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_mount.h" |
28 | #include "xfs_error.h" | |
29 | #include "xfs_bmap_btree.h" | |
a844f451 NS |
30 | #include "xfs_alloc_btree.h" |
31 | #include "xfs_ialloc_btree.h" | |
ee1a47ab | 32 | #include "xfs_btree.h" |
1da177e4 | 33 | #include "xfs_dinode.h" |
1da177e4 | 34 | #include "xfs_inode.h" |
a844f451 | 35 | #include "xfs_inode_item.h" |
a844f451 | 36 | #include "xfs_alloc.h" |
1da177e4 LT |
37 | #include "xfs_ialloc.h" |
38 | #include "xfs_log_priv.h" | |
39 | #include "xfs_buf_item.h" | |
1da177e4 LT |
40 | #include "xfs_log_recover.h" |
41 | #include "xfs_extfree_item.h" | |
42 | #include "xfs_trans_priv.h" | |
1da177e4 | 43 | #include "xfs_quota.h" |
43355099 | 44 | #include "xfs_utils.h" |
0e446be4 | 45 | #include "xfs_cksum.h" |
0b1b213f | 46 | #include "xfs_trace.h" |
33479e05 | 47 | #include "xfs_icache.h" |
f948dd76 | 48 | #include "xfs_symlink.h" |
1da177e4 | 49 | |
9a8d2fdb MT |
50 | STATIC int |
51 | xlog_find_zeroed( | |
52 | struct xlog *, | |
53 | xfs_daddr_t *); | |
54 | STATIC int | |
55 | xlog_clear_stale_blocks( | |
56 | struct xlog *, | |
57 | xfs_lsn_t); | |
1da177e4 | 58 | #if defined(DEBUG) |
9a8d2fdb MT |
59 | STATIC void |
60 | xlog_recover_check_summary( | |
61 | struct xlog *); | |
1da177e4 LT |
62 | #else |
63 | #define xlog_recover_check_summary(log) | |
1da177e4 LT |
64 | #endif |
65 | ||
d5689eaa CH |
66 | /* |
67 | * This structure is used during recovery to record the buf log items which | |
68 | * have been canceled and should not be replayed. | |
69 | */ | |
70 | struct xfs_buf_cancel { | |
71 | xfs_daddr_t bc_blkno; | |
72 | uint bc_len; | |
73 | int bc_refcount; | |
74 | struct list_head bc_list; | |
75 | }; | |
76 | ||
1da177e4 LT |
77 | /* |
78 | * Sector aligned buffer routines for buffer create/read/write/access | |
79 | */ | |
80 | ||
ff30a622 AE |
81 | /* |
82 | * Verify the given count of basic blocks is valid number of blocks | |
83 | * to specify for an operation involving the given XFS log buffer. | |
84 | * Returns nonzero if the count is valid, 0 otherwise. | |
85 | */ | |
86 | ||
87 | static inline int | |
88 | xlog_buf_bbcount_valid( | |
9a8d2fdb | 89 | struct xlog *log, |
ff30a622 AE |
90 | int bbcount) |
91 | { | |
92 | return bbcount > 0 && bbcount <= log->l_logBBsize; | |
93 | } | |
94 | ||
36adecff AE |
95 | /* |
96 | * Allocate a buffer to hold log data. The buffer needs to be able | |
97 | * to map to a range of nbblks basic blocks at any valid (basic | |
98 | * block) offset within the log. | |
99 | */ | |
5d77c0dc | 100 | STATIC xfs_buf_t * |
1da177e4 | 101 | xlog_get_bp( |
9a8d2fdb | 102 | struct xlog *log, |
3228149c | 103 | int nbblks) |
1da177e4 | 104 | { |
c8da0faf CH |
105 | struct xfs_buf *bp; |
106 | ||
ff30a622 | 107 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 108 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
109 | nbblks); |
110 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
111 | return NULL; |
112 | } | |
1da177e4 | 113 | |
36adecff AE |
114 | /* |
115 | * We do log I/O in units of log sectors (a power-of-2 | |
116 | * multiple of the basic block size), so we round up the | |
25985edc | 117 | * requested size to accommodate the basic blocks required |
36adecff AE |
118 | * for complete log sectors. |
119 | * | |
120 | * In addition, the buffer may be used for a non-sector- | |
121 | * aligned block offset, in which case an I/O of the | |
122 | * requested size could extend beyond the end of the | |
123 | * buffer. If the requested size is only 1 basic block it | |
124 | * will never straddle a sector boundary, so this won't be | |
125 | * an issue. Nor will this be a problem if the log I/O is | |
126 | * done in basic blocks (sector size 1). But otherwise we | |
127 | * extend the buffer by one extra log sector to ensure | |
25985edc | 128 | * there's space to accommodate this possibility. |
36adecff | 129 | */ |
69ce58f0 AE |
130 | if (nbblks > 1 && log->l_sectBBsize > 1) |
131 | nbblks += log->l_sectBBsize; | |
132 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
36adecff | 133 | |
e70b73f8 | 134 | bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0); |
c8da0faf CH |
135 | if (bp) |
136 | xfs_buf_unlock(bp); | |
137 | return bp; | |
1da177e4 LT |
138 | } |
139 | ||
5d77c0dc | 140 | STATIC void |
1da177e4 LT |
141 | xlog_put_bp( |
142 | xfs_buf_t *bp) | |
143 | { | |
144 | xfs_buf_free(bp); | |
145 | } | |
146 | ||
48389ef1 AE |
147 | /* |
148 | * Return the address of the start of the given block number's data | |
149 | * in a log buffer. The buffer covers a log sector-aligned region. | |
150 | */ | |
076e6acb CH |
151 | STATIC xfs_caddr_t |
152 | xlog_align( | |
9a8d2fdb | 153 | struct xlog *log, |
076e6acb CH |
154 | xfs_daddr_t blk_no, |
155 | int nbblks, | |
9a8d2fdb | 156 | struct xfs_buf *bp) |
076e6acb | 157 | { |
fdc07f44 | 158 | xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1); |
076e6acb | 159 | |
4e94b71b | 160 | ASSERT(offset + nbblks <= bp->b_length); |
62926044 | 161 | return bp->b_addr + BBTOB(offset); |
076e6acb CH |
162 | } |
163 | ||
1da177e4 LT |
164 | |
165 | /* | |
166 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. | |
167 | */ | |
076e6acb CH |
168 | STATIC int |
169 | xlog_bread_noalign( | |
9a8d2fdb | 170 | struct xlog *log, |
1da177e4 LT |
171 | xfs_daddr_t blk_no, |
172 | int nbblks, | |
9a8d2fdb | 173 | struct xfs_buf *bp) |
1da177e4 LT |
174 | { |
175 | int error; | |
176 | ||
ff30a622 | 177 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 178 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
179 | nbblks); |
180 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
181 | return EFSCORRUPTED; |
182 | } | |
183 | ||
69ce58f0 AE |
184 | blk_no = round_down(blk_no, log->l_sectBBsize); |
185 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
186 | |
187 | ASSERT(nbblks > 0); | |
4e94b71b | 188 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
189 | |
190 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
191 | XFS_BUF_READ(bp); | |
aa0e8833 | 192 | bp->b_io_length = nbblks; |
0e95f19a | 193 | bp->b_error = 0; |
1da177e4 LT |
194 | |
195 | xfsbdstrat(log->l_mp, bp); | |
1a1a3e97 | 196 | error = xfs_buf_iowait(bp); |
d64e31a2 | 197 | if (error) |
901796af | 198 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
199 | return error; |
200 | } | |
201 | ||
076e6acb CH |
202 | STATIC int |
203 | xlog_bread( | |
9a8d2fdb | 204 | struct xlog *log, |
076e6acb CH |
205 | xfs_daddr_t blk_no, |
206 | int nbblks, | |
9a8d2fdb | 207 | struct xfs_buf *bp, |
076e6acb CH |
208 | xfs_caddr_t *offset) |
209 | { | |
210 | int error; | |
211 | ||
212 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
213 | if (error) | |
214 | return error; | |
215 | ||
216 | *offset = xlog_align(log, blk_no, nbblks, bp); | |
217 | return 0; | |
218 | } | |
219 | ||
44396476 DC |
220 | /* |
221 | * Read at an offset into the buffer. Returns with the buffer in it's original | |
222 | * state regardless of the result of the read. | |
223 | */ | |
224 | STATIC int | |
225 | xlog_bread_offset( | |
9a8d2fdb | 226 | struct xlog *log, |
44396476 DC |
227 | xfs_daddr_t blk_no, /* block to read from */ |
228 | int nbblks, /* blocks to read */ | |
9a8d2fdb | 229 | struct xfs_buf *bp, |
44396476 DC |
230 | xfs_caddr_t offset) |
231 | { | |
62926044 | 232 | xfs_caddr_t orig_offset = bp->b_addr; |
4e94b71b | 233 | int orig_len = BBTOB(bp->b_length); |
44396476 DC |
234 | int error, error2; |
235 | ||
02fe03d9 | 236 | error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks)); |
44396476 DC |
237 | if (error) |
238 | return error; | |
239 | ||
240 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
241 | ||
242 | /* must reset buffer pointer even on error */ | |
02fe03d9 | 243 | error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len); |
44396476 DC |
244 | if (error) |
245 | return error; | |
246 | return error2; | |
247 | } | |
248 | ||
1da177e4 LT |
249 | /* |
250 | * Write out the buffer at the given block for the given number of blocks. | |
251 | * The buffer is kept locked across the write and is returned locked. | |
252 | * This can only be used for synchronous log writes. | |
253 | */ | |
ba0f32d4 | 254 | STATIC int |
1da177e4 | 255 | xlog_bwrite( |
9a8d2fdb | 256 | struct xlog *log, |
1da177e4 LT |
257 | xfs_daddr_t blk_no, |
258 | int nbblks, | |
9a8d2fdb | 259 | struct xfs_buf *bp) |
1da177e4 LT |
260 | { |
261 | int error; | |
262 | ||
ff30a622 | 263 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 264 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
265 | nbblks); |
266 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
267 | return EFSCORRUPTED; |
268 | } | |
269 | ||
69ce58f0 AE |
270 | blk_no = round_down(blk_no, log->l_sectBBsize); |
271 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
272 | |
273 | ASSERT(nbblks > 0); | |
4e94b71b | 274 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
275 | |
276 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
277 | XFS_BUF_ZEROFLAGS(bp); | |
72790aa1 | 278 | xfs_buf_hold(bp); |
0c842ad4 | 279 | xfs_buf_lock(bp); |
aa0e8833 | 280 | bp->b_io_length = nbblks; |
0e95f19a | 281 | bp->b_error = 0; |
1da177e4 | 282 | |
c2b006c1 | 283 | error = xfs_bwrite(bp); |
901796af CH |
284 | if (error) |
285 | xfs_buf_ioerror_alert(bp, __func__); | |
c2b006c1 | 286 | xfs_buf_relse(bp); |
1da177e4 LT |
287 | return error; |
288 | } | |
289 | ||
1da177e4 LT |
290 | #ifdef DEBUG |
291 | /* | |
292 | * dump debug superblock and log record information | |
293 | */ | |
294 | STATIC void | |
295 | xlog_header_check_dump( | |
296 | xfs_mount_t *mp, | |
297 | xlog_rec_header_t *head) | |
298 | { | |
a0fa2b67 | 299 | xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n", |
03daa57c | 300 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); |
a0fa2b67 | 301 | xfs_debug(mp, " log : uuid = %pU, fmt = %d\n", |
03daa57c | 302 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); |
1da177e4 LT |
303 | } |
304 | #else | |
305 | #define xlog_header_check_dump(mp, head) | |
306 | #endif | |
307 | ||
308 | /* | |
309 | * check log record header for recovery | |
310 | */ | |
311 | STATIC int | |
312 | xlog_header_check_recover( | |
313 | xfs_mount_t *mp, | |
314 | xlog_rec_header_t *head) | |
315 | { | |
69ef921b | 316 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 LT |
317 | |
318 | /* | |
319 | * IRIX doesn't write the h_fmt field and leaves it zeroed | |
320 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover | |
321 | * a dirty log created in IRIX. | |
322 | */ | |
69ef921b | 323 | if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) { |
a0fa2b67 DC |
324 | xfs_warn(mp, |
325 | "dirty log written in incompatible format - can't recover"); | |
1da177e4 LT |
326 | xlog_header_check_dump(mp, head); |
327 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", | |
328 | XFS_ERRLEVEL_HIGH, mp); | |
329 | return XFS_ERROR(EFSCORRUPTED); | |
330 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
a0fa2b67 DC |
331 | xfs_warn(mp, |
332 | "dirty log entry has mismatched uuid - can't recover"); | |
1da177e4 LT |
333 | xlog_header_check_dump(mp, head); |
334 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", | |
335 | XFS_ERRLEVEL_HIGH, mp); | |
336 | return XFS_ERROR(EFSCORRUPTED); | |
337 | } | |
338 | return 0; | |
339 | } | |
340 | ||
341 | /* | |
342 | * read the head block of the log and check the header | |
343 | */ | |
344 | STATIC int | |
345 | xlog_header_check_mount( | |
346 | xfs_mount_t *mp, | |
347 | xlog_rec_header_t *head) | |
348 | { | |
69ef921b | 349 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 LT |
350 | |
351 | if (uuid_is_nil(&head->h_fs_uuid)) { | |
352 | /* | |
353 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If | |
354 | * h_fs_uuid is nil, we assume this log was last mounted | |
355 | * by IRIX and continue. | |
356 | */ | |
a0fa2b67 | 357 | xfs_warn(mp, "nil uuid in log - IRIX style log"); |
1da177e4 | 358 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
a0fa2b67 | 359 | xfs_warn(mp, "log has mismatched uuid - can't recover"); |
1da177e4 LT |
360 | xlog_header_check_dump(mp, head); |
361 | XFS_ERROR_REPORT("xlog_header_check_mount", | |
362 | XFS_ERRLEVEL_HIGH, mp); | |
363 | return XFS_ERROR(EFSCORRUPTED); | |
364 | } | |
365 | return 0; | |
366 | } | |
367 | ||
368 | STATIC void | |
369 | xlog_recover_iodone( | |
370 | struct xfs_buf *bp) | |
371 | { | |
5a52c2a5 | 372 | if (bp->b_error) { |
1da177e4 LT |
373 | /* |
374 | * We're not going to bother about retrying | |
375 | * this during recovery. One strike! | |
376 | */ | |
901796af | 377 | xfs_buf_ioerror_alert(bp, __func__); |
ebad861b DC |
378 | xfs_force_shutdown(bp->b_target->bt_mount, |
379 | SHUTDOWN_META_IO_ERROR); | |
1da177e4 | 380 | } |
cb669ca5 | 381 | bp->b_iodone = NULL; |
1a1a3e97 | 382 | xfs_buf_ioend(bp, 0); |
1da177e4 LT |
383 | } |
384 | ||
385 | /* | |
386 | * This routine finds (to an approximation) the first block in the physical | |
387 | * log which contains the given cycle. It uses a binary search algorithm. | |
388 | * Note that the algorithm can not be perfect because the disk will not | |
389 | * necessarily be perfect. | |
390 | */ | |
a8272ce0 | 391 | STATIC int |
1da177e4 | 392 | xlog_find_cycle_start( |
9a8d2fdb MT |
393 | struct xlog *log, |
394 | struct xfs_buf *bp, | |
1da177e4 LT |
395 | xfs_daddr_t first_blk, |
396 | xfs_daddr_t *last_blk, | |
397 | uint cycle) | |
398 | { | |
399 | xfs_caddr_t offset; | |
400 | xfs_daddr_t mid_blk; | |
e3bb2e30 | 401 | xfs_daddr_t end_blk; |
1da177e4 LT |
402 | uint mid_cycle; |
403 | int error; | |
404 | ||
e3bb2e30 AE |
405 | end_blk = *last_blk; |
406 | mid_blk = BLK_AVG(first_blk, end_blk); | |
407 | while (mid_blk != first_blk && mid_blk != end_blk) { | |
076e6acb CH |
408 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
409 | if (error) | |
1da177e4 | 410 | return error; |
03bea6fe | 411 | mid_cycle = xlog_get_cycle(offset); |
e3bb2e30 AE |
412 | if (mid_cycle == cycle) |
413 | end_blk = mid_blk; /* last_half_cycle == mid_cycle */ | |
414 | else | |
415 | first_blk = mid_blk; /* first_half_cycle == mid_cycle */ | |
416 | mid_blk = BLK_AVG(first_blk, end_blk); | |
1da177e4 | 417 | } |
e3bb2e30 AE |
418 | ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) || |
419 | (mid_blk == end_blk && mid_blk-1 == first_blk)); | |
420 | ||
421 | *last_blk = end_blk; | |
1da177e4 LT |
422 | |
423 | return 0; | |
424 | } | |
425 | ||
426 | /* | |
3f943d85 AE |
427 | * Check that a range of blocks does not contain stop_on_cycle_no. |
428 | * Fill in *new_blk with the block offset where such a block is | |
429 | * found, or with -1 (an invalid block number) if there is no such | |
430 | * block in the range. The scan needs to occur from front to back | |
431 | * and the pointer into the region must be updated since a later | |
432 | * routine will need to perform another test. | |
1da177e4 LT |
433 | */ |
434 | STATIC int | |
435 | xlog_find_verify_cycle( | |
9a8d2fdb | 436 | struct xlog *log, |
1da177e4 LT |
437 | xfs_daddr_t start_blk, |
438 | int nbblks, | |
439 | uint stop_on_cycle_no, | |
440 | xfs_daddr_t *new_blk) | |
441 | { | |
442 | xfs_daddr_t i, j; | |
443 | uint cycle; | |
444 | xfs_buf_t *bp; | |
445 | xfs_daddr_t bufblks; | |
446 | xfs_caddr_t buf = NULL; | |
447 | int error = 0; | |
448 | ||
6881a229 AE |
449 | /* |
450 | * Greedily allocate a buffer big enough to handle the full | |
451 | * range of basic blocks we'll be examining. If that fails, | |
452 | * try a smaller size. We need to be able to read at least | |
453 | * a log sector, or we're out of luck. | |
454 | */ | |
1da177e4 | 455 | bufblks = 1 << ffs(nbblks); |
81158e0c DC |
456 | while (bufblks > log->l_logBBsize) |
457 | bufblks >>= 1; | |
1da177e4 | 458 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1da177e4 | 459 | bufblks >>= 1; |
69ce58f0 | 460 | if (bufblks < log->l_sectBBsize) |
1da177e4 LT |
461 | return ENOMEM; |
462 | } | |
463 | ||
464 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { | |
465 | int bcount; | |
466 | ||
467 | bcount = min(bufblks, (start_blk + nbblks - i)); | |
468 | ||
076e6acb CH |
469 | error = xlog_bread(log, i, bcount, bp, &buf); |
470 | if (error) | |
1da177e4 LT |
471 | goto out; |
472 | ||
1da177e4 | 473 | for (j = 0; j < bcount; j++) { |
03bea6fe | 474 | cycle = xlog_get_cycle(buf); |
1da177e4 LT |
475 | if (cycle == stop_on_cycle_no) { |
476 | *new_blk = i+j; | |
477 | goto out; | |
478 | } | |
479 | ||
480 | buf += BBSIZE; | |
481 | } | |
482 | } | |
483 | ||
484 | *new_blk = -1; | |
485 | ||
486 | out: | |
487 | xlog_put_bp(bp); | |
488 | return error; | |
489 | } | |
490 | ||
491 | /* | |
492 | * Potentially backup over partial log record write. | |
493 | * | |
494 | * In the typical case, last_blk is the number of the block directly after | |
495 | * a good log record. Therefore, we subtract one to get the block number | |
496 | * of the last block in the given buffer. extra_bblks contains the number | |
497 | * of blocks we would have read on a previous read. This happens when the | |
498 | * last log record is split over the end of the physical log. | |
499 | * | |
500 | * extra_bblks is the number of blocks potentially verified on a previous | |
501 | * call to this routine. | |
502 | */ | |
503 | STATIC int | |
504 | xlog_find_verify_log_record( | |
9a8d2fdb | 505 | struct xlog *log, |
1da177e4 LT |
506 | xfs_daddr_t start_blk, |
507 | xfs_daddr_t *last_blk, | |
508 | int extra_bblks) | |
509 | { | |
510 | xfs_daddr_t i; | |
511 | xfs_buf_t *bp; | |
512 | xfs_caddr_t offset = NULL; | |
513 | xlog_rec_header_t *head = NULL; | |
514 | int error = 0; | |
515 | int smallmem = 0; | |
516 | int num_blks = *last_blk - start_blk; | |
517 | int xhdrs; | |
518 | ||
519 | ASSERT(start_blk != 0 || *last_blk != start_blk); | |
520 | ||
521 | if (!(bp = xlog_get_bp(log, num_blks))) { | |
522 | if (!(bp = xlog_get_bp(log, 1))) | |
523 | return ENOMEM; | |
524 | smallmem = 1; | |
525 | } else { | |
076e6acb CH |
526 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
527 | if (error) | |
1da177e4 | 528 | goto out; |
1da177e4 LT |
529 | offset += ((num_blks - 1) << BBSHIFT); |
530 | } | |
531 | ||
532 | for (i = (*last_blk) - 1; i >= 0; i--) { | |
533 | if (i < start_blk) { | |
534 | /* valid log record not found */ | |
a0fa2b67 DC |
535 | xfs_warn(log->l_mp, |
536 | "Log inconsistent (didn't find previous header)"); | |
1da177e4 LT |
537 | ASSERT(0); |
538 | error = XFS_ERROR(EIO); | |
539 | goto out; | |
540 | } | |
541 | ||
542 | if (smallmem) { | |
076e6acb CH |
543 | error = xlog_bread(log, i, 1, bp, &offset); |
544 | if (error) | |
1da177e4 | 545 | goto out; |
1da177e4 LT |
546 | } |
547 | ||
548 | head = (xlog_rec_header_t *)offset; | |
549 | ||
69ef921b | 550 | if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
1da177e4 LT |
551 | break; |
552 | ||
553 | if (!smallmem) | |
554 | offset -= BBSIZE; | |
555 | } | |
556 | ||
557 | /* | |
558 | * We hit the beginning of the physical log & still no header. Return | |
559 | * to caller. If caller can handle a return of -1, then this routine | |
560 | * will be called again for the end of the physical log. | |
561 | */ | |
562 | if (i == -1) { | |
563 | error = -1; | |
564 | goto out; | |
565 | } | |
566 | ||
567 | /* | |
