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" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
a844f451 | 25 | #include "xfs_inum.h" |
a844f451 NS |
26 | #include "xfs_sb.h" |
27 | #include "xfs_ag.h" | |
1da177e4 | 28 | #include "xfs_mount.h" |
57062787 | 29 | #include "xfs_da_format.h" |
1da177e4 | 30 | #include "xfs_inode.h" |
239880ef | 31 | #include "xfs_trans.h" |
239880ef | 32 | #include "xfs_log.h" |
1da177e4 | 33 | #include "xfs_log_priv.h" |
1da177e4 | 34 | #include "xfs_log_recover.h" |
a4fbe6ab | 35 | #include "xfs_inode_item.h" |
1da177e4 LT |
36 | #include "xfs_extfree_item.h" |
37 | #include "xfs_trans_priv.h" | |
a4fbe6ab DC |
38 | #include "xfs_alloc.h" |
39 | #include "xfs_ialloc.h" | |
1da177e4 | 40 | #include "xfs_quota.h" |
0e446be4 | 41 | #include "xfs_cksum.h" |
0b1b213f | 42 | #include "xfs_trace.h" |
33479e05 | 43 | #include "xfs_icache.h" |
a4fbe6ab DC |
44 | #include "xfs_bmap_btree.h" |
45 | #include "xfs_dinode.h" | |
46 | #include "xfs_error.h" | |
2b9ab5ab | 47 | #include "xfs_dir2.h" |
1da177e4 | 48 | |
fc06c6d0 DC |
49 | #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1) |
50 | ||
9a8d2fdb MT |
51 | STATIC int |
52 | xlog_find_zeroed( | |
53 | struct xlog *, | |
54 | xfs_daddr_t *); | |
55 | STATIC int | |
56 | xlog_clear_stale_blocks( | |
57 | struct xlog *, | |
58 | xfs_lsn_t); | |
1da177e4 | 59 | #if defined(DEBUG) |
9a8d2fdb MT |
60 | STATIC void |
61 | xlog_recover_check_summary( | |
62 | struct xlog *); | |
1da177e4 LT |
63 | #else |
64 | #define xlog_recover_check_summary(log) | |
1da177e4 LT |
65 | #endif |
66 | ||
d5689eaa CH |
67 | /* |
68 | * This structure is used during recovery to record the buf log items which | |
69 | * have been canceled and should not be replayed. | |
70 | */ | |
71 | struct xfs_buf_cancel { | |
72 | xfs_daddr_t bc_blkno; | |
73 | uint bc_len; | |
74 | int bc_refcount; | |
75 | struct list_head bc_list; | |
76 | }; | |
77 | ||
1da177e4 LT |
78 | /* |
79 | * Sector aligned buffer routines for buffer create/read/write/access | |
80 | */ | |
81 | ||
ff30a622 AE |
82 | /* |
83 | * Verify the given count of basic blocks is valid number of blocks | |
84 | * to specify for an operation involving the given XFS log buffer. | |
85 | * Returns nonzero if the count is valid, 0 otherwise. | |
86 | */ | |
87 | ||
88 | static inline int | |
89 | xlog_buf_bbcount_valid( | |
9a8d2fdb | 90 | struct xlog *log, |
ff30a622 AE |
91 | int bbcount) |
92 | { | |
93 | return bbcount > 0 && bbcount <= log->l_logBBsize; | |
94 | } | |
95 | ||
36adecff AE |
96 | /* |
97 | * Allocate a buffer to hold log data. The buffer needs to be able | |
98 | * to map to a range of nbblks basic blocks at any valid (basic | |
99 | * block) offset within the log. | |
100 | */ | |
5d77c0dc | 101 | STATIC xfs_buf_t * |
1da177e4 | 102 | xlog_get_bp( |
9a8d2fdb | 103 | struct xlog *log, |
3228149c | 104 | int nbblks) |
1da177e4 | 105 | { |
c8da0faf CH |
106 | struct xfs_buf *bp; |
107 | ||
ff30a622 | 108 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 109 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
110 | nbblks); |
111 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
112 | return NULL; |
113 | } | |
1da177e4 | 114 | |
36adecff AE |
115 | /* |
116 | * We do log I/O in units of log sectors (a power-of-2 | |
117 | * multiple of the basic block size), so we round up the | |
25985edc | 118 | * requested size to accommodate the basic blocks required |
36adecff AE |
119 | * for complete log sectors. |
120 | * | |
121 | * In addition, the buffer may be used for a non-sector- | |
122 | * aligned block offset, in which case an I/O of the | |
123 | * requested size could extend beyond the end of the | |
124 | * buffer. If the requested size is only 1 basic block it | |
125 | * will never straddle a sector boundary, so this won't be | |
126 | * an issue. Nor will this be a problem if the log I/O is | |
127 | * done in basic blocks (sector size 1). But otherwise we | |
128 | * extend the buffer by one extra log sector to ensure | |
25985edc | 129 | * there's space to accommodate this possibility. |
36adecff | 130 | */ |
69ce58f0 AE |
131 | if (nbblks > 1 && log->l_sectBBsize > 1) |
132 | nbblks += log->l_sectBBsize; | |
133 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
36adecff | 134 | |
e70b73f8 | 135 | bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0); |
c8da0faf CH |
136 | if (bp) |
137 | xfs_buf_unlock(bp); | |
138 | return bp; | |
1da177e4 LT |
139 | } |
140 | ||
5d77c0dc | 141 | STATIC void |
1da177e4 LT |
142 | xlog_put_bp( |
143 | xfs_buf_t *bp) | |
144 | { | |
145 | xfs_buf_free(bp); | |
146 | } | |
147 | ||
48389ef1 AE |
148 | /* |
149 | * Return the address of the start of the given block number's data | |
150 | * in a log buffer. The buffer covers a log sector-aligned region. | |
151 | */ | |
076e6acb CH |
152 | STATIC xfs_caddr_t |
153 | xlog_align( | |
9a8d2fdb | 154 | struct xlog *log, |
076e6acb CH |
155 | xfs_daddr_t blk_no, |
156 | int nbblks, | |
9a8d2fdb | 157 | struct xfs_buf *bp) |
076e6acb | 158 | { |
fdc07f44 | 159 | xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1); |
076e6acb | 160 | |
4e94b71b | 161 | ASSERT(offset + nbblks <= bp->b_length); |
62926044 | 162 | return bp->b_addr + BBTOB(offset); |
076e6acb CH |
163 | } |
164 | ||
1da177e4 LT |
165 | |
166 | /* | |
167 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. | |
168 | */ | |
076e6acb CH |
169 | STATIC int |
170 | xlog_bread_noalign( | |
9a8d2fdb | 171 | struct xlog *log, |
1da177e4 LT |
172 | xfs_daddr_t blk_no, |
173 | int nbblks, | |
9a8d2fdb | 174 | struct xfs_buf *bp) |
1da177e4 LT |
175 | { |
176 | int error; | |
177 | ||
ff30a622 | 178 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 179 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
180 | nbblks); |
181 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
2451337d | 182 | return -EFSCORRUPTED; |
3228149c DC |
183 | } |
184 | ||
69ce58f0 AE |
185 | blk_no = round_down(blk_no, log->l_sectBBsize); |
186 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
187 | |
188 | ASSERT(nbblks > 0); | |
4e94b71b | 189 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
190 | |
191 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
192 | XFS_BUF_READ(bp); | |
aa0e8833 | 193 | bp->b_io_length = nbblks; |
0e95f19a | 194 | bp->b_error = 0; |
1da177e4 | 195 | |
83a0adc3 | 196 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) |
2451337d | 197 | return -EIO; |
83a0adc3 CH |
198 | |
199 | xfs_buf_iorequest(bp); | |
1a1a3e97 | 200 | error = xfs_buf_iowait(bp); |
d64e31a2 | 201 | if (error) |
901796af | 202 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
203 | return error; |
204 | } | |
205 | ||
076e6acb CH |
206 | STATIC int |
207 | xlog_bread( | |
9a8d2fdb | 208 | struct xlog *log, |
076e6acb CH |
209 | xfs_daddr_t blk_no, |
210 | int nbblks, | |
9a8d2fdb | 211 | struct xfs_buf *bp, |
076e6acb CH |
212 | xfs_caddr_t *offset) |
213 | { | |
214 | int error; | |
215 | ||
216 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
217 | if (error) | |
218 | return error; | |
219 | ||
220 | *offset = xlog_align(log, blk_no, nbblks, bp); | |
221 | return 0; | |
222 | } | |
223 | ||
44396476 DC |
224 | /* |
225 | * Read at an offset into the buffer. Returns with the buffer in it's original | |
226 | * state regardless of the result of the read. | |
227 | */ | |
228 | STATIC int | |
229 | xlog_bread_offset( | |
9a8d2fdb | 230 | struct xlog *log, |
44396476 DC |
231 | xfs_daddr_t blk_no, /* block to read from */ |
232 | int nbblks, /* blocks to read */ | |
9a8d2fdb | 233 | struct xfs_buf *bp, |
44396476 DC |
234 | xfs_caddr_t offset) |
235 | { | |
62926044 | 236 | xfs_caddr_t orig_offset = bp->b_addr; |
4e94b71b | 237 | int orig_len = BBTOB(bp->b_length); |
44396476 DC |
238 | int error, error2; |
239 | ||
02fe03d9 | 240 | error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks)); |
44396476 DC |
241 | if (error) |
242 | return error; | |
243 | ||
244 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
245 | ||
246 | /* must reset buffer pointer even on error */ | |
02fe03d9 | 247 | error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len); |
44396476 DC |
248 | if (error) |
249 | return error; | |
250 | return error2; | |
251 | } | |
252 | ||
1da177e4 LT |
253 | /* |
254 | * Write out the buffer at the given block for the given number of blocks. | |
255 | * The buffer is kept locked across the write and is returned locked. | |
256 | * This can only be used for synchronous log writes. | |
257 | */ | |
ba0f32d4 | 258 | STATIC int |
1da177e4 | 259 | xlog_bwrite( |
9a8d2fdb | 260 | struct xlog *log, |
1da177e4 LT |
261 | xfs_daddr_t blk_no, |
262 | int nbblks, | |
9a8d2fdb | 263 | struct xfs_buf *bp) |
1da177e4 LT |
264 | { |
265 | int error; | |
266 | ||
ff30a622 | 267 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 268 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
269 | nbblks); |
270 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
2451337d | 271 | return -EFSCORRUPTED; |
3228149c DC |
272 | } |
273 | ||
69ce58f0 AE |
274 | blk_no = round_down(blk_no, log->l_sectBBsize); |
275 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
276 | |
277 | ASSERT(nbblks > 0); | |
4e94b71b | 278 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
279 | |
280 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
281 | XFS_BUF_ZEROFLAGS(bp); | |
72790aa1 | 282 | xfs_buf_hold(bp); |
0c842ad4 | 283 | xfs_buf_lock(bp); |
aa0e8833 | 284 | bp->b_io_length = nbblks; |
0e95f19a | 285 | bp->b_error = 0; |
1da177e4 | 286 | |
c2b006c1 | 287 | error = xfs_bwrite(bp); |
901796af CH |
288 | if (error) |
289 | xfs_buf_ioerror_alert(bp, __func__); | |
c2b006c1 | 290 | xfs_buf_relse(bp); |
1da177e4 LT |
291 | return error; |
292 | } | |
293 | ||
1da177e4 LT |
294 | #ifdef DEBUG |
295 | /* | |
296 | * dump debug superblock and log record information | |
297 | */ | |
298 | STATIC void | |
299 | xlog_header_check_dump( | |
300 | xfs_mount_t *mp, | |
301 | xlog_rec_header_t *head) | |
302 | { | |
08e96e1a | 303 | xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d", |
03daa57c | 304 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); |
08e96e1a | 305 | xfs_debug(mp, " log : uuid = %pU, fmt = %d", |
03daa57c | 306 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); |
1da177e4 LT |
307 | } |
308 | #else | |
309 | #define xlog_header_check_dump(mp, head) | |
310 | #endif | |
311 | ||
312 | /* | |
313 | * check log record header for recovery | |
314 | */ | |
315 | STATIC int | |
316 | xlog_header_check_recover( | |
317 | xfs_mount_t *mp, | |
318 | xlog_rec_header_t *head) | |
319 | { | |
69ef921b | 320 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 LT |
321 | |
322 | /* | |
323 | * IRIX doesn't write the h_fmt field and leaves it zeroed | |
324 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover | |
325 | * a dirty log created in IRIX. | |
326 | */ | |
69ef921b | 327 | if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) { |
a0fa2b67 DC |
328 | xfs_warn(mp, |
329 | "dirty log written in incompatible format - can't recover"); | |
1da177e4 LT |
330 | xlog_header_check_dump(mp, head); |
331 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", | |
332 | XFS_ERRLEVEL_HIGH, mp); | |
2451337d | 333 | return -EFSCORRUPTED; |
1da177e4 | 334 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
a0fa2b67 DC |
335 | xfs_warn(mp, |
336 | "dirty log entry has mismatched uuid - can't recover"); | |
1da177e4 LT |
337 | xlog_header_check_dump(mp, head); |
338 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", | |
339 | XFS_ERRLEVEL_HIGH, mp); | |
2451337d | 340 | return -EFSCORRUPTED; |
1da177e4 LT |
341 | } |
342 | return 0; | |
343 | } | |
344 | ||
345 | /* | |
346 | * read the head block of the log and check the header | |
347 | */ | |
348 | STATIC int | |
349 | xlog_header_check_mount( | |
350 | xfs_mount_t *mp, | |
351 | xlog_rec_header_t *head) | |
352 | { | |
69ef921b | 353 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 LT |
354 | |
355 | if (uuid_is_nil(&head->h_fs_uuid)) { | |
356 | /* | |
357 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If | |
358 | * h_fs_uuid is nil, we assume this log was last mounted | |
359 | * by IRIX and continue. | |
360 | */ | |
a0fa2b67 | 361 | xfs_warn(mp, "nil uuid in log - IRIX style log"); |
1da177e4 | 362 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
a0fa2b67 | 363 | xfs_warn(mp, "log has mismatched uuid - can't recover"); |
1da177e4 LT |
364 | xlog_header_check_dump(mp, head); |
365 | XFS_ERROR_REPORT("xlog_header_check_mount", | |
366 | XFS_ERRLEVEL_HIGH, mp); | |
2451337d | 367 | return -EFSCORRUPTED; |
1da177e4 LT |
368 | } |
369 | return 0; | |
370 | } | |
371 | ||
372 | STATIC void | |
373 | xlog_recover_iodone( | |
374 | struct xfs_buf *bp) | |
375 | { | |
5a52c2a5 | 376 | if (bp->b_error) { |
1da177e4 LT |
377 | /* |
378 | * We're not going to bother about retrying | |
379 | * this during recovery. One strike! | |
380 | */ | |
901796af | 381 | xfs_buf_ioerror_alert(bp, __func__); |
ebad861b DC |
382 | xfs_force_shutdown(bp->b_target->bt_mount, |
383 | SHUTDOWN_META_IO_ERROR); | |
1da177e4 | 384 | } |
cb669ca5 | 385 | bp->b_iodone = NULL; |
1a1a3e97 | 386 | xfs_buf_ioend(bp, 0); |
1da177e4 LT |
387 | } |
388 | ||
389 | /* | |
390 | * This routine finds (to an approximation) the first block in the physical | |
391 | * log which contains the given cycle. It uses a binary search algorithm. | |
392 | * Note that the algorithm can not be perfect because the disk will not | |
393 | * necessarily be perfect. | |
394 | */ | |
a8272ce0 | 395 | STATIC int |
1da177e4 | 396 | xlog_find_cycle_start( |
9a8d2fdb MT |
397 | struct xlog *log, |
398 | struct xfs_buf *bp, | |
1da177e4 LT |
399 | xfs_daddr_t first_blk, |
400 | xfs_daddr_t *last_blk, | |
401 | uint cycle) | |
402 | { | |
403 | xfs_caddr_t offset; | |
404 | xfs_daddr_t mid_blk; | |
e3bb2e30 | 405 | xfs_daddr_t end_blk; |
1da177e4 LT |
406 | uint mid_cycle; |
407 | int error; | |
408 | ||
e3bb2e30 AE |
409 | end_blk = *last_blk; |
410 | mid_blk = BLK_AVG(first_blk, end_blk); | |
411 | while (mid_blk != first_blk && mid_blk != end_blk) { | |
076e6acb CH |
412 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
413 | if (error) | |
1da177e4 | 414 | return error; |
03bea6fe | 415 | mid_cycle = xlog_get_cycle(offset); |
e3bb2e30 AE |
416 | if (mid_cycle == cycle) |
417 | end_blk = mid_blk; /* last_half_cycle == mid_cycle */ | |
418 | else | |
419 | first_blk = mid_blk; /* first_half_cycle == mid_cycle */ | |
420 | mid_blk = BLK_AVG(first_blk, end_blk); | |
1da177e4 | 421 | } |
e3bb2e30 AE |
422 | ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) || |
423 | (mid_blk == end_blk && mid_blk-1 == first_blk)); | |
424 | ||
425 | *last_blk = end_blk; | |
1da177e4 LT |
426 | |
427 | return 0; | |
428 | } | |
429 | ||
430 | /* | |
3f943d85 AE |
431 | * Check that a range of blocks does not contain stop_on_cycle_no. |
432 | * Fill in *new_blk with the block offset where such a block is | |
433 | * found, or with -1 (an invalid block number) if there is no such | |
434 | * block in the range. The scan needs to occur from front to back | |
435 | * and the pointer into the region must be updated since a later | |
436 | * routine will need to perform another test. | |
1da177e4 LT |
437 | */ |
438 | STATIC int | |
439 | xlog_find_verify_cycle( | |
9a8d2fdb | 440 | struct xlog *log, |
1da177e4 LT |
441 | xfs_daddr_t start_blk, |
442 | int nbblks, | |
443 | uint stop_on_cycle_no, | |
444 | xfs_daddr_t *new_blk) | |
445 | { | |
446 | xfs_daddr_t i, j; | |
447 | uint cycle; | |
448 | xfs_buf_t *bp; | |
449 | xfs_daddr_t bufblks; | |
450 | xfs_caddr_t buf = NULL; | |
451 | int error = 0; | |
452 | ||
6881a229 AE |
453 | /* |
454 | * Greedily allocate a buffer big enough to handle the full | |
455 | * range of basic blocks we'll be examining. If that fails, | |
456 | * try a smaller size. We need to be able to read at least | |
457 | * a log sector, or we're out of luck. | |
458 | */ | |
1da177e4 | 459 | bufblks = 1 << ffs(nbblks); |
81158e0c DC |
460 | while (bufblks > log->l_logBBsize) |
461 | bufblks >>= 1; | |
1da177e4 | 462 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1da177e4 | 463 | bufblks >>= 1; |
69ce58f0 | 464 | if (bufblks < log->l_sectBBsize) |
2451337d | 465 | return -ENOMEM; |
1da177e4 LT |
466 | } |
467 | ||
468 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { | |
469 | int bcount; | |
470 | ||
471 | bcount = min(bufblks, (start_blk + nbblks - i)); | |
472 | ||
076e6acb CH |
473 | error = xlog_bread(log, i, bcount, bp, &buf); |
474 | if (error) | |
1da177e4 LT |
475 | goto out; |
476 | ||
1da177e4 | 477 | for (j = 0; j < bcount; j++) { |
03bea6fe | 478 | cycle = xlog_get_cycle(buf); |
1da177e4 LT |
479 | if (cycle == stop_on_cycle_no) { |
480 | *new_blk = i+j; | |
481 | goto out; | |
482 | } | |
483 | ||
484 | buf += BBSIZE; | |
485 | } | |
486 | } | |
487 | ||
488 | *new_blk = -1; | |
489 | ||
490 | out: | |
491 | xlog_put_bp(bp); | |
492 | return error; | |
493 | } | |
494 | ||
495 | /* | |
496 | * Potentially backup over partial log record write. | |
497 | * | |
498 | * In the typical case, last_blk is the number of the block directly after | |
499 | * a good log record. Therefore, we subtract one to get the block number | |
500 | * of the last block in the given buffer. extra_bblks contains the number | |
501 | * of blocks we would have read on a previous read. This happens when the | |
502 | * last log record is split over the end of the physical log. | |
503 | * | |
504 | * extra_bblks is the number of blocks potentially verified on a previous | |
505 | * call to this routine. | |
506 | */ | |
507 | STATIC int | |
508 | xlog_find_verify_log_record( | |
9a8d2fdb | 509 | struct xlog *log, |
1da177e4 LT |
510 | xfs_daddr_t start_blk, |
511 | xfs_daddr_t *last_blk, | |
512 | int extra_bblks) | |
513 | { | |
514 | xfs_daddr_t i; | |
515 | xfs_buf_t *bp; | |
516 | xfs_caddr_t offset = NULL; | |
517 | xlog_rec_header_t *head = NULL; | |
518 | int error = 0; | |
519 | int smallmem = 0; | |
520 | int num_blks = *last_blk - start_blk; | |
521 | int xhdrs; | |
522 | ||
523 | ASSERT(start_blk != 0 || *last_blk != start_blk); | |
524 | ||
525 | if (!(bp = xlog_get_bp(log, num_blks))) { | |
526 | if (!(bp = xlog_get_bp(log, 1))) | |
2451337d | 527 | return -ENOMEM; |
1da177e4 LT |
528 | smallmem = 1; |
529 | } else { | |
076e6acb CH |
530 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
531 | if (error) | |
1da177e4 | 532 | goto out; |
1da177e4 LT |
533 | offset += ((num_blks - 1) << BBSHIFT); |
534 | } | |
535 | ||
536 | for (i = (*last_blk) - 1; i >= 0; i--) { | |
537 | if (i < start_blk) { | |
538 | /* valid log record not found */ | |
a0fa2b67 DC |
539 | xfs_warn(log->l_mp, |
540 | "Log inconsistent (didn't find previous header)"); | |
1da177e4 | 541 | ASSERT(0); |
2451337d | 542 | error = -EIO; |
1da177e4 LT |
543 | goto out; |
544 | } | |
545 | ||
546 | if (smallmem) { | |
076e6acb CH |
547 | error = xlog_bread(log, i, 1, bp, &offset); |
548 | if (error) | |
1da177e4 | 549 | goto out; |
1da177e4 LT |
550 | } |
551 | ||
552 | head = (xlog_rec_header_t *)offset; | |
553 | ||
69ef921b | 554 | if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
1da177e4 LT |
555 | break; |
556 | ||
557 | if (!smallmem) | |
558 | offset -= BBSIZE; | |
559 | } | |
560 | ||
561 | /* | |
562 | * We hit the beginning of the physical log & still no header. Return | |
563 | * to caller. If caller can handle a return of -1, then this routine | |
564 | * will be called again for the end of the physical log. | |
565 | */ | |
566 | if (i == -1) { | |
2451337d | 567 | error = 1; |
1da177e4 LT |
568 | goto out; |
569 | } | |
570 | ||
571 | /* | |
572 | * We have the final block of the good log (the first block | |
573 | * of the log record _before_ the head. So we check the uuid. | |
574 | */ | |
575 | if ((error = xlog_header_check_mount(log->l_mp, head))) | |
576 | goto out; | |
577 | ||
578 | /* | |
579 | * We may have found a log record header before we expected one. | |
580 | * last_blk will be the 1st block # with a given cycle #. We may end | |
581 | * up reading an entire log record. In this case, we don't want to | |
582 | * reset last_blk. Only when last_blk points in the middle of a log | |
583 | * record do we update last_blk. | |
584 | */ | |
62118709 | 585 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d | 586 | uint h_size = be32_to_cpu(head->h_size); |
1da177e4 LT |
587 | |
588 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; | |
589 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
590 | xhdrs++; | |
591 | } else { | |
592 | xhdrs = 1; | |
593 | } | |
594 | ||
b53e675d CH |
595 | if (*last_blk - i + extra_bblks != |
596 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) | |
1da177e4 LT |
597 | *last_blk = i; |
598 | ||
599 | out: | |
600 | xlog_put_bp(bp); | |
601 | return error; | |
602 | } | |
603 | ||
604 | /* | |
605 | * Head is defined to be the point of the log where the next log write | |
0a94da24 | 606 | * could go. This means that incomplete LR writes at the end are |
1da177e4 LT |
607 | * eliminated when calculating the head. We aren't guaranteed that previous |
608 | * LR have complete transactions. We only know that a cycle number of | |
609 | * current cycle number -1 won't be present in the log if we start writing | |
610 | * from our current block number. | |
611 | * | |
612 | * last_blk contains the block number of the first block with a given | |
613 | * cycle number. | |
614 | * | |
615 | * Return: zero if normal, non-zero if error. | |
616 | */ | |
ba0f32d4 | 617 | STATIC int |
1da177e4 | 618 | xlog_find_head( |
9a8d2fdb | 619 | struct xlog *log, |
1da177e4 LT |
620 | xfs_daddr_t *return_head_blk) |
621 | { | |
622 | xfs_buf_t *bp; | |
623 | xfs_caddr_t offset; | |
624 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; | |
625 | int num_scan_bblks; | |
626 | uint first_half_cycle, last_half_cycle; | |
627 | uint stop_on_cycle; | |
628 | int error, log_bbnum = log->l_logBBsize; | |
629 | ||
630 | /* Is the end of the log device zeroed? */ | |
2451337d DC |
631 | error = xlog_find_zeroed(log, &first_blk); |
632 | if (error < 0) { | |
633 | xfs_warn(log->l_mp, "empty log check failed"); | |
634 | return error; | |
635 | } | |
636 | if (error == 1) { | |
1da177e4 LT |
637 | *return_head_blk = first_blk; |
638 | ||
639 | /* Is the whole lot zeroed? */ | |
640 | if (!first_blk) { | |
641 | /* Linux XFS shouldn't generate totally zeroed logs - | |
642 | * mkfs etc write a dummy unmount record to a fresh | |
643 | * log so we can store the uuid in there | |
644 | */ | |
a0fa2b67 | 645 | xfs_warn(log->l_mp, "totally zeroed log"); |
1da177e4 LT |
646 | } |
647 | ||
648 | return 0; | |
1da177e4 LT |
649 | } |
650 | ||
651 | first_blk = 0; /* get cycle # of 1st block */ | |
652 | bp = xlog_get_bp(log, 1); | |
653 | if (!bp) | |
2451337d | 654 | return -ENOMEM; |
076e6acb CH |
655 | |
656 | error = xlog_bread(log, 0, 1, bp, &offset); | |
657 | if (error) | |
1da177e4 | 658 | goto bp_err; |
076e6acb | 659 | |
