]>
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_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
57062787 | 27 | #include "xfs_da_format.h" |
9a2cc41c | 28 | #include "xfs_da_btree.h" |
1da177e4 | 29 | #include "xfs_inode.h" |
239880ef | 30 | #include "xfs_trans.h" |
239880ef | 31 | #include "xfs_log.h" |
1da177e4 | 32 | #include "xfs_log_priv.h" |
1da177e4 | 33 | #include "xfs_log_recover.h" |
a4fbe6ab | 34 | #include "xfs_inode_item.h" |
1da177e4 LT |
35 | #include "xfs_extfree_item.h" |
36 | #include "xfs_trans_priv.h" | |
a4fbe6ab DC |
37 | #include "xfs_alloc.h" |
38 | #include "xfs_ialloc.h" | |
1da177e4 | 39 | #include "xfs_quota.h" |
0e446be4 | 40 | #include "xfs_cksum.h" |
0b1b213f | 41 | #include "xfs_trace.h" |
33479e05 | 42 | #include "xfs_icache.h" |
a4fbe6ab | 43 | #include "xfs_bmap_btree.h" |
a4fbe6ab | 44 | #include "xfs_error.h" |
2b9ab5ab | 45 | #include "xfs_dir2.h" |
1da177e4 | 46 | |
fc06c6d0 DC |
47 | #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1) |
48 | ||
9a8d2fdb MT |
49 | STATIC int |
50 | xlog_find_zeroed( | |
51 | struct xlog *, | |
52 | xfs_daddr_t *); | |
53 | STATIC int | |
54 | xlog_clear_stale_blocks( | |
55 | struct xlog *, | |
56 | xfs_lsn_t); | |
1da177e4 | 57 | #if defined(DEBUG) |
9a8d2fdb MT |
58 | STATIC void |
59 | xlog_recover_check_summary( | |
60 | struct xlog *); | |
1da177e4 LT |
61 | #else |
62 | #define xlog_recover_check_summary(log) | |
1da177e4 | 63 | #endif |
7088c413 BF |
64 | STATIC int |
65 | xlog_do_recovery_pass( | |
66 | struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *); | |
1da177e4 | 67 | |
d5689eaa CH |
68 | /* |
69 | * This structure is used during recovery to record the buf log items which | |
70 | * have been canceled and should not be replayed. | |
71 | */ | |
72 | struct xfs_buf_cancel { | |
73 | xfs_daddr_t bc_blkno; | |
74 | uint bc_len; | |
75 | int bc_refcount; | |
76 | struct list_head bc_list; | |
77 | }; | |
78 | ||
1da177e4 LT |
79 | /* |
80 | * Sector aligned buffer routines for buffer create/read/write/access | |
81 | */ | |
82 | ||
ff30a622 AE |
83 | /* |
84 | * Verify the given count of basic blocks is valid number of blocks | |
85 | * to specify for an operation involving the given XFS log buffer. | |
86 | * Returns nonzero if the count is valid, 0 otherwise. | |
87 | */ | |
88 | ||
89 | static inline int | |
90 | xlog_buf_bbcount_valid( | |
9a8d2fdb | 91 | struct xlog *log, |
ff30a622 AE |
92 | int bbcount) |
93 | { | |
94 | return bbcount > 0 && bbcount <= log->l_logBBsize; | |
95 | } | |
96 | ||
36adecff AE |
97 | /* |
98 | * Allocate a buffer to hold log data. The buffer needs to be able | |
99 | * to map to a range of nbblks basic blocks at any valid (basic | |
100 | * block) offset within the log. | |
101 | */ | |
5d77c0dc | 102 | STATIC xfs_buf_t * |
1da177e4 | 103 | xlog_get_bp( |
9a8d2fdb | 104 | struct xlog *log, |
3228149c | 105 | int nbblks) |
1da177e4 | 106 | { |
c8da0faf CH |
107 | struct xfs_buf *bp; |
108 | ||
ff30a622 | 109 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 110 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
111 | nbblks); |
112 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
113 | return NULL; |
114 | } | |
1da177e4 | 115 | |
36adecff AE |
116 | /* |
117 | * We do log I/O in units of log sectors (a power-of-2 | |
118 | * multiple of the basic block size), so we round up the | |
25985edc | 119 | * requested size to accommodate the basic blocks required |
36adecff AE |
120 | * for complete log sectors. |
121 | * | |
122 | * In addition, the buffer may be used for a non-sector- | |
123 | * aligned block offset, in which case an I/O of the | |
124 | * requested size could extend beyond the end of the | |
125 | * buffer. If the requested size is only 1 basic block it | |
126 | * will never straddle a sector boundary, so this won't be | |
127 | * an issue. Nor will this be a problem if the log I/O is | |
128 | * done in basic blocks (sector size 1). But otherwise we | |
129 | * extend the buffer by one extra log sector to ensure | |
25985edc | 130 | * there's space to accommodate this possibility. |
36adecff | 131 | */ |
69ce58f0 AE |
132 | if (nbblks > 1 && log->l_sectBBsize > 1) |
133 | nbblks += log->l_sectBBsize; | |
134 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
36adecff | 135 | |
e70b73f8 | 136 | bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0); |
c8da0faf CH |
137 | if (bp) |
138 | xfs_buf_unlock(bp); | |
139 | return bp; | |
1da177e4 LT |
140 | } |
141 | ||
5d77c0dc | 142 | STATIC void |
1da177e4 LT |
143 | xlog_put_bp( |
144 | xfs_buf_t *bp) | |
145 | { | |
146 | xfs_buf_free(bp); | |
147 | } | |
148 | ||
48389ef1 AE |
149 | /* |
150 | * Return the address of the start of the given block number's data | |
151 | * in a log buffer. The buffer covers a log sector-aligned region. | |
152 | */ | |
b2a922cd | 153 | STATIC char * |
076e6acb | 154 | xlog_align( |
9a8d2fdb | 155 | struct xlog *log, |
076e6acb CH |
156 | xfs_daddr_t blk_no, |
157 | int nbblks, | |
9a8d2fdb | 158 | struct xfs_buf *bp) |
076e6acb | 159 | { |
fdc07f44 | 160 | xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1); |
076e6acb | 161 | |
4e94b71b | 162 | ASSERT(offset + nbblks <= bp->b_length); |
62926044 | 163 | return bp->b_addr + BBTOB(offset); |
076e6acb CH |
164 | } |
165 | ||
1da177e4 LT |
166 | |
167 | /* | |
168 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. | |
169 | */ | |
076e6acb CH |
170 | STATIC int |
171 | xlog_bread_noalign( | |
9a8d2fdb | 172 | struct xlog *log, |
1da177e4 LT |
173 | xfs_daddr_t blk_no, |
174 | int nbblks, | |
9a8d2fdb | 175 | struct xfs_buf *bp) |
1da177e4 LT |
176 | { |
177 | int error; | |
178 | ||
ff30a622 | 179 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 180 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
181 | nbblks); |
182 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
2451337d | 183 | return -EFSCORRUPTED; |
3228149c DC |
184 | } |
185 | ||
69ce58f0 AE |
186 | blk_no = round_down(blk_no, log->l_sectBBsize); |
187 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
188 | |
189 | ASSERT(nbblks > 0); | |
4e94b71b | 190 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
191 | |
192 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
193 | XFS_BUF_READ(bp); | |
aa0e8833 | 194 | bp->b_io_length = nbblks; |
0e95f19a | 195 | bp->b_error = 0; |
1da177e4 | 196 | |
595bff75 DC |
197 | error = xfs_buf_submit_wait(bp); |
198 | if (error && !XFS_FORCED_SHUTDOWN(log->l_mp)) | |
901796af | 199 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
200 | return error; |
201 | } | |
202 | ||
076e6acb CH |
203 | STATIC int |
204 | xlog_bread( | |
