]>
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" |
1da177e4 | 48 | |
9a8d2fdb MT |
49 | STATIC int |
50 | xlog_find_zeroed( | |
51 | struct xlog *, | |
52 | xfs_daddr_t *); | |
53 | STATIC int | |
54 | xlog_clear_stale_blocks( | |
55 | struct xlog *, | |
56 | xfs_lsn_t); | |
1da177e4 | 57 | #if defined(DEBUG) |
9a8d2fdb MT |
58 | STATIC void |
59 | xlog_recover_check_summary( | |
60 | struct xlog *); | |
1da177e4 LT |
61 | #else |
62 | #define xlog_recover_check_summary(log) | |
1da177e4 LT |
63 | #endif |
64 | ||
d5689eaa CH |
65 | /* |
66 | * This structure is used during recovery to record the buf log items which | |
67 | * have been canceled and should not be replayed. | |
68 | */ | |
69 | struct xfs_buf_cancel { | |
70 | xfs_daddr_t bc_blkno; | |
71 | uint bc_len; | |
72 | int bc_refcount; | |
73 | struct list_head bc_list; | |
74 | }; | |
75 | ||
1da177e4 LT |
76 | /* |
77 | * Sector aligned buffer routines for buffer create/read/write/access | |
78 | */ | |
79 | ||
ff30a622 AE |
80 | /* |
81 | * Verify the given count of basic blocks is valid number of blocks | |
82 | * to specify for an operation involving the given XFS log buffer. | |
83 | * Returns nonzero if the count is valid, 0 otherwise. | |
84 | */ | |
85 | ||
86 | static inline int | |
87 | xlog_buf_bbcount_valid( | |
9a8d2fdb | 88 | struct xlog *log, |
ff30a622 AE |
89 | int bbcount) |
90 | { | |
91 | return bbcount > 0 && bbcount <= log->l_logBBsize; | |
92 | } | |
93 | ||
36adecff AE |
94 | /* |
95 | * Allocate a buffer to hold log data. The buffer needs to be able | |
96 | * to map to a range of nbblks basic blocks at any valid (basic | |
97 | * block) offset within the log. | |
98 | */ | |
5d77c0dc | 99 | STATIC xfs_buf_t * |
1da177e4 | 100 | xlog_get_bp( |
9a8d2fdb | 101 | struct xlog *log, |
3228149c | 102 | int nbblks) |
1da177e4 | 103 | { |
c8da0faf CH |
104 | struct xfs_buf *bp; |
105 | ||
ff30a622 | 106 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 107 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
108 | nbblks); |
109 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
110 | return NULL; |
111 | } | |
1da177e4 | 112 | |
36adecff AE |
113 | /* |
114 | * We do log I/O in units of log sectors (a power-of-2 | |
115 | * multiple of the basic block size), so we round up the | |
25985edc | 116 | * requested size to accommodate the basic blocks required |
36adecff AE |
117 | * for complete log sectors. |
118 | * | |
119 | * In addition, the buffer may be used for a non-sector- | |
120 | * aligned block offset, in which case an I/O of the | |
121 | * requested size could extend beyond the end of the | |
122 | * buffer. If the requested size is only 1 basic block it | |
123 | * will never straddle a sector boundary, so this won't be | |
124 | * an issue. Nor will this be a problem if the log I/O is | |
125 | * done in basic blocks (sector size 1). But otherwise we | |
126 | * extend the buffer by one extra log sector to ensure | |
25985edc | 127 | * there's space to accommodate this possibility. |
36adecff | 128 | */ |
69ce58f0 AE |
129 | if (nbblks > 1 && log->l_sectBBsize > 1) |
130 | nbblks += log->l_sectBBsize; | |
131 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
36adecff | 132 | |
e70b73f8 | 133 | bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0); |
c8da0faf CH |
134 | if (bp) |
135 | xfs_buf_unlock(bp); | |
136 | return bp; | |
1da177e4 LT |
137 | } |
138 | ||
5d77c0dc | 139 | STATIC void |
1da177e4 LT |
140 | xlog_put_bp( |
141 | xfs_buf_t *bp) | |
142 | { | |
143 | xfs_buf_free(bp); | |
144 | } | |
145 | ||
48389ef1 AE |
146 | /* |
147 | * Return the address of the start of the given block number's data | |
148 | * in a log buffer. The buffer covers a log sector-aligned region. | |
149 | */ | |
076e6acb CH |
150 | STATIC xfs_caddr_t |
151 | xlog_align( | |
9a8d2fdb | 152 | struct xlog *log, |
076e6acb CH |
153 | xfs_daddr_t blk_no, |
154 | int nbblks, | |
9a8d2fdb | 155 | struct xfs_buf *bp) |
076e6acb | 156 | { |
fdc07f44 | 157 | xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1); |
076e6acb | 158 | |
4e94b71b | 159 | ASSERT(offset + nbblks <= bp->b_length); |
62926044 | 160 | return bp->b_addr + BBTOB(offset); |
076e6acb CH |
161 | } |
162 | ||
1da177e4 LT |
163 | |
164 | /* | |
165 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. | |
166 | */ | |
076e6acb CH |
167 | STATIC int |
168 | xlog_bread_noalign( | |
9a8d2fdb | 169 | struct xlog *log, |
1da177e4 LT |
170 | xfs_daddr_t blk_no, |
171 | int nbblks, | |
9a8d2fdb | 172 | struct xfs_buf *bp) |
1da177e4 LT |
173 | { |
174 | int error; | |
175 | ||
ff30a622 | 176 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 177 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
178 | nbblks); |
179 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
180 | return EFSCORRUPTED; |
181 | } | |
182 | ||
69ce58f0 AE |
183 | blk_no = round_down(blk_no, log->l_sectBBsize); |
184 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
185 | |
186 | ASSERT(nbblks > 0); | |
4e94b71b | 187 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
188 | |
189 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
190 | XFS_BUF_READ(bp); | |
aa0e8833 | 191 | bp->b_io_length = nbblks; |
0e95f19a | 192 | bp->b_error = 0; |
1da177e4 LT |
193 | |
194 | xfsbdstrat(log->l_mp, bp); | |
1a1a3e97 | 195 | error = xfs_buf_iowait(bp); |
d64e31a2 | 196 | if (error) |
901796af | 197 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
198 | return error; |
199 | } | |
200 | ||
076e6acb CH |
201 | STATIC int |
202 | xlog_bread( | |
9a8d2fdb | 203 | struct xlog *log, |
076e6acb CH |
204 | xfs_daddr_t blk_no, |
205 | int nbblks, | |
9a8d2fdb | 206 | struct xfs_buf *bp, |
076e6acb CH |
207 | xfs_caddr_t *offset) |
208 | { | |
209 | int error; | |
210 | ||
211 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
212 | if (error) | |
213 | return error; | |
214 | ||
215 | *offset = xlog_align(log, blk_no, nbblks, bp); | |
216 | return 0; | |
217 | } | |
218 | ||
44396476 DC |
219 | /* |
220 | * Read at an offset into the buffer. Returns with the buffer in it's original | |
221 | * state regardless of the result of the read. | |
222 | */ | |
223 | STATIC int | |
224 | xlog_bread_offset( | |
9a8d2fdb | 225 | struct xlog *log, |
44396476 DC |
226 | xfs_daddr_t blk_no, /* block to read from */ |
227 | int nbblks, /* blocks to read */ | |
9a8d2fdb | 228 | struct xfs_buf *bp, |
44396476 DC |
229 | xfs_caddr_t offset) |
230 | { | |
62926044 | 231 | xfs_caddr_t orig_offset = bp->b_addr; |
4e94b71b | 232 | int orig_len = BBTOB(bp->b_length); |
44396476 DC |
233 | int error, error2; |
234 | ||
02fe03d9 | 235 | error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks)); |
44396476 DC |
236 | if (error) |
237 | return error; | |
238 | ||
239 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
240 | ||
241 | /* must reset buffer pointer even on error */ | |
02fe03d9 | 242 | error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len); |
44396476 DC |
243 | if (error) |
244 | return error; | |
245 | return error2; | |
246 | } | |
247 | ||
1da177e4 LT |
248 | /* |
249 | * Write out the buffer at the given block for the given number of blocks. | |
250 | * The buffer is kept locked across the write and is returned locked. | |
251 | * This can only be used for synchronous log writes. | |
252 | */ | |
ba0f32d4 | 253 | STATIC int |
1da177e4 | 254 | xlog_bwrite( |
9a8d2fdb | 255 | struct xlog *log, |
1da177e4 LT |
256 | xfs_daddr_t blk_no, |
257 | int nbblks, | |
9a8d2fdb | 258 | struct xfs_buf *bp) |
1da177e4 LT |
259 | { |
260 | int error; | |
261 | ||
ff30a622 | 262 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 263 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
264 | nbblks); |
265 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
266 | return EFSCORRUPTED; |
267 | } | |
268 | ||
69ce58f0 AE |
269 | blk_no = round_down(blk_no, log->l_sectBBsize); |
270 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
271 | |
272 | ASSERT(nbblks > 0); | |
4e94b71b | 273 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
274 | |
275 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
276 | XFS_BUF_ZEROFLAGS(bp); | |
72790aa1 | 277 | xfs_buf_hold(bp); |
0c842ad4 | 278 | xfs_buf_lock(bp); |
aa0e8833 | 279 | bp->b_io_length = nbblks; |
0e95f19a | 280 | bp->b_error = 0; |
1da177e4 | 281 | |
c2b006c1 | 282 | error = xfs_bwrite(bp); |
901796af CH |
283 | if (error) |
284 | xfs_buf_ioerror_alert(bp, __func__); | |
c2b006c1 | 285 | xfs_buf_relse(bp); |
1da177e4 LT |
286 | return error; |
287 | } | |
288 | ||
1da177e4 LT |
289 | #ifdef DEBUG |
290 | /* | |
291 | * dump debug superblock and log record information | |
292 | */ | |
293 | STATIC void | |
294 | xlog_header_check_dump( | |
295 | xfs_mount_t *mp, | |
296 | xlog_rec_header_t *head) | |
297 | { | |
a0fa2b67 | 298 | xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n", |
03daa57c | 299 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); |
a0fa2b67 | 300 | xfs_debug(mp, " log : uuid = %pU, fmt = %d\n", |
03daa57c | 301 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); |
1da177e4 LT |
302 | } |
303 | #else | |
304 | #define xlog_header_check_dump(mp, head) | |
305 | #endif | |
306 | ||
307 | /* | |
308 | * check log record header for recovery | |
309 | */ | |
310 | STATIC int | |
311 | xlog_header_check_recover( | |
312 | xfs_mount_t *mp, | |
313 | xlog_rec_header_t *head) | |
314 | { | |
69ef921b | 315 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 LT |
316 | |
317 | /* | |
318 | * IRIX doesn't write the h_fmt field and leaves it zeroed | |
319 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover | |
320 | * a dirty log created in IRIX. | |
321 | */ | |
69ef921b | 322 | if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) { |
a0fa2b67 DC |
323 | xfs_warn(mp, |
324 | "dirty log written in incompatible format - can't recover"); | |
1da177e4 LT |
325 | xlog_header_check_dump(mp, head); |
326 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", | |
327 | XFS_ERRLEVEL_HIGH, mp); | |
328 | return XFS_ERROR(EFSCORRUPTED); | |
329 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
a0fa2b67 DC |
330 | xfs_warn(mp, |
331 | "dirty log entry has mismatched uuid - can't recover"); | |
1da177e4 LT |
332 | xlog_header_check_dump(mp, head); |
333 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", | |
334 | XFS_ERRLEVEL_HIGH, mp); | |
335 | return XFS_ERROR(EFSCORRUPTED); | |
336 | } | |
337 | return 0; | |
338 | } | |
339 | ||
340 | /* | |
341 | * read the head block of the log and check the header | |
342 | */ | |
343 | STATIC int | |
344 | xlog_header_check_mount( | |
345 | xfs_mount_t *mp, | |
346 | xlog_rec_header_t *head) | |
347 | { | |
69ef921b | 348 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 LT |
349 | |
350 | if (uuid_is_nil(&head->h_fs_uuid)) { | |
351 | /* | |
352 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If | |
353 | * h_fs_uuid is nil, we assume this log was last mounted | |
354 | * by IRIX and continue. | |
355 | */ | |
a0fa2b67 | 356 | xfs_warn(mp, "nil uuid in log - IRIX style log"); |
1da177e4 | 357 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
a0fa2b67 | 358 | xfs_warn(mp, "log has mismatched uuid - can't recover"); |
1da177e4 LT |
359 | xlog_header_check_dump(mp, head); |
360 | XFS_ERROR_REPORT("xlog_header_check_mount", | |
361 | XFS_ERRLEVEL_HIGH, mp); | |
362 | return XFS_ERROR(EFSCORRUPTED); | |
363 | } | |
364 | return 0; | |
365 | } | |
366 | ||
367 | STATIC void | |
368 | xlog_recover_iodone( | |
369 | struct xfs_buf *bp) | |
370 | { | |
5a52c2a5 | 371 | if (bp->b_error) { |
1da177e4 LT |
372 | /* |
373 | * We're not going to bother about retrying | |
374 | * this during recovery. One strike! | |
375 | */ | |
901796af | 376 | xfs_buf_ioerror_alert(bp, __func__); |
ebad861b DC |
377 | xfs_force_shutdown(bp->b_target->bt_mount, |
378 | SHUTDOWN_META_IO_ERROR); | |
1da177e4 | 379 | } |
cb669ca5 | 380 | bp->b_iodone = NULL; |
1a1a3e97 | 381 | xfs_buf_ioend(bp, 0); |
1da177e4 LT |
382 | } |
383 | ||
384 | /* | |
385 | * This routine finds (to an approximation) the first block in the physical | |
386 | * log which contains the given cycle. It uses a binary search algorithm. | |
387 | * Note that the algorithm can not be perfect because the disk will not | |
388 | * necessarily be perfect. | |
389 | */ | |
a8272ce0 | 390 | STATIC int |
1da177e4 | 391 | xlog_find_cycle_start( |
9a8d2fdb MT |
392 | struct xlog *log, |
393 | struct xfs_buf *bp, | |
1da177e4 LT |
394 | xfs_daddr_t first_blk, |
395 | xfs_daddr_t *last_blk, | |
396 | uint cycle) | |
397 | { | |
398 | xfs_caddr_t offset; | |
399 | xfs_daddr_t mid_blk; | |
e3bb2e30 | 400 | xfs_daddr_t end_blk; |
1da177e4 LT |
401 | uint mid_cycle; |
402 | int error; | |
403 | ||
e3bb2e30 AE |
404 | end_blk = *last_blk; |
405 | mid_blk = BLK_AVG(first_blk, end_blk); | |
406 | while (mid_blk != first_blk && mid_blk != end_blk) { | |
076e6acb CH |
407 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
408 | if (error) | |
1da177e4 | 409 | return error; |
03bea6fe | 410 | mid_cycle = xlog_get_cycle(offset); |
e3bb2e30 AE |
411 | if (mid_cycle == cycle) |
412 | end_blk = mid_blk; /* last_half_cycle == mid_cycle */ | |
413 | else | |
414 | first_blk = mid_blk; /* first_half_cycle == mid_cycle */ | |
415 | mid_blk = BLK_AVG(first_blk, end_blk); | |
1da177e4 | 416 | } |
e3bb2e30 AE |
417 | ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) || |
418 | (mid_blk == end_blk && mid_blk-1 == first_blk)); | |
419 | ||
420 | *last_blk = end_blk; | |
1da177e4 LT |
421 | |
422 | return 0; | |
423 | } | |
424 | ||
425 | /* | |
3f943d85 AE |
426 | * Check that a range of blocks does not contain stop_on_cycle_no. |
427 | * Fill in *new_blk with the block offset where such a block is | |
428 | * found, or with -1 (an invalid block number) if there is no such | |
429 | * block in the range. The scan needs to occur from front to back | |
430 | * and the pointer into the region must be updated since a later | |
431 | * routine will need to perform another test. | |
1da177e4 LT |
432 | */ |
433 | STATIC int | |
434 | xlog_find_verify_cycle( | |
9a8d2fdb | 435 | struct xlog *log, |
1da177e4 LT |
436 | xfs_daddr_t start_blk, |
437 | int nbblks, | |
438 | uint stop_on_cycle_no, | |
439 | xfs_daddr_t *new_blk) | |
440 | { | |
441 | xfs_daddr_t i, j; | |
442 | uint cycle; | |
443 | xfs_buf_t *bp; | |
444 | xfs_daddr_t bufblks; | |
445 | xfs_caddr_t buf = NULL; | |
446 | int error = 0; | |
447 | ||
6881a229 AE |
448 | /* |
449 | * Greedily allocate a buffer big enough to handle the full | |
450 | * range of basic blocks we'll be examining. If that fails, | |
451 | * try a smaller size. We need to be able to read at least | |
452 | * a log sector, or we're out of luck. | |
453 | */ | |
1da177e4 | 454 | bufblks = 1 << ffs(nbblks); |
81158e0c DC |
455 | while (bufblks > log->l_logBBsize) |
456 | bufblks >>= 1; | |
1da177e4 | 457 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1da177e4 | 458 | bufblks >>= 1; |
69ce58f0 | 459 | if (bufblks < log->l_sectBBsize) |
1da177e4 LT |
460 | return ENOMEM; |
461 | } | |
462 | ||
463 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { | |
464 | int bcount; | |
465 | ||
466 | bcount = min(bufblks, (start_blk + nbblks - i)); | |
467 | ||
076e6acb CH |
468 | error = xlog_bread(log, i, bcount, bp, &buf); |
469 | if (error) | |
1da177e4 LT |
470 | goto out; |
471 | ||
1da177e4 | 472 | for (j = 0; j < bcount; j++) { |
03bea6fe | 473 | cycle = xlog_get_cycle(buf); |
1da177e4 LT |
474 | if (cycle == stop_on_cycle_no) { |
475 | *new_blk = i+j; | |
476 | goto out; | |
477 | } | |
478 | ||
479 | buf += BBSIZE; | |
480 | } | |
481 | } | |
482 | ||
483 | *new_blk = -1; | |
484 | ||
485 | out: | |
486 | xlog_put_bp(bp); | |
487 | return error; | |
488 | } | |
489 | ||
490 | /* | |
491 | * Potentially backup over partial log record write. | |
492 | * | |
493 | * In the typical case, last_blk is the number of the block directly after | |
494 | * a good log record. Therefore, we subtract one to get the block number | |
495 | * of the last block in the given buffer. extra_bblks contains the number | |
496 | * of blocks we would have read on a previous read. This happens when the | |
497 | * last log record is split over the end of the physical log. | |
498 | * | |
499 | * extra_bblks is the number of blocks potentially verified on a previous | |
500 | * call to this routine. | |
501 | */ | |
502 | STATIC int | |
503 | xlog_find_verify_log_record( | |
9a8d2fdb | 504 | struct xlog *log, |
1da177e4 LT |
505 | xfs_daddr_t start_blk, |
506 | xfs_daddr_t *last_blk, | |
507 | int extra_bblks) | |
508 | { | |
509 | xfs_daddr_t i; | |
510 | xfs_buf_t *bp; | |
511 | xfs_caddr_t offset = NULL; | |
512 | xlog_rec_header_t *head = NULL; | |
513 | int error = 0; | |
514 | int smallmem = 0; | |
515 | int num_blks = *last_blk - start_blk; | |
516 | int xhdrs; | |
517 | ||
518 | ASSERT(start_blk != 0 || *last_blk != start_blk); | |
519 | ||
520 | if (!(bp = xlog_get_bp(log, num_blks))) { | |
521 | if (!(bp = xlog_get_bp(log, 1))) | |
522 | return ENOMEM; | |
523 | smallmem = 1; | |
524 | } else { | |
076e6acb CH |
525 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
526 | if (error) | |
1da177e4 | 527 | goto out; |
1da177e4 LT |
528 | offset += ((num_blks - 1) << BBSHIFT); |
529 | } | |
530 | ||
531 | for (i = (*last_blk) - 1; i >= 0; i--) { | |
532 | if (i < start_blk) { | |
533 | /* valid log record not found */ | |
a0fa2b67 DC |
534 | xfs_warn(log->l_mp, |
535 | "Log inconsistent (didn't find previous header)"); | |
1da177e4 LT |
536 | ASSERT(0); |
537 | error = XFS_ERROR(EIO); | |
538 | goto out; | |
539 | } | |
540 | ||
541 | if (smallmem) { | |
076e6acb CH |
542 | error = xlog_bread(log, i, 1, bp, &offset); |
543 | if (error) | |
1da177e4 | 544 | goto out; |
1da177e4 LT |
545 | } |
546 | ||
547 | head = (xlog_rec_header_t *)offset; | |
548 | ||
69ef921b | 549 | if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
1da177e4 LT |
550 | break; |
551 | ||
552 | if (!smallmem) | |
553 | offset -= BBSIZE; | |
554 | } | |
555 | ||
556 | /* | |
557 | * We hit the beginning of the physical log & still no header. Return | |
558 | * to caller. If caller can handle a return of -1, then this routine | |
559 | * will be called again for the end of the physical log. | |
560 | */ | |
561 | if (i == -1) { | |
562 | error = -1; | |
563 | goto out; | |
564 | } | |
565 | ||
566 | /* | |
567 | * We have the final block of the good log (the first block | |
568 | * of the log record _before_ the head. So we check the uuid. | |
569 | */ | |
570 | if ((error = xlog_header_check_mount(log->l_mp, head))) | |
571 | goto out; | |
572 | ||
573 | /* | |
574 | * We may have found a log record header before we expected one. | |
575 | * last_blk will be the 1st block # with a given cycle #. We may end | |
576 | * up reading an entire log record. In this case, we don't want to | |
577 | * reset last_blk. Only when last_blk points in the middle of a log | |
578 | * record do we update last_blk. | |