03bea6fe | 660 | first_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
661 | |
662 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ | |
076e6acb CH |
663 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
664 | if (error) | |
1da177e4 | 665 | goto bp_err; |
076e6acb | 666 | |
03bea6fe | 667 | last_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
668 | ASSERT(last_half_cycle != 0); |
669 | ||
670 | /* | |
671 | * If the 1st half cycle number is equal to the last half cycle number, | |
672 | * then the entire log is stamped with the same cycle number. In this | |
673 | * case, head_blk can't be set to zero (which makes sense). The below | |
674 | * math doesn't work out properly with head_blk equal to zero. Instead, | |
675 | * we set it to log_bbnum which is an invalid block number, but this | |
676 | * value makes the math correct. If head_blk doesn't changed through | |
677 | * all the tests below, *head_blk is set to zero at the very end rather | |
678 | * than log_bbnum. In a sense, log_bbnum and zero are the same block | |
679 | * in a circular file. | |
680 | */ | |
681 | if (first_half_cycle == last_half_cycle) { | |
682 | /* | |
683 | * In this case we believe that the entire log should have | |
684 | * cycle number last_half_cycle. We need to scan backwards | |
685 | * from the end verifying that there are no holes still | |
686 | * containing last_half_cycle - 1. If we find such a hole, | |
687 | * then the start of that hole will be the new head. The | |
688 | * simple case looks like | |
689 | * x | x ... | x - 1 | x | |
690 | * Another case that fits this picture would be | |
691 | * x | x + 1 | x ... | x | |
c41564b5 | 692 | * In this case the head really is somewhere at the end of the |
1da177e4 LT |
693 | * log, as one of the latest writes at the beginning was |
694 | * incomplete. | |
695 | * One more case is | |
696 | * x | x + 1 | x ... | x - 1 | x | |
697 | * This is really the combination of the above two cases, and | |
698 | * the head has to end up at the start of the x-1 hole at the | |
699 | * end of the log. | |
700 | * | |
701 | * In the 256k log case, we will read from the beginning to the | |
702 | * end of the log and search for cycle numbers equal to x-1. | |
703 | * We don't worry about the x+1 blocks that we encounter, | |
704 | * because we know that they cannot be the head since the log | |
705 | * started with x. | |
706 | */ | |
707 | head_blk = log_bbnum; | |
708 | stop_on_cycle = last_half_cycle - 1; | |
709 | } else { | |
710 | /* | |
711 | * In this case we want to find the first block with cycle | |
712 | * number matching last_half_cycle. We expect the log to be | |
713 | * some variation on | |
3f943d85 | 714 | * x + 1 ... | x ... | x |
1da177e4 LT |
715 | * The first block with cycle number x (last_half_cycle) will |
716 | * be where the new head belongs. First we do a binary search | |
717 | * for the first occurrence of last_half_cycle. The binary | |
718 | * search may not be totally accurate, so then we scan back | |
719 | * from there looking for occurrences of last_half_cycle before | |
720 | * us. If that backwards scan wraps around the beginning of | |
721 | * the log, then we look for occurrences of last_half_cycle - 1 | |
722 | * at the end of the log. The cases we're looking for look | |
723 | * like | |
3f943d85 AE |
724 | * v binary search stopped here |
725 | * x + 1 ... | x | x + 1 | x ... | x | |
726 | * ^ but we want to locate this spot | |
1da177e4 | 727 | * or |
1da177e4 | 728 | * <---------> less than scan distance |
3f943d85 AE |
729 | * x + 1 ... | x ... | x - 1 | x |
730 | * ^ we want to locate this spot | |
1da177e4 LT |
731 | */ |
732 | stop_on_cycle = last_half_cycle; | |
733 | if ((error = xlog_find_cycle_start(log, bp, first_blk, | |
734 | &head_blk, last_half_cycle))) | |
735 | goto bp_err; | |
736 | } | |
737 | ||
738 | /* | |
739 | * Now validate the answer. Scan back some number of maximum possible | |
740 | * blocks and make sure each one has the expected cycle number. The | |
741 | * maximum is determined by the total possible amount of buffering | |
742 | * in the in-core log. The following number can be made tighter if | |
743 | * we actually look at the block size of the filesystem. | |
744 | */ | |
745 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
746 | if (head_blk >= num_scan_bblks) { | |
747 | /* | |
748 | * We are guaranteed that the entire check can be performed | |
749 | * in one buffer. | |
750 | */ | |
751 | start_blk = head_blk - num_scan_bblks; | |
752 | if ((error = xlog_find_verify_cycle(log, | |
753 | start_blk, num_scan_bblks, | |
754 | stop_on_cycle, &new_blk))) | |
755 | goto bp_err; | |
756 | if (new_blk != -1) | |
757 | head_blk = new_blk; | |
758 | } else { /* need to read 2 parts of log */ | |
759 | /* | |
760 | * We are going to scan backwards in the log in two parts. | |
761 | * First we scan the physical end of the log. In this part | |
762 | * of the log, we are looking for blocks with cycle number | |
763 | * last_half_cycle - 1. | |
764 | * If we find one, then we know that the log starts there, as | |
765 | * we've found a hole that didn't get written in going around | |
766 | * the end of the physical log. The simple case for this is | |
767 | * x + 1 ... | x ... | x - 1 | x | |
768 | * <---------> less than scan distance | |
769 | * If all of the blocks at the end of the log have cycle number | |
770 | * last_half_cycle, then we check the blocks at the start of | |
771 | * the log looking for occurrences of last_half_cycle. If we | |
772 | * find one, then our current estimate for the location of the | |
773 | * first occurrence of last_half_cycle is wrong and we move | |
774 | * back to the hole we've found. This case looks like | |
775 | * x + 1 ... | x | x + 1 | x ... | |
776 | * ^ binary search stopped here | |
777 | * Another case we need to handle that only occurs in 256k | |
778 | * logs is | |
779 | * x + 1 ... | x ... | x+1 | x ... | |
780 | * ^ binary search stops here | |
781 | * In a 256k log, the scan at the end of the log will see the | |
782 | * x + 1 blocks. We need to skip past those since that is | |
783 | * certainly not the head of the log. By searching for | |
784 | * last_half_cycle-1 we accomplish that. | |
785 | */ | |
1da177e4 | 786 | ASSERT(head_blk <= INT_MAX && |
3f943d85 AE |
787 | (xfs_daddr_t) num_scan_bblks >= head_blk); |
788 | start_blk = log_bbnum - (num_scan_bblks - head_blk); | |
1da177e4 LT |
789 | if ((error = xlog_find_verify_cycle(log, start_blk, |
790 | num_scan_bblks - (int)head_blk, | |
791 | (stop_on_cycle - 1), &new_blk))) | |
792 | goto bp_err; | |
793 | if (new_blk != -1) { | |
794 | head_blk = new_blk; | |
9db127ed | 795 | goto validate_head; |
1da177e4 LT |
796 | } |
797 | ||
798 | /* | |
799 | * Scan beginning of log now. The last part of the physical | |
800 | * log is good. This scan needs to verify that it doesn't find | |
801 | * the last_half_cycle. | |
802 | */ | |
803 | start_blk = 0; | |
804 | ASSERT(head_blk <= INT_MAX); | |
805 | if ((error = xlog_find_verify_cycle(log, | |
806 | start_blk, (int)head_blk, | |
807 | stop_on_cycle, &new_blk))) | |
808 | goto bp_err; | |
809 | if (new_blk != -1) | |
810 | head_blk = new_blk; | |
811 | } | |
812 | ||
9db127ed | 813 | validate_head: |
1da177e4 LT |
814 | /* |
815 | * Now we need to make sure head_blk is not pointing to a block in | |
816 | * the middle of a log record. | |
817 | */ | |
818 | num_scan_bblks = XLOG_REC_SHIFT(log); | |
819 | if (head_blk >= num_scan_bblks) { | |
820 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ | |
821 | ||
822 | /* start ptr at last block ptr before head_blk */ | |
2451337d DC |
823 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
824 | if (error == 1) | |
825 | error = -EIO; | |
826 | if (error) | |
1da177e4 LT |
827 | goto bp_err; |
828 | } else { | |
829 | start_blk = 0; | |
830 | ASSERT(head_blk <= INT_MAX); | |
2451337d DC |
831 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
832 | if (error < 0) | |
833 | goto bp_err; | |
834 | if (error == 1) { | |
1da177e4 | 835 | /* We hit the beginning of the log during our search */ |
3f943d85 | 836 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
1da177e4 LT |
837 | new_blk = log_bbnum; |
838 | ASSERT(start_blk <= INT_MAX && | |
839 | (xfs_daddr_t) log_bbnum-start_blk >= 0); | |
840 | ASSERT(head_blk <= INT_MAX); | |
2451337d DC |
841 | error = xlog_find_verify_log_record(log, start_blk, |
842 | &new_blk, (int)head_blk); | |
843 | if (error == 1) | |
844 | error = -EIO; | |
845 | if (error) | |
1da177e4 LT |
846 | goto bp_err; |
847 | if (new_blk != log_bbnum) | |
848 | head_blk = new_blk; | |
849 | } else if (error) | |
850 | goto bp_err; | |
851 | } | |
852 | ||
853 | xlog_put_bp(bp); | |
854 | if (head_blk == log_bbnum) | |
855 | *return_head_blk = 0; | |
856 | else | |
857 | *return_head_blk = head_blk; | |
858 | /* | |
859 | * When returning here, we have a good block number. Bad block | |
860 | * means that during a previous crash, we didn't have a clean break | |
861 | * from cycle number N to cycle number N-1. In this case, we need | |
862 | * to find the first block with cycle number N-1. | |
863 | */ | |
864 | return 0; | |
865 | ||
866 | bp_err: | |
867 | xlog_put_bp(bp); | |
868 | ||
869 | if (error) | |
a0fa2b67 | 870 | xfs_warn(log->l_mp, "failed to find log head"); |
1da177e4 LT |
871 | return error; |
872 | } | |
873 | ||
874 | /* | |
875 | * Find the sync block number or the tail of the log. | |
876 | * | |
877 | * This will be the block number of the last record to have its | |
878 | * associated buffers synced to disk. Every log record header has | |
879 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy | |
880 | * to get a sync block number. The only concern is to figure out which | |
881 | * log record header to believe. | |
882 | * | |
883 | * The following algorithm uses the log record header with the largest | |
884 | * lsn. The entire log record does not need to be valid. We only care | |
885 | * that the header is valid. | |
886 | * | |
887 | * We could speed up search by using current head_blk buffer, but it is not | |
888 | * available. | |
889 | */ | |
5d77c0dc | 890 | STATIC int |
1da177e4 | 891 | xlog_find_tail( |
9a8d2fdb | 892 | struct xlog *log, |
1da177e4 | 893 | xfs_daddr_t *head_blk, |
65be6054 | 894 | xfs_daddr_t *tail_blk) |
1da177e4 LT |
895 | { |
896 | xlog_rec_header_t *rhead; | |
897 | xlog_op_header_t *op_head; | |
898 | xfs_caddr_t offset = NULL; | |
899 | xfs_buf_t *bp; | |
900 | int error, i, found; | |
901 | xfs_daddr_t umount_data_blk; | |
902 | xfs_daddr_t after_umount_blk; | |
903 | xfs_lsn_t tail_lsn; | |
904 | int hblks; | |
905 | ||
906 | found = 0; | |
907 | ||
908 | /* | |
909 | * Find previous log record | |
910 | */ | |
911 | if ((error = xlog_find_head(log, head_blk))) | |
912 | return error; | |
913 | ||
914 | bp = xlog_get_bp(log, 1); | |
915 | if (!bp) | |
2451337d | 916 | return -ENOMEM; |
1da177e4 | 917 | if (*head_blk == 0) { /* special case */ |
076e6acb CH |
918 | error = xlog_bread(log, 0, 1, bp, &offset); |
919 | if (error) | |
9db127ed | 920 | goto done; |
076e6acb | 921 | |
03bea6fe | 922 | if (xlog_get_cycle(offset) == 0) { |
1da177e4 LT |
923 | *tail_blk = 0; |
924 | /* leave all other log inited values alone */ | |
9db127ed | 925 | goto done; |
1da177e4 LT |
926 | } |
927 | } | |
928 | ||
929 | /* | |
930 | * Search backwards looking for log record header block | |
931 | */ | |
932 | ASSERT(*head_blk < INT_MAX); | |
933 | for (i = (int)(*head_blk) - 1; i >= 0; i--) { | |
076e6acb CH |
934 | error = xlog_bread(log, i, 1, bp, &offset); |
935 | if (error) | |
9db127ed | 936 | goto done; |
076e6acb | 937 | |
69ef921b | 938 | if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
1da177e4 LT |
939 | found = 1; |
940 | break; | |
941 | } | |
942 | } | |
943 | /* | |
944 | * If we haven't found the log record header block, start looking | |
945 | * again from the end of the physical log. XXXmiken: There should be | |
946 | * a check here to make sure we didn't search more than N blocks in | |
947 | * the previous code. | |
948 | */ | |
949 | if (!found) { | |
950 | for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) { | |
076e6acb CH |
951 | error = xlog_bread(log, i, 1, bp, &offset); |
952 | if (error) | |
9db127ed | 953 | goto done; |
076e6acb | 954 | |
69ef921b CH |
955 | if (*(__be32 *)offset == |
956 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
1da177e4 LT |
957 | found = 2; |
958 | break; | |
959 | } | |
960 | } | |
961 | } | |
962 | if (!found) { | |
a0fa2b67 | 963 | xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__); |
050a1952 | 964 | xlog_put_bp(bp); |
1da177e4 | 965 | ASSERT(0); |
2451337d | 966 | return -EIO; |
1da177e4 LT |
967 | } |
968 | ||
969 | /* find blk_no of tail of log */ | |
970 | rhead = (xlog_rec_header_t *)offset; | |
b53e675d | 971 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); |
1da177e4 LT |
972 | |
973 | /* | |
974 | * Reset log values according to the state of the log when we | |
975 | * crashed. In the case where head_blk == 0, we bump curr_cycle | |
976 | * one because the next write starts a new cycle rather than | |
977 | * continuing the cycle of the last good log record. At this | |
978 | * point we have guaranteed that all partial log records have been | |
979 | * accounted for. Therefore, we know that the last good log record | |
980 | * written was complete and ended exactly on the end boundary | |
981 | * of the physical log. | |
982 | */ | |
983 | log->l_prev_block = i; | |
984 | log->l_curr_block = (int)*head_blk; | |
b53e675d | 985 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); |
1da177e4 LT |
986 | if (found == 2) |
987 | log->l_curr_cycle++; | |
1c3cb9ec | 988 | atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn)); |
84f3c683 | 989 | atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn)); |
28496968 | 990 | xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle, |
a69ed03c | 991 | BBTOB(log->l_curr_block)); |
28496968 | 992 | xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle, |
a69ed03c | 993 | BBTOB(log->l_curr_block)); |
1da177e4 LT |
994 | |
995 | /* | |
996 | * Look for unmount record. If we find it, then we know there | |
997 | * was a clean unmount. Since 'i' could be the last block in | |
998 | * the physical log, we convert to a log block before comparing | |
999 | * to the head_blk. | |
1000 | * | |
1001 | * Save the current tail lsn to use to pass to | |
1002 | * xlog_clear_stale_blocks() below. We won't want to clear the | |
1003 | * unmount record if there is one, so we pass the lsn of the | |
1004 | * unmount record rather than the block after it. | |
1005 | */ | |
62118709 | 1006 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d CH |
1007 | int h_size = be32_to_cpu(rhead->h_size); |
1008 | int h_version = be32_to_cpu(rhead->h_version); | |
1da177e4 LT |
1009 | |
1010 | if ((h_version & XLOG_VERSION_2) && | |
1011 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
1012 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
1013 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
1014 | hblks++; | |
1015 | } else { | |
1016 | hblks = 1; | |
1017 | } | |
1018 | } else { | |
1019 | hblks = 1; | |
1020 | } | |
1021 | after_umount_blk = (i + hblks + (int) | |
b53e675d | 1022 | BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize; |
1c3cb9ec | 1023 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
1da177e4 | 1024 | if (*head_blk == after_umount_blk && |
b53e675d | 1025 | be32_to_cpu(rhead->h_num_logops) == 1) { |
1da177e4 | 1026 | umount_data_blk = (i + hblks) % log->l_logBBsize; |
076e6acb CH |
1027 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); |
1028 | if (error) | |
9db127ed | 1029 | goto done; |
076e6acb | 1030 | |
1da177e4 LT |
1031 | op_head = (xlog_op_header_t *)offset; |
1032 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { | |
1033 | /* | |
1034 | * Set tail and last sync so that newly written | |
1035 | * log records will point recovery to after the | |
1036 | * current unmount record. | |
1037 | */ | |
1c3cb9ec DC |
1038 | xlog_assign_atomic_lsn(&log->l_tail_lsn, |
1039 | log->l_curr_cycle, after_umount_blk); | |
1040 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, | |
1041 | log->l_curr_cycle, after_umount_blk); | |
1da177e4 | 1042 | *tail_blk = after_umount_blk; |
92821e2b DC |
1043 | |
1044 | /* | |
1045 | * Note that the unmount was clean. If the unmount | |
1046 | * was not clean, we need to know this to rebuild the | |
1047 | * superblock counters from the perag headers if we | |
1048 | * have a filesystem using non-persistent counters. | |
1049 | */ | |
1050 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; | |
1da177e4 LT |
1051 | } |
1052 | } | |
1053 | ||
1054 | /* | |
1055 | * Make sure that there are no blocks in front of the head | |
1056 | * with the same cycle number as the head. This can happen | |
1057 | * because we allow multiple outstanding log writes concurrently, | |
1058 | * and the later writes might make it out before earlier ones. | |
1059 | * | |
1060 | * We use the lsn from before modifying it so that we'll never | |
1061 | * overwrite the unmount record after a clean unmount. | |
1062 | * | |
1063 | * Do this only if we are going to recover the filesystem | |
1064 | * | |
1065 | * NOTE: This used to say "if (!readonly)" | |
1066 | * However on Linux, we can & do recover a read-only filesystem. | |
1067 | * We only skip recovery if NORECOVERY is specified on mount, | |
1068 | * in which case we would not be here. | |
1069 | * | |
1070 | * But... if the -device- itself is readonly, just skip this. | |
1071 | * We can't recover this device anyway, so it won't matter. | |
1072 | */ | |
9db127ed | 1073 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) |
1da177e4 | 1074 | error = xlog_clear_stale_blocks(log, tail_lsn); |
1da177e4 | 1075 | |
9db127ed | 1076 | done: |
1da177e4 LT |
1077 | xlog_put_bp(bp); |
1078 | ||
1079 | if (error) | |
a0fa2b67 | 1080 | xfs_warn(log->l_mp, "failed to locate log tail"); |
1da177e4 LT |
1081 | return error; |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * Is the log zeroed at all? | |
1086 | * | |
1087 | * The last binary search should be changed to perform an X block read | |
1088 | * once X becomes small enough. You can then search linearly through | |
1089 | * the X blocks. This will cut down on the number of reads we need to do. | |
1090 | * | |
1091 | * If the log is partially zeroed, this routine will pass back the blkno | |
1092 | * of the first block with cycle number 0. It won't have a complete LR | |
1093 | * preceding it. | |
1094 | * | |
1095 | * Return: | |
1096 | * 0 => the log is completely written to | |
2451337d DC |
1097 | * 1 => use *blk_no as the first block of the log |
1098 | * <0 => error has occurred | |
1da177e4 | 1099 | */ |
a8272ce0 | 1100 | STATIC int |
1da177e4 | 1101 | xlog_find_zeroed( |
9a8d2fdb | 1102 | struct xlog *log, |
1da177e4 LT |
1103 | xfs_daddr_t *blk_no) |
1104 | { | |
1105 | xfs_buf_t *bp; | |
1106 | xfs_caddr_t offset; | |
1107 | uint first_cycle, last_cycle; | |
1108 | xfs_daddr_t new_blk, last_blk, start_blk; | |
1109 | xfs_daddr_t num_scan_bblks; | |
1110 | int error, log_bbnum = log->l_logBBsize; | |
1111 | ||
6fdf8ccc NS |
1112 | *blk_no = 0; |
1113 | ||
1da177e4 LT |
1114 | /* check totally zeroed log */ |
1115 | bp = xlog_get_bp(log, 1); | |
1116 | if (!bp) | |
2451337d | 1117 | return -ENOMEM; |
076e6acb CH |
1118 | error = xlog_bread(log, 0, 1, bp, &offset); |
1119 | if (error) | |
1da177e4 | 1120 | goto bp_err; |
076e6acb | 1121 | |
03bea6fe | 1122 | first_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1123 | if (first_cycle == 0) { /* completely zeroed log */ |
1124 | *blk_no = 0; | |
1125 | xlog_put_bp(bp); | |
2451337d | 1126 | return 1; |
1da177e4 LT |
1127 | } |
1128 | ||
1129 | /* check partially zeroed log */ | |
076e6acb CH |
1130 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
1131 | if (error) | |
1da177e4 | 1132 | goto bp_err; |
076e6acb | 1133 | |
03bea6fe | 1134 | last_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1135 | if (last_cycle != 0) { /* log completely written to */ |
1136 | xlog_put_bp(bp); | |
1137 | return 0; | |
1138 | } else if (first_cycle != 1) { | |
1139 | /* | |
1140 | * If the cycle of the last block is zero, the cycle of | |
1141 | * the first block must be 1. If it's not, maybe we're | |
1142 | * not looking at a log... Bail out. | |
1143 | */ | |
a0fa2b67 DC |
1144 | xfs_warn(log->l_mp, |
1145 | "Log inconsistent or not a log (last==0, first!=1)"); | |
2451337d | 1146 | error = -EINVAL; |
5d0a6549 | 1147 | goto bp_err; |
1da177e4 LT |
1148 | } |
1149 | ||
1150 | /* we have a partially zeroed log */ | |
1151 | last_blk = log_bbnum-1; | |
1152 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) | |
1153 | goto bp_err; | |
1154 | ||
1155 | /* | |
1156 | * Validate the answer. Because there is no way to guarantee that | |
1157 | * the entire log is made up of log records which are the same size, | |
1158 | * we scan over the defined maximum blocks. At this point, the maximum | |
1159 | * is not chosen to mean anything special. XXXmiken | |
1160 | */ | |
1161 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
1162 | ASSERT(num_scan_bblks <= INT_MAX); | |
1163 | ||
1164 | if (last_blk < num_scan_bblks) | |
1165 | num_scan_bblks = last_blk; | |
1166 | start_blk = last_blk - num_scan_bblks; | |
1167 | ||
1168 | /* | |
1169 | * We search for any instances of cycle number 0 that occur before | |
1170 | * our current estimate of the head. What we're trying to detect is | |
1171 | * 1 ... | 0 | 1 | 0... | |
1172 | * ^ binary search ends here | |
1173 | */ | |
1174 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
1175 | (int)num_scan_bblks, 0, &new_blk))) | |
1176 | goto bp_err; | |
1177 | if (new_blk != -1) | |
1178 | last_blk = new_blk; | |
1179 | ||
1180 | /* | |
1181 | * Potentially backup over partial log record write. We don't need | |
1182 | * to search the end of the log because we know it is zero. | |
1183 | */ | |
2451337d DC |
1184 | error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0); |
1185 | if (error == 1) | |
1186 | error = -EIO; | |
1187 | if (error) | |
1188 | goto bp_err; | |
1da177e4 LT |
1189 | |
1190 | *blk_no = last_blk; | |
1191 | bp_err: | |
1192 | xlog_put_bp(bp); | |
1193 | if (error) | |
1194 | return error; | |
2451337d | 1195 | return 1; |
1da177e4 LT |
1196 | } |
1197 | ||
1198 | /* | |
1199 | * These are simple subroutines used by xlog_clear_stale_blocks() below | |
1200 | * to initialize a buffer full of empty log record headers and write | |
1201 | * them into the log. | |
1202 | */ | |
1203 | STATIC void | |
1204 | xlog_add_record( | |
9a8d2fdb | 1205 | struct xlog *log, |
1da177e4 LT |
1206 | xfs_caddr_t buf, |
1207 | int cycle, | |
1208 | int block, | |
1209 | int tail_cycle, | |
1210 | int tail_block) | |
1211 | { | |
1212 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; | |
1213 | ||
1214 | memset(buf, 0, BBSIZE); | |
b53e675d CH |
1215 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1216 | recp->h_cycle = cpu_to_be32(cycle); | |