9a8d2fdb | 205 | struct xlog *log, |
076e6acb CH |
206 | xfs_daddr_t blk_no, |
207 | int nbblks, | |
9a8d2fdb | 208 | struct xfs_buf *bp, |
b2a922cd | 209 | char **offset) |
076e6acb CH |
210 | { |
211 | int error; | |
212 | ||
213 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
214 | if (error) | |
215 | return error; | |
216 | ||
217 | *offset = xlog_align(log, blk_no, nbblks, bp); | |
218 | return 0; | |
219 | } | |
220 | ||
44396476 DC |
221 | /* |
222 | * Read at an offset into the buffer. Returns with the buffer in it's original | |
223 | * state regardless of the result of the read. | |
224 | */ | |
225 | STATIC int | |
226 | xlog_bread_offset( | |
9a8d2fdb | 227 | struct xlog *log, |
44396476 DC |
228 | xfs_daddr_t blk_no, /* block to read from */ |
229 | int nbblks, /* blocks to read */ | |
9a8d2fdb | 230 | struct xfs_buf *bp, |
b2a922cd | 231 | char *offset) |
44396476 | 232 | { |
b2a922cd | 233 | char *orig_offset = bp->b_addr; |
4e94b71b | 234 | int orig_len = BBTOB(bp->b_length); |
44396476 DC |
235 | int error, error2; |
236 | ||
02fe03d9 | 237 | error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks)); |
44396476 DC |
238 | if (error) |
239 | return error; | |
240 | ||
241 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
242 | ||
243 | /* must reset buffer pointer even on error */ | |
02fe03d9 | 244 | error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len); |
44396476 DC |
245 | if (error) |
246 | return error; | |
247 | return error2; | |
248 | } | |
249 | ||
1da177e4 LT |
250 | /* |
251 | * Write out the buffer at the given block for the given number of blocks. | |
252 | * The buffer is kept locked across the write and is returned locked. | |
253 | * This can only be used for synchronous log writes. | |
254 | */ | |
ba0f32d4 | 255 | STATIC int |
1da177e4 | 256 | xlog_bwrite( |
9a8d2fdb | 257 | struct xlog *log, |
1da177e4 LT |
258 | xfs_daddr_t blk_no, |
259 | int nbblks, | |
9a8d2fdb | 260 | struct xfs_buf *bp) |
1da177e4 LT |
261 | { |
262 | int error; | |
263 | ||
ff30a622 | 264 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
a0fa2b67 | 265 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
266 | nbblks); |
267 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
2451337d | 268 | return -EFSCORRUPTED; |
3228149c DC |
269 | } |
270 | ||
69ce58f0 AE |
271 | blk_no = round_down(blk_no, log->l_sectBBsize); |
272 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
273 | |
274 | ASSERT(nbblks > 0); | |
4e94b71b | 275 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
276 | |
277 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
278 | XFS_BUF_ZEROFLAGS(bp); | |
72790aa1 | 279 | xfs_buf_hold(bp); |
0c842ad4 | 280 | xfs_buf_lock(bp); |
aa0e8833 | 281 | bp->b_io_length = nbblks; |
0e95f19a | 282 | bp->b_error = 0; |
1da177e4 | 283 | |
c2b006c1 | 284 | error = xfs_bwrite(bp); |
901796af CH |
285 | if (error) |
286 | xfs_buf_ioerror_alert(bp, __func__); | |
c2b006c1 | 287 | xfs_buf_relse(bp); |
1da177e4 LT |
288 | return error; |
289 | } | |
290 | ||
1da177e4 LT |
291 | #ifdef DEBUG |
292 | /* | |
293 | * dump debug superblock and log record information | |
294 | */ | |
295 | STATIC void | |
296 | xlog_header_check_dump( | |
297 | xfs_mount_t *mp, | |
298 | xlog_rec_header_t *head) | |
299 | { | |
08e96e1a | 300 | xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d", |
03daa57c | 301 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); |
08e96e1a | 302 | xfs_debug(mp, " log : uuid = %pU, fmt = %d", |
03daa57c | 303 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); |
1da177e4 LT |
304 | } |
305 | #else | |
306 | #define xlog_header_check_dump(mp, head) | |
307 | #endif | |
308 | ||
309 | /* | |
310 | * check log record header for recovery | |
311 | */ | |
312 | STATIC int | |
313 | xlog_header_check_recover( | |
314 | xfs_mount_t *mp, | |
315 | xlog_rec_header_t *head) | |
316 | { | |
69ef921b | 317 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 LT |
318 | |
319 | /* | |
320 | * IRIX doesn't write the h_fmt field and leaves it zeroed | |
321 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover | |
322 | * a dirty log created in IRIX. | |
323 | */ | |
69ef921b | 324 | if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) { |
a0fa2b67 DC |
325 | xfs_warn(mp, |
326 | "dirty log written in incompatible format - can't recover"); | |
1da177e4 LT |
327 | xlog_header_check_dump(mp, head); |
328 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", | |
329 | XFS_ERRLEVEL_HIGH, mp); | |
2451337d | 330 | return -EFSCORRUPTED; |
1da177e4 | 331 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
a0fa2b67 DC |
332 | xfs_warn(mp, |
333 | "dirty log entry has mismatched uuid - can't recover"); | |
1da177e4 LT |
334 | xlog_header_check_dump(mp, head); |
335 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", | |
336 | XFS_ERRLEVEL_HIGH, mp); | |
2451337d | 337 | return -EFSCORRUPTED; |
1da177e4 LT |
338 | } |
339 | return 0; | |
340 | } | |
341 | ||
342 | /* | |
343 | * read the head block of the log and check the header | |
344 | */ | |
345 | STATIC int | |
346 | xlog_header_check_mount( | |
347 | xfs_mount_t *mp, | |
348 | xlog_rec_header_t *head) | |
349 | { | |
69ef921b | 350 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 LT |
351 | |
352 | if (uuid_is_nil(&head->h_fs_uuid)) { | |
353 | /* | |
354 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If | |
355 | * h_fs_uuid is nil, we assume this log was last mounted | |
356 | * by IRIX and continue. | |
357 | */ | |
a0fa2b67 | 358 | xfs_warn(mp, "nil uuid in log - IRIX style log"); |
1da177e4 | 359 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
a0fa2b67 | 360 | xfs_warn(mp, "log has mismatched uuid - can't recover"); |
1da177e4 LT |
361 | xlog_header_check_dump(mp, head); |
362 | XFS_ERROR_REPORT("xlog_header_check_mount", | |
363 | XFS_ERRLEVEL_HIGH, mp); | |
2451337d | 364 | return -EFSCORRUPTED; |
1da177e4 LT |
365 | } |
366 | return 0; | |
367 | } | |
368 | ||
369 | STATIC void | |
370 | xlog_recover_iodone( | |
371 | struct xfs_buf *bp) | |
372 | { | |
5a52c2a5 | 373 | if (bp->b_error) { |
1da177e4 LT |
374 | /* |
375 | * We're not going to bother about retrying | |
376 | * this during recovery. One strike! | |
377 | */ | |
595bff75 DC |
378 | if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
379 | xfs_buf_ioerror_alert(bp, __func__); | |
380 | xfs_force_shutdown(bp->b_target->bt_mount, | |
381 | SHUTDOWN_META_IO_ERROR); | |
382 | } | |
1da177e4 | 383 | } |
cb669ca5 | 384 | bp->b_iodone = NULL; |
e8aaba9a | 385 | xfs_buf_ioend(bp); |
1da177e4 LT |
386 | } |
387 | ||
388 | /* | |
389 | * This routine finds (to an approximation) the first block in the physical | |