579 | */ | |
62118709 | 580 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d | 581 | uint h_size = be32_to_cpu(head->h_size); |
1da177e4 LT |
582 | |
583 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; | |
584 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
585 | xhdrs++; | |
586 | } else { | |
587 | xhdrs = 1; | |
588 | } | |
589 | ||
b53e675d CH |
590 | if (*last_blk - i + extra_bblks != |
591 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) | |
1da177e4 LT |
592 | *last_blk = i; |
593 | ||
594 | out: | |
595 | xlog_put_bp(bp); | |
596 | return error; | |
597 | } | |
598 | ||
599 | /* | |
600 | * Head is defined to be the point of the log where the next log write | |
601 | * write could go. This means that incomplete LR writes at the end are | |
602 | * eliminated when calculating the head. We aren't guaranteed that previous | |
603 | * LR have complete transactions. We only know that a cycle number of | |
604 | * current cycle number -1 won't be present in the log if we start writing | |
605 | * from our current block number. | |
606 | * | |
607 | * last_blk contains the block number of the first block with a given | |
608 | * cycle number. | |
609 | * | |
610 | * Return: zero if normal, non-zero if error. | |
611 | */ | |
ba0f32d4 | 612 | STATIC int |
1da177e4 | 613 | xlog_find_head( |
9a8d2fdb | 614 | struct xlog *log, |
1da177e4 LT |
615 | xfs_daddr_t *return_head_blk) |
616 | { | |
617 | xfs_buf_t *bp; | |
618 | xfs_caddr_t offset; | |
619 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; | |
620 | int num_scan_bblks; | |
621 | uint first_half_cycle, last_half_cycle; | |
622 | uint stop_on_cycle; | |
623 | int error, log_bbnum = log->l_logBBsize; | |
624 | ||
625 | /* Is the end of the log device zeroed? */ | |
626 | if ((error = xlog_find_zeroed(log, &first_blk)) == -1) { | |
627 | *return_head_blk = first_blk; | |
628 | ||
629 | /* Is the whole lot zeroed? */ | |
630 | if (!first_blk) { | |
631 | /* Linux XFS shouldn't generate totally zeroed logs - | |
632 | * mkfs etc write a dummy unmount record to a fresh | |
633 | * log so we can store the uuid in there | |
634 | */ | |
a0fa2b67 | 635 | xfs_warn(log->l_mp, "totally zeroed log"); |
1da177e4 LT |
636 | } |
637 | ||
638 | return 0; | |
639 | } else if (error) { | |
a0fa2b67 | 640 | xfs_warn(log->l_mp, "empty log check failed"); |
1da177e4 LT |
641 | return error; |
642 | } | |
643 | ||
644 | first_blk = 0; /* get cycle # of 1st block */ | |
645 | bp = xlog_get_bp(log, 1); | |
646 | if (!bp) | |
647 | return ENOMEM; | |
076e6acb CH |
648 | |
649 | error = xlog_bread(log, 0, 1, bp, &offset); | |
650 | if (error) | |
1da177e4 | 651 | goto bp_err; |
076e6acb | 652 | |
03bea6fe | 653 | first_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
654 | |
655 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ | |
076e6acb CH |
656 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
657 | if (error) | |
1da177e4 | 658 | goto bp_err; |
076e6acb | 659 | |
03bea6fe | 660 | last_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
661 | ASSERT(last_half_cycle != 0); |
662 | ||
663 | /* | |
664 | * If the 1st half cycle number is equal to the last half cycle number, | |
665 | * then the entire log is stamped with the same cycle number. In this | |
666 | * case, head_blk can't be set to zero (which makes sense). The below | |
667 | * math doesn't work out properly with head_blk equal to zero. Instead, | |
668 | * we set it to log_bbnum which is an invalid block number, but this | |
669 | * value makes the math correct. If head_blk doesn't changed through | |
670 | * all the tests below, *head_blk is set to zero at the very end rather | |
671 | * than log_bbnum. In a sense, log_bbnum and zero are the same block | |
672 | * in a circular file. | |
673 | */ | |
674 | if (first_half_cycle == last_half_cycle) { | |
675 | /* | |
676 | * In this case we believe that the entire log should have | |
677 | * cycle number last_half_cycle. We need to scan backwards | |
678 | * from the end verifying that there are no holes still | |
679 | * containing last_half_cycle - 1. If we find such a hole, | |
680 | * then the start of that hole will be the new head. The | |
681 | * simple case looks like | |
682 | * x | x ... | x - 1 | x | |
683 | * Another case that fits this picture would be | |
684 | * x | x + 1 | x ... | x | |
c41564b5 | 685 | * In this case the head really is somewhere at the end of the |
1da177e4 LT |
686 | * log, as one of the latest writes at the beginning was |
687 | * incomplete. | |
688 | * One more case is | |
689 | * x | x + 1 | x ... | x - 1 | x | |
690 | * This is really the combination of the above two cases, and | |
691 | * the head has to end up at the start of the x-1 hole at the | |
692 | * end of the log. | |
693 | * | |
694 | * In the 256k log case, we will read from the beginning to the | |
695 | * end of the log and search for cycle numbers equal to x-1. | |
696 | * We don't worry about the x+1 blocks that we encounter, | |
697 | * because we know that they cannot be the head since the log | |
698 | * started with x. | |
699 | */ | |
700 | head_blk = log_bbnum; | |
701 | stop_on_cycle = last_half_cycle - 1; | |
702 | } else { | |
703 | /* | |
704 | * In this case we want to find the first block with cycle | |
705 | * number matching last_half_cycle. We expect the log to be | |
706 | * some variation on | |
3f943d85 | 707 | * x + 1 ... | x ... | x |
1da177e4 LT |
708 | * The first block with cycle number x (last_half_cycle) will |
709 | * be where the new head belongs. First we do a binary search | |
710 | * for the first occurrence of last_half_cycle. The binary | |
711 | * search may not be totally accurate, so then we scan back | |
712 | * from there looking for occurrences of last_half_cycle before | |
713 | * us. If that backwards scan wraps around the beginning of | |
714 | * the log, then we look for occurrences of last_half_cycle - 1 | |
715 | * at the end of the log. The cases we're looking for look | |
716 | * like | |
3f943d85 AE |
717 | * v binary search stopped here |
718 | * x + 1 ... | x | x + 1 | x ... | x | |
719 | * ^ but we want to locate this spot | |
1da177e4 | 720 | * or |
1da177e4 | 721 | * <---------> less than scan distance |
3f943d85 AE |
722 | * x + 1 ... | x ... | x - 1 | x |
723 | * ^ we want to locate this spot | |
1da177e4 LT |
724 | */ |
725 | stop_on_cycle = last_half_cycle; | |
726 | if ((error = xlog_find_cycle_start(log, bp, first_blk, | |
727 | &head_blk, last_half_cycle))) | |
728 | goto bp_err; | |
729 | } | |
730 | ||
731 | /* | |
732 | * Now validate the answer. Scan back some number of maximum possible | |
733 | * blocks and make sure each one has the expected cycle number. The | |
734 | * maximum is determined by the total possible amount of buffering | |
735 | * in the in-core log. The following number can be made tighter if | |
736 | * we actually look at the block size of the filesystem. | |
737 | */ | |
738 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
739 | if (head_blk >= num_scan_bblks) { | |
740 | /* | |
741 | * We are guaranteed that the entire check can be performed | |
742 | * in one buffer. | |
743 | */ | |
744 | start_blk = head_blk - num_scan_bblks; | |
745 | if ((error = xlog_find_verify_cycle(log, | |
746 | start_blk, num_scan_bblks, | |
747 | stop_on_cycle, &new_blk))) | |
748 | goto bp_err; | |
749 | if (new_blk != -1) | |
750 | head_blk = new_blk; | |
751 | } else { /* need to read 2 parts of log */ | |
752 | /* | |
753 | * We are going to scan backwards in the log in two parts. | |
754 | * First we scan the physical end of the log. In this part | |
755 | * of the log, we are looking for blocks with cycle number | |
756 | * last_half_cycle - 1. | |
757 | * If we find one, then we know that the log starts there, as | |
758 | * we've found a hole that didn't get written in going around | |
759 | * the end of the physical log. The simple case for this is | |
760 | * x + 1 ... | x ... | x - 1 | x | |
761 | * <---------> less than scan distance | |
762 | * If all of the blocks at the end of the log have cycle number | |
763 | * last_half_cycle, then we check the blocks at the start of | |
764 | * the log looking for occurrences of last_half_cycle. If we | |
765 | * find one, then our current estimate for the location of the | |
766 | * first occurrence of last_half_cycle is wrong and we move | |
767 | * back to the hole we've found. This case looks like | |
768 | * x + 1 ... | x | x + 1 | x ... | |
769 | * ^ binary search stopped here | |
770 | * Another case we need to handle that only occurs in 256k | |
771 | * logs is | |
772 | * x + 1 ... | x ... | x+1 | x ... | |
773 | * ^ binary search stops here | |
774 | * In a 256k log, the scan at the end of the log will see the | |
775 | * x + 1 blocks. We need to skip past those since that is | |
776 | * certainly not the head of the log. By searching for | |
777 | * last_half_cycle-1 we accomplish that. | |
778 | */ | |
1da177e4 | 779 | ASSERT(head_blk <= INT_MAX && |
3f943d85 AE |
780 | (xfs_daddr_t) num_scan_bblks >= head_blk); |
781 | start_blk = log_bbnum - (num_scan_bblks - head_blk); | |
1da177e4 LT |
782 | if ((error = xlog_find_verify_cycle(log, start_blk, |
783 | num_scan_bblks - (int)head_blk, | |
784 | (stop_on_cycle - 1), &new_blk))) | |
785 | goto bp_err; | |
786 | if (new_blk != -1) { | |
787 | head_blk = new_blk; | |
9db127ed | 788 | goto validate_head; |
1da177e4 LT |
789 | } |
790 | ||
791 | /* | |
792 | * Scan beginning of log now. The last part of the physical | |
793 | * log is good. This scan needs to verify that it doesn't find | |
794 | * the last_half_cycle. | |
795 | */ | |
796 | start_blk = 0; | |
797 | ASSERT(head_blk <= INT_MAX); | |
798 | if ((error = xlog_find_verify_cycle(log, | |
799 | start_blk, (int)head_blk, | |
800 | stop_on_cycle, &new_blk))) | |
801 | goto bp_err; | |
802 | if (new_blk != -1) | |
803 | head_blk = new_blk; | |
804 | } | |
805 | ||
9db127ed | 806 | validate_head: |
1da177e4 LT |
807 | /* |
808 | * Now we need to make sure head_blk is not pointing to a block in | |
809 | * the middle of a log record. | |
810 | */ | |
811 | num_scan_bblks = XLOG_REC_SHIFT(log); | |
812 | if (head_blk >= num_scan_bblks) { | |
813 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ | |
814 | ||
815 | /* start ptr at last block ptr before head_blk */ | |
816 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
817 | &head_blk, 0)) == -1) { | |
818 | error = XFS_ERROR(EIO); | |
819 | goto bp_err; | |
820 | } else if (error) | |
821 | goto bp_err; | |
822 | } else { | |
823 | start_blk = 0; | |
824 | ASSERT(head_blk <= INT_MAX); | |
825 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
826 | &head_blk, 0)) == -1) { | |
827 | /* We hit the beginning of the log during our search */ | |
3f943d85 | 828 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
1da177e4 LT |
829 | new_blk = log_bbnum; |
830 | ASSERT(start_blk <= INT_MAX && | |
831 | (xfs_daddr_t) log_bbnum-start_blk >= 0); | |
832 | ASSERT(head_blk <= INT_MAX); | |
833 | if ((error = xlog_find_verify_log_record(log, | |
834 | start_blk, &new_blk, | |
835 | (int)head_blk)) == -1) { | |
836 | error = XFS_ERROR(EIO); | |
837 | goto bp_err; | |
838 | } else if (error) | |
839 | goto bp_err; | |
840 | if (new_blk != log_bbnum) | |
841 | head_blk = new_blk; | |
842 | } else if (error) | |
843 | goto bp_err; | |
844 | } | |
845 | ||
846 | xlog_put_bp(bp); | |
847 | if (head_blk == log_bbnum) | |
848 | *return_head_blk = 0; | |
849 | else | |
850 | *return_head_blk = head_blk; | |
851 | /* | |
852 | * When returning here, we have a good block number. Bad block | |
853 | * means that during a previous crash, we didn't have a clean break | |
854 | * from cycle number N to cycle number N-1. In this case, we need | |
855 | * to find the first block with cycle number N-1. | |
856 | */ | |
857 | return 0; | |
858 | ||
859 | bp_err: | |
860 | xlog_put_bp(bp); | |
861 | ||
862 | if (error) | |
a0fa2b67 | 863 | xfs_warn(log->l_mp, "failed to find log head"); |
1da177e4 LT |
864 | return error; |
865 | } | |
866 | ||
867 | /* | |
868 | * Find the sync block number or the tail of the log. | |
869 | * | |
870 | * This will be the block number of the last record to have its | |
871 | * associated buffers synced to disk. Every log record header has | |
872 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy | |
873 | * to get a sync block number. The only concern is to figure out which | |
874 | * log record header to believe. | |
875 | * | |
876 | * The following algorithm uses the log record header with the largest | |
877 | * lsn. The entire log record does not need to be valid. We only care | |
878 | * that the header is valid. | |
879 | * | |
880 | * We could speed up search by using current head_blk buffer, but it is not | |
881 | * available. | |
882 | */ | |
5d77c0dc | 883 | STATIC int |
1da177e4 | 884 | xlog_find_tail( |
9a8d2fdb | 885 | struct xlog *log, |
1da177e4 | 886 | xfs_daddr_t *head_blk, |
65be6054 | 887 | xfs_daddr_t *tail_blk) |
1da177e4 LT |
888 | { |
889 | xlog_rec_header_t *rhead; | |
890 | xlog_op_header_t *op_head; | |
891 | xfs_caddr_t offset = NULL; | |
892 | xfs_buf_t *bp; | |
893 | int error, i, found; | |
894 | xfs_daddr_t umount_data_blk; | |
895 | xfs_daddr_t after_umount_blk; | |
896 | xfs_lsn_t tail_lsn; | |
897 | int hblks; | |
898 | ||
899 | found = 0; | |
900 | ||
901 | /* | |
902 | * Find previous log record | |
903 | */ | |
904 | if ((error = xlog_find_head(log, head_blk))) | |
905 | return error; | |
906 | ||
907 | bp = xlog_get_bp(log, 1); | |
908 | if (!bp) | |
909 | return ENOMEM; | |
910 | if (*head_blk == 0) { /* special case */ | |
076e6acb CH |
911 | error = xlog_bread(log, 0, 1, bp, &offset); |
912 | if (error) | |
9db127ed | 913 | goto done; |
076e6acb | 914 | |
03bea6fe | 915 | if (xlog_get_cycle(offset) == 0) { |
1da177e4 LT |
916 | *tail_blk = 0; |
917 | /* leave all other log inited values alone */ | |
9db127ed | 918 | goto done; |
1da177e4 LT |
919 | } |
920 | } | |
921 | ||
922 | /* | |
923 | * Search backwards looking for log record header block | |
924 | */ | |
925 | ASSERT(*head_blk < INT_MAX); | |
926 | for (i = (int)(*head_blk) - 1; i >= 0; i--) { | |
076e6acb CH |
927 | error = xlog_bread(log, i, 1, bp, &offset); |
928 | if (error) | |
9db127ed | 929 | goto done; |
076e6acb | 930 | |
69ef921b | 931 | if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
1da177e4 LT |
932 | found = 1; |
933 | break; | |
934 | } | |
935 | } | |
936 | /* | |
937 | * If we haven't found the log record header block, start looking | |
938 | * again from the end of the physical log. XXXmiken: There should be | |
939 | * a check here to make sure we didn't search more than N blocks in | |
940 | * the previous code. | |
941 | */ | |
942 | if (!found) { | |
943 | for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) { | |
076e6acb CH |
944 | error = xlog_bread(log, i, 1, bp, &offset); |
945 | if (error) | |
9db127ed | 946 | goto done; |
076e6acb | 947 | |
69ef921b CH |
948 | if (*(__be32 *)offset == |
949 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
1da177e4 LT |
950 | found = 2; |
951 | break; | |
952 | } | |
953 | } | |
954 | } | |
955 | if (!found) { | |
a0fa2b67 | 956 | xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__); |
1da177e4 LT |
957 | ASSERT(0); |
958 | return XFS_ERROR(EIO); | |
959 | } | |
960 | ||
961 | /* find blk_no of tail of log */ | |
962 | rhead = (xlog_rec_header_t *)offset; | |
b53e675d | 963 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); |
1da177e4 LT |
964 | |
965 | /* | |
966 | * Reset log values according to the state of the log when we | |
967 | * crashed. In the case where head_blk == 0, we bump curr_cycle | |
968 | * one because the next write starts a new cycle rather than | |
969 | * continuing the cycle of the last good log record. At this | |
970 | * point we have guaranteed that all partial log records have been | |
971 | * accounted for. Therefore, we know that the last good log record | |
972 | * written was complete and ended exactly on the end boundary | |
973 | * of the physical log. | |
974 | */ | |
975 | log->l_prev_block = i; | |
976 | log->l_curr_block = (int)*head_blk; | |
b53e675d | 977 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); |
1da177e4 LT |
978 | if (found == 2) |
979 | log->l_curr_cycle++; | |
1c3cb9ec | 980 | atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn)); |
84f3c683 | 981 | atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn)); |
28496968 | 982 | xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle, |
a69ed03c | 983 | BBTOB(log->l_curr_block)); |
28496968 | 984 | xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle, |
a69ed03c | 985 | BBTOB(log->l_curr_block)); |
1da177e4 LT |
986 | |
987 | /* | |
988 | * Look for unmount record. If we find it, then we know there | |
989 | * was a clean unmount. Since 'i' could be the last block in | |
990 | * the physical log, we convert to a log block before comparing | |
991 | * to the head_blk. | |
992 | * | |
993 | * Save the current tail lsn to use to pass to | |
994 | * xlog_clear_stale_blocks() below. We won't want to clear the | |
995 | * unmount record if there is one, so we pass the lsn of the | |
996 | * unmount record rather than the block after it. | |
997 | */ | |
62118709 | 998 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d CH |
999 | int h_size = be32_to_cpu(rhead->h_size); |
1000 | int h_version = be32_to_cpu(rhead->h_version); | |
1da177e4 LT |
1001 | |
1002 | if ((h_version & XLOG_VERSION_2) && | |
1003 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
1004 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
1005 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
1006 | hblks++; | |
1007 | } else { | |
1008 | hblks = 1; | |
1009 | } | |
1010 | } else { | |
1011 | hblks = 1; | |
1012 | } | |
1013 | after_umount_blk = (i + hblks + (int) | |
b53e675d | 1014 | BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize; |
1c3cb9ec | 1015 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
1da177e4 | 1016 | if (*head_blk == after_umount_blk && |
b53e675d | 1017 | be32_to_cpu(rhead->h_num_logops) == 1) { |
1da177e4 | 1018 | umount_data_blk = (i + hblks) % log->l_logBBsize; |
076e6acb CH |
1019 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); |
1020 | if (error) | |
9db127ed | 1021 | goto done; |
076e6acb | 1022 | |
1da177e4 LT |
1023 | op_head = (xlog_op_header_t *)offset; |
1024 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { | |
1025 | /* | |
1026 | * Set tail and last sync so that newly written | |
1027 | * log records will point recovery to after the | |
1028 | * current unmount record. | |
1029 | */ | |
1c3cb9ec DC |
1030 | xlog_assign_atomic_lsn(&log->l_tail_lsn, |
1031 | log->l_curr_cycle, after_umount_blk); | |
1032 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, | |
1033 | log->l_curr_cycle, after_umount_blk); | |
1da177e4 | 1034 | *tail_blk = after_umount_blk; |
92821e2b DC |
1035 | |
1036 | /* | |