1217 | recp->h_version = cpu_to_be32( | |
62118709 | 1218 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d CH |
1219 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
1220 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); | |
1221 | recp->h_fmt = cpu_to_be32(XLOG_FMT); | |
1da177e4 LT |
1222 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1223 | } | |
1224 | ||
1225 | STATIC int | |
1226 | xlog_write_log_records( | |
9a8d2fdb | 1227 | struct xlog *log, |
1da177e4 LT |
1228 | int cycle, |
1229 | int start_block, | |
1230 | int blocks, | |
1231 | int tail_cycle, | |
1232 | int tail_block) | |
1233 | { | |
1234 | xfs_caddr_t offset; | |
1235 | xfs_buf_t *bp; | |
1236 | int balign, ealign; | |
69ce58f0 | 1237 | int sectbb = log->l_sectBBsize; |
1da177e4 LT |
1238 | int end_block = start_block + blocks; |
1239 | int bufblks; | |
1240 | int error = 0; | |
1241 | int i, j = 0; | |
1242 | ||
6881a229 AE |
1243 | /* |
1244 | * Greedily allocate a buffer big enough to handle the full | |
1245 | * range of basic blocks to be written. If that fails, try | |
1246 | * a smaller size. We need to be able to write at least a | |
1247 | * log sector, or we're out of luck. | |
1248 | */ | |
1da177e4 | 1249 | bufblks = 1 << ffs(blocks); |
81158e0c DC |
1250 | while (bufblks > log->l_logBBsize) |
1251 | bufblks >>= 1; | |
1da177e4 LT |
1252 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1253 | bufblks >>= 1; | |
69ce58f0 | 1254 | if (bufblks < sectbb) |
2451337d | 1255 | return -ENOMEM; |
1da177e4 LT |
1256 | } |
1257 | ||
1258 | /* We may need to do a read at the start to fill in part of | |
1259 | * the buffer in the starting sector not covered by the first | |
1260 | * write below. | |
1261 | */ | |
5c17f533 | 1262 | balign = round_down(start_block, sectbb); |
1da177e4 | 1263 | if (balign != start_block) { |
076e6acb CH |
1264 | error = xlog_bread_noalign(log, start_block, 1, bp); |
1265 | if (error) | |
1266 | goto out_put_bp; | |
1267 | ||
1da177e4 LT |
1268 | j = start_block - balign; |
1269 | } | |
1270 | ||
1271 | for (i = start_block; i < end_block; i += bufblks) { | |
1272 | int bcount, endcount; | |
1273 | ||
1274 | bcount = min(bufblks, end_block - start_block); | |
1275 | endcount = bcount - j; | |
1276 | ||
1277 | /* We may need to do a read at the end to fill in part of | |
1278 | * the buffer in the final sector not covered by the write. | |
1279 | * If this is the same sector as the above read, skip it. | |
1280 | */ | |
5c17f533 | 1281 | ealign = round_down(end_block, sectbb); |
1da177e4 | 1282 | if (j == 0 && (start_block + endcount > ealign)) { |
62926044 | 1283 | offset = bp->b_addr + BBTOB(ealign - start_block); |
44396476 DC |
1284 | error = xlog_bread_offset(log, ealign, sectbb, |
1285 | bp, offset); | |
076e6acb CH |
1286 | if (error) |
1287 | break; | |
1288 | ||
1da177e4 LT |
1289 | } |
1290 | ||
1291 | offset = xlog_align(log, start_block, endcount, bp); | |
1292 | for (; j < endcount; j++) { | |
1293 | xlog_add_record(log, offset, cycle, i+j, | |
1294 | tail_cycle, tail_block); | |
1295 | offset += BBSIZE; | |
1296 | } | |
1297 | error = xlog_bwrite(log, start_block, endcount, bp); | |
1298 | if (error) | |
1299 | break; | |
1300 | start_block += endcount; | |
1301 | j = 0; | |
1302 | } | |
076e6acb CH |
1303 | |
1304 | out_put_bp: | |
1da177e4 LT |
1305 | xlog_put_bp(bp); |
1306 | return error; | |
1307 | } | |
1308 | ||
1309 | /* | |
1310 | * This routine is called to blow away any incomplete log writes out | |
1311 | * in front of the log head. We do this so that we won't become confused | |
1312 | * if we come up, write only a little bit more, and then crash again. | |
1313 | * If we leave the partial log records out there, this situation could | |
1314 | * cause us to think those partial writes are valid blocks since they | |
1315 | * have the current cycle number. We get rid of them by overwriting them | |
1316 | * with empty log records with the old cycle number rather than the | |
1317 | * current one. | |
1318 | * | |
1319 | * The tail lsn is passed in rather than taken from | |
1320 | * the log so that we will not write over the unmount record after a | |
1321 | * clean unmount in a 512 block log. Doing so would leave the log without | |
1322 | * any valid log records in it until a new one was written. If we crashed | |
1323 | * during that time we would not be able to recover. | |
1324 | */ | |
1325 | STATIC int | |
1326 | xlog_clear_stale_blocks( | |
9a8d2fdb | 1327 | struct xlog *log, |
1da177e4 LT |
1328 | xfs_lsn_t tail_lsn) |
1329 | { | |
1330 | int tail_cycle, head_cycle; | |
1331 | int tail_block, head_block; | |
1332 | int tail_distance, max_distance; | |
1333 | int distance; | |
1334 | int error; | |
1335 | ||
1336 | tail_cycle = CYCLE_LSN(tail_lsn); | |
1337 | tail_block = BLOCK_LSN(tail_lsn); | |
1338 | head_cycle = log->l_curr_cycle; | |
1339 | head_block = log->l_curr_block; | |
1340 | ||
1341 | /* | |
1342 | * Figure out the distance between the new head of the log | |
1343 | * and the tail. We want to write over any blocks beyond the | |
1344 | * head that we may have written just before the crash, but | |
1345 | * we don't want to overwrite the tail of the log. | |
1346 | */ | |
1347 | if (head_cycle == tail_cycle) { | |
1348 | /* | |
1349 | * The tail is behind the head in the physical log, | |
1350 | * so the distance from the head to the tail is the | |
1351 | * distance from the head to the end of the log plus | |
1352 | * the distance from the beginning of the log to the | |
1353 | * tail. | |
1354 | */ | |
1355 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { | |
1356 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", | |
1357 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 1358 | return -EFSCORRUPTED; |
1da177e4 LT |
1359 | } |
1360 | tail_distance = tail_block + (log->l_logBBsize - head_block); | |
1361 | } else { | |
1362 | /* | |
1363 | * The head is behind the tail in the physical log, | |
1364 | * so the distance from the head to the tail is just | |
1365 | * the tail block minus the head block. | |
1366 | */ | |
1367 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ | |
1368 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", | |
1369 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 1370 | return -EFSCORRUPTED; |
1da177e4 LT |
1371 | } |
1372 | tail_distance = tail_block - head_block; | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * If the head is right up against the tail, we can't clear | |
1377 | * anything. | |
1378 | */ | |
1379 | if (tail_distance <= 0) { | |
1380 | ASSERT(tail_distance == 0); | |
1381 | return 0; | |
1382 | } | |
1383 | ||
1384 | max_distance = XLOG_TOTAL_REC_SHIFT(log); | |
1385 | /* | |
1386 | * Take the smaller of the maximum amount of outstanding I/O | |
1387 | * we could have and the distance to the tail to clear out. | |
1388 | * We take the smaller so that we don't overwrite the tail and | |
1389 | * we don't waste all day writing from the head to the tail | |
1390 | * for no reason. | |
1391 | */ | |
1392 | max_distance = MIN(max_distance, tail_distance); | |
1393 | ||
1394 | if ((head_block + max_distance) <= log->l_logBBsize) { | |
1395 | /* | |
1396 | * We can stomp all the blocks we need to without | |
1397 | * wrapping around the end of the log. Just do it | |
1398 | * in a single write. Use the cycle number of the | |
1399 | * current cycle minus one so that the log will look like: | |
1400 | * n ... | n - 1 ... | |
1401 | */ | |
1402 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1403 | head_block, max_distance, tail_cycle, | |
1404 | tail_block); | |
1405 | if (error) | |
1406 | return error; | |
1407 | } else { | |
1408 | /* | |
1409 | * We need to wrap around the end of the physical log in | |
1410 | * order to clear all the blocks. Do it in two separate | |
1411 | * I/Os. The first write should be from the head to the | |
1412 | * end of the physical log, and it should use the current | |
1413 | * cycle number minus one just like above. | |
1414 | */ | |
1415 | distance = log->l_logBBsize - head_block; | |
1416 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1417 | head_block, distance, tail_cycle, | |
1418 | tail_block); | |
1419 | ||
1420 | if (error) | |
1421 | return error; | |
1422 | ||
1423 | /* | |
1424 | * Now write the blocks at the start of the physical log. | |
1425 | * This writes the remainder of the blocks we want to clear. | |
1426 | * It uses the current cycle number since we're now on the | |
1427 | * same cycle as the head so that we get: | |
1428 | * n ... n ... | n - 1 ... | |
1429 | * ^^^^^ blocks we're writing | |
1430 | */ | |
1431 | distance = max_distance - (log->l_logBBsize - head_block); | |
1432 | error = xlog_write_log_records(log, head_cycle, 0, distance, | |
1433 | tail_cycle, tail_block); | |
1434 | if (error) | |
1435 | return error; | |
1436 | } | |
1437 | ||
1438 | return 0; | |
1439 | } | |
1440 | ||
1441 | /****************************************************************************** | |
1442 | * | |
1443 | * Log recover routines | |
1444 | * | |
1445 | ****************************************************************************** | |
1446 | */ | |
1447 | ||
f0a76953 | 1448 | /* |
a775ad77 DC |
1449 | * Sort the log items in the transaction. |
1450 | * | |
1451 | * The ordering constraints are defined by the inode allocation and unlink | |
1452 | * behaviour. The rules are: | |
1453 | * | |
1454 | * 1. Every item is only logged once in a given transaction. Hence it | |
1455 | * represents the last logged state of the item. Hence ordering is | |
1456 | * dependent on the order in which operations need to be performed so | |
1457 | * required initial conditions are always met. | |
1458 | * | |
1459 | * 2. Cancelled buffers are recorded in pass 1 in a separate table and | |
1460 | * there's nothing to replay from them so we can simply cull them | |
1461 | * from the transaction. However, we can't do that until after we've | |
1462 | * replayed all the other items because they may be dependent on the | |
1463 | * cancelled buffer and replaying the cancelled buffer can remove it | |
1464 | * form the cancelled buffer table. Hence they have tobe done last. | |
1465 | * | |
1466 | * 3. Inode allocation buffers must be replayed before inode items that | |
28c8e41a DC |
1467 | * read the buffer and replay changes into it. For filesystems using the |
1468 | * ICREATE transactions, this means XFS_LI_ICREATE objects need to get | |
1469 | * treated the same as inode allocation buffers as they create and | |
1470 | * initialise the buffers directly. | |
a775ad77 DC |
1471 | * |
1472 | * 4. Inode unlink buffers must be replayed after inode items are replayed. | |
1473 | * This ensures that inodes are completely flushed to the inode buffer | |
1474 | * in a "free" state before we remove the unlinked inode list pointer. | |
1475 | * | |
1476 | * Hence the ordering needs to be inode allocation buffers first, inode items | |
1477 | * second, inode unlink buffers third and cancelled buffers last. | |
1478 | * | |
1479 | * But there's a problem with that - we can't tell an inode allocation buffer | |
1480 | * apart from a regular buffer, so we can't separate them. We can, however, | |
1481 | * tell an inode unlink buffer from the others, and so we can separate them out | |
1482 | * from all the other buffers and move them to last. | |
1483 | * | |
1484 | * Hence, 4 lists, in order from head to tail: | |
28c8e41a DC |
1485 | * - buffer_list for all buffers except cancelled/inode unlink buffers |
1486 | * - item_list for all non-buffer items | |
1487 | * - inode_buffer_list for inode unlink buffers | |
1488 | * - cancel_list for the cancelled buffers | |
1489 | * | |
1490 | * Note that we add objects to the tail of the lists so that first-to-last | |
1491 | * ordering is preserved within the lists. Adding objects to the head of the | |
1492 | * list means when we traverse from the head we walk them in last-to-first | |
1493 | * order. For cancelled buffers and inode unlink buffers this doesn't matter, | |
1494 | * but for all other items there may be specific ordering that we need to | |
1495 | * preserve. | |
f0a76953 | 1496 | */ |
1da177e4 LT |
1497 | STATIC int |
1498 | xlog_recover_reorder_trans( | |
ad223e60 MT |
1499 | struct xlog *log, |
1500 | struct xlog_recover *trans, | |
9abbc539 | 1501 | int pass) |
1da177e4 | 1502 | { |
f0a76953 | 1503 | xlog_recover_item_t *item, *n; |
2a84108f | 1504 | int error = 0; |
f0a76953 | 1505 | LIST_HEAD(sort_list); |
a775ad77 DC |
1506 | LIST_HEAD(cancel_list); |
1507 | LIST_HEAD(buffer_list); | |
1508 | LIST_HEAD(inode_buffer_list); | |
1509 | LIST_HEAD(inode_list); | |
f0a76953 DC |
1510 | |
1511 | list_splice_init(&trans->r_itemq, &sort_list); | |
1512 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { | |
4e0d5f92 | 1513 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
1da177e4 | 1514 | |
f0a76953 | 1515 | switch (ITEM_TYPE(item)) { |
28c8e41a DC |
1516 | case XFS_LI_ICREATE: |
1517 | list_move_tail(&item->ri_list, &buffer_list); | |
1518 | break; | |
1da177e4 | 1519 | case XFS_LI_BUF: |
a775ad77 | 1520 | if (buf_f->blf_flags & XFS_BLF_CANCEL) { |
9abbc539 DC |
1521 | trace_xfs_log_recover_item_reorder_head(log, |
1522 | trans, item, pass); | |
a775ad77 | 1523 | list_move(&item->ri_list, &cancel_list); |
1da177e4 LT |
1524 | break; |
1525 | } | |
a775ad77 DC |
1526 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
1527 | list_move(&item->ri_list, &inode_buffer_list); | |
1528 | break; | |
1529 | } | |
1530 | list_move_tail(&item->ri_list, &buffer_list); | |
1531 | break; | |
1da177e4 | 1532 | case XFS_LI_INODE: |
1da177e4 LT |
1533 | case XFS_LI_DQUOT: |
1534 | case XFS_LI_QUOTAOFF: | |
1535 | case XFS_LI_EFD: | |
1536 | case XFS_LI_EFI: | |
9abbc539 DC |
1537 | trace_xfs_log_recover_item_reorder_tail(log, |
1538 | trans, item, pass); | |
a775ad77 | 1539 | list_move_tail(&item->ri_list, &inode_list); |
1da177e4 LT |
1540 | break; |
1541 | default: | |
a0fa2b67 DC |
1542 | xfs_warn(log->l_mp, |
1543 | "%s: unrecognized type of log operation", | |
1544 | __func__); | |
1da177e4 | 1545 | ASSERT(0); |
2a84108f MT |
1546 | /* |
1547 | * return the remaining items back to the transaction | |
1548 | * item list so they can be freed in caller. | |
1549 | */ | |
1550 | if (!list_empty(&sort_list)) | |
1551 | list_splice_init(&sort_list, &trans->r_itemq); | |
2451337d | 1552 | error = -EIO; |
2a84108f | 1553 | goto out; |
1da177e4 | 1554 | } |
f0a76953 | 1555 | } |
2a84108f | 1556 | out: |
f0a76953 | 1557 | ASSERT(list_empty(&sort_list)); |
a775ad77 DC |
1558 | if (!list_empty(&buffer_list)) |
1559 | list_splice(&buffer_list, &trans->r_itemq); | |
1560 | if (!list_empty(&inode_list)) | |
1561 | list_splice_tail(&inode_list, &trans->r_itemq); | |
1562 | if (!list_empty(&inode_buffer_list)) | |
1563 | list_splice_tail(&inode_buffer_list, &trans->r_itemq); | |
1564 | if (!list_empty(&cancel_list)) | |
1565 | list_splice_tail(&cancel_list, &trans->r_itemq); | |
2a84108f | 1566 | return error; |
1da177e4 LT |
1567 | } |
1568 | ||
1569 | /* | |
1570 | * Build up the table of buf cancel records so that we don't replay | |
1571 | * cancelled data in the second pass. For buffer records that are | |
1572 | * not cancel records, there is nothing to do here so we just return. | |
1573 | * | |
1574 | * If we get a cancel record which is already in the table, this indicates | |
1575 | * that the buffer was cancelled multiple times. In order to ensure | |
1576 | * that during pass 2 we keep the record in the table until we reach its | |
1577 | * last occurrence in the log, we keep a reference count in the cancel | |
1578 | * record in the table to tell us how many times we expect to see this | |
1579 | * record during the second pass. | |
1580 | */ | |
c9f71f5f CH |
1581 | STATIC int |
1582 | xlog_recover_buffer_pass1( | |
ad223e60 MT |
1583 | struct xlog *log, |
1584 | struct xlog_recover_item *item) | |
1da177e4 | 1585 | { |
c9f71f5f | 1586 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
d5689eaa CH |
1587 | struct list_head *bucket; |
1588 | struct xfs_buf_cancel *bcp; | |
1da177e4 LT |
1589 | |
1590 | /* | |
1591 | * If this isn't a cancel buffer item, then just return. | |
1592 | */ | |
e2714bf8 | 1593 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
9abbc539 | 1594 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); |
c9f71f5f | 1595 | return 0; |
9abbc539 | 1596 | } |
1da177e4 LT |
1597 | |
1598 | /* | |
d5689eaa CH |
1599 | * Insert an xfs_buf_cancel record into the hash table of them. |
1600 | * If there is already an identical record, bump its reference count. | |
1da177e4 | 1601 | */ |
d5689eaa CH |
1602 | bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno); |
1603 | list_for_each_entry(bcp, bucket, bc_list) { | |
1604 | if (bcp->bc_blkno == buf_f->blf_blkno && | |
1605 | bcp->bc_len == buf_f->blf_len) { | |
1606 | bcp->bc_refcount++; | |
9abbc539 | 1607 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
c9f71f5f | 1608 | return 0; |
1da177e4 | 1609 | } |
d5689eaa CH |
1610 | } |
1611 | ||
1612 | bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP); | |
1613 | bcp->bc_blkno = buf_f->blf_blkno; | |
1614 | bcp->bc_len = buf_f->blf_len; | |
1da177e4 | 1615 | bcp->bc_refcount = 1; |
d5689eaa CH |
1616 | list_add_tail(&bcp->bc_list, bucket); |
1617 | ||
9abbc539 | 1618 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
c9f71f5f | 1619 | return 0; |
1da177e4 LT |
1620 | } |
1621 | ||
1622 | /* | |
1623 | * Check to see whether the buffer being recovered has a corresponding | |
84a5b730 DC |
1624 | * entry in the buffer cancel record table. If it is, return the cancel |
1625 | * buffer structure to the caller. | |
1da177e4 | 1626 | */ |
84a5b730 DC |
1627 | STATIC struct xfs_buf_cancel * |
1628 | xlog_peek_buffer_cancelled( | |
ad223e60 | 1629 | struct xlog *log, |
1da177e4 LT |
1630 | xfs_daddr_t blkno, |
1631 | uint len, | |
1632 | ushort flags) | |
1633 | { | |
d5689eaa CH |
1634 | struct list_head *bucket; |
1635 | struct xfs_buf_cancel *bcp; | |
1da177e4 | 1636 | |
84a5b730 DC |
1637 | if (!log->l_buf_cancel_table) { |
1638 | /* empty table means no cancelled buffers in the log */ | |
c1155410 | 1639 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
84a5b730 | 1640 | return NULL; |
1da177e4 LT |
1641 | } |
1642 | ||
d5689eaa CH |
1643 | bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno); |
1644 | list_for_each_entry(bcp, bucket, bc_list) { | |
1645 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) | |
84a5b730 | 1646 | return bcp; |
1da177e4 | 1647 | } |
d5689eaa | 1648 | |
1da177e4 | 1649 | /* |
d5689eaa CH |
1650 | * We didn't find a corresponding entry in the table, so return 0 so |
1651 | * that the buffer is NOT cancelled. | |
1da177e4 | 1652 | */ |
c1155410 | 1653 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
84a5b730 DC |
1654 | return NULL; |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * If the buffer is being cancelled then return 1 so that it will be cancelled, | |
1659 | * otherwise return 0. If the buffer is actually a buffer cancel item | |
1660 | * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the | |
1661 | * table and remove it from the table if this is the last reference. | |
1662 | * | |
1663 | * We remove the cancel record from the table when we encounter its last | |
1664 | * occurrence in the log so that if the same buffer is re-used again after its | |
1665 | * last cancellation we actually replay the changes made at that point. | |
1666 | */ | |
1667 | STATIC int | |
1668 | xlog_check_buffer_cancelled( | |
1669 | struct xlog *log, | |
1670 | xfs_daddr_t blkno, | |
1671 | uint len, | |
1672 | ushort flags) | |
1673 | { | |
1674 | struct xfs_buf_cancel *bcp; | |
1675 | ||
1676 | bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags); | |
1677 | if (!bcp) | |
1678 | return 0; | |
d5689eaa | 1679 | |
d5689eaa CH |
1680 | /* |
1681 | * We've go a match, so return 1 so that the recovery of this buffer | |
1682 | * is cancelled. If this buffer is actually a buffer cancel log | |
1683 | * item, then decrement the refcount on the one in the table and | |
1684 | * remove it if this is the last reference. | |
1685 | */ | |
1686 | if (flags & XFS_BLF_CANCEL) { | |
1687 | if (--bcp->bc_refcount == 0) { | |
1688 | list_del(&bcp->bc_list); | |
1689 | kmem_free(bcp); | |
1690 | } | |
1691 | } | |
1692 | return 1; | |
1da177e4 LT |
1693 | } |
1694 | ||
1da177e4 | 1695 | /* |
e2714bf8 CH |
1696 | * Perform recovery for a buffer full of inodes. In these buffers, the only |
1697 | * data which should be recovered is that which corresponds to the | |
1698 | * di_next_unlinked pointers in the on disk inode structures. The rest of the | |
1699 | * data for the inodes is always logged through the inodes themselves rather | |
1700 | * than the inode buffer and is recovered in xlog_recover_inode_pass2(). | |
1da177e4 | 1701 | * |
e2714bf8 CH |
1702 | * The only time when buffers full of inodes are fully recovered is when the |
1703 | * buffer is full of newly allocated inodes. In this case the buffer will | |
1704 | * not be marked as an inode buffer and so will be sent to | |
1705 | * xlog_recover_do_reg_buffer() below during recovery. | |
1da177e4 LT |
1706 | */ |
1707 | STATIC int | |
1708 | xlog_recover_do_inode_buffer( | |
e2714bf8 | 1709 | struct xfs_mount *mp, |
1da177e4 | 1710 | xlog_recover_item_t *item, |
e2714bf8 | 1711 | struct xfs_buf *bp, |
1da177e4 LT |
1712 | xfs_buf_log_format_t *buf_f) |
1713 | { | |
1714 | int i; | |
e2714bf8 CH |
1715 | int item_index = 0; |
1716 | int bit = 0; | |
1717 | int nbits = 0; | |
1718 | int reg_buf_offset = 0; | |
1719 | int reg_buf_bytes = 0; | |
1da177e4 LT |
1720 | int next_unlinked_offset; |
1721 | int inodes_per_buf; | |
1722 | xfs_agino_t *logged_nextp; | |
1723 | xfs_agino_t *buffer_nextp; | |
1da177e4 | 1724 | |
9abbc539 | 1725 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
9222a9cf DC |
1726 | |
1727 | /* | |
1728 | * Post recovery validation only works properly on CRC enabled | |
1729 | * filesystems. | |
1730 | */ | |
1731 | if (xfs_sb_version_hascrc(&mp->m_sb)) | |
1732 | bp->b_ops = &xfs_inode_buf_ops; | |
9abbc539 | 1733 | |
aa0e8833 | 1734 | inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog; |
1da177e4 LT |
1735 | for (i = 0; i < inodes_per_buf; i++) { |
1736 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + | |
1737 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1738 | ||
1739 | while (next_unlinked_offset >= | |
1740 | (reg_buf_offset + reg_buf_bytes)) { | |
1741 | /* | |
1742 | * The next di_next_unlinked field is beyond | |
1743 | * the current logged region. Find the next | |
1744 | * logged region that contains or is beyond | |
1745 | * the current di_next_unlinked field. | |
1746 | */ | |
1747 | bit += nbits; | |
e2714bf8 CH |
1748 | bit = xfs_next_bit(buf_f->blf_data_map, |
1749 | buf_f->blf_map_size, bit); | |
1da177e4 LT |
1750 | |
1751 | /* | |
1752 | * If there are no more logged regions in the | |
1753 | * buffer, then we're done. | |
1754 | */ | |