568 | * We have the final block of the good log (the first block | |
569 | * of the log record _before_ the head. So we check the uuid. | |
570 | */ | |
571 | if ((error = xlog_header_check_mount(log->l_mp, head))) | |
572 | goto out; | |
573 | ||
574 | /* | |
575 | * We may have found a log record header before we expected one. | |
576 | * last_blk will be the 1st block # with a given cycle #. We may end | |
577 | * up reading an entire log record. In this case, we don't want to | |
578 | * reset last_blk. Only when last_blk points in the middle of a log | |
579 | * record do we update last_blk. | |
580 | */ | |
62118709 | 581 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d | 582 | uint h_size = be32_to_cpu(head->h_size); |
1da177e4 LT |
583 | |
584 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; | |
585 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
586 | xhdrs++; | |
587 | } else { | |
588 | xhdrs = 1; | |
589 | } | |
590 | ||
b53e675d CH |
591 | if (*last_blk - i + extra_bblks != |
592 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) | |
1da177e4 LT |
593 | *last_blk = i; |
594 | ||
595 | out: | |
596 | xlog_put_bp(bp); | |
597 | return error; | |
598 | } | |
599 | ||
600 | /* | |
601 | * Head is defined to be the point of the log where the next log write | |
602 | * write could go. This means that incomplete LR writes at the end are | |
603 | * eliminated when calculating the head. We aren't guaranteed that previous | |
604 | * LR have complete transactions. We only know that a cycle number of | |
605 | * current cycle number -1 won't be present in the log if we start writing | |
606 | * from our current block number. | |
607 | * | |
608 | * last_blk contains the block number of the first block with a given | |
609 | * cycle number. | |
610 | * | |
611 | * Return: zero if normal, non-zero if error. | |
612 | */ | |
ba0f32d4 | 613 | STATIC int |
1da177e4 | 614 | xlog_find_head( |
9a8d2fdb | 615 | struct xlog *log, |
1da177e4 LT |
616 | xfs_daddr_t *return_head_blk) |
617 | { | |
618 | xfs_buf_t *bp; | |
619 | xfs_caddr_t offset; | |
620 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; | |
621 | int num_scan_bblks; | |
622 | uint first_half_cycle, last_half_cycle; | |
623 | uint stop_on_cycle; | |
624 | int error, log_bbnum = log->l_logBBsize; | |
625 | ||
626 | /* Is the end of the log device zeroed? */ | |
627 | if ((error = xlog_find_zeroed(log, &first_blk)) == -1) { | |
628 | *return_head_blk = first_blk; | |
629 | ||
630 | /* Is the whole lot zeroed? */ | |
631 | if (!first_blk) { | |
632 | /* Linux XFS shouldn't generate totally zeroed logs - | |
633 | * mkfs etc write a dummy unmount record to a fresh | |
634 | * log so we can store the uuid in there | |
635 | */ | |
a0fa2b67 | 636 | xfs_warn(log->l_mp, "totally zeroed log"); |
1da177e4 LT |
637 | } |
638 | ||
639 | return 0; | |
640 | } else if (error) { | |
a0fa2b67 | 641 | xfs_warn(log->l_mp, "empty log check failed"); |
1da177e4 LT |
642 | return error; |
643 | } | |
644 | ||
645 | first_blk = 0; /* get cycle # of 1st block */ | |
646 | bp = xlog_get_bp(log, 1); | |
647 | if (!bp) | |
648 | return ENOMEM; | |
076e6acb CH |
649 | |
650 | error = xlog_bread(log, 0, 1, bp, &offset); | |
651 | if (error) | |
1da177e4 | 652 | goto bp_err; |
076e6acb | 653 | |
03bea6fe | 654 | first_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
655 | |
656 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ | |
076e6acb CH |
657 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
658 | if (error) | |
1da177e4 | 659 | goto bp_err; |
076e6acb | 660 | |
03bea6fe | 661 | last_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
662 | ASSERT(last_half_cycle != 0); |
663 | ||
664 | /* | |
665 | * If the 1st half cycle number is equal to the last half cycle number, | |
666 | * then the entire log is stamped with the same cycle number. In this | |
667 | * case, head_blk can't be set to zero (which makes sense). The below | |
668 | * math doesn't work out properly with head_blk equal to zero. Instead, | |
669 | * we set it to log_bbnum which is an invalid block number, but this | |
670 | * value makes the math correct. If head_blk doesn't changed through | |
671 | * all the tests below, *head_blk is set to zero at the very end rather | |
672 | * than log_bbnum. In a sense, log_bbnum and zero are the same block | |
673 | * in a circular file. | |
674 | */ | |
675 | if (first_half_cycle == last_half_cycle) { | |
676 | /* | |
677 | * In this case we believe that the entire log should have | |
678 | * cycle number last_half_cycle. We need to scan backwards | |
679 | * from the end verifying that there are no holes still | |
680 | * containing last_half_cycle - 1. If we find such a hole, | |
681 | * then the start of that hole will be the new head. The | |
682 | * simple case looks like | |
683 | * x | x ... | x - 1 | x | |
684 | * Another case that fits this picture would be | |
685 | * x | x + 1 | x ... | x | |
c41564b5 | 686 | * In this case the head really is somewhere at the end of the |
1da177e4 LT |
687 | * log, as one of the latest writes at the beginning was |
688 | * incomplete. | |
689 | * One more case is | |
690 | * x | x + 1 | x ... | x - 1 | x | |
691 | * This is really the combination of the above two cases, and | |
692 | * the head has to end up at the start of the x-1 hole at the | |
693 | * end of the log. | |
694 | * | |
695 | * In the 256k log case, we will read from the beginning to the | |
696 | * end of the log and search for cycle numbers equal to x-1. | |
697 | * We don't worry about the x+1 blocks that we encounter, | |
698 | * because we know that they cannot be the head since the log | |
699 | * started with x. | |
700 | */ | |
701 | head_blk = log_bbnum; | |
702 | stop_on_cycle = last_half_cycle - 1; | |
703 | } else { | |
704 | /* | |
705 | * In this case we want to find the first block with cycle | |
706 | * number matching last_half_cycle. We expect the log to be | |
707 | * some variation on | |
3f943d85 | 708 | * x + 1 ... | x ... | x |
1da177e4 LT |
709 | * The first block with cycle number x (last_half_cycle) will |
710 | * be where the new head belongs. First we do a binary search | |
711 | * for the first occurrence of last_half_cycle. The binary | |
712 | * search may not be totally accurate, so then we scan back | |
713 | * from there looking for occurrences of last_half_cycle before | |
714 | * us. If that backwards scan wraps around the beginning of | |
715 | * the log, then we look for occurrences of last_half_cycle - 1 | |
716 | * at the end of the log. The cases we're looking for look | |
717 | * like | |
3f943d85 AE |
718 | * v binary search stopped here |
719 | * x + 1 ... | x | x + 1 | x ... | x | |
720 | * ^ but we want to locate this spot | |
1da177e4 | 721 | * or |
1da177e4 | 722 | * <---------> less than scan distance |
3f943d85 AE |
723 | * x + 1 ... | x ... | x - 1 | x |
724 | * ^ we want to locate this spot | |
1da177e4 LT |
725 | */ |
726 | stop_on_cycle = last_half_cycle; | |
727 | if ((error = xlog_find_cycle_start(log, bp, first_blk, | |
728 | &head_blk, last_half_cycle))) | |
729 | goto bp_err; | |
730 | } | |
731 | ||
732 | /* | |
733 | * Now validate the answer. Scan back some number of maximum possible | |
734 | * blocks and make sure each one has the expected cycle number. The | |
735 | * maximum is determined by the total possible amount of buffering | |
736 | * in the in-core log. The following number can be made tighter if | |
737 | * we actually look at the block size of the filesystem. | |
738 | */ | |
739 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
740 | if (head_blk >= num_scan_bblks) { | |
741 | /* | |
742 | * We are guaranteed that the entire check can be performed | |
743 | * in one buffer. | |
744 | */ | |
745 | start_blk = head_blk - num_scan_bblks; | |
746 | if ((error = xlog_find_verify_cycle(log, | |
747 | start_blk, num_scan_bblks, | |
748 | stop_on_cycle, &new_blk))) | |
749 | goto bp_err; | |
750 | if (new_blk != -1) | |
751 | head_blk = new_blk; | |
752 | } else { /* need to read 2 parts of log */ | |
753 | /* | |
754 | * We are going to scan backwards in the log in two parts. | |
755 | * First we scan the physical end of the log. In this part | |
756 | * of the log, we are looking for blocks with cycle number | |
757 | * last_half_cycle - 1. | |
758 | * If we find one, then we know that the log starts there, as | |
759 | * we've found a hole that didn't get written in going around | |
760 | * the end of the physical log. The simple case for this is | |
761 | * x + 1 ... | x ... | x - 1 | x | |
762 | * <---------> less than scan distance | |
763 | * If all of the blocks at the end of the log have cycle number | |
764 | * last_half_cycle, then we check the blocks at the start of | |
765 | * the log looking for occurrences of last_half_cycle. If we | |
766 | * find one, then our current estimate for the location of the | |
767 | * first occurrence of last_half_cycle is wrong and we move | |
768 | * back to the hole we've found. This case looks like | |
769 | * x + 1 ... | x | x + 1 | x ... | |
770 | * ^ binary search stopped here | |
771 | * Another case we need to handle that only occurs in 256k | |
772 | * logs is | |
773 | * x + 1 ... | x ... | x+1 | x ... | |
774 | * ^ binary search stops here | |
775 | * In a 256k log, the scan at the end of the log will see the | |
776 | * x + 1 blocks. We need to skip past those since that is | |
777 | * certainly not the head of the log. By searching for | |
778 | * last_half_cycle-1 we accomplish that. | |
779 | */ | |
1da177e4 | 780 | ASSERT(head_blk <= INT_MAX && |
3f943d85 AE |
781 | (xfs_daddr_t) num_scan_bblks >= head_blk); |
782 | start_blk = log_bbnum - (num_scan_bblks - head_blk); | |
1da177e4 LT |
783 | if ((error = xlog_find_verify_cycle(log, start_blk, |
784 | num_scan_bblks - (int)head_blk, | |
785 | (stop_on_cycle - 1), &new_blk))) | |
786 | goto bp_err; | |
787 | if (new_blk != -1) { | |
788 | head_blk = new_blk; | |
9db127ed | 789 | goto validate_head; |
1da177e4 LT |
790 | } |
791 | ||
792 | /* | |
793 | * Scan beginning of log now. The last part of the physical | |
794 | * log is good. This scan needs to verify that it doesn't find | |
795 | * the last_half_cycle. | |
796 | */ | |
797 | start_blk = 0; | |
798 | ASSERT(head_blk <= INT_MAX); | |
799 | if ((error = xlog_find_verify_cycle(log, | |
800 | start_blk, (int)head_blk, | |
801 | stop_on_cycle, &new_blk))) | |
802 | goto bp_err; | |
803 | if (new_blk != -1) | |
804 | head_blk = new_blk; | |
805 | } | |
806 | ||
9db127ed | 807 | validate_head: |
1da177e4 LT |
808 | /* |
809 | * Now we need to make sure head_blk is not pointing to a block in | |
810 | * the middle of a log record. | |
811 | */ | |
812 | num_scan_bblks = XLOG_REC_SHIFT(log); | |
813 | if (head_blk >= num_scan_bblks) { | |
814 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ | |
815 | ||
816 | /* start ptr at last block ptr before head_blk */ | |
817 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
818 | &head_blk, 0)) == -1) { | |
819 | error = XFS_ERROR(EIO); | |
820 | goto bp_err; | |
821 | } else if (error) | |
822 | goto bp_err; | |
823 | } else { | |
824 | start_blk = 0; | |
825 | ASSERT(head_blk <= INT_MAX); | |
826 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
827 | &head_blk, 0)) == -1) { | |
828 | /* We hit the beginning of the log during our search */ | |
3f943d85 | 829 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
1da177e4 LT |
830 | new_blk = log_bbnum; |
831 | ASSERT(start_blk <= INT_MAX && | |
832 | (xfs_daddr_t) log_bbnum-start_blk >= 0); | |
833 | ASSERT(head_blk <= INT_MAX); | |
834 | if ((error = xlog_find_verify_log_record(log, | |
835 | start_blk, &new_blk, | |
836 | (int)head_blk)) == -1) { | |
837 | error = XFS_ERROR(EIO); | |
838 | goto bp_err; | |
839 | } else if (error) | |
840 | goto bp_err; | |
841 | if (new_blk != log_bbnum) | |
842 | head_blk = new_blk; | |
843 | } else if (error) | |
844 | goto bp_err; | |
845 | } | |
846 | ||
847 | xlog_put_bp(bp); | |
848 | if (head_blk == log_bbnum) | |
849 | *return_head_blk = 0; | |
850 | else | |
851 | *return_head_blk = head_blk; | |
852 | /* | |
853 | * When returning here, we have a good block number. Bad block | |
854 | * means that during a previous crash, we didn't have a clean break | |
855 | * from cycle number N to cycle number N-1. In this case, we need | |
856 | * to find the first block with cycle number N-1. | |
857 | */ | |
858 | return 0; | |
859 | ||
860 | bp_err: | |
861 | xlog_put_bp(bp); | |
862 | ||
863 | if (error) | |
a0fa2b67 | 864 | xfs_warn(log->l_mp, "failed to find log head"); |
1da177e4 LT |
865 | return error; |
866 | } | |
867 | ||
868 | /* | |
869 | * Find the sync block number or the tail of the log. | |
870 | * | |
871 | * This will be the block number of the last record to have its | |
872 | * associated buffers synced to disk. Every log record header has | |
873 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy | |
874 | * to get a sync block number. The only concern is to figure out which | |
875 | * log record header to believe. | |
876 | * | |
877 | * The following algorithm uses the log record header with the largest | |
878 | * lsn. The entire log record does not need to be valid. We only care | |
879 | * that the header is valid. | |
880 | * | |
881 | * We could speed up search by using current head_blk buffer, but it is not | |
882 | * available. | |
883 | */ | |
5d77c0dc | 884 | STATIC int |
1da177e4 | 885 | xlog_find_tail( |
9a8d2fdb | 886 | struct xlog *log, |
1da177e4 | 887 | xfs_daddr_t *head_blk, |
65be6054 | 888 | xfs_daddr_t *tail_blk) |
1da177e4 LT |
889 | { |
890 | xlog_rec_header_t *rhead; | |
891 | xlog_op_header_t *op_head; | |
892 | xfs_caddr_t offset = NULL; | |
893 | xfs_buf_t *bp; | |
894 | int error, i, found; | |
895 | xfs_daddr_t umount_data_blk; | |
896 | xfs_daddr_t after_umount_blk; | |
897 | xfs_lsn_t tail_lsn; | |
898 | int hblks; | |
899 | ||
900 | found = 0; | |
901 | ||
902 | /* | |
903 | * Find previous log record | |
904 | */ | |
905 | if ((error = xlog_find_head(log, head_blk))) | |
906 | return error; | |
907 | ||
908 | bp = xlog_get_bp(log, 1); | |
909 | if (!bp) | |
910 | return ENOMEM; | |
911 | if (*head_blk == 0) { /* special case */ | |
076e6acb CH |
912 | error = xlog_bread(log, 0, 1, bp, &offset); |
913 | if (error) | |
9db127ed | 914 | goto done; |
076e6acb | 915 | |
03bea6fe | 916 | if (xlog_get_cycle(offset) == 0) { |
1da177e4 LT |
917 | *tail_blk = 0; |
918 | /* leave all other log inited values alone */ | |
9db127ed | 919 | goto done; |
1da177e4 LT |
920 | } |
921 | } | |
922 | ||
923 | /* | |
924 | * Search backwards looking for log record header block | |
925 | */ | |
926 | ASSERT(*head_blk < INT_MAX); | |
927 | for (i = (int)(*head_blk) - 1; i >= 0; i--) { | |
076e6acb CH |
928 | error = xlog_bread(log, i, 1, bp, &offset); |
929 | if (error) | |
9db127ed | 930 | goto done; |
076e6acb | 931 | |
69ef921b | 932 | if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
1da177e4 LT |
933 | found = 1; |
934 | break; | |
935 | } | |
936 | } | |
937 | /* | |
938 | * If we haven't found the log record header block, start looking | |
939 | * again from the end of the physical log. XXXmiken: There should be | |
940 | * a check here to make sure we didn't search more than N blocks in | |
941 | * the previous code. | |
942 | */ | |
943 | if (!found) { | |
944 | for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) { | |
076e6acb CH |
945 | error = xlog_bread(log, i, 1, bp, &offset); |
946 | if (error) | |
9db127ed | 947 | goto done; |
076e6acb | 948 | |
69ef921b CH |
949 | if (*(__be32 *)offset == |
950 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
1da177e4 LT |
951 | found = 2; |
952 | break; | |
953 | } | |
954 | } | |
955 | } | |
956 | if (!found) { | |
a0fa2b67 | 957 | xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__); |
1da177e4 LT |
958 | ASSERT(0); |
959 | return XFS_ERROR(EIO); | |
960 | } | |
961 | ||
962 | /* find blk_no of tail of log */ | |
963 | rhead = (xlog_rec_header_t *)offset; | |
b53e675d | 964 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); |
1da177e4 LT |
965 | |
966 | /* | |
967 | * Reset log values according to the state of the log when we | |
968 | * crashed. In the case where head_blk == 0, we bump curr_cycle | |
969 | * one because the next write starts a new cycle rather than | |
970 | * continuing the cycle of the last good log record. At this | |
971 | * point we have guaranteed that all partial log records have been | |
972 | * accounted for. Therefore, we know that the last good log record | |
973 | * written was complete and ended exactly on the end boundary | |
974 | * of the physical log. | |
975 | */ | |
976 | log->l_prev_block = i; | |
977 | log->l_curr_block = (int)*head_blk; | |
b53e675d | 978 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); |
1da177e4 LT |
979 | if (found == 2) |
980 | log->l_curr_cycle++; | |
1c3cb9ec | 981 | atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn)); |
84f3c683 | 982 | atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn)); |
28496968 | 983 | xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle, |
a69ed03c | 984 | BBTOB(log->l_curr_block)); |
28496968 | 985 | xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle, |
a69ed03c | 986 | BBTOB(log->l_curr_block)); |
1da177e4 LT |
987 | |
988 | /* | |
989 | * Look for unmount record. If we find it, then we know there | |
990 | * was a clean unmount. Since 'i' could be the last block in | |
991 | * the physical log, we convert to a log block before comparing | |
992 | * to the head_blk. | |
993 | * | |
994 | * Save the current tail lsn to use to pass to | |
995 | * xlog_clear_stale_blocks() below. We won't want to clear the | |
996 | * unmount record if there is one, so we pass the lsn of the | |
997 | * unmount record rather than the block after it. | |
998 | */ | |
62118709 | 999 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d CH |
1000 | int h_size = be32_to_cpu(rhead->h_size); |
1001 | int h_version = be32_to_cpu(rhead->h_version); | |
1da177e4 LT |
1002 | |
1003 | if ((h_version & XLOG_VERSION_2) && | |
1004 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
1005 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
1006 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
1007 | hblks++; | |
1008 | } else { | |
1009 | hblks = 1; | |
1010 | } | |
1011 | } else { | |
1012 | hblks = 1; | |
1013 | } | |
1014 | after_umount_blk = (i + hblks + (int) | |
b53e675d | 1015 | BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize; |
1c3cb9ec | 1016 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
1da177e4 | 1017 | if (*head_blk == after_umount_blk && |
b53e675d | 1018 | be32_to_cpu(rhead->h_num_logops) == 1) { |
1da177e4 | 1019 | umount_data_blk = (i + hblks) % log->l_logBBsize; |
076e6acb CH |
1020 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); |
1021 | if (error) | |
9db127ed | 1022 | goto done; |
076e6acb | 1023 | |
1da177e4 LT |
1024 | op_head = (xlog_op_header_t *)offset; |
1025 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { | |
1026 | /* | |
1027 | * Set tail and last sync so that newly written | |
1028 | * log records will point recovery to after the | |
1029 | * current unmount record. | |
1030 | */ | |
1c3cb9ec DC |