1037 | * Note that the unmount was clean. If the unmount | |
1038 | * was not clean, we need to know this to rebuild the | |
1039 | * superblock counters from the perag headers if we | |
1040 | * have a filesystem using non-persistent counters. | |
1041 | */ | |
1042 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; | |
1da177e4 LT |
1043 | } |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * Make sure that there are no blocks in front of the head | |
1048 | * with the same cycle number as the head. This can happen | |
1049 | * because we allow multiple outstanding log writes concurrently, | |
1050 | * and the later writes might make it out before earlier ones. | |
1051 | * | |
1052 | * We use the lsn from before modifying it so that we'll never | |
1053 | * overwrite the unmount record after a clean unmount. | |
1054 | * | |
1055 | * Do this only if we are going to recover the filesystem | |
1056 | * | |
1057 | * NOTE: This used to say "if (!readonly)" | |
1058 | * However on Linux, we can & do recover a read-only filesystem. | |
1059 | * We only skip recovery if NORECOVERY is specified on mount, | |
1060 | * in which case we would not be here. | |
1061 | * | |
1062 | * But... if the -device- itself is readonly, just skip this. | |
1063 | * We can't recover this device anyway, so it won't matter. | |
1064 | */ | |
9db127ed | 1065 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) |
1da177e4 | 1066 | error = xlog_clear_stale_blocks(log, tail_lsn); |
1da177e4 | 1067 | |
9db127ed | 1068 | done: |
1da177e4 LT |
1069 | xlog_put_bp(bp); |
1070 | ||
1071 | if (error) | |
a0fa2b67 | 1072 | xfs_warn(log->l_mp, "failed to locate log tail"); |
1da177e4 LT |
1073 | return error; |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * Is the log zeroed at all? | |
1078 | * | |
1079 | * The last binary search should be changed to perform an X block read | |
1080 | * once X becomes small enough. You can then search linearly through | |
1081 | * the X blocks. This will cut down on the number of reads we need to do. | |
1082 | * | |
1083 | * If the log is partially zeroed, this routine will pass back the blkno | |
1084 | * of the first block with cycle number 0. It won't have a complete LR | |
1085 | * preceding it. | |
1086 | * | |
1087 | * Return: | |
1088 | * 0 => the log is completely written to | |
1089 | * -1 => use *blk_no as the first block of the log | |
1090 | * >0 => error has occurred | |
1091 | */ | |
a8272ce0 | 1092 | STATIC int |
1da177e4 | 1093 | xlog_find_zeroed( |
9a8d2fdb | 1094 | struct xlog *log, |
1da177e4 LT |
1095 | xfs_daddr_t *blk_no) |
1096 | { | |
1097 | xfs_buf_t *bp; | |
1098 | xfs_caddr_t offset; | |
1099 | uint first_cycle, last_cycle; | |
1100 | xfs_daddr_t new_blk, last_blk, start_blk; | |
1101 | xfs_daddr_t num_scan_bblks; | |
1102 | int error, log_bbnum = log->l_logBBsize; | |
1103 | ||
6fdf8ccc NS |
1104 | *blk_no = 0; |
1105 | ||
1da177e4 LT |
1106 | /* check totally zeroed log */ |
1107 | bp = xlog_get_bp(log, 1); | |
1108 | if (!bp) | |
1109 | return ENOMEM; | |
076e6acb CH |
1110 | error = xlog_bread(log, 0, 1, bp, &offset); |
1111 | if (error) | |
1da177e4 | 1112 | goto bp_err; |
076e6acb | 1113 | |
03bea6fe | 1114 | first_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1115 | if (first_cycle == 0) { /* completely zeroed log */ |
1116 | *blk_no = 0; | |
1117 | xlog_put_bp(bp); | |
1118 | return -1; | |
1119 | } | |
1120 | ||
1121 | /* check partially zeroed log */ | |
076e6acb CH |
1122 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
1123 | if (error) | |
1da177e4 | 1124 | goto bp_err; |
076e6acb | 1125 | |
03bea6fe | 1126 | last_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1127 | if (last_cycle != 0) { /* log completely written to */ |
1128 | xlog_put_bp(bp); | |
1129 | return 0; | |
1130 | } else if (first_cycle != 1) { | |
1131 | /* | |
1132 | * If the cycle of the last block is zero, the cycle of | |
1133 | * the first block must be 1. If it's not, maybe we're | |
1134 | * not looking at a log... Bail out. | |
1135 | */ | |
a0fa2b67 DC |
1136 | xfs_warn(log->l_mp, |
1137 | "Log inconsistent or not a log (last==0, first!=1)"); | |
1da177e4 LT |
1138 | return XFS_ERROR(EINVAL); |
1139 | } | |
1140 | ||
1141 | /* we have a partially zeroed log */ | |
1142 | last_blk = log_bbnum-1; | |
1143 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) | |
1144 | goto bp_err; | |
1145 | ||
1146 | /* | |
1147 | * Validate the answer. Because there is no way to guarantee that | |
1148 | * the entire log is made up of log records which are the same size, | |
1149 | * we scan over the defined maximum blocks. At this point, the maximum | |
1150 | * is not chosen to mean anything special. XXXmiken | |
1151 | */ | |
1152 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
1153 | ASSERT(num_scan_bblks <= INT_MAX); | |
1154 | ||
1155 | if (last_blk < num_scan_bblks) | |
1156 | num_scan_bblks = last_blk; | |
1157 | start_blk = last_blk - num_scan_bblks; | |
1158 | ||
1159 | /* | |
1160 | * We search for any instances of cycle number 0 that occur before | |
1161 | * our current estimate of the head. What we're trying to detect is | |
1162 | * 1 ... | 0 | 1 | 0... | |
1163 | * ^ binary search ends here | |
1164 | */ | |
1165 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
1166 | (int)num_scan_bblks, 0, &new_blk))) | |
1167 | goto bp_err; | |
1168 | if (new_blk != -1) | |
1169 | last_blk = new_blk; | |
1170 | ||
1171 | /* | |
1172 | * Potentially backup over partial log record write. We don't need | |
1173 | * to search the end of the log because we know it is zero. | |
1174 | */ | |
1175 | if ((error = xlog_find_verify_log_record(log, start_blk, | |
1176 | &last_blk, 0)) == -1) { | |
1177 | error = XFS_ERROR(EIO); | |
1178 | goto bp_err; | |
1179 | } else if (error) | |
1180 | goto bp_err; | |
1181 | ||
1182 | *blk_no = last_blk; | |
1183 | bp_err: | |
1184 | xlog_put_bp(bp); | |
1185 | if (error) | |
1186 | return error; | |
1187 | return -1; | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * These are simple subroutines used by xlog_clear_stale_blocks() below | |
1192 | * to initialize a buffer full of empty log record headers and write | |
1193 | * them into the log. | |
1194 | */ | |
1195 | STATIC void | |
1196 | xlog_add_record( | |
9a8d2fdb | 1197 | struct xlog *log, |
1da177e4 LT |
1198 | xfs_caddr_t buf, |
1199 | int cycle, | |
1200 | int block, | |
1201 | int tail_cycle, | |
1202 | int tail_block) | |
1203 | { | |
1204 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; | |
1205 | ||
1206 | memset(buf, 0, BBSIZE); | |
b53e675d CH |
1207 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1208 | recp->h_cycle = cpu_to_be32(cycle); | |
1209 | recp->h_version = cpu_to_be32( | |
62118709 | 1210 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d CH |
1211 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
1212 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); | |
1213 | recp->h_fmt = cpu_to_be32(XLOG_FMT); | |
1da177e4 LT |
1214 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1215 | } | |
1216 | ||
1217 | STATIC int | |
1218 | xlog_write_log_records( | |
9a8d2fdb | 1219 | struct xlog *log, |
1da177e4 LT |
1220 | int cycle, |
1221 | int start_block, | |
1222 | int blocks, | |
1223 | int tail_cycle, | |
1224 | int tail_block) | |
1225 | { | |
1226 | xfs_caddr_t offset; | |
1227 | xfs_buf_t *bp; | |
1228 | int balign, ealign; | |
69ce58f0 | 1229 | int sectbb = log->l_sectBBsize; |
1da177e4 LT |
1230 | int end_block = start_block + blocks; |
1231 | int bufblks; | |
1232 | int error = 0; | |
1233 | int i, j = 0; | |
1234 | ||
6881a229 AE |
1235 | /* |
1236 | * Greedily allocate a buffer big enough to handle the full | |
1237 | * range of basic blocks to be written. If that fails, try | |
1238 | * a smaller size. We need to be able to write at least a | |
1239 | * log sector, or we're out of luck. | |
1240 | */ | |
1da177e4 | 1241 | bufblks = 1 << ffs(blocks); |
81158e0c DC |
1242 | while (bufblks > log->l_logBBsize) |
1243 | bufblks >>= 1; | |
1da177e4 LT |
1244 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1245 | bufblks >>= 1; | |
69ce58f0 | 1246 | if (bufblks < sectbb) |
1da177e4 LT |
1247 | return ENOMEM; |
1248 | } | |
1249 | ||
1250 | /* We may need to do a read at the start to fill in part of | |
1251 | * the buffer in the starting sector not covered by the first | |
1252 | * write below. | |
1253 | */ | |
5c17f533 | 1254 | balign = round_down(start_block, sectbb); |
1da177e4 | 1255 | if (balign != start_block) { |
076e6acb CH |
1256 | error = xlog_bread_noalign(log, start_block, 1, bp); |
1257 | if (error) | |
1258 | goto out_put_bp; | |
1259 | ||
1da177e4 LT |
1260 | j = start_block - balign; |
1261 | } | |
1262 | ||
1263 | for (i = start_block; i < end_block; i += bufblks) { | |
1264 | int bcount, endcount; | |
1265 | ||
1266 | bcount = min(bufblks, end_block - start_block); | |
1267 | endcount = bcount - j; | |
1268 | ||
1269 | /* We may need to do a read at the end to fill in part of | |
1270 | * the buffer in the final sector not covered by the write. | |
1271 | * If this is the same sector as the above read, skip it. | |
1272 | */ | |
5c17f533 | 1273 | ealign = round_down(end_block, sectbb); |
1da177e4 | 1274 | if (j == 0 && (start_block + endcount > ealign)) { |
62926044 | 1275 | offset = bp->b_addr + BBTOB(ealign - start_block); |
44396476 DC |
1276 | error = xlog_bread_offset(log, ealign, sectbb, |
1277 | bp, offset); | |
076e6acb CH |
1278 | if (error) |
1279 | break; | |
1280 | ||
1da177e4 LT |
1281 | } |
1282 | ||
1283 | offset = xlog_align(log, start_block, endcount, bp); | |
1284 | for (; j < endcount; j++) { | |
1285 | xlog_add_record(log, offset, cycle, i+j, | |
1286 | tail_cycle, tail_block); | |
1287 | offset += BBSIZE; | |
1288 | } | |
1289 | error = xlog_bwrite(log, start_block, endcount, bp); | |
1290 | if (error) | |
1291 | break; | |
1292 | start_block += endcount; | |
1293 | j = 0; | |
1294 | } | |
076e6acb CH |
1295 | |
1296 | out_put_bp: | |
1da177e4 LT |
1297 | xlog_put_bp(bp); |
1298 | return error; | |
1299 | } | |
1300 | ||
1301 | /* | |
1302 | * This routine is called to blow away any incomplete log writes out | |
1303 | * in front of the log head. We do this so that we won't become confused | |
1304 | * if we come up, write only a little bit more, and then crash again. | |
1305 | * If we leave the partial log records out there, this situation could | |
1306 | * cause us to think those partial writes are valid blocks since they | |
1307 | * have the current cycle number. We get rid of them by overwriting them | |
1308 | * with empty log records with the old cycle number rather than the | |
1309 | * current one. | |
1310 | * | |
1311 | * The tail lsn is passed in rather than taken from | |
1312 | * the log so that we will not write over the unmount record after a | |
1313 | * clean unmount in a 512 block log. Doing so would leave the log without | |
1314 | * any valid log records in it until a new one was written. If we crashed | |
1315 | * during that time we would not be able to recover. | |
1316 | */ | |
1317 | STATIC int | |
1318 | xlog_clear_stale_blocks( | |
9a8d2fdb | 1319 | struct xlog *log, |
1da177e4 LT |
1320 | xfs_lsn_t tail_lsn) |
1321 | { | |
1322 | int tail_cycle, head_cycle; | |
1323 | int tail_block, head_block; | |
1324 | int tail_distance, max_distance; | |
1325 | int distance; | |
1326 | int error; | |
1327 | ||
1328 | tail_cycle = CYCLE_LSN(tail_lsn); | |
1329 | tail_block = BLOCK_LSN(tail_lsn); | |
1330 | head_cycle = log->l_curr_cycle; | |
1331 | head_block = log->l_curr_block; | |
1332 | ||
1333 | /* | |
1334 | * Figure out the distance between the new head of the log | |
1335 | * and the tail. We want to write over any blocks beyond the | |
1336 | * head that we may have written just before the crash, but | |
1337 | * we don't want to overwrite the tail of the log. | |
1338 | */ | |
1339 | if (head_cycle == tail_cycle) { | |
1340 | /* | |
1341 | * The tail is behind the head in the physical log, | |
1342 | * so the distance from the head to the tail is the | |
1343 | * distance from the head to the end of the log plus | |
1344 | * the distance from the beginning of the log to the | |
1345 | * tail. | |
1346 | */ | |
1347 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { | |
1348 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", | |
1349 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1350 | return XFS_ERROR(EFSCORRUPTED); | |
1351 | } | |
1352 | tail_distance = tail_block + (log->l_logBBsize - head_block); | |
1353 | } else { | |
1354 | /* | |
1355 | * The head is behind the tail in the physical log, | |
1356 | * so the distance from the head to the tail is just | |
1357 | * the tail block minus the head block. | |
1358 | */ | |
1359 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ | |
1360 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", | |
1361 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1362 | return XFS_ERROR(EFSCORRUPTED); | |
1363 | } | |
1364 | tail_distance = tail_block - head_block; | |
1365 | } | |
1366 | ||
1367 | /* | |
1368 | * If the head is right up against the tail, we can't clear | |
1369 | * anything. | |
1370 | */ | |
1371 | if (tail_distance <= 0) { | |
1372 | ASSERT(tail_distance == 0); | |
1373 | return 0; | |
1374 | } | |
1375 | ||
1376 | max_distance = XLOG_TOTAL_REC_SHIFT(log); | |
1377 | /* | |
1378 | * Take the smaller of the maximum amount of outstanding I/O | |
1379 | * we could have and the distance to the tail to clear out. | |
1380 | * We take the smaller so that we don't overwrite the tail and | |
1381 | * we don't waste all day writing from the head to the tail | |
1382 | * for no reason. | |
1383 | */ | |
1384 | max_distance = MIN(max_distance, tail_distance); | |
1385 | ||
1386 | if ((head_block + max_distance) <= log->l_logBBsize) { | |
1387 | /* | |
1388 | * We can stomp all the blocks we need to without | |
1389 | * wrapping around the end of the log. Just do it | |
1390 | * in a single write. Use the cycle number of the | |
1391 | * current cycle minus one so that the log will look like: | |
1392 | * n ... | n - 1 ... | |
1393 | */ | |
1394 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1395 | head_block, max_distance, tail_cycle, | |
1396 | tail_block); | |
1397 | if (error) | |
1398 | return error; | |
1399 | } else { | |
1400 | /* | |
1401 | * We need to wrap around the end of the physical log in | |
1402 | * order to clear all the blocks. Do it in two separate | |
1403 | * I/Os. The first write should be from the head to the | |
1404 | * end of the physical log, and it should use the current | |
1405 | * cycle number minus one just like above. | |
1406 | */ | |
1407 | distance = log->l_logBBsize - head_block; | |
1408 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1409 | head_block, distance, tail_cycle, | |
1410 | tail_block); | |
1411 | ||
1412 | if (error) | |
1413 | return error; | |
1414 | ||
1415 | /* | |
1416 | * Now write the blocks at the start of the physical log. | |
1417 | * This writes the remainder of the blocks we want to clear. | |
1418 | * It uses the current cycle number since we're now on the | |
1419 | * same cycle as the head so that we get: | |
1420 | * n ... n ... | n - 1 ... | |
1421 | * ^^^^^ blocks we're writing | |
1422 | */ | |
1423 | distance = max_distance - (log->l_logBBsize - head_block); | |
1424 | error = xlog_write_log_records(log, head_cycle, 0, distance, | |
1425 | tail_cycle, tail_block); | |
1426 | if (error) | |
1427 | return error; | |
1428 | } | |
1429 | ||
1430 | return 0; | |
1431 | } | |
1432 | ||
1433 | /****************************************************************************** | |
1434 | * | |
1435 | * Log recover routines | |
1436 | * | |
1437 | ****************************************************************************** | |
1438 | */ | |
1439 | ||
1440 | STATIC xlog_recover_t * | |
1441 | xlog_recover_find_tid( | |
f0a76953 | 1442 | struct hlist_head *head, |
1da177e4 LT |
1443 | xlog_tid_t tid) |
1444 | { | |
f0a76953 | 1445 | xlog_recover_t *trans; |
1da177e4 | 1446 | |
b67bfe0d | 1447 | hlist_for_each_entry(trans, head, r_list) { |
f0a76953 DC |
1448 | if (trans->r_log_tid == tid) |
1449 | return trans; | |
1da177e4 | 1450 | } |
f0a76953 | 1451 | return NULL; |
1da177e4 LT |
1452 | } |
1453 | ||
1454 | STATIC void | |
f0a76953 DC |
1455 | xlog_recover_new_tid( |
1456 | struct hlist_head *head, | |
1457 | xlog_tid_t tid, | |
1458 | xfs_lsn_t lsn) | |
1da177e4 | 1459 | { |
f0a76953 DC |
1460 | xlog_recover_t *trans; |
1461 | ||
1462 | trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP); | |
1463 | trans->r_log_tid = tid; | |
1464 | trans->r_lsn = lsn; | |
1465 | INIT_LIST_HEAD(&trans->r_itemq); | |
1466 | ||
1467 | INIT_HLIST_NODE(&trans->r_list); | |
1468 | hlist_add_head(&trans->r_list, head); | |
1da177e4 LT |
1469 | } |
1470 | ||
1471 | STATIC void | |
1472 | xlog_recover_add_item( | |
f0a76953 | 1473 | struct list_head *head) |
1da177e4 LT |
1474 | { |
1475 | xlog_recover_item_t *item; | |
1476 | ||
1477 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); | |
f0a76953 DC |
1478 | INIT_LIST_HEAD(&item->ri_list); |
1479 | list_add_tail(&item->ri_list, head); | |
1da177e4 LT |
1480 | } |
1481 | ||
1482 | STATIC int | |
1483 | xlog_recover_add_to_cont_trans( | |
ad223e60 MT |
1484 | struct xlog *log, |
1485 | struct xlog_recover *trans, | |
1da177e4 LT |
1486 | xfs_caddr_t dp, |
1487 | int len) | |
1488 | { | |
1489 | xlog_recover_item_t *item; | |
1490 | xfs_caddr_t ptr, old_ptr; | |
1491 | int old_len; | |
1492 | ||
f0a76953 | 1493 | if (list_empty(&trans->r_itemq)) { |
1da177e4 LT |
1494 | /* finish copying rest of trans header */ |
1495 | xlog_recover_add_item(&trans->r_itemq); | |
1496 | ptr = (xfs_caddr_t) &trans->r_theader + | |
1497 | sizeof(xfs_trans_header_t) - len; | |
1498 | memcpy(ptr, dp, len); /* d, s, l */ | |
1499 | return 0; | |
1500 | } | |
f0a76953 DC |
1501 | /* take the tail entry */ |
1502 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
1da177e4 LT |
1503 | |
1504 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; | |
1505 | old_len = item->ri_buf[item->ri_cnt-1].i_len; | |
1506 | ||
45053603 | 1507 | ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP); |
1da177e4 LT |
1508 | memcpy(&ptr[old_len], dp, len); /* d, s, l */ |
1509 | item->ri_buf[item->ri_cnt-1].i_len += len; | |
1510 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; | |
9abbc539 | 1511 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); |
1da177e4 LT |
1512 | return 0; |
1513 | } | |
1514 | ||
1515 | /* | |
1516 | * The next region to add is the start of a new region. It could be | |
1517 | * a whole region or it could be the first part of a new region. Because | |
1518 | * of this, the assumption here is that the type and size fields of all | |
1519 | * format structures fit into the first 32 bits of the structure. | |
1520 | * | |
1521 | * This works because all regions must be 32 bit aligned. Therefore, we | |
1522 | * either have both fields or we have neither field. In the case we have | |