e2714bf8 | 1755 | if (bit == -1) |
1da177e4 | 1756 | return 0; |
1da177e4 | 1757 | |
e2714bf8 CH |
1758 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
1759 | buf_f->blf_map_size, bit); | |
1da177e4 | 1760 | ASSERT(nbits > 0); |
c1155410 DC |
1761 | reg_buf_offset = bit << XFS_BLF_SHIFT; |
1762 | reg_buf_bytes = nbits << XFS_BLF_SHIFT; | |
1da177e4 LT |
1763 | item_index++; |
1764 | } | |
1765 | ||
1766 | /* | |
1767 | * If the current logged region starts after the current | |
1768 | * di_next_unlinked field, then move on to the next | |
1769 | * di_next_unlinked field. | |
1770 | */ | |
e2714bf8 | 1771 | if (next_unlinked_offset < reg_buf_offset) |
1da177e4 | 1772 | continue; |
1da177e4 LT |
1773 | |
1774 | ASSERT(item->ri_buf[item_index].i_addr != NULL); | |
c1155410 | 1775 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0); |
aa0e8833 DC |
1776 | ASSERT((reg_buf_offset + reg_buf_bytes) <= |
1777 | BBTOB(bp->b_io_length)); | |
1da177e4 LT |
1778 | |
1779 | /* | |
1780 | * The current logged region contains a copy of the | |
1781 | * current di_next_unlinked field. Extract its value | |
1782 | * and copy it to the buffer copy. | |
1783 | */ | |
4e0d5f92 CH |
1784 | logged_nextp = item->ri_buf[item_index].i_addr + |
1785 | next_unlinked_offset - reg_buf_offset; | |
1da177e4 | 1786 | if (unlikely(*logged_nextp == 0)) { |
a0fa2b67 DC |
1787 | xfs_alert(mp, |
1788 | "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). " | |
1789 | "Trying to replay bad (0) inode di_next_unlinked field.", | |
1da177e4 LT |
1790 | item, bp); |
1791 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", | |
1792 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 1793 | return -EFSCORRUPTED; |
1da177e4 LT |
1794 | } |
1795 | ||
1796 | buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp, | |
1797 | next_unlinked_offset); | |
87c199c2 | 1798 | *buffer_nextp = *logged_nextp; |
0a32c26e DC |
1799 | |
1800 | /* | |
1801 | * If necessary, recalculate the CRC in the on-disk inode. We | |
1802 | * have to leave the inode in a consistent state for whoever | |
1803 | * reads it next.... | |
1804 | */ | |
1805 | xfs_dinode_calc_crc(mp, (struct xfs_dinode *) | |
1806 | xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize)); | |
1807 | ||
1da177e4 LT |
1808 | } |
1809 | ||
1810 | return 0; | |
1811 | } | |
1812 | ||
50d5c8d8 DC |
1813 | /* |
1814 | * V5 filesystems know the age of the buffer on disk being recovered. We can | |
1815 | * have newer objects on disk than we are replaying, and so for these cases we | |
1816 | * don't want to replay the current change as that will make the buffer contents | |
1817 | * temporarily invalid on disk. | |
1818 | * | |
1819 | * The magic number might not match the buffer type we are going to recover | |
1820 | * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence | |
1821 | * extract the LSN of the existing object in the buffer based on it's current | |
1822 | * magic number. If we don't recognise the magic number in the buffer, then | |
1823 | * return a LSN of -1 so that the caller knows it was an unrecognised block and | |
1824 | * so can recover the buffer. | |
566055d3 DC |
1825 | * |
1826 | * Note: we cannot rely solely on magic number matches to determine that the | |
1827 | * buffer has a valid LSN - we also need to verify that it belongs to this | |
1828 | * filesystem, so we need to extract the object's LSN and compare it to that | |
1829 | * which we read from the superblock. If the UUIDs don't match, then we've got a | |
1830 | * stale metadata block from an old filesystem instance that we need to recover | |
1831 | * over the top of. | |
50d5c8d8 DC |
1832 | */ |
1833 | static xfs_lsn_t | |
1834 | xlog_recover_get_buf_lsn( | |
1835 | struct xfs_mount *mp, | |
1836 | struct xfs_buf *bp) | |
1837 | { | |
1838 | __uint32_t magic32; | |
1839 | __uint16_t magic16; | |
1840 | __uint16_t magicda; | |
1841 | void *blk = bp->b_addr; | |
566055d3 DC |
1842 | uuid_t *uuid; |
1843 | xfs_lsn_t lsn = -1; | |
50d5c8d8 DC |
1844 | |
1845 | /* v4 filesystems always recover immediately */ | |
1846 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
1847 | goto recover_immediately; | |
1848 | ||
1849 | magic32 = be32_to_cpu(*(__be32 *)blk); | |
1850 | switch (magic32) { | |
1851 | case XFS_ABTB_CRC_MAGIC: | |
1852 | case XFS_ABTC_CRC_MAGIC: | |
1853 | case XFS_ABTB_MAGIC: | |
1854 | case XFS_ABTC_MAGIC: | |
1855 | case XFS_IBT_CRC_MAGIC: | |
566055d3 DC |
1856 | case XFS_IBT_MAGIC: { |
1857 | struct xfs_btree_block *btb = blk; | |
1858 | ||
1859 | lsn = be64_to_cpu(btb->bb_u.s.bb_lsn); | |
1860 | uuid = &btb->bb_u.s.bb_uuid; | |
1861 | break; | |
1862 | } | |
50d5c8d8 | 1863 | case XFS_BMAP_CRC_MAGIC: |
566055d3 DC |
1864 | case XFS_BMAP_MAGIC: { |
1865 | struct xfs_btree_block *btb = blk; | |
1866 | ||
1867 | lsn = be64_to_cpu(btb->bb_u.l.bb_lsn); | |
1868 | uuid = &btb->bb_u.l.bb_uuid; | |
1869 | break; | |
1870 | } | |
50d5c8d8 | 1871 | case XFS_AGF_MAGIC: |
566055d3 DC |
1872 | lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn); |
1873 | uuid = &((struct xfs_agf *)blk)->agf_uuid; | |
1874 | break; | |
50d5c8d8 | 1875 | case XFS_AGFL_MAGIC: |
566055d3 DC |
1876 | lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn); |
1877 | uuid = &((struct xfs_agfl *)blk)->agfl_uuid; | |
1878 | break; | |
50d5c8d8 | 1879 | case XFS_AGI_MAGIC: |
566055d3 DC |
1880 | lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn); |
1881 | uuid = &((struct xfs_agi *)blk)->agi_uuid; | |
1882 | break; | |
50d5c8d8 | 1883 | case XFS_SYMLINK_MAGIC: |
566055d3 DC |
1884 | lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn); |
1885 | uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid; | |
1886 | break; | |
50d5c8d8 DC |
1887 | case XFS_DIR3_BLOCK_MAGIC: |
1888 | case XFS_DIR3_DATA_MAGIC: | |
1889 | case XFS_DIR3_FREE_MAGIC: | |
566055d3 DC |
1890 | lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn); |
1891 | uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid; | |
1892 | break; | |
50d5c8d8 | 1893 | case XFS_ATTR3_RMT_MAGIC: |
566055d3 DC |
1894 | lsn = be64_to_cpu(((struct xfs_attr3_rmt_hdr *)blk)->rm_lsn); |
1895 | uuid = &((struct xfs_attr3_rmt_hdr *)blk)->rm_uuid; | |
1896 | break; | |
50d5c8d8 | 1897 | case XFS_SB_MAGIC: |
566055d3 DC |
1898 | lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn); |
1899 | uuid = &((struct xfs_dsb *)blk)->sb_uuid; | |
1900 | break; | |
50d5c8d8 DC |
1901 | default: |
1902 | break; | |
1903 | } | |
1904 | ||
566055d3 DC |
1905 | if (lsn != (xfs_lsn_t)-1) { |
1906 | if (!uuid_equal(&mp->m_sb.sb_uuid, uuid)) | |
1907 | goto recover_immediately; | |
1908 | return lsn; | |
1909 | } | |
1910 | ||
50d5c8d8 DC |
1911 | magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic); |
1912 | switch (magicda) { | |
1913 | case XFS_DIR3_LEAF1_MAGIC: | |
1914 | case XFS_DIR3_LEAFN_MAGIC: | |
1915 | case XFS_DA3_NODE_MAGIC: | |
566055d3 DC |
1916 | lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn); |
1917 | uuid = &((struct xfs_da3_blkinfo *)blk)->uuid; | |
1918 | break; | |
50d5c8d8 DC |
1919 | default: |
1920 | break; | |
1921 | } | |
1922 | ||
566055d3 DC |
1923 | if (lsn != (xfs_lsn_t)-1) { |
1924 | if (!uuid_equal(&mp->m_sb.sb_uuid, uuid)) | |
1925 | goto recover_immediately; | |
1926 | return lsn; | |
1927 | } | |
1928 | ||
50d5c8d8 DC |
1929 | /* |
1930 | * We do individual object checks on dquot and inode buffers as they | |
1931 | * have their own individual LSN records. Also, we could have a stale | |
1932 | * buffer here, so we have to at least recognise these buffer types. | |
1933 | * | |
1934 | * A notd complexity here is inode unlinked list processing - it logs | |
1935 | * the inode directly in the buffer, but we don't know which inodes have | |
1936 | * been modified, and there is no global buffer LSN. Hence we need to | |
1937 | * recover all inode buffer types immediately. This problem will be | |
1938 | * fixed by logical logging of the unlinked list modifications. | |
1939 | */ | |
1940 | magic16 = be16_to_cpu(*(__be16 *)blk); | |
1941 | switch (magic16) { | |
1942 | case XFS_DQUOT_MAGIC: | |
1943 | case XFS_DINODE_MAGIC: | |
1944 | goto recover_immediately; | |
1945 | default: | |
1946 | break; | |
1947 | } | |
1948 | ||
1949 | /* unknown buffer contents, recover immediately */ | |
1950 | ||
1951 | recover_immediately: | |
1952 | return (xfs_lsn_t)-1; | |
1953 | ||
1954 | } | |
1955 | ||
1da177e4 | 1956 | /* |
d75afeb3 DC |
1957 | * Validate the recovered buffer is of the correct type and attach the |
1958 | * appropriate buffer operations to them for writeback. Magic numbers are in a | |
1959 | * few places: | |
1960 | * the first 16 bits of the buffer (inode buffer, dquot buffer), | |
1961 | * the first 32 bits of the buffer (most blocks), | |
1962 | * inside a struct xfs_da_blkinfo at the start of the buffer. | |
1da177e4 | 1963 | */ |
d75afeb3 | 1964 | static void |
50d5c8d8 | 1965 | xlog_recover_validate_buf_type( |
9abbc539 | 1966 | struct xfs_mount *mp, |
e2714bf8 | 1967 | struct xfs_buf *bp, |
1da177e4 LT |
1968 | xfs_buf_log_format_t *buf_f) |
1969 | { | |
d75afeb3 DC |
1970 | struct xfs_da_blkinfo *info = bp->b_addr; |
1971 | __uint32_t magic32; | |
1972 | __uint16_t magic16; | |
1973 | __uint16_t magicda; | |
1974 | ||
67dc288c DC |
1975 | /* |
1976 | * We can only do post recovery validation on items on CRC enabled | |
1977 | * fielsystems as we need to know when the buffer was written to be able | |
1978 | * to determine if we should have replayed the item. If we replay old | |
1979 | * metadata over a newer buffer, then it will enter a temporarily | |
1980 | * inconsistent state resulting in verification failures. Hence for now | |
1981 | * just avoid the verification stage for non-crc filesystems | |
1982 | */ | |
1983 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
1984 | return; | |
1985 | ||
d75afeb3 DC |
1986 | magic32 = be32_to_cpu(*(__be32 *)bp->b_addr); |
1987 | magic16 = be16_to_cpu(*(__be16*)bp->b_addr); | |
1988 | magicda = be16_to_cpu(info->magic); | |
61fe135c DC |
1989 | switch (xfs_blft_from_flags(buf_f)) { |
1990 | case XFS_BLFT_BTREE_BUF: | |
d75afeb3 | 1991 | switch (magic32) { |
ee1a47ab CH |
1992 | case XFS_ABTB_CRC_MAGIC: |
1993 | case XFS_ABTC_CRC_MAGIC: | |
1994 | case XFS_ABTB_MAGIC: | |
1995 | case XFS_ABTC_MAGIC: | |
1996 | bp->b_ops = &xfs_allocbt_buf_ops; | |
1997 | break; | |
1998 | case XFS_IBT_CRC_MAGIC: | |
aafc3c24 | 1999 | case XFS_FIBT_CRC_MAGIC: |
ee1a47ab | 2000 | case XFS_IBT_MAGIC: |
aafc3c24 | 2001 | case XFS_FIBT_MAGIC: |
ee1a47ab CH |
2002 | bp->b_ops = &xfs_inobt_buf_ops; |
2003 | break; | |
2004 | case XFS_BMAP_CRC_MAGIC: | |
2005 | case XFS_BMAP_MAGIC: | |
2006 | bp->b_ops = &xfs_bmbt_buf_ops; | |
2007 | break; | |
2008 | default: | |
2009 | xfs_warn(mp, "Bad btree block magic!"); | |
2010 | ASSERT(0); | |
2011 | break; | |
2012 | } | |
2013 | break; | |
61fe135c | 2014 | case XFS_BLFT_AGF_BUF: |
d75afeb3 | 2015 | if (magic32 != XFS_AGF_MAGIC) { |
4e0e6040 DC |
2016 | xfs_warn(mp, "Bad AGF block magic!"); |
2017 | ASSERT(0); | |
2018 | break; | |
2019 | } | |
2020 | bp->b_ops = &xfs_agf_buf_ops; | |
2021 | break; | |
61fe135c | 2022 | case XFS_BLFT_AGFL_BUF: |
d75afeb3 | 2023 | if (magic32 != XFS_AGFL_MAGIC) { |
77c95bba CH |
2024 | xfs_warn(mp, "Bad AGFL block magic!"); |
2025 | ASSERT(0); | |
2026 | break; | |
2027 | } | |
2028 | bp->b_ops = &xfs_agfl_buf_ops; | |
2029 | break; | |
61fe135c | 2030 | case XFS_BLFT_AGI_BUF: |
d75afeb3 | 2031 | if (magic32 != XFS_AGI_MAGIC) { |
983d09ff DC |
2032 | xfs_warn(mp, "Bad AGI block magic!"); |
2033 | ASSERT(0); | |
2034 | break; | |
2035 | } | |
2036 | bp->b_ops = &xfs_agi_buf_ops; | |
2037 | break; | |
61fe135c DC |
2038 | case XFS_BLFT_UDQUOT_BUF: |
2039 | case XFS_BLFT_PDQUOT_BUF: | |
2040 | case XFS_BLFT_GDQUOT_BUF: | |
123887e8 | 2041 | #ifdef CONFIG_XFS_QUOTA |
d75afeb3 | 2042 | if (magic16 != XFS_DQUOT_MAGIC) { |
3fe58f30 CH |
2043 | xfs_warn(mp, "Bad DQUOT block magic!"); |
2044 | ASSERT(0); | |
2045 | break; | |
2046 | } | |
2047 | bp->b_ops = &xfs_dquot_buf_ops; | |
123887e8 DC |
2048 | #else |
2049 | xfs_alert(mp, | |
2050 | "Trying to recover dquots without QUOTA support built in!"); | |
2051 | ASSERT(0); | |
2052 | #endif | |
3fe58f30 | 2053 | break; |
61fe135c | 2054 | case XFS_BLFT_DINO_BUF: |
d75afeb3 | 2055 | if (magic16 != XFS_DINODE_MAGIC) { |
93848a99 CH |
2056 | xfs_warn(mp, "Bad INODE block magic!"); |
2057 | ASSERT(0); | |
2058 | break; | |
2059 | } | |
2060 | bp->b_ops = &xfs_inode_buf_ops; | |
2061 | break; | |
61fe135c | 2062 | case XFS_BLFT_SYMLINK_BUF: |
d75afeb3 | 2063 | if (magic32 != XFS_SYMLINK_MAGIC) { |
f948dd76 DC |
2064 | xfs_warn(mp, "Bad symlink block magic!"); |
2065 | ASSERT(0); | |
2066 | break; | |
2067 | } | |
2068 | bp->b_ops = &xfs_symlink_buf_ops; | |
2069 | break; | |
61fe135c | 2070 | case XFS_BLFT_DIR_BLOCK_BUF: |
d75afeb3 DC |
2071 | if (magic32 != XFS_DIR2_BLOCK_MAGIC && |
2072 | magic32 != XFS_DIR3_BLOCK_MAGIC) { | |
2073 | xfs_warn(mp, "Bad dir block magic!"); | |
2074 | ASSERT(0); | |
2075 | break; | |
2076 | } | |
2077 | bp->b_ops = &xfs_dir3_block_buf_ops; | |
2078 | break; | |
61fe135c | 2079 | case XFS_BLFT_DIR_DATA_BUF: |
d75afeb3 DC |
2080 | if (magic32 != XFS_DIR2_DATA_MAGIC && |
2081 | magic32 != XFS_DIR3_DATA_MAGIC) { | |
2082 | xfs_warn(mp, "Bad dir data magic!"); | |
2083 | ASSERT(0); | |
2084 | break; | |
2085 | } | |
2086 | bp->b_ops = &xfs_dir3_data_buf_ops; | |
2087 | break; | |
61fe135c | 2088 | case XFS_BLFT_DIR_FREE_BUF: |
d75afeb3 DC |
2089 | if (magic32 != XFS_DIR2_FREE_MAGIC && |
2090 | magic32 != XFS_DIR3_FREE_MAGIC) { | |
2091 | xfs_warn(mp, "Bad dir3 free magic!"); | |
2092 | ASSERT(0); | |
2093 | break; | |
2094 | } | |
2095 | bp->b_ops = &xfs_dir3_free_buf_ops; | |
2096 | break; | |
61fe135c | 2097 | case XFS_BLFT_DIR_LEAF1_BUF: |
d75afeb3 DC |
2098 | if (magicda != XFS_DIR2_LEAF1_MAGIC && |
2099 | magicda != XFS_DIR3_LEAF1_MAGIC) { | |
2100 | xfs_warn(mp, "Bad dir leaf1 magic!"); | |
2101 | ASSERT(0); | |
2102 | break; | |
2103 | } | |
2104 | bp->b_ops = &xfs_dir3_leaf1_buf_ops; | |
2105 | break; | |
61fe135c | 2106 | case XFS_BLFT_DIR_LEAFN_BUF: |
d75afeb3 DC |
2107 | if (magicda != XFS_DIR2_LEAFN_MAGIC && |
2108 | magicda != XFS_DIR3_LEAFN_MAGIC) { | |
2109 | xfs_warn(mp, "Bad dir leafn magic!"); | |
2110 | ASSERT(0); | |
2111 | break; | |
2112 | } | |
2113 | bp->b_ops = &xfs_dir3_leafn_buf_ops; | |
2114 | break; | |
61fe135c | 2115 | case XFS_BLFT_DA_NODE_BUF: |
d75afeb3 DC |
2116 | if (magicda != XFS_DA_NODE_MAGIC && |
2117 | magicda != XFS_DA3_NODE_MAGIC) { | |
2118 | xfs_warn(mp, "Bad da node magic!"); | |
2119 | ASSERT(0); | |
2120 | break; | |
2121 | } | |
2122 | bp->b_ops = &xfs_da3_node_buf_ops; | |
2123 | break; | |
61fe135c | 2124 | case XFS_BLFT_ATTR_LEAF_BUF: |
d75afeb3 DC |
2125 | if (magicda != XFS_ATTR_LEAF_MAGIC && |
2126 | magicda != XFS_ATTR3_LEAF_MAGIC) { | |
2127 | xfs_warn(mp, "Bad attr leaf magic!"); | |
2128 | ASSERT(0); | |
2129 | break; | |
2130 | } | |
2131 | bp->b_ops = &xfs_attr3_leaf_buf_ops; | |
2132 | break; | |
61fe135c | 2133 | case XFS_BLFT_ATTR_RMT_BUF: |
cab09a81 | 2134 | if (magic32 != XFS_ATTR3_RMT_MAGIC) { |
d75afeb3 DC |
2135 | xfs_warn(mp, "Bad attr remote magic!"); |
2136 | ASSERT(0); | |
2137 | break; | |
2138 | } | |
2139 | bp->b_ops = &xfs_attr3_rmt_buf_ops; | |
2140 | break; | |
04a1e6c5 DC |
2141 | case XFS_BLFT_SB_BUF: |
2142 | if (magic32 != XFS_SB_MAGIC) { | |
2143 | xfs_warn(mp, "Bad SB block magic!"); | |
2144 | ASSERT(0); | |
2145 | break; | |
2146 | } | |
2147 | bp->b_ops = &xfs_sb_buf_ops; | |
2148 | break; | |
ee1a47ab | 2149 | default: |
61fe135c DC |
2150 | xfs_warn(mp, "Unknown buffer type %d!", |
2151 | xfs_blft_from_flags(buf_f)); | |
ee1a47ab CH |
2152 | break; |
2153 | } | |
1da177e4 LT |
2154 | } |
2155 | ||
d75afeb3 DC |
2156 | /* |
2157 | * Perform a 'normal' buffer recovery. Each logged region of the | |
2158 | * buffer should be copied over the corresponding region in the | |
2159 | * given buffer. The bitmap in the buf log format structure indicates | |
2160 | * where to place the logged data. | |
2161 | */ | |
2162 | STATIC void | |
2163 | xlog_recover_do_reg_buffer( | |
2164 | struct xfs_mount *mp, | |
2165 | xlog_recover_item_t *item, | |
2166 | struct xfs_buf *bp, | |
2167 | xfs_buf_log_format_t *buf_f) | |
2168 | { | |
2169 | int i; | |
2170 | int bit; | |
2171 | int nbits; | |
2172 | int error; | |
2173 | ||
2174 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); | |
2175 | ||
2176 | bit = 0; | |
2177 | i = 1; /* 0 is the buf format structure */ | |
2178 | while (1) { | |
2179 | bit = xfs_next_bit(buf_f->blf_data_map, | |
2180 | buf_f->blf_map_size, bit); | |
2181 | if (bit == -1) | |
2182 | break; | |
2183 | nbits = xfs_contig_bits(buf_f->blf_data_map, | |
2184 | buf_f->blf_map_size, bit); | |
2185 | ASSERT(nbits > 0); | |
2186 | ASSERT(item->ri_buf[i].i_addr != NULL); | |
2187 | ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0); | |
2188 | ASSERT(BBTOB(bp->b_io_length) >= | |
2189 | ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT)); | |
2190 | ||
709da6a6 DC |
2191 | /* |
2192 | * The dirty regions logged in the buffer, even though | |
2193 | * contiguous, may span multiple chunks. This is because the | |
2194 | * dirty region may span a physical page boundary in a buffer | |
2195 | * and hence be split into two separate vectors for writing into | |
2196 | * the log. Hence we need to trim nbits back to the length of | |
2197 | * the current region being copied out of the log. | |
2198 | */ | |
2199 | if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT)) | |
2200 | nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT; | |
2201 | ||
d75afeb3 DC |
2202 | /* |
2203 | * Do a sanity check if this is a dquot buffer. Just checking | |
2204 | * the first dquot in the buffer should do. XXXThis is | |
2205 | * probably a good thing to do for other buf types also. | |
2206 | */ | |
2207 | error = 0; | |
2208 | if (buf_f->blf_flags & | |
2209 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { | |
2210 | if (item->ri_buf[i].i_addr == NULL) { | |
2211 | xfs_alert(mp, | |
2212 | "XFS: NULL dquot in %s.", __func__); | |
2213 | goto next; | |
2214 | } | |
2215 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { | |
2216 | xfs_alert(mp, | |
2217 | "XFS: dquot too small (%d) in %s.", | |
2218 | item->ri_buf[i].i_len, __func__); | |
2219 | goto next; | |
2220 | } | |
9aede1d8 | 2221 | error = xfs_dqcheck(mp, item->ri_buf[i].i_addr, |
d75afeb3 DC |
2222 | -1, 0, XFS_QMOPT_DOWARN, |
2223 | "dquot_buf_recover"); | |
2224 | if (error) | |
2225 | goto next; | |
2226 | } | |
2227 | ||
2228 | memcpy(xfs_buf_offset(bp, | |
2229 | (uint)bit << XFS_BLF_SHIFT), /* dest */ | |
2230 | item->ri_buf[i].i_addr, /* source */ | |
2231 | nbits<<XFS_BLF_SHIFT); /* length */ | |
2232 | next: | |
2233 | i++; | |
2234 | bit += nbits; | |
2235 | } | |
2236 | ||
2237 | /* Shouldn't be any more regions */ | |
2238 | ASSERT(i == item->ri_total); | |
2239 | ||
67dc288c | 2240 | xlog_recover_validate_buf_type(mp, bp, buf_f); |
d75afeb3 DC |
2241 | } |
2242 | ||
1da177e4 LT |
2243 | /* |
2244 | * Perform a dquot buffer recovery. | |
8ba701ee | 2245 | * Simple algorithm: if we have found a QUOTAOFF log item of the same type |
1da177e4 LT |
2246 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. |
2247 | * Else, treat it as a regular buffer and do recovery. | |
ad3714b8 DC |
2248 | * |
2249 | * Return false if the buffer was tossed and true if we recovered the buffer to | |
2250 | * indicate to the caller if the buffer needs writing. | |
1da177e4 | 2251 | */ |
ad3714b8 | 2252 | STATIC bool |
1da177e4 | 2253 | xlog_recover_do_dquot_buffer( |
9a8d2fdb MT |
2254 | struct xfs_mount *mp, |
2255 | struct xlog *log, | |
2256 | struct xlog_recover_item *item, | |
2257 | struct xfs_buf *bp, | |
2258 | struct xfs_buf_log_format *buf_f) | |
1da177e4 LT |
2259 | { |
2260 | uint type; | |
2261 | ||
9abbc539 DC |
2262 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
2263 | ||
1da177e4 LT |
2264 | /* |
2265 | * Filesystems are required to send in quota flags at mount time. | |
2266 | */ | |
ad3714b8 DC |
2267 | if (!mp->m_qflags) |
2268 | return false; | |
1da177e4 LT |
2269 | |
2270 | type = 0; | |
c1155410 | 2271 | if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF) |
1da177e4 | 2272 | type |= XFS_DQ_USER; |
c1155410 | 2273 | if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF) |
c8ad20ff | 2274 | type |= XFS_DQ_PROJ; |
c1155410 | 2275 | if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF) |
1da177e4 LT |
2276 | type |= XFS_DQ_GROUP; |
2277 | /* | |
2278 | * This type of quotas was turned off, so ignore this buffer | |
2279 | */ | |
2280 | if (log->l_quotaoffs_flag & type) | |
ad3714b8 | 2281 | return false; |
1da177e4 | 2282 | |
9abbc539 | 2283 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
ad3714b8 | 2284 | return true; |
1da177e4 LT |
2285 | } |
2286 | ||
2287 | /* | |
2288 | * This routine replays a modification made to a buffer at runtime. | |
2289 | * There are actually two types of buffer, regular and inode, which | |
2290 | * are handled differently. Inode buffers are handled differently | |
2291 | * in that we only recover a specific set of data from them, namely | |
2292 | * the inode di_next_unlinked fields. This is because all other inode | |
2293 | * data is actually logged via inode records and any data we replay | |
2294 | * here which overlaps that may be stale. | |
2295 | * | |
2296 | * When meta-data buffers are freed at run time we log a buffer item | |
c1155410 | 2297 | * with the XFS_BLF_CANCEL bit set to indicate that previous copies |
1da177e4 LT |
2298 | * of the buffer in the log should not be replayed at recovery time. |
2299 | * This is so that if the blocks covered by the buffer are reused for | |
2300 | * file data before we crash we don't end up replaying old, freed | |
2301 | * meta-data into a user's file. | |
2302 | * | |
2303 | * To handle the cancellation of buffer log items, we make two passes | |
2304 | * over the log during recovery. During the first we build a table of | |
2305 | * those buffers which have been cancelled, and during the second we | |
2306 | * only replay those buffers which do not have corresponding cancel | |
34be5ff3 | 2307 | * records in the table. See xlog_recover_buffer_pass[1,2] above |
1da177e4 LT |
2308 | * for more details on the implementation of the table of cancel records. |
2309 | */ | |
2310 | STATIC int | |
c9f71f5f | 2311 | xlog_recover_buffer_pass2( |
9a8d2fdb MT |
2312 | struct xlog *log, |
2313 | struct list_head *buffer_list, | |
50d5c8d8 DC |
2314 | struct xlog_recover_item *item, |
2315 | xfs_lsn_t current_lsn) | |
1da177e4 | 2316 | { |
4e0d5f92 | 2317 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
e2714bf8 | 2318 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2319 | xfs_buf_t *bp; |
2320 | int error; | |
6ad112bf | 2321 | uint buf_flags; |
50d5c8d8 | 2322 | xfs_lsn_t lsn; |
1da177e4 | 2323 | |
c9f71f5f CH |
2324 | /* |
2325 | * In this pass we only want to recover all the buffers which have | |
2326 | * not been cancelled and are not cancellation buffers themselves. | |