390 | * log which contains the given cycle. It uses a binary search algorithm. | |
391 | * Note that the algorithm can not be perfect because the disk will not | |
392 | * necessarily be perfect. | |
393 | */ | |
a8272ce0 | 394 | STATIC int |
1da177e4 | 395 | xlog_find_cycle_start( |
9a8d2fdb MT |
396 | struct xlog *log, |
397 | struct xfs_buf *bp, | |
1da177e4 LT |
398 | xfs_daddr_t first_blk, |
399 | xfs_daddr_t *last_blk, | |
400 | uint cycle) | |
401 | { | |
b2a922cd | 402 | char *offset; |
1da177e4 | 403 | xfs_daddr_t mid_blk; |
e3bb2e30 | 404 | xfs_daddr_t end_blk; |
1da177e4 LT |
405 | uint mid_cycle; |
406 | int error; | |
407 | ||
e3bb2e30 AE |
408 | end_blk = *last_blk; |
409 | mid_blk = BLK_AVG(first_blk, end_blk); | |
410 | while (mid_blk != first_blk && mid_blk != end_blk) { | |
076e6acb CH |
411 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
412 | if (error) | |
1da177e4 | 413 | return error; |
03bea6fe | 414 | mid_cycle = xlog_get_cycle(offset); |
e3bb2e30 AE |
415 | if (mid_cycle == cycle) |
416 | end_blk = mid_blk; /* last_half_cycle == mid_cycle */ | |
417 | else | |
418 | first_blk = mid_blk; /* first_half_cycle == mid_cycle */ | |
419 | mid_blk = BLK_AVG(first_blk, end_blk); | |
1da177e4 | 420 | } |
e3bb2e30 AE |
421 | ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) || |
422 | (mid_blk == end_blk && mid_blk-1 == first_blk)); | |
423 | ||
424 | *last_blk = end_blk; | |
1da177e4 LT |
425 | |
426 | return 0; | |
427 | } | |
428 | ||
429 | /* | |
3f943d85 AE |
430 | * Check that a range of blocks does not contain stop_on_cycle_no. |
431 | * Fill in *new_blk with the block offset where such a block is | |
432 | * found, or with -1 (an invalid block number) if there is no such | |
433 | * block in the range. The scan needs to occur from front to back | |
434 | * and the pointer into the region must be updated since a later | |
435 | * routine will need to perform another test. | |
1da177e4 LT |
436 | */ |
437 | STATIC int | |
438 | xlog_find_verify_cycle( | |
9a8d2fdb | 439 | struct xlog *log, |
1da177e4 LT |
440 | xfs_daddr_t start_blk, |
441 | int nbblks, | |
442 | uint stop_on_cycle_no, | |
443 | xfs_daddr_t *new_blk) | |
444 | { | |
445 | xfs_daddr_t i, j; | |
446 | uint cycle; | |
447 | xfs_buf_t *bp; | |
448 | xfs_daddr_t bufblks; | |
b2a922cd | 449 | char *buf = NULL; |
1da177e4 LT |
450 | int error = 0; |
451 | ||
6881a229 AE |
452 | /* |
453 | * Greedily allocate a buffer big enough to handle the full | |
454 | * range of basic blocks we'll be examining. If that fails, | |
455 | * try a smaller size. We need to be able to read at least | |
456 | * a log sector, or we're out of luck. | |
457 | */ | |
1da177e4 | 458 | bufblks = 1 << ffs(nbblks); |
81158e0c DC |
459 | while (bufblks > log->l_logBBsize) |
460 | bufblks >>= 1; | |
1da177e4 | 461 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1da177e4 | 462 | bufblks >>= 1; |
69ce58f0 | 463 | if (bufblks < log->l_sectBBsize) |
2451337d | 464 | return -ENOMEM; |
1da177e4 LT |
465 | } |
466 | ||
467 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { | |
468 | int bcount; | |
469 | ||
470 | bcount = min(bufblks, (start_blk + nbblks - i)); | |
471 | ||
076e6acb CH |
472 | error = xlog_bread(log, i, bcount, bp, &buf); |
473 | if (error) | |
1da177e4 LT |
474 | goto out; |
475 | ||
1da177e4 | 476 | for (j = 0; j < bcount; j++) { |
03bea6fe | 477 | cycle = xlog_get_cycle(buf); |
1da177e4 LT |
478 | if (cycle == stop_on_cycle_no) { |
479 | *new_blk = i+j; | |
480 | goto out; | |
481 | } | |
482 | ||
483 | buf += BBSIZE; | |
484 | } | |
485 | } | |
486 | ||
487 | *new_blk = -1; | |
488 | ||
489 | out: | |
490 | xlog_put_bp(bp); | |
491 | return error; | |
492 | } | |
493 | ||
494 | /* | |
495 | * Potentially backup over partial log record write. | |
496 | * | |
497 | * In the typical case, last_blk is the number of the block directly after | |
498 | * a good log record. Therefore, we subtract one to get the block number | |
499 | * of the last block in the given buffer. extra_bblks contains the number | |
500 | * of blocks we would have read on a previous read. This happens when the | |
501 | * last log record is split over the end of the physical log. | |
502 | * | |
503 | * extra_bblks is the number of blocks potentially verified on a previous | |
504 | * call to this routine. | |
505 | */ | |
506 | STATIC int | |
507 | xlog_find_verify_log_record( | |
9a8d2fdb | 508 | struct xlog *log, |
1da177e4 LT |
509 | xfs_daddr_t start_blk, |
510 | xfs_daddr_t *last_blk, | |
511 | int extra_bblks) | |
512 | { | |
513 | xfs_daddr_t i; | |
514 | xfs_buf_t *bp; | |
b2a922cd | 515 | char *offset = NULL; |
1da177e4 LT |
516 | xlog_rec_header_t *head = NULL; |
517 | int error = 0; | |
518 | int smallmem = 0; | |
519 | int num_blks = *last_blk - start_blk; | |
520 | int xhdrs; | |
521 | ||
522 | ASSERT(start_blk != 0 || *last_blk != start_blk); | |
523 | ||
524 | if (!(bp = xlog_get_bp(log, num_blks))) { | |
525 | if (!(bp = xlog_get_bp(log, 1))) | |
2451337d | 526 | return -ENOMEM; |
1da177e4 LT |
527 | smallmem = 1; |
528 | } else { | |
076e6acb CH |
529 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
530 | if (error) | |
1da177e4 | 531 | goto out; |
1da177e4 LT |
532 | offset += ((num_blks - 1) << BBSHIFT); |
533 | } | |
534 | ||
535 | for (i = (*last_blk) - 1; i >= 0; i--) { | |
536 | if (i < start_blk) { | |
537 | /* valid log record not found */ | |
a0fa2b67 DC |
538 | xfs_warn(log->l_mp, |
539 | "Log inconsistent (didn't find previous header)"); | |
1da177e4 | 540 | ASSERT(0); |
2451337d | 541 | error = -EIO; |
1da177e4 LT |
542 | goto out; |
543 | } | |
544 | ||
545 | if (smallmem) { | |
076e6acb CH |
546 | error = xlog_bread(log, i, 1, bp, &offset); |
547 | if (error) | |
1da177e4 | 548 | goto out; |
1da177e4 LT |
549 | } |
550 | ||
551 | head = (xlog_rec_header_t *)offset; | |
552 | ||
69ef921b | 553 | if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
1da177e4 LT |
554 | break; |
555 | ||
556 | if (!smallmem) | |
557 | offset -= BBSIZE; | |
558 | } | |
559 | ||
560 | /* | |
561 | * We hit the beginning of the physical log & still no header. Return | |
562 | * to caller. If caller can handle a return of -1, then this routine | |
563 | * will be called again for the end of the physical log. | |
564 | */ | |
565 | if (i == -1) { | |
2451337d | 566 | error = 1; |
1da177e4 LT |
567 | goto out; |
568 | } | |
569 | ||
570 | /* | |
571 | * We have the final block of the good log (the first block | |
572 | * of the log record _before_ the head. So we check the uuid. | |
573 | */ | |
574 | if ((error = xlog_header_check_mount(log->l_mp, head))) | |
575 | goto out; | |
576 | ||