1523 | * neither field, the data part of the region is zero length. We only have | |
1524 | * a log_op_header and can throw away the header since a new one will appear | |
1525 | * later. If we have at least 4 bytes, then we can determine how many regions | |
1526 | * will appear in the current log item. | |
1527 | */ | |
1528 | STATIC int | |
1529 | xlog_recover_add_to_trans( | |
ad223e60 MT |
1530 | struct xlog *log, |
1531 | struct xlog_recover *trans, | |
1da177e4 LT |
1532 | xfs_caddr_t dp, |
1533 | int len) | |
1534 | { | |
1535 | xfs_inode_log_format_t *in_f; /* any will do */ | |
1536 | xlog_recover_item_t *item; | |
1537 | xfs_caddr_t ptr; | |
1538 | ||
1539 | if (!len) | |
1540 | return 0; | |
f0a76953 | 1541 | if (list_empty(&trans->r_itemq)) { |
5a792c45 DC |
1542 | /* we need to catch log corruptions here */ |
1543 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { | |
a0fa2b67 DC |
1544 | xfs_warn(log->l_mp, "%s: bad header magic number", |
1545 | __func__); | |
5a792c45 DC |
1546 | ASSERT(0); |
1547 | return XFS_ERROR(EIO); | |
1548 | } | |
1da177e4 LT |
1549 | if (len == sizeof(xfs_trans_header_t)) |
1550 | xlog_recover_add_item(&trans->r_itemq); | |
1551 | memcpy(&trans->r_theader, dp, len); /* d, s, l */ | |
1552 | return 0; | |
1553 | } | |
1554 | ||
1555 | ptr = kmem_alloc(len, KM_SLEEP); | |
1556 | memcpy(ptr, dp, len); | |
1557 | in_f = (xfs_inode_log_format_t *)ptr; | |
1558 | ||
f0a76953 DC |
1559 | /* take the tail entry */ |
1560 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
1561 | if (item->ri_total != 0 && | |
1562 | item->ri_total == item->ri_cnt) { | |
1563 | /* tail item is in use, get a new one */ | |
1da177e4 | 1564 | xlog_recover_add_item(&trans->r_itemq); |
f0a76953 DC |
1565 | item = list_entry(trans->r_itemq.prev, |
1566 | xlog_recover_item_t, ri_list); | |
1da177e4 | 1567 | } |
1da177e4 LT |
1568 | |
1569 | if (item->ri_total == 0) { /* first region to be added */ | |
e8fa6b48 CH |
1570 | if (in_f->ilf_size == 0 || |
1571 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { | |
a0fa2b67 DC |
1572 | xfs_warn(log->l_mp, |
1573 | "bad number of regions (%d) in inode log format", | |
e8fa6b48 CH |
1574 | in_f->ilf_size); |
1575 | ASSERT(0); | |
1576 | return XFS_ERROR(EIO); | |
1577 | } | |
1578 | ||
1579 | item->ri_total = in_f->ilf_size; | |
1580 | item->ri_buf = | |
1581 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), | |
1582 | KM_SLEEP); | |
1da177e4 LT |
1583 | } |
1584 | ASSERT(item->ri_total > item->ri_cnt); | |
1585 | /* Description region is ri_buf[0] */ | |
1586 | item->ri_buf[item->ri_cnt].i_addr = ptr; | |
1587 | item->ri_buf[item->ri_cnt].i_len = len; | |
1588 | item->ri_cnt++; | |
9abbc539 | 1589 | trace_xfs_log_recover_item_add(log, trans, item, 0); |
1da177e4 LT |
1590 | return 0; |
1591 | } | |
1592 | ||
f0a76953 DC |
1593 | /* |
1594 | * Sort the log items in the transaction. Cancelled buffers need | |
1595 | * to be put first so they are processed before any items that might | |
1596 | * modify the buffers. If they are cancelled, then the modifications | |
1597 | * don't need to be replayed. | |
1598 | */ | |
1da177e4 LT |
1599 | STATIC int |
1600 | xlog_recover_reorder_trans( | |
ad223e60 MT |
1601 | struct xlog *log, |
1602 | struct xlog_recover *trans, | |
9abbc539 | 1603 | int pass) |
1da177e4 | 1604 | { |
f0a76953 DC |
1605 | xlog_recover_item_t *item, *n; |
1606 | LIST_HEAD(sort_list); | |
1607 | ||
1608 | list_splice_init(&trans->r_itemq, &sort_list); | |
1609 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { | |
4e0d5f92 | 1610 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
1da177e4 | 1611 | |
f0a76953 | 1612 | switch (ITEM_TYPE(item)) { |
1da177e4 | 1613 | case XFS_LI_BUF: |
c1155410 | 1614 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
9abbc539 DC |
1615 | trace_xfs_log_recover_item_reorder_head(log, |
1616 | trans, item, pass); | |
f0a76953 | 1617 | list_move(&item->ri_list, &trans->r_itemq); |
1da177e4 LT |
1618 | break; |
1619 | } | |
1620 | case XFS_LI_INODE: | |
1da177e4 LT |
1621 | case XFS_LI_DQUOT: |
1622 | case XFS_LI_QUOTAOFF: | |
1623 | case XFS_LI_EFD: | |
1624 | case XFS_LI_EFI: | |
9abbc539 DC |
1625 | trace_xfs_log_recover_item_reorder_tail(log, |
1626 | trans, item, pass); | |
f0a76953 | 1627 | list_move_tail(&item->ri_list, &trans->r_itemq); |
1da177e4 LT |
1628 | break; |
1629 | default: | |
a0fa2b67 DC |
1630 | xfs_warn(log->l_mp, |
1631 | "%s: unrecognized type of log operation", | |
1632 | __func__); | |
1da177e4 LT |
1633 | ASSERT(0); |
1634 | return XFS_ERROR(EIO); | |
1635 | } | |
f0a76953 DC |
1636 | } |
1637 | ASSERT(list_empty(&sort_list)); | |
1da177e4 LT |
1638 | return 0; |
1639 | } | |
1640 | ||
1641 | /* | |
1642 | * Build up the table of buf cancel records so that we don't replay | |
1643 | * cancelled data in the second pass. For buffer records that are | |
1644 | * not cancel records, there is nothing to do here so we just return. | |
1645 | * | |
1646 | * If we get a cancel record which is already in the table, this indicates | |
1647 | * that the buffer was cancelled multiple times. In order to ensure | |
1648 | * that during pass 2 we keep the record in the table until we reach its | |
1649 | * last occurrence in the log, we keep a reference count in the cancel | |
1650 | * record in the table to tell us how many times we expect to see this | |
1651 | * record during the second pass. | |
1652 | */ | |
c9f71f5f CH |
1653 | STATIC int |
1654 | xlog_recover_buffer_pass1( | |
ad223e60 MT |
1655 | struct xlog *log, |
1656 | struct xlog_recover_item *item) | |
1da177e4 | 1657 | { |
c9f71f5f | 1658 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
d5689eaa CH |
1659 | struct list_head *bucket; |
1660 | struct xfs_buf_cancel *bcp; | |
1da177e4 LT |
1661 | |
1662 | /* | |
1663 | * If this isn't a cancel buffer item, then just return. | |
1664 | */ | |
e2714bf8 | 1665 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
9abbc539 | 1666 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); |
c9f71f5f | 1667 | return 0; |
9abbc539 | 1668 | } |
1da177e4 LT |
1669 | |
1670 | /* | |
d5689eaa CH |
1671 | * Insert an xfs_buf_cancel record into the hash table of them. |
1672 | * If there is already an identical record, bump its reference count. | |
1da177e4 | 1673 | */ |
d5689eaa CH |
1674 | bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno); |
1675 | list_for_each_entry(bcp, bucket, bc_list) { | |
1676 | if (bcp->bc_blkno == buf_f->blf_blkno && | |
1677 | bcp->bc_len == buf_f->blf_len) { | |
1678 | bcp->bc_refcount++; | |
9abbc539 | 1679 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
c9f71f5f | 1680 | return 0; |
1da177e4 | 1681 | } |
d5689eaa CH |
1682 | } |
1683 | ||
1684 | bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP); | |
1685 | bcp->bc_blkno = buf_f->blf_blkno; | |
1686 | bcp->bc_len = buf_f->blf_len; | |
1da177e4 | 1687 | bcp->bc_refcount = 1; |
d5689eaa CH |
1688 | list_add_tail(&bcp->bc_list, bucket); |
1689 | ||
9abbc539 | 1690 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
c9f71f5f | 1691 | return 0; |
1da177e4 LT |
1692 | } |
1693 | ||
1694 | /* | |
1695 | * Check to see whether the buffer being recovered has a corresponding | |
1696 | * entry in the buffer cancel record table. If it does then return 1 | |
1697 | * so that it will be cancelled, otherwise return 0. If the buffer is | |
c1155410 | 1698 | * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement |
1da177e4 LT |
1699 | * the refcount on the entry in the table and remove it from the table |
1700 | * if this is the last reference. | |
1701 | * | |
1702 | * We remove the cancel record from the table when we encounter its | |
1703 | * last occurrence in the log so that if the same buffer is re-used | |
1704 | * again after its last cancellation we actually replay the changes | |
1705 | * made at that point. | |
1706 | */ | |
1707 | STATIC int | |
1708 | xlog_check_buffer_cancelled( | |
ad223e60 | 1709 | struct xlog *log, |
1da177e4 LT |
1710 | xfs_daddr_t blkno, |
1711 | uint len, | |
1712 | ushort flags) | |
1713 | { | |
d5689eaa CH |
1714 | struct list_head *bucket; |
1715 | struct xfs_buf_cancel *bcp; | |
1da177e4 LT |
1716 | |
1717 | if (log->l_buf_cancel_table == NULL) { | |
1718 | /* | |
1719 | * There is nothing in the table built in pass one, | |
1720 | * so this buffer must not be cancelled. | |
1721 | */ | |
c1155410 | 1722 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
1da177e4 LT |
1723 | return 0; |
1724 | } | |
1725 | ||
1da177e4 | 1726 | /* |
d5689eaa | 1727 | * Search for an entry in the cancel table that matches our buffer. |
1da177e4 | 1728 | */ |
d5689eaa CH |
1729 | bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno); |
1730 | list_for_each_entry(bcp, bucket, bc_list) { | |
1731 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) | |
1732 | goto found; | |
1da177e4 | 1733 | } |
d5689eaa | 1734 | |
1da177e4 | 1735 | /* |
d5689eaa CH |
1736 | * We didn't find a corresponding entry in the table, so return 0 so |
1737 | * that the buffer is NOT cancelled. | |
1da177e4 | 1738 | */ |
c1155410 | 1739 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
1da177e4 | 1740 | return 0; |
d5689eaa CH |
1741 | |
1742 | found: | |
1743 | /* | |
1744 | * We've go a match, so return 1 so that the recovery of this buffer | |
1745 | * is cancelled. If this buffer is actually a buffer cancel log | |
1746 | * item, then decrement the refcount on the one in the table and | |
1747 | * remove it if this is the last reference. | |
1748 | */ | |
1749 | if (flags & XFS_BLF_CANCEL) { | |
1750 | if (--bcp->bc_refcount == 0) { | |
1751 | list_del(&bcp->bc_list); | |
1752 | kmem_free(bcp); | |
1753 | } | |
1754 | } | |
1755 | return 1; | |
1da177e4 LT |
1756 | } |
1757 | ||
1da177e4 | 1758 | /* |
e2714bf8 CH |
1759 | * Perform recovery for a buffer full of inodes. In these buffers, the only |
1760 | * data which should be recovered is that which corresponds to the | |
1761 | * di_next_unlinked pointers in the on disk inode structures. The rest of the | |
1762 | * data for the inodes is always logged through the inodes themselves rather | |
1763 | * than the inode buffer and is recovered in xlog_recover_inode_pass2(). | |
1da177e4 | 1764 | * |
e2714bf8 CH |
1765 | * The only time when buffers full of inodes are fully recovered is when the |
1766 | * buffer is full of newly allocated inodes. In this case the buffer will | |
1767 | * not be marked as an inode buffer and so will be sent to | |
1768 | * xlog_recover_do_reg_buffer() below during recovery. | |
1da177e4 LT |
1769 | */ |
1770 | STATIC int | |
1771 | xlog_recover_do_inode_buffer( | |
e2714bf8 | 1772 | struct xfs_mount *mp, |
1da177e4 | 1773 | xlog_recover_item_t *item, |
e2714bf8 | 1774 | struct xfs_buf *bp, |
1da177e4 LT |
1775 | xfs_buf_log_format_t *buf_f) |
1776 | { | |
1777 | int i; | |
e2714bf8 CH |
1778 | int item_index = 0; |
1779 | int bit = 0; | |
1780 | int nbits = 0; | |
1781 | int reg_buf_offset = 0; | |
1782 | int reg_buf_bytes = 0; | |
1da177e4 LT |
1783 | int next_unlinked_offset; |
1784 | int inodes_per_buf; | |
1785 | xfs_agino_t *logged_nextp; | |
1786 | xfs_agino_t *buffer_nextp; | |
1da177e4 | 1787 | |
9abbc539 DC |
1788 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
1789 | ||
aa0e8833 | 1790 | inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog; |
1da177e4 LT |
1791 | for (i = 0; i < inodes_per_buf; i++) { |
1792 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + | |
1793 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1794 | ||
1795 | while (next_unlinked_offset >= | |
1796 | (reg_buf_offset + reg_buf_bytes)) { | |
1797 | /* | |
1798 | * The next di_next_unlinked field is beyond | |
1799 | * the current logged region. Find the next | |
1800 | * logged region that contains or is beyond | |
1801 | * the current di_next_unlinked field. | |
1802 | */ | |
1803 | bit += nbits; | |
e2714bf8 CH |
1804 | bit = xfs_next_bit(buf_f->blf_data_map, |
1805 | buf_f->blf_map_size, bit); | |
1da177e4 LT |
1806 | |
1807 | /* | |
1808 | * If there are no more logged regions in the | |
1809 | * buffer, then we're done. | |
1810 | */ | |
e2714bf8 | 1811 | if (bit == -1) |
1da177e4 | 1812 | return 0; |
1da177e4 | 1813 | |
e2714bf8 CH |
1814 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
1815 | buf_f->blf_map_size, bit); | |
1da177e4 | 1816 | ASSERT(nbits > 0); |
c1155410 DC |
1817 | reg_buf_offset = bit << XFS_BLF_SHIFT; |
1818 | reg_buf_bytes = nbits << XFS_BLF_SHIFT; | |
1da177e4 LT |
1819 | item_index++; |
1820 | } | |
1821 | ||
1822 | /* | |
1823 | * If the current logged region starts after the current | |
1824 | * di_next_unlinked field, then move on to the next | |
1825 | * di_next_unlinked field. | |
1826 | */ | |
e2714bf8 | 1827 | if (next_unlinked_offset < reg_buf_offset) |
1da177e4 | 1828 | continue; |
1da177e4 LT |
1829 | |
1830 | ASSERT(item->ri_buf[item_index].i_addr != NULL); | |
c1155410 | 1831 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0); |
aa0e8833 DC |
1832 | ASSERT((reg_buf_offset + reg_buf_bytes) <= |
1833 | BBTOB(bp->b_io_length)); | |
1da177e4 LT |
1834 | |
1835 | /* | |
1836 | * The current logged region contains a copy of the | |
1837 | * current di_next_unlinked field. Extract its value | |
1838 | * and copy it to the buffer copy. | |
1839 | */ | |
4e0d5f92 CH |
1840 | logged_nextp = item->ri_buf[item_index].i_addr + |
1841 | next_unlinked_offset - reg_buf_offset; | |
1da177e4 | 1842 | if (unlikely(*logged_nextp == 0)) { |
a0fa2b67 DC |
1843 | xfs_alert(mp, |
1844 | "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). " | |
1845 | "Trying to replay bad (0) inode di_next_unlinked field.", | |
1da177e4 LT |
1846 | item, bp); |
1847 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", | |
1848 | XFS_ERRLEVEL_LOW, mp); | |
1849 | return XFS_ERROR(EFSCORRUPTED); | |
1850 | } | |
1851 | ||
1852 | buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp, | |
1853 | next_unlinked_offset); | |
87c199c2 | 1854 | *buffer_nextp = *logged_nextp; |
1da177e4 LT |
1855 | } |
1856 | ||
1857 | return 0; | |
1858 | } | |
1859 | ||
1860 | /* | |
1861 | * Perform a 'normal' buffer recovery. Each logged region of the | |
1862 | * buffer should be copied over the corresponding region in the | |
1863 | * given buffer. The bitmap in the buf log format structure indicates | |
1864 | * where to place the logged data. | |
1865 | */ | |
1da177e4 LT |
1866 | STATIC void |
1867 | xlog_recover_do_reg_buffer( | |
9abbc539 | 1868 | struct xfs_mount *mp, |
1da177e4 | 1869 | xlog_recover_item_t *item, |
e2714bf8 | 1870 | struct xfs_buf *bp, |
1da177e4 LT |
1871 | xfs_buf_log_format_t *buf_f) |
1872 | { | |
1873 | int i; | |
1874 | int bit; | |
1875 | int nbits; | |
1da177e4 LT |
1876 | int error; |
1877 | ||
9abbc539 DC |
1878 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); |
1879 | ||
1da177e4 LT |
1880 | bit = 0; |
1881 | i = 1; /* 0 is the buf format structure */ | |
1882 | while (1) { | |
e2714bf8 CH |
1883 | bit = xfs_next_bit(buf_f->blf_data_map, |
1884 | buf_f->blf_map_size, bit); | |
1da177e4 LT |
1885 | if (bit == -1) |
1886 | break; | |
e2714bf8 CH |
1887 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
1888 | buf_f->blf_map_size, bit); | |
1da177e4 | 1889 | ASSERT(nbits > 0); |
4b80916b | 1890 | ASSERT(item->ri_buf[i].i_addr != NULL); |
c1155410 | 1891 | ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0); |
aa0e8833 DC |
1892 | ASSERT(BBTOB(bp->b_io_length) >= |
1893 | ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT)); | |
1da177e4 LT |
1894 | |
1895 | /* | |
1896 | * Do a sanity check if this is a dquot buffer. Just checking | |
1897 | * the first dquot in the buffer should do. XXXThis is | |
1898 | * probably a good thing to do for other buf types also. | |
1899 | */ | |
1900 | error = 0; | |
c8ad20ff | 1901 | if (buf_f->blf_flags & |
c1155410 | 1902 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
0c5e1ce8 | 1903 | if (item->ri_buf[i].i_addr == NULL) { |
a0fa2b67 | 1904 | xfs_alert(mp, |
0c5e1ce8 CH |
1905 | "XFS: NULL dquot in %s.", __func__); |
1906 | goto next; | |
1907 | } | |
8ec6dba2 | 1908 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { |
a0fa2b67 | 1909 | xfs_alert(mp, |
0c5e1ce8 CH |
1910 | "XFS: dquot too small (%d) in %s.", |
1911 | item->ri_buf[i].i_len, __func__); | |
1912 | goto next; | |
1913 | } | |
a0fa2b67 | 1914 | error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr, |
1da177e4 LT |
1915 | -1, 0, XFS_QMOPT_DOWARN, |
1916 | "dquot_buf_recover"); | |
0c5e1ce8 CH |
1917 | if (error) |
1918 | goto next; | |
1da177e4 | 1919 | } |
0c5e1ce8 CH |
1920 | |
1921 | memcpy(xfs_buf_offset(bp, | |
c1155410 | 1922 | (uint)bit << XFS_BLF_SHIFT), /* dest */ |
0c5e1ce8 | 1923 | item->ri_buf[i].i_addr, /* source */ |
c1155410 | 1924 | nbits<<XFS_BLF_SHIFT); /* length */ |
0c5e1ce8 | 1925 | next: |
1da177e4 LT |
1926 | i++; |
1927 | bit += nbits; | |
1928 | } | |
1929 | ||
1930 | /* Shouldn't be any more regions */ | |
1931 | ASSERT(i == item->ri_total); | |
ee1a47ab CH |
1932 | |
1933 | switch (buf_f->blf_flags & XFS_BLF_TYPE_MASK) { | |
1934 | case XFS_BLF_BTREE_BUF: | |
1935 | switch (be32_to_cpu(*(__be32 *)bp->b_addr)) { | |
1936 | case XFS_ABTB_CRC_MAGIC: | |
1937 | case XFS_ABTC_CRC_MAGIC: | |
1938 | case XFS_ABTB_MAGIC: | |
1939 | case XFS_ABTC_MAGIC: | |
1940 | bp->b_ops = &xfs_allocbt_buf_ops; | |
1941 | break; | |
1942 | case XFS_IBT_CRC_MAGIC: | |
1943 | case XFS_IBT_MAGIC: | |
1944 | bp->b_ops = &xfs_inobt_buf_ops; | |
1945 | break; | |
1946 | case XFS_BMAP_CRC_MAGIC: | |
1947 | case XFS_BMAP_MAGIC: | |
1948 | bp->b_ops = &xfs_bmbt_buf_ops; | |
1949 | break; | |
1950 | default: | |
1951 | xfs_warn(mp, "Bad btree block magic!"); | |
1952 | ASSERT(0); | |
1953 | break; | |
1954 | } | |
1955 | break; | |
4e0e6040 DC |
1956 | case XFS_BLF_AGF_BUF: |
1957 | if (*(__be32 *)bp->b_addr != cpu_to_be32(XFS_AGF_MAGIC)) { | |
1958 | xfs_warn(mp, "Bad AGF block magic!"); | |
1959 | ASSERT(0); | |
1960 | break; | |
1961 | } | |
1962 | bp->b_ops = &xfs_agf_buf_ops; | |
1963 | break; | |
77c95bba CH |
1964 | case XFS_BLF_AGFL_BUF: |
1965 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
1966 | break; | |
1967 | if (*(__be32 *)bp->b_addr != cpu_to_be32(XFS_AGFL_MAGIC)) { | |
1968 | xfs_warn(mp, "Bad AGFL block magic!"); | |
1969 | ASSERT(0); | |
1970 | break; | |
1971 | } | |
1972 | bp->b_ops = &xfs_agfl_buf_ops; | |
1973 | break; | |
ee1a47ab CH |
1974 | default: |
1975 | break; | |
1976 | } | |
1da177e4 LT |
1977 | } |
1978 | ||
1979 | /* | |
1980 | * Do some primitive error checking on ondisk dquot data structures. | |
1981 | */ | |
1982 | int | |
1983 | xfs_qm_dqcheck( | |
a0fa2b67 | 1984 | struct xfs_mount *mp, |
1da177e4 LT |
1985 | xfs_disk_dquot_t *ddq, |
1986 | xfs_dqid_t id, | |
1987 | uint type, /* used only when IO_dorepair is true */ | |
1988 | uint flags, | |
1989 | char *str) | |
1990 | { | |
1991 | xfs_dqblk_t *d = (xfs_dqblk_t *)ddq; | |
1992 | int errs = 0; | |
1993 | ||
1994 | /* | |
