { |
3674 | /* This log record is split across the | |
3675 | * physical end of log */ | |
fc5bc4c8 | 3676 | offset = XFS_BUF_PTR(dbp); |
1da177e4 LT |
3677 | split_bblks = 0; |
3678 | if (blk_no != log->l_logBBsize) { | |
3679 | /* some data is before the physical | |
3680 | * end of log */ | |
3681 | ASSERT(!wrapped_hblks); | |
3682 | ASSERT(blk_no <= INT_MAX); | |
3683 | split_bblks = | |
3684 | log->l_logBBsize - (int)blk_no; | |
3685 | ASSERT(split_bblks > 0); | |
076e6acb CH |
3686 | error = xlog_bread(log, blk_no, |
3687 | split_bblks, dbp, | |
3688 | &offset); | |
3689 | if (error) | |
1da177e4 | 3690 | goto bread_err2; |
1da177e4 | 3691 | } |
076e6acb | 3692 | |
1da177e4 LT |
3693 | /* |
3694 | * Note: this black magic still works with | |
3695 | * large sector sizes (non-512) only because: | |
3696 | * - we increased the buffer size originally | |
3697 | * by 1 sector giving us enough extra space | |
3698 | * for the second read; | |
3699 | * - the log start is guaranteed to be sector | |
3700 | * aligned; | |
3701 | * - we read the log end (LR header start) | |
3702 | * _first_, then the log start (LR header end) | |
3703 | * - order is important. | |
3704 | */ | |
234f56ac | 3705 | error = XFS_BUF_SET_PTR(dbp, |
fc5bc4c8 | 3706 | offset + BBTOB(split_bblks), |
1da177e4 | 3707 | BBTOB(bblks - split_bblks)); |
234f56ac | 3708 | if (error) |
1da177e4 | 3709 | goto bread_err2; |
076e6acb CH |
3710 | |
3711 | error = xlog_bread_noalign(log, wrapped_hblks, | |
3712 | bblks - split_bblks, | |
3713 | dbp); | |
3714 | if (error) | |
3715 | goto bread_err2; | |
3716 | ||
fc5bc4c8 | 3717 | error = XFS_BUF_SET_PTR(dbp, offset, h_size); |
076e6acb CH |
3718 | if (error) |
3719 | goto bread_err2; | |
1da177e4 LT |
3720 | } |
3721 | xlog_unpack_data(rhead, offset, log); | |
3722 | if ((error = xlog_recover_process_data(log, rhash, | |
3723 | rhead, offset, pass))) | |
3724 | goto bread_err2; | |
3725 | blk_no += bblks; | |
3726 | } | |
3727 | ||
3728 | ASSERT(blk_no >= log->l_logBBsize); | |
3729 | blk_no -= log->l_logBBsize; | |
3730 | ||
3731 | /* read first part of physical log */ | |
3732 | while (blk_no < head_blk) { | |
076e6acb CH |
3733 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
3734 | if (error) | |
1da177e4 | 3735 | goto bread_err2; |
076e6acb | 3736 | |
1da177e4 LT |
3737 | rhead = (xlog_rec_header_t *)offset; |
3738 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3739 | if (error) | |
3740 | goto bread_err2; | |
076e6acb | 3741 | |
b53e675d | 3742 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
3743 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, |
3744 | &offset); | |
3745 | if (error) | |
1da177e4 | 3746 | goto bread_err2; |
076e6acb | 3747 | |
1da177e4 LT |
3748 | xlog_unpack_data(rhead, offset, log); |
3749 | if ((error = xlog_recover_process_data(log, rhash, | |
3750 | rhead, offset, pass))) | |
3751 | goto bread_err2; | |
3752 | blk_no += bblks + hblks; | |
3753 | } | |
3754 | } | |
3755 | ||
3756 | bread_err2: | |
3757 | xlog_put_bp(dbp); | |
3758 | bread_err1: | |
3759 | xlog_put_bp(hbp); | |
3760 | return error; | |
3761 | } | |
3762 | ||
3763 | /* | |
3764 | * Do the recovery of the log. We actually do this in two phases. | |
3765 | * The two passes are necessary in order to implement the function | |
3766 | * of cancelling a record written into the log. The first pass | |
3767 | * determines those things which have been cancelled, and the | |
3768 | * second pass replays log items normally except for those which | |
3769 | * have been cancelled. The handling of the replay and cancellations | |
3770 | * takes place in the log item type specific routines. | |
3771 | * | |
3772 | * The table of items which have cancel records in the log is allocated | |