2327 | */ | |
2328 | if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno, | |
2329 | buf_f->blf_len, buf_f->blf_flags)) { | |
2330 | trace_xfs_log_recover_buf_cancel(log, buf_f); | |
1da177e4 | 2331 | return 0; |
1da177e4 | 2332 | } |
c9f71f5f | 2333 | |
9abbc539 | 2334 | trace_xfs_log_recover_buf_recover(log, buf_f); |
1da177e4 | 2335 | |
a8acad70 | 2336 | buf_flags = 0; |
611c9946 DC |
2337 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) |
2338 | buf_flags |= XBF_UNMAPPED; | |
6ad112bf | 2339 | |
e2714bf8 | 2340 | bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len, |
c3f8fc73 | 2341 | buf_flags, NULL); |
ac4d6888 | 2342 | if (!bp) |
2451337d | 2343 | return -ENOMEM; |
e5702805 | 2344 | error = bp->b_error; |
5a52c2a5 | 2345 | if (error) { |
901796af | 2346 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)"); |
50d5c8d8 | 2347 | goto out_release; |
1da177e4 LT |
2348 | } |
2349 | ||
50d5c8d8 | 2350 | /* |
67dc288c | 2351 | * Recover the buffer only if we get an LSN from it and it's less than |
50d5c8d8 | 2352 | * the lsn of the transaction we are replaying. |
67dc288c DC |
2353 | * |
2354 | * Note that we have to be extremely careful of readahead here. | |
2355 | * Readahead does not attach verfiers to the buffers so if we don't | |
2356 | * actually do any replay after readahead because of the LSN we found | |
2357 | * in the buffer if more recent than that current transaction then we | |
2358 | * need to attach the verifier directly. Failure to do so can lead to | |
2359 | * future recovery actions (e.g. EFI and unlinked list recovery) can | |
2360 | * operate on the buffers and they won't get the verifier attached. This | |
2361 | * can lead to blocks on disk having the correct content but a stale | |
2362 | * CRC. | |
2363 | * | |
2364 | * It is safe to assume these clean buffers are currently up to date. | |
2365 | * If the buffer is dirtied by a later transaction being replayed, then | |
2366 | * the verifier will be reset to match whatever recover turns that | |
2367 | * buffer into. | |
50d5c8d8 DC |
2368 | */ |
2369 | lsn = xlog_recover_get_buf_lsn(mp, bp); | |
67dc288c DC |
2370 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { |
2371 | xlog_recover_validate_buf_type(mp, bp, buf_f); | |
50d5c8d8 | 2372 | goto out_release; |
67dc288c | 2373 | } |
50d5c8d8 | 2374 | |
e2714bf8 | 2375 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
1da177e4 | 2376 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); |
ad3714b8 DC |
2377 | if (error) |
2378 | goto out_release; | |
e2714bf8 | 2379 | } else if (buf_f->blf_flags & |
c1155410 | 2380 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
ad3714b8 DC |
2381 | bool dirty; |
2382 | ||
2383 | dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); | |
2384 | if (!dirty) | |
2385 | goto out_release; | |
1da177e4 | 2386 | } else { |
9abbc539 | 2387 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 | 2388 | } |
1da177e4 LT |
2389 | |
2390 | /* | |
2391 | * Perform delayed write on the buffer. Asynchronous writes will be | |
2392 | * slower when taking into account all the buffers to be flushed. | |
2393 | * | |
2394 | * Also make sure that only inode buffers with good sizes stay in | |
2395 | * the buffer cache. The kernel moves inodes in buffers of 1 block | |
0f49efd8 | 2396 | * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode |
1da177e4 LT |
2397 | * buffers in the log can be a different size if the log was generated |
2398 | * by an older kernel using unclustered inode buffers or a newer kernel | |
2399 | * running with a different inode cluster size. Regardless, if the | |
0f49efd8 JL |
2400 | * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size) |
2401 | * for *our* value of mp->m_inode_cluster_size, then we need to keep | |
1da177e4 LT |
2402 | * the buffer out of the buffer cache so that the buffer won't |
2403 | * overlap with future reads of those inodes. | |
2404 | */ | |
2405 | if (XFS_DINODE_MAGIC == | |
b53e675d | 2406 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
aa0e8833 | 2407 | (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize, |
0f49efd8 | 2408 | (__uint32_t)log->l_mp->m_inode_cluster_size))) { |
c867cb61 | 2409 | xfs_buf_stale(bp); |
c2b006c1 | 2410 | error = xfs_bwrite(bp); |
1da177e4 | 2411 | } else { |
ebad861b | 2412 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2413 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2414 | xfs_buf_delwri_queue(bp, buffer_list); |
1da177e4 LT |
2415 | } |
2416 | ||
50d5c8d8 | 2417 | out_release: |
c2b006c1 CH |
2418 | xfs_buf_relse(bp); |
2419 | return error; | |
1da177e4 LT |
2420 | } |
2421 | ||
638f4416 DC |
2422 | /* |
2423 | * Inode fork owner changes | |
2424 | * | |
2425 | * If we have been told that we have to reparent the inode fork, it's because an | |
2426 | * extent swap operation on a CRC enabled filesystem has been done and we are | |
2427 | * replaying it. We need to walk the BMBT of the appropriate fork and change the | |
2428 | * owners of it. | |
2429 | * | |
2430 | * The complexity here is that we don't have an inode context to work with, so | |
2431 | * after we've replayed the inode we need to instantiate one. This is where the | |
2432 | * fun begins. | |
2433 | * | |
2434 | * We are in the middle of log recovery, so we can't run transactions. That | |
2435 | * means we cannot use cache coherent inode instantiation via xfs_iget(), as | |
2436 | * that will result in the corresponding iput() running the inode through | |
2437 | * xfs_inactive(). If we've just replayed an inode core that changes the link | |
2438 | * count to zero (i.e. it's been unlinked), then xfs_inactive() will run | |
2439 | * transactions (bad!). | |
2440 | * | |
2441 | * So, to avoid this, we instantiate an inode directly from the inode core we've | |
2442 | * just recovered. We have the buffer still locked, and all we really need to | |
2443 | * instantiate is the inode core and the forks being modified. We can do this | |
2444 | * manually, then run the inode btree owner change, and then tear down the | |
2445 | * xfs_inode without having to run any transactions at all. | |
2446 | * | |
2447 | * Also, because we don't have a transaction context available here but need to | |
2448 | * gather all the buffers we modify for writeback so we pass the buffer_list | |
2449 | * instead for the operation to use. | |
2450 | */ | |
2451 | ||
2452 | STATIC int | |
2453 | xfs_recover_inode_owner_change( | |
2454 | struct xfs_mount *mp, | |
2455 | struct xfs_dinode *dip, | |
2456 | struct xfs_inode_log_format *in_f, | |
2457 | struct list_head *buffer_list) | |
2458 | { | |
2459 | struct xfs_inode *ip; | |
2460 | int error; | |
2461 | ||
2462 | ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)); | |
2463 | ||
2464 | ip = xfs_inode_alloc(mp, in_f->ilf_ino); | |
2465 | if (!ip) | |
2451337d | 2466 | return -ENOMEM; |
638f4416 DC |
2467 | |
2468 | /* instantiate the inode */ | |
2469 | xfs_dinode_from_disk(&ip->i_d, dip); | |
2470 | ASSERT(ip->i_d.di_version >= 3); | |
2471 | ||
2472 | error = xfs_iformat_fork(ip, dip); | |
2473 | if (error) | |
2474 | goto out_free_ip; | |
2475 | ||
2476 | ||
2477 | if (in_f->ilf_fields & XFS_ILOG_DOWNER) { | |
2478 | ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT); | |
2479 | error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK, | |
2480 | ip->i_ino, buffer_list); | |
2481 | if (error) | |
2482 | goto out_free_ip; | |
2483 | } | |
2484 | ||
2485 | if (in_f->ilf_fields & XFS_ILOG_AOWNER) { | |
2486 | ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT); | |
2487 | error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK, | |
2488 | ip->i_ino, buffer_list); | |
2489 | if (error) | |
2490 | goto out_free_ip; | |
2491 | } | |
2492 | ||
2493 | out_free_ip: | |
2494 | xfs_inode_free(ip); | |
2495 | return error; | |
2496 | } | |
2497 | ||
1da177e4 | 2498 | STATIC int |
c9f71f5f | 2499 | xlog_recover_inode_pass2( |
9a8d2fdb MT |
2500 | struct xlog *log, |
2501 | struct list_head *buffer_list, | |
50d5c8d8 DC |
2502 | struct xlog_recover_item *item, |
2503 | xfs_lsn_t current_lsn) | |
1da177e4 LT |
2504 | { |
2505 | xfs_inode_log_format_t *in_f; | |
c9f71f5f | 2506 | xfs_mount_t *mp = log->l_mp; |
1da177e4 | 2507 | xfs_buf_t *bp; |
1da177e4 | 2508 | xfs_dinode_t *dip; |
1da177e4 LT |
2509 | int len; |
2510 | xfs_caddr_t src; | |
2511 | xfs_caddr_t dest; | |
2512 | int error; | |
2513 | int attr_index; | |
2514 | uint fields; | |
347d1c01 | 2515 | xfs_icdinode_t *dicp; |
93848a99 | 2516 | uint isize; |
6d192a9b | 2517 | int need_free = 0; |
1da177e4 | 2518 | |
6d192a9b | 2519 | if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) { |
4e0d5f92 | 2520 | in_f = item->ri_buf[0].i_addr; |
6d192a9b | 2521 | } else { |
4e0d5f92 | 2522 | in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP); |
6d192a9b TS |
2523 | need_free = 1; |
2524 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); | |
2525 | if (error) | |
2526 | goto error; | |
2527 | } | |
1da177e4 LT |
2528 | |
2529 | /* | |
2530 | * Inode buffers can be freed, look out for it, | |
2531 | * and do not replay the inode. | |
2532 | */ | |
a1941895 CH |
2533 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
2534 | in_f->ilf_len, 0)) { | |
6d192a9b | 2535 | error = 0; |
9abbc539 | 2536 | trace_xfs_log_recover_inode_cancel(log, in_f); |
6d192a9b TS |
2537 | goto error; |
2538 | } | |
9abbc539 | 2539 | trace_xfs_log_recover_inode_recover(log, in_f); |
1da177e4 | 2540 | |
c3f8fc73 | 2541 | bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0, |
93848a99 | 2542 | &xfs_inode_buf_ops); |
ac4d6888 | 2543 | if (!bp) { |
2451337d | 2544 | error = -ENOMEM; |
ac4d6888 CS |
2545 | goto error; |
2546 | } | |
e5702805 | 2547 | error = bp->b_error; |
5a52c2a5 | 2548 | if (error) { |
901796af | 2549 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)"); |
638f4416 | 2550 | goto out_release; |
1da177e4 | 2551 | } |
1da177e4 | 2552 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); |
a1941895 | 2553 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset); |
1da177e4 LT |
2554 | |
2555 | /* | |
2556 | * Make sure the place we're flushing out to really looks | |
2557 | * like an inode! | |
2558 | */ | |
69ef921b | 2559 | if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) { |
a0fa2b67 DC |
2560 | xfs_alert(mp, |
2561 | "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld", | |
2562 | __func__, dip, bp, in_f->ilf_ino); | |
c9f71f5f | 2563 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)", |
1da177e4 | 2564 | XFS_ERRLEVEL_LOW, mp); |
2451337d | 2565 | error = -EFSCORRUPTED; |
638f4416 | 2566 | goto out_release; |
1da177e4 | 2567 | } |
4e0d5f92 | 2568 | dicp = item->ri_buf[1].i_addr; |
1da177e4 | 2569 | if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) { |
a0fa2b67 DC |
2570 | xfs_alert(mp, |
2571 | "%s: Bad inode log record, rec ptr 0x%p, ino %Ld", | |
2572 | __func__, item, in_f->ilf_ino); | |
c9f71f5f | 2573 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)", |
1da177e4 | 2574 | XFS_ERRLEVEL_LOW, mp); |
2451337d | 2575 | error = -EFSCORRUPTED; |
638f4416 | 2576 | goto out_release; |
1da177e4 LT |
2577 | } |
2578 | ||
50d5c8d8 DC |
2579 | /* |
2580 | * If the inode has an LSN in it, recover the inode only if it's less | |
638f4416 DC |
2581 | * than the lsn of the transaction we are replaying. Note: we still |
2582 | * need to replay an owner change even though the inode is more recent | |
2583 | * than the transaction as there is no guarantee that all the btree | |
2584 | * blocks are more recent than this transaction, too. | |
50d5c8d8 DC |
2585 | */ |
2586 | if (dip->di_version >= 3) { | |
2587 | xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn); | |
2588 | ||
2589 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { | |
2590 | trace_xfs_log_recover_inode_skip(log, in_f); | |
2591 | error = 0; | |
638f4416 | 2592 | goto out_owner_change; |
50d5c8d8 DC |
2593 | } |
2594 | } | |
2595 | ||
e60896d8 DC |
2596 | /* |
2597 | * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes | |
2598 | * are transactional and if ordering is necessary we can determine that | |
2599 | * more accurately by the LSN field in the V3 inode core. Don't trust | |
2600 | * the inode versions we might be changing them here - use the | |
2601 | * superblock flag to determine whether we need to look at di_flushiter | |
2602 | * to skip replay when the on disk inode is newer than the log one | |
2603 | */ | |
2604 | if (!xfs_sb_version_hascrc(&mp->m_sb) && | |
2605 | dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) { | |
1da177e4 LT |
2606 | /* |
2607 | * Deal with the wrap case, DI_MAX_FLUSH is less | |
2608 | * than smaller numbers | |
2609 | */ | |
81591fe2 | 2610 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
347d1c01 | 2611 | dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
1da177e4 LT |
2612 | /* do nothing */ |
2613 | } else { | |
9abbc539 | 2614 | trace_xfs_log_recover_inode_skip(log, in_f); |
6d192a9b | 2615 | error = 0; |
638f4416 | 2616 | goto out_release; |
1da177e4 LT |
2617 | } |
2618 | } | |
e60896d8 | 2619 | |
1da177e4 LT |
2620 | /* Take the opportunity to reset the flush iteration count */ |
2621 | dicp->di_flushiter = 0; | |
2622 | ||
abbede1b | 2623 | if (unlikely(S_ISREG(dicp->di_mode))) { |
1da177e4 LT |
2624 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
2625 | (dicp->di_format != XFS_DINODE_FMT_BTREE)) { | |
c9f71f5f | 2626 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)", |
1da177e4 | 2627 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
2628 | xfs_alert(mp, |
2629 | "%s: Bad regular inode log record, rec ptr 0x%p, " | |
2630 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2631 | __func__, item, dip, bp, in_f->ilf_ino); | |
2451337d | 2632 | error = -EFSCORRUPTED; |
638f4416 | 2633 | goto out_release; |
1da177e4 | 2634 | } |
abbede1b | 2635 | } else if (unlikely(S_ISDIR(dicp->di_mode))) { |
1da177e4 LT |
2636 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
2637 | (dicp->di_format != XFS_DINODE_FMT_BTREE) && | |
2638 | (dicp->di_format != XFS_DINODE_FMT_LOCAL)) { | |
c9f71f5f | 2639 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)", |
1da177e4 | 2640 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
2641 | xfs_alert(mp, |
2642 | "%s: Bad dir inode log record, rec ptr 0x%p, " | |
2643 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
2644 | __func__, item, dip, bp, in_f->ilf_ino); | |
2451337d | 2645 | error = -EFSCORRUPTED; |
638f4416 | 2646 | goto out_release; |
1da177e4 LT |
2647 | } |
2648 | } | |
2649 | if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){ | |
c9f71f5f | 2650 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)", |
1da177e4 | 2651 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
2652 | xfs_alert(mp, |
2653 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
2654 | "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld", | |
2655 | __func__, item, dip, bp, in_f->ilf_ino, | |
1da177e4 LT |
2656 | dicp->di_nextents + dicp->di_anextents, |
2657 | dicp->di_nblocks); | |
2451337d | 2658 | error = -EFSCORRUPTED; |
638f4416 | 2659 | goto out_release; |
1da177e4 LT |
2660 | } |
2661 | if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) { | |
c9f71f5f | 2662 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)", |
1da177e4 | 2663 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
2664 | xfs_alert(mp, |
2665 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
2666 | "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__, | |
c9f71f5f | 2667 | item, dip, bp, in_f->ilf_ino, dicp->di_forkoff); |
2451337d | 2668 | error = -EFSCORRUPTED; |
638f4416 | 2669 | goto out_release; |
1da177e4 | 2670 | } |
93848a99 CH |
2671 | isize = xfs_icdinode_size(dicp->di_version); |
2672 | if (unlikely(item->ri_buf[1].i_len > isize)) { | |
c9f71f5f | 2673 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)", |
1da177e4 | 2674 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
2675 | xfs_alert(mp, |
2676 | "%s: Bad inode log record length %d, rec ptr 0x%p", | |
2677 | __func__, item->ri_buf[1].i_len, item); | |
2451337d | 2678 | error = -EFSCORRUPTED; |
638f4416 | 2679 | goto out_release; |
1da177e4 LT |
2680 | } |
2681 | ||
2682 | /* The core is in in-core format */ | |
93848a99 | 2683 | xfs_dinode_to_disk(dip, dicp); |
1da177e4 LT |
2684 | |
2685 | /* the rest is in on-disk format */ | |
93848a99 CH |
2686 | if (item->ri_buf[1].i_len > isize) { |
2687 | memcpy((char *)dip + isize, | |
2688 | item->ri_buf[1].i_addr + isize, | |
2689 | item->ri_buf[1].i_len - isize); | |
1da177e4 LT |
2690 | } |
2691 | ||
2692 | fields = in_f->ilf_fields; | |
2693 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { | |
2694 | case XFS_ILOG_DEV: | |
81591fe2 | 2695 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
1da177e4 LT |
2696 | break; |
2697 | case XFS_ILOG_UUID: | |
81591fe2 CH |
2698 | memcpy(XFS_DFORK_DPTR(dip), |
2699 | &in_f->ilf_u.ilfu_uuid, | |
2700 | sizeof(uuid_t)); | |
1da177e4 LT |
2701 | break; |
2702 | } | |
2703 | ||
2704 | if (in_f->ilf_size == 2) | |
638f4416 | 2705 | goto out_owner_change; |
1da177e4 LT |
2706 | len = item->ri_buf[2].i_len; |
2707 | src = item->ri_buf[2].i_addr; | |
2708 | ASSERT(in_f->ilf_size <= 4); | |
2709 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); | |
2710 | ASSERT(!(fields & XFS_ILOG_DFORK) || | |
2711 | (len == in_f->ilf_dsize)); | |
2712 | ||
2713 | switch (fields & XFS_ILOG_DFORK) { | |
2714 | case XFS_ILOG_DDATA: | |
2715 | case XFS_ILOG_DEXT: | |
81591fe2 | 2716 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
1da177e4 LT |
2717 | break; |
2718 | ||
2719 | case XFS_ILOG_DBROOT: | |
7cc95a82 | 2720 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
81591fe2 | 2721 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
1da177e4 LT |
2722 | XFS_DFORK_DSIZE(dip, mp)); |
2723 | break; | |
2724 | ||
2725 | default: | |
2726 | /* | |
2727 | * There are no data fork flags set. | |
2728 | */ | |
2729 | ASSERT((fields & XFS_ILOG_DFORK) == 0); | |
2730 | break; | |
2731 | } | |
2732 | ||
2733 | /* | |
2734 | * If we logged any attribute data, recover it. There may or | |
2735 | * may not have been any other non-core data logged in this | |
2736 | * transaction. | |
2737 | */ | |
2738 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2739 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { | |
2740 | attr_index = 3; | |
2741 | } else { | |
2742 | attr_index = 2; | |
2743 | } | |
2744 | len = item->ri_buf[attr_index].i_len; | |
2745 | src = item->ri_buf[attr_index].i_addr; | |
2746 | ASSERT(len == in_f->ilf_asize); | |
2747 | ||
2748 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
2749 | case XFS_ILOG_ADATA: | |
2750 | case XFS_ILOG_AEXT: | |
2751 | dest = XFS_DFORK_APTR(dip); | |
2752 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); | |
2753 | memcpy(dest, src, len); | |
2754 | break; | |
2755 | ||
2756 | case XFS_ILOG_ABROOT: | |
2757 | dest = XFS_DFORK_APTR(dip); | |
7cc95a82 CH |
2758 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
2759 | len, (xfs_bmdr_block_t*)dest, | |
1da177e4 LT |
2760 | XFS_DFORK_ASIZE(dip, mp)); |
2761 | break; | |
2762 | ||
2763 | default: | |
a0fa2b67 | 2764 | xfs_warn(log->l_mp, "%s: Invalid flag", __func__); |
1da177e4 | 2765 | ASSERT(0); |
2451337d | 2766 | error = -EIO; |
638f4416 | 2767 | goto out_release; |
1da177e4 LT |
2768 | } |
2769 | } | |
2770 | ||
638f4416 DC |
2771 | out_owner_change: |
2772 | if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) | |
2773 | error = xfs_recover_inode_owner_change(mp, dip, in_f, | |
2774 | buffer_list); | |
93848a99 CH |
2775 | /* re-generate the checksum. */ |
2776 | xfs_dinode_calc_crc(log->l_mp, dip); | |
2777 | ||
ebad861b | 2778 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2779 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2780 | xfs_buf_delwri_queue(bp, buffer_list); |
50d5c8d8 DC |
2781 | |
2782 | out_release: | |
61551f1e | 2783 | xfs_buf_relse(bp); |
6d192a9b TS |
2784 | error: |
2785 | if (need_free) | |
f0e2d93c | 2786 | kmem_free(in_f); |
b474c7ae | 2787 | return error; |
1da177e4 LT |
2788 | } |
2789 | ||
2790 | /* | |
9a8d2fdb | 2791 | * Recover QUOTAOFF records. We simply make a note of it in the xlog |
1da177e4 LT |
2792 | * structure, so that we know not to do any dquot item or dquot buffer recovery, |
2793 | * of that type. | |
2794 | */ | |
2795 | STATIC int | |
c9f71f5f | 2796 | xlog_recover_quotaoff_pass1( |
9a8d2fdb MT |
2797 | struct xlog *log, |
2798 | struct xlog_recover_item *item) | |
1da177e4 | 2799 | { |
c9f71f5f | 2800 | xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr; |
1da177e4 LT |
2801 | ASSERT(qoff_f); |
2802 | ||
2803 | /* | |
2804 | * The logitem format's flag tells us if this was user quotaoff, | |
77a7cce4 | 2805 | * group/project quotaoff or both. |
1da177e4 LT |
2806 | */ |
2807 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) | |
2808 | log->l_quotaoffs_flag |= XFS_DQ_USER; | |
77a7cce4 NS |
2809 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
2810 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; | |
1da177e4 LT |
2811 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
2812 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; | |
2813 | ||
d99831ff | 2814 | return 0; |
1da177e4 LT |
2815 | } |
2816 | ||
2817 | /* | |
2818 | * Recover a dquot record | |
2819 | */ | |
2820 | STATIC int | |
c9f71f5f | 2821 | xlog_recover_dquot_pass2( |
9a8d2fdb MT |
2822 | struct xlog *log, |
2823 | struct list_head *buffer_list, | |
50d5c8d8 DC |
2824 | struct xlog_recover_item *item, |
2825 | xfs_lsn_t current_lsn) | |
1da177e4 | 2826 | { |
c9f71f5f | 2827 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2828 | xfs_buf_t *bp; |
2829 | struct xfs_disk_dquot *ddq, *recddq; | |
2830 | int error; | |
2831 | xfs_dq_logformat_t *dq_f; | |
2832 | uint type; | |
2833 | ||
1da177e4 LT |
2834 | |
2835 | /* | |
2836 | * Filesystems are required to send in quota flags at mount time. | |
2837 | */ | |
2838 | if (mp->m_qflags == 0) | |
d99831ff | 2839 | return 0; |
1da177e4 | 2840 | |
4e0d5f92 CH |
2841 | recddq = item->ri_buf[1].i_addr; |
2842 | if (recddq == NULL) { | |
a0fa2b67 | 2843 | xfs_alert(log->l_mp, "NULL dquot in %s.", __func__); |
2451337d | 2844 | return -EIO; |
0c5e1ce8 | 2845 | } |
8ec6dba2 | 2846 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
a0fa2b67 | 2847 | xfs_alert(log->l_mp, "dquot too small (%d) in %s.", |
0c5e1ce8 | 2848 | item->ri_buf[1].i_len, __func__); |
2451337d | 2849 | return -EIO; |
0c5e1ce8 CH |
2850 | } |
2851 | ||
1da177e4 LT |
2852 | /* |
2853 | * This type of quotas was turned off, so ignore this record. | |
2854 | */ | |
b53e675d | 2855 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
1da177e4 LT |
2856 | ASSERT(type); |
2857 | if (log->l_quotaoffs_flag & type) | |
d99831ff | 2858 | return 0; |
1da177e4 LT |
2859 | |
2860 | /* | |
2861 | * At this point we know that quota was _not_ turned off. | |
2862 | * Since the mount flags are not indicating to us otherwise, this | |
2863 | * must mean that quota is on, and the dquot needs to be replayed. | |