1995 | * We can encounter an uninitialized dquot buffer for 2 reasons: | |
1996 | * 1. If we crash while deleting the quotainode(s), and those blks got | |
1997 | * used for user data. This is because we take the path of regular | |
1998 | * file deletion; however, the size field of quotainodes is never | |
1999 | * updated, so all the tricks that we play in itruncate_finish | |
2000 | * don't quite matter. | |
2001 | * | |
2002 | * 2. We don't play the quota buffers when there's a quotaoff logitem. | |
2003 | * But the allocation will be replayed so we'll end up with an | |
2004 | * uninitialized quota block. | |
2005 | * | |
2006 | * This is all fine; things are still consistent, and we haven't lost | |
2007 | * any quota information. Just don't complain about bad dquot blks. | |
2008 | */ | |
69ef921b | 2009 | if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) { |
1da177e4 | 2010 | if (flags & XFS_QMOPT_DOWARN) |
a0fa2b67 | 2011 | xfs_alert(mp, |
1da177e4 | 2012 | "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x", |
1149d96a | 2013 | str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC); |
1da177e4 LT |
2014 | errs++; |
2015 | } | |
1149d96a | 2016 | if (ddq->d_version != XFS_DQUOT_VERSION) { |
1da177e4 | 2017 | if (flags & XFS_QMOPT_DOWARN) |
a0fa2b67 | 2018 | xfs_alert(mp, |
1da177e4 | 2019 | "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x", |
1149d96a | 2020 | str, id, ddq->d_version, XFS_DQUOT_VERSION); |
1da177e4 LT |
2021 | errs++; |
2022 | } | |
2023 | ||
1149d96a CH |
2024 | if (ddq->d_flags != XFS_DQ_USER && |
2025 | ddq->d_flags != XFS_DQ_PROJ && | |
2026 | ddq->d_flags != XFS_DQ_GROUP) { | |
1da177e4 | 2027 | if (flags & XFS_QMOPT_DOWARN) |
a0fa2b67 | 2028 | xfs_alert(mp, |
1da177e4 | 2029 | "%s : XFS dquot ID 0x%x, unknown flags 0x%x", |
1149d96a | 2030 | str, id, ddq->d_flags); |
1da177e4 LT |
2031 | errs++; |
2032 | } | |
2033 | ||
1149d96a | 2034 | if (id != -1 && id != be32_to_cpu(ddq->d_id)) { |
1da177e4 | 2035 | if (flags & XFS_QMOPT_DOWARN) |
a0fa2b67 | 2036 | xfs_alert(mp, |
1da177e4 LT |
2037 | "%s : ondisk-dquot 0x%p, ID mismatch: " |
2038 | "0x%x expected, found id 0x%x", | |
1149d96a | 2039 | str, ddq, id, be32_to_cpu(ddq->d_id)); |
1da177e4 LT |
2040 | errs++; |
2041 | } | |
2042 | ||
2043 | if (!errs && ddq->d_id) { | |
1149d96a | 2044 | if (ddq->d_blk_softlimit && |
d0a3fe67 | 2045 | be64_to_cpu(ddq->d_bcount) > |
1149d96a | 2046 | be64_to_cpu(ddq->d_blk_softlimit)) { |
1da177e4 LT |
2047 | if (!ddq->d_btimer) { |
2048 | if (flags & XFS_QMOPT_DOWARN) | |
a0fa2b67 DC |
2049 | xfs_alert(mp, |
2050 | "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED", | |
1149d96a | 2051 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2052 | errs++; |
2053 | } | |
2054 | } | |
1149d96a | 2055 | if (ddq->d_ino_softlimit && |
d0a3fe67 | 2056 | be64_to_cpu(ddq->d_icount) > |
1149d96a | 2057 | be64_to_cpu(ddq->d_ino_softlimit)) { |
1da177e4 LT |
2058 | if (!ddq->d_itimer) { |
2059 | if (flags & XFS_QMOPT_DOWARN) | |
a0fa2b67 DC |
2060 | xfs_alert(mp, |
2061 | "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED", | |
1149d96a | 2062 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2063 | errs++; |
2064 | } | |
2065 | } | |
1149d96a | 2066 | if (ddq->d_rtb_softlimit && |
d0a3fe67 | 2067 | be64_to_cpu(ddq->d_rtbcount) > |
1149d96a | 2068 | be64_to_cpu(ddq->d_rtb_softlimit)) { |
1da177e4 LT |
2069 | if (!ddq->d_rtbtimer) { |
2070 | if (flags & XFS_QMOPT_DOWARN) | |
a0fa2b67 DC |
2071 | xfs_alert(mp, |
2072 | "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED", | |
1149d96a | 2073 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
1da177e4 LT |
2074 | errs++; |
2075 | } | |
2076 | } | |
2077 | } | |
2078 | ||
2079 | if (!errs || !(flags & XFS_QMOPT_DQREPAIR)) | |
2080 | return errs; | |
2081 | ||
2082 | if (flags & XFS_QMOPT_DOWARN) | |
a0fa2b67 | 2083 | xfs_notice(mp, "Re-initializing dquot ID 0x%x", id); |
1da177e4 LT |
2084 | |
2085 | /* | |
2086 | * Typically, a repair is only requested by quotacheck. | |
2087 | */ | |
2088 | ASSERT(id != -1); | |
2089 | ASSERT(flags & XFS_QMOPT_DQREPAIR); | |
2090 | memset(d, 0, sizeof(xfs_dqblk_t)); | |
1149d96a CH |
2091 | |
2092 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
2093 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
2094 | d->dd_diskdq.d_flags = type; | |
2095 | d->dd_diskdq.d_id = cpu_to_be32(id); | |
1da177e4 LT |
2096 | |
2097 | return errs; | |
2098 | } | |
2099 | ||
2100 | /* | |
2101 | * Perform a dquot buffer recovery. | |
2102 | * Simple algorithm: if we have found a QUOTAOFF logitem of the same type | |
2103 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. | |
2104 | * Else, treat it as a regular buffer and do recovery. | |
2105 | */ | |
2106 | STATIC void | |
2107 | xlog_recover_do_dquot_buffer( | |
9a8d2fdb MT |
2108 | struct xfs_mount *mp, |
2109 | struct xlog *log, | |
2110 | struct xlog_recover_item *item, | |
2111 | struct xfs_buf *bp, | |
2112 | struct xfs_buf_log_format *buf_f) | |
1da177e4 LT |
2113 | { |
2114 | uint type; | |
2115 | ||
9abbc539 DC |
2116 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
2117 | ||
1da177e4 LT |
2118 | /* |
2119 | * Filesystems are required to send in quota flags at mount time. | |
2120 | */ | |
2121 | if (mp->m_qflags == 0) { | |
2122 | return; | |
2123 | } | |
2124 | ||
2125 | type = 0; | |
c1155410 | 2126 | if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF) |
1da177e4 | 2127 | type |= XFS_DQ_USER; |
c1155410 | 2128 | if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF) |
c8ad20ff | 2129 | type |= XFS_DQ_PROJ; |
c1155410 | 2130 | if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF) |
1da177e4 LT |
2131 | type |= XFS_DQ_GROUP; |
2132 | /* | |
2133 | * This type of quotas was turned off, so ignore this buffer | |
2134 | */ | |
2135 | if (log->l_quotaoffs_flag & type) | |
2136 | return; | |
2137 | ||
9abbc539 | 2138 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 LT |
2139 | } |
2140 | ||
2141 | /* | |
2142 | * This routine replays a modification made to a buffer at runtime. | |
2143 | * There are actually two types of buffer, regular and inode, which | |
2144 | * are handled differently. Inode buffers are handled differently | |
2145 | * in that we only recover a specific set of data from them, namely | |
2146 | * the inode di_next_unlinked fields. This is because all other inode | |
2147 | * data is actually logged via inode records and any data we replay | |
2148 | * here which overlaps that may be stale. | |
2149 | * | |
2150 | * When meta-data buffers are freed at run time we log a buffer item | |
c1155410 | 2151 | * with the XFS_BLF_CANCEL bit set to indicate that previous copies |
1da177e4 LT |
2152 | * of the buffer in the log should not be replayed at recovery time. |
2153 | * This is so that if the blocks covered by the buffer are reused for | |
2154 | * file data before we crash we don't end up replaying old, freed | |
2155 | * meta-data into a user's file. | |
2156 | * | |
2157 | * To handle the cancellation of buffer log items, we make two passes | |
2158 | * over the log during recovery. During the first we build a table of | |
2159 | * those buffers which have been cancelled, and during the second we | |
2160 | * only replay those buffers which do not have corresponding cancel | |
2161 | * records in the table. See xlog_recover_do_buffer_pass[1,2] above | |
2162 | * for more details on the implementation of the table of cancel records. | |
2163 | */ | |
2164 | STATIC int | |
c9f71f5f | 2165 | xlog_recover_buffer_pass2( |
9a8d2fdb MT |
2166 | struct xlog *log, |
2167 | struct list_head *buffer_list, | |
2168 | struct xlog_recover_item *item) | |
1da177e4 | 2169 | { |
4e0d5f92 | 2170 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
e2714bf8 | 2171 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2172 | xfs_buf_t *bp; |
2173 | int error; | |
6ad112bf | 2174 | uint buf_flags; |
1da177e4 | 2175 | |
c9f71f5f CH |
2176 | /* |
2177 | * In this pass we only want to recover all the buffers which have | |
2178 | * not been cancelled and are not cancellation buffers themselves. | |
2179 | */ | |
2180 | if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno, | |
2181 | buf_f->blf_len, buf_f->blf_flags)) { | |
2182 | trace_xfs_log_recover_buf_cancel(log, buf_f); | |
1da177e4 | 2183 | return 0; |
1da177e4 | 2184 | } |
c9f71f5f | 2185 | |
9abbc539 | 2186 | trace_xfs_log_recover_buf_recover(log, buf_f); |
1da177e4 | 2187 | |
a8acad70 | 2188 | buf_flags = 0; |
611c9946 DC |
2189 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) |
2190 | buf_flags |= XBF_UNMAPPED; | |
6ad112bf | 2191 | |
e2714bf8 | 2192 | bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len, |
c3f8fc73 | 2193 | buf_flags, NULL); |
ac4d6888 CS |
2194 | if (!bp) |
2195 | return XFS_ERROR(ENOMEM); | |
e5702805 | 2196 | error = bp->b_error; |
5a52c2a5 | 2197 | if (error) { |
901796af | 2198 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)"); |
1da177e4 LT |
2199 | xfs_buf_relse(bp); |
2200 | return error; | |
2201 | } | |
2202 | ||
e2714bf8 | 2203 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
1da177e4 | 2204 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); |
e2714bf8 | 2205 | } else if (buf_f->blf_flags & |
c1155410 | 2206 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
1da177e4 LT |
2207 | xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); |
2208 | } else { | |
9abbc539 | 2209 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 LT |
2210 | } |
2211 | if (error) | |
2212 | return XFS_ERROR(error); | |
2213 | ||
2214 | /* | |
2215 | * Perform delayed write on the buffer. Asynchronous writes will be | |
2216 | * slower when taking into account all the buffers to be flushed. | |
2217 | * | |
2218 | * Also make sure that only inode buffers with good sizes stay in | |
2219 | * the buffer cache. The kernel moves inodes in buffers of 1 block | |
2220 | * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode | |
2221 | * buffers in the log can be a different size if the log was generated | |
2222 | * by an older kernel using unclustered inode buffers or a newer kernel | |
2223 | * running with a different inode cluster size. Regardless, if the | |
2224 | * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE) | |
2225 | * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep | |
2226 | * the buffer out of the buffer cache so that the buffer won't | |
2227 | * overlap with future reads of those inodes. | |
2228 | */ | |
2229 | if (XFS_DINODE_MAGIC == | |
b53e675d | 2230 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
aa0e8833 | 2231 | (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize, |
1da177e4 | 2232 | (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) { |
c867cb61 | 2233 | xfs_buf_stale(bp); |
c2b006c1 | 2234 | error = xfs_bwrite(bp); |
1da177e4 | 2235 | } else { |
ebad861b | 2236 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2237 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2238 | xfs_buf_delwri_queue(bp, buffer_list); |
1da177e4 LT |
2239 | } |
2240 | ||
c2b006c1 CH |
2241 | xfs_buf_relse(bp); |
2242 | return error; | |
1da177e4 LT |
2243 | } |
2244 | ||
2245 | STATIC int | |
c9f71f5f | 2246 | xlog_recover_inode_pass2( |
9a8d2fdb MT |
2247 | struct xlog *log, |
2248 | struct list_head *buffer_list, | |
2249 | struct xlog_recover_item *item) | |
1da177e4 LT |
2250 | { |
2251 | xfs_inode_log_format_t *in_f; | |
c9f71f5f | 2252 | xfs_mount_t *mp = log->l_mp; |
1da177e4 | 2253 | xfs_buf_t *bp; |
1da177e4 | 2254 | xfs_dinode_t *dip; |
1da177e4 LT |
2255 | int len; |
2256 | xfs_caddr_t src; | |
2257 | xfs_caddr_t dest; | |
2258 | int error; | |
2259 | int attr_index; | |
2260 | uint fields; | |
347d1c01 | 2261 | xfs_icdinode_t *dicp; |
6d192a9b | 2262 | int need_free = 0; |
1da177e4 | 2263 | |
6d192a9b | 2264 | if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) { |
4e0d5f92 | 2265 | in_f = item->ri_buf[0].i_addr; |
6d192a9b | 2266 | } else { |
4e0d5f92 | 2267 | in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP); |
6d192a9b TS |
2268 | need_free = 1; |
2269 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); | |
2270 | if (error) | |
2271 | goto error; | |
2272 | } | |
1da177e4 LT |
2273 | |
2274 | /* | |
2275 | * Inode buffers can be freed, look out for it, | |
2276 | * and do not replay the inode. | |
2277 | */ | |
a1941895 CH |
2278 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
2279 | in_f->ilf_len, 0)) { | |
6d192a9b | 2280 | error = 0; |
9abbc539 | 2281 | trace_xfs_log_recover_inode_cancel(log, in_f); |
6d192a9b TS |
2282 | goto error; |
2283 | } | |
9abbc539 | 2284 | trace_xfs_log_recover_inode_recover(log, in_f); |
1da177e4 | 2285 | |
c3f8fc73 DC |
2286 | bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0, |
2287 | NULL); | |
ac4d6888 CS |
2288 | if (!bp) { |
2289 | error = ENOMEM; | |
2290 | goto error; | |
2291 | } | |
e5702805 | 2292 | error = bp->b_error; |
5a52c2a5 | 2293 | if (error) { |
901796af | 2294 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)"); |
1da177e4 | 2295 | xfs_buf_relse(bp); |
6d192a9b | 2296 | goto error; |
1da177e4 | 2297 | } |
1da177e4 | 2298 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); |
a1941895 | 2299 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset); |
1da177e4 LT |
2300 | |
2301 | /* | |
2302 | * Make sure the place we're flushing out to really looks | |
2303 | * like an inode! | |
2304 | */ | |
69ef921b | 2305 | if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) { |
1da177e4 | 2306 | xfs_buf_relse(bp); |
a0fa2b67 DC |
2307 | xfs_alert(mp, |
2308 | "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld", | |
2309 | __func__, dip, bp, in_f->ilf_ino); | |
c9f71f5f | 2310 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)", |
1da177e4 | 2311 | XFS_ERRLEVEL_LOW, mp); |
6d192a9b TS |
2312 | error = EFSCORRUPTED; |
2313 | goto error; | |
1da177e4 | 2314 | } |
4e0d5f92 | 2315 | dicp = item->ri_buf[1].i_addr; |
1da177e4 LT |
2316 | if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) { |
2317 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2318 | xfs_alert(mp, |
2319 | "%s: Bad inode log record, rec ptr 0x%p, ino %Ld", | |
2320 | __func__, item, in_f->ilf_ino); | |
c9f71f5f | 2321 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)", |
1da177e4 | 2322 | XFS_ERRLEVEL_LOW, mp); |
6d192a9b TS |
2323 | error = EFSCORRUPTED; |
2324 | goto error; | |
1da177e4 LT |
2325 | } |
2326 | ||
2327 | /* Skip replay when the on disk inode is newer than the log one */ | |
81591fe2 | 2328 | if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) { |
1da177e4 LT |
2329 | /* |
2330 | * Deal with the wrap case, DI_MAX_FLUSH is less | |
2331 | * than smaller numbers | |
2332 | */ | |
81591fe2 | 2333 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
347d1c01 | 2334 | dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
1da177e4 LT |
2335 | /* do nothing */ |
2336 | } else { | |
2337 | xfs_buf_relse(bp); | |
9abbc539 | 2338 | trace_xfs_log_recover_inode_skip(log, in_f); |
6d192a9b TS |
2339 | error = 0; |
2340 | goto error; | |
1da177e4 LT |
2341 | } |
2342 | } | |
2343 | /* Take the opportunity to reset the flush iteration count */ | |
2344 | dicp->di_flushiter = 0; | |
2345 | ||
abbede1b | 2346 | if (unlikely(S_ISREG(dicp->di_mode))) { |
1da177e4 LT |
2347 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
2348 | (dicp->di_format != XFS_DINODE_FMT_BTREE)) { | |
c9f71f5f | 2349 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)", |
1da177e4 LT |
2350 | XFS_ERRLEVEL_LOW, mp, dicp); |
2351 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2352 | xfs_alert(mp, |
2353 | "%s: Bad regular inode log record, rec ptr 0x%p, " | |
2354 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2355 | __func__, item, dip, bp, in_f->ilf_ino); | |
6d192a9b TS |
2356 | error = EFSCORRUPTED; |
2357 | goto error; | |
1da177e4 | 2358 | } |
abbede1b | 2359 | } else if (unlikely(S_ISDIR(dicp->di_mode))) { |
1da177e4 LT |
2360 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
2361 | (dicp->di_format != XFS_DINODE_FMT_BTREE) && | |
2362 | (dicp->di_format != XFS_DINODE_FMT_LOCAL)) { | |
c9f71f5f | 2363 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)", |
1da177e4 LT |
2364 | XFS_ERRLEVEL_LOW, mp, dicp); |
2365 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2366 | xfs_alert(mp, |
2367 | "%s: Bad dir inode log record, rec ptr 0x%p, " | |
2368 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2369 | __func__, item, dip, bp, in_f->ilf_ino); | |
6d192a9b TS |
2370 | error = EFSCORRUPTED; |
2371 | goto error; | |
1da177e4 LT |
2372 | } |
2373 | } | |
2374 | if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){ | |
c9f71f5f | 2375 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)", |
1da177e4 LT |
2376 | XFS_ERRLEVEL_LOW, mp, dicp); |
2377 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2378 | xfs_alert(mp, |
2379 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
2380 | "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld", | |
2381 | __func__, item, dip, bp, in_f->ilf_ino, | |
1da177e4 LT |
2382 | dicp->di_nextents + dicp->di_anextents, |
2383 | dicp->di_nblocks); | |
6d192a9b TS |
2384 | error = EFSCORRUPTED; |
2385 | goto error; | |
1da177e4 LT |
2386 | } |
2387 | if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) { | |
c9f71f5f | 2388 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)", |
1da177e4 LT |
2389 | XFS_ERRLEVEL_LOW, mp, dicp); |
2390 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2391 | xfs_alert(mp, |
2392 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
2393 | "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__, | |
c9f71f5f | 2394 | item, dip, bp, in_f->ilf_ino, dicp->di_forkoff); |
6d192a9b TS |
2395 | error = EFSCORRUPTED; |
2396 | goto error; | |
1da177e4 | 2397 | } |
81591fe2 | 2398 | if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) { |
c9f71f5f | 2399 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)", |
1da177e4 LT |
2400 | XFS_ERRLEVEL_LOW, mp, dicp); |
2401 | xfs_buf_relse(bp); | |
a0fa2b67 DC |
2402 | xfs_alert(mp, |
2403 | "%s: Bad inode log record length %d, rec ptr 0x%p", | |
2404 | __func__, item->ri_buf[1].i_len, item); | |
6d192a9b TS |
2405 | error = EFSCORRUPTED; |
2406 | goto error; | |
1da177e4 LT |
2407 | } |
2408 | ||
2409 | /* The core is in in-core format */ | |
4e0d5f92 | 2410 | xfs_dinode_to_disk(dip, item->ri_buf[1].i_addr); |
1da177e4 LT |
2411 | |
2412 | /* the rest is in on-disk format */ | |
81591fe2 CH |
2413 | if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) { |
2414 | memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode), | |
2415 | item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode), | |
2416 | item->ri_buf[1].i_len - sizeof(struct xfs_icdinode)); | |
1da177e4 LT |
2417 | } |
2418 | ||
2419 | fields = in_f->ilf_fields; | |
2420 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { | |
2421 | case XFS_ILOG_DEV: | |
81591fe2 | 2422 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
1da177e4 LT |
2423 | break; |