1031 | xlog_assign_atomic_lsn(&log->l_tail_lsn, |
1032 | log->l_curr_cycle, after_umount_blk); | |
1033 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, | |
1034 | log->l_curr_cycle, after_umount_blk); | |
1da177e4 | 1035 | *tail_blk = after_umount_blk; |
92821e2b DC |
1036 | |
1037 | /* | |
1038 | * Note that the unmount was clean. If the unmount | |
1039 | * was not clean, we need to know this to rebuild the | |
1040 | * superblock counters from the perag headers if we | |
1041 | * have a filesystem using non-persistent counters. | |
1042 | */ | |
1043 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; | |
1da177e4 LT |
1044 | } |
1045 | } | |
1046 | ||
1047 | /* | |
1048 | * Make sure that there are no blocks in front of the head | |
1049 | * with the same cycle number as the head. This can happen | |
1050 | * because we allow multiple outstanding log writes concurrently, | |
1051 | * and the later writes might make it out before earlier ones. | |
1052 | * | |
1053 | * We use the lsn from before modifying it so that we'll never | |
1054 | * overwrite the unmount record after a clean unmount. | |
1055 | * | |
1056 | * Do this only if we are going to recover the filesystem | |
1057 | * | |
1058 | * NOTE: This used to say "if (!readonly)" | |
1059 | * However on Linux, we can & do recover a read-only filesystem. | |
1060 | * We only skip recovery if NORECOVERY is specified on mount, | |
1061 | * in which case we would not be here. | |
1062 | * | |
1063 | * But... if the -device- itself is readonly, just skip this. | |
1064 | * We can't recover this device anyway, so it won't matter. | |
1065 | */ | |
9db127ed | 1066 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) |
1da177e4 | 1067 | error = xlog_clear_stale_blocks(log, tail_lsn); |
1da177e4 | 1068 | |
9db127ed | 1069 | done: |
1da177e4 LT |
1070 | xlog_put_bp(bp); |
1071 | ||
1072 | if (error) | |
a0fa2b67 | 1073 | xfs_warn(log->l_mp, "failed to locate log tail"); |
1da177e4 LT |
1074 | return error; |
1075 | } | |
1076 | ||
1077 | /* | |
1078 | * Is the log zeroed at all? | |
1079 | * | |
1080 | * The last binary search should be changed to perform an X block read | |
1081 | * once X becomes small enough. You can then search linearly through | |
1082 | * the X blocks. This will cut down on the number of reads we need to do. | |
1083 | * | |
1084 | * If the log is partially zeroed, this routine will pass back the blkno | |
1085 | * of the first block with cycle number 0. It won't have a complete LR | |
1086 | * preceding it. | |
1087 | * | |
1088 | * Return: | |
1089 | * 0 => the log is completely written to | |
1090 | * -1 => use *blk_no as the first block of the log | |
1091 | * >0 => error has occurred | |
1092 | */ | |
a8272ce0 | 1093 | STATIC int |
1da177e4 | 1094 | xlog_find_zeroed( |
9a8d2fdb | 1095 | struct xlog *log, |
1da177e4 LT |
1096 | xfs_daddr_t *blk_no) |
1097 | { | |
1098 | xfs_buf_t *bp; | |
1099 | xfs_caddr_t offset; | |
1100 | uint first_cycle, last_cycle; | |
1101 | xfs_daddr_t new_blk, last_blk, start_blk; | |
1102 | xfs_daddr_t num_scan_bblks; | |
1103 | int error, log_bbnum = log->l_logBBsize; | |
1104 | ||
6fdf8ccc NS |
1105 | *blk_no = 0; |
1106 | ||
1da177e4 LT |
1107 | /* check totally zeroed log */ |
1108 | bp = xlog_get_bp(log, 1); | |
1109 | if (!bp) | |
1110 | return ENOMEM; | |
076e6acb CH |
1111 | error = xlog_bread(log, 0, 1, bp, &offset); |
1112 | if (error) | |
1da177e4 | 1113 | goto bp_err; |
076e6acb | 1114 | |
03bea6fe | 1115 | first_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1116 | if (first_cycle == 0) { /* completely zeroed log */ |
1117 | *blk_no = 0; | |
1118 | xlog_put_bp(bp); | |
1119 | return -1; | |
1120 | } | |
1121 | ||
1122 | /* check partially zeroed log */ | |
076e6acb CH |
1123 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
1124 | if (error) | |
1da177e4 | 1125 | goto bp_err; |
076e6acb | 1126 | |
03bea6fe | 1127 | last_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1128 | if (last_cycle != 0) { /* log completely written to */ |
1129 | xlog_put_bp(bp); | |
1130 | return 0; | |
1131 | } else if (first_cycle != 1) { | |
1132 | /* | |
1133 | * If the cycle of the last block is zero, the cycle of | |
1134 | * the first block must be 1. If it's not, maybe we're | |
1135 | * not looking at a log... Bail out. | |
1136 | */ | |
a0fa2b67 DC |
1137 | xfs_warn(log->l_mp, |
1138 | "Log inconsistent or not a log (last==0, first!=1)"); | |
1da177e4 LT |
1139 | return XFS_ERROR(EINVAL); |
1140 | } | |
1141 | ||
1142 | /* we have a partially zeroed log */ | |
1143 | last_blk = log_bbnum-1; | |
1144 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) | |
1145 | goto bp_err; | |
1146 | ||
1147 | /* | |
1148 | * Validate the answer. Because there is no way to guarantee that | |
1149 | * the entire log is made up of log records which are the same size, | |
1150 | * we scan over the defined maximum blocks. At this point, the maximum | |
1151 | * is not chosen to mean anything special. XXXmiken | |
1152 | */ | |
1153 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
1154 | ASSERT(num_scan_bblks <= INT_MAX); | |
1155 | ||
1156 | if (last_blk < num_scan_bblks) | |
1157 | num_scan_bblks = last_blk; | |
1158 | start_blk = last_blk - num_scan_bblks; | |
1159 | ||
1160 | /* | |
1161 | * We search for any instances of cycle number 0 that occur before | |
1162 | * our current estimate of the head. What we're trying to detect is | |
1163 | * 1 ... | 0 | 1 | 0... | |
1164 | * ^ binary search ends here | |
1165 | */ | |
1166 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
1167 | (int)num_scan_bblks, 0, &new_blk))) | |
1168 | goto bp_err; | |
1169 | if (new_blk != -1) | |
1170 | last_blk = new_blk; | |
1171 | ||
1172 | /* | |
1173 | * Potentially backup over partial log record write. We don't need | |
1174 | * to search the end of the log because we know it is zero. | |
1175 | */ | |
1176 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
1177 | &last_blk, 0)) == -1) { | |
1178 | error = XFS_ERROR(EIO); | |
1179 | goto bp_err; | |
1180 | } else if (error) | |
1181 | goto bp_err; | |
1182 | ||
1183 | *blk_no = last_blk; | |
1184 | bp_err: | |
1185 | xlog_put_bp(bp); | |
1186 | if (error) | |
1187 | return error; | |
1188 | return -1; | |
1189 | } | |
1190 | ||
1191 | /* | |
1192 | * These are simple subroutines used by xlog_clear_stale_blocks() below | |
1193 | * to initialize a buffer full of empty log record headers and write | |
1194 | * them into the log. | |
1195 | */ | |
1196 | STATIC void | |
1197 | xlog_add_record( | |
9a8d2fdb | 1198 | struct xlog *log, |
1da177e4 LT |
1199 | xfs_caddr_t buf, |
1200 | int cycle, | |
1201 | int block, | |
1202 | int tail_cycle, | |
1203 | int tail_block) | |
1204 | { | |
1205 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; | |
1206 | ||
1207 | memset(buf, 0, BBSIZE); | |
b53e675d CH |
1208 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1209 | recp->h_cycle = cpu_to_be32(cycle); | |
1210 | recp->h_version = cpu_to_be32( | |
62118709 | 1211 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d CH |
1212 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
1213 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); | |
1214 | recp->h_fmt = cpu_to_be32(XLOG_FMT); | |
1da177e4 LT |
1215 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1216 | } | |
1217 | ||
1218 | STATIC int | |
1219 | xlog_write_log_records( | |
9a8d2fdb | 1220 | struct xlog *log, |
1da177e4 LT |
1221 | int cycle, |
1222 | int start_block, | |
1223 | int blocks, | |
1224 | int tail_cycle, | |
1225 | int tail_block) | |
1226 | { | |
1227 | xfs_caddr_t offset; | |
1228 | xfs_buf_t *bp; | |
1229 | int balign, ealign; | |
69ce58f0 | 1230 | int sectbb = log->l_sectBBsize; |
1da177e4 LT |
1231 | int end_block = start_block + blocks; |
1232 | int bufblks; | |
1233 | int error = 0; | |
1234 | int i, j = 0; | |
1235 | ||
6881a229 AE |
1236 | /* |
1237 | * Greedily allocate a buffer big enough to handle the full | |
1238 | * range of basic blocks to be written. If that fails, try | |
1239 | * a smaller size. We need to be able to write at least a | |
1240 | * log sector, or we're out of luck. | |
1241 | */ | |
1da177e4 | 1242 | bufblks = 1 << ffs(blocks); |
81158e0c DC |
1243 | while (bufblks > log->l_logBBsize) |
1244 | bufblks >>= 1; | |
1da177e4 LT |
1245 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1246 | bufblks >>= 1; | |
69ce58f0 | 1247 | if (bufblks < sectbb) |
1da177e4 LT |
1248 | return ENOMEM; |
1249 | } | |
1250 | ||
1251 | /* We may need to do a read at the start to fill in part of | |
1252 | * the buffer in the starting sector not covered by the first | |
1253 | * write below. | |
1254 | */ | |
5c17f533 | 1255 | balign = round_down(start_block, sectbb); |
1da177e4 | 1256 | if (balign != start_block) { |
076e6acb CH |
1257 | error = xlog_bread_noalign(log, start_block, 1, bp); |
1258 | if (error) | |
1259 | goto out_put_bp; | |
1260 | ||
1da177e4 LT |
1261 | j = start_block - balign; |
1262 | } | |
1263 | ||
1264 | for (i = start_block; i < end_block; i += bufblks) { | |
1265 | int bcount, endcount; | |
1266 | ||
1267 | bcount = min(bufblks, end_block - start_block); | |
1268 | endcount = bcount - j; | |
1269 | ||
1270 | /* We may need to do a read at the end to fill in part of | |
1271 | * the buffer in the final sector not covered by the write. | |
1272 | * If this is the same sector as the above read, skip it. | |
1273 | */ | |
5c17f533 | 1274 | ealign = round_down(end_block, sectbb); |
1da177e4 | 1275 | if (j == 0 && (start_block + endcount > ealign)) { |
62926044 | 1276 | offset = bp->b_addr + BBTOB(ealign - start_block); |
44396476 DC |
1277 | error = xlog_bread_offset(log, ealign, sectbb, |
1278 | bp, offset); | |
076e6acb CH |
1279 | if (error) |
1280 | break; | |
1281 | ||
1da177e4 LT |
1282 | } |
1283 | ||
1284 | offset = xlog_align(log, start_block, endcount, bp); | |
1285 | for (; j < endcount; j++) { | |
1286 | xlog_add_record(log, offset, cycle, i+j, | |
1287 | tail_cycle, tail_block); | |
1288 | offset += BBSIZE; | |
1289 | } | |
1290 | error = xlog_bwrite(log, start_block, endcount, bp); | |
1291 | if (error) | |
1292 | break; | |
1293 | start_block += endcount; | |
1294 | j = 0; | |
1295 | } | |
076e6acb CH |
1296 | |
1297 | out_put_bp: | |
1da177e4 LT |
1298 | xlog_put_bp(bp); |
1299 | return error; | |
1300 | } | |
1301 | ||
1302 | /* | |
1303 | * This routine is called to blow away any incomplete log writes out | |
1304 | * in front of the log head. We do this so that we won't become confused | |
1305 | * if we come up, write only a little bit more, and then crash again. | |
1306 | * If we leave the partial log records out there, this situation could | |
1307 | * cause us to think those partial writes are valid blocks since they | |
1308 | * have the current cycle number. We get rid of them by overwriting them | |
1309 | * with empty log records with the old cycle number rather than the | |
1310 | * current one. | |
1311 | * | |
1312 | * The tail lsn is passed in rather than taken from | |
1313 | * the log so that we will not write over the unmount record after a | |
1314 | * clean unmount in a 512 block log. Doing so would leave the log without | |
1315 | * any valid log records in it until a new one was written. If we crashed | |
1316 | * during that time we would not be able to recover. | |
1317 | */ | |
1318 | STATIC int | |
1319 | xlog_clear_stale_blocks( | |
9a8d2fdb | 1320 | struct xlog *log, |
1da177e4 LT |
1321 | xfs_lsn_t tail_lsn) |
1322 | { | |
1323 | int tail_cycle, head_cycle; | |
1324 | int tail_block, head_block; | |
1325 | int tail_distance, max_distance; | |
1326 | int distance; | |
1327 | int error; | |
1328 | ||
1329 | tail_cycle = CYCLE_LSN(tail_lsn); | |
1330 | tail_block = BLOCK_LSN(tail_lsn); | |
1331 | head_cycle = log->l_curr_cycle; | |
1332 | head_block = log->l_curr_block; | |
1333 | ||
1334 | /* | |
1335 | * Figure out the distance between the new head of the log | |
1336 | * and the tail. We want to write over any blocks beyond the | |
1337 | * head that we may have written just before the crash, but | |
1338 | * we don't want to overwrite the tail of the log. | |
1339 | */ | |
1340 | if (head_cycle == tail_cycle) { | |
1341 | /* | |
1342 | * The tail is behind the head in the physical log, | |
1343 | * so the distance from the head to the tail is the | |
1344 | * distance from the head to the end of the log plus | |
1345 | * the distance from the beginning of the log to the | |
1346 | * tail. | |
1347 | */ | |
1348 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { | |
1349 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", | |
1350 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1351 | return XFS_ERROR(EFSCORRUPTED); | |
1352 | } | |
1353 | tail_distance = tail_block + (log->l_logBBsize - head_block); | |
1354 | } else { | |
1355 | /* | |
1356 | * The head is behind the tail in the physical log, | |
1357 | * so the distance from the head to the tail is just | |
1358 | * the tail block minus the head block. | |
1359 | */ | |
1360 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ | |
1361 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", | |
1362 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1363 | return XFS_ERROR(EFSCORRUPTED); | |
1364 | } | |
1365 | tail_distance = tail_block - head_block; | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * If the head is right up against the tail, we can't clear | |
1370 | * anything. | |
1371 | */ | |
1372 | if (tail_distance <= 0) { | |
1373 | ASSERT(tail_distance == 0); | |
1374 | return 0; | |
1375 | } | |
1376 | ||
1377 | max_distance = XLOG_TOTAL_REC_SHIFT(log); | |
1378 | /* | |
1379 | * Take the smaller of the maximum amount of outstanding I/O | |
1380 | * we could have and the distance to the tail to clear out. | |
1381 | * We take the smaller so that we don't overwrite the tail and | |
1382 | * we don't waste all day writing from the head to the tail | |
1383 | * for no reason. | |
1384 | */ | |
1385 | max_distance = MIN(max_distance, tail_distance); | |
1386 | ||
1387 | if ((head_block + max_distance) <= log->l_logBBsize) { | |
1388 | /* | |
1389 | * We can stomp all the blocks we need to without | |
1390 | * wrapping around the end of the log. Just do it | |
1391 | * in a single write. Use the cycle number of the | |
1392 | * current cycle minus one so that the log will look like: | |
1393 | * n ... | n - 1 ... | |
1394 | */ | |
1395 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1396 | head_block, max_distance, tail_cycle, | |
1397 | tail_block); | |
1398 | if (error) | |
1399 | return error; | |
1400 | } else { | |
1401 | /* | |
1402 | * We need to wrap around the end of the physical log in | |
1403 | * order to clear all the blocks. Do it in two separate | |
1404 | * I/Os. The first write should be from the head to the | |
1405 | * end of the physical log, and it should use the current | |
1406 | * cycle number minus one just like above. | |
1407 | */ | |
1408 | distance = log->l_logBBsize - head_block; | |
1409 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1410 | head_block, distance, tail_cycle, | |
1411 | tail_block); | |
1412 | ||
1413 | if (error) | |
1414 | return error; | |
1415 | ||
1416 | /* | |
1417 | * Now write the blocks at the start of the physical log. | |
1418 | * This writes the remainder of the blocks we want to clear. | |
1419 | * It uses the current cycle number since we're now on the | |
1420 | * same cycle as the head so that we get: | |
1421 | * n ... n ... | n - 1 ... | |
1422 | * ^^^^^ blocks we're writing | |
1423 | */ | |
1424 | distance = max_distance - (log->l_logBBsize - head_block); | |
1425 | error = xlog_write_log_records(log, head_cycle, 0, distance, | |
1426 | tail_cycle, tail_block); | |
1427 | if (error) | |
1428 | return error; | |
1429 | } | |
1430 | ||
1431 | return 0; | |
1432 | } | |
1433 | ||
1434 | /****************************************************************************** | |
1435 | * | |
1436 | * Log recover routines | |
1437 | * | |
1438 | ****************************************************************************** | |
1439 | */ | |
1440 | ||
1441 | STATIC xlog_recover_t * | |
1442 | xlog_recover_find_tid( | |
f0a76953 | 1443 | struct hlist_head *head, |
1da177e4 LT |
1444 | xlog_tid_t tid) |
1445 | { | |
f0a76953 | 1446 | xlog_recover_t *trans; |
1da177e4 | 1447 | |
b67bfe0d | 1448 | hlist_for_each_entry(trans, head, r_list) { |
f0a76953 DC |
1449 | if (trans->r_log_tid == tid) |
1450 | return trans; | |
1da177e4 | 1451 | } |
f0a76953 | 1452 | return NULL; |
1da177e4 LT |
1453 | } |
1454 | ||
1455 | STATIC void | |
f0a76953 DC |
1456 | xlog_recover_new_tid( |
1457 | struct hlist_head *head, | |
1458 | xlog_tid_t tid, | |
1459 | xfs_lsn_t lsn) | |
1da177e4 | 1460 | { |
f0a76953 DC |
1461 | xlog_recover_t *trans; |
1462 | ||
1463 | trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP); | |
1464 | trans->r_log_tid = tid; | |
1465 | trans->r_lsn = lsn; | |
1466 | INIT_LIST_HEAD(&trans->r_itemq); | |
1467 | ||
1468 | INIT_HLIST_NODE(&trans->r_list); | |
1469 | hlist_add_head(&trans->r_list, head); | |
1da177e4 LT |
1470 | } |
1471 | ||
1472 | STATIC void | |
1473 | xlog_recover_add_item( | |
f0a76953 | 1474 | struct list_head *head) |
1da177e4 LT |
1475 | { |
1476 | xlog_recover_item_t *item; | |
1477 | ||
1478 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); | |
f0a76953 DC |
1479 | INIT_LIST_HEAD(&item->ri_list); |
1480 | list_add_tail(&item->ri_list, head); | |
1da177e4 LT |
1481 | } |
1482 | ||
1483 | STATIC int | |
1484 | xlog_recover_add_to_cont_trans( | |
ad223e60 MT |
1485 | struct xlog *log, |
1486 | struct xlog_recover *trans, | |
1da177e4 LT |
1487 | xfs_caddr_t dp, |
1488 | int len) | |
1489 | { | |
1490 | xlog_recover_item_t *item; | |
1491 | xfs_caddr_t ptr, old_ptr; | |
1492 | int old_len; | |
1493 | ||
f0a76953 | 1494 | if (list_empty(&trans->r_itemq)) { |
1da177e4 LT |
1495 | /* finish copying rest of trans header */ |
1496 | xlog_recover_add_item(&trans->r_itemq); | |
1497 | ptr = (xfs_caddr_t) &trans->r_theader + | |
1498 | sizeof(xfs_trans_header_t) - len; | |
1499 | memcpy(ptr, dp, len); /* d, s, l */ | |
1500 | return 0; | |
1501 | } | |
f0a76953 DC |
1502 | /* take the tail entry */ |
1503 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
1da177e4 LT |
1504 | |
1505 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; | |
1506 | old_len = item->ri_buf[item->ri_cnt-1].i_len; | |
1507 | ||
45053603 | 1508 | ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP); |
1da177e4 LT |
1509 | memcpy(&ptr[old_len], dp, len); /* d, s, l */ |
1510 | item->ri_buf[item->ri_cnt-1].i_len += len; | |
1511 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; | |
9abbc539 | 1512 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); |
1da177e4 LT |
1513 | return 0; |
1514 | } | |
1515 | ||
1516 | /* | |
1517 | * The next region to add is the start of a new region. It could be | |
1518 | * a whole region or it could be the first part of a new region. Because | |
1519 | * of this, the assumption here is that the type and size fields of all | |
1520 | * format structures fit into the first 32 bits of the structure. | |
1521 | * | |
1522 | * This works because all regions must be 32 bit aligned. Therefore, we | |