577 | /* | |
578 | * We may have found a log record header before we expected one. | |
579 | * last_blk will be the 1st block # with a given cycle #. We may end | |
580 | * up reading an entire log record. In this case, we don't want to | |
581 | * reset last_blk. Only when last_blk points in the middle of a log | |
582 | * record do we update last_blk. | |
583 | */ | |
62118709 | 584 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d | 585 | uint h_size = be32_to_cpu(head->h_size); |
1da177e4 LT |
586 | |
587 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; | |
588 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
589 | xhdrs++; | |
590 | } else { | |
591 | xhdrs = 1; | |
592 | } | |
593 | ||
b53e675d CH |
594 | if (*last_blk - i + extra_bblks != |
595 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) | |
1da177e4 LT |
596 | *last_blk = i; |
597 | ||
598 | out: | |
599 | xlog_put_bp(bp); | |
600 | return error; | |
601 | } | |
602 | ||
603 | /* | |
604 | * Head is defined to be the point of the log where the next log write | |
0a94da24 | 605 | * could go. This means that incomplete LR writes at the end are |
1da177e4 LT |
606 | * eliminated when calculating the head. We aren't guaranteed that previous |
607 | * LR have complete transactions. We only know that a cycle number of | |
608 | * current cycle number -1 won't be present in the log if we start writing | |
609 | * from our current block number. | |
610 | * | |
611 | * last_blk contains the block number of the first block with a given | |
612 | * cycle number. | |
613 | * | |
614 | * Return: zero if normal, non-zero if error. | |
615 | */ | |
ba0f32d4 | 616 | STATIC int |
1da177e4 | 617 | xlog_find_head( |
9a8d2fdb | 618 | struct xlog *log, |
1da177e4 LT |
619 | xfs_daddr_t *return_head_blk) |
620 | { | |
621 | xfs_buf_t *bp; | |
b2a922cd | 622 | char *offset; |
1da177e4 LT |
623 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; |
624 | int num_scan_bblks; | |
625 | uint first_half_cycle, last_half_cycle; | |
626 | uint stop_on_cycle; | |
627 | int error, log_bbnum = log->l_logBBsize; | |
628 | ||
629 | /* Is the end of the log device zeroed? */ | |
2451337d DC |
630 | error = xlog_find_zeroed(log, &first_blk); |
631 | if (error < 0) { | |
632 | xfs_warn(log->l_mp, "empty log check failed"); | |
633 | return error; | |
634 | } | |
635 | if (error == 1) { | |
1da177e4 LT |
636 | *return_head_blk = first_blk; |
637 | ||
638 | /* Is the whole lot zeroed? */ | |
639 | if (!first_blk) { | |
640 | /* Linux XFS shouldn't generate totally zeroed logs - | |
641 | * mkfs etc write a dummy unmount record to a fresh | |
642 | * log so we can store the uuid in there | |
643 | */ | |
a0fa2b67 | 644 | xfs_warn(log->l_mp, "totally zeroed log"); |
1da177e4 LT |
645 | } |
646 | ||
647 | return 0; | |
1da177e4 LT |
648 | } |
649 | ||
650 | first_blk = 0; /* get cycle # of 1st block */ | |
651 | bp = xlog_get_bp(log, 1); | |
652 | if (!bp) | |
2451337d | 653 | return -ENOMEM; |
076e6acb CH |
654 | |
655 | error = xlog_bread(log, 0, 1, bp, &offset); | |
656 | if (error) | |
1da177e4 | 657 | goto bp_err; |
076e6acb | 658 | |
03bea6fe | 659 | first_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
660 | |
661 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ | |
076e6acb CH |
662 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
663 | if (error) | |
1da177e4 | 664 | goto bp_err; |
076e6acb | 665 | |
03bea6fe | 666 | last_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
667 | ASSERT(last_half_cycle != 0); |
668 | ||
669 | /* | |
670 | * If the 1st half cycle number is equal to the last half cycle number, | |
671 | * then the entire log is stamped with the same cycle number. In this | |
672 | * case, head_blk can't be set to zero (which makes sense). The below | |
673 | * math doesn't work out properly with head_blk equal to zero. Instead, | |
674 | * we set it to log_bbnum which is an invalid block number, but this | |
675 | * value makes the math correct. If head_blk doesn't changed through | |
676 | * all the tests below, *head_blk is set to zero at the very end rather | |
677 | * than log_bbnum. In a sense, log_bbnum and zero are the same block | |
678 | * in a circular file. | |
679 | */ | |
680 | if (first_half_cycle == last_half_cycle) { | |
681 | /* | |
682 | * In this case we believe that the entire log should have | |
683 | * cycle number last_half_cycle. We need to scan backwards | |
684 | * from the end verifying that there are no holes still | |
685 | * containing last_half_cycle - 1. If we find such a hole, | |
686 | * then the start of that hole will be the new head. The | |
687 | * simple case looks like | |
688 | * x | x ... | x - 1 | x | |
689 | * Another case that fits this picture would be | |
690 | * x | x + 1 | x ... | x | |
c41564b5 | 691 | * In this case the head really is somewhere at the end of the |
1da177e4 LT |
692 | * log, as one of the latest writes at the beginning was |
693 | * incomplete. | |
694 | * One more case is | |
695 | * x | x + 1 | x ... | x - 1 | x | |
696 | * This is really the combination of the above two cases, and | |
697 | * the head has to end up at the start of the x-1 hole at the | |
698 | * end of the log. | |
699 | * | |
700 | * In the 256k log case, we will read from the beginning to the | |
701 | * end of the log and search for cycle numbers equal to x-1. | |
702 | * We don't worry about the x+1 blocks that we encounter, | |
703 | * because we know that they cannot be the head since the log | |
704 | * started with x. | |
705 | */ | |
706 | head_blk = log_bbnum; | |
707 | stop_on_cycle = last_half_cycle - 1; | |
708 | } else { | |
709 | /* | |
710 | * In this case we want to find the first block with cycle | |
711 | * number matching last_half_cycle. We expect the log to be | |
712 | * some variation on | |
3f943d85 | 713 | * x + 1 ... | x ... | x |
1da177e4 LT |
714 | * The first block with cycle number x (last_half_cycle) will |
715 | * be where the new head belongs. First we do a binary search | |
716 | * for the first occurrence of last_half_cycle. The binary | |
717 | * search may not be totally accurate, so then we scan back | |
718 | * from there looking for occurrences of last_half_cycle before | |
719 | * us. If that backwards scan wraps around the beginning of | |
720 | * the log, then we look for occurrences of last_half_cycle - 1 | |
721 | * at the end of the log. The cases we're looking for look | |
722 | * like | |
3f943d85 AE |
723 | * v binary search stopped here |
724 | * x + 1 ... | x | x + 1 | x ... | x | |
725 | * ^ but we want to locate this spot | |
1da177e4 | 726 | * or |
1da177e4 | 727 | * <---------> less than scan distance |