2424 | case XFS_ILOG_UUID: | |
81591fe2 CH |
2425 | memcpy(XFS_DFORK_DPTR(dip), |
2426 | &in_f->ilf_u.ilfu_uuid, | |
2427 | sizeof(uuid_t)); | |
1da177e4 LT |
2428 | break; |
2429 | } | |
2430 | ||
2431 | if (in_f->ilf_size == 2) | |
2432 | goto write_inode_buffer; | |
2433 | len = item->ri_buf[2].i_len; | |
2434 | src = item->ri_buf[2].i_addr; | |
2435 | ASSERT(in_f->ilf_size <= 4); | |
2436 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); | |
2437 | ASSERT(!(fields & XFS_ILOG_DFORK) || | |
2438 | (len == in_f->ilf_dsize)); | |
2439 | ||
2440 | switch (fields & XFS_ILOG_DFORK) { | |
2441 | case XFS_ILOG_DDATA: | |
2442 | case XFS_ILOG_DEXT: | |
81591fe2 | 2443 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
1da177e4 LT |
2444 | break; |
2445 | ||
2446 | case XFS_ILOG_DBROOT: | |
7cc95a82 | 2447 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
81591fe2 | 2448 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
1da177e4 LT |
2449 | XFS_DFORK_DSIZE(dip, mp)); |
2450 | break; | |
2451 | ||
2452 | default: | |
2453 | /* | |
2454 | * There are no data fork flags set. | |
2455 | */ | |
2456 | ASSERT((fields & XFS_ILOG_DFORK) == 0); | |
2457 | break; | |
2458 | } | |
2459 | ||
2460 | /* | |
2461 | * If we logged any attribute data, recover it. There may or | |
2462 | * may not have been any other non-core data logged in this | |
2463 | * transaction. | |
2464 | */ | |
2465 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2466 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { | |
2467 | attr_index = 3; | |
2468 | } else { | |
2469 | attr_index = 2; | |
2470 | } | |
2471 | len = item->ri_buf[attr_index].i_len; | |
2472 | src = item->ri_buf[attr_index].i_addr; | |
2473 | ASSERT(len == in_f->ilf_asize); | |
2474 | ||
2475 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2476 | case XFS_ILOG_ADATA: | |
2477 | case XFS_ILOG_AEXT: | |
2478 | dest = XFS_DFORK_APTR(dip); | |
2479 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); | |
2480 | memcpy(dest, src, len); | |
2481 | break; | |
2482 | ||
2483 | case XFS_ILOG_ABROOT: | |
2484 | dest = XFS_DFORK_APTR(dip); | |
7cc95a82 CH |
2485 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
2486 | len, (xfs_bmdr_block_t*)dest, | |
1da177e4 LT |
2487 | XFS_DFORK_ASIZE(dip, mp)); |
2488 | break; | |
2489 | ||
2490 | default: | |
a0fa2b67 | 2491 | xfs_warn(log->l_mp, "%s: Invalid flag", __func__); |
1da177e4 LT |
2492 | ASSERT(0); |
2493 | xfs_buf_relse(bp); | |
6d192a9b TS |
2494 | error = EIO; |
2495 | goto error; | |
1da177e4 LT |
2496 | } |
2497 | } | |
2498 | ||
2499 | write_inode_buffer: | |
ebad861b | 2500 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2501 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2502 | xfs_buf_delwri_queue(bp, buffer_list); |
61551f1e | 2503 | xfs_buf_relse(bp); |
6d192a9b TS |
2504 | error: |
2505 | if (need_free) | |
f0e2d93c | 2506 | kmem_free(in_f); |
6d192a9b | 2507 | return XFS_ERROR(error); |
1da177e4 LT |
2508 | } |
2509 | ||
2510 | /* | |
9a8d2fdb | 2511 | * Recover QUOTAOFF records. We simply make a note of it in the xlog |
1da177e4 LT |
2512 | * structure, so that we know not to do any dquot item or dquot buffer recovery, |
2513 | * of that type. | |
2514 | */ | |
2515 | STATIC int | |
c9f71f5f | 2516 | xlog_recover_quotaoff_pass1( |
9a8d2fdb MT |
2517 | struct xlog *log, |
2518 | struct xlog_recover_item *item) | |
1da177e4 | 2519 | { |
c9f71f5f | 2520 | xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr; |
1da177e4 LT |
2521 | ASSERT(qoff_f); |
2522 | ||
2523 | /* | |
2524 | * The logitem format's flag tells us if this was user quotaoff, | |
77a7cce4 | 2525 | * group/project quotaoff or both. |
1da177e4 LT |
2526 | */ |
2527 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) | |
2528 | log->l_quotaoffs_flag |= XFS_DQ_USER; | |
77a7cce4 NS |
2529 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
2530 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; | |
1da177e4 LT |
2531 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
2532 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; | |
2533 | ||
2534 | return (0); | |
2535 | } | |
2536 | ||
2537 | /* | |
2538 | * Recover a dquot record | |
2539 | */ | |
2540 | STATIC int | |
c9f71f5f | 2541 | xlog_recover_dquot_pass2( |
9a8d2fdb MT |
2542 | struct xlog *log, |
2543 | struct list_head *buffer_list, | |
2544 | struct xlog_recover_item *item) | |
1da177e4 | 2545 | { |
c9f71f5f | 2546 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2547 | xfs_buf_t *bp; |
2548 | struct xfs_disk_dquot *ddq, *recddq; | |
2549 | int error; | |
2550 | xfs_dq_logformat_t *dq_f; | |
2551 | uint type; | |
2552 | ||
1da177e4 LT |
2553 | |
2554 | /* | |
2555 | * Filesystems are required to send in quota flags at mount time. | |
2556 | */ | |
2557 | if (mp->m_qflags == 0) | |
2558 | return (0); | |
2559 | ||
4e0d5f92 CH |
2560 | recddq = item->ri_buf[1].i_addr; |
2561 | if (recddq == NULL) { | |
a0fa2b67 | 2562 | xfs_alert(log->l_mp, "NULL dquot in %s.", __func__); |
0c5e1ce8 CH |
2563 | return XFS_ERROR(EIO); |
2564 | } | |
8ec6dba2 | 2565 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
a0fa2b67 | 2566 | xfs_alert(log->l_mp, "dquot too small (%d) in %s.", |
0c5e1ce8 CH |
2567 | item->ri_buf[1].i_len, __func__); |
2568 | return XFS_ERROR(EIO); | |
2569 | } | |
2570 | ||
1da177e4 LT |
2571 | /* |
2572 | * This type of quotas was turned off, so ignore this record. | |
2573 | */ | |
b53e675d | 2574 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
1da177e4 LT |
2575 | ASSERT(type); |
2576 | if (log->l_quotaoffs_flag & type) | |
2577 | return (0); | |
2578 | ||
2579 | /* | |
2580 | * At this point we know that quota was _not_ turned off. | |
2581 | * Since the mount flags are not indicating to us otherwise, this | |
2582 | * must mean that quota is on, and the dquot needs to be replayed. | |
2583 | * Remember that we may not have fully recovered the superblock yet, | |
2584 | * so we can't do the usual trick of looking at the SB quota bits. | |
2585 | * | |
2586 | * The other possibility, of course, is that the quota subsystem was | |
2587 | * removed since the last mount - ENOSYS. | |
2588 | */ | |
4e0d5f92 | 2589 | dq_f = item->ri_buf[0].i_addr; |
1da177e4 | 2590 | ASSERT(dq_f); |
a0fa2b67 DC |
2591 | error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, |
2592 | "xlog_recover_dquot_pass2 (log copy)"); | |
2593 | if (error) | |
1da177e4 | 2594 | return XFS_ERROR(EIO); |
1da177e4 LT |
2595 | ASSERT(dq_f->qlf_len == 1); |
2596 | ||
7ca790a5 | 2597 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno, |
c3f8fc73 DC |
2598 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp, |
2599 | NULL); | |
7ca790a5 | 2600 | if (error) |
1da177e4 | 2601 | return error; |
7ca790a5 | 2602 | |
1da177e4 LT |
2603 | ASSERT(bp); |
2604 | ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset); | |
2605 | ||
2606 | /* | |
2607 | * At least the magic num portion should be on disk because this | |
2608 | * was among a chunk of dquots created earlier, and we did some | |
2609 | * minimal initialization then. | |
2610 | */ | |
a0fa2b67 DC |
2611 | error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, |
2612 | "xlog_recover_dquot_pass2"); | |
2613 | if (error) { | |
1da177e4 LT |
2614 | xfs_buf_relse(bp); |
2615 | return XFS_ERROR(EIO); | |
2616 | } | |
2617 | ||
2618 | memcpy(ddq, recddq, item->ri_buf[1].i_len); | |
2619 | ||
2620 | ASSERT(dq_f->qlf_size == 2); | |
ebad861b | 2621 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2622 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2623 | xfs_buf_delwri_queue(bp, buffer_list); |
61551f1e | 2624 | xfs_buf_relse(bp); |
1da177e4 LT |
2625 | |
2626 | return (0); | |
2627 | } | |
2628 | ||
2629 | /* | |
2630 | * This routine is called to create an in-core extent free intent | |
2631 | * item from the efi format structure which was logged on disk. | |
2632 | * It allocates an in-core efi, copies the extents from the format | |
2633 | * structure into it, and adds the efi to the AIL with the given | |
2634 | * LSN. | |
2635 | */ | |
6d192a9b | 2636 | STATIC int |
c9f71f5f | 2637 | xlog_recover_efi_pass2( |
9a8d2fdb MT |
2638 | struct xlog *log, |
2639 | struct xlog_recover_item *item, | |
2640 | xfs_lsn_t lsn) | |
1da177e4 | 2641 | { |
6d192a9b | 2642 | int error; |
c9f71f5f | 2643 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2644 | xfs_efi_log_item_t *efip; |
2645 | xfs_efi_log_format_t *efi_formatp; | |
1da177e4 | 2646 | |
4e0d5f92 | 2647 | efi_formatp = item->ri_buf[0].i_addr; |
1da177e4 | 2648 | |
1da177e4 | 2649 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); |
6d192a9b TS |
2650 | if ((error = xfs_efi_copy_format(&(item->ri_buf[0]), |
2651 | &(efip->efi_format)))) { | |
2652 | xfs_efi_item_free(efip); | |
2653 | return error; | |
2654 | } | |
b199c8a4 | 2655 | atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents); |
1da177e4 | 2656 | |
a9c21c1b | 2657 | spin_lock(&log->l_ailp->xa_lock); |
1da177e4 | 2658 | /* |
783a2f65 | 2659 | * xfs_trans_ail_update() drops the AIL lock. |
1da177e4 | 2660 | */ |
e6059949 | 2661 | xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn); |
6d192a9b | 2662 | return 0; |
1da177e4 LT |
2663 | } |
2664 | ||
2665 | ||
2666 | /* | |
2667 | * This routine is called when an efd format structure is found in | |
2668 | * a committed transaction in the log. It's purpose is to cancel | |
2669 | * the corresponding efi if it was still in the log. To do this | |
2670 | * it searches the AIL for the efi with an id equal to that in the | |
2671 | * efd format structure. If we find it, we remove the efi from the | |
2672 | * AIL and free it. | |
2673 | */ | |
c9f71f5f CH |
2674 | STATIC int |
2675 | xlog_recover_efd_pass2( | |
9a8d2fdb MT |
2676 | struct xlog *log, |
2677 | struct xlog_recover_item *item) | |
1da177e4 | 2678 | { |
1da177e4 LT |
2679 | xfs_efd_log_format_t *efd_formatp; |
2680 | xfs_efi_log_item_t *efip = NULL; | |
2681 | xfs_log_item_t *lip; | |
1da177e4 | 2682 | __uint64_t efi_id; |
27d8d5fe | 2683 | struct xfs_ail_cursor cur; |
783a2f65 | 2684 | struct xfs_ail *ailp = log->l_ailp; |
1da177e4 | 2685 | |
4e0d5f92 | 2686 | efd_formatp = item->ri_buf[0].i_addr; |
6d192a9b TS |
2687 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
2688 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || | |
2689 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + | |
2690 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); | |
1da177e4 LT |
2691 | efi_id = efd_formatp->efd_efi_id; |
2692 | ||
2693 | /* | |
2694 | * Search for the efi with the id in the efd format structure | |
2695 | * in the AIL. | |
2696 | */ | |
a9c21c1b DC |
2697 | spin_lock(&ailp->xa_lock); |
2698 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
2699 | while (lip != NULL) { |
2700 | if (lip->li_type == XFS_LI_EFI) { | |
2701 | efip = (xfs_efi_log_item_t *)lip; | |
2702 | if (efip->efi_format.efi_id == efi_id) { | |
2703 | /* | |
783a2f65 | 2704 | * xfs_trans_ail_delete() drops the |
1da177e4 LT |
2705 | * AIL lock. |
2706 | */ | |
04913fdd DC |
2707 | xfs_trans_ail_delete(ailp, lip, |
2708 | SHUTDOWN_CORRUPT_INCORE); | |
8ae2c0f6 | 2709 | xfs_efi_item_free(efip); |
a9c21c1b | 2710 | spin_lock(&ailp->xa_lock); |
27d8d5fe | 2711 | break; |
1da177e4 LT |
2712 | } |
2713 | } | |
a9c21c1b | 2714 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 2715 | } |
a9c21c1b DC |
2716 | xfs_trans_ail_cursor_done(ailp, &cur); |
2717 | spin_unlock(&ailp->xa_lock); | |
c9f71f5f CH |
2718 | |
2719 | return 0; | |
1da177e4 LT |
2720 | } |
2721 | ||
1da177e4 LT |
2722 | /* |
2723 | * Free up any resources allocated by the transaction | |
2724 | * | |
2725 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. | |
2726 | */ | |
2727 | STATIC void | |
2728 | xlog_recover_free_trans( | |
d0450948 | 2729 | struct xlog_recover *trans) |
1da177e4 | 2730 | { |
f0a76953 | 2731 | xlog_recover_item_t *item, *n; |
1da177e4 LT |
2732 | int i; |
2733 | ||
f0a76953 DC |
2734 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { |
2735 | /* Free the regions in the item. */ | |
2736 | list_del(&item->ri_list); | |
2737 | for (i = 0; i < item->ri_cnt; i++) | |
2738 | kmem_free(item->ri_buf[i].i_addr); | |
1da177e4 | 2739 | /* Free the item itself */ |
f0a76953 DC |
2740 | kmem_free(item->ri_buf); |
2741 | kmem_free(item); | |
2742 | } | |
1da177e4 | 2743 | /* Free the transaction recover structure */ |
f0e2d93c | 2744 | kmem_free(trans); |
1da177e4 LT |
2745 | } |
2746 | ||
d0450948 | 2747 | STATIC int |
c9f71f5f | 2748 | xlog_recover_commit_pass1( |
ad223e60 MT |
2749 | struct xlog *log, |
2750 | struct xlog_recover *trans, | |
2751 | struct xlog_recover_item *item) | |
d0450948 | 2752 | { |
c9f71f5f | 2753 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1); |
d0450948 CH |
2754 | |
2755 | switch (ITEM_TYPE(item)) { | |
2756 | case XFS_LI_BUF: | |
c9f71f5f CH |
2757 | return xlog_recover_buffer_pass1(log, item); |
2758 | case XFS_LI_QUOTAOFF: | |
2759 | return xlog_recover_quotaoff_pass1(log, item); | |
d0450948 | 2760 | case XFS_LI_INODE: |
d0450948 | 2761 | case XFS_LI_EFI: |
d0450948 | 2762 | case XFS_LI_EFD: |
c9f71f5f CH |
2763 | case XFS_LI_DQUOT: |
2764 | /* nothing to do in pass 1 */ | |
d0450948 | 2765 | return 0; |
c9f71f5f | 2766 | default: |
a0fa2b67 DC |
2767 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
2768 | __func__, ITEM_TYPE(item)); | |
c9f71f5f CH |
2769 | ASSERT(0); |
2770 | return XFS_ERROR(EIO); | |
2771 | } | |
2772 | } | |
2773 | ||
2774 | STATIC int | |
2775 | xlog_recover_commit_pass2( | |
ad223e60 MT |
2776 | struct xlog *log, |
2777 | struct xlog_recover *trans, | |
2778 | struct list_head *buffer_list, | |
2779 | struct xlog_recover_item *item) | |
c9f71f5f CH |
2780 | { |
2781 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2); | |
2782 | ||
2783 | switch (ITEM_TYPE(item)) { | |
2784 | case XFS_LI_BUF: | |
43ff2122 | 2785 | return xlog_recover_buffer_pass2(log, buffer_list, item); |
c9f71f5f | 2786 | case XFS_LI_INODE: |
43ff2122 | 2787 | return xlog_recover_inode_pass2(log, buffer_list, item); |
c9f71f5f CH |
2788 | case XFS_LI_EFI: |
2789 | return xlog_recover_efi_pass2(log, item, trans->r_lsn); | |
2790 | case XFS_LI_EFD: | |
2791 | return xlog_recover_efd_pass2(log, item); | |
d0450948 | 2792 | case XFS_LI_DQUOT: |
43ff2122 | 2793 | return xlog_recover_dquot_pass2(log, buffer_list, item); |
d0450948 | 2794 | case XFS_LI_QUOTAOFF: |
c9f71f5f CH |
2795 | /* nothing to do in pass2 */ |
2796 | return 0; | |
d0450948 | 2797 | default: |
a0fa2b67 DC |
2798 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
2799 | __func__, ITEM_TYPE(item)); | |
d0450948 CH |
2800 | ASSERT(0); |
2801 | return XFS_ERROR(EIO); | |
2802 | } | |
2803 | } | |
2804 | ||
2805 | /* | |
2806 | * Perform the transaction. | |
2807 | * | |
2808 | * If the transaction modifies a buffer or inode, do it now. Otherwise, | |
2809 | * EFIs and EFDs get queued up by adding entries into the AIL for them. | |
2810 | */ | |
1da177e4 LT |
2811 | STATIC int |
2812 | xlog_recover_commit_trans( | |
ad223e60 | 2813 | struct xlog *log, |
d0450948 | 2814 | struct xlog_recover *trans, |
1da177e4 LT |
2815 | int pass) |
2816 | { | |
43ff2122 | 2817 | int error = 0, error2; |
d0450948 | 2818 | xlog_recover_item_t *item; |
43ff2122 | 2819 | LIST_HEAD (buffer_list); |
1da177e4 | 2820 | |
f0a76953 | 2821 | hlist_del(&trans->r_list); |
d0450948 CH |
2822 | |
2823 | error = xlog_recover_reorder_trans(log, trans, pass); | |
2824 | if (error) | |
1da177e4 | 2825 | return error; |
d0450948 CH |
2826 | |
2827 | list_for_each_entry(item, &trans->r_itemq, ri_list) { | |
43ff2122 CH |
2828 | switch (pass) { |
2829 | case XLOG_RECOVER_PASS1: | |
c9f71f5f | 2830 | error = xlog_recover_commit_pass1(log, trans, item); |
43ff2122 CH |
2831 | break; |
2832 | case XLOG_RECOVER_PASS2: | |
2833 | error = xlog_recover_commit_pass2(log, trans, | |
2834 | &buffer_list, item); | |
2835 | break; | |
2836 | default: | |
2837 | ASSERT(0); | |
2838 | } | |
2839 | ||
d0450948 | 2840 | if (error) |
43ff2122 | 2841 | goto out; |
d0450948 CH |
2842 | } |
2843 | ||
2844 | xlog_recover_free_trans(trans); | |
43ff2122 CH |
2845 | |
2846 | out: | |
2847 | error2 = xfs_buf_delwri_submit(&buffer_list); | |
2848 | return error ? error : error2; | |
1da177e4 LT |
2849 | } |
2850 | ||
2851 | STATIC int | |
2852 | xlog_recover_unmount_trans( | |
ad223e60 MT |
2853 | struct xlog *log, |
2854 | struct xlog_recover *trans) | |
1da177e4 LT |
2855 | { |
2856 | /* Do nothing now */ | |
a0fa2b67 | 2857 | xfs_warn(log->l_mp, "%s: Unmount LR", __func__); |
1da177e4 LT |
2858 | return 0; |
2859 | } | |
2860 | ||
2861 | /* | |
2862 | * There are two valid states of the r_state field. 0 indicates that the | |
2863 | * transaction structure is in a normal state. We have either seen the | |
2864 | * start of the transaction or the last operation we added was not a partial | |
2865 | * operation. If the last operation we added to the transaction was a | |
2866 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. | |
2867 | * | |
2868 | * NOTE: skip LRs with 0 data length. | |
2869 | */ | |
2870 | STATIC int | |
2871 | xlog_recover_process_data( | |
9a8d2fdb | 2872 | struct xlog *log, |
f0a76953 | 2873 | struct hlist_head rhash[], |
9a8d2fdb | 2874 | struct xlog_rec_header *rhead, |
1da177e4 LT |
2875 | xfs_caddr_t dp, |
2876 | int pass) | |
2877 | { | |
2878 | xfs_caddr_t lp; | |
2879 | int num_logops; | |
2880 | xlog_op_header_t *ohead; | |
2881 | xlog_recover_t *trans; | |
2882 | xlog_tid_t tid; | |
2883 | int error; | |
2884 | unsigned long hash; | |
2885 | uint flags; | |
2886 | ||
b53e675d CH |
2887 | lp = dp + be32_to_cpu(rhead->h_len); |
2888 | num_logops = be32_to_cpu(rhead->h_num_logops); | |
1da177e4 LT |
2889 | |
2890 | /* check the log format matches our own - else we can't recover */ | |
2891 | if (xlog_header_check_recover(log->l_mp, rhead)) | |
2892 | return (XFS_ERROR(EIO)); | |
2893 | ||
2894 | while ((dp < lp) && num_logops) { | |
2895 | ASSERT(dp + sizeof(xlog_op_header_t) <= lp); | |
2896 | ohead = (xlog_op_header_t *)dp; | |
2897 | dp += sizeof(xlog_op_header_t); | |
2898 | if (ohead->oh_clientid != XFS_TRANSACTION && | |
2899 | ohead->oh_clientid != XFS_LOG) { | |
a0fa2b67 DC |
2900 | xfs_warn(log->l_mp, "%s: bad clientid 0x%x", |
2901 | __func__, ohead->oh_clientid); | |
1da177e4 LT |
2902 | ASSERT(0); |
2903 | return (XFS_ERROR(EIO)); | |
2904 | } | |
67fcb7bf | 2905 | tid = be32_to_cpu(ohead->oh_tid); |
1da177e4 | 2906 | hash = XLOG_RHASH(tid); |
f0a76953 | 2907 | trans = xlog_recover_find_tid(&rhash[hash], tid); |
1da177e4 LT |
2908 | if (trans == NULL) { /* not found; add new tid */ |
2909 | if (ohead->oh_flags & XLOG_START_TRANS) | |
2910 | xlog_recover_new_tid(&rhash[hash], tid, | |
b53e675d | 2911 | be64_to_cpu(rhead->h_lsn)); |
1da177e4 | 2912 | } else { |
9742bb93 | 2913 | if (dp + be32_to_cpu(ohead->oh_len) > lp) { |
a0fa2b67 DC |
2914 | xfs_warn(log->l_mp, "%s: bad length 0x%x", |
2915 | __func__, be32_to_cpu(ohead->oh_len)); | |
9742bb93 LM |