1523 | * either have both fields or we have neither field. In the case we have | |
1524 | * neither field, the data part of the region is zero length. We only have | |
1525 | * a log_op_header and can throw away the header since a new one will appear | |
1526 | * later. If we have at least 4 bytes, then we can determine how many regions | |
1527 | * will appear in the current log item. | |
1528 | */ | |
1529 | STATIC int | |
1530 | xlog_recover_add_to_trans( | |
ad223e60 MT |
1531 | struct xlog *log, |
1532 | struct xlog_recover *trans, | |
1da177e4 LT |
1533 | xfs_caddr_t dp, |
1534 | int len) | |
1535 | { | |
1536 | xfs_inode_log_format_t *in_f; /* any will do */ | |
1537 | xlog_recover_item_t *item; | |
1538 | xfs_caddr_t ptr; | |
1539 | ||
1540 | if (!len) | |
1541 | return 0; | |
f0a76953 | 1542 | if (list_empty(&trans->r_itemq)) { |
5a792c45 DC |
1543 | /* we need to catch log corruptions here */ |
1544 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { | |
a0fa2b67 DC |
1545 | xfs_warn(log->l_mp, "%s: bad header magic number", |
1546 | __func__); | |
5a792c45 DC |
1547 | ASSERT(0); |
1548 | return XFS_ERROR(EIO); | |
1549 | } | |
1da177e4 LT |
1550 | if (len == sizeof(xfs_trans_header_t)) |
1551 | xlog_recover_add_item(&trans->r_itemq); | |
1552 | memcpy(&trans->r_theader, dp, len); /* d, s, l */ | |
1553 | return 0; | |
1554 | } | |
1555 | ||
1556 | ptr = kmem_alloc(len, KM_SLEEP); | |
1557 | memcpy(ptr, dp, len); | |
1558 | in_f = (xfs_inode_log_format_t *)ptr; | |
1559 | ||
f0a76953 DC |
1560 | /* take the tail entry */ |
1561 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
1562 | if (item->ri_total != 0 && | |
1563 | item->ri_total == item->ri_cnt) { | |
1564 | /* tail item is in use, get a new one */ | |
1da177e4 | 1565 | xlog_recover_add_item(&trans->r_itemq); |
f0a76953 DC |
1566 | item = list_entry(trans->r_itemq.prev, |
1567 | xlog_recover_item_t, ri_list); | |
1da177e4 | 1568 | } |
1da177e4 LT |
1569 | |
1570 | if (item->ri_total == 0) { /* first region to be added */ | |
e8fa6b48 CH |
1571 | if (in_f->ilf_size == 0 || |
1572 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { | |
a0fa2b67 DC |
1573 | xfs_warn(log->l_mp, |
1574 | "bad number of regions (%d) in inode log format", | |
e8fa6b48 CH |
1575 | in_f->ilf_size); |
1576 | ASSERT(0); | |
1577 | return XFS_ERROR(EIO); | |
1578 | } | |
1579 | ||
1580 | item->ri_total = in_f->ilf_size; | |
1581 | item->ri_buf = | |
1582 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), | |
1583 | KM_SLEEP); | |
1da177e4 LT |
1584 | } |
1585 | ASSERT(item->ri_total > item->ri_cnt); | |
1586 | /* Description region is ri_buf[0] */ | |
1587 | item->ri_buf[item->ri_cnt].i_addr = ptr; | |
1588 | item->ri_buf[item->ri_cnt].i_len = len; | |
1589 | item->ri_cnt++; | |
9abbc539 | 1590 | trace_xfs_log_recover_item_add(log, trans, item, 0); |
1da177e4 LT |
1591 | return 0; |
1592 | } | |
1593 | ||
f0a76953 DC |
1594 | /* |
1595 | * Sort the log items in the transaction. Cancelled buffers need | |
1596 | * to be put first so they are processed before any items that might | |
1597 | * modify the buffers. If they are cancelled, then the modifications | |
1598 | * don't need to be replayed. | |
1599 | */ | |
1da177e4 LT |
1600 | STATIC int |
1601 | xlog_recover_reorder_trans( | |
ad223e60 MT |
1602 | struct xlog *log, |
1603 | struct xlog_recover *trans, | |
9abbc539 | 1604 | int pass) |
1da177e4 | 1605 | { |
f0a76953 DC |
1606 | xlog_recover_item_t *item, *n; |
1607 | LIST_HEAD(sort_list); | |
1608 | ||
1609 | list_splice_init(&trans->r_itemq, &sort_list); | |
1610 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { | |
4e0d5f92 | 1611 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
1da177e4 | 1612 | |
f0a76953 | 1613 | switch (ITEM_TYPE(item)) { |
1da177e4 | 1614 | case XFS_LI_BUF: |
c1155410 | 1615 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
9abbc539 DC |
1616 | trace_xfs_log_recover_item_reorder_head(log, |
1617 | trans, item, pass); | |
f0a76953 | 1618 | list_move(&item->ri_list, &trans->r_itemq); |
1da177e4 LT |
1619 | break; |
1620 | } | |
1621 | case XFS_LI_INODE: | |
1da177e4 LT |
1622 | case XFS_LI_DQUOT: |
1623 | case XFS_LI_QUOTAOFF: | |
1624 | case XFS_LI_EFD: | |
1625 | case XFS_LI_EFI: | |
9abbc539 DC |
1626 | trace_xfs_log_recover_item_reorder_tail(log, |
1627 | trans, item, pass); | |
f0a76953 | 1628 | list_move_tail(&item->ri_list, &trans->r_itemq); |
1da177e4 LT |
1629 | break; |
1630 | default: | |
a0fa2b67 DC |
1631 | xfs_warn(log->l_mp, |
1632 | "%s: unrecognized type of log operation", | |
1633 | __func__); | |
1da177e4 LT |
1634 | ASSERT(0); |
1635 | return XFS_ERROR(EIO); | |
1636 | } | |
f0a76953 DC |
1637 | } |
1638 | ASSERT(list_empty(&sort_list)); | |
1da177e4 LT |
1639 | return 0; |
1640 | } | |
1641 | ||
1642 | /* | |
1643 | * Build up the table of buf cancel records so that we don't replay | |
1644 | * cancelled data in the second pass. For buffer records that are | |
1645 | * not cancel records, there is nothing to do here so we just return. | |
1646 | * | |
1647 | * If we get a cancel record which is already in the table, this indicates | |
1648 | * that the buffer was cancelled multiple times. In order to ensure | |
1649 | * that during pass 2 we keep the record in the table until we reach its | |
1650 | * last occurrence in the log, we keep a reference count in the cancel | |
1651 | * record in the table to tell us how many times we expect to see this | |
1652 | * record during the second pass. | |
1653 | */ | |
c9f71f5f CH |
1654 | STATIC int |
1655 | xlog_recover_buffer_pass1( | |
ad223e60 MT |
1656 | struct xlog *log, |
1657 | struct xlog_recover_item *item) | |
1da177e4 | 1658 | { |
c9f71f5f | 1659 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
d5689eaa CH |
1660 | struct list_head *bucket; |
1661 | struct xfs_buf_cancel *bcp; | |
1da177e4 LT |
1662 | |
1663 | /* | |
1664 | * If this isn't a cancel buffer item, then just return. | |
1665 | */ | |
e2714bf8 | 1666 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
9abbc539 | 1667 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); |
c9f71f5f | 1668 | return 0; |
9abbc539 | 1669 | } |
1da177e4 LT |
1670 | |
1671 | /* | |
d5689eaa CH |
1672 | * Insert an xfs_buf_cancel record into the hash table of them. |
1673 | * If there is already an identical record, bump its reference count. | |
1da177e4 | 1674 | */ |
d5689eaa CH |
1675 | bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno); |
1676 | list_for_each_entry(bcp, bucket, bc_list) { | |
1677 | if (bcp->bc_blkno == buf_f->blf_blkno && | |
1678 | bcp->bc_len == buf_f->blf_len) { | |
1679 | bcp->bc_refcount++; | |
9abbc539 | 1680 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
c9f71f5f | 1681 | return 0; |
1da177e4 | 1682 | } |
d5689eaa CH |
1683 | } |
1684 | ||
1685 | bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP); | |
1686 | bcp->bc_blkno = buf_f->blf_blkno; | |
1687 | bcp->bc_len = buf_f->blf_len; | |
1da177e4 | 1688 | bcp->bc_refcount = 1; |
d5689eaa CH |
1689 | list_add_tail(&bcp->bc_list, bucket); |
1690 | ||
9abbc539 | 1691 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
c9f71f5f | 1692 | return 0; |
1da177e4 LT |
1693 | } |
1694 | ||
1695 | /* | |
1696 | * Check to see whether the buffer being recovered has a corresponding | |
1697 | * entry in the buffer cancel record table. If it does then return 1 | |
1698 | * so that it will be cancelled, otherwise return 0. If the buffer is | |
c1155410 | 1699 | * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement |
1da177e4 LT |
1700 | * the refcount on the entry in the table and remove it from the table |
1701 | * if this is the last reference. | |
1702 | * | |
1703 | * We remove the cancel record from the table when we encounter its | |
1704 | * last occurrence in the log so that if the same buffer is re-used | |
1705 | * again after its last cancellation we actually replay the changes | |
1706 | * made at that point. | |
1707 | */ | |
1708 | STATIC int | |
1709 | xlog_check_buffer_cancelled( | |
ad223e60 | 1710 | struct xlog *log, |
1da177e4 LT |
1711 | xfs_daddr_t blkno, |
1712 | uint len, | |
1713 | ushort flags) | |
1714 | { | |
d5689eaa CH |
1715 | struct list_head *bucket; |
1716 | struct xfs_buf_cancel *bcp; | |
1da177e4 LT |
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 | */ | |
c1155410 | 1723 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
1da177e4 LT |
1724 | return 0; |
1725 | } | |
1726 | ||
1da177e4 | 1727 | /* |
d5689eaa | 1728 | * Search for an entry in the cancel table that matches our buffer. |
1da177e4 | 1729 | */ |
d5689eaa CH |
1730 | bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno); |
1731 | list_for_each_entry(bcp, bucket, bc_list) { | |
1732 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) | |
1733 | goto found; | |
1da177e4 | 1734 | } |
d5689eaa | 1735 | |
1da177e4 | 1736 | /* |
d5689eaa CH |
1737 | * We didn't find a corresponding entry in the table, so return 0 so |
1738 | * that the buffer is NOT cancelled. | |
1da177e4 | 1739 | */ |
c1155410 | 1740 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
1da177e4 | 1741 | return 0; |
d5689eaa CH |
1742 | |
1743 | found: | |
1744 | /* | |
1745 | * We've go a match, so return 1 so that the recovery of this buffer | |
1746 | * is cancelled. If this buffer is actually a buffer cancel log | |
1747 | * item, then decrement the refcount on the one in the table and | |
1748 | * remove it if this is the last reference. | |
1749 | */ | |
1750 | if (flags & XFS_BLF_CANCEL) { | |
1751 | if (--bcp->bc_refcount == 0) { | |
1752 | list_del(&bcp->bc_list); | |
1753 | kmem_free(bcp); | |
1754 | } | |
1755 | } | |
1756 | return 1; | |
1da177e4 LT |
1757 | } |
1758 | ||
1da177e4 | 1759 | /* |
e2714bf8 CH |
1760 | * Perform recovery for a buffer full of inodes. In these buffers, the only |
1761 | * data which should be recovered is that which corresponds to the | |
1762 | * di_next_unlinked pointers in the on disk inode structures. The rest of the | |
1763 | * data for the inodes is always logged through the inodes themselves rather | |
1764 | * than the inode buffer and is recovered in xlog_recover_inode_pass2(). | |
1da177e4 | 1765 | * |
e2714bf8 CH |
1766 | * The only time when buffers full of inodes are fully recovered is when the |
1767 | * buffer is full of newly allocated inodes. In this case the buffer will | |
1768 | * not be marked as an inode buffer and so will be sent to | |
1769 | * xlog_recover_do_reg_buffer() below during recovery. | |
1da177e4 LT |
1770 | */ |
1771 | STATIC int | |
1772 | xlog_recover_do_inode_buffer( | |
e2714bf8 | 1773 | struct xfs_mount *mp, |
1da177e4 | 1774 | xlog_recover_item_t *item, |
e2714bf8 | 1775 | struct xfs_buf *bp, |
1da177e4 LT |
1776 | xfs_buf_log_format_t *buf_f) |
1777 | { | |
1778 | int i; | |
e2714bf8 CH |
1779 | int item_index = 0; |
1780 | int bit = 0; | |
1781 | int nbits = 0; | |
1782 | int reg_buf_offset = 0; | |
1783 | int reg_buf_bytes = 0; | |
1da177e4 LT |
1784 | int next_unlinked_offset; |
1785 | int inodes_per_buf; | |
1786 | xfs_agino_t *logged_nextp; | |
1787 | xfs_agino_t *buffer_nextp; | |
1da177e4 | 1788 | |
9abbc539 | 1789 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
93848a99 | 1790 | bp->b_ops = &xfs_inode_buf_ops; |
9abbc539 | 1791 | |
aa0e8833 | 1792 | inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog; |
1da177e4 LT |
1793 | for (i = 0; i < inodes_per_buf; i++) { |
1794 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + | |
1795 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1796 | ||
1797 | while (next_unlinked_offset >= | |
1798 | (reg_buf_offset + reg_buf_bytes)) { | |
1799 | /* | |
1800 | * The next di_next_unlinked field is beyond | |
1801 | * the current logged region. Find the next | |
1802 | * logged region that contains or is beyond | |
1803 | * the current di_next_unlinked field. | |
1804 | */ | |
1805 | bit += nbits; | |
e2714bf8 CH |
1806 | bit = xfs_next_bit(buf_f->blf_data_map, |
1807 | buf_f->blf_map_size, bit); | |
1da177e4 LT |
1808 | |
1809 | /* | |
1810 | * If there are no more logged regions in the | |
1811 | * buffer, then we're done. | |
1812 | */ | |
e2714bf8 | 1813 | if (bit == -1) |
1da177e4 | 1814 | return 0; |
1da177e4 | 1815 | |
e2714bf8 CH |
1816 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
1817 | buf_f->blf_map_size, bit); | |
1da177e4 | 1818 | ASSERT(nbits > 0); |
c1155410 DC |
1819 | reg_buf_offset = bit << XFS_BLF_SHIFT; |
1820 | reg_buf_bytes = nbits << XFS_BLF_SHIFT; | |
1da177e4 LT |
1821 | item_index++; |
1822 | } | |
1823 | ||
1824 | /* | |
1825 | * If the current logged region starts after the current | |
1826 | * di_next_unlinked field, then move on to the next | |
1827 | * di_next_unlinked field. | |
1828 | */ | |
e2714bf8 | 1829 | if (next_unlinked_offset < reg_buf_offset) |
1da177e4 | 1830 | continue; |
1da177e4 LT |
1831 | |
1832 | ASSERT(item->ri_buf[item_index].i_addr != NULL); | |
c1155410 | 1833 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0); |
aa0e8833 DC |
1834 | ASSERT((reg_buf_offset + reg_buf_bytes) <= |
1835 | BBTOB(bp->b_io_length)); | |
1da177e4 LT |
1836 | |
1837 | /* | |
1838 | * The current logged region contains a copy of the | |
1839 | * current di_next_unlinked field. Extract its value | |
1840 | * and copy it to the buffer copy. | |
1841 | */ | |
4e0d5f92 CH |
1842 | logged_nextp = item->ri_buf[item_index].i_addr + |
1843 | next_unlinked_offset - reg_buf_offset; | |
1da177e4 | 1844 | if (unlikely(*logged_nextp == 0)) { |
a0fa2b67 DC |
1845 | xfs_alert(mp, |
1846 | "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). " | |
1847 | "Trying to replay bad (0) inode di_next_unlinked field.", | |
1da177e4 LT |
1848 | item, bp); |
1849 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", | |
1850 | XFS_ERRLEVEL_LOW, mp); | |
1851 | return XFS_ERROR(EFSCORRUPTED); | |
1852 | } | |
1853 | ||
1854 | buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp, | |
1855 | next_unlinked_offset); | |
87c199c2 | 1856 | *buffer_nextp = *logged_nextp; |
1da177e4 LT |
1857 | } |
1858 | ||
1859 | return 0; | |
1860 | } | |
1861 | ||
1862 | /* | |
1863 | * Perform a 'normal' buffer recovery. Each logged region of the | |
1864 | * buffer should be copied over the corresponding region in the | |
1865 | * given buffer. The bitmap in the buf log format structure indicates | |
1866 | * where to place the logged data. | |
1867 | */ | |
1da177e4 LT |
1868 | STATIC void |
1869 | xlog_recover_do_reg_buffer( | |
9abbc539 | 1870 | struct xfs_mount *mp, |
1da177e4 | 1871 | xlog_recover_item_t *item, |
e2714bf8 | 1872 | struct xfs_buf *bp, |
1da177e4 LT |
1873 | xfs_buf_log_format_t *buf_f) |
1874 | { | |
1875 | int i; | |
1876 | int bit; | |
1877 | int nbits; | |
1da177e4 LT |
1878 | int error; |
1879 | ||
9abbc539 DC |
1880 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); |
1881 | ||
1da177e4 LT |
1882 | bit = 0; |
1883 | i = 1; /* 0 is the buf format structure */ | |
1884 | while (1) { | |
e2714bf8 CH |
1885 | bit = xfs_next_bit(buf_f->blf_data_map, |
1886 | buf_f->blf_map_size, bit); | |
1da177e4 LT |
1887 | if (bit == -1) |
1888 | break; | |
e2714bf8 CH |
1889 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
1890 | buf_f->blf_map_size, bit); | |
1da177e4 | 1891 | ASSERT(nbits > 0); |
4b80916b | 1892 | ASSERT(item->ri_buf[i].i_addr != NULL); |
c1155410 | 1893 | ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0); |
aa0e8833 DC |
1894 | ASSERT(BBTOB(bp->b_io_length) >= |
1895 | ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT)); | |
1da177e4 LT |
1896 | |
1897 | /* | |
1898 | * Do a sanity check if this is a dquot buffer. Just checking | |
1899 | * the first dquot in the buffer should do. XXXThis is | |
1900 | * probably a good thing to do for other buf types also. | |
1901 | */ | |
1902 | error = 0; | |
c8ad20ff | 1903 | if (buf_f->blf_flags & |
c1155410 | 1904 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
0c5e1ce8 | 1905 | if (item->ri_buf[i].i_addr == NULL) { |
a0fa2b67 | 1906 | xfs_alert(mp, |
0c5e1ce8 CH |
1907 | "XFS: NULL dquot in %s.", __func__); |
1908 | goto next; | |
1909 | } | |
8ec6dba2 | 1910 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { |
a0fa2b67 | 1911 | xfs_alert(mp, |
0c5e1ce8 CH |
1912 | "XFS: dquot too small (%d) in %s.", |
1913 | item->ri_buf[i].i_len, __func__); | |
1914 | goto next; | |
1915 | } | |
a0fa2b67 | 1916 | error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr, |
1da177e4 LT |
1917 | -1, 0, XFS_QMOPT_DOWARN, |
1918 | "dquot_buf_recover"); | |
0c5e1ce8 CH |
1919 | if (error) |
1920 | goto next; | |
1da177e4 | 1921 | } |
0c5e1ce8 CH |
1922 | |
1923 | memcpy(xfs_buf_offset(bp, | |
c1155410 | 1924 | (uint)bit << XFS_BLF_SHIFT), /* dest */ |
0c5e1ce8 | 1925 | item->ri_buf[i].i_addr, /* source */ |
c1155410 | 1926 | nbits<<XFS_BLF_SHIFT); /* length */ |
0c5e1ce8 | 1927 | next: |
1da177e4 LT |
1928 | i++; |
1929 | bit += nbits; | |
1930 | } | |
1931 | ||
1932 | /* Shouldn't be any more regions */ | |
1933 | ASSERT(i == item->ri_total); | |
ee1a47ab CH |
1934 | |
1935 | switch (buf_f->blf_flags & XFS_BLF_TYPE_MASK) { | |
1936 | case XFS_BLF_BTREE_BUF: | |
1937 | switch (be32_to_cpu(*(__be32 *)bp->b_addr)) { | |
1938 | case XFS_ABTB_CRC_MAGIC: | |
1939 | case XFS_ABTC_CRC_MAGIC: | |
1940 | case XFS_ABTB_MAGIC: | |
1941 | case XFS_ABTC_MAGIC: | |
1942 | bp->b_ops = &xfs_allocbt_buf_ops; | |
1943 | break; | |
1944 | case XFS_IBT_CRC_MAGIC: | |
1945 | case XFS_IBT_MAGIC: | |
1946 | bp->b_ops = &xfs_inobt_buf_ops; | |
1947 | break; | |
1948 | case XFS_BMAP_CRC_MAGIC: | |
1949 | case XFS_BMAP_MAGIC: | |
1950 | bp->b_ops = &xfs_bmbt_buf_ops; | |
1951 | break; | |
1952 | default: | |
1953 | xfs_warn(mp, "Bad btree block magic!"); | |
1954 | ASSERT(0); | |
1955 | break; | |
1956 | } | |
1957 | break; | |
4e0e6040 DC |
1958 | case XFS_BLF_AGF_BUF: |
1959 | if (*(__be32 *)bp->b_addr != cpu_to_be32(XFS_AGF_MAGIC)) { | |
1960 | xfs_warn(mp, "Bad AGF block magic!"); | |
1961 | ASSERT(0); | |
1962 | break; | |
1963 | } | |
1964 | bp->b_ops = &xfs_agf_buf_ops; | |
1965 | break; | |
77c95bba CH |
1966 | case XFS_BLF_AGFL_BUF: |
1967 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
1968 | break; | |
1969 | if (*(__be32 *)bp->b_addr != cpu_to_be32(XFS_AGFL_MAGIC)) { | |
1970 | xfs_warn(mp, "Bad AGFL block magic!"); | |
1971 | ASSERT(0); | |
1972 | break; | |
1973 | } | |
1974 | bp->b_ops = &xfs_agfl_buf_ops; | |
1975 | break; | |
983d09ff DC |
1976 | case XFS_BLF_AGI_BUF: |
1977 | if (*(__be32 *)bp->b_addr != cpu_to_be32(XFS_AGI_MAGIC)) { | |
1978 | xfs_warn(mp, "Bad AGI block magic!"); | |
1979 | ASSERT(0); | |
1980 | break; | |
1981 | } | |
1982 | bp->b_ops = &xfs_agi_buf_ops; | |
1983 | break; | |
3fe58f30 CH |
1984 | case XFS_BLF_UDQUOT_BUF: |
1985 | case XFS_BLF_PDQUOT_BUF: | |
1986 | case XFS_BLF_GDQUOT_BUF: | |
1987 | if (*(__be16 *)bp->b_addr != cpu_to_be16(XFS_DQUOT_MAGIC)) { | |
1988 | xfs_warn(mp, "Bad DQUOT block magic!"); | |
1989 | ASSERT(0); | |
1990 | break; | |
1991 | } | |
1992 | bp->b_ops = &xfs_dquot_buf_ops; | |
1993 | break; | |
93848a99 CH |
1994 | case XFS_BLF_DINO_BUF: |
1995 | /* | |
1996 | * we get here with inode allocation buffers, not buffers that | |
1997 | * track unlinked list changes. | |
1998 | */ | |
1999 | if (*(__be16 *)bp->b_addr != cpu_to_be16(XFS_DINODE_MAGIC)) { | |
2000 | xfs_warn(mp, "Bad INODE block magic!"); | |