3f943d85 AE |
728 | * x + 1 ... | x ... | x - 1 | x |
729 | * ^ we want to locate this spot | |
1da177e4 LT |
730 | */ |
731 | stop_on_cycle = last_half_cycle; | |
732 | if ((error = xlog_find_cycle_start(log, bp, first_blk, | |
733 | &head_blk, last_half_cycle))) | |
734 | goto bp_err; | |
735 | } | |
736 | ||
737 | /* | |
738 | * Now validate the answer. Scan back some number of maximum possible | |
739 | * blocks and make sure each one has the expected cycle number. The | |
740 | * maximum is determined by the total possible amount of buffering | |
741 | * in the in-core log. The following number can be made tighter if | |
742 | * we actually look at the block size of the filesystem. | |
743 | */ | |
744 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
745 | if (head_blk >= num_scan_bblks) { | |
746 | /* | |
747 | * We are guaranteed that the entire check can be performed | |
748 | * in one buffer. | |
749 | */ | |
750 | start_blk = head_blk - num_scan_bblks; | |
751 | if ((error = xlog_find_verify_cycle(log, | |
752 | start_blk, num_scan_bblks, | |
753 | stop_on_cycle, &new_blk))) | |
754 | goto bp_err; | |
755 | if (new_blk != -1) | |
756 | head_blk = new_blk; | |
757 | } else { /* need to read 2 parts of log */ | |
758 | /* | |
759 | * We are going to scan backwards in the log in two parts. | |
760 | * First we scan the physical end of the log. In this part | |
761 | * of the log, we are looking for blocks with cycle number | |
762 | * last_half_cycle - 1. | |
763 | * If we find one, then we know that the log starts there, as | |
764 | * we've found a hole that didn't get written in going around | |
765 | * the end of the physical log. The simple case for this is | |
766 | * x + 1 ... | x ... | x - 1 | x | |
767 | * <---------> less than scan distance | |
768 | * If all of the blocks at the end of the log have cycle number | |
769 | * last_half_cycle, then we check the blocks at the start of | |
770 | * the log looking for occurrences of last_half_cycle. If we | |
771 | * find one, then our current estimate for the location of the | |
772 | * first occurrence of last_half_cycle is wrong and we move | |
773 | * back to the hole we've found. This case looks like | |
774 | * x + 1 ... | x | x + 1 | x ... | |
775 | * ^ binary search stopped here | |
776 | * Another case we need to handle that only occurs in 256k | |
777 | * logs is | |
778 | * x + 1 ... | x ... | x+1 | x ... | |
779 | * ^ binary search stops here | |
780 | * In a 256k log, the scan at the end of the log will see the | |
781 | * x + 1 blocks. We need to skip past those since that is | |
782 | * certainly not the head of the log. By searching for | |
783 | * last_half_cycle-1 we accomplish that. | |
784 | */ | |
1da177e4 | 785 | ASSERT(head_blk <= INT_MAX && |
3f943d85 AE |
786 | (xfs_daddr_t) num_scan_bblks >= head_blk); |
787 | start_blk = log_bbnum - (num_scan_bblks - head_blk); | |
1da177e4 LT |
788 | if ((error = xlog_find_verify_cycle(log, start_blk, |
789 | num_scan_bblks - (int)head_blk, | |
790 | (stop_on_cycle - 1), &new_blk))) | |
791 | goto bp_err; | |
792 | if (new_blk != -1) { | |
793 | head_blk = new_blk; | |
9db127ed | 794 | goto validate_head; |
1da177e4 LT |
795 | } |
796 | ||
797 | /* | |
798 | * Scan beginning of log now. The last part of the physical | |
799 | * log is good. This scan needs to verify that it doesn't find | |
800 | * the last_half_cycle. | |
801 | */ | |
802 | start_blk = 0; | |
803 | ASSERT(head_blk <= INT_MAX); | |
804 | if ((error = xlog_find_verify_cycle(log, | |
805 | start_blk, (int)head_blk, | |
806 | stop_on_cycle, &new_blk))) | |
807 | goto bp_err; | |
808 | if (new_blk != -1) | |
809 | head_blk = new_blk; | |
810 | } | |
811 | ||
9db127ed | 812 | validate_head: |
1da177e4 LT |
813 | /* |
814 | * Now we need to make sure head_blk is not pointing to a block in | |
815 | * the middle of a log record. | |
816 | */ | |
817 | num_scan_bblks = XLOG_REC_SHIFT(log); | |
818 | if (head_blk >= num_scan_bblks) { | |
819 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ | |
820 | ||
821 | /* start ptr at last block ptr before head_blk */ | |
2451337d DC |
822 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
823 | if (error == 1) | |
824 | error = -EIO; | |
825 | if (error) | |
1da177e4 LT |
826 | goto bp_err; |
827 | } else { | |
828 | start_blk = 0; | |
829 | ASSERT(head_blk <= INT_MAX); | |
2451337d DC |
830 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
831 | if (error < 0) | |
832 | goto bp_err; | |
833 | if (error == 1) { | |
1da177e4 | 834 | /* We hit the beginning of the log during our search */ |
3f943d85 | 835 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
1da177e4 LT |
836 | new_blk = log_bbnum; |
837 | ASSERT(start_blk <= INT_MAX && | |
838 | (xfs_daddr_t) log_bbnum-start_blk >= 0); | |
839 | ASSERT(head_blk <= INT_MAX); | |
2451337d DC |
840 | error = xlog_find_verify_log_record(log, start_blk, |
841 | &new_blk, (int)head_blk); | |
842 | if (error == 1) | |
843 | error = -EIO; | |
844 | if (error) | |
1da177e4 LT |
845 | goto bp_err; |
846 | if (new_blk != log_bbnum) | |
847 | head_blk = new_blk; | |
848 | } else if (error) | |
849 | goto bp_err; | |
850 | } | |
851 | ||
852 | xlog_put_bp(bp); | |
853 | if (head_blk == log_bbnum) | |
854 | *return_head_blk = 0; | |
855 | else | |
856 | *return_head_blk = head_blk; | |
857 | /* | |
858 | * When returning here, we have a good block number. Bad block | |
859 | * means that during a previous crash, we didn't have a clean break | |
860 | * from cycle number N to cycle number N-1. In this case, we need | |
861 | * to find the first block with cycle number N-1. | |
862 | */ | |
863 | return 0; | |
864 | ||
865 | bp_err: | |
866 | xlog_put_bp(bp); | |
867 | ||
868 | if (error) | |
a0fa2b67 | 869 | xfs_warn(log->l_mp, "failed to find log head"); |
1da177e4 LT |
870 | return error; |
871 | } | |
872 | ||
eed6b462 BF |
873 | /* |
874 | * Seek backwards in the log for log record headers. | |
875 | * | |
876 | * Given a starting log block, walk backwards until we find the provided number | |
877 | * of records or hit the provided tail block. The return value is the number of | |
878 | * records encountered or a negative error code. The log block and buffer | |
879 | * pointer of the last record seen are returned in rblk and rhead respectively. | |
880 | */ | |
881 | STATIC int | |
882 | xlog_rseek_logrec_hdr( | |
883 | struct xlog *log, | |
884 | xfs_daddr_t head_blk, | |
885 | xfs_daddr_t tail_blk, | |
886 | int count, | |
887 | struct xfs_buf *bp, | |
888 | xfs_daddr_t *rblk, | |
889 | struct xlog_rec_header **rhead, | |
890 | bool *wrapped) | |
891 | { | |
892 | int i; | |
893 | int error; | |