2916 | WARN_ON(1); |
2917 | return (XFS_ERROR(EIO)); | |
2918 | } | |
1da177e4 LT |
2919 | flags = ohead->oh_flags & ~XLOG_END_TRANS; |
2920 | if (flags & XLOG_WAS_CONT_TRANS) | |
2921 | flags &= ~XLOG_CONTINUE_TRANS; | |
2922 | switch (flags) { | |
2923 | case XLOG_COMMIT_TRANS: | |
2924 | error = xlog_recover_commit_trans(log, | |
f0a76953 | 2925 | trans, pass); |
1da177e4 LT |
2926 | break; |
2927 | case XLOG_UNMOUNT_TRANS: | |
a0fa2b67 | 2928 | error = xlog_recover_unmount_trans(log, trans); |
1da177e4 LT |
2929 | break; |
2930 | case XLOG_WAS_CONT_TRANS: | |
9abbc539 DC |
2931 | error = xlog_recover_add_to_cont_trans(log, |
2932 | trans, dp, | |
2933 | be32_to_cpu(ohead->oh_len)); | |
1da177e4 LT |
2934 | break; |
2935 | case XLOG_START_TRANS: | |
a0fa2b67 DC |
2936 | xfs_warn(log->l_mp, "%s: bad transaction", |
2937 | __func__); | |
1da177e4 LT |
2938 | ASSERT(0); |
2939 | error = XFS_ERROR(EIO); | |
2940 | break; | |
2941 | case 0: | |
2942 | case XLOG_CONTINUE_TRANS: | |
9abbc539 | 2943 | error = xlog_recover_add_to_trans(log, trans, |
67fcb7bf | 2944 | dp, be32_to_cpu(ohead->oh_len)); |
1da177e4 LT |
2945 | break; |
2946 | default: | |
a0fa2b67 DC |
2947 | xfs_warn(log->l_mp, "%s: bad flag 0x%x", |
2948 | __func__, flags); | |
1da177e4 LT |
2949 | ASSERT(0); |
2950 | error = XFS_ERROR(EIO); | |
2951 | break; | |
2952 | } | |
2953 | if (error) | |
2954 | return error; | |
2955 | } | |
67fcb7bf | 2956 | dp += be32_to_cpu(ohead->oh_len); |
1da177e4 LT |
2957 | num_logops--; |
2958 | } | |
2959 | return 0; | |
2960 | } | |
2961 | ||
2962 | /* | |
2963 | * Process an extent free intent item that was recovered from | |
2964 | * the log. We need to free the extents that it describes. | |
2965 | */ | |
3c1e2bbe | 2966 | STATIC int |
1da177e4 LT |
2967 | xlog_recover_process_efi( |
2968 | xfs_mount_t *mp, | |
2969 | xfs_efi_log_item_t *efip) | |
2970 | { | |
2971 | xfs_efd_log_item_t *efdp; | |
2972 | xfs_trans_t *tp; | |
2973 | int i; | |
3c1e2bbe | 2974 | int error = 0; |
1da177e4 LT |
2975 | xfs_extent_t *extp; |
2976 | xfs_fsblock_t startblock_fsb; | |
2977 | ||
b199c8a4 | 2978 | ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)); |
1da177e4 LT |
2979 | |
2980 | /* | |
2981 | * First check the validity of the extents described by the | |
2982 | * EFI. If any are bad, then assume that all are bad and | |
2983 | * just toss the EFI. | |
2984 | */ | |
2985 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
2986 | extp = &(efip->efi_format.efi_extents[i]); | |
2987 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
2988 | XFS_FSB_TO_DADDR(mp, extp->ext_start)); | |
2989 | if ((startblock_fsb == 0) || | |
2990 | (extp->ext_len == 0) || | |
2991 | (startblock_fsb >= mp->m_sb.sb_dblocks) || | |
2992 | (extp->ext_len >= mp->m_sb.sb_agblocks)) { | |
2993 | /* | |
2994 | * This will pull the EFI from the AIL and | |
2995 | * free the memory associated with it. | |
2996 | */ | |
666d644c | 2997 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
1da177e4 | 2998 | xfs_efi_release(efip, efip->efi_format.efi_nextents); |
3c1e2bbe | 2999 | return XFS_ERROR(EIO); |
1da177e4 LT |
3000 | } |
3001 | } | |
3002 | ||
3003 | tp = xfs_trans_alloc(mp, 0); | |
3c1e2bbe | 3004 | error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0); |
fc6149d8 DC |
3005 | if (error) |
3006 | goto abort_error; | |
1da177e4 LT |
3007 | efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents); |
3008 | ||
3009 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3010 | extp = &(efip->efi_format.efi_extents[i]); | |
fc6149d8 DC |
3011 | error = xfs_free_extent(tp, extp->ext_start, extp->ext_len); |
3012 | if (error) | |
3013 | goto abort_error; | |
1da177e4 LT |
3014 | xfs_trans_log_efd_extent(tp, efdp, extp->ext_start, |
3015 | extp->ext_len); | |
3016 | } | |
3017 | ||
b199c8a4 | 3018 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
e5720eec | 3019 | error = xfs_trans_commit(tp, 0); |
3c1e2bbe | 3020 | return error; |
fc6149d8 DC |
3021 | |
3022 | abort_error: | |
3023 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3024 | return error; | |
1da177e4 LT |
3025 | } |
3026 | ||
1da177e4 LT |
3027 | /* |
3028 | * When this is called, all of the EFIs which did not have | |
3029 | * corresponding EFDs should be in the AIL. What we do now | |
3030 | * is free the extents associated with each one. | |
3031 | * | |
3032 | * Since we process the EFIs in normal transactions, they | |
3033 | * will be removed at some point after the commit. This prevents | |
3034 | * us from just walking down the list processing each one. | |
3035 | * We'll use a flag in the EFI to skip those that we've already | |
3036 | * processed and use the AIL iteration mechanism's generation | |
3037 | * count to try to speed this up at least a bit. | |
3038 | * | |
3039 | * When we start, we know that the EFIs are the only things in | |
3040 | * the AIL. As we process them, however, other items are added | |
3041 | * to the AIL. Since everything added to the AIL must come after | |
3042 | * everything already in the AIL, we stop processing as soon as | |
3043 | * we see something other than an EFI in the AIL. | |
3044 | */ | |
3c1e2bbe | 3045 | STATIC int |
1da177e4 | 3046 | xlog_recover_process_efis( |
9a8d2fdb | 3047 | struct xlog *log) |
1da177e4 LT |
3048 | { |
3049 | xfs_log_item_t *lip; | |
3050 | xfs_efi_log_item_t *efip; | |
3c1e2bbe | 3051 | int error = 0; |
27d8d5fe | 3052 | struct xfs_ail_cursor cur; |
a9c21c1b | 3053 | struct xfs_ail *ailp; |
1da177e4 | 3054 | |
a9c21c1b DC |
3055 | ailp = log->l_ailp; |
3056 | spin_lock(&ailp->xa_lock); | |
3057 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
3058 | while (lip != NULL) { |
3059 | /* | |
3060 | * We're done when we see something other than an EFI. | |
27d8d5fe | 3061 | * There should be no EFIs left in the AIL now. |
1da177e4 LT |
3062 | */ |
3063 | if (lip->li_type != XFS_LI_EFI) { | |
27d8d5fe | 3064 | #ifdef DEBUG |
a9c21c1b | 3065 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
27d8d5fe DC |
3066 | ASSERT(lip->li_type != XFS_LI_EFI); |
3067 | #endif | |
1da177e4 LT |
3068 | break; |
3069 | } | |
3070 | ||
3071 | /* | |
3072 | * Skip EFIs that we've already processed. | |
3073 | */ | |
3074 | efip = (xfs_efi_log_item_t *)lip; | |
b199c8a4 | 3075 | if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) { |
a9c21c1b | 3076 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 LT |
3077 | continue; |
3078 | } | |
3079 | ||
a9c21c1b DC |
3080 | spin_unlock(&ailp->xa_lock); |
3081 | error = xlog_recover_process_efi(log->l_mp, efip); | |
3082 | spin_lock(&ailp->xa_lock); | |
27d8d5fe DC |
3083 | if (error) |
3084 | goto out; | |
a9c21c1b | 3085 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 3086 | } |
27d8d5fe | 3087 | out: |
a9c21c1b DC |
3088 | xfs_trans_ail_cursor_done(ailp, &cur); |
3089 | spin_unlock(&ailp->xa_lock); | |
3c1e2bbe | 3090 | return error; |
1da177e4 LT |
3091 | } |
3092 | ||
3093 | /* | |
3094 | * This routine performs a transaction to null out a bad inode pointer | |
3095 | * in an agi unlinked inode hash bucket. | |
3096 | */ | |
3097 | STATIC void | |
3098 | xlog_recover_clear_agi_bucket( | |
3099 | xfs_mount_t *mp, | |
3100 | xfs_agnumber_t agno, | |
3101 | int bucket) | |
3102 | { | |
3103 | xfs_trans_t *tp; | |
3104 | xfs_agi_t *agi; | |
3105 | xfs_buf_t *agibp; | |
3106 | int offset; | |
3107 | int error; | |
3108 | ||
3109 | tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET); | |
5e1be0fb CH |
3110 | error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), |
3111 | 0, 0, 0); | |
e5720eec DC |
3112 | if (error) |
3113 | goto out_abort; | |
1da177e4 | 3114 | |
5e1be0fb CH |
3115 | error = xfs_read_agi(mp, tp, agno, &agibp); |
3116 | if (error) | |
e5720eec | 3117 | goto out_abort; |
1da177e4 | 3118 | |
5e1be0fb | 3119 | agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 3120 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
3121 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
3122 | (sizeof(xfs_agino_t) * bucket); | |
3123 | xfs_trans_log_buf(tp, agibp, offset, | |
3124 | (offset + sizeof(xfs_agino_t) - 1)); | |
3125 | ||
e5720eec DC |
3126 | error = xfs_trans_commit(tp, 0); |
3127 | if (error) | |
3128 | goto out_error; | |
3129 | return; | |
3130 | ||
3131 | out_abort: | |
3132 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3133 | out_error: | |
a0fa2b67 | 3134 | xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno); |
e5720eec | 3135 | return; |
1da177e4 LT |
3136 | } |
3137 | ||
23fac50f CH |
3138 | STATIC xfs_agino_t |
3139 | xlog_recover_process_one_iunlink( | |
3140 | struct xfs_mount *mp, | |
3141 | xfs_agnumber_t agno, | |
3142 | xfs_agino_t agino, | |
3143 | int bucket) | |
3144 | { | |
3145 | struct xfs_buf *ibp; | |
3146 | struct xfs_dinode *dip; | |
3147 | struct xfs_inode *ip; | |
3148 | xfs_ino_t ino; | |
3149 | int error; | |
3150 | ||
3151 | ino = XFS_AGINO_TO_INO(mp, agno, agino); | |
7b6259e7 | 3152 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip); |
23fac50f CH |
3153 | if (error) |
3154 | goto fail; | |
3155 | ||
3156 | /* | |
3157 | * Get the on disk inode to find the next inode in the bucket. | |
3158 | */ | |
475ee413 | 3159 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0); |
23fac50f | 3160 | if (error) |
0e446673 | 3161 | goto fail_iput; |
23fac50f | 3162 | |
23fac50f | 3163 | ASSERT(ip->i_d.di_nlink == 0); |
0e446673 | 3164 | ASSERT(ip->i_d.di_mode != 0); |
23fac50f CH |
3165 | |
3166 | /* setup for the next pass */ | |
3167 | agino = be32_to_cpu(dip->di_next_unlinked); | |
3168 | xfs_buf_relse(ibp); | |
3169 | ||
3170 | /* | |
3171 | * Prevent any DMAPI event from being sent when the reference on | |
3172 | * the inode is dropped. | |
3173 | */ | |
3174 | ip->i_d.di_dmevmask = 0; | |
3175 | ||
0e446673 | 3176 | IRELE(ip); |
23fac50f CH |
3177 | return agino; |
3178 | ||
0e446673 CH |
3179 | fail_iput: |
3180 | IRELE(ip); | |
23fac50f CH |
3181 | fail: |
3182 | /* | |
3183 | * We can't read in the inode this bucket points to, or this inode | |
3184 | * is messed up. Just ditch this bucket of inodes. We will lose | |
3185 | * some inodes and space, but at least we won't hang. | |
3186 | * | |
3187 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to | |
3188 | * clear the inode pointer in the bucket. | |
3189 | */ | |
3190 | xlog_recover_clear_agi_bucket(mp, agno, bucket); | |
3191 | return NULLAGINO; | |
3192 | } | |
3193 | ||
1da177e4 LT |
3194 | /* |
3195 | * xlog_iunlink_recover | |
3196 | * | |
3197 | * This is called during recovery to process any inodes which | |
3198 | * we unlinked but not freed when the system crashed. These | |
3199 | * inodes will be on the lists in the AGI blocks. What we do | |
3200 | * here is scan all the AGIs and fully truncate and free any | |
3201 | * inodes found on the lists. Each inode is removed from the | |
3202 | * lists when it has been fully truncated and is freed. The | |
3203 | * freeing of the inode and its removal from the list must be | |
3204 | * atomic. | |
3205 | */ | |
d96f8f89 | 3206 | STATIC void |
1da177e4 | 3207 | xlog_recover_process_iunlinks( |
9a8d2fdb | 3208 | struct xlog *log) |
1da177e4 LT |
3209 | { |
3210 | xfs_mount_t *mp; | |
3211 | xfs_agnumber_t agno; | |
3212 | xfs_agi_t *agi; | |
3213 | xfs_buf_t *agibp; | |
1da177e4 | 3214 | xfs_agino_t agino; |
1da177e4 LT |
3215 | int bucket; |
3216 | int error; | |
3217 | uint mp_dmevmask; | |
3218 | ||
3219 | mp = log->l_mp; | |
3220 | ||
3221 | /* | |
3222 | * Prevent any DMAPI event from being sent while in this function. | |
3223 | */ | |
3224 | mp_dmevmask = mp->m_dmevmask; | |
3225 | mp->m_dmevmask = 0; | |
3226 | ||
3227 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
3228 | /* | |
3229 | * Find the agi for this ag. | |
3230 | */ | |
5e1be0fb CH |
3231 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
3232 | if (error) { | |
3233 | /* | |
3234 | * AGI is b0rked. Don't process it. | |
3235 | * | |
3236 | * We should probably mark the filesystem as corrupt | |
3237 | * after we've recovered all the ag's we can.... | |
3238 | */ | |
3239 | continue; | |
1da177e4 | 3240 | } |
d97d32ed JK |
3241 | /* |
3242 | * Unlock the buffer so that it can be acquired in the normal | |
3243 | * course of the transaction to truncate and free each inode. | |
3244 | * Because we are not racing with anyone else here for the AGI | |
3245 | * buffer, we don't even need to hold it locked to read the | |
3246 | * initial unlinked bucket entries out of the buffer. We keep | |
3247 | * buffer reference though, so that it stays pinned in memory | |
3248 | * while we need the buffer. | |
3249 | */ | |
1da177e4 | 3250 | agi = XFS_BUF_TO_AGI(agibp); |
d97d32ed | 3251 | xfs_buf_unlock(agibp); |
1da177e4 LT |
3252 | |
3253 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { | |
16259e7d | 3254 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
1da177e4 | 3255 | while (agino != NULLAGINO) { |
23fac50f CH |
3256 | agino = xlog_recover_process_one_iunlink(mp, |
3257 | agno, agino, bucket); | |
1da177e4 LT |
3258 | } |
3259 | } | |
d97d32ed | 3260 | xfs_buf_rele(agibp); |
1da177e4 LT |
3261 | } |
3262 | ||
3263 | mp->m_dmevmask = mp_dmevmask; | |
3264 | } | |
3265 | ||
1da177e4 | 3266 | /* |
0e446be4 CH |
3267 | * Upack the log buffer data and crc check it. If the check fails, issue a |
3268 | * warning if and only if the CRC in the header is non-zero. This makes the | |
3269 | * check an advisory warning, and the zero CRC check will prevent failure | |
3270 | * warnings from being emitted when upgrading the kernel from one that does not | |
3271 | * add CRCs by default. | |
3272 | * | |
3273 | * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log | |
3274 | * corruption failure | |
1da177e4 | 3275 | */ |
0e446be4 CH |
3276 | STATIC int |
3277 | xlog_unpack_data_crc( | |
3278 | struct xlog_rec_header *rhead, | |
3279 | xfs_caddr_t dp, | |
3280 | struct xlog *log) | |
1da177e4 | 3281 | { |
f9668a09 | 3282 | __le32 crc; |
0e446be4 CH |
3283 | |
3284 | crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len)); | |
3285 | if (crc != rhead->h_crc) { | |
3286 | if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) { | |
3287 | xfs_alert(log->l_mp, | |
3288 | "log record CRC mismatch: found 0x%x, expected 0x%x.\n", | |
f9668a09 DC |
3289 | le32_to_cpu(rhead->h_crc), |
3290 | le32_to_cpu(crc)); | |
0e446be4 | 3291 | xfs_hex_dump(dp, 32); |
1da177e4 LT |
3292 | } |
3293 | ||
0e446be4 CH |
3294 | /* |
3295 | * If we've detected a log record corruption, then we can't | |
3296 | * recover past this point. Abort recovery if we are enforcing | |
3297 | * CRC protection by punting an error back up the stack. | |
3298 | */ | |
3299 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) | |
3300 | return EFSCORRUPTED; | |
1da177e4 | 3301 | } |
0e446be4 CH |
3302 | |
3303 | return 0; | |
1da177e4 LT |
3304 | } |
3305 | ||
0e446be4 | 3306 | STATIC int |
1da177e4 | 3307 | xlog_unpack_data( |
9a8d2fdb | 3308 | struct xlog_rec_header *rhead, |
1da177e4 | 3309 | xfs_caddr_t dp, |
9a8d2fdb | 3310 | struct xlog *log) |
1da177e4 LT |
3311 | { |
3312 | int i, j, k; | |
0e446be4 CH |
3313 | int error; |
3314 | ||
3315 | error = xlog_unpack_data_crc(rhead, dp, log); | |
3316 | if (error) | |
3317 | return error; | |
1da177e4 | 3318 | |
b53e675d | 3319 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
1da177e4 | 3320 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
b53e675d | 3321 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
1da177e4 LT |
3322 | dp += BBSIZE; |
3323 | } | |
3324 | ||
62118709 | 3325 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b28708d6 | 3326 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
b53e675d | 3327 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
1da177e4 LT |
3328 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
3329 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3330 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
1da177e4 LT |
3331 | dp += BBSIZE; |
3332 | } | |
3333 | } | |
0e446be4 CH |
3334 | |
3335 | return 0; | |
1da177e4 LT |
3336 | } |
3337 | ||
3338 | STATIC int | |
3339 | xlog_valid_rec_header( | |
9a8d2fdb MT |
3340 | struct xlog *log, |
3341 | struct xlog_rec_header *rhead, | |
1da177e4 LT |
3342 | xfs_daddr_t blkno) |
3343 | { | |
3344 | int hlen; | |
3345 | ||
69ef921b | 3346 | if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) { |
1da177e4 LT |
3347 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
3348 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3349 | return XFS_ERROR(EFSCORRUPTED); | |
3350 | } | |
3351 | if (unlikely( | |
3352 | (!rhead->h_version || | |
b53e675d | 3353 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
a0fa2b67 | 3354 | xfs_warn(log->l_mp, "%s: unrecognised log version (%d).", |
34a622b2 | 3355 | __func__, be32_to_cpu(rhead->h_version)); |
1da177e4 LT |
3356 | return XFS_ERROR(EIO); |
3357 | } | |
3358 | ||
3359 | /* LR body must have data or it wouldn't have been written */ | |
b53e675d | 3360 | hlen = be32_to_cpu(rhead->h_len); |
1da177e4 LT |
3361 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
3362 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", | |
3363 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3364 | return XFS_ERROR(EFSCORRUPTED); | |
3365 | } | |
3366 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { | |
3367 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", | |
3368 | XFS_ERRLEVEL_LOW, log->l_mp); | |
3369 | return XFS_ERROR(EFSCORRUPTED); | |
3370 | } | |
3371 | return 0; | |
3372 | } | |
3373 | ||
3374 | /* | |
3375 | * Read the log from tail to head and process the log records found. | |
3376 | * Handle the two cases where the tail and head are in the same cycle | |
3377 | * and where the active portion of the log wraps around the end of | |
3378 | * the physical log separately. The pass parameter is passed through | |
3379 | * to the routines called to process the data and is not looked at | |
3380 | * here. | |
3381 | */ | |
3382 | STATIC int | |
3383 | xlog_do_recovery_pass( | |
9a8d2fdb | 3384 | struct xlog *log, |
1da177e4 LT |
3385 | xfs_daddr_t head_blk, |
3386 | xfs_daddr_t tail_blk, | |
3387 | int pass) | |
3388 | { | |
3389 | xlog_rec_header_t *rhead; | |
3390 | xfs_daddr_t blk_no; | |
fc5bc4c8 | 3391 | xfs_caddr_t offset; |
1da177e4 LT |
3392 | xfs_buf_t *hbp, *dbp; |
3393 | int error = 0, h_size; | |
3394 | int bblks, split_bblks; | |
3395 | int hblks, split_hblks, wrapped_hblks; | |
f0a76953 | 3396 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
1da177e4 LT |
3397 | |
3398 | ASSERT(head_blk != tail_blk); | |
3399 | ||
3400 | /* | |
3401 | * Read the header of the tail block and get the iclog buffer size from | |
3402 | * h_size. Use this to tell how many sectors make up the log header. | |
3403 | */ | |
62118709 | 3404 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
1da177e4 LT |
3405 | /* |
3406 | * When using variable length iclogs, read first sector of | |
3407 | * iclog header and extract the header size from it. Get a | |