2001 | ASSERT(0); | |
2002 | break; | |
2003 | } | |
2004 | bp->b_ops = &xfs_inode_buf_ops; | |
2005 | break; | |
f948dd76 DC |
2006 | case XFS_BLF_SYMLINK_BUF: |
2007 | if (*(__be32 *)bp->b_addr != cpu_to_be32(XFS_SYMLINK_MAGIC)) { | |
2008 | xfs_warn(mp, "Bad symlink block magic!"); | |
2009 | ASSERT(0); | |
2010 | break; | |
2011 | } | |
2012 | bp->b_ops = &xfs_symlink_buf_ops; | |
2013 | break; | |
ee1a47ab CH |
2014 | default: |
2015 | break; | |
2016 | } | |
1da177e4 LT |
2017 | } |
2018 | ||
2019 | /* | |
2020 | * Do some primitive error checking on ondisk dquot data structures. | |
2021 | */ | |
2022 | int | |
2023 | xfs_qm_dqcheck( | |
a0fa2b67 | 2024 | struct xfs_mount *mp, |
1da177e4 LT |
2025 | xfs_disk_dquot_t *ddq, |
2026 | xfs_dqid_t id, | |
2027 | uint type, /* used only when IO_dorepair is true */ | |
2028 | uint flags, | |
2029 | char *str) | |
2030 | { | |
2031 | xfs_dqblk_t *d = (xfs_dqblk_t *)ddq; | |
2032 | int errs = 0; | |
2033 | ||
2034 | /* | |
2035 | * We can encounter an uninitialized dquot buffer for 2 reasons: | |
2036 | * 1. If we crash while deleting the quotainode(s), and those blks got | |
2037 | * used for user data. This is because we take the path of regular | |
2038 | * file deletion; however, the size field of quotainodes is never | |
2039 | * updated, so all the tricks that we play in itruncate_finish | |
2040 | * don't quite matter. | |
2041 | * | |
2042 | * 2. We don't play the quota buffers when there's a quotaoff logitem. | |
2043 | * But the allocation will be replayed so we'll end up with an | |
2044 | * uninitialized quota block. | |
2045 | * | |
2046 | * This is all fine; things are still consistent, and we haven't lost | |
2047 | * any quota information. Just don't complain about bad dquot blks. | |
2048 | */ | |
69ef921b | 2049 | if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) { |
1da177e4 | 2050 | if (flags & XFS_QMOPT_DOWARN) |
a0fa2b67 | 2051 | xfs_alert(mp, |
1da177e4 | 2052 | "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x", |
1149d96a | 2053 | str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC); |
1da177e4 LT |
2054 | errs++; |
2055 | } | |
1149d96a | 2056 | if (ddq->d_version != XFS_DQUOT_VERSION) { |
1da177e4 | 2057 | if (flags & XFS_QMOPT_DOWARN) |
a0fa2b67 | 2058 | xfs_alert(mp, |
1da177e4 | 2059 | "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x", |
1149d96a | 2060 | str, id, ddq->d_version, XFS_DQUOT_VERSION); |
1da177e4 LT |
2061 | errs++; |
2062 | } | |
2063 | ||
1149d96a CH |
2064 | if (ddq->d_flags != XFS_DQ_USER && |
2065 | ddq->d_flags != XFS_DQ_PROJ && | |
2066 | ddq->d_flags != XFS_DQ_GROUP) { | |
1da177e4 | 2067 | if (flags & XFS_QMOPT_DOWARN) |
a0fa2b67 | 2068 | xfs_alert(mp, |
1da177e4 | 2069 | "%s : XFS dquot ID 0x%x, unknown flags 0x%x", |
1149d96a | 2070 | str, id, ddq->d_flags); |
1da177e4 LT |
2071 | errs++; |
2072 | } | |
2073 | ||
1149d96a | 2074 | if (id != -1 && id != be32_to_cpu(ddq->d_id)) { |
1da177e4 | 2075 | if (flags & XFS_QMOPT_DOWARN) |
a0fa2b67 | 2076 | xfs_alert(mp, |
1da177e4 LT |
2077 | "%s : ondisk-dquot 0x%p, ID mismatch: " |
2078 | "0x%x expected, found id 0x%x", | |
1149d96a | 2079 | str, ddq, id, be32_to_cpu(ddq->d_id)); |
1da177e4 LT |
2080 | errs++; |
2081 | } | |
2082 | ||
2083 | if (!errs && ddq->d_id) { | |
1149d96a | 2084 | if (ddq->d_blk_softlimit && |
d0a3fe67 | 2085 | be64_to_cpu(ddq->d_bcount) > |
1149d96a | 2086 | be64_to_cpu(ddq->d_blk_softlimit)) { |
1da177e4 LT |
2087 | if (!ddq->d_btimer) { |
2088 | if (flags & XFS_QMOPT_DOWARN) | |
a0fa2b67 DC |
2089 | xfs_alert(mp, |
2090 | "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED", | |
1149d96a | 2091 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2092 | errs++; |
2093 | } | |
2094 | } | |
1149d96a | 2095 | if (ddq->d_ino_softlimit && |
d0a3fe67 | 2096 | be64_to_cpu(ddq->d_icount) > |
1149d96a | 2097 | be64_to_cpu(ddq->d_ino_softlimit)) { |
1da177e4 LT |
2098 | if (!ddq->d_itimer) { |
2099 | if (flags & XFS_QMOPT_DOWARN) | |
a0fa2b67 DC |
2100 | xfs_alert(mp, |
2101 | "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED", | |
1149d96a | 2102 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2103 | errs++; |
2104 | } | |
2105 | } | |
1149d96a | 2106 | if (ddq->d_rtb_softlimit && |
d0a3fe67 | 2107 | be64_to_cpu(ddq->d_rtbcount) > |
1149d96a | 2108 | be64_to_cpu(ddq->d_rtb_softlimit)) { |
1da177e4 LT |
2109 | if (!ddq->d_rtbtimer) { |
2110 | if (flags & XFS_QMOPT_DOWARN) | |
a0fa2b67 DC |
2111 | xfs_alert(mp, |
2112 | "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED", | |
1149d96a | 2113 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2114 | errs++; |
2115 | } | |
2116 | } | |
2117 | } | |
2118 | ||
2119 | if (!errs || !(flags & XFS_QMOPT_DQREPAIR)) | |
2120 | return errs; | |
2121 | ||
2122 | if (flags & XFS_QMOPT_DOWARN) | |
a0fa2b67 | 2123 | xfs_notice(mp, "Re-initializing dquot ID 0x%x", id); |
1da177e4 LT |
2124 | |
2125 | /* | |
2126 | * Typically, a repair is only requested by quotacheck. | |
2127 | */ | |
2128 | ASSERT(id != -1); | |
2129 | ASSERT(flags & XFS_QMOPT_DQREPAIR); | |
2130 | memset(d, 0, sizeof(xfs_dqblk_t)); | |
1149d96a CH |
2131 | |
2132 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
2133 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
2134 | d->dd_diskdq.d_flags = type; | |
2135 | d->dd_diskdq.d_id = cpu_to_be32(id); | |
1da177e4 LT |
2136 | |
2137 | return errs; | |
2138 | } | |
2139 | ||
2140 | /* | |
2141 | * Perform a dquot buffer recovery. | |
2142 | * Simple algorithm: if we have found a QUOTAOFF logitem of the same type | |
2143 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. | |
2144 | * Else, treat it as a regular buffer and do recovery. | |
2145 | */ | |
2146 | STATIC void | |
2147 | xlog_recover_do_dquot_buffer( | |
9a8d2fdb MT |
2148 | struct xfs_mount *mp, |
2149 | struct xlog *log, | |
2150 | struct xlog_recover_item *item, | |
2151 | struct xfs_buf *bp, | |
2152 | struct xfs_buf_log_format *buf_f) | |
1da177e4 LT |
2153 | { |
2154 | uint type; | |
2155 | ||
9abbc539 DC |
2156 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
2157 | ||
1da177e4 LT |
2158 | /* |
2159 | * Filesystems are required to send in quota flags at mount time. | |
2160 | */ | |
2161 | if (mp->m_qflags == 0) { | |
2162 | return; | |
2163 | } | |
2164 | ||
2165 | type = 0; | |
c1155410 | 2166 | if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF) |
1da177e4 | 2167 | type |= XFS_DQ_USER; |
c1155410 | 2168 | if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF) |
c8ad20ff | 2169 | type |= XFS_DQ_PROJ; |
c1155410 | 2170 | if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF) |
1da177e4 LT |
2171 | type |= XFS_DQ_GROUP; |
2172 | /* | |
2173 | * This type of quotas was turned off, so ignore this buffer | |
2174 | */ | |
2175 | if (log->l_quotaoffs_flag & type) | |
2176 | return; | |
2177 | ||
9abbc539 | 2178 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 LT |
2179 | } |
2180 | ||
2181 | /* | |
2182 | * This routine replays a modification made to a buffer at runtime. | |
2183 | * There are actually two types of buffer, regular and inode, which | |
2184 | * are handled differently. Inode buffers are handled differently | |
2185 | * in that we only recover a specific set of data from them, namely | |
2186 | * the inode di_next_unlinked fields. This is because all other inode | |
2187 | * data is actually logged via inode records and any data we replay | |
2188 | * here which overlaps that may be stale. | |
2189 | * | |
2190 | * When meta-data buffers are freed at run time we log a buffer item | |
c1155410 | 2191 | * with the XFS_BLF_CANCEL bit set to indicate that previous copies |
1da177e4 LT |
2192 | * of the buffer in the log should not be replayed at recovery time. |
2193 | * This is so that if the blocks covered by the buffer are reused for | |
2194 | * file data before we crash we don't end up replaying old, freed | |
2195 | * meta-data into a user's file. | |
2196 | * | |
2197 | * To handle the cancellation of buffer log items, we make two passes | |
2198 | * over the log during recovery. During the first we build a table of | |
2199 | * those buffers which have been cancelled, and during the second we | |
2200 | * only replay those buffers which do not have corresponding cancel | |
2201 | * records in the table. See xlog_recover_do_buffer_pass[1,2] above | |
2202 | * for more details on the implementation of the table of cancel records. | |
2203 | */ | |
2204 | STATIC int | |
c9f71f5f | 2205 | xlog_recover_buffer_pass2( |
9a8d2fdb MT |
2206 | struct xlog *log, |
2207 | struct list_head *buffer_list, | |
2208 | struct xlog_recover_item *item) | |
1da177e4 | 2209 | { |
4e0d5f92 | 2210 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
e2714bf8 | 2211 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2212 | xfs_buf_t *bp; |
2213 | int error; | |
6ad112bf | 2214 | uint buf_flags; |
1da177e4 | 2215 | |
c9f71f5f CH |
2216 | /* |
2217 | * In this pass we only want to recover all the buffers which have | |
2218 | * not been cancelled and are not cancellation buffers themselves. | |
2219 | */ | |
2220 | if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno, | |
2221 | buf_f->blf_len, buf_f->blf_flags)) { | |
2222 | trace_xfs_log_recover_buf_cancel(log, buf_f); | |
1da177e4 | 2223 | return 0; |
1da177e4 | 2224 | } |
c9f71f5f | 2225 | |
9abbc539 | 2226 | trace_xfs_log_recover_buf_recover(log, buf_f); |
1da177e4 | 2227 | |
a8acad70 | 2228 | buf_flags = 0; |
611c9946 DC |
2229 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) |
2230 | buf_flags |= XBF_UNMAPPED; | |
6ad112bf | 2231 | |
e2714bf8 | 2232 | bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len, |
c3f8fc73 | 2233 | buf_flags, NULL); |
ac4d6888 CS |
2234 | if (!bp) |
2235 | return XFS_ERROR(ENOMEM); | |
e5702805 | 2236 | error = bp->b_error; |
5a52c2a5 | 2237 | if (error) { |
901796af | 2238 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)"); |
1da177e4 LT |
2239 | xfs_buf_relse(bp); |
2240 | return error; | |
2241 | } | |
2242 | ||
e2714bf8 | 2243 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
1da177e4 | 2244 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); |
e2714bf8 | 2245 | } else if (buf_f->blf_flags & |
c1155410 | 2246 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
1da177e4 LT |
2247 | xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); |
2248 | } else { | |
9abbc539 | 2249 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 LT |
2250 | } |
2251 | if (error) | |
2252 | return XFS_ERROR(error); | |
2253 | ||
2254 | /* | |
2255 | * Perform delayed write on the buffer. Asynchronous writes will be | |
2256 | * slower when taking into account all the buffers to be flushed. | |
2257 | * | |
2258 | * Also make sure that only inode buffers with good sizes stay in | |
2259 | * the buffer cache. The kernel moves inodes in buffers of 1 block | |
2260 | * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode | |
2261 | * buffers in the log can be a different size if the log was generated | |
2262 | * by an older kernel using unclustered inode buffers or a newer kernel | |
2263 | * running with a different inode cluster size. Regardless, if the | |
2264 | * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE) | |
2265 | * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep | |
2266 | * the buffer out of the buffer cache so that the buffer won't | |
2267 | * overlap with future reads of those inodes. | |
2268 | */ | |
2269 | if (XFS_DINODE_MAGIC == | |
b53e675d | 2270 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
aa0e8833 | 2271 | (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize, |
1da177e4 | 2272 | (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) { |
c867cb61 | 2273 | xfs_buf_stale(bp); |
c2b006c1 | 2274 | error = xfs_bwrite(bp); |
1da177e4 | 2275 | } else { |
ebad861b | 2276 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2277 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2278 | xfs_buf_delwri_queue(bp, buffer_list); |
1da177e4 LT |
2279 | } |
2280 | ||
c2b006c1 CH |
2281 | xfs_buf_relse(bp); |
2282 | return error; | |
1da177e4 LT |
2283 | } |
2284 | ||
2285 | STATIC int | |
c9f71f5f | 2286 | xlog_recover_inode_pass2( |
9a8d2fdb MT |
2287 | struct xlog *log, |
2288 | struct list_head *buffer_list, | |
2289 | struct xlog_recover_item *item) | |
1da177e4 LT |
2290 | { |
2291 | xfs_inode_log_format_t *in_f; | |
c9f71f5f | 2292 | xfs_mount_t *mp = log->l_mp; |
1da177e4 | 2293 | xfs_buf_t *bp; |
1da177e4 | 2294 | xfs_dinode_t *dip; |
1da177e4 LT |
2295 | int len; |
2296 | xfs_caddr_t src; | |
2297 | xfs_caddr_t dest; | |
2298 | int error; | |
2299 | int attr_index; | |
2300 | uint fields; | |
347d1c01 | 2301 | xfs_icdinode_t *dicp; |
93848a99 | 2302 | uint isize; |
6d192a9b | 2303 | int need_free = 0; |
1da177e4 | 2304 | |
6d192a9b | 2305 | if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) { |
4e0d5f92 | 2306 | in_f = item->ri_buf[0].i_addr; |
6d192a9b | 2307 | } else { |
4e0d5f92 | 2308 | in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP); |
6d192a9b TS |
2309 | need_free = 1; |
2310 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); | |
2311 | if (error) | |
2312 | goto error; | |
2313 | } | |
1da177e4 LT |
2314 | |
2315 | /* | |
2316 | * Inode buffers can be freed, look out for it, | |
2317 | * and do not replay the inode. | |
2318 | */ | |
a1941895 CH |
2319 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
2320 | in_f->ilf_len, 0)) { | |
6d192a9b | 2321 | error = 0; |
9abbc539 | 2322 | trace_xfs_log_recover_inode_cancel(log, in_f); |
6d192a9b TS |
2323 | goto error; |
2324 | } | |
9abbc539 | 2325 | trace_xfs_log_recover_inode_recover(log, in_f); |
1da177e4 | 2326 | |
c3f8fc73 | 2327 | bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0, |
93848a99 | 2328 | &xfs_inode_buf_ops); |
ac4d6888 CS |
2329 | if (!bp) { |
2330 | error = ENOMEM; | |
2331 | goto error; | |
2332 | } | |
e5702805 | 2333 | error = bp->b_error; |
5a52c2a5 | 2334 | if (error) { |
901796af | 2335 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)"); |
1da177e4 | 2336 | xfs_buf_relse(bp); |
6d192a9b | 2337 | goto error; |
1da177e4 | 2338 | } |
1da177e4 | 2339 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); |
a1941895 | 2340 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset); |
1da177e4 LT |
2341 | |
2342 | /* | |
2343 | * Make sure the place we're flushing out to really looks | |
2344 | * like an inode! | |
2345 | */ | |
69ef921b | 2346 | if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) { |
1da177e4 | 2347 | xfs_buf_relse(bp); |
a0fa2b67 DC |
2348 | xfs_alert(mp, |
2349 | "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld", | |
2350 | __func__, dip, bp, in_f->ilf_ino); | |
c9f71f5f | 2351 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)", |
1da177e4 | 2352 | XFS_ERRLEVEL_LOW, mp); |
6d192a9b TS |
2353 | error = EFSCORRUPTED; |
2354 | goto error; | |
1da177e4 | 2355 | } |
4e0d5f92 | 2356 | dicp = item->ri_buf[1].i_addr; |
1da177e4 LT |
2357 | if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) { |
2358 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2359 | xfs_alert(mp, |
2360 | "%s: Bad inode log record, rec ptr 0x%p, ino %Ld", | |
2361 | __func__, item, in_f->ilf_ino); | |
c9f71f5f | 2362 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)", |
1da177e4 | 2363 | XFS_ERRLEVEL_LOW, mp); |
6d192a9b TS |
2364 | error = EFSCORRUPTED; |
2365 | goto error; | |
1da177e4 LT |
2366 | } |
2367 | ||
2368 | /* Skip replay when the on disk inode is newer than the log one */ | |
81591fe2 | 2369 | if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) { |
1da177e4 LT |
2370 | /* |
2371 | * Deal with the wrap case, DI_MAX_FLUSH is less | |
2372 | * than smaller numbers | |
2373 | */ | |
81591fe2 | 2374 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
347d1c01 | 2375 | dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
1da177e4 LT |
2376 | /* do nothing */ |
2377 | } else { | |
2378 | xfs_buf_relse(bp); | |
9abbc539 | 2379 | trace_xfs_log_recover_inode_skip(log, in_f); |
6d192a9b TS |
2380 | error = 0; |
2381 | goto error; | |
1da177e4 LT |
2382 | } |
2383 | } | |
2384 | /* Take the opportunity to reset the flush iteration count */ | |
2385 | dicp->di_flushiter = 0; | |
2386 | ||
abbede1b | 2387 | if (unlikely(S_ISREG(dicp->di_mode))) { |
1da177e4 LT |
2388 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
2389 | (dicp->di_format != XFS_DINODE_FMT_BTREE)) { | |
c9f71f5f | 2390 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)", |
1da177e4 LT |
2391 | XFS_ERRLEVEL_LOW, mp, dicp); |
2392 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2393 | xfs_alert(mp, |
2394 | "%s: Bad regular inode log record, rec ptr 0x%p, " | |
2395 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2396 | __func__, item, dip, bp, in_f->ilf_ino); | |
6d192a9b TS |
2397 | error = EFSCORRUPTED; |
2398 | goto error; | |
1da177e4 | 2399 | } |
abbede1b | 2400 | } else if (unlikely(S_ISDIR(dicp->di_mode))) { |
1da177e4 LT |
2401 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
2402 | (dicp->di_format != XFS_DINODE_FMT_BTREE) && | |
2403 | (dicp->di_format != XFS_DINODE_FMT_LOCAL)) { | |
c9f71f5f | 2404 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)", |
1da177e4 LT |
2405 | XFS_ERRLEVEL_LOW, mp, dicp); |
2406 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2407 | xfs_alert(mp, |
2408 | "%s: Bad dir inode log record, rec ptr 0x%p, " | |
2409 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2410 | __func__, item, dip, bp, in_f->ilf_ino); | |
6d192a9b TS |
2411 | error = EFSCORRUPTED; |
2412 | goto error; | |
1da177e4 LT |
2413 | } |
2414 | } | |
2415 | if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){ | |
c9f71f5f | 2416 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)", |
1da177e4 LT |
2417 | XFS_ERRLEVEL_LOW, mp, dicp); |
2418 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2419 | xfs_alert(mp, |
2420 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
2421 | "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld", | |
2422 | __func__, item, dip, bp, in_f->ilf_ino, | |
1da177e4 LT |
2423 | dicp->di_nextents + dicp->di_anextents, |
2424 | dicp->di_nblocks); | |
6d192a9b TS |
2425 | error = EFSCORRUPTED; |
2426 | goto error; | |
1da177e4 LT |
2427 | } |
2428 | if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) { | |
c9f71f5f | 2429 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)", |
1da177e4 LT |
2430 | XFS_ERRLEVEL_LOW, mp, dicp); |
2431 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2432 | xfs_alert(mp, |
2433 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
2434 | "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__, | |
c9f71f5f | 2435 | item, dip, bp, in_f->ilf_ino, dicp->di_forkoff); |
6d192a9b TS |
2436 | error = EFSCORRUPTED; |
2437 | goto error; | |
1da177e4 | 2438 | } |
93848a99 CH |
2439 | isize = xfs_icdinode_size(dicp->di_version); |
2440 | if (unlikely(item->ri_buf[1].i_len > isize)) { | |
c9f71f5f | 2441 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)", |
1da177e4 LT |
2442 | XFS_ERRLEVEL_LOW, mp, dicp); |
2443 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2444 | xfs_alert(mp, |
2445 | "%s: Bad inode log record length %d, rec ptr 0x%p", | |
2446 | __func__, item->ri_buf[1].i_len, item); | |
6d192a9b TS |