894 | int found = 0; | |
895 | char *offset = NULL; | |
896 | xfs_daddr_t end_blk; | |
897 | ||
898 | *wrapped = false; | |
899 | ||
900 | /* | |
901 | * Walk backwards from the head block until we hit the tail or the first | |
902 | * block in the log. | |
903 | */ | |
904 | end_blk = head_blk > tail_blk ? tail_blk : 0; | |
905 | for (i = (int) head_blk - 1; i >= end_blk; i--) { | |
906 | error = xlog_bread(log, i, 1, bp, &offset); | |
907 | if (error) | |
908 | goto out_error; | |
909 | ||
910 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
911 | *rblk = i; | |
912 | *rhead = (struct xlog_rec_header *) offset; | |
913 | if (++found == count) | |
914 | break; | |
915 | } | |
916 | } | |
917 | ||
918 | /* | |
919 | * If we haven't hit the tail block or the log record header count, | |
920 | * start looking again from the end of the physical log. Note that | |
921 | * callers can pass head == tail if the tail is not yet known. | |
922 | */ | |
923 | if (tail_blk >= head_blk && found != count) { | |
924 | for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) { | |
925 | error = xlog_bread(log, i, 1, bp, &offset); | |
926 | if (error) | |
927 | goto out_error; | |
928 | ||
929 | if (*(__be32 *)offset == | |
930 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
931 | *wrapped = true; | |
932 | *rblk = i; | |
933 | *rhead = (struct xlog_rec_header *) offset; | |
934 | if (++found == count) | |
935 | break; | |
936 | } | |
937 | } | |
938 | } | |
939 | ||
940 | return found; | |
941 | ||
942 | out_error: | |
943 | return error; | |
944 | } | |
945 | ||
7088c413 BF |
946 | /* |
947 | * Seek forward in the log for log record headers. | |
948 | * | |
949 | * Given head and tail blocks, walk forward from the tail block until we find | |
950 | * the provided number of records or hit the head block. The return value is the | |
951 | * number of records encountered or a negative error code. The log block and | |
952 | * buffer pointer of the last record seen are returned in rblk and rhead | |
953 | * respectively. | |
954 | */ | |
955 | STATIC int | |
956 | xlog_seek_logrec_hdr( | |
957 | struct xlog *log, | |
958 | xfs_daddr_t head_blk, | |
959 | xfs_daddr_t tail_blk, | |
960 | int count, | |
961 | struct xfs_buf *bp, | |
962 | xfs_daddr_t *rblk, | |
963 | struct xlog_rec_header **rhead, | |
964 | bool *wrapped) | |
965 | { | |
966 | int i; | |
967 | int error; | |
968 | int found = 0; | |
969 | char *offset = NULL; | |
970 | xfs_daddr_t end_blk; | |
971 | ||
972 | *wrapped = false; | |
973 | ||
974 | /* | |
975 | * Walk forward from the tail block until we hit the head or the last | |
976 | * block in the log. | |
977 | */ | |
978 | end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1; | |
979 | for (i = (int) tail_blk; i <= end_blk; i++) { | |
980 | error = xlog_bread(log, i, 1, bp, &offset); | |
981 | if (error) | |
982 | goto out_error; | |
983 | ||
984 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
985 | *rblk = i; | |
986 | *rhead = (struct xlog_rec_header *) offset; | |
987 | if (++found == count) | |
988 | break; | |
989 | } | |
990 | } | |
991 | ||
992 | /* | |
993 | * If we haven't hit the head block or the log record header count, | |
994 | * start looking again from the start of the physical log. | |
995 | */ | |
996 | if (tail_blk > head_blk && found != count) { | |
997 | for (i = 0; i < (int) head_blk; i++) { | |
998 | error = xlog_bread(log, i, 1, bp, &offset); | |
999 | if (error) | |
1000 | goto out_error; | |
1001 | ||
1002 | if (*(__be32 *)offset == | |
1003 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
1004 | *wrapped = true; | |
1005 | *rblk = i; | |
1006 | *rhead = (struct xlog_rec_header *) offset; | |
1007 | if (++found == count) | |
1008 | break; | |
1009 | } | |
1010 | } | |
1011 | } | |
1012 | ||
1013 | return found; | |
1014 | ||
1015 | out_error: | |
1016 | return error; | |
1017 | } | |
1018 | ||
1019 | /* | |
1020 | * Check the log tail for torn writes. This is required when torn writes are | |
1021 | * detected at the head and the head had to be walked back to a previous record. | |
1022 | * The tail of the previous record must now be verified to ensure the torn | |
1023 | * writes didn't corrupt the previous tail. | |
1024 | * | |
1025 | * Return an error if CRC verification fails as recovery cannot proceed. | |
1026 | */ | |
1027 | STATIC int | |
1028 | xlog_verify_tail( | |
1029 | struct xlog *log, | |
1030 | xfs_daddr_t head_blk, | |
1031 | xfs_daddr_t tail_blk) | |
1032 | { | |
1033 | struct xlog_rec_header *thead; | |
1034 | struct xfs_buf *bp; | |
1035 | xfs_daddr_t first_bad; | |
1036 | int count; | |
1037 | int error = 0; | |
1038 | bool wrapped; | |
1039 | xfs_daddr_t tmp_head; | |
1040 | ||
1041 | bp = xlog_get_bp(log, 1); | |
1042 | if (!bp) | |
1043 | return -ENOMEM; | |
1044 | ||
1045 | /* | |
1046 | * Seek XLOG_MAX_ICLOGS + 1 records past the current tail record to get | |
1047 | * a temporary head block that points after the last possible | |
1048 | * concurrently written record of the tail. | |
1049 | */ | |
1050 | count = xlog_seek_logrec_hdr(log, head_blk, tail_blk, | |
1051 | XLOG_MAX_ICLOGS + 1, bp, &tmp_head, &thead, | |
1052 | &wrapped); | |
1053 | if (count < 0) { | |
1054 | error = count; | |
1055 | goto out; | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * If the call above didn't find XLOG_MAX_ICLOGS + 1 records, we ran | |
1060 | * into the actual log head. tmp_head points to the start of the record | |
1061 | * so update it to the actual head block. | |
1062 | */ | |
1063 | if (count < XLOG_MAX_ICLOGS + 1) | |
1064 | tmp_head = head_blk; | |
1065 | ||
1066 | /* | |
1067 | * We now have a tail and temporary head block that covers at least | |
1068 | * XLOG_MAX_ICLOGS records from the tail. We need to verify that these | |
1069 | * records were completely written. Run a CRC verification pass from | |
1070 | * tail to head and return the result. | |
1071 | */ | |
1072 | error = xlog_do_recovery_pass(log, tmp_head, tail_blk, | |
1073 | XLOG_RECOVER_CRCPASS, &first_bad); | |
1074 | ||
1075 | out: | |
1076 | xlog_put_bp(bp); | |
1077 | return error; | |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * Detect and trim torn writes from the head of the log. | |
1082 | * | |
1083 | * Storage without sector atomicity guarantees can result in torn writes in the | |
1084 | * log in the event of a crash. Our only means to detect this scenario is via | |
1085 | * CRC verification. While we can't always be certain that CRC verification | |
1086 | * failure is due to a torn write vs. an unrelated corruption, we do know that | |
1087 | * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at | |