2864 | * Remember that we may not have fully recovered the superblock yet, | |
2865 | * so we can't do the usual trick of looking at the SB quota bits. | |
2866 | * | |
2867 | * The other possibility, of course, is that the quota subsystem was | |
2868 | * removed since the last mount - ENOSYS. | |
2869 | */ | |
4e0d5f92 | 2870 | dq_f = item->ri_buf[0].i_addr; |
1da177e4 | 2871 | ASSERT(dq_f); |
9aede1d8 | 2872 | error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, |
a0fa2b67 DC |
2873 | "xlog_recover_dquot_pass2 (log copy)"); |
2874 | if (error) | |
2451337d | 2875 | return -EIO; |
1da177e4 LT |
2876 | ASSERT(dq_f->qlf_len == 1); |
2877 | ||
ad3714b8 DC |
2878 | /* |
2879 | * At this point we are assuming that the dquots have been allocated | |
2880 | * and hence the buffer has valid dquots stamped in it. It should, | |
2881 | * therefore, pass verifier validation. If the dquot is bad, then the | |
2882 | * we'll return an error here, so we don't need to specifically check | |
2883 | * the dquot in the buffer after the verifier has run. | |
2884 | */ | |
7ca790a5 | 2885 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno, |
c3f8fc73 | 2886 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp, |
ad3714b8 | 2887 | &xfs_dquot_buf_ops); |
7ca790a5 | 2888 | if (error) |
1da177e4 | 2889 | return error; |
7ca790a5 | 2890 | |
1da177e4 LT |
2891 | ASSERT(bp); |
2892 | ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset); | |
2893 | ||
50d5c8d8 DC |
2894 | /* |
2895 | * If the dquot has an LSN in it, recover the dquot only if it's less | |
2896 | * than the lsn of the transaction we are replaying. | |
2897 | */ | |
2898 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
2899 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq; | |
2900 | xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn); | |
2901 | ||
2902 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { | |
2903 | goto out_release; | |
2904 | } | |
2905 | } | |
2906 | ||
1da177e4 | 2907 | memcpy(ddq, recddq, item->ri_buf[1].i_len); |
6fcdc59d DC |
2908 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
2909 | xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk), | |
2910 | XFS_DQUOT_CRC_OFF); | |
2911 | } | |
1da177e4 LT |
2912 | |
2913 | ASSERT(dq_f->qlf_size == 2); | |
ebad861b | 2914 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2915 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2916 | xfs_buf_delwri_queue(bp, buffer_list); |
1da177e4 | 2917 | |
50d5c8d8 DC |
2918 | out_release: |
2919 | xfs_buf_relse(bp); | |
2920 | return 0; | |
1da177e4 LT |
2921 | } |
2922 | ||
2923 | /* | |
2924 | * This routine is called to create an in-core extent free intent | |
2925 | * item from the efi format structure which was logged on disk. | |
2926 | * It allocates an in-core efi, copies the extents from the format | |
2927 | * structure into it, and adds the efi to the AIL with the given | |
2928 | * LSN. | |
2929 | */ | |
6d192a9b | 2930 | STATIC int |
c9f71f5f | 2931 | xlog_recover_efi_pass2( |
9a8d2fdb MT |
2932 | struct xlog *log, |
2933 | struct xlog_recover_item *item, | |
2934 | xfs_lsn_t lsn) | |
1da177e4 | 2935 | { |
6d192a9b | 2936 | int error; |
c9f71f5f | 2937 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2938 | xfs_efi_log_item_t *efip; |
2939 | xfs_efi_log_format_t *efi_formatp; | |
1da177e4 | 2940 | |
4e0d5f92 | 2941 | efi_formatp = item->ri_buf[0].i_addr; |
1da177e4 | 2942 | |
1da177e4 | 2943 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); |
6d192a9b TS |
2944 | if ((error = xfs_efi_copy_format(&(item->ri_buf[0]), |
2945 | &(efip->efi_format)))) { | |
2946 | xfs_efi_item_free(efip); | |
2947 | return error; | |
2948 | } | |
b199c8a4 | 2949 | atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents); |
1da177e4 | 2950 | |
a9c21c1b | 2951 | spin_lock(&log->l_ailp->xa_lock); |
1da177e4 | 2952 | /* |
783a2f65 | 2953 | * xfs_trans_ail_update() drops the AIL lock. |
1da177e4 | 2954 | */ |
e6059949 | 2955 | xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn); |
6d192a9b | 2956 | return 0; |
1da177e4 LT |
2957 | } |
2958 | ||
2959 | ||
2960 | /* | |
2961 | * This routine is called when an efd format structure is found in | |
2962 | * a committed transaction in the log. It's purpose is to cancel | |
2963 | * the corresponding efi if it was still in the log. To do this | |
2964 | * it searches the AIL for the efi with an id equal to that in the | |
2965 | * efd format structure. If we find it, we remove the efi from the | |
2966 | * AIL and free it. | |
2967 | */ | |
c9f71f5f CH |
2968 | STATIC int |
2969 | xlog_recover_efd_pass2( | |
9a8d2fdb MT |
2970 | struct xlog *log, |
2971 | struct xlog_recover_item *item) | |
1da177e4 | 2972 | { |
1da177e4 LT |
2973 | xfs_efd_log_format_t *efd_formatp; |
2974 | xfs_efi_log_item_t *efip = NULL; | |
2975 | xfs_log_item_t *lip; | |
1da177e4 | 2976 | __uint64_t efi_id; |
27d8d5fe | 2977 | struct xfs_ail_cursor cur; |
783a2f65 | 2978 | struct xfs_ail *ailp = log->l_ailp; |
1da177e4 | 2979 | |
4e0d5f92 | 2980 | efd_formatp = item->ri_buf[0].i_addr; |
6d192a9b TS |
2981 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
2982 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || | |
2983 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + | |
2984 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); | |
1da177e4 LT |
2985 | efi_id = efd_formatp->efd_efi_id; |
2986 | ||
2987 | /* | |
2988 | * Search for the efi with the id in the efd format structure | |
2989 | * in the AIL. | |
2990 | */ | |
a9c21c1b DC |
2991 | spin_lock(&ailp->xa_lock); |
2992 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
2993 | while (lip != NULL) { |
2994 | if (lip->li_type == XFS_LI_EFI) { | |
2995 | efip = (xfs_efi_log_item_t *)lip; | |
2996 | if (efip->efi_format.efi_id == efi_id) { | |
2997 | /* | |
783a2f65 | 2998 | * xfs_trans_ail_delete() drops the |
1da177e4 LT |
2999 | * AIL lock. |
3000 | */ | |
04913fdd DC |
3001 | xfs_trans_ail_delete(ailp, lip, |
3002 | SHUTDOWN_CORRUPT_INCORE); | |
8ae2c0f6 | 3003 | xfs_efi_item_free(efip); |
a9c21c1b | 3004 | spin_lock(&ailp->xa_lock); |
27d8d5fe | 3005 | break; |
1da177e4 LT |
3006 | } |
3007 | } | |
a9c21c1b | 3008 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 3009 | } |
e4a1e29c | 3010 | xfs_trans_ail_cursor_done(&cur); |
a9c21c1b | 3011 | spin_unlock(&ailp->xa_lock); |
c9f71f5f CH |
3012 | |
3013 | return 0; | |
1da177e4 LT |
3014 | } |
3015 | ||
28c8e41a DC |
3016 | /* |
3017 | * This routine is called when an inode create format structure is found in a | |
3018 | * committed transaction in the log. It's purpose is to initialise the inodes | |
3019 | * being allocated on disk. This requires us to get inode cluster buffers that | |
3020 | * match the range to be intialised, stamped with inode templates and written | |
3021 | * by delayed write so that subsequent modifications will hit the cached buffer | |
3022 | * and only need writing out at the end of recovery. | |
3023 | */ | |
3024 | STATIC int | |
3025 | xlog_recover_do_icreate_pass2( | |
3026 | struct xlog *log, | |
3027 | struct list_head *buffer_list, | |
3028 | xlog_recover_item_t *item) | |
3029 | { | |
3030 | struct xfs_mount *mp = log->l_mp; | |
3031 | struct xfs_icreate_log *icl; | |
3032 | xfs_agnumber_t agno; | |
3033 | xfs_agblock_t agbno; | |
3034 | unsigned int count; | |
3035 | unsigned int isize; | |
3036 | xfs_agblock_t length; | |
3037 | ||
3038 | icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr; | |
3039 | if (icl->icl_type != XFS_LI_ICREATE) { | |
3040 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type"); | |
2451337d | 3041 | return -EINVAL; |
28c8e41a DC |
3042 | } |
3043 | ||
3044 | if (icl->icl_size != 1) { | |
3045 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size"); | |
2451337d | 3046 | return -EINVAL; |
28c8e41a DC |
3047 | } |
3048 | ||
3049 | agno = be32_to_cpu(icl->icl_ag); | |
3050 | if (agno >= mp->m_sb.sb_agcount) { | |
3051 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno"); | |
2451337d | 3052 | return -EINVAL; |
28c8e41a DC |
3053 | } |
3054 | agbno = be32_to_cpu(icl->icl_agbno); | |
3055 | if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) { | |
3056 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno"); | |
2451337d | 3057 | return -EINVAL; |
28c8e41a DC |
3058 | } |
3059 | isize = be32_to_cpu(icl->icl_isize); | |
3060 | if (isize != mp->m_sb.sb_inodesize) { | |
3061 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize"); | |
2451337d | 3062 | return -EINVAL; |
28c8e41a DC |
3063 | } |
3064 | count = be32_to_cpu(icl->icl_count); | |
3065 | if (!count) { | |
3066 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count"); | |
2451337d | 3067 | return -EINVAL; |
28c8e41a DC |
3068 | } |
3069 | length = be32_to_cpu(icl->icl_length); | |
3070 | if (!length || length >= mp->m_sb.sb_agblocks) { | |
3071 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length"); | |
2451337d | 3072 | return -EINVAL; |
28c8e41a DC |
3073 | } |
3074 | ||
3075 | /* existing allocation is fixed value */ | |
71783438 | 3076 | ASSERT(count == mp->m_ialloc_inos); |
126cd105 | 3077 | ASSERT(length == mp->m_ialloc_blks); |
71783438 | 3078 | if (count != mp->m_ialloc_inos || |
126cd105 | 3079 | length != mp->m_ialloc_blks) { |
28c8e41a | 3080 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2"); |
2451337d | 3081 | return -EINVAL; |
28c8e41a DC |
3082 | } |
3083 | ||
3084 | /* | |
3085 | * Inode buffers can be freed. Do not replay the inode initialisation as | |
3086 | * we could be overwriting something written after this inode buffer was | |
3087 | * cancelled. | |
3088 | * | |
3089 | * XXX: we need to iterate all buffers and only init those that are not | |
3090 | * cancelled. I think that a more fine grained factoring of | |
3091 | * xfs_ialloc_inode_init may be appropriate here to enable this to be | |
3092 | * done easily. | |
3093 | */ | |
3094 | if (xlog_check_buffer_cancelled(log, | |
3095 | XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0)) | |
3096 | return 0; | |
3097 | ||
3098 | xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length, | |
3099 | be32_to_cpu(icl->icl_gen)); | |
3100 | return 0; | |
3101 | } | |
3102 | ||
00574da1 ZYW |
3103 | STATIC void |
3104 | xlog_recover_buffer_ra_pass2( | |
3105 | struct xlog *log, | |
3106 | struct xlog_recover_item *item) | |
3107 | { | |
3108 | struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr; | |
3109 | struct xfs_mount *mp = log->l_mp; | |
3110 | ||
84a5b730 | 3111 | if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno, |
00574da1 ZYW |
3112 | buf_f->blf_len, buf_f->blf_flags)) { |
3113 | return; | |
3114 | } | |
3115 | ||
3116 | xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno, | |
3117 | buf_f->blf_len, NULL); | |
3118 | } | |
3119 | ||
3120 | STATIC void | |
3121 | xlog_recover_inode_ra_pass2( | |
3122 | struct xlog *log, | |
3123 | struct xlog_recover_item *item) | |
3124 | { | |
3125 | struct xfs_inode_log_format ilf_buf; | |
3126 | struct xfs_inode_log_format *ilfp; | |
3127 | struct xfs_mount *mp = log->l_mp; | |
3128 | int error; | |
3129 | ||
3130 | if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) { | |
3131 | ilfp = item->ri_buf[0].i_addr; | |
3132 | } else { | |
3133 | ilfp = &ilf_buf; | |
3134 | memset(ilfp, 0, sizeof(*ilfp)); | |
3135 | error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp); | |
3136 | if (error) | |
3137 | return; | |
3138 | } | |
3139 | ||
84a5b730 | 3140 | if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0)) |
00574da1 ZYW |
3141 | return; |
3142 | ||
3143 | xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno, | |
d8914002 | 3144 | ilfp->ilf_len, &xfs_inode_buf_ra_ops); |
00574da1 ZYW |
3145 | } |
3146 | ||
3147 | STATIC void | |
3148 | xlog_recover_dquot_ra_pass2( | |
3149 | struct xlog *log, | |
3150 | struct xlog_recover_item *item) | |
3151 | { | |
3152 | struct xfs_mount *mp = log->l_mp; | |
3153 | struct xfs_disk_dquot *recddq; | |
3154 | struct xfs_dq_logformat *dq_f; | |
3155 | uint type; | |
3156 | ||
3157 | ||
3158 | if (mp->m_qflags == 0) | |
3159 | return; | |
3160 | ||
3161 | recddq = item->ri_buf[1].i_addr; | |
3162 | if (recddq == NULL) | |
3163 | return; | |
3164 | if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot)) | |
3165 | return; | |
3166 | ||
3167 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); | |
3168 | ASSERT(type); | |
3169 | if (log->l_quotaoffs_flag & type) | |
3170 | return; | |
3171 | ||
3172 | dq_f = item->ri_buf[0].i_addr; | |
3173 | ASSERT(dq_f); | |
3174 | ASSERT(dq_f->qlf_len == 1); | |
3175 | ||
3176 | xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, | |
0f0d3345 | 3177 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL); |
00574da1 ZYW |
3178 | } |
3179 | ||
3180 | STATIC void | |
3181 | xlog_recover_ra_pass2( | |
3182 | struct xlog *log, | |
3183 | struct xlog_recover_item *item) | |
3184 | { | |
3185 | switch (ITEM_TYPE(item)) { | |
3186 | case XFS_LI_BUF: | |
3187 | xlog_recover_buffer_ra_pass2(log, item); | |
3188 | break; | |
3189 | case XFS_LI_INODE: | |
3190 | xlog_recover_inode_ra_pass2(log, item); | |
3191 | break; | |
3192 | case XFS_LI_DQUOT: | |
3193 | xlog_recover_dquot_ra_pass2(log, item); | |
3194 | break; | |
3195 | case XFS_LI_EFI: | |
3196 | case XFS_LI_EFD: | |
3197 | case XFS_LI_QUOTAOFF: | |
3198 | default: | |
3199 | break; | |
3200 | } | |
3201 | } | |
3202 | ||
d0450948 | 3203 | STATIC int |
c9f71f5f | 3204 | xlog_recover_commit_pass1( |
ad223e60 MT |
3205 | struct xlog *log, |
3206 | struct xlog_recover *trans, | |
3207 | struct xlog_recover_item *item) | |
d0450948 | 3208 | { |
c9f71f5f | 3209 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1); |
d0450948 CH |
3210 | |
3211 | switch (ITEM_TYPE(item)) { | |
3212 | case XFS_LI_BUF: | |
c9f71f5f CH |
3213 | return xlog_recover_buffer_pass1(log, item); |
3214 | case XFS_LI_QUOTAOFF: | |
3215 | return xlog_recover_quotaoff_pass1(log, item); | |
d0450948 | 3216 | case XFS_LI_INODE: |
d0450948 | 3217 | case XFS_LI_EFI: |
d0450948 | 3218 | case XFS_LI_EFD: |
c9f71f5f | 3219 | case XFS_LI_DQUOT: |
28c8e41a | 3220 | case XFS_LI_ICREATE: |
c9f71f5f | 3221 | /* nothing to do in pass 1 */ |
d0450948 | 3222 | return 0; |
c9f71f5f | 3223 | default: |
a0fa2b67 DC |
3224 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
3225 | __func__, ITEM_TYPE(item)); | |
c9f71f5f | 3226 | ASSERT(0); |
2451337d | 3227 | return -EIO; |
c9f71f5f CH |
3228 | } |
3229 | } | |
3230 | ||
3231 | STATIC int | |
3232 | xlog_recover_commit_pass2( | |
ad223e60 MT |
3233 | struct xlog *log, |
3234 | struct xlog_recover *trans, | |
3235 | struct list_head *buffer_list, | |
3236 | struct xlog_recover_item *item) | |
c9f71f5f CH |
3237 | { |
3238 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2); | |
3239 | ||
3240 | switch (ITEM_TYPE(item)) { | |
3241 | case XFS_LI_BUF: | |
50d5c8d8 DC |
3242 | return xlog_recover_buffer_pass2(log, buffer_list, item, |
3243 | trans->r_lsn); | |
c9f71f5f | 3244 | case XFS_LI_INODE: |
50d5c8d8 DC |
3245 | return xlog_recover_inode_pass2(log, buffer_list, item, |
3246 | trans->r_lsn); | |
c9f71f5f CH |
3247 | case XFS_LI_EFI: |
3248 | return xlog_recover_efi_pass2(log, item, trans->r_lsn); | |
3249 | case XFS_LI_EFD: | |
3250 | return xlog_recover_efd_pass2(log, item); | |
d0450948 | 3251 | case XFS_LI_DQUOT: |
50d5c8d8 DC |
3252 | return xlog_recover_dquot_pass2(log, buffer_list, item, |
3253 | trans->r_lsn); | |
28c8e41a DC |
3254 | case XFS_LI_ICREATE: |
3255 | return xlog_recover_do_icreate_pass2(log, buffer_list, item); | |
d0450948 | 3256 | case XFS_LI_QUOTAOFF: |
c9f71f5f CH |
3257 | /* nothing to do in pass2 */ |
3258 | return 0; | |
d0450948 | 3259 | default: |
a0fa2b67 DC |
3260 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
3261 | __func__, ITEM_TYPE(item)); | |
d0450948 | 3262 | ASSERT(0); |
2451337d | 3263 | return -EIO; |
d0450948 CH |
3264 | } |
3265 | } | |
3266 | ||
00574da1 ZYW |
3267 | STATIC int |
3268 | xlog_recover_items_pass2( | |
3269 | struct xlog *log, | |
3270 | struct xlog_recover *trans, | |
3271 | struct list_head *buffer_list, | |
3272 | struct list_head *item_list) | |
3273 | { | |
3274 | struct xlog_recover_item *item; | |
3275 | int error = 0; | |
3276 | ||
3277 | list_for_each_entry(item, item_list, ri_list) { | |
3278 | error = xlog_recover_commit_pass2(log, trans, | |
3279 | buffer_list, item); | |
3280 | if (error) | |
3281 | return error; | |
3282 | } | |
3283 | ||
3284 | return error; | |
3285 | } | |
3286 | ||
d0450948 CH |
3287 | /* |
3288 | * Perform the transaction. | |
3289 | * | |
3290 | * If the transaction modifies a buffer or inode, do it now. Otherwise, | |
3291 | * EFIs and EFDs get queued up by adding entries into the AIL for them. | |
3292 | */ | |
1da177e4 LT |
3293 | STATIC int |
3294 | xlog_recover_commit_trans( | |
ad223e60 | 3295 | struct xlog *log, |
d0450948 | 3296 | struct xlog_recover *trans, |
1da177e4 LT |
3297 | int pass) |
3298 | { | |
00574da1 ZYW |
3299 | int error = 0; |
3300 | int error2; | |
3301 | int items_queued = 0; | |
3302 | struct xlog_recover_item *item; | |
3303 | struct xlog_recover_item *next; | |
3304 | LIST_HEAD (buffer_list); | |
3305 | LIST_HEAD (ra_list); | |
3306 | LIST_HEAD (done_list); | |
3307 | ||
3308 | #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100 | |
1da177e4 | 3309 | |
f0a76953 | 3310 | hlist_del(&trans->r_list); |
d0450948 CH |
3311 | |
3312 | error = xlog_recover_reorder_trans(log, trans, pass); | |
3313 | if (error) | |
1da177e4 | 3314 | return error; |
d0450948 | 3315 | |
00574da1 | 3316 | list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) { |
43ff2122 CH |
3317 | switch (pass) { |
3318 | case XLOG_RECOVER_PASS1: | |
c9f71f5f | 3319 | error = xlog_recover_commit_pass1(log, trans, item); |
43ff2122 CH |
3320 | break; |
3321 | case XLOG_RECOVER_PASS2: | |
00574da1 ZYW |
3322 | xlog_recover_ra_pass2(log, item); |
3323 | list_move_tail(&item->ri_list, &ra_list); | |
3324 | items_queued++; | |
3325 | if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) { | |
3326 | error = xlog_recover_items_pass2(log, trans, | |
3327 | &buffer_list, &ra_list); | |
3328 | list_splice_tail_init(&ra_list, &done_list); | |
3329 | items_queued = 0; | |
3330 | } | |
3331 | ||
43ff2122 CH |
3332 | break; |
3333 | default: | |
3334 | ASSERT(0); | |
3335 | } | |
3336 | ||
d0450948 | 3337 | if (error) |
43ff2122 | 3338 | goto out; |
d0450948 CH |
3339 | } |
3340 | ||
00574da1 ZYW |
3341 | out: |
3342 | if (!list_empty(&ra_list)) { | |
3343 | if (!error) | |
3344 | error = xlog_recover_items_pass2(log, trans, | |
3345 | &buffer_list, &ra_list); | |
3346 | list_splice_tail_init(&ra_list, &done_list); | |
3347 | } | |
3348 | ||
3349 | if (!list_empty(&done_list)) | |
3350 | list_splice_init(&done_list, &trans->r_itemq); | |
3351 | ||
43ff2122 CH |
3352 | error2 = xfs_buf_delwri_submit(&buffer_list); |
3353 | return error ? error : error2; | |
1da177e4 LT |
3354 | } |
3355 | ||
76560669 DC |
3356 | |
3357 | STATIC xlog_recover_t * | |
3358 | xlog_recover_find_tid( | |
3359 | struct hlist_head *head, | |
3360 | xlog_tid_t tid) | |
3361 | { | |
3362 | xlog_recover_t *trans; | |
3363 | ||
3364 | hlist_for_each_entry(trans, head, r_list) { | |
3365 | if (trans->r_log_tid == tid) | |
3366 | return trans; | |
3367 | } | |
3368 | return NULL; | |
3369 | } | |
3370 | ||
3371 | STATIC void | |
3372 | xlog_recover_new_tid( | |
3373 | struct hlist_head *head, | |
3374 | xlog_tid_t tid, | |
3375 | xfs_lsn_t lsn) | |
3376 | { | |
3377 | xlog_recover_t *trans; | |
3378 | ||
3379 | trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP); | |
3380 | trans->r_log_tid = tid; | |
3381 | trans->r_lsn = lsn; | |
3382 | INIT_LIST_HEAD(&trans->r_itemq); | |
3383 | ||
3384 | INIT_HLIST_NODE(&trans->r_list); | |
3385 | hlist_add_head(&trans->r_list, head); | |
3386 | } | |
3387 | ||
3388 | STATIC void | |
3389 | xlog_recover_add_item( | |
3390 | struct list_head *head) | |
3391 | { | |
3392 | xlog_recover_item_t *item; | |
3393 | ||
3394 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); | |
3395 | INIT_LIST_HEAD(&item->ri_list); | |
3396 | list_add_tail(&item->ri_list, head); | |
3397 | } | |
3398 | ||
3399 | STATIC int | |
3400 | xlog_recover_add_to_cont_trans( | |
3401 | struct xlog *log, | |
3402 | struct xlog_recover *trans, | |
3403 | xfs_caddr_t dp, | |
3404 | int len) | |
3405 | { | |
3406 | xlog_recover_item_t *item; | |
3407 | xfs_caddr_t ptr, old_ptr; | |
3408 | int old_len; | |
3409 | ||
3410 | if (list_empty(&trans->r_itemq)) { | |
3411 | /* finish copying rest of trans header */ | |
3412 | xlog_recover_add_item(&trans->r_itemq); | |
3413 | ptr = (xfs_caddr_t) &trans->r_theader + | |
3414 | sizeof(xfs_trans_header_t) - len; | |
3415 | memcpy(ptr, dp, len); | |
3416 | return 0; | |
3417 | } | |
3418 | /* take the tail entry */ | |
3419 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
3420 | ||
3421 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; | |
3422 | old_len = item->ri_buf[item->ri_cnt-1].i_len; | |
3423 | ||
3424 | ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP); | |
3425 | memcpy(&ptr[old_len], dp, len); | |
3426 | item->ri_buf[item->ri_cnt-1].i_len += len; | |
3427 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; | |
3428 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); | |
3429 | return 0; | |
3430 | } | |
3431 | ||
3432 | /* | |
3433 | * The next region to add is the start of a new region. It could be | |
3434 | * a whole region or it could be the first part of a new region. Because | |
3435 | * of this, the assumption here is that the type and size fields of all | |
3436 | * format structures fit into the first 32 bits of the structure. | |
3437 | * | |
3438 | * This works because all regions must be 32 bit aligned. Therefore, we | |
3439 | * either have both fields or we have neither field. In the case we have | |
3440 | * neither field, the data part of the region is zero length. We only have | |
3441 | * a log_op_header and can throw away the header since a new one will appear | |
3442 | * later. If we have at least 4 bytes, then we can determine how many regions | |
3443 | * will appear in the current log item. | |
3444 | */ | |
3445 | STATIC int | |
3446 | xlog_recover_add_to_trans( | |
3447 | struct xlog *log, | |
3448 | struct xlog_recover *trans, | |
3449 | xfs_caddr_t dp, | |
3450 | int len) | |
3451 | { | |
3452 | xfs_inode_log_format_t *in_f; /* any will do */ | |
3453 | xlog_recover_item_t *item; | |
3454 | xfs_caddr_t ptr; | |
3455 | ||
3456 | if (!len) | |
3457 | return 0; | |
3458 | if (list_empty(&trans->r_itemq)) { | |
3459 | /* we need to catch log corruptions here */ | |
3460 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { | |
3461 | xfs_warn(log->l_mp, "%s: bad header magic number", | |
3462 | __func__); | |
3463 | ASSERT(0); | |
3464 | return -EIO; | |
3465 | } | |
3466 | if (len == sizeof(xfs_trans_header_t)) | |
3467 | xlog_recover_add_item(&trans->r_itemq); | |
3468 | memcpy(&trans->r_theader, dp, len); | |
3469 | return 0; | |
3470 | } | |
3471 | ||
3472 | ptr = kmem_alloc(len, KM_SLEEP); | |
3473 | memcpy(ptr, dp, len); | |