3773 | * and freed at this level, since only here do we know when all of | |
3774 | * the log recovery has been completed. | |
3775 | */ | |
3776 | STATIC int | |
3777 | xlog_do_log_recovery( | |
3778 | xlog_t *log, | |
3779 | xfs_daddr_t head_blk, | |
3780 | xfs_daddr_t tail_blk) | |
3781 | { | |
3782 | int error; | |
3783 | ||
3784 | ASSERT(head_blk != tail_blk); | |
3785 | ||
3786 | /* | |
3787 | * First do a pass to find all of the cancelled buf log items. | |
3788 | * Store them in the buf_cancel_table for use in the second pass. | |
3789 | */ | |
3790 | log->l_buf_cancel_table = | |
3791 | (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE * | |
3792 | sizeof(xfs_buf_cancel_t*), | |
3793 | KM_SLEEP); | |
3794 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
3795 | XLOG_RECOVER_PASS1); | |
3796 | if (error != 0) { | |
f0e2d93c | 3797 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
3798 | log->l_buf_cancel_table = NULL; |
3799 | return error; | |
3800 | } | |
3801 | /* | |
3802 | * Then do a second pass to actually recover the items in the log. | |
3803 | * When it is complete free the table of buf cancel items. | |
3804 | */ | |
3805 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
3806 | XLOG_RECOVER_PASS2); | |
3807 | #ifdef DEBUG | |
6d192a9b | 3808 | if (!error) { |
1da177e4 LT |
3809 | int i; |
3810 | ||
3811 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
3812 | ASSERT(log->l_buf_cancel_table[i] == NULL); | |
3813 | } | |
3814 | #endif /* DEBUG */ | |
3815 | ||
f0e2d93c | 3816 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
3817 | log->l_buf_cancel_table = NULL; |
3818 | ||
3819 | return error; | |
3820 | } | |
3821 | ||
3822 | /* | |
3823 | * Do the actual recovery | |
3824 | */ | |
3825 | STATIC int | |
3826 | xlog_do_recover( | |
3827 | xlog_t *log, | |
3828 | xfs_daddr_t head_blk, | |
3829 | xfs_daddr_t tail_blk) | |
3830 | { | |
3831 | int error; | |
3832 | xfs_buf_t *bp; | |
3833 | xfs_sb_t *sbp; | |
3834 | ||
3835 | /* | |
3836 | * First replay the images in the log. | |
3837 | */ | |
3838 | error = xlog_do_log_recovery(log, head_blk, tail_blk); | |
3839 | if (error) { | |
3840 | return error; | |
3841 | } | |
3842 | ||
3843 | XFS_bflush(log->l_mp->m_ddev_targp); | |
3844 | ||
3845 | /* | |
3846 | * If IO errors happened during recovery, bail out. | |
3847 | */ | |
3848 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) { | |
3849 | return (EIO); | |
3850 | } | |
3851 | ||
3852 | /* | |
3853 | * We now update the tail_lsn since much of the recovery has completed | |
3854 | * and there may be space available to use. If there were no extent | |
3855 | * or iunlinks, we can free up the entire log and set the tail_lsn to | |
3856 | * be the last_sync_lsn. This was set in xlog_find_tail to be the | |
3857 | * lsn of the last known good LR on disk. If there are extent frees | |
3858 | * or iunlinks they will have some entries in the AIL; so we look at | |
3859 | * the AIL to determine how to set the tail_lsn. | |
3860 | */ | |
3861 | xlog_assign_tail_lsn(log->l_mp); | |
3862 | ||
3863 | /* | |
3864 | * Now that we've finished replaying all buffer and inode | |
3865 | * updates, re-read in the superblock. | |
3866 | */ | |
3867 | bp = xfs_getsb(log->l_mp, 0); | |
3868 | XFS_BUF_UNDONE(bp); | |
bebf963f LM |
3869 | ASSERT(!(XFS_BUF_ISWRITE(bp))); |
3870 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); | |
1da177e4 | 3871 | XFS_BUF_READ(bp); |
bebf963f | 3872 | XFS_BUF_UNASYNC(bp); |
1da177e4 | 3873 | xfsbdstrat(log->l_mp, bp); |
d64e31a2 DC |
3874 | error = xfs_iowait(bp); |
3875 | if (error) { | |
1da177e4 LT |
3876 | xfs_ioerror_alert("xlog_do_recover", |
3877 | log->l_mp, bp, XFS_BUF_ADDR(bp)); | |
3878 | ASSERT(0); | |
3879 | xfs_buf_relse(bp); | |