1088 | * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of | |
1089 | * the log and treat failures in this range as torn writes as a matter of | |
1090 | * policy. In the event of CRC failure, the head is walked back to the last good | |
1091 | * record in the log and the tail is updated from that record and verified. | |
1092 | */ | |
1093 | STATIC int | |
1094 | xlog_verify_head( | |
1095 | struct xlog *log, | |
1096 | xfs_daddr_t *head_blk, /* in/out: unverified head */ | |
1097 | xfs_daddr_t *tail_blk, /* out: tail block */ | |
1098 | struct xfs_buf *bp, | |
1099 | xfs_daddr_t *rhead_blk, /* start blk of last record */ | |
1100 | struct xlog_rec_header **rhead, /* ptr to last record */ | |
1101 | bool *wrapped) /* last rec. wraps phys. log */ | |
1102 | { | |
1103 | struct xlog_rec_header *tmp_rhead; | |
1104 | struct xfs_buf *tmp_bp; | |
1105 | xfs_daddr_t first_bad; | |
1106 | xfs_daddr_t tmp_rhead_blk; | |
1107 | int found; | |
1108 | int error; | |
1109 | bool tmp_wrapped; | |
1110 | ||
1111 | /* | |
82ff6cc2 BF |
1112 | * Check the head of the log for torn writes. Search backwards from the |
1113 | * head until we hit the tail or the maximum number of log record I/Os | |
1114 | * that could have been in flight at one time. Use a temporary buffer so | |
1115 | * we don't trash the rhead/bp pointers from the caller. | |
7088c413 BF |
1116 | */ |
1117 | tmp_bp = xlog_get_bp(log, 1); | |
1118 | if (!tmp_bp) | |
1119 | return -ENOMEM; | |
1120 | error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk, | |
1121 | XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk, | |
1122 | &tmp_rhead, &tmp_wrapped); | |
1123 | xlog_put_bp(tmp_bp); | |
1124 | if (error < 0) | |
1125 | return error; | |
1126 | ||
1127 | /* | |
1128 | * Now run a CRC verification pass over the records starting at the | |
1129 | * block found above to the current head. If a CRC failure occurs, the | |
1130 | * log block of the first bad record is saved in first_bad. | |
1131 | */ | |
1132 | error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk, | |
1133 | XLOG_RECOVER_CRCPASS, &first_bad); | |
1134 | if (error == -EFSBADCRC) { | |
1135 | /* | |
1136 | * We've hit a potential torn write. Reset the error and warn | |
1137 | * about it. | |
1138 | */ | |
1139 | error = 0; | |
1140 | xfs_warn(log->l_mp, | |
1141 | "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.", | |
1142 | first_bad, *head_blk); | |
1143 | ||
1144 | /* | |
1145 | * Get the header block and buffer pointer for the last good | |
1146 | * record before the bad record. | |
1147 | * | |
1148 | * Note that xlog_find_tail() clears the blocks at the new head | |
1149 | * (i.e., the records with invalid CRC) if the cycle number | |
1150 | * matches the the current cycle. | |
1151 | */ | |
1152 | found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp, | |
1153 | rhead_blk, rhead, wrapped); | |
1154 | if (found < 0) | |
1155 | return found; | |
1156 | if (found == 0) /* XXX: right thing to do here? */ | |
1157 | return -EIO; | |
1158 | ||
1159 | /* | |
1160 | * Reset the head block to the starting block of the first bad | |
1161 | * log record and set the tail block based on the last good | |
1162 | * record. | |
1163 | * | |
1164 | * Bail out if the updated head/tail match as this indicates | |
1165 | * possible corruption outside of the acceptable | |
1166 | * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair... | |
1167 | */ | |
1168 | *head_blk = first_bad; | |
1169 | *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn)); | |
1170 | if (*head_blk == *tail_blk) { | |
1171 | ASSERT(0); | |
1172 | return 0; | |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * Now verify the tail based on the updated head. This is | |
1177 | * required because the torn writes trimmed from the head could | |
1178 | * have been written over the tail of a previous record. Return | |
1179 | * any errors since recovery cannot proceed if the tail is | |
1180 | * corrupt. | |
1181 | * | |
1182 | * XXX: This leaves a gap in truly robust protection from torn | |
1183 | * writes in the log. If the head is behind the tail, the tail | |
1184 | * pushes forward to create some space and then a crash occurs | |
1185 | * causing the writes into the previous record's tail region to | |
1186 | * tear, log recovery isn't able to recover. | |
1187 | * | |
1188 | * How likely is this to occur? If possible, can we do something | |
1189 | * more intelligent here? Is it safe to push the tail forward if | |
1190 | * we can determine that the tail is within the range of the | |
1191 | * torn write (e.g., the kernel can only overwrite the tail if | |
1192 | * it has actually been pushed forward)? Alternatively, could we | |
1193 | * somehow prevent this condition at runtime? | |
1194 | */ | |
1195 | error = xlog_verify_tail(log, *head_blk, *tail_blk); | |
1196 | } | |
1197 | ||
1198 | return error; | |
1199 | } | |
1200 | ||
65b99a08 BF |
1201 | /* |
1202 | * Check whether the head of the log points to an unmount record. In other | |
1203 | * words, determine whether the log is clean. If so, update the in-core state | |
1204 | * appropriately. | |
1205 | */ | |
1206 | static int | |
1207 | xlog_check_unmount_rec( | |
1208 | struct xlog *log, | |
1209 | xfs_daddr_t *head_blk, | |
1210 | xfs_daddr_t *tail_blk, | |
1211 | struct xlog_rec_header *rhead, | |
1212 | xfs_daddr_t rhead_blk, | |
1213 | struct xfs_buf *bp, | |
1214 | bool *clean) | |
1215 | { | |
1216 | struct xlog_op_header *op_head; | |
1217 | xfs_daddr_t umount_data_blk; | |
1218 | xfs_daddr_t after_umount_blk; | |
1219 | int hblks; | |
1220 | int error; | |
1221 | char *offset; | |
1222 | ||
1223 | *clean = false; | |
1224 | ||
1225 | /* | |
1226 | * Look for unmount record. If we find it, then we know there was a | |
1227 | * clean unmount. Since 'i' could be the last block in the physical | |
1228 | * log, we convert to a log block before comparing to the head_blk. | |
1229 | * | |
1230 | * Save the current tail lsn to use to pass to xlog_clear_stale_blocks() | |
1231 | * below. We won't want to clear the unmount record if there is one, so | |
1232 | * we pass the lsn of the unmount record rather than the block after it. | |
1233 | */ | |
1234 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1235 | int h_size = be32_to_cpu(rhead->h_size); | |
1236 | int h_version = be32_to_cpu(rhead->h_version); | |
1237 | ||
1238 | if ((h_version & XLOG_VERSION_2) && | |
1239 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
1240 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
1241 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
1242 | hblks++; | |
1243 | } else { | |
1244 | hblks = 1; | |
1245 | } | |