2447 | error = EFSCORRUPTED; |
2448 | goto error; | |
1da177e4 LT |
2449 | } |
2450 | ||
2451 | /* The core is in in-core format */ | |
93848a99 | 2452 | xfs_dinode_to_disk(dip, dicp); |
1da177e4 LT |
2453 | |
2454 | /* the rest is in on-disk format */ | |
93848a99 CH |
2455 | if (item->ri_buf[1].i_len > isize) { |
2456 | memcpy((char *)dip + isize, | |
2457 | item->ri_buf[1].i_addr + isize, | |
2458 | item->ri_buf[1].i_len - isize); | |
1da177e4 LT |
2459 | } |
2460 | ||
2461 | fields = in_f->ilf_fields; | |
2462 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { | |
2463 | case XFS_ILOG_DEV: | |
81591fe2 | 2464 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
1da177e4 LT |
2465 | break; |
2466 | case XFS_ILOG_UUID: | |
81591fe2 CH |
2467 | memcpy(XFS_DFORK_DPTR(dip), |
2468 | &in_f->ilf_u.ilfu_uuid, | |
2469 | sizeof(uuid_t)); | |
1da177e4 LT |
2470 | break; |
2471 | } | |
2472 | ||
2473 | if (in_f->ilf_size == 2) | |
2474 | goto write_inode_buffer; | |
2475 | len = item->ri_buf[2].i_len; | |
2476 | src = item->ri_buf[2].i_addr; | |
2477 | ASSERT(in_f->ilf_size <= 4); | |
2478 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); | |
2479 | ASSERT(!(fields & XFS_ILOG_DFORK) || | |
2480 | (len == in_f->ilf_dsize)); | |
2481 | ||
2482 | switch (fields & XFS_ILOG_DFORK) { | |
2483 | case XFS_ILOG_DDATA: | |
2484 | case XFS_ILOG_DEXT: | |
81591fe2 | 2485 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
1da177e4 LT |
2486 | break; |
2487 | ||
2488 | case XFS_ILOG_DBROOT: | |
7cc95a82 | 2489 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
81591fe2 | 2490 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
1da177e4 LT |
2491 | XFS_DFORK_DSIZE(dip, mp)); |
2492 | break; | |
2493 | ||
2494 | default: | |
2495 | /* | |
2496 | * There are no data fork flags set. | |
2497 | */ | |
2498 | ASSERT((fields & XFS_ILOG_DFORK) == 0); | |
2499 | break; | |
2500 | } | |
2501 | ||
2502 | /* | |
2503 | * If we logged any attribute data, recover it. There may or | |
2504 | * may not have been any other non-core data logged in this | |
2505 | * transaction. | |
2506 | */ | |
2507 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2508 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { | |
2509 | attr_index = 3; | |
2510 | } else { | |
2511 | attr_index = 2; | |
2512 | } | |
2513 | len = item->ri_buf[attr_index].i_len; | |
2514 | src = item->ri_buf[attr_index].i_addr; | |
2515 | ASSERT(len == in_f->ilf_asize); | |
2516 | ||
2517 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2518 | case XFS_ILOG_ADATA: | |
2519 | case XFS_ILOG_AEXT: | |
2520 | dest = XFS_DFORK_APTR(dip); | |
2521 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); | |
2522 | memcpy(dest, src, len); | |
2523 | break; | |
2524 | ||
2525 | case XFS_ILOG_ABROOT: | |
2526 | dest = XFS_DFORK_APTR(dip); | |
7cc95a82 CH |
2527 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
2528 | len, (xfs_bmdr_block_t*)dest, | |
1da177e4 LT |
2529 | XFS_DFORK_ASIZE(dip, mp)); |
2530 | break; | |
2531 | ||
2532 | default: | |
a0fa2b67 | 2533 | xfs_warn(log->l_mp, "%s: Invalid flag", __func__); |
1da177e4 LT |
2534 | ASSERT(0); |
2535 | xfs_buf_relse(bp); | |
6d192a9b TS |
2536 | error = EIO; |
2537 | goto error; | |
1da177e4 LT |
2538 | } |
2539 | } | |
2540 | ||
2541 | write_inode_buffer: | |
93848a99 CH |
2542 | /* re-generate the checksum. */ |
2543 | xfs_dinode_calc_crc(log->l_mp, dip); | |
2544 | ||
ebad861b | 2545 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2546 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2547 | xfs_buf_delwri_queue(bp, buffer_list); |
61551f1e | 2548 | xfs_buf_relse(bp); |
6d192a9b TS |
2549 | error: |
2550 | if (need_free) | |
f0e2d93c | 2551 | kmem_free(in_f); |
6d192a9b | 2552 | return XFS_ERROR(error); |
1da177e4 LT |
2553 | } |
2554 | ||
2555 | /* | |
9a8d2fdb | 2556 | * Recover QUOTAOFF records. We simply make a note of it in the xlog |
1da177e4 LT |
2557 | * structure, so that we know not to do any dquot item or dquot buffer recovery, |
2558 | * of that type. | |
2559 | */ | |
2560 | STATIC int | |
c9f71f5f | 2561 | xlog_recover_quotaoff_pass1( |
9a8d2fdb MT |
2562 | struct xlog *log, |
2563 | struct xlog_recover_item *item) | |
1da177e4 | 2564 | { |
c9f71f5f | 2565 | xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr; |
1da177e4 LT |
2566 | ASSERT(qoff_f); |
2567 | ||
2568 | /* | |
2569 | * The logitem format's flag tells us if this was user quotaoff, | |
77a7cce4 | 2570 | * group/project quotaoff or both. |
1da177e4 LT |
2571 | */ |
2572 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) | |
2573 | log->l_quotaoffs_flag |= XFS_DQ_USER; | |
77a7cce4 NS |
2574 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
2575 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; | |
1da177e4 LT |
2576 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
2577 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; | |
2578 | ||
2579 | return (0); | |
2580 | } | |
2581 | ||
2582 | /* | |
2583 | * Recover a dquot record | |
2584 | */ | |
2585 | STATIC int | |
c9f71f5f | 2586 | xlog_recover_dquot_pass2( |
9a8d2fdb MT |
2587 | struct xlog *log, |
2588 | struct list_head *buffer_list, | |
2589 | struct xlog_recover_item *item) | |
1da177e4 | 2590 | { |
c9f71f5f | 2591 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2592 | xfs_buf_t *bp; |
2593 | struct xfs_disk_dquot *ddq, *recddq; | |
2594 | int error; | |
2595 | xfs_dq_logformat_t *dq_f; | |
2596 | uint type; | |
2597 | ||
1da177e4 LT |
2598 | |
2599 | /* | |
2600 | * Filesystems are required to send in quota flags at mount time. | |
2601 | */ | |
2602 | if (mp->m_qflags == 0) | |
2603 | return (0); | |
2604 | ||
4e0d5f92 CH |
2605 | recddq = item->ri_buf[1].i_addr; |
2606 | if (recddq == NULL) { | |
a0fa2b67 | 2607 | xfs_alert(log->l_mp, "NULL dquot in %s.", __func__); |
0c5e1ce8 CH |
2608 | return XFS_ERROR(EIO); |
2609 | } | |
8ec6dba2 | 2610 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
a0fa2b67 | 2611 | xfs_alert(log->l_mp, "dquot too small (%d) in %s.", |
0c5e1ce8 CH |
2612 | item->ri_buf[1].i_len, __func__); |
2613 | return XFS_ERROR(EIO); | |
2614 | } | |
2615 | ||
1da177e4 LT |
2616 | /* |
2617 | * This type of quotas was turned off, so ignore this record. | |
2618 | */ | |
b53e675d | 2619 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
1da177e4 LT |
2620 | ASSERT(type); |
2621 | if (log->l_quotaoffs_flag & type) | |
2622 | return (0); | |
2623 | ||
2624 | /* | |
2625 | * At this point we know that quota was _not_ turned off. | |
2626 | * Since the mount flags are not indicating to us otherwise, this | |
2627 | * must mean that quota is on, and the dquot needs to be replayed. | |
2628 | * Remember that we may not have fully recovered the superblock yet, | |
2629 | * so we can't do the usual trick of looking at the SB quota bits. | |
2630 | * | |
2631 | * The other possibility, of course, is that the quota subsystem was | |
2632 | * removed since the last mount - ENOSYS. | |
2633 | */ | |
4e0d5f92 | 2634 | dq_f = item->ri_buf[0].i_addr; |
1da177e4 | 2635 | ASSERT(dq_f); |
a0fa2b67 DC |
2636 | error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, |
2637 | "xlog_recover_dquot_pass2 (log copy)"); | |
2638 | if (error) | |
1da177e4 | 2639 | return XFS_ERROR(EIO); |
1da177e4 LT |
2640 | ASSERT(dq_f->qlf_len == 1); |
2641 | ||
7ca790a5 | 2642 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno, |
c3f8fc73 DC |
2643 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp, |
2644 | NULL); | |
7ca790a5 | 2645 | if (error) |
1da177e4 | 2646 | return error; |
7ca790a5 | 2647 | |
1da177e4 LT |
2648 | ASSERT(bp); |
2649 | ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset); | |
2650 | ||
2651 | /* | |
2652 | * At least the magic num portion should be on disk because this | |
2653 | * was among a chunk of dquots created earlier, and we did some | |
2654 | * minimal initialization then. | |
2655 | */ | |
a0fa2b67 DC |
2656 | error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, |
2657 | "xlog_recover_dquot_pass2"); | |
2658 | if (error) { | |
1da177e4 LT |
2659 | xfs_buf_relse(bp); |
2660 | return XFS_ERROR(EIO); | |
2661 | } | |
2662 | ||
2663 | memcpy(ddq, recddq, item->ri_buf[1].i_len); | |
2664 | ||
2665 | ASSERT(dq_f->qlf_size == 2); | |
ebad861b | 2666 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2667 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2668 | xfs_buf_delwri_queue(bp, buffer_list); |
61551f1e | 2669 | xfs_buf_relse(bp); |
1da177e4 LT |
2670 | |
2671 | return (0); | |
2672 | } | |
2673 | ||
2674 | /* | |
2675 | * This routine is called to create an in-core extent free intent | |
2676 | * item from the efi format structure which was logged on disk. | |
2677 | * It allocates an in-core efi, copies the extents from the format | |
2678 | * structure into it, and adds the efi to the AIL with the given | |
2679 | * LSN. | |
2680 | */ | |
6d192a9b | 2681 | STATIC int |
c9f71f5f | 2682 | xlog_recover_efi_pass2( |
9a8d2fdb MT |
2683 | struct xlog *log, |
2684 | struct xlog_recover_item *item, | |
2685 | xfs_lsn_t lsn) | |
1da177e4 | 2686 | { |
6d192a9b | 2687 | int error; |
c9f71f5f | 2688 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2689 | xfs_efi_log_item_t *efip; |
2690 | xfs_efi_log_format_t *efi_formatp; | |
1da177e4 | 2691 | |
4e0d5f92 | 2692 | efi_formatp = item->ri_buf[0].i_addr; |
1da177e4 | 2693 | |
1da177e4 | 2694 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); |
6d192a9b TS |
2695 | if ((error = xfs_efi_copy_format(&(item->ri_buf[0]), |
2696 | &(efip->efi_format)))) { | |
2697 | xfs_efi_item_free(efip); | |
2698 | return error; | |
2699 | } | |
b199c8a4 | 2700 | atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents); |
1da177e4 | 2701 | |
a9c21c1b | 2702 | spin_lock(&log->l_ailp->xa_lock); |
1da177e4 | 2703 | /* |
783a2f65 | 2704 | * xfs_trans_ail_update() drops the AIL lock. |
1da177e4 | 2705 | */ |
e6059949 | 2706 | xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn); |
6d192a9b | 2707 | return 0; |
1da177e4 LT |
2708 | } |
2709 | ||
2710 | ||
2711 | /* | |
2712 | * This routine is called when an efd format structure is found in | |
2713 | * a committed transaction in the log. It's purpose is to cancel | |
2714 | * the corresponding efi if it was still in the log. To do this | |
2715 | * it searches the AIL for the efi with an id equal to that in the | |
2716 | * efd format structure. If we find it, we remove the efi from the | |
2717 | * AIL and free it. | |
2718 | */ | |
c9f71f5f CH |
2719 | STATIC int |
2720 | xlog_recover_efd_pass2( | |
9a8d2fdb MT |
2721 | struct xlog *log, |
2722 | struct xlog_recover_item *item) | |
1da177e4 | 2723 | { |
1da177e4 LT |
2724 | xfs_efd_log_format_t *efd_formatp; |
2725 | xfs_efi_log_item_t *efip = NULL; | |
2726 | xfs_log_item_t *lip; | |
1da177e4 | 2727 | __uint64_t efi_id; |
27d8d5fe | 2728 | struct xfs_ail_cursor cur; |
783a2f65 | 2729 | struct xfs_ail *ailp = log->l_ailp; |
1da177e4 | 2730 | |
4e0d5f92 | 2731 | efd_formatp = item->ri_buf[0].i_addr; |
6d192a9b TS |
2732 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
2733 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || | |
2734 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + | |
2735 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); | |
1da177e4 LT |
2736 | efi_id = efd_formatp->efd_efi_id; |
2737 | ||
2738 | /* | |
2739 | * Search for the efi with the id in the efd format structure | |
2740 | * in the AIL. | |
2741 | */ | |
a9c21c1b DC |
2742 | spin_lock(&ailp->xa_lock); |
2743 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
2744 | while (lip != NULL) { |
2745 | if (lip->li_type == XFS_LI_EFI) { | |
2746 | efip = (xfs_efi_log_item_t *)lip; | |
2747 | if (efip->efi_format.efi_id == efi_id) { | |
2748 | /* | |
783a2f65 | 2749 | * xfs_trans_ail_delete() drops the |
1da177e4 LT |
2750 | * AIL lock. |
2751 | */ | |
04913fdd DC |
2752 | xfs_trans_ail_delete(ailp, lip, |
2753 | SHUTDOWN_CORRUPT_INCORE); | |
8ae2c0f6 | 2754 | xfs_efi_item_free(efip); |
a9c21c1b | 2755 | spin_lock(&ailp->xa_lock); |
27d8d5fe | 2756 | break; |
1da177e4 LT |
2757 | } |
2758 | } | |
a9c21c1b | 2759 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 2760 | } |
a9c21c1b DC |
2761 | xfs_trans_ail_cursor_done(ailp, &cur); |
2762 | spin_unlock(&ailp->xa_lock); | |
c9f71f5f CH |
2763 | |
2764 | return 0; | |
1da177e4 LT |
2765 | } |
2766 | ||
1da177e4 LT |
2767 | /* |
2768 | * Free up any resources allocated by the transaction | |
2769 | * | |
2770 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. | |
2771 | */ | |
2772 | STATIC void | |
2773 | xlog_recover_free_trans( | |
d0450948 | 2774 | struct xlog_recover *trans) |
1da177e4 | 2775 | { |
f0a76953 | 2776 | xlog_recover_item_t *item, *n; |
1da177e4 LT |
2777 | int i; |
2778 | ||
f0a76953 DC |
2779 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { |
2780 | /* Free the regions in the item. */ | |
2781 | list_del(&item->ri_list); | |
2782 | for (i = 0; i < item->ri_cnt; i++) | |
2783 | kmem_free(item->ri_buf[i].i_addr); | |
1da177e4 | 2784 | /* Free the item itself */ |
f0a76953 DC |
2785 | kmem_free(item->ri_buf); |
2786 | kmem_free(item); | |
2787 | } | |
1da177e4 | 2788 | /* Free the transaction recover structure */ |
f0e2d93c | 2789 | kmem_free(trans); |
1da177e4 LT |
2790 | } |
2791 | ||
d0450948 | 2792 | STATIC int |
c9f71f5f | 2793 | xlog_recover_commit_pass1( |
ad223e60 MT |
2794 | struct xlog *log, |
2795 | struct xlog_recover *trans, | |
2796 | struct xlog_recover_item *item) | |
d0450948 | 2797 | { |
c9f71f5f | 2798 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1); |
d0450948 CH |
2799 | |
2800 | switch (ITEM_TYPE(item)) { | |
2801 | case XFS_LI_BUF: | |
c9f71f5f CH |
2802 | return xlog_recover_buffer_pass1(log, item); |
2803 | case XFS_LI_QUOTAOFF: | |
2804 | return xlog_recover_quotaoff_pass1(log, item); | |
d0450948 | 2805 | case XFS_LI_INODE: |
d0450948 | 2806 | case XFS_LI_EFI: |
d0450948 | 2807 | case XFS_LI_EFD: |
c9f71f5f CH |
2808 | case XFS_LI_DQUOT: |
2809 | /* nothing to do in pass 1 */ | |
d0450948 | 2810 | return 0; |
c9f71f5f | 2811 | default: |
a0fa2b67 DC |
2812 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
2813 | __func__, ITEM_TYPE(item)); | |
c9f71f5f CH |
2814 | ASSERT(0); |
2815 | return XFS_ERROR(EIO); | |
2816 | } | |
2817 | } | |
2818 | ||
2819 | STATIC int | |
2820 | xlog_recover_commit_pass2( | |
ad223e60 MT |
2821 | struct xlog *log, |
2822 | struct xlog_recover *trans, | |
2823 | struct list_head *buffer_list, | |
2824 | struct xlog_recover_item *item) | |
c9f71f5f CH |
2825 | { |
2826 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2); | |
2827 | ||
2828 | switch (ITEM_TYPE(item)) { | |
2829 | case XFS_LI_BUF: | |
43ff2122 | 2830 | return xlog_recover_buffer_pass2(log, buffer_list, item); |
c9f71f5f | 2831 | case XFS_LI_INODE: |
43ff2122 | 2832 | return xlog_recover_inode_pass2(log, buffer_list, item); |
c9f71f5f CH |
2833 | case XFS_LI_EFI: |
2834 | return xlog_recover_efi_pass2(log, item, trans->r_lsn); | |
2835 | case XFS_LI_EFD: | |
2836 | return xlog_recover_efd_pass2(log, item); | |
d0450948 | 2837 | case XFS_LI_DQUOT: |
43ff2122 | 2838 | return xlog_recover_dquot_pass2(log, buffer_list, item); |
d0450948 | 2839 | case XFS_LI_QUOTAOFF: |
c9f71f5f CH |
2840 | /* nothing to do in pass2 */ |
2841 | return 0; | |
d0450948 | 2842 | default: |
a0fa2b67 DC |
2843 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
2844 | __func__, ITEM_TYPE(item)); | |
d0450948 CH |
2845 | ASSERT(0); |
2846 | return XFS_ERROR(EIO); | |
2847 | } | |
2848 | } | |
2849 | ||
2850 | /* | |
2851 | * Perform the transaction. | |
2852 | * | |
2853 | * If the transaction modifies a buffer or inode, do it now. Otherwise, | |
2854 | * EFIs and EFDs get queued up by adding entries into the AIL for them. | |
2855 | */ | |
1da177e4 LT |
2856 | STATIC int |
2857 | xlog_recover_commit_trans( | |
ad223e60 | 2858 | struct xlog *log, |
d0450948 | 2859 | struct xlog_recover *trans, |
1da177e4 LT |
2860 | int pass) |
2861 | { | |
43ff2122 | 2862 | int error = 0, error2; |
d0450948 | 2863 | xlog_recover_item_t *item; |
43ff2122 | 2864 | LIST_HEAD (buffer_list); |
1da177e4 | 2865 | |
f0a76953 | 2866 | hlist_del(&trans->r_list); |
d0450948 CH |
2867 | |
2868 | error = xlog_recover_reorder_trans(log, trans, pass); | |
2869 | if (error) | |
1da177e4 | 2870 | return error; |
d0450948 CH |
2871 | |
2872 | list_for_each_entry(item, &trans->r_itemq, ri_list) { | |
43ff2122 CH |
2873 | switch (pass) { |
2874 | case XLOG_RECOVER_PASS1: | |
c9f71f5f | 2875 | error = xlog_recover_commit_pass1(log, trans, item); |
43ff2122 CH |
2876 | break; |
2877 | case XLOG_RECOVER_PASS2: | |
2878 | error = xlog_recover_commit_pass2(log, trans, | |
2879 | &buffer_list, item); | |
2880 | break; | |
2881 | default: | |
2882 | ASSERT(0); | |
2883 | } | |
2884 | ||
d0450948 | 2885 | if (error) |
43ff2122 | 2886 | goto out; |
d0450948 CH |
2887 | } |
2888 | ||
2889 | xlog_recover_free_trans(trans); | |
43ff2122 CH |
2890 | |
2891 | out: | |
2892 | error2 = xfs_buf_delwri_submit(&buffer_list); | |
2893 | return error ? error : error2; | |
1da177e4 LT |
2894 | } |
2895 | ||
2896 | STATIC int | |
2897 | xlog_recover_unmount_trans( | |
ad223e60 MT |
2898 | struct xlog *log, |
2899 | struct xlog_recover *trans) | |
1da177e4 LT |
2900 | { |
2901 | /* Do nothing now */ | |
a0fa2b67 | 2902 | xfs_warn(log->l_mp, "%s: Unmount LR", __func__); |
1da177e4 LT |
2903 | return 0; |
2904 | } | |
2905 | ||
2906 | /* | |
2907 | * There are two valid states of the r_state field. 0 indicates that the | |
2908 | * transaction structure is in a normal state. We have either seen the | |
2909 | * start of the transaction or the last operation we added was not a partial | |
2910 | * operation. If the last operation we added to the transaction was a | |
2911 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. | |
2912 | * | |
2913 | * NOTE: skip LRs with 0 data length. | |
2914 | */ | |
2915 | STATIC int | |
2916 | xlog_recover_process_data( | |
9a8d2fdb | 2917 | struct xlog *log, |
f0a76953 | 2918 | struct hlist_head rhash[], |
9a8d2fdb | 2919 | struct xlog_rec_header *rhead, |
1da177e4 LT |
2920 | xfs_caddr_t dp, |
2921 | int pass) | |
2922 | { | |
2923 | xfs_caddr_t lp; | |
2924 | int num_logops; | |
2925 | xlog_op_header_t *ohead; | |
2926 | xlog_recover_t *trans; | |
2927 | xlog_tid_t tid; | |
2928 | int error; | |
2929 | unsigned long hash; | |
2930 | uint flags; | |
2931 | ||
b53e675d CH |
2932 | lp = dp + be32_to_cpu(rhead->h_len); |
2933 | num_logops = be32_to_cpu(rhead->h_num_logops); | |
1da177e4 LT |
2934 | |
2935 | /* check the log format matches our own - else we can't recover */ | |
2936 | if (xlog_header_check_recover(log->l_mp, rhead)) | |
2937 | return (XFS_ERROR(EIO)); | |
2938 | ||
2939 | while ((dp < lp) && num_logops) { | |
2940 | ASSERT(dp + sizeof(xlog_op_header_t) <= lp); | |
2941 | ohead = (xlog_op_header_t *)dp; | |
2942 | dp += sizeof(xlog_op_header_t); | |
2943 | if (ohead->oh_clientid != XFS_TRANSACTION && | |
2944 | ohead->oh_clientid != XFS_LOG) { | |
a0fa2b67 DC |
2945 | xfs_warn(log->l_mp, "%s: bad clientid 0x%x", |
2946 | __func__, ohead->oh_clientid); | |
1da177e4 LT |
2947 | ASSERT(0); |
2948 | return (XFS_ERROR(EIO)); | |
2949 | } | |
67fcb7bf | 2950 | tid = be32_to_cpu(ohead->oh_tid); |
1da177e4 | 2951 | hash = XLOG_RHASH(tid); |