3474 | in_f = (xfs_inode_log_format_t *)ptr; | |
3475 | ||
3476 | /* take the tail entry */ | |
3477 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
3478 | if (item->ri_total != 0 && | |
3479 | item->ri_total == item->ri_cnt) { | |
3480 | /* tail item is in use, get a new one */ | |
3481 | xlog_recover_add_item(&trans->r_itemq); | |
3482 | item = list_entry(trans->r_itemq.prev, | |
3483 | xlog_recover_item_t, ri_list); | |
3484 | } | |
3485 | ||
3486 | if (item->ri_total == 0) { /* first region to be added */ | |
3487 | if (in_f->ilf_size == 0 || | |
3488 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { | |
3489 | xfs_warn(log->l_mp, | |
3490 | "bad number of regions (%d) in inode log format", | |
3491 | in_f->ilf_size); | |
3492 | ASSERT(0); | |
3493 | kmem_free(ptr); | |
3494 | return -EIO; | |
3495 | } | |
3496 | ||
3497 | item->ri_total = in_f->ilf_size; | |
3498 | item->ri_buf = | |
3499 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), | |
3500 | KM_SLEEP); | |
3501 | } | |
3502 | ASSERT(item->ri_total > item->ri_cnt); | |
3503 | /* Description region is ri_buf[0] */ | |
3504 | item->ri_buf[item->ri_cnt].i_addr = ptr; | |
3505 | item->ri_buf[item->ri_cnt].i_len = len; | |
3506 | item->ri_cnt++; | |
3507 | trace_xfs_log_recover_item_add(log, trans, item, 0); | |
3508 | return 0; | |
3509 | } | |
3510 | /* | |
3511 | * Free up any resources allocated by the transaction | |
3512 | * | |
3513 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. | |
3514 | */ | |
3515 | STATIC void | |
3516 | xlog_recover_free_trans( | |
3517 | struct xlog_recover *trans) | |
3518 | { | |
3519 | xlog_recover_item_t *item, *n; | |
3520 | int i; | |
3521 | ||
3522 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { | |
3523 | /* Free the regions in the item. */ | |
3524 | list_del(&item->ri_list); | |
3525 | for (i = 0; i < item->ri_cnt; i++) | |
3526 | kmem_free(item->ri_buf[i].i_addr); | |
3527 | /* Free the item itself */ | |
3528 | kmem_free(item->ri_buf); | |
3529 | kmem_free(item); | |
3530 | } | |
3531 | /* Free the transaction recover structure */ | |
3532 | kmem_free(trans); | |
3533 | } | |
3534 | ||
e9131e50 DC |
3535 | /* |
3536 | * On error or completion, trans is freed. | |
3537 | */ | |
1da177e4 | 3538 | STATIC int |
eeb11688 DC |
3539 | xlog_recovery_process_trans( |
3540 | struct xlog *log, | |
3541 | struct xlog_recover *trans, | |
3542 | xfs_caddr_t dp, | |
3543 | unsigned int len, | |
3544 | unsigned int flags, | |
3545 | int pass) | |
1da177e4 | 3546 | { |
e9131e50 DC |
3547 | int error = 0; |
3548 | bool freeit = false; | |
eeb11688 DC |
3549 | |
3550 | /* mask off ophdr transaction container flags */ | |
3551 | flags &= ~XLOG_END_TRANS; | |
3552 | if (flags & XLOG_WAS_CONT_TRANS) | |
3553 | flags &= ~XLOG_CONTINUE_TRANS; | |
3554 | ||
88b863db DC |
3555 | /* |
3556 | * Callees must not free the trans structure. We'll decide if we need to | |
3557 | * free it or not based on the operation being done and it's result. | |
3558 | */ | |
eeb11688 DC |
3559 | switch (flags) { |
3560 | /* expected flag values */ | |
3561 | case 0: | |
3562 | case XLOG_CONTINUE_TRANS: | |
3563 | error = xlog_recover_add_to_trans(log, trans, dp, len); | |
3564 | break; | |
3565 | case XLOG_WAS_CONT_TRANS: | |
3566 | error = xlog_recover_add_to_cont_trans(log, trans, dp, len); | |
3567 | break; | |
3568 | case XLOG_COMMIT_TRANS: | |
3569 | error = xlog_recover_commit_trans(log, trans, pass); | |
88b863db DC |
3570 | /* success or fail, we are now done with this transaction. */ |
3571 | freeit = true; | |
eeb11688 DC |
3572 | break; |
3573 | ||
3574 | /* unexpected flag values */ | |
3575 | case XLOG_UNMOUNT_TRANS: | |
e9131e50 | 3576 | /* just skip trans */ |
eeb11688 | 3577 | xfs_warn(log->l_mp, "%s: Unmount LR", __func__); |
e9131e50 | 3578 | freeit = true; |
eeb11688 DC |
3579 | break; |
3580 | case XLOG_START_TRANS: | |
eeb11688 DC |
3581 | default: |
3582 | xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags); | |
3583 | ASSERT(0); | |
e9131e50 | 3584 | error = -EIO; |
eeb11688 DC |
3585 | break; |
3586 | } | |
e9131e50 DC |
3587 | if (error || freeit) |
3588 | xlog_recover_free_trans(trans); | |
eeb11688 DC |
3589 | return error; |
3590 | } | |
3591 | ||
3592 | STATIC struct xlog_recover * | |
3593 | xlog_recover_ophdr_to_trans( | |
3594 | struct hlist_head rhash[], | |
3595 | struct xlog_rec_header *rhead, | |
3596 | struct xlog_op_header *ohead) | |
3597 | { | |
3598 | struct xlog_recover *trans; | |
3599 | xlog_tid_t tid; | |
3600 | struct hlist_head *rhp; | |
3601 | ||
3602 | tid = be32_to_cpu(ohead->oh_tid); | |
3603 | rhp = &rhash[XLOG_RHASH(tid)]; | |
3604 | trans = xlog_recover_find_tid(rhp, tid); | |
3605 | if (trans) | |
3606 | return trans; | |
3607 | ||
3608 | /* | |
3609 | * If this is a new transaction, the ophdr only contains the | |
3610 | * start record. In that case, the only processing we need to do | |
3611 | * on this opheader is allocate a new recovery container to hold | |
3612 | * the recovery ops that will follow. | |
3613 | */ | |
3614 | if (ohead->oh_flags & XLOG_START_TRANS) { | |
3615 | ASSERT(be32_to_cpu(ohead->oh_len) == 0); | |
3616 | xlog_recover_new_tid(rhp, tid, be64_to_cpu(rhead->h_lsn)); | |
3617 | } | |
3618 | return NULL; | |
3619 | } | |
3620 | ||
3621 | STATIC int | |
3622 | xlog_recover_process_ophdr( | |
3623 | struct xlog *log, | |
3624 | struct hlist_head rhash[], | |
3625 | struct xlog_rec_header *rhead, | |
3626 | struct xlog_op_header *ohead, | |
3627 | xfs_caddr_t dp, | |
3628 | xfs_caddr_t end, | |
3629 | int pass) | |
3630 | { | |
3631 | struct xlog_recover *trans; | |
eeb11688 DC |
3632 | unsigned int len; |
3633 | ||
3634 | /* Do we understand who wrote this op? */ | |
3635 | if (ohead->oh_clientid != XFS_TRANSACTION && | |
3636 | ohead->oh_clientid != XFS_LOG) { | |
3637 | xfs_warn(log->l_mp, "%s: bad clientid 0x%x", | |
3638 | __func__, ohead->oh_clientid); | |
3639 | ASSERT(0); | |
3640 | return -EIO; | |
3641 | } | |
3642 | ||
3643 | /* | |
3644 | * Check the ophdr contains all the data it is supposed to contain. | |
3645 | */ | |
3646 | len = be32_to_cpu(ohead->oh_len); | |
3647 | if (dp + len > end) { | |
3648 | xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len); | |
3649 | WARN_ON(1); | |
3650 | return -EIO; | |
3651 | } | |
3652 | ||
3653 | trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead); | |
3654 | if (!trans) { | |
3655 | /* nothing to do, so skip over this ophdr */ | |
3656 | return 0; | |
3657 | } | |
3658 | ||
e9131e50 DC |
3659 | return xlog_recovery_process_trans(log, trans, dp, len, |
3660 | ohead->oh_flags, pass); | |
1da177e4 LT |
3661 | } |
3662 | ||
3663 | /* | |
3664 | * There are two valid states of the r_state field. 0 indicates that the | |
3665 | * transaction structure is in a normal state. We have either seen the | |
3666 | * start of the transaction or the last operation we added was not a partial | |
3667 | * operation. If the last operation we added to the transaction was a | |
3668 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. | |
3669 | * | |
3670 | * NOTE: skip LRs with 0 data length. | |
3671 | */ | |
3672 | STATIC int | |
3673 | xlog_recover_process_data( | |
9a8d2fdb | 3674 | struct xlog *log, |
f0a76953 | 3675 | struct hlist_head rhash[], |
9a8d2fdb | 3676 | struct xlog_rec_header *rhead, |
1da177e4 LT |
3677 | xfs_caddr_t dp, |
3678 | int pass) | |
3679 | { | |
eeb11688 DC |
3680 | struct xlog_op_header *ohead; |
3681 | xfs_caddr_t end; | |
1da177e4 | 3682 | int num_logops; |
1da177e4 | 3683 | int error; |
1da177e4 | 3684 | |
eeb11688 | 3685 | end = dp + be32_to_cpu(rhead->h_len); |
b53e675d | 3686 | num_logops = be32_to_cpu(rhead->h_num_logops); |
1da177e4 LT |
3687 | |
3688 | /* check the log format matches our own - else we can't recover */ | |
3689 | if (xlog_header_check_recover(log->l_mp, rhead)) | |
2451337d | 3690 | return -EIO; |
1da177e4 | 3691 | |
eeb11688 DC |
3692 | while ((dp < end) && num_logops) { |
3693 | ||
3694 | ohead = (struct xlog_op_header *)dp; | |
3695 | dp += sizeof(*ohead); | |
3696 | ASSERT(dp <= end); | |
3697 | ||
3698 | /* errors will abort recovery */ | |
3699 | error = xlog_recover_process_ophdr(log, rhash, rhead, ohead, | |
3700 | dp, end, pass); | |
3701 | if (error) | |
3702 | return error; | |
3703 | ||
67fcb7bf | 3704 | dp += be32_to_cpu(ohead->oh_len); |
1da177e4 LT |
3705 | num_logops--; |
3706 | } | |
3707 | return 0; | |
3708 | } | |
3709 | ||
3710 | /* | |
3711 | * Process an extent free intent item that was recovered from | |
3712 | * the log. We need to free the extents that it describes. | |
3713 | */ | |
3c1e2bbe | 3714 | STATIC int |
1da177e4 LT |
3715 | xlog_recover_process_efi( |
3716 | xfs_mount_t *mp, | |
3717 | xfs_efi_log_item_t *efip) | |
3718 | { | |
3719 | xfs_efd_log_item_t *efdp; | |
3720 | xfs_trans_t *tp; | |
3721 | int i; | |
3c1e2bbe | 3722 | int error = 0; |
1da177e4 LT |
3723 | xfs_extent_t *extp; |
3724 | xfs_fsblock_t startblock_fsb; | |
3725 | ||
b199c8a4 | 3726 | ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)); |
1da177e4 LT |
3727 | |
3728 | /* | |
3729 | * First check the validity of the extents described by the | |
3730 | * EFI. If any are bad, then assume that all are bad and | |
3731 | * just toss the EFI. | |
3732 | */ | |
3733 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3734 | extp = &(efip->efi_format.efi_extents[i]); | |
3735 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
3736 | XFS_FSB_TO_DADDR(mp, extp->ext_start)); | |
3737 | if ((startblock_fsb == 0) || | |
3738 | (extp->ext_len == 0) || | |
3739 | (startblock_fsb >= mp->m_sb.sb_dblocks) || | |
3740 | (extp->ext_len >= mp->m_sb.sb_agblocks)) { | |
3741 | /* | |
3742 | * This will pull the EFI from the AIL and | |
3743 | * free the memory associated with it. | |
3744 | */ | |
666d644c | 3745 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
1da177e4 | 3746 | xfs_efi_release(efip, efip->efi_format.efi_nextents); |
2451337d | 3747 | return -EIO; |
1da177e4 LT |
3748 | } |
3749 | } | |
3750 | ||
3751 | tp = xfs_trans_alloc(mp, 0); | |
3d3c8b52 | 3752 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); |
fc6149d8 DC |
3753 | if (error) |
3754 | goto abort_error; | |
1da177e4 LT |
3755 | efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents); |
3756 | ||
3757 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
3758 | extp = &(efip->efi_format.efi_extents[i]); | |
fc6149d8 DC |
3759 | error = xfs_free_extent(tp, extp->ext_start, extp->ext_len); |
3760 | if (error) | |
3761 | goto abort_error; | |
1da177e4 LT |
3762 | xfs_trans_log_efd_extent(tp, efdp, extp->ext_start, |
3763 | extp->ext_len); | |
3764 | } | |
3765 | ||
b199c8a4 | 3766 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
e5720eec | 3767 | error = xfs_trans_commit(tp, 0); |
3c1e2bbe | 3768 | return error; |
fc6149d8 DC |
3769 | |
3770 | abort_error: | |
3771 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3772 | return error; | |
1da177e4 LT |
3773 | } |
3774 | ||
1da177e4 LT |
3775 | /* |
3776 | * When this is called, all of the EFIs which did not have | |
3777 | * corresponding EFDs should be in the AIL. What we do now | |
3778 | * is free the extents associated with each one. | |
3779 | * | |
3780 | * Since we process the EFIs in normal transactions, they | |
3781 | * will be removed at some point after the commit. This prevents | |
3782 | * us from just walking down the list processing each one. | |
3783 | * We'll use a flag in the EFI to skip those that we've already | |
3784 | * processed and use the AIL iteration mechanism's generation | |
3785 | * count to try to speed this up at least a bit. | |
3786 | * | |
3787 | * When we start, we know that the EFIs are the only things in | |
3788 | * the AIL. As we process them, however, other items are added | |
3789 | * to the AIL. Since everything added to the AIL must come after | |
3790 | * everything already in the AIL, we stop processing as soon as | |
3791 | * we see something other than an EFI in the AIL. | |
3792 | */ | |
3c1e2bbe | 3793 | STATIC int |
1da177e4 | 3794 | xlog_recover_process_efis( |
9a8d2fdb | 3795 | struct xlog *log) |
1da177e4 LT |
3796 | { |
3797 | xfs_log_item_t *lip; | |
3798 | xfs_efi_log_item_t *efip; | |
3c1e2bbe | 3799 | int error = 0; |
27d8d5fe | 3800 | struct xfs_ail_cursor cur; |
a9c21c1b | 3801 | struct xfs_ail *ailp; |
1da177e4 | 3802 | |
a9c21c1b DC |
3803 | ailp = log->l_ailp; |
3804 | spin_lock(&ailp->xa_lock); | |
3805 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
3806 | while (lip != NULL) { |
3807 | /* | |
3808 | * We're done when we see something other than an EFI. | |
27d8d5fe | 3809 | * There should be no EFIs left in the AIL now. |
1da177e4 LT |
3810 | */ |
3811 | if (lip->li_type != XFS_LI_EFI) { | |
27d8d5fe | 3812 | #ifdef DEBUG |
a9c21c1b | 3813 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
27d8d5fe DC |
3814 | ASSERT(lip->li_type != XFS_LI_EFI); |
3815 | #endif | |
1da177e4 LT |
3816 | break; |
3817 | } | |
3818 | ||
3819 | /* | |
3820 | * Skip EFIs that we've already processed. | |
3821 | */ | |
3822 | efip = (xfs_efi_log_item_t *)lip; | |
b199c8a4 | 3823 | if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) { |
a9c21c1b | 3824 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 LT |
3825 | continue; |
3826 | } | |
3827 | ||
a9c21c1b DC |
3828 | spin_unlock(&ailp->xa_lock); |
3829 | error = xlog_recover_process_efi(log->l_mp, efip); | |
3830 | spin_lock(&ailp->xa_lock); | |
27d8d5fe DC |
3831 | if (error) |
3832 | goto out; | |
a9c21c1b | 3833 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 3834 | } |
27d8d5fe | 3835 | out: |
e4a1e29c | 3836 | xfs_trans_ail_cursor_done(&cur); |
a9c21c1b | 3837 | spin_unlock(&ailp->xa_lock); |
3c1e2bbe | 3838 | return error; |
1da177e4 LT |
3839 | } |
3840 | ||
3841 | /* | |
3842 | * This routine performs a transaction to null out a bad inode pointer | |
3843 | * in an agi unlinked inode hash bucket. | |
3844 | */ | |
3845 | STATIC void | |
3846 | xlog_recover_clear_agi_bucket( | |
3847 | xfs_mount_t *mp, | |
3848 | xfs_agnumber_t agno, | |
3849 | int bucket) | |
3850 | { | |
3851 | xfs_trans_t *tp; | |
3852 | xfs_agi_t *agi; | |
3853 | xfs_buf_t *agibp; | |
3854 | int offset; | |
3855 | int error; | |
3856 | ||
3857 | tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET); | |
3d3c8b52 | 3858 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0); |
e5720eec DC |
3859 | if (error) |
3860 | goto out_abort; | |
1da177e4 | 3861 | |
5e1be0fb CH |
3862 | error = xfs_read_agi(mp, tp, agno, &agibp); |
3863 | if (error) | |
e5720eec | 3864 | goto out_abort; |
1da177e4 | 3865 | |
5e1be0fb | 3866 | agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 3867 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
3868 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
3869 | (sizeof(xfs_agino_t) * bucket); | |
3870 | xfs_trans_log_buf(tp, agibp, offset, | |
3871 | (offset + sizeof(xfs_agino_t) - 1)); | |
3872 | ||
e5720eec DC |
3873 | error = xfs_trans_commit(tp, 0); |
3874 | if (error) | |
3875 | goto out_error; | |
3876 | return; | |
3877 | ||
3878 | out_abort: | |
3879 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); | |
3880 | out_error: | |
a0fa2b67 | 3881 | xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno); |
e5720eec | 3882 | return; |
1da177e4 LT |
3883 | } |
3884 | ||
23fac50f CH |
3885 | STATIC xfs_agino_t |
3886 | xlog_recover_process_one_iunlink( | |
3887 | struct xfs_mount *mp, | |
3888 | xfs_agnumber_t agno, | |
3889 | xfs_agino_t agino, | |
3890 | int bucket) | |
3891 | { | |
3892 | struct xfs_buf *ibp; | |
3893 | struct xfs_dinode *dip; | |
3894 | struct xfs_inode *ip; | |
3895 | xfs_ino_t ino; | |
3896 | int error; | |
3897 | ||
3898 | ino = XFS_AGINO_TO_INO(mp, agno, agino); | |
7b6259e7 | 3899 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip); |
23fac50f CH |
3900 | if (error) |
3901 | goto fail; | |
3902 | ||
3903 | /* | |
3904 | * Get the on disk inode to find the next inode in the bucket. | |
3905 | */ | |
475ee413 | 3906 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0); |
23fac50f | 3907 | if (error) |
0e446673 | 3908 | goto fail_iput; |
23fac50f | 3909 | |
23fac50f | 3910 | ASSERT(ip->i_d.di_nlink == 0); |
0e446673 | 3911 | ASSERT(ip->i_d.di_mode != 0); |
23fac50f CH |
3912 | |
3913 | /* setup for the next pass */ | |
3914 | agino = be32_to_cpu(dip->di_next_unlinked); | |
3915 | xfs_buf_relse(ibp); | |
3916 | ||
3917 | /* | |
3918 | * Prevent any DMAPI event from being sent when the reference on | |
3919 | * the inode is dropped. | |
3920 | */ | |
3921 | ip->i_d.di_dmevmask = 0; | |
3922 | ||
0e446673 | 3923 | IRELE(ip); |
23fac50f CH |
3924 | return agino; |
3925 | ||
0e446673 CH |
3926 | fail_iput: |
3927 | IRELE(ip); | |
23fac50f CH |
3928 | fail: |
3929 | /* | |
3930 | * We can't read in the inode this bucket points to, or this inode | |
3931 | * is messed up. Just ditch this bucket of inodes. We will lose | |
3932 | * some inodes and space, but at least we won't hang. | |
3933 | * | |
3934 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to | |
3935 | * clear the inode pointer in the bucket. | |
3936 | */ | |
3937 | xlog_recover_clear_agi_bucket(mp, agno, bucket); | |
3938 | return NULLAGINO; | |
3939 | } | |
3940 | ||
1da177e4 LT |
3941 | /* |
3942 | * xlog_iunlink_recover | |
3943 | * | |
3944 | * This is called during recovery to process any inodes which | |
3945 | * we unlinked but not freed when the system crashed. These | |
3946 | * inodes will be on the lists in the AGI blocks. What we do | |
3947 | * here is scan all the AGIs and fully truncate and free any | |
3948 | * inodes found on the lists. Each inode is removed from the | |
3949 | * lists when it has been fully truncated and is freed. The | |
3950 | * freeing of the inode and its removal from the list must be | |
3951 | * atomic. | |
3952 | */ | |
d96f8f89 | 3953 | STATIC void |
1da177e4 | 3954 | xlog_recover_process_iunlinks( |
9a8d2fdb | 3955 | struct xlog *log) |
1da177e4 LT |
3956 | { |
3957 | xfs_mount_t *mp; | |
3958 | xfs_agnumber_t agno; | |
3959 | xfs_agi_t *agi; | |
3960 | xfs_buf_t *agibp; | |
1da177e4 | 3961 | xfs_agino_t agino; |
1da177e4 LT |
3962 | int bucket; |
3963 | int error; | |
3964 | uint mp_dmevmask; | |
3965 | ||
3966 | mp = log->l_mp; | |
3967 | ||
3968 | /* | |
3969 | * Prevent any DMAPI event from being sent while in this function. | |
3970 | */ | |
3971 | mp_dmevmask = mp->m_dmevmask; | |
3972 | mp->m_dmevmask = 0; | |
3973 | ||
3974 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
3975 | /* | |
3976 | * Find the agi for this ag. | |
3977 | */ | |
5e1be0fb CH |
3978 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
3979 | if (error) { | |
3980 | /* | |
3981 | * AGI is b0rked. Don't process it. | |
3982 | * | |
3983 | * We should probably mark the filesystem as corrupt | |
3984 | * after we've recovered all the ag's we can.... | |
3985 | */ | |
3986 | continue; | |
1da177e4 | 3987 | } |
d97d32ed JK |
3988 | /* |
3989 | * Unlock the buffer so that it can be acquired in the normal | |
3990 | * course of the transaction to truncate and free each inode. | |
3991 | * Because we are not racing with anyone else here for the AGI | |
3992 | * buffer, we don't even need to hold it locked to read the | |
3993 | * initial unlinked bucket entries out of the buffer. We keep | |
3994 | * buffer reference though, so that it stays pinned in memory | |
3995 | * while we need the buffer. | |
3996 | */ | |
1da177e4 | 3997 | agi = XFS_BUF_TO_AGI(agibp); |
d97d32ed | 3998 | xfs_buf_unlock(agibp); |
1da177e4 LT |
3999 | |
4000 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { | |
16259e7d | 4001 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
1da177e4 | 4002 | while (agino != NULLAGINO) { |
23fac50f CH |
4003 | agino = xlog_recover_process_one_iunlink(mp, |
4004 | agno, agino, bucket); | |
1da177e4 LT |
4005 | } |
4006 | } | |
d97d32ed | 4007 | xfs_buf_rele(agibp); |
1da177e4 LT |
4008 | } |
4009 | ||
4010 | mp->m_dmevmask = mp_dmevmask; | |
4011 | } | |
4012 | ||
1da177e4 | 4013 | /* |
0e446be4 CH |
4014 | * Upack the log buffer data and crc check it. If the check fails, issue a |
4015 | * warning if and only if the CRC in the header is non-zero. This makes the | |
4016 | * check an advisory warning, and the zero CRC check will prevent failure | |
4017 | * warnings from being emitted when upgrading the kernel from one that does not | |
4018 | * add CRCs by default. | |
4019 | * | |
4020 | * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log | |
4021 | * corruption failure | |
1da177e4 | 4022 | */ |
0e446be4 CH |
4023 | STATIC int |
4024 | xlog_unpack_data_crc( | |
4025 | struct xlog_rec_header *rhead, | |
4026 | xfs_caddr_t dp, | |
4027 | struct xlog *log) | |
1da177e4 | 4028 | { |
f9668a09 | 4029 | __le32 crc; |
0e446be4 CH |
4030 | |
4031 | crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len)); | |
4032 | if (crc != rhead->h_crc) { | |
4033 | if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) { | |
4034 | xfs_alert(log->l_mp, | |
08e96e1a | 4035 | "log record CRC mismatch: found 0x%x, expected 0x%x.", |
f9668a09 DC |
4036 | le32_to_cpu(rhead->h_crc), |
4037 | le32_to_cpu(crc)); | |
0e446be4 | 4038 | xfs_hex_dump(dp, 32); |
1da177e4 LT |
4039 | } |
4040 | ||
0e446be4 CH |
4041 | /* |
4042 | * If we've detected a log record corruption, then we can't | |
4043 | * recover past this point. Abort recovery if we are enforcing | |
4044 | * CRC protection by punting an error back up the stack. | |
4045 | */ | |
4046 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) | |
2451337d | 4047 | return -EFSCORRUPTED; |
1da177e4 | 4048 | } |
0e446be4 CH |
4049 | |
4050 | return 0; | |
1da177e4 LT |
4051 | } |
4052 | ||
0e446be4 | 4053 | STATIC int |
1da177e4 | 4054 | xlog_unpack_data( |
9a8d2fdb | 4055 | struct xlog_rec_header *rhead, |
1da177e4 | 4056 | xfs_caddr_t dp, |
9a8d2fdb | 4057 | struct xlog *log) |
1da177e4 LT |
4058 | { |
4059 | int i, j, k; | |
0e446be4 CH |
4060 | int error; |
4061 | ||
4062 | error = xlog_unpack_data_crc(rhead, dp, log); | |
4063 | if (error) | |
4064 | return error; | |
1da177e4 | 4065 | |
b53e675d | 4066 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
1da177e4 | 4067 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
b53e675d | 4068 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
1da177e4 LT |
4069 | dp += BBSIZE; |
4070 | } | |
4071 | ||
62118709 | 4072 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b28708d6 | 4073 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
b53e675d | 4074 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
1da177e4 LT |
4075 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
4076 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 4077 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
1da177e4 LT |
4078 | dp += BBSIZE; |
4079 | } | |
4080 | } | |
0e446be4 CH |
4081 | |
4082 | return 0; | |
1da177e4 LT |
4083 | } |
4084 | ||
4085 | STATIC int | |
4086 | xlog_valid_rec_header( | |
9a8d2fdb MT |
4087 | struct xlog *log, |
4088 | struct xlog_rec_header *rhead, | |
1da177e4 LT |
4089 | xfs_daddr_t blkno) |
4090 | { | |