3880 | return error; | |
3881 | } | |
3882 | ||
3883 | /* Convert superblock from on-disk format */ | |
3884 | sbp = &log->l_mp->m_sb; | |
2bdf7cd0 | 3885 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 3886 | ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC); |
62118709 | 3887 | ASSERT(xfs_sb_good_version(sbp)); |
1da177e4 LT |
3888 | xfs_buf_relse(bp); |
3889 | ||
5478eead LM |
3890 | /* We've re-read the superblock so re-initialize per-cpu counters */ |
3891 | xfs_icsb_reinit_counters(log->l_mp); | |
3892 | ||
1da177e4 LT |
3893 | xlog_recover_check_summary(log); |
3894 | ||
3895 | /* Normal transactions can now occur */ | |
3896 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
3897 | return 0; | |
3898 | } | |
3899 | ||
3900 | /* | |
3901 | * Perform recovery and re-initialize some log variables in xlog_find_tail. | |
3902 | * | |
3903 | * Return error or zero. | |
3904 | */ | |
3905 | int | |
3906 | xlog_recover( | |
65be6054 | 3907 | xlog_t *log) |
1da177e4 LT |
3908 | { |
3909 | xfs_daddr_t head_blk, tail_blk; | |
3910 | int error; | |
3911 | ||
3912 | /* find the tail of the log */ | |
65be6054 | 3913 | if ((error = xlog_find_tail(log, &head_blk, &tail_blk))) |
1da177e4 LT |
3914 | return error; |
3915 | ||
3916 | if (tail_blk != head_blk) { | |
3917 | /* There used to be a comment here: | |
3918 | * | |
3919 | * disallow recovery on read-only mounts. note -- mount | |
3920 | * checks for ENOSPC and turns it into an intelligent | |
3921 | * error message. | |
3922 | * ...but this is no longer true. Now, unless you specify | |
3923 | * NORECOVERY (in which case this function would never be | |
3924 | * called), we just go ahead and recover. We do this all | |
3925 | * under the vfs layer, so we can get away with it unless | |
3926 | * the device itself is read-only, in which case we fail. | |
3927 | */ | |
3a02ee18 | 3928 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
1da177e4 LT |
3929 | return error; |
3930 | } | |
3931 | ||
3932 | cmn_err(CE_NOTE, | |
fc1f8c1c NS |
3933 | "Starting XFS recovery on filesystem: %s (logdev: %s)", |
3934 | log->l_mp->m_fsname, log->l_mp->m_logname ? | |
3935 | log->l_mp->m_logname : "internal"); | |
1da177e4 LT |
3936 | |
3937 | error = xlog_do_recover(log, head_blk, tail_blk); | |
3938 | log->l_flags |= XLOG_RECOVERY_NEEDED; | |
3939 | } | |
3940 | return error; | |
3941 | } | |
3942 | ||
3943 | /* | |
3944 | * In the first part of recovery we replay inodes and buffers and build | |
3945 | * up the list of extent free items which need to be processed. Here | |
3946 | * we process the extent free items and clean up the on disk unlinked | |
3947 | * inode lists. This is separated from the first part of recovery so | |
3948 | * that the root and real-time bitmap inodes can be read in from disk in | |
3949 | * between the two stages. This is necessary so that we can free space | |
3950 | * in the real-time portion of the file system. | |
3951 | */ | |
3952 | int | |
3953 | xlog_recover_finish( | |
4249023a | 3954 | xlog_t *log) |
1da177e4 LT |
3955 | { |
3956 | /* | |
3957 | * Now we're ready to do the transactions needed for the | |
3958 | * rest of recovery. Start with completing all the extent | |
3959 | * free intent records and then process the unlinked inode | |
3960 | * lists. At this point, we essentially run in normal mode | |
3961 | * except that we're still performing recovery actions | |
3962 | * rather than accepting new requests. | |
3963 | */ | |
3964 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { | |
3c1e2bbe DC |
3965 | int error; |
3966 | error = xlog_recover_process_efis(log); | |
3967 | if (error) { | |
3968 | cmn_err(CE_ALERT, | |
3969 | "Failed to recover EFIs on filesystem: %s", | |
3970 | log->l_mp->m_fsname); | |