f0a76953 | 2952 | trans = xlog_recover_find_tid(&rhash[hash], tid); |
1da177e4 LT |
2953 | if (trans == NULL) { /* not found; add new tid */ |
2954 | if (ohead->oh_flags & XLOG_START_TRANS) | |
2955 | xlog_recover_new_tid(&rhash[hash], tid, | |
b53e675d | 2956 | be64_to_cpu(rhead->h_lsn)); |
1da177e4 | 2957 | } else { |
9742bb93 | 2958 | if (dp + be32_to_cpu(ohead->oh_len) > lp) { |
a0fa2b67 DC |
2959 | xfs_warn(log->l_mp, "%s: bad length 0x%x", |
2960 | __func__, be32_to_cpu(ohead->oh_len)); | |
9742bb93 LM |
2961 | WARN_ON(1); |
2962 | return (XFS_ERROR(EIO)); | |
2963 | } | |
1da177e4 LT |
2964 | flags = ohead->oh_flags & ~XLOG_END_TRANS; |
2965 | if (flags & XLOG_WAS_CONT_TRANS) | |
2966 | flags &= ~XLOG_CONTINUE_TRANS; | |
2967 | switch (flags) { | |
2968 | case XLOG_COMMIT_TRANS: | |
2969 | error = xlog_recover_commit_trans(log, | |
f0a76953 | 2970 | trans, pass); |
1da177e4 LT |
2971 | break; |
2972 | case XLOG_UNMOUNT_TRANS: | |
a0fa2b67 | 2973 | error = xlog_recover_unmount_trans(log, trans); |
1da177e4 LT |
2974 | break; |
2975 | case XLOG_WAS_CONT_TRANS: | |
9abbc539 DC |
2976 | error = xlog_recover_add_to_cont_trans(log, |
2977 | trans, dp, | |
2978 | be32_to_cpu(ohead->oh_len)); | |
1da177e4 LT |
2979 | break; |
2980 | case XLOG_START_TRANS: | |
a0fa2b67 DC |
2981 | xfs_warn(log->l_mp, "%s: bad transaction", |
2982 | __func__); | |
1da177e4 LT |
2983 | ASSERT(0); |
2984 | error = XFS_ERROR(EIO); | |
2985 | break; | |
2986 | case 0: | |
2987 | case XLOG_CONTINUE_TRANS: | |
9abbc539 | 2988 | error = xlog_recover_add_to_trans(log, trans, |
67fcb7bf | 2989 | dp, be32_to_cpu(ohead->oh_len)); |
1da177e4 LT |
2990 | break; |
2991 | default: | |
a0fa2b67 DC |
2992 | xfs_warn(log->l_mp, "%s: bad flag 0x%x", |
2993 | __func__, flags); | |
1da177e4 LT |
2994 | ASSERT(0); |
2995 | error = XFS_ERROR(EIO); | |
2996 | break; | |
2997 | } | |
2998 | if (error) | |
2999 | return error; | |
3000 | } | |
67fcb7bf | 3001 | dp += be32_to_cpu(ohead->oh_len); |
1da177e4 LT |
3002 | num_logops--; |
3003 | } | |
3004 | return 0; | |
3005 | } | |
3006 | ||
3007 | /* | |
3008 | * Process an extent free intent item that was recovered from | |
3009 | * the log. We need to free the extents that it describes. | |
3010 | */ | |
3c1e2bbe | 3011 | STATIC int |
1da177e4 LT |
3012 | xlog_recover_process_efi( |
3013 | xfs_mount_t *mp, | |
3014 | xfs_efi_log_item_t *efip) | |
3015 | { | |
3016 | xfs_efd_log_item_t *efdp; | |
3017 | xfs_trans_t *tp; | |
3018 | int i; | |
3c1e2bbe | 3019 | int error = 0; |
1da177e4 LT |
3020 | xfs_extent_t *extp; |
3021 | xfs_fsblock_t startblock_fsb; | |
3022 | ||
b199c8a4 | 3023 | ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)); |
1da177e4 LT |
3024 | |
3025 | /* | |
3026 | * First check the validity of the extents described by the | |
3027 | * EFI. If any are bad, then assume that all are bad and | |
3028 | * just toss the EFI. | |
3029 | */ | |
3030 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3031 | extp = &(efip->efi_format.efi_extents[i]); | |
3032 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
3033 | XFS_FSB_TO_DADDR(mp, extp->ext_start)); | |
3034 | if ((startblock_fsb == 0) || | |
3035 | (extp->ext_len == 0) || | |
3036 | (startblock_fsb >= mp->m_sb.sb_dblocks) || | |
3037 | (extp->ext_len >= mp->m_sb.sb_agblocks)) { | |
3038 | /* | |
3039 | * This will pull the EFI from the AIL and | |
3040 | * free the memory associated with it. | |
3041 | */ | |
666d644c | 3042 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
1da177e4 | 3043 | xfs_efi_release(efip, efip->efi_format.efi_nextents); |
3c1e2bbe | 3044 | return XFS_ERROR(EIO); |
1da177e4 LT |
3045 | } |
3046 | } | |
3047 | ||
3048 | tp = xfs_trans_alloc(mp, 0); | |
3c1e2bbe | 3049 | error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0); |
fc6149d8 DC |
3050 | if (error) |
3051 | goto abort_error; | |
1da177e4 LT |
3052 | efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents); |
3053 | ||
3054 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3055 | extp = &(efip->efi_format.efi_extents[i]); | |
fc6149d8 DC |
3056 | error = xfs_free_extent(tp, extp->ext_start, extp->ext_len); |
3057 | if (error) | |
3058 | goto abort_error; | |
1da177e4 LT |
3059 | xfs_trans_log_efd_extent(tp, efdp, extp->ext_start, |
3060 | extp->ext_len); | |
3061 | } | |
3062 | ||
b199c8a4 | 3063 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
e5720eec | 3064 | error = xfs_trans_commit(tp, 0); |
3c1e2bbe | 3065 | return error; |
fc6149d8 DC |
3066 | |
3067 | abort_error: | |
3068 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3069 | return error; | |
1da177e4 LT |
3070 | } |
3071 | ||
1da177e4 LT |
3072 | /* |
3073 | * When this is called, all of the EFIs which did not have | |
3074 | * corresponding EFDs should be in the AIL. What we do now | |
3075 | * is free the extents associated with each one. | |
3076 | * | |
3077 | * Since we process the EFIs in normal transactions, they | |
3078 | * will be removed at some point after the commit. This prevents | |
3079 | * us from just walking down the list processing each one. | |
3080 | * We'll use a flag in the EFI to skip those that we've already | |
3081 | * processed and use the AIL iteration mechanism's generation | |
3082 | * count to try to speed this up at least a bit. | |
3083 | * | |
3084 | * When we start, we know that the EFIs are the only things in | |
3085 | * the AIL. As we process them, however, other items are added | |
3086 | * to the AIL. Since everything added to the AIL must come after | |
3087 | * everything already in the AIL, we stop processing as soon as | |
3088 | * we see something other than an EFI in the AIL. | |
3089 | */ | |
3c1e2bbe | 3090 | STATIC int |
1da177e4 | 3091 | xlog_recover_process_efis( |
9a8d2fdb | 3092 | struct xlog *log) |
1da177e4 LT |
3093 | { |
3094 | xfs_log_item_t *lip; | |
3095 | xfs_efi_log_item_t *efip; | |
3c1e2bbe | 3096 | int error = 0; |
27d8d5fe | 3097 | struct xfs_ail_cursor cur; |
a9c21c1b | 3098 | struct xfs_ail *ailp; |
1da177e4 | 3099 | |
a9c21c1b DC |
3100 | ailp = log->l_ailp; |
3101 | spin_lock(&ailp->xa_lock); | |
3102 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
3103 | while (lip != NULL) { |
3104 | /* | |
3105 | * We're done when we see something other than an EFI. | |
27d8d5fe | 3106 | * There should be no EFIs left in the AIL now. |
1da177e4 LT |
3107 | */ |
3108 | if (lip->li_type != XFS_LI_EFI) { | |
27d8d5fe | 3109 | #ifdef DEBUG |
a9c21c1b | 3110 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
27d8d5fe DC |
3111 | ASSERT(lip->li_type != XFS_LI_EFI); |
3112 | #endif | |
1da177e4 LT |
3113 | break; |
3114 | } | |
3115 | ||
3116 | /* | |
3117 | * Skip EFIs that we've already processed. | |
3118 | */ | |
3119 | efip = (xfs_efi_log_item_t *)lip; | |
b199c8a4 | 3120 | if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) { |
a9c21c1b | 3121 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 LT |
3122 | continue; |
3123 | } | |
3124 | ||
a9c21c1b DC |
3125 | spin_unlock(&ailp->xa_lock); |
3126 | error = xlog_recover_process_efi(log->l_mp, efip); | |
3127 | spin_lock(&ailp->xa_lock); | |
27d8d5fe DC |
3128 | if (error) |
3129 | goto out; | |
a9c21c1b | 3130 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 3131 | } |
27d8d5fe | 3132 | out: |
a9c21c1b DC |
3133 | xfs_trans_ail_cursor_done(ailp, &cur); |
3134 | spin_unlock(&ailp->xa_lock); | |
3c1e2bbe | 3135 | return error; |
1da177e4 LT |
3136 | } |
3137 | ||
3138 | /* | |
3139 | * This routine performs a transaction to null out a bad inode pointer | |
3140 | * in an agi unlinked inode hash bucket. | |
3141 | */ | |
3142 | STATIC void | |
3143 | xlog_recover_clear_agi_bucket( | |
3144 | xfs_mount_t *mp, | |
3145 | xfs_agnumber_t agno, | |
3146 | int bucket) | |
3147 | { | |
3148 | xfs_trans_t *tp; | |
3149 | xfs_agi_t *agi; | |
3150 | xfs_buf_t *agibp; | |
3151 | int offset; | |
3152 | int error; | |
3153 | ||
3154 | tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET); | |
5e1be0fb CH |
3155 | error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), |
3156 | 0, 0, 0); | |
e5720eec DC |
3157 | if (error) |
3158 | goto out_abort; | |
1da177e4 | 3159 | |
5e1be0fb CH |
3160 | error = xfs_read_agi(mp, tp, agno, &agibp); |
3161 | if (error) | |
e5720eec | 3162 | goto out_abort; |
1da177e4 | 3163 | |
5e1be0fb | 3164 | agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 3165 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
3166 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
3167 | (sizeof(xfs_agino_t) * bucket); | |
3168 | xfs_trans_log_buf(tp, agibp, offset, | |
3169 | (offset + sizeof(xfs_agino_t) - 1)); | |
3170 | ||
e5720eec DC |
3171 | error = xfs_trans_commit(tp, 0); |
3172 | if (error) | |
3173 | goto out_error; | |
3174 | return; | |
3175 | ||
3176 | out_abort: | |
3177 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3178 | out_error: | |
a0fa2b67 | 3179 | xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno); |
e5720eec | 3180 | return; |
1da177e4 LT |
3181 | } |
3182 | ||
23fac50f CH |
3183 | STATIC xfs_agino_t |
3184 | xlog_recover_process_one_iunlink( | |
3185 | struct xfs_mount *mp, | |
3186 | xfs_agnumber_t agno, | |
3187 | xfs_agino_t agino, | |
3188 | int bucket) | |
3189 | { | |
3190 | struct xfs_buf *ibp; | |
3191 | struct xfs_dinode *dip; | |
3192 | struct xfs_inode *ip; | |
3193 | xfs_ino_t ino; | |
3194 | int error; | |
3195 | ||
3196 | ino = XFS_AGINO_TO_INO(mp, agno, agino); | |
7b6259e7 | 3197 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip); |
23fac50f CH |
3198 | if (error) |
3199 | goto fail; | |
3200 | ||
3201 | /* | |
3202 | * Get the on disk inode to find the next inode in the bucket. | |
3203 | */ | |
475ee413 | 3204 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0); |
23fac50f | 3205 | if (error) |
0e446673 | 3206 | goto fail_iput; |
23fac50f | 3207 | |
23fac50f | 3208 | ASSERT(ip->i_d.di_nlink == 0); |
0e446673 | 3209 | ASSERT(ip->i_d.di_mode != 0); |
23fac50f CH |
3210 | |
3211 | /* setup for the next pass */ | |
3212 | agino = be32_to_cpu(dip->di_next_unlinked); | |
3213 | xfs_buf_relse(ibp); | |
3214 | ||
3215 | /* | |
3216 | * Prevent any DMAPI event from being sent when the reference on | |
3217 | * the inode is dropped. | |
3218 | */ | |
3219 | ip->i_d.di_dmevmask = 0; | |
3220 | ||
0e446673 | 3221 | IRELE(ip); |
23fac50f CH |
3222 | return agino; |
3223 | ||
0e446673 CH |
3224 | fail_iput: |
3225 | IRELE(ip); | |
23fac50f CH |
3226 | fail: |
3227 | /* | |
3228 | * We can't read in the inode this bucket points to, or this inode | |
3229 | * is messed up. Just ditch this bucket of inodes. We will lose | |
3230 | * some inodes and space, but at least we won't hang. | |
3231 | * | |
3232 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to | |
3233 | * clear the inode pointer in the bucket. | |
3234 | */ | |
3235 | xlog_recover_clear_agi_bucket(mp, agno, bucket); | |
3236 | return NULLAGINO; | |
3237 | } | |
3238 | ||
1da177e4 LT |
3239 | /* |
3240 | * xlog_iunlink_recover | |
3241 | * | |
3242 | * This is called during recovery to process any inodes which | |
3243 | * we unlinked but not freed when the system crashed. These | |
3244 | * inodes will be on the lists in the AGI blocks. What we do | |
3245 | * here is scan all the AGIs and fully truncate and free any | |
3246 | * inodes found on the lists. Each inode is removed from the | |
3247 | * lists when it has been fully truncated and is freed. The | |
3248 | * freeing of the inode and its removal from the list must be | |
3249 | * atomic. | |
3250 | */ | |
d96f8f89 | 3251 | STATIC void |
1da177e4 | 3252 | xlog_recover_process_iunlinks( |
9a8d2fdb | 3253 | struct xlog *log) |
1da177e4 LT |
3254 | { |
3255 | xfs_mount_t *mp; | |
3256 | xfs_agnumber_t agno; | |
3257 | xfs_agi_t *agi; | |
3258 | xfs_buf_t *agibp; | |
1da177e4 | 3259 | xfs_agino_t agino; |
1da177e4 LT |
3260 | int bucket; |
3261 | int error; | |
3262 | uint mp_dmevmask; | |
3263 | ||
3264 | mp = log->l_mp; | |
3265 | ||
3266 | /* | |
3267 | * Prevent any DMAPI event from being sent while in this function. | |
3268 | */ | |
3269 | mp_dmevmask = mp->m_dmevmask; | |
3270 | mp->m_dmevmask = 0; | |
3271 | ||
3272 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
3273 | /* | |
3274 | * Find the agi for this ag. | |
3275 | */ | |
5e1be0fb CH |
3276 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
3277 | if (error) { | |
3278 | /* | |
3279 | * AGI is b0rked. Don't process it. | |
3280 | * | |
3281 | * We should probably mark the filesystem as corrupt | |
3282 | * after we've recovered all the ag's we can.... | |
3283 | */ | |
3284 | continue; | |
1da177e4 | 3285 | } |
d97d32ed JK |
3286 | /* |
3287 | * Unlock the buffer so that it can be acquired in the normal | |
3288 | * course of the transaction to truncate and free each inode. | |
3289 | * Because we are not racing with anyone else here for the AGI | |
3290 | * buffer, we don't even need to hold it locked to read the | |
3291 | * initial unlinked bucket entries out of the buffer. We keep | |
3292 | * buffer reference though, so that it stays pinned in memory | |
3293 | * while we need the buffer. | |
3294 | */ | |
1da177e4 | 3295 | agi = XFS_BUF_TO_AGI(agibp); |
d97d32ed | 3296 | xfs_buf_unlock(agibp); |
1da177e4 LT |
3297 | |
3298 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { | |
16259e7d | 3299 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
1da177e4 | 3300 | while (agino != NULLAGINO) { |
23fac50f CH |
3301 | agino = xlog_recover_process_one_iunlink(mp, |
3302 | agno, agino, bucket); | |
1da177e4 LT |
3303 | } |
3304 | } | |
d97d32ed | 3305 | xfs_buf_rele(agibp); |
1da177e4 LT |
3306 | } |
3307 | ||
3308 | mp->m_dmevmask = mp_dmevmask; | |
3309 | } | |
3310 | ||
1da177e4 | 3311 | /* |
0e446be4 CH |
3312 | * Upack the log buffer data and crc check it. If the check fails, issue a |
3313 | * warning if and only if the CRC in the header is non-zero. This makes the | |
3314 | * check an advisory warning, and the zero CRC check will prevent failure | |
3315 | * warnings from being emitted when upgrading the kernel from one that does not | |
3316 | * add CRCs by default. | |
3317 | * | |
3318 | * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log | |
3319 | * corruption failure | |
1da177e4 | 3320 | */ |
0e446be4 CH |
3321 | STATIC int |
3322 | xlog_unpack_data_crc( | |
3323 | struct xlog_rec_header *rhead, | |
3324 | xfs_caddr_t dp, | |
3325 | struct xlog *log) | |
1da177e4 | 3326 | { |
f9668a09 | 3327 | __le32 crc; |
0e446be4 CH |
3328 | |
3329 | crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len)); | |
3330 | if (crc != rhead->h_crc) { | |
3331 | if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) { | |
3332 | xfs_alert(log->l_mp, | |
3333 | "log record CRC mismatch: found 0x%x, expected 0x%x.\n", | |
f9668a09 DC |
3334 | le32_to_cpu(rhead->h_crc), |
3335 | le32_to_cpu(crc)); | |
0e446be4 | 3336 | xfs_hex_dump(dp, 32); |
1da177e4 LT |
3337 | } |
3338 | ||
0e446be4 CH |
3339 | /* |
3340 | * If we've detected a log record corruption, then we can't | |
3341 | * recover past this point. Abort recovery if we are enforcing | |
3342 | * CRC protection by punting an error back up the stack. | |
3343 | */ | |
3344 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) | |
3345 | return EFSCORRUPTED; | |
1da177e4 | 3346 | } |
0e446be4 CH |
3347 | |
3348 | return 0; | |
1da177e4 LT |
3349 | } |
3350 | ||
0e446be4 | 3351 | STATIC int |
1da177e4 | 3352 | xlog_unpack_data( |
9a8d2fdb | 3353 | struct xlog_rec_header *rhead, |
1da177e4 | 3354 | xfs_caddr_t dp, |
9a8d2fdb | 3355 | struct xlog *log) |
1da177e4 LT |
3356 | { |
3357 | int i, j, k; | |
0e446be4 CH |
3358 | int error; |
3359 | ||
3360 | error = xlog_unpack_data_crc(rhead, dp, log); | |
3361 | if (error) | |
3362 | return error; | |
1da177e4 | 3363 | |
b53e675d | 3364 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
1da177e4 | 3365 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
b53e675d | 3366 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
1da177e4 LT |
3367 | dp += BBSIZE; |
3368 | } | |
3369 | ||
62118709 | 3370 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b28708d6 | 3371 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
b53e675d | 3372 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
1da177e4 LT |
3373 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
3374 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3375 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
1da177e4 LT |
3376 | dp += BBSIZE; |
3377 | } | |
3378 | } | |
0e446be4 CH |
3379 | |
3380 | return 0; | |
1da177e4 LT |
3381 | } |
3382 | ||
3383 | STATIC int | |
3384 | xlog_valid_rec_header( | |
9a8d2fdb MT |
3385 | struct xlog *log, |
3386 | struct xlog_rec_header *rhead, | |
1da177e4 LT |
3387 | xfs_daddr_t blkno) |
3388 | { | |
3389 | int hlen; | |
3390 | ||
69ef921b | 3391 | if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) { |
1da177e4 LT |
3392 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
3393 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3394 | return XFS_ERROR(EFSCORRUPTED); | |
3395 | } | |
3396 | if (unlikely( | |
3397 | (!rhead->h_version || | |
b53e675d | 3398 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
a0fa2b67 | 3399 | xfs_warn(log->l_mp, "%s: unrecognised log version (%d).", |
34a622b2 | 3400 | __func__, be32_to_cpu(rhead->h_version)); |
1da177e4 LT |
3401 | return XFS_ERROR(EIO); |
3402 | } | |
3403 | ||
3404 | /* LR body must have data or it wouldn't have been written */ | |
b53e675d | 3405 | hlen = be32_to_cpu(rhead->h_len); |
1da177e4 LT |
3406 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
3407 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", | |
3408 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3409 | return XFS_ERROR(EFSCORRUPTED); | |
3410 | } | |
3411 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { | |
3412 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", | |
3413 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3414 | return XFS_ERROR(EFSCORRUPTED); | |
3415 | } | |
3416 | return 0; | |
3417 | } | |
3418 | ||
3419 | /* | |
3420 | * Read the log from tail to head and process the log records found. | |
3421 | * Handle the two cases where the tail and head are in the same cycle | |
3422 | * and where the active portion of the log wraps around the end of | |
3423 | * the physical log separately. The pass parameter is passed through | |
3424 | * to the routines called to process the data and is not looked at | |
3425 | * here. | |
3426 | */ | |
3427 | STATIC int | |
3428 | xlog_do_recovery_pass( | |
9a8d2fdb | 3429 | struct xlog *log, |
1da177e4 LT |
3430 | xfs_daddr_t head_blk, |
3431 | xfs_daddr_t tail_blk, | |
3432 | int pass) | |
3433 | { | |
3434 | xlog_rec_header_t *rhead; | |
3435 | xfs_daddr_t blk_no; | |
fc5bc4c8 | 3436 | xfs_caddr_t offset; |
1da177e4 LT |
3437 | xfs_buf_t *hbp, *dbp; |
3438 | int error = 0, h_size; | |
3439 | int bblks, split_bblks; | |
3440 | int hblks, split_hblks, wrapped_hblks; | |
f0a76953 | 3441 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
1da177e4 LT |
3442 | |
3443 | ASSERT(head_blk != tail_blk); | |
3444 | ||
3445 | /* | |
3446 | * Read the header of the tail block and get the iclog buffer size from | |
3447 | * h_size. Use this to tell how many sectors make up the log header. | |