4091 | int hlen; | |
4092 | ||
69ef921b | 4093 | if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) { |
1da177e4 LT |
4094 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
4095 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 4096 | return -EFSCORRUPTED; |
1da177e4 LT |
4097 | } |
4098 | if (unlikely( | |
4099 | (!rhead->h_version || | |
b53e675d | 4100 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
a0fa2b67 | 4101 | xfs_warn(log->l_mp, "%s: unrecognised log version (%d).", |
34a622b2 | 4102 | __func__, be32_to_cpu(rhead->h_version)); |
2451337d | 4103 | return -EIO; |
1da177e4 LT |
4104 | } |
4105 | ||
4106 | /* LR body must have data or it wouldn't have been written */ | |
b53e675d | 4107 | hlen = be32_to_cpu(rhead->h_len); |
1da177e4 LT |
4108 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
4109 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", | |
4110 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 4111 | return -EFSCORRUPTED; |
1da177e4 LT |
4112 | } |
4113 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { | |
4114 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", | |
4115 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 4116 | return -EFSCORRUPTED; |
1da177e4 LT |
4117 | } |
4118 | return 0; | |
4119 | } | |
4120 | ||
4121 | /* | |
4122 | * Read the log from tail to head and process the log records found. | |
4123 | * Handle the two cases where the tail and head are in the same cycle | |
4124 | * and where the active portion of the log wraps around the end of | |
4125 | * the physical log separately. The pass parameter is passed through | |
4126 | * to the routines called to process the data and is not looked at | |
4127 | * here. | |
4128 | */ | |
4129 | STATIC int | |
4130 | xlog_do_recovery_pass( | |
9a8d2fdb | 4131 | struct xlog *log, |
1da177e4 LT |
4132 | xfs_daddr_t head_blk, |
4133 | xfs_daddr_t tail_blk, | |
4134 | int pass) | |
4135 | { | |
4136 | xlog_rec_header_t *rhead; | |
4137 | xfs_daddr_t blk_no; | |
fc5bc4c8 | 4138 | xfs_caddr_t offset; |
1da177e4 LT |
4139 | xfs_buf_t *hbp, *dbp; |
4140 | int error = 0, h_size; | |
4141 | int bblks, split_bblks; | |
4142 | int hblks, split_hblks, wrapped_hblks; | |
f0a76953 | 4143 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
1da177e4 LT |
4144 | |
4145 | ASSERT(head_blk != tail_blk); | |
4146 | ||
4147 | /* | |
4148 | * Read the header of the tail block and get the iclog buffer size from | |
4149 | * h_size. Use this to tell how many sectors make up the log header. | |
4150 | */ | |
62118709 | 4151 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
1da177e4 LT |
4152 | /* |
4153 | * When using variable length iclogs, read first sector of | |
4154 | * iclog header and extract the header size from it. Get a | |
4155 | * new hbp that is the correct size. | |
4156 | */ | |
4157 | hbp = xlog_get_bp(log, 1); | |
4158 | if (!hbp) | |
2451337d | 4159 | return -ENOMEM; |
076e6acb CH |
4160 | |
4161 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); | |
4162 | if (error) | |
1da177e4 | 4163 | goto bread_err1; |
076e6acb | 4164 | |
1da177e4 LT |
4165 | rhead = (xlog_rec_header_t *)offset; |
4166 | error = xlog_valid_rec_header(log, rhead, tail_blk); | |
4167 | if (error) | |
4168 | goto bread_err1; | |
b53e675d CH |
4169 | h_size = be32_to_cpu(rhead->h_size); |
4170 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && | |
1da177e4 LT |
4171 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
4172 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
4173 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
4174 | hblks++; | |
4175 | xlog_put_bp(hbp); | |
4176 | hbp = xlog_get_bp(log, hblks); | |
4177 | } else { | |
4178 | hblks = 1; | |
4179 | } | |
4180 | } else { | |
69ce58f0 | 4181 | ASSERT(log->l_sectBBsize == 1); |
1da177e4 LT |
4182 | hblks = 1; |
4183 | hbp = xlog_get_bp(log, 1); | |
4184 | h_size = XLOG_BIG_RECORD_BSIZE; | |
4185 | } | |
4186 | ||
4187 | if (!hbp) | |
2451337d | 4188 | return -ENOMEM; |
1da177e4 LT |
4189 | dbp = xlog_get_bp(log, BTOBB(h_size)); |
4190 | if (!dbp) { | |
4191 | xlog_put_bp(hbp); | |
2451337d | 4192 | return -ENOMEM; |
1da177e4 LT |
4193 | } |
4194 | ||
4195 | memset(rhash, 0, sizeof(rhash)); | |
4196 | if (tail_blk <= head_blk) { | |
4197 | for (blk_no = tail_blk; blk_no < head_blk; ) { | |
076e6acb CH |
4198 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
4199 | if (error) | |
1da177e4 | 4200 | goto bread_err2; |
076e6acb | 4201 | |
1da177e4 LT |
4202 | rhead = (xlog_rec_header_t *)offset; |
4203 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
4204 | if (error) | |
4205 | goto bread_err2; | |
4206 | ||
4207 | /* blocks in data section */ | |
b53e675d | 4208 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
4209 | error = xlog_bread(log, blk_no + hblks, bblks, dbp, |
4210 | &offset); | |
1da177e4 LT |
4211 | if (error) |
4212 | goto bread_err2; | |
076e6acb | 4213 | |
0e446be4 CH |
4214 | error = xlog_unpack_data(rhead, offset, log); |
4215 | if (error) | |
4216 | goto bread_err2; | |
4217 | ||
4218 | error = xlog_recover_process_data(log, | |
4219 | rhash, rhead, offset, pass); | |
4220 | if (error) | |
1da177e4 LT |
4221 | goto bread_err2; |
4222 | blk_no += bblks + hblks; | |
4223 | } | |
4224 | } else { | |
4225 | /* | |
4226 | * Perform recovery around the end of the physical log. | |
4227 | * When the head is not on the same cycle number as the tail, | |
4228 | * we can't do a sequential recovery as above. | |
4229 | */ | |
4230 | blk_no = tail_blk; | |
4231 | while (blk_no < log->l_logBBsize) { | |
4232 | /* | |
4233 | * Check for header wrapping around physical end-of-log | |
4234 | */ | |
62926044 | 4235 | offset = hbp->b_addr; |
1da177e4 LT |
4236 | split_hblks = 0; |
4237 | wrapped_hblks = 0; | |
4238 | if (blk_no + hblks <= log->l_logBBsize) { | |
4239 | /* Read header in one read */ | |
076e6acb CH |
4240 | error = xlog_bread(log, blk_no, hblks, hbp, |
4241 | &offset); | |
1da177e4 LT |
4242 | if (error) |
4243 | goto bread_err2; | |
1da177e4 LT |
4244 | } else { |
4245 | /* This LR is split across physical log end */ | |
4246 | if (blk_no != log->l_logBBsize) { | |
4247 | /* some data before physical log end */ | |
4248 | ASSERT(blk_no <= INT_MAX); | |
4249 | split_hblks = log->l_logBBsize - (int)blk_no; | |
4250 | ASSERT(split_hblks > 0); | |
076e6acb CH |
4251 | error = xlog_bread(log, blk_no, |
4252 | split_hblks, hbp, | |
4253 | &offset); | |
4254 | if (error) | |
1da177e4 | 4255 | goto bread_err2; |
1da177e4 | 4256 | } |
076e6acb | 4257 | |
1da177e4 LT |
4258 | /* |
4259 | * Note: this black magic still works with | |
4260 | * large sector sizes (non-512) only because: | |
4261 | * - we increased the buffer size originally | |
4262 | * by 1 sector giving us enough extra space | |
4263 | * for the second read; | |
4264 | * - the log start is guaranteed to be sector | |
4265 | * aligned; | |
4266 | * - we read the log end (LR header start) | |
4267 | * _first_, then the log start (LR header end) | |
4268 | * - order is important. | |
4269 | */ | |
234f56ac | 4270 | wrapped_hblks = hblks - split_hblks; |
44396476 DC |
4271 | error = xlog_bread_offset(log, 0, |
4272 | wrapped_hblks, hbp, | |
4273 | offset + BBTOB(split_hblks)); | |
1da177e4 LT |
4274 | if (error) |
4275 | goto bread_err2; | |
1da177e4 LT |
4276 | } |
4277 | rhead = (xlog_rec_header_t *)offset; | |
4278 | error = xlog_valid_rec_header(log, rhead, | |
4279 | split_hblks ? blk_no : 0); | |
4280 | if (error) | |
4281 | goto bread_err2; | |
4282 | ||
b53e675d | 4283 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
1da177e4 LT |
4284 | blk_no += hblks; |
4285 | ||
4286 | /* Read in data for log record */ | |
4287 | if (blk_no + bblks <= log->l_logBBsize) { | |
076e6acb CH |
4288 | error = xlog_bread(log, blk_no, bblks, dbp, |
4289 | &offset); | |
1da177e4 LT |
4290 | if (error) |
4291 | goto bread_err2; | |
1da177e4 LT |
4292 | } else { |
4293 | /* This log record is split across the | |
4294 | * physical end of log */ | |
62926044 | 4295 | offset = dbp->b_addr; |
1da177e4 LT |
4296 | split_bblks = 0; |
4297 | if (blk_no != log->l_logBBsize) { | |
4298 | /* some data is before the physical | |
4299 | * end of log */ | |
4300 | ASSERT(!wrapped_hblks); | |
4301 | ASSERT(blk_no <= INT_MAX); | |
4302 | split_bblks = | |
4303 | log->l_logBBsize - (int)blk_no; | |
4304 | ASSERT(split_bblks > 0); | |
076e6acb CH |
4305 | error = xlog_bread(log, blk_no, |
4306 | split_bblks, dbp, | |
4307 | &offset); | |
4308 | if (error) | |
1da177e4 | 4309 | goto bread_err2; |
1da177e4 | 4310 | } |
076e6acb | 4311 | |
1da177e4 LT |
4312 | /* |
4313 | * Note: this black magic still works with | |
4314 | * large sector sizes (non-512) only because: | |
4315 | * - we increased the buffer size originally | |
4316 | * by 1 sector giving us enough extra space | |
4317 | * for the second read; | |
4318 | * - the log start is guaranteed to be sector | |
4319 | * aligned; | |
4320 | * - we read the log end (LR header start) | |
4321 | * _first_, then the log start (LR header end) | |
4322 | * - order is important. | |
4323 | */ | |
44396476 | 4324 | error = xlog_bread_offset(log, 0, |
009507b0 | 4325 | bblks - split_bblks, dbp, |
44396476 | 4326 | offset + BBTOB(split_bblks)); |
076e6acb CH |
4327 | if (error) |
4328 | goto bread_err2; | |
1da177e4 | 4329 | } |
0e446be4 CH |
4330 | |
4331 | error = xlog_unpack_data(rhead, offset, log); | |
4332 | if (error) | |
4333 | goto bread_err2; | |
4334 | ||
4335 | error = xlog_recover_process_data(log, rhash, | |
4336 | rhead, offset, pass); | |
4337 | if (error) | |
1da177e4 LT |
4338 | goto bread_err2; |
4339 | blk_no += bblks; | |
4340 | } | |
4341 | ||
4342 | ASSERT(blk_no >= log->l_logBBsize); | |
4343 | blk_no -= log->l_logBBsize; | |
4344 | ||
4345 | /* read first part of physical log */ | |
4346 | while (blk_no < head_blk) { | |
076e6acb CH |
4347 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
4348 | if (error) | |
1da177e4 | 4349 | goto bread_err2; |
076e6acb | 4350 | |
1da177e4 LT |
4351 | rhead = (xlog_rec_header_t *)offset; |
4352 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
4353 | if (error) | |
4354 | goto bread_err2; | |
076e6acb | 4355 | |
b53e675d | 4356 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
076e6acb CH |
4357 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, |
4358 | &offset); | |
4359 | if (error) | |
1da177e4 | 4360 | goto bread_err2; |
076e6acb | 4361 | |
0e446be4 CH |
4362 | error = xlog_unpack_data(rhead, offset, log); |
4363 | if (error) | |
4364 | goto bread_err2; | |
4365 | ||
4366 | error = xlog_recover_process_data(log, rhash, | |
4367 | rhead, offset, pass); | |
4368 | if (error) | |
1da177e4 LT |
4369 | goto bread_err2; |
4370 | blk_no += bblks + hblks; | |
4371 | } | |
4372 | } | |
4373 | ||
4374 | bread_err2: | |
4375 | xlog_put_bp(dbp); | |
4376 | bread_err1: | |
4377 | xlog_put_bp(hbp); | |
4378 | return error; | |
4379 | } | |
4380 | ||
4381 | /* | |
4382 | * Do the recovery of the log. We actually do this in two phases. | |
4383 | * The two passes are necessary in order to implement the function | |
4384 | * of cancelling a record written into the log. The first pass | |
4385 | * determines those things which have been cancelled, and the | |
4386 | * second pass replays log items normally except for those which | |
4387 | * have been cancelled. The handling of the replay and cancellations | |
4388 | * takes place in the log item type specific routines. | |
4389 | * | |
4390 | * The table of items which have cancel records in the log is allocated | |
4391 | * and freed at this level, since only here do we know when all of | |
4392 | * the log recovery has been completed. | |
4393 | */ | |
4394 | STATIC int | |
4395 | xlog_do_log_recovery( | |
9a8d2fdb | 4396 | struct xlog *log, |
1da177e4 LT |
4397 | xfs_daddr_t head_blk, |
4398 | xfs_daddr_t tail_blk) | |
4399 | { | |
d5689eaa | 4400 | int error, i; |
1da177e4 LT |
4401 | |
4402 | ASSERT(head_blk != tail_blk); | |
4403 | ||
4404 | /* | |
4405 | * First do a pass to find all of the cancelled buf log items. | |
4406 | * Store them in the buf_cancel_table for use in the second pass. | |
4407 | */ | |
d5689eaa CH |
4408 | log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE * |
4409 | sizeof(struct list_head), | |
1da177e4 | 4410 | KM_SLEEP); |
d5689eaa CH |
4411 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
4412 | INIT_LIST_HEAD(&log->l_buf_cancel_table[i]); | |
4413 | ||
1da177e4 LT |
4414 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
4415 | XLOG_RECOVER_PASS1); | |
4416 | if (error != 0) { | |
f0e2d93c | 4417 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
4418 | log->l_buf_cancel_table = NULL; |
4419 | return error; | |
4420 | } | |
4421 | /* | |
4422 | * Then do a second pass to actually recover the items in the log. | |
4423 | * When it is complete free the table of buf cancel items. | |
4424 | */ | |
4425 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
4426 | XLOG_RECOVER_PASS2); | |
4427 | #ifdef DEBUG | |
6d192a9b | 4428 | if (!error) { |
1da177e4 LT |
4429 | int i; |
4430 | ||
4431 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
d5689eaa | 4432 | ASSERT(list_empty(&log->l_buf_cancel_table[i])); |
1da177e4 LT |
4433 | } |
4434 | #endif /* DEBUG */ | |
4435 | ||
f0e2d93c | 4436 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
4437 | log->l_buf_cancel_table = NULL; |
4438 | ||
4439 | return error; | |
4440 | } | |
4441 | ||
4442 | /* | |
4443 | * Do the actual recovery | |
4444 | */ | |
4445 | STATIC int | |
4446 | xlog_do_recover( | |
9a8d2fdb | 4447 | struct xlog *log, |
1da177e4 LT |
4448 | xfs_daddr_t head_blk, |
4449 | xfs_daddr_t tail_blk) | |
4450 | { | |
4451 | int error; | |
4452 | xfs_buf_t *bp; | |
4453 | xfs_sb_t *sbp; | |
4454 | ||
4455 | /* | |
4456 | * First replay the images in the log. | |
4457 | */ | |
4458 | error = xlog_do_log_recovery(log, head_blk, tail_blk); | |
43ff2122 | 4459 | if (error) |
1da177e4 | 4460 | return error; |
1da177e4 LT |
4461 | |
4462 | /* | |
4463 | * If IO errors happened during recovery, bail out. | |
4464 | */ | |
4465 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) { | |
2451337d | 4466 | return -EIO; |
1da177e4 LT |
4467 | } |
4468 | ||
4469 | /* | |
4470 | * We now update the tail_lsn since much of the recovery has completed | |
4471 | * and there may be space available to use. If there were no extent | |
4472 | * or iunlinks, we can free up the entire log and set the tail_lsn to | |
4473 | * be the last_sync_lsn. This was set in xlog_find_tail to be the | |
4474 | * lsn of the last known good LR on disk. If there are extent frees | |
4475 | * or iunlinks they will have some entries in the AIL; so we look at | |
4476 | * the AIL to determine how to set the tail_lsn. | |
4477 | */ | |
4478 | xlog_assign_tail_lsn(log->l_mp); | |
4479 | ||
4480 | /* | |
4481 | * Now that we've finished replaying all buffer and inode | |
98021821 | 4482 | * updates, re-read in the superblock and reverify it. |
1da177e4 LT |
4483 | */ |
4484 | bp = xfs_getsb(log->l_mp, 0); | |
4485 | XFS_BUF_UNDONE(bp); | |
bebf963f | 4486 | ASSERT(!(XFS_BUF_ISWRITE(bp))); |
1da177e4 | 4487 | XFS_BUF_READ(bp); |
bebf963f | 4488 | XFS_BUF_UNASYNC(bp); |
1813dd64 | 4489 | bp->b_ops = &xfs_sb_buf_ops; |
83a0adc3 CH |
4490 | |
4491 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) { | |
4492 | xfs_buf_relse(bp); | |
2451337d | 4493 | return -EIO; |
83a0adc3 CH |
4494 | } |
4495 | ||
4496 | xfs_buf_iorequest(bp); | |
1a1a3e97 | 4497 | error = xfs_buf_iowait(bp); |
d64e31a2 | 4498 | if (error) { |
901796af | 4499 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
4500 | ASSERT(0); |
4501 | xfs_buf_relse(bp); | |
4502 | return error; | |
4503 | } | |
4504 | ||
4505 | /* Convert superblock from on-disk format */ | |
4506 | sbp = &log->l_mp->m_sb; | |
98021821 | 4507 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 4508 | ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC); |
62118709 | 4509 | ASSERT(xfs_sb_good_version(sbp)); |
1da177e4 LT |
4510 | xfs_buf_relse(bp); |
4511 | ||
5478eead LM |
4512 | /* We've re-read the superblock so re-initialize per-cpu counters */ |
4513 | xfs_icsb_reinit_counters(log->l_mp); | |
4514 | ||
1da177e4 LT |
4515 | xlog_recover_check_summary(log); |
4516 | ||
4517 | /* Normal transactions can now occur */ | |
4518 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
4519 | return 0; | |
4520 | } | |
4521 | ||
4522 | /* | |
4523 | * Perform recovery and re-initialize some log variables in xlog_find_tail. | |
4524 | * | |
4525 | * Return error or zero. | |
4526 | */ | |
4527 | int | |
4528 | xlog_recover( | |
9a8d2fdb | 4529 | struct xlog *log) |
1da177e4 LT |
4530 | { |
4531 | xfs_daddr_t head_blk, tail_blk; | |
4532 | int error; | |
4533 | ||
4534 | /* find the tail of the log */ | |
65be6054 | 4535 | if ((error = xlog_find_tail(log, &head_blk, &tail_blk))) |
1da177e4 LT |
4536 | return error; |
4537 | ||
4538 | if (tail_blk != head_blk) { | |
4539 | /* There used to be a comment here: | |
4540 | * | |
4541 | * disallow recovery on read-only mounts. note -- mount | |
4542 | * checks for ENOSPC and turns it into an intelligent | |
4543 | * error message. | |
4544 | * ...but this is no longer true. Now, unless you specify | |
4545 | * NORECOVERY (in which case this function would never be | |
4546 | * called), we just go ahead and recover. We do this all | |
4547 | * under the vfs layer, so we can get away with it unless | |
4548 | * the device itself is read-only, in which case we fail. | |
4549 | */ | |
3a02ee18 | 4550 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
1da177e4 LT |
4551 | return error; |
4552 | } | |
4553 | ||
e721f504 DC |
4554 | /* |
4555 | * Version 5 superblock log feature mask validation. We know the | |
4556 | * log is dirty so check if there are any unknown log features | |
4557 | * in what we need to recover. If there are unknown features | |
4558 | * (e.g. unsupported transactions, then simply reject the | |
4559 | * attempt at recovery before touching anything. | |
4560 | */ | |
4561 | if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 && | |
4562 | xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb, | |
4563 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) { | |
4564 | xfs_warn(log->l_mp, | |
4565 | "Superblock has unknown incompatible log features (0x%x) enabled.\n" | |
4566 | "The log can not be fully and/or safely recovered by this kernel.\n" | |
4567 | "Please recover the log on a kernel that supports the unknown features.", | |
4568 | (log->l_mp->m_sb.sb_features_log_incompat & | |
4569 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); | |
2451337d | 4570 | return -EINVAL; |
e721f504 DC |
4571 | } |
4572 | ||
a0fa2b67 DC |
4573 | xfs_notice(log->l_mp, "Starting recovery (logdev: %s)", |
4574 | log->l_mp->m_logname ? log->l_mp->m_logname | |
4575 | : "internal"); | |
1da177e4 LT |
4576 | |
4577 | error = xlog_do_recover(log, head_blk, tail_blk); | |
4578 | log->l_flags |= XLOG_RECOVERY_NEEDED; | |
4579 | } | |
4580 | return error; | |
4581 | } | |
4582 | ||
4583 | /* | |
4584 | * In the first part of recovery we replay inodes and buffers and build | |
4585 | * up the list of extent free items which need to be processed. Here | |
4586 | * we process the extent free items and clean up the on disk unlinked | |
4587 | * inode lists. This is separated from the first part of recovery so | |
4588 | * that the root and real-time bitmap inodes can be read in from disk in | |
4589 | * between the two stages. This is necessary so that we can free space | |
4590 | * in the real-time portion of the file system. | |
4591 | */ | |
4592 | int | |
4593 | xlog_recover_finish( | |
9a8d2fdb | 4594 | struct xlog *log) |
1da177e4 LT |
4595 | { |
4596 | /* | |
4597 | * Now we're ready to do the transactions needed for the | |
4598 | * rest of recovery. Start with completing all the extent | |
4599 | * free intent records and then process the unlinked inode | |
4600 | * lists. At this point, we essentially run in normal mode | |
4601 | * except that we're still performing recovery actions | |
4602 | * rather than accepting new requests. | |
4603 | */ | |
4604 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { | |
3c1e2bbe DC |
4605 | int error; |
4606 | error = xlog_recover_process_efis(log); | |
4607 | if (error) { | |
a0fa2b67 | 4608 | xfs_alert(log->l_mp, "Failed to recover EFIs"); |
3c1e2bbe DC |
4609 | return error; |
4610 | } | |
1da177e4 LT |
4611 | /* |
4612 | * Sync the log to get all the EFIs out of the AIL. | |
4613 | * This isn't absolutely necessary, but it helps in | |
4614 | * case the unlink transactions would have problems | |
4615 | * pushing the EFIs out of the way. | |
4616 | */ | |
a14a348b | 4617 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
1da177e4 | 4618 | |
4249023a | 4619 | xlog_recover_process_iunlinks(log); |
1da177e4 LT |
4620 | |
4621 | xlog_recover_check_summary(log); | |
4622 | ||
a0fa2b67 DC |
4623 | xfs_notice(log->l_mp, "Ending recovery (logdev: %s)", |
4624 | log->l_mp->m_logname ? log->l_mp->m_logname | |
4625 | : "internal"); | |
1da177e4 LT |
4626 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
4627 | } else { | |
a0fa2b67 | 4628 | xfs_info(log->l_mp, "Ending clean mount"); |
1da177e4 LT |
4629 | } |
4630 | return 0; | |
4631 | } | |
4632 | ||
4633 | ||
4634 | #if defined(DEBUG) | |
4635 | /* | |
4636 | * Read all of the agf and agi counters and check that they | |
4637 | * are consistent with the superblock counters. | |
4638 | */ | |
4639 | void | |
4640 | xlog_recover_check_summary( | |
9a8d2fdb | 4641 | struct xlog *log) |
1da177e4 LT |
4642 | { |
4643 | xfs_mount_t *mp; | |
4644 | xfs_agf_t *agfp; | |
1da177e4 LT |
4645 | xfs_buf_t *agfbp; |
4646 | xfs_buf_t *agibp; | |
1da177e4 LT |
4647 | xfs_agnumber_t agno; |
4648 | __uint64_t freeblks; | |
4649 | __uint64_t itotal; | |
4650 | __uint64_t ifree; | |
5e1be0fb | 4651 | int error; |
1da177e4 LT |
4652 | |
4653 | mp = log->l_mp; | |
4654 | ||
4655 | freeblks = 0LL; | |
4656 | itotal = 0LL; | |
4657 | ifree = 0LL; | |
4658 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
4805621a FCH |
4659 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
4660 | if (error) { | |
a0fa2b67 DC |
4661 | xfs_alert(mp, "%s agf read failed agno %d error %d", |
4662 | __func__, agno, error); | |
4805621a FCH |
4663 | } else { |
4664 | agfp = XFS_BUF_TO_AGF(agfbp); | |
4665 | freeblks += be32_to_cpu(agfp->agf_freeblks) + | |
4666 | be32_to_cpu(agfp->agf_flcount); | |
4667 | xfs_buf_relse(agfbp); | |
1da177e4 | 4668 | } |
1da177e4 | 4669 | |
5e1be0fb | 4670 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
a0fa2b67 DC |
4671 | if (error) { |
4672 | xfs_alert(mp, "%s agi read failed agno %d error %d", | |
4673 | __func__, agno, error); | |
4674 | } else { | |
5e1be0fb | 4675 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 4676 | |
5e1be0fb CH |
4677 | itotal += be32_to_cpu(agi->agi_count); |
4678 | ifree += be32_to_cpu(agi->agi_freecount); | |
4679 | xfs_buf_relse(agibp); | |
4680 | } | |
1da177e4 | 4681 | } |
1da177e4 LT |
4682 | } |
4683 | #endif /* DEBUG */ |