3971 | return error; | |
3972 | } | |
1da177e4 LT |
3973 | /* |
3974 | * Sync the log to get all the EFIs out of the AIL. | |
3975 | * This isn't absolutely necessary, but it helps in | |
3976 | * case the unlink transactions would have problems | |
3977 | * pushing the EFIs out of the way. | |
3978 | */ | |
a14a348b | 3979 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
1da177e4 | 3980 | |
4249023a | 3981 | xlog_recover_process_iunlinks(log); |
1da177e4 LT |
3982 | |
3983 | xlog_recover_check_summary(log); | |
3984 | ||
3985 | cmn_err(CE_NOTE, | |
fc1f8c1c NS |
3986 | "Ending XFS recovery on filesystem: %s (logdev: %s)", |
3987 | log->l_mp->m_fsname, log->l_mp->m_logname ? | |
3988 | log->l_mp->m_logname : "internal"); | |
1da177e4 LT |
3989 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
3990 | } else { | |
3991 | cmn_err(CE_DEBUG, | |
b6574520 | 3992 | "!Ending clean XFS mount for filesystem: %s\n", |
1da177e4 LT |
3993 | log->l_mp->m_fsname); |
3994 | } | |
3995 | return 0; | |
3996 | } | |
3997 | ||
3998 | ||
3999 | #if defined(DEBUG) | |
4000 | /* | |
4001 | * Read all of the agf and agi counters and check that they | |
4002 | * are consistent with the superblock counters. | |
4003 | */ | |
4004 | void | |
4005 | xlog_recover_check_summary( | |
4006 | xlog_t *log) | |
4007 | { | |
4008 | xfs_mount_t *mp; | |
4009 | xfs_agf_t *agfp; | |
1da177e4 LT |
4010 | xfs_buf_t *agfbp; |
4011 | xfs_buf_t *agibp; | |
1da177e4 LT |
4012 | xfs_buf_t *sbbp; |
4013 | #ifdef XFS_LOUD_RECOVERY | |
4014 | xfs_sb_t *sbp; | |
4015 | #endif | |
4016 | xfs_agnumber_t agno; | |
4017 | __uint64_t freeblks; | |
4018 | __uint64_t itotal; | |
4019 | __uint64_t ifree; | |
5e1be0fb | 4020 | int error; |
1da177e4 LT |
4021 | |
4022 | mp = log->l_mp; | |
4023 | ||
4024 | freeblks = 0LL; | |
4025 | itotal = 0LL; | |
4026 | ifree = 0LL; | |
4027 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
4805621a FCH |
4028 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
4029 | if (error) { | |
4030 | xfs_fs_cmn_err(CE_ALERT, mp, | |
4031 | "xlog_recover_check_summary(agf)" | |
4032 | "agf read failed agno %d error %d", | |
4033 | agno, error); | |
4034 | } else { | |
4035 | agfp = XFS_BUF_TO_AGF(agfbp); | |
4036 | freeblks += be32_to_cpu(agfp->agf_freeblks) + | |
4037 | be32_to_cpu(agfp->agf_flcount); | |
4038 | xfs_buf_relse(agfbp); | |
1da177e4 | 4039 | } |
1da177e4 | 4040 | |
5e1be0fb CH |
4041 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
4042 | if (!error) { | |
4043 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); | |
16259e7d | 4044 | |
5e1be0fb CH |
4045 | itotal += be32_to_cpu(agi->agi_count); |
4046 | ifree += be32_to_cpu(agi->agi_freecount); | |
4047 | xfs_buf_relse(agibp); | |
4048 | } | |
1da177e4 LT |
4049 | } |
4050 | ||
4051 | sbbp = xfs_getsb(mp, 0); | |
4052 | #ifdef XFS_LOUD_RECOVERY | |
4053 | sbp = &mp->m_sb; | |
2bdf7cd0 | 4054 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp)); |
1da177e4 LT |
4055 | cmn_err(CE_NOTE, |
4056 | "xlog_recover_check_summary: sb_icount %Lu itotal %Lu", | |
4057 | sbp->sb_icount, itotal); | |
4058 | cmn_err(CE_NOTE, | |
4059 | "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu", | |
4060 | sbp->sb_ifree, ifree); | |
4061 | cmn_err(CE_NOTE, | |
4062 | "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu", | |
4063 | sbp->sb_fdblocks, freeblks); | |
4064 | #if 0 | |
4065 | /* | |
4066 | * This is turned off until I account for the allocation | |
4067 | * btree blocks which live in free space. | |
4068 | */ | |
4069 | ASSERT(sbp->sb_icount == itotal); | |
4070 | ASSERT(sbp->sb_ifree == ifree); | |
4071 | ASSERT(sbp->sb_fdblocks == freeblks); | |
4072 | #endif | |
4073 | #endif | |
4074 | xfs_buf_relse(sbbp); | |
4075 | } | |
4076 | #endif /* DEBUG */ |