3408 | * new hbp that is the correct size. | |
3409 | */ | |
3410 | hbp = xlog_get_bp(log, 1); | |
3411 | if (!hbp) | |
3412 | return ENOMEM; | |
076e6acb CH |
3413 | |
3414 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); | |
3415 | if (error) | |
1da177e4 | 3416 | goto bread_err1; |
076e6acb | 3417 | |
1da177e4 LT |
3418 | rhead = (xlog_rec_header_t *)offset; |
3419 | error = xlog_valid_rec_header(log, rhead, tail_blk); | |
3420 | if (error) | |
3421 | goto bread_err1; | |
b53e675d CH |
3422 | h_size = be32_to_cpu(rhead->h_size); |
3423 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && | |
1da177e4 LT |
3424 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
3425 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
3426 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
3427 | hblks++; | |
3428 | xlog_put_bp(hbp); | |
3429 | hbp = xlog_get_bp(log, hblks); | |
3430 | } else { | |
3431 | hblks = 1; | |
3432 | } | |
3433 | } else { | |
69ce58f0 | 3434 | ASSERT(log->l_sectBBsize == 1); |
1da177e4 LT |
3435 | hblks = 1; |
3436 | hbp = xlog_get_bp(log, 1); | |
3437 | h_size = XLOG_BIG_RECORD_BSIZE; | |
3438 | } | |
3439 | ||
3440 | if (!hbp) | |
3441 | return ENOMEM; | |
3442 | dbp = xlog_get_bp(log, BTOBB(h_size)); | |
3443 | if (!dbp) { | |
3444 | xlog_put_bp(hbp); | |
3445 | return ENOMEM; | |
3446 | } | |
3447 | ||
3448 | memset(rhash, 0, sizeof(rhash)); | |
3449 | if (tail_blk <= head_blk) { | |
3450 | for (blk_no = tail_blk; blk_no < head_blk; ) { | |
076e6acb CH |
3451 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
3452 | if (error) | |
1da177e4 | 3453 | goto bread_err2; |
076e6acb | 3454 | |
1da177e4 LT |
3455 | rhead = (xlog_rec_header_t *)offset; |
3456 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3457 | if (error) | |
3458 | goto bread_err2; | |
3459 | ||
3460 | /* blocks in data section */ | |
b53e675d | 3461 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
3462 | error = xlog_bread(log, blk_no + hblks, bblks, dbp, |
3463 | &offset); | |
1da177e4 LT |
3464 | if (error) |
3465 | goto bread_err2; | |
076e6acb | 3466 | |
0e446be4 CH |
3467 | error = xlog_unpack_data(rhead, offset, log); |
3468 | if (error) | |
3469 | goto bread_err2; | |
3470 | ||
3471 | error = xlog_recover_process_data(log, | |
3472 | rhash, rhead, offset, pass); | |
3473 | if (error) | |
1da177e4 LT |
3474 | goto bread_err2; |
3475 | blk_no += bblks + hblks; | |
3476 | } | |
3477 | } else { | |
3478 | /* | |
3479 | * Perform recovery around the end of the physical log. | |
3480 | * When the head is not on the same cycle number as the tail, | |
3481 | * we can't do a sequential recovery as above. | |
3482 | */ | |
3483 | blk_no = tail_blk; | |
3484 | while (blk_no < log->l_logBBsize) { | |
3485 | /* | |
3486 | * Check for header wrapping around physical end-of-log | |
3487 | */ | |
62926044 | 3488 | offset = hbp->b_addr; |
1da177e4 LT |
3489 | split_hblks = 0; |
3490 | wrapped_hblks = 0; | |
3491 | if (blk_no + hblks <= log->l_logBBsize) { | |
3492 | /* Read header in one read */ | |
076e6acb CH |
3493 | error = xlog_bread(log, blk_no, hblks, hbp, |
3494 | &offset); | |
1da177e4 LT |
3495 | if (error) |
3496 | goto bread_err2; | |
1da177e4 LT |
3497 | } else { |
3498 | /* This LR is split across physical log end */ | |
3499 | if (blk_no != log->l_logBBsize) { | |
3500 | /* some data before physical log end */ | |
3501 | ASSERT(blk_no <= INT_MAX); | |
3502 | split_hblks = log->l_logBBsize - (int)blk_no; | |
3503 | ASSERT(split_hblks > 0); | |
076e6acb CH |
3504 | error = xlog_bread(log, blk_no, |
3505 | split_hblks, hbp, | |
3506 | &offset); | |
3507 | if (error) | |
1da177e4 | 3508 | goto bread_err2; |
1da177e4 | 3509 | } |
076e6acb | 3510 | |
1da177e4 LT |
3511 | /* |
3512 | * Note: this black magic still works with | |
3513 | * large sector sizes (non-512) only because: | |
3514 | * - we increased the buffer size originally | |
3515 | * by 1 sector giving us enough extra space | |
3516 | * for the second read; | |
3517 | * - the log start is guaranteed to be sector | |
3518 | * aligned; | |
3519 | * - we read the log end (LR header start) | |
3520 | * _first_, then the log start (LR header end) | |
3521 | * - order is important. | |
3522 | */ | |
234f56ac | 3523 | wrapped_hblks = hblks - split_hblks; |
44396476 DC |
3524 | error = xlog_bread_offset(log, 0, |
3525 | wrapped_hblks, hbp, | |
3526 | offset + BBTOB(split_hblks)); | |
1da177e4 LT |
3527 | if (error) |
3528 | goto bread_err2; | |
1da177e4 LT |
3529 | } |
3530 | rhead = (xlog_rec_header_t *)offset; | |
3531 | error = xlog_valid_rec_header(log, rhead, | |
3532 | split_hblks ? blk_no : 0); | |
3533 | if (error) | |
3534 | goto bread_err2; | |
3535 | ||
b53e675d | 3536 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
1da177e4 LT |
3537 | blk_no += hblks; |
3538 | ||
3539 | /* Read in data for log record */ | |
3540 | if (blk_no + bblks <= log->l_logBBsize) { | |
076e6acb CH |
3541 | error = xlog_bread(log, blk_no, bblks, dbp, |
3542 | &offset); | |
1da177e4 LT |
3543 | if (error) |
3544 | goto bread_err2; | |
1da177e4 LT |
3545 | } else { |
3546 | /* This log record is split across the | |
3547 | * physical end of log */ | |
62926044 | 3548 | offset = dbp->b_addr; |
1da177e4 LT |
3549 | split_bblks = 0; |
3550 | if (blk_no != log->l_logBBsize) { | |
3551 | /* some data is before the physical | |
3552 | * end of log */ | |
3553 | ASSERT(!wrapped_hblks); | |
3554 | ASSERT(blk_no <= INT_MAX); | |
3555 | split_bblks = | |
3556 | log->l_logBBsize - (int)blk_no; | |
3557 | ASSERT(split_bblks > 0); | |
076e6acb CH |
3558 | error = xlog_bread(log, blk_no, |
3559 | split_bblks, dbp, | |
3560 | &offset); | |
3561 | if (error) | |
1da177e4 | 3562 | goto bread_err2; |
1da177e4 | 3563 | } |
076e6acb | 3564 | |
1da177e4 LT |
3565 | /* |
3566 | * Note: this black magic still works with | |
3567 | * large sector sizes (non-512) only because: | |
3568 | * - we increased the buffer size originally | |
3569 | * by 1 sector giving us enough extra space | |
3570 | * for the second read; | |
3571 | * - the log start is guaranteed to be sector | |
3572 | * aligned; | |
3573 | * - we read the log end (LR header start) | |
3574 | * _first_, then the log start (LR header end) | |
3575 | * - order is important. | |
3576 | */ | |
44396476 | 3577 | error = xlog_bread_offset(log, 0, |
009507b0 | 3578 | bblks - split_bblks, dbp, |
44396476 | 3579 | offset + BBTOB(split_bblks)); |
076e6acb CH |
3580 | if (error) |
3581 | goto bread_err2; | |
1da177e4 | 3582 | } |
0e446be4 CH |
3583 | |
3584 | error = xlog_unpack_data(rhead, offset, log); | |
3585 | if (error) | |
3586 | goto bread_err2; | |
3587 | ||
3588 | error = xlog_recover_process_data(log, rhash, | |
3589 | rhead, offset, pass); | |
3590 | if (error) | |
1da177e4 LT |
3591 | goto bread_err2; |
3592 | blk_no += bblks; | |
3593 | } | |
3594 | ||
3595 | ASSERT(blk_no >= log->l_logBBsize); | |
3596 | blk_no -= log->l_logBBsize; | |
3597 | ||
3598 | /* read first part of physical log */ | |
3599 | while (blk_no < head_blk) { | |
076e6acb CH |
3600 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
3601 | if (error) | |
1da177e4 | 3602 | goto bread_err2; |
076e6acb | 3603 | |
1da177e4 LT |
3604 | rhead = (xlog_rec_header_t *)offset; |
3605 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
3606 | if (error) | |
3607 | goto bread_err2; | |
076e6acb | 3608 | |
b53e675d | 3609 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
3610 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, |
3611 | &offset); | |
3612 | if (error) | |
1da177e4 | 3613 | goto bread_err2; |
076e6acb | 3614 | |
0e446be4 CH |
3615 | error = xlog_unpack_data(rhead, offset, log); |
3616 | if (error) | |
3617 | goto bread_err2; | |
3618 | ||
3619 | error = xlog_recover_process_data(log, rhash, | |
3620 | rhead, offset, pass); | |
3621 | if (error) | |
1da177e4 LT |
3622 | goto bread_err2; |
3623 | blk_no += bblks + hblks; | |
3624 | } | |
3625 | } | |
3626 | ||
3627 | bread_err2: | |
3628 | xlog_put_bp(dbp); | |
3629 | bread_err1: | |
3630 | xlog_put_bp(hbp); | |
3631 | return error; | |
3632 | } | |
3633 | ||
3634 | /* | |
3635 | * Do the recovery of the log. We actually do this in two phases. | |
3636 | * The two passes are necessary in order to implement the function | |
3637 | * of cancelling a record written into the log. The first pass | |
3638 | * determines those things which have been cancelled, and the | |
3639 | * second pass replays log items normally except for those which | |
3640 | * have been cancelled. The handling of the replay and cancellations | |
3641 | * takes place in the log item type specific routines. | |
3642 | * | |
3643 | * The table of items which have cancel records in the log is allocated | |
3644 | * and freed at this level, since only here do we know when all of | |
3645 | * the log recovery has been completed. | |
3646 | */ | |
3647 | STATIC int | |
3648 | xlog_do_log_recovery( | |
9a8d2fdb | 3649 | struct xlog *log, |
1da177e4 LT |
3650 | xfs_daddr_t head_blk, |
3651 | xfs_daddr_t tail_blk) | |
3652 | { | |
d5689eaa | 3653 | int error, i; |
1da177e4 LT |
3654 | |
3655 | ASSERT(head_blk != tail_blk); | |
3656 | ||
3657 | /* | |
3658 | * First do a pass to find all of the cancelled buf log items. | |
3659 | * Store them in the buf_cancel_table for use in the second pass. | |
3660 | */ | |
d5689eaa CH |
3661 | log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE * |
3662 | sizeof(struct list_head), | |
1da177e4 | 3663 | KM_SLEEP); |
d5689eaa CH |
3664 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
3665 | INIT_LIST_HEAD(&log->l_buf_cancel_table[i]); | |
3666 | ||
1da177e4 LT |
3667 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
3668 | XLOG_RECOVER_PASS1); | |
3669 | if (error != 0) { | |
f0e2d93c | 3670 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
3671 | log->l_buf_cancel_table = NULL; |
3672 | return error; | |
3673 | } | |
3674 | /* | |
3675 | * Then do a second pass to actually recover the items in the log. | |
3676 | * When it is complete free the table of buf cancel items. | |
3677 | */ | |
3678 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
3679 | XLOG_RECOVER_PASS2); | |
3680 | #ifdef DEBUG | |
6d192a9b | 3681 | if (!error) { |
1da177e4 LT |
3682 | int i; |
3683 | ||
3684 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
d5689eaa | 3685 | ASSERT(list_empty(&log->l_buf_cancel_table[i])); |
1da177e4 LT |
3686 | } |
3687 | #endif /* DEBUG */ | |
3688 | ||
f0e2d93c | 3689 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
3690 | log->l_buf_cancel_table = NULL; |
3691 | ||
3692 | return error; | |
3693 | } | |
3694 | ||
3695 | /* | |
3696 | * Do the actual recovery | |
3697 | */ | |
3698 | STATIC int | |
3699 | xlog_do_recover( | |
9a8d2fdb | 3700 | struct xlog *log, |
1da177e4 LT |
3701 | xfs_daddr_t head_blk, |
3702 | xfs_daddr_t tail_blk) | |
3703 | { | |
3704 | int error; | |
3705 | xfs_buf_t *bp; | |
3706 | xfs_sb_t *sbp; | |
3707 | ||
3708 | /* | |
3709 | * First replay the images in the log. | |
3710 | */ | |
3711 | error = xlog_do_log_recovery(log, head_blk, tail_blk); | |
43ff2122 | 3712 | if (error) |
1da177e4 | 3713 | return error; |
1da177e4 LT |
3714 | |
3715 | /* | |
3716 | * If IO errors happened during recovery, bail out. | |
3717 | */ | |
3718 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) { | |
3719 | return (EIO); | |
3720 | } | |
3721 | ||
3722 | /* | |
3723 | * We now update the tail_lsn since much of the recovery has completed | |
3724 | * and there may be space available to use. If there were no extent | |
3725 | * or iunlinks, we can free up the entire log and set the tail_lsn to | |
3726 | * be the last_sync_lsn. This was set in xlog_find_tail to be the | |
3727 | * lsn of the last known good LR on disk. If there are extent frees | |
3728 | * or iunlinks they will have some entries in the AIL; so we look at | |
3729 | * the AIL to determine how to set the tail_lsn. | |
3730 | */ | |
3731 | xlog_assign_tail_lsn(log->l_mp); | |
3732 | ||
3733 | /* | |
3734 | * Now that we've finished replaying all buffer and inode | |
98021821 | 3735 | * updates, re-read in the superblock and reverify it. |
1da177e4 LT |
3736 | */ |
3737 | bp = xfs_getsb(log->l_mp, 0); | |
3738 | XFS_BUF_UNDONE(bp); | |
bebf963f | 3739 | ASSERT(!(XFS_BUF_ISWRITE(bp))); |
1da177e4 | 3740 | XFS_BUF_READ(bp); |
bebf963f | 3741 | XFS_BUF_UNASYNC(bp); |
1813dd64 | 3742 | bp->b_ops = &xfs_sb_buf_ops; |
1da177e4 | 3743 | xfsbdstrat(log->l_mp, bp); |
1a1a3e97 | 3744 | error = xfs_buf_iowait(bp); |
d64e31a2 | 3745 | if (error) { |
901796af | 3746 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
3747 | ASSERT(0); |
3748 | xfs_buf_relse(bp); | |
3749 | return error; | |
3750 | } | |
3751 | ||
3752 | /* Convert superblock from on-disk format */ | |
3753 | sbp = &log->l_mp->m_sb; | |
98021821 | 3754 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 3755 | ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC); |
62118709 | 3756 | ASSERT(xfs_sb_good_version(sbp)); |
1da177e4 LT |
3757 | xfs_buf_relse(bp); |
3758 | ||
5478eead LM |
3759 | /* We've re-read the superblock so re-initialize per-cpu counters */ |
3760 | xfs_icsb_reinit_counters(log->l_mp); | |
3761 | ||
1da177e4 LT |
3762 | xlog_recover_check_summary(log); |
3763 | ||
3764 | /* Normal transactions can now occur */ | |
3765 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
3766 | return 0; | |
3767 | } | |
3768 | ||
3769 | /* | |
3770 | * Perform recovery and re-initialize some log variables in xlog_find_tail. | |
3771 | * | |
3772 | * Return error or zero. | |
3773 | */ | |
3774 | int | |
3775 | xlog_recover( | |
9a8d2fdb | 3776 | struct xlog *log) |
1da177e4 LT |
3777 | { |
3778 | xfs_daddr_t head_blk, tail_blk; | |
3779 | int error; | |
3780 | ||
3781 | /* find the tail of the log */ | |
65be6054 | 3782 | if ((error = xlog_find_tail(log, &head_blk, &tail_blk))) |
1da177e4 LT |
3783 | return error; |
3784 | ||
3785 | if (tail_blk != head_blk) { | |
3786 | /* There used to be a comment here: | |
3787 | * | |
3788 | * disallow recovery on read-only mounts. note -- mount | |
3789 | * checks for ENOSPC and turns it into an intelligent | |
3790 | * error message. | |
3791 | * ...but this is no longer true. Now, unless you specify | |
3792 | * NORECOVERY (in which case this function would never be | |
3793 | * called), we just go ahead and recover. We do this all | |
3794 | * under the vfs layer, so we can get away with it unless | |
3795 | * the device itself is read-only, in which case we fail. | |
3796 | */ | |
3a02ee18 | 3797 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
1da177e4 LT |
3798 | return error; |
3799 | } | |
3800 | ||
a0fa2b67 DC |
3801 | xfs_notice(log->l_mp, "Starting recovery (logdev: %s)", |
3802 | log->l_mp->m_logname ? log->l_mp->m_logname | |
3803 | : "internal"); | |
1da177e4 LT |
3804 | |
3805 | error = xlog_do_recover(log, head_blk, tail_blk); | |
3806 | log->l_flags |= XLOG_RECOVERY_NEEDED; | |
3807 | } | |
3808 | return error; | |
3809 | } | |
3810 | ||
3811 | /* | |
3812 | * In the first part of recovery we replay inodes and buffers and build | |
3813 | * up the list of extent free items which need to be processed. Here | |
3814 | * we process the extent free items and clean up the on disk unlinked | |
3815 | * inode lists. This is separated from the first part of recovery so | |
3816 | * that the root and real-time bitmap inodes can be read in from disk in | |
3817 | * between the two stages. This is necessary so that we can free space | |
3818 | * in the real-time portion of the file system. | |
3819 | */ | |
3820 | int | |
3821 | xlog_recover_finish( | |
9a8d2fdb | 3822 | struct xlog *log) |
1da177e4 LT |
3823 | { |
3824 | /* | |
3825 | * Now we're ready to do the transactions needed for the | |
3826 | * rest of recovery. Start with completing all the extent | |
3827 | * free intent records and then process the unlinked inode | |
3828 | * lists. At this point, we essentially run in normal mode | |
3829 | * except that we're still performing recovery actions | |
3830 | * rather than accepting new requests. | |
3831 | */ | |
3832 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { | |
3c1e2bbe DC |
3833 | int error; |
3834 | error = xlog_recover_process_efis(log); | |
3835 | if (error) { | |
a0fa2b67 | 3836 | xfs_alert(log->l_mp, "Failed to recover EFIs"); |
3c1e2bbe DC |
3837 | return error; |
3838 | } | |
1da177e4 LT |
3839 | /* |
3840 | * Sync the log to get all the EFIs out of the AIL. | |
3841 | * This isn't absolutely necessary, but it helps in | |
3842 | * case the unlink transactions would have problems | |
3843 | * pushing the EFIs out of the way. | |
3844 | */ | |
a14a348b | 3845 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
1da177e4 | 3846 | |
4249023a | 3847 | xlog_recover_process_iunlinks(log); |
1da177e4 LT |
3848 | |
3849 | xlog_recover_check_summary(log); | |
3850 | ||
a0fa2b67 DC |
3851 | xfs_notice(log->l_mp, "Ending recovery (logdev: %s)", |
3852 | log->l_mp->m_logname ? log->l_mp->m_logname | |
3853 | : "internal"); | |
1da177e4 LT |
3854 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
3855 | } else { | |
a0fa2b67 | 3856 | xfs_info(log->l_mp, "Ending clean mount"); |
1da177e4 LT |
3857 | } |
3858 | return 0; | |
3859 | } | |
3860 | ||
3861 | ||
3862 | #if defined(DEBUG) | |
3863 | /* | |
3864 | * Read all of the agf and agi counters and check that they | |
3865 | * are consistent with the superblock counters. | |
3866 | */ | |
3867 | void | |
3868 | xlog_recover_check_summary( | |
9a8d2fdb | 3869 | struct xlog *log) |
1da177e4 LT |
3870 | { |
3871 | xfs_mount_t *mp; | |
3872 | xfs_agf_t *agfp; | |
1da177e4 LT |
3873 | xfs_buf_t *agfbp; |
3874 | xfs_buf_t *agibp; | |
1da177e4 LT |
3875 | xfs_agnumber_t agno; |
3876 | __uint64_t freeblks; | |
3877 | __uint64_t itotal; | |
3878 | __uint64_t ifree; | |
5e1be0fb | 3879 | int error; |
1da177e4 LT |
3880 | |
3881 | mp = log->l_mp; | |
3882 | ||
3883 | freeblks = 0LL; | |
3884 | itotal = 0LL; | |
3885 | ifree = 0LL; | |
3886 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
4805621a CH |
3887 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
3888 | if (error) { | |
a0fa2b67 DC |
3889 | xfs_alert(mp, "%s agf read failed agno %d error %d", |
3890 | __func__, agno, error); | |
4805621a CH |
3891 | } else { |
3892 | agfp = XFS_BUF_TO_AGF(agfbp); | |
3893 | freeblks += be32_to_cpu(agfp->agf_freeblks) + | |
3894 | be32_to_cpu(agfp->agf_flcount); | |
3895 | xfs_buf_relse(agfbp); | |
1da177e4 | 3896 | } |
1da177e4 | 3897 | |
5e1be0fb | 3898 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
a0fa2b67 DC |
3899 | if (error) { |
3900 | xfs_alert(mp, "%s agi read failed agno %d error %d", | |
3901 | __func__, agno, error); | |
3902 | } else { | |
5e1be0fb | 3903 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 3904 | |
5e1be0fb CH |
3905 | itotal += be32_to_cpu(agi->agi_count); |
3906 | ifree += be32_to_cpu(agi->agi_freecount); | |
3907 | xfs_buf_relse(agibp); | |
3908 | } | |
1da177e4 | 3909 | } |
1da177e4 LT |
3910 | } |
3911 | #endif /* DEBUG */ |