3448 | */ | |
62118709 | 3449 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
1da177e4 LT |
3450 | /* |
3451 | * When using variable length iclogs, read first sector of | |
3452 | * iclog header and extract the header size from it. Get a | |
3453 | * new hbp that is the correct size. | |
3454 | */ | |
3455 | hbp = xlog_get_bp(log, 1); | |
3456 | if (!hbp) | |
3457 | return ENOMEM; | |
076e6acb CH |
3458 | |
3459 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); | |
3460 | if (error) | |
1da177e4 | 3461 | goto bread_err1; |
076e6acb | 3462 | |
1da177e4 LT |
3463 | rhead = (xlog_rec_header_t *)offset; |
3464 | error = xlog_valid_rec_header(log, rhead, tail_blk); | |
3465 | if (error) | |
3466 | goto bread_err1; | |
b53e675d CH |
3467 | h_size = be32_to_cpu(rhead->h_size); |
3468 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && | |
1da177e4 LT |
3469 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
3470 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
3471 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
3472 | hblks++; | |
3473 | xlog_put_bp(hbp); | |
3474 | hbp = xlog_get_bp(log, hblks); | |
3475 | } else { | |
3476 | hblks = 1; | |
3477 | } | |
3478 | } else { | |
69ce58f0 | 3479 | ASSERT(log->l_sectBBsize == 1); |
1da177e4 LT |
3480 | hblks = 1; |
3481 | hbp = xlog_get_bp(log, 1); | |
3482 | h_size = XLOG_BIG_RECORD_BSIZE; | |
3483 | } | |
3484 | ||
3485 | if (!hbp) | |
3486 | return ENOMEM; | |
3487 | dbp = xlog_get_bp(log, BTOBB(h_size)); | |
3488 | if (!dbp) { | |
3489 | xlog_put_bp(hbp); | |
3490 | return ENOMEM; | |
3491 | } | |
3492 | ||
3493 | memset(rhash, 0, sizeof(rhash)); | |
3494 | if (tail_blk <= head_blk) { | |
3495 | for (blk_no = tail_blk; blk_no < head_blk; ) { | |
076e6acb CH |
3496 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
3497 | if (error) | |
1da177e4 | 3498 | goto bread_err2; |
076e6acb | 3499 | |
1da177e4 LT |
3500 | rhead = (xlog_rec_header_t *)offset; |
3501 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3502 | if (error) | |
3503 | goto bread_err2; | |
3504 | ||
3505 | /* blocks in data section */ | |
b53e675d | 3506 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
3507 | error = xlog_bread(log, blk_no + hblks, bblks, dbp, |
3508 | &offset); | |
1da177e4 LT |
3509 | if (error) |
3510 | goto bread_err2; | |
076e6acb | 3511 | |
0e446be4 CH |
3512 | error = xlog_unpack_data(rhead, offset, log); |
3513 | if (error) | |
3514 | goto bread_err2; | |
3515 | ||
3516 | error = xlog_recover_process_data(log, | |
3517 | rhash, rhead, offset, pass); | |
3518 | if (error) | |
1da177e4 LT |
3519 | goto bread_err2; |
3520 | blk_no += bblks + hblks; | |
3521 | } | |
3522 | } else { | |
3523 | /* | |
3524 | * Perform recovery around the end of the physical log. | |
3525 | * When the head is not on the same cycle number as the tail, | |
3526 | * we can't do a sequential recovery as above. | |
3527 | */ | |
3528 | blk_no = tail_blk; | |
3529 | while (blk_no < log->l_logBBsize) { | |
3530 | /* | |
3531 | * Check for header wrapping around physical end-of-log | |
3532 | */ | |
62926044 | 3533 | offset = hbp->b_addr; |
1da177e4 LT |
3534 | split_hblks = 0; |
3535 | wrapped_hblks = 0; | |
3536 | if (blk_no + hblks <= log->l_logBBsize) { | |
3537 | /* Read header in one read */ | |
076e6acb CH |
3538 | error = xlog_bread(log, blk_no, hblks, hbp, |
3539 | &offset); | |
1da177e4 LT |
3540 | if (error) |
3541 | goto bread_err2; | |
1da177e4 LT |
3542 | } else { |
3543 | /* This LR is split across physical log end */ | |
3544 | if (blk_no != log->l_logBBsize) { | |
3545 | /* some data before physical log end */ | |
3546 | ASSERT(blk_no <= INT_MAX); | |
3547 | split_hblks = log->l_logBBsize - (int)blk_no; | |
3548 | ASSERT(split_hblks > 0); | |
076e6acb CH |
3549 | error = xlog_bread(log, blk_no, |
3550 | split_hblks, hbp, | |
3551 | &offset); | |
3552 | if (error) | |
1da177e4 | 3553 | goto bread_err2; |
1da177e4 | 3554 | } |
076e6acb | 3555 | |
1da177e4 LT |
3556 | /* |
3557 | * Note: this black magic still works with | |
3558 | * large sector sizes (non-512) only because: | |
3559 | * - we increased the buffer size originally | |
3560 | * by 1 sector giving us enough extra space | |
3561 | * for the second read; | |
3562 | * - the log start is guaranteed to be sector | |
3563 | * aligned; | |
3564 | * - we read the log end (LR header start) | |
3565 | * _first_, then the log start (LR header end) | |
3566 | * - order is important. | |
3567 | */ | |
234f56ac | 3568 | wrapped_hblks = hblks - split_hblks; |
44396476 DC |
3569 | error = xlog_bread_offset(log, 0, |
3570 | wrapped_hblks, hbp, | |
3571 | offset + BBTOB(split_hblks)); | |
1da177e4 LT |
3572 | if (error) |
3573 | goto bread_err2; | |
1da177e4 LT |
3574 | } |
3575 | rhead = (xlog_rec_header_t *)offset; | |
3576 | error = xlog_valid_rec_header(log, rhead, | |
3577 | split_hblks ? blk_no : 0); | |
3578 | if (error) | |
3579 | goto bread_err2; | |
3580 | ||
b53e675d | 3581 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
1da177e4 LT |
3582 | blk_no += hblks; |
3583 | ||
3584 | /* Read in data for log record */ | |
3585 | if (blk_no + bblks <= log->l_logBBsize) { | |
076e6acb CH |
3586 | error = xlog_bread(log, blk_no, bblks, dbp, |
3587 | &offset); | |
1da177e4 LT |
3588 | if (error) |
3589 | goto bread_err2; | |
1da177e4 LT |
3590 | } else { |
3591 | /* This log record is split across the | |
3592 | * physical end of log */ | |
62926044 | 3593 | offset = dbp->b_addr; |
1da177e4 LT |
3594 | split_bblks = 0; |
3595 | if (blk_no != log->l_logBBsize) { | |
3596 | /* some data is before the physical | |
3597 | * end of log */ | |
3598 | ASSERT(!wrapped_hblks); | |
3599 | ASSERT(blk_no <= INT_MAX); | |
3600 | split_bblks = | |
3601 | log->l_logBBsize - (int)blk_no; | |
3602 | ASSERT(split_bblks > 0); | |
076e6acb CH |
3603 | error = xlog_bread(log, blk_no, |
3604 | split_bblks, dbp, | |
3605 | &offset); | |
3606 | if (error) | |
1da177e4 | 3607 | goto bread_err2; |
1da177e4 | 3608 | } |
076e6acb | 3609 | |
1da177e4 LT |
3610 | /* |
3611 | * Note: this black magic still works with | |
3612 | * large sector sizes (non-512) only because: | |
3613 | * - we increased the buffer size originally | |
3614 | * by 1 sector giving us enough extra space | |
3615 | * for the second read; | |
3616 | * - the log start is guaranteed to be sector | |
3617 | * aligned; | |
3618 | * - we read the log end (LR header start) | |
3619 | * _first_, then the log start (LR header end) | |
3620 | * - order is important. | |
3621 | */ | |
44396476 | 3622 | error = xlog_bread_offset(log, 0, |
009507b0 | 3623 | bblks - split_bblks, dbp, |
44396476 | 3624 | offset + BBTOB(split_bblks)); |
076e6acb CH |
3625 | if (error) |
3626 | goto bread_err2; | |
1da177e4 | 3627 | } |
0e446be4 CH |
3628 | |
3629 | error = xlog_unpack_data(rhead, offset, log); | |
3630 | if (error) | |
3631 | goto bread_err2; | |
3632 | ||
3633 | error = xlog_recover_process_data(log, rhash, | |
3634 | rhead, offset, pass); | |
3635 | if (error) | |
1da177e4 LT |
3636 | goto bread_err2; |
3637 | blk_no += bblks; | |
3638 | } | |
3639 | ||
3640 | ASSERT(blk_no >= log->l_logBBsize); | |
3641 | blk_no -= log->l_logBBsize; | |
3642 | ||
3643 | /* read first part of physical log */ | |
3644 | while (blk_no < head_blk) { | |
076e6acb CH |
3645 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
3646 | if (error) | |
1da177e4 | 3647 | goto bread_err2; |
076e6acb | 3648 | |
1da177e4 LT |
3649 | rhead = (xlog_rec_header_t *)offset; |
3650 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3651 | if (error) | |
3652 | goto bread_err2; | |
076e6acb | 3653 | |
b53e675d | 3654 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
3655 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, |
3656 | &offset); | |
3657 | if (error) | |
1da177e4 | 3658 | goto bread_err2; |
076e6acb | 3659 | |
0e446be4 CH |
3660 | error = xlog_unpack_data(rhead, offset, log); |
3661 | if (error) | |
3662 | goto bread_err2; | |
3663 | ||
3664 | error = xlog_recover_process_data(log, rhash, | |
3665 | rhead, offset, pass); | |
3666 | if (error) | |
1da177e4 LT |
3667 | goto bread_err2; |
3668 | blk_no += bblks + hblks; | |
3669 | } | |
3670 | } | |
3671 | ||
3672 | bread_err2: | |
3673 | xlog_put_bp(dbp); | |
3674 | bread_err1: | |
3675 | xlog_put_bp(hbp); | |
3676 | return error; | |
3677 | } | |
3678 | ||
3679 | /* | |
3680 | * Do the recovery of the log. We actually do this in two phases. | |
3681 | * The two passes are necessary in order to implement the function | |
3682 | * of cancelling a record written into the log. The first pass | |
3683 | * determines those things which have been cancelled, and the | |
3684 | * second pass replays log items normally except for those which | |
3685 | * have been cancelled. The handling of the replay and cancellations | |
3686 | * takes place in the log item type specific routines. | |
3687 | * | |
3688 | * The table of items which have cancel records in the log is allocated | |
3689 | * and freed at this level, since only here do we know when all of | |
3690 | * the log recovery has been completed. | |
3691 | */ | |
3692 | STATIC int | |
3693 | xlog_do_log_recovery( | |
9a8d2fdb | 3694 | struct xlog *log, |
1da177e4 LT |
3695 | xfs_daddr_t head_blk, |
3696 | xfs_daddr_t tail_blk) | |
3697 | { | |
d5689eaa | 3698 | int error, i; |
1da177e4 LT |
3699 | |
3700 | ASSERT(head_blk != tail_blk); | |
3701 | ||
3702 | /* | |
3703 | * First do a pass to find all of the cancelled buf log items. | |
3704 | * Store them in the buf_cancel_table for use in the second pass. | |
3705 | */ | |
d5689eaa CH |
3706 | log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE * |
3707 | sizeof(struct list_head), | |
1da177e4 | 3708 | KM_SLEEP); |
d5689eaa CH |
3709 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
3710 | INIT_LIST_HEAD(&log->l_buf_cancel_table[i]); | |
3711 | ||
1da177e4 LT |
3712 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
3713 | XLOG_RECOVER_PASS1); | |
3714 | if (error != 0) { | |
f0e2d93c | 3715 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
3716 | log->l_buf_cancel_table = NULL; |
3717 | return error; | |
3718 | } | |
3719 | /* | |
3720 | * Then do a second pass to actually recover the items in the log. | |
3721 | * When it is complete free the table of buf cancel items. | |
3722 | */ | |
3723 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
3724 | XLOG_RECOVER_PASS2); | |
3725 | #ifdef DEBUG | |
6d192a9b | 3726 | if (!error) { |
1da177e4 LT |
3727 | int i; |
3728 | ||
3729 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
d5689eaa | 3730 | ASSERT(list_empty(&log->l_buf_cancel_table[i])); |
1da177e4 LT |
3731 | } |
3732 | #endif /* DEBUG */ | |
3733 | ||
f0e2d93c | 3734 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
3735 | log->l_buf_cancel_table = NULL; |
3736 | ||
3737 | return error; | |
3738 | } | |
3739 | ||
3740 | /* | |
3741 | * Do the actual recovery | |
3742 | */ | |
3743 | STATIC int | |
3744 | xlog_do_recover( | |
9a8d2fdb | 3745 | struct xlog *log, |
1da177e4 LT |
3746 | xfs_daddr_t head_blk, |
3747 | xfs_daddr_t tail_blk) | |
3748 | { | |
3749 | int error; | |
3750 | xfs_buf_t *bp; | |
3751 | xfs_sb_t *sbp; | |
3752 | ||
3753 | /* | |
3754 | * First replay the images in the log. | |
3755 | */ | |
3756 | error = xlog_do_log_recovery(log, head_blk, tail_blk); | |
43ff2122 | 3757 | if (error) |
1da177e4 | 3758 | return error; |
1da177e4 LT |
3759 | |
3760 | /* | |
3761 | * If IO errors happened during recovery, bail out. | |
3762 | */ | |
3763 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) { | |
3764 | return (EIO); | |
3765 | } | |
3766 | ||
3767 | /* | |
3768 | * We now update the tail_lsn since much of the recovery has completed | |
3769 | * and there may be space available to use. If there were no extent | |
3770 | * or iunlinks, we can free up the entire log and set the tail_lsn to | |
3771 | * be the last_sync_lsn. This was set in xlog_find_tail to be the | |
3772 | * lsn of the last known good LR on disk. If there are extent frees | |
3773 | * or iunlinks they will have some entries in the AIL; so we look at | |
3774 | * the AIL to determine how to set the tail_lsn. | |
3775 | */ | |
3776 | xlog_assign_tail_lsn(log->l_mp); | |
3777 | ||
3778 | /* | |
3779 | * Now that we've finished replaying all buffer and inode | |
98021821 | 3780 | * updates, re-read in the superblock and reverify it. |
1da177e4 LT |
3781 | */ |
3782 | bp = xfs_getsb(log->l_mp, 0); | |
3783 | XFS_BUF_UNDONE(bp); | |
bebf963f | 3784 | ASSERT(!(XFS_BUF_ISWRITE(bp))); |
1da177e4 | 3785 | XFS_BUF_READ(bp); |
bebf963f | 3786 | XFS_BUF_UNASYNC(bp); |
1813dd64 | 3787 | bp->b_ops = &xfs_sb_buf_ops; |
1da177e4 | 3788 | xfsbdstrat(log->l_mp, bp); |
1a1a3e97 | 3789 | error = xfs_buf_iowait(bp); |
d64e31a2 | 3790 | if (error) { |
901796af | 3791 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
3792 | ASSERT(0); |
3793 | xfs_buf_relse(bp); | |
3794 | return error; | |
3795 | } | |
3796 | ||
3797 | /* Convert superblock from on-disk format */ | |
3798 | sbp = &log->l_mp->m_sb; | |
98021821 | 3799 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 3800 | ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC); |
62118709 | 3801 | ASSERT(xfs_sb_good_version(sbp)); |
1da177e4 LT |
3802 | xfs_buf_relse(bp); |
3803 | ||
5478eead LM |
3804 | /* We've re-read the superblock so re-initialize per-cpu counters */ |
3805 | xfs_icsb_reinit_counters(log->l_mp); | |
3806 | ||
1da177e4 LT |
3807 | xlog_recover_check_summary(log); |
3808 | ||
3809 | /* Normal transactions can now occur */ | |
3810 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
3811 | return 0; | |
3812 | } | |
3813 | ||
3814 | /* | |
3815 | * Perform recovery and re-initialize some log variables in xlog_find_tail. | |
3816 | * | |
3817 | * Return error or zero. | |
3818 | */ | |
3819 | int | |
3820 | xlog_recover( | |
9a8d2fdb | 3821 | struct xlog *log) |
1da177e4 LT |
3822 | { |
3823 | xfs_daddr_t head_blk, tail_blk; | |
3824 | int error; | |
3825 | ||
3826 | /* find the tail of the log */ | |
65be6054 | 3827 | if ((error = xlog_find_tail(log, &head_blk, &tail_blk))) |
1da177e4 LT |
3828 | return error; |
3829 | ||
3830 | if (tail_blk != head_blk) { | |
3831 | /* There used to be a comment here: | |
3832 | * | |
3833 | * disallow recovery on read-only mounts. note -- mount | |
3834 | * checks for ENOSPC and turns it into an intelligent | |
3835 | * error message. | |
3836 | * ...but this is no longer true. Now, unless you specify | |
3837 | * NORECOVERY (in which case this function would never be | |
3838 | * called), we just go ahead and recover. We do this all | |
3839 | * under the vfs layer, so we can get away with it unless | |
3840 | * the device itself is read-only, in which case we fail. | |
3841 | */ | |
3a02ee18 | 3842 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
1da177e4 LT |
3843 | return error; |
3844 | } | |
3845 | ||
a0fa2b67 DC |
3846 | xfs_notice(log->l_mp, "Starting recovery (logdev: %s)", |
3847 | log->l_mp->m_logname ? log->l_mp->m_logname | |
3848 | : "internal"); | |
1da177e4 LT |
3849 | |
3850 | error = xlog_do_recover(log, head_blk, tail_blk); | |
3851 | log->l_flags |= XLOG_RECOVERY_NEEDED; | |
3852 | } | |
3853 | return error; | |
3854 | } | |
3855 | ||
3856 | /* | |
3857 | * In the first part of recovery we replay inodes and buffers and build | |
3858 | * up the list of extent free items which need to be processed. Here | |
3859 | * we process the extent free items and clean up the on disk unlinked | |
3860 | * inode lists. This is separated from the first part of recovery so | |
3861 | * that the root and real-time bitmap inodes can be read in from disk in | |
3862 | * between the two stages. This is necessary so that we can free space | |
3863 | * in the real-time portion of the file system. | |
3864 | */ | |
3865 | int | |
3866 | xlog_recover_finish( | |
9a8d2fdb | 3867 | struct xlog *log) |
1da177e4 LT |
3868 | { |
3869 | /* | |
3870 | * Now we're ready to do the transactions needed for the | |
3871 | * rest of recovery. Start with completing all the extent | |
3872 | * free intent records and then process the unlinked inode | |
3873 | * lists. At this point, we essentially run in normal mode | |
3874 | * except that we're still performing recovery actions | |
3875 | * rather than accepting new requests. | |
3876 | */ | |
3877 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { | |
3c1e2bbe DC |
3878 | int error; |
3879 | error = xlog_recover_process_efis(log); | |
3880 | if (error) { | |
a0fa2b67 | 3881 | xfs_alert(log->l_mp, "Failed to recover EFIs"); |
3c1e2bbe DC |
3882 | return error; |
3883 | } | |
1da177e4 LT |
3884 | /* |
3885 | * Sync the log to get all the EFIs out of the AIL. | |
3886 | * This isn't absolutely necessary, but it helps in | |
3887 | * case the unlink transactions would have problems | |
3888 | * pushing the EFIs out of the way. | |
3889 | */ | |
a14a348b | 3890 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
1da177e4 | 3891 | |
4249023a | 3892 | xlog_recover_process_iunlinks(log); |
1da177e4 LT |
3893 | |
3894 | xlog_recover_check_summary(log); | |
3895 | ||
a0fa2b67 DC |
3896 | xfs_notice(log->l_mp, "Ending recovery (logdev: %s)", |
3897 | log->l_mp->m_logname ? log->l_mp->m_logname | |
3898 | : "internal"); | |
1da177e4 LT |
3899 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
3900 | } else { | |
a0fa2b67 | 3901 | xfs_info(log->l_mp, "Ending clean mount"); |
1da177e4 LT |
3902 | } |
3903 | return 0; | |
3904 | } | |
3905 | ||
3906 | ||
3907 | #if defined(DEBUG) | |
3908 | /* | |
3909 | * Read all of the agf and agi counters and check that they | |
3910 | * are consistent with the superblock counters. | |
3911 | */ | |
3912 | void | |
3913 | xlog_recover_check_summary( | |
9a8d2fdb | 3914 | struct xlog *log) |
1da177e4 LT |
3915 | { |
3916 | xfs_mount_t *mp; | |
3917 | xfs_agf_t *agfp; | |
1da177e4 LT |
3918 | xfs_buf_t *agfbp; |
3919 | xfs_buf_t *agibp; | |
1da177e4 LT |
3920 | xfs_agnumber_t agno; |
3921 | __uint64_t freeblks; | |
3922 | __uint64_t itotal; | |
3923 | __uint64_t ifree; | |
5e1be0fb | 3924 | int error; |
1da177e4 LT |
3925 | |
3926 | mp = log->l_mp; | |
3927 | ||
3928 | freeblks = 0LL; | |
3929 | itotal = 0LL; | |
3930 | ifree = 0LL; | |
3931 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
4805621a FCH |
3932 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
3933 | if (error) { | |
a0fa2b67 DC |
3934 | xfs_alert(mp, "%s agf read failed agno %d error %d", |
3935 | __func__, agno, error); | |
4805621a FCH |
3936 | } else { |
3937 | agfp = XFS_BUF_TO_AGF(agfbp); | |
3938 | freeblks += be32_to_cpu(agfp->agf_freeblks) + | |
3939 | be32_to_cpu(agfp->agf_flcount); | |
3940 | xfs_buf_relse(agfbp); | |
1da177e4 | 3941 | } |
1da177e4 | 3942 | |
5e1be0fb | 3943 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
a0fa2b67 DC |
3944 | if (error) { |
3945 | xfs_alert(mp, "%s agi read failed agno %d error %d", | |
3946 | __func__, agno, error); | |
3947 | } else { | |
5e1be0fb | 3948 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 3949 | |
5e1be0fb CH |
3950 | itotal += be32_to_cpu(agi->agi_count); |
3951 | ifree += be32_to_cpu(agi->agi_freecount); | |
3952 | xfs_buf_relse(agibp); | |
3953 | } | |
1da177e4 | 3954 | } |
1da177e4 LT |
3955 | } |
3956 | #endif /* DEBUG */ |