1246 | } else { | |
1247 | hblks = 1; | |
1248 | } | |
1249 | after_umount_blk = rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)); | |
1250 | after_umount_blk = do_mod(after_umount_blk, log->l_logBBsize); | |
1251 | if (*head_blk == after_umount_blk && | |
1252 | be32_to_cpu(rhead->h_num_logops) == 1) { | |
1253 | umount_data_blk = rhead_blk + hblks; | |
1254 | umount_data_blk = do_mod(umount_data_blk, log->l_logBBsize); | |
1255 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); | |
1256 | if (error) | |
1257 | return error; | |
1258 | ||
1259 | op_head = (struct xlog_op_header *)offset; | |
1260 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { | |
1261 | /* | |
1262 | * Set tail and last sync so that newly written log | |
1263 | * records will point recovery to after the current | |
1264 | * unmount record. | |
1265 | */ | |
1266 | xlog_assign_atomic_lsn(&log->l_tail_lsn, | |
1267 | log->l_curr_cycle, after_umount_blk); | |
1268 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, | |
1269 | log->l_curr_cycle, after_umount_blk); | |
1270 | *tail_blk = after_umount_blk; | |
1271 | ||
1272 | *clean = true; | |
1273 | } | |
1274 | } | |
1275 | ||
1276 | return 0; | |
1277 | } | |
1278 | ||
717bc0eb BF |
1279 | static void |
1280 | xlog_set_state( | |
1281 | struct xlog *log, | |
1282 | xfs_daddr_t head_blk, | |
1283 | struct xlog_rec_header *rhead, | |
1284 | xfs_daddr_t rhead_blk, | |
1285 | bool bump_cycle) | |
1286 | { | |
1287 | /* | |
1288 | * Reset log values according to the state of the log when we | |
1289 | * crashed. In the case where head_blk == 0, we bump curr_cycle | |
1290 | * one because the next write starts a new cycle rather than | |
1291 | * continuing the cycle of the last good log record. At this | |
1292 | * point we have guaranteed that all partial log records have been | |
1293 | * accounted for. Therefore, we know that the last good log record | |
1294 | * written was complete and ended exactly on the end boundary | |
1295 | * of the physical log. | |
1296 | */ | |
1297 | log->l_prev_block = rhead_blk; | |
1298 | log->l_curr_block = (int)head_blk; | |
1299 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); | |
1300 | if (bump_cycle) | |
1301 | log->l_curr_cycle++; | |
1302 | atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn)); | |
1303 | atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn)); | |
1304 | xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle, | |
1305 | BBTOB(log->l_curr_block)); | |
1306 | xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle, | |
1307 | BBTOB(log->l_curr_block)); | |
1308 | } | |
1309 | ||
1da177e4 LT |
1310 | /* |
1311 | * Find the sync block number or the tail of the log. | |
1312 | * | |
1313 | * This will be the block number of the last record to have its | |
1314 | * associated buffers synced to disk. Every log record header has | |
1315 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy | |
1316 | * to get a sync block number. The only concern is to figure out which | |
1317 | * log record header to believe. | |
1318 | * | |
1319 | * The following algorithm uses the log record header with the largest | |
1320 | * lsn. The entire log record does not need to be valid. We only care | |
1321 | * that the header is valid. | |
1322 | * | |
1323 | * We could speed up search by using current head_blk buffer, but it is not | |
1324 | * available. | |
1325 | */ | |
5d77c0dc | 1326 | STATIC int |
1da177e4 | 1327 | xlog_find_tail( |
9a8d2fdb | 1328 | struct xlog *log, |
1da177e4 | 1329 | xfs_daddr_t *head_blk, |
65be6054 | 1330 | xfs_daddr_t *tail_blk) |
1da177e4 LT |
1331 | { |
1332 | xlog_rec_header_t *rhead; | |
b2a922cd | 1333 | char *offset = NULL; |
1da177e4 | 1334 | xfs_buf_t *bp; |
7088c413 | 1335 | int error; |
7088c413 | 1336 | xfs_daddr_t rhead_blk; |
1da177e4 | 1337 | xfs_lsn_t tail_lsn; |
eed6b462 | 1338 | bool wrapped = false; |
65b99a08 | 1339 | bool clean = false; |
1da177e4 LT |
1340 | |
1341 | /* | |
1342 | * Find previous log record | |
1343 | */ | |
1344 | if ((error = xlog_find_head(log, head_blk))) | |
1345 | return error; | |
82ff6cc2 | 1346 | ASSERT(*head_blk < INT_MAX); |
1da177e4 LT |
1347 | |
1348 | bp = xlog_get_bp(log, 1); | |
1349 | if (!bp) | |
2451337d | 1350 | return -ENOMEM; |
1da177e4 | 1351 | if (*head_blk == 0) { /* special case */ |
076e6acb CH |
1352 | error = xlog_bread(log, 0, 1, bp, &offset); |
1353 | if (error) | |
9db127ed | 1354 | goto done; |
076e6acb | 1355 | |
03bea6fe | 1356 | if (xlog_get_cycle(offset) == 0) { |
1da177e4 LT |
1357 | *tail_blk = 0; |
1358 | /* leave all other log inited values alone */ | |
9db127ed | 1359 | goto done; |
1da177e4 LT |
1360 | } |
1361 | } | |
1362 | ||
82ff6cc2 BF |
1363 | /* |
1364 | * Search backwards through the log looking for the log record header | |
1365 | * block. This wraps all the way back around to the head so something is | |
1366 | * seriously wrong if we can't find it. | |
1367 | */ | |
1368 | error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp, | |
1369 | &rhead_blk, &rhead, &wrapped); | |
1370 | if (error < 0) | |
1371 | return error; | |
1372 | if (!error) { | |
1373 | xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__); | |
1374 | return -EIO; | |
1375 | } | |
1376 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); | |
1377 | ||
1da177e4 | 1378 | /* |
7088c413 BF |
1379 | * Trim the head block back to skip over torn records. We can have |
1380 | * multiple log I/Os in flight at any time, so we assume CRC failures | |
1381 | * back through the previous several records are torn writes and skip | |
1382 | * them. | |
1da177e4 | 1383 | */ |
7088c413 BF |
1384 | error = xlog_verify_head(log, head_blk, tail_blk, bp, &rhead_blk, |
1385 | &rhead, &wrapped); | |
1386 | if (error) | |
eed6b462 | 1387 | goto done; |
1da177e4 LT |
1388 | |
1389 | /* | |
717bc0eb | 1390 | * Set the log state based on the current head record. |
1da177e4 | 1391 | */ |
717bc0eb | 1392 | xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped); |
65b99a08 | 1393 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
1da177e4 LT |
1394 | |
1395 | /* | |
65b99a08 BF |
1396 | * Look for an unmount record at the head of the log. This sets the log |
1397 | * state to determine whether recovery is necessary. | |
1da177e4 | 1398 | */ |
65b99a08 BF |
1399 | error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead, |
1400 | rhead_blk, bp, &clean); | |
1401 | if (error) | |
1402 | goto done; | |
92821e2b | 1403 | |
65b99a08 BF |
1404 | /* |
1405 | * Note that the unmount was clean. If the unmount was not clean, we | |
1406 | * need to know this to rebuild the superblock counters from the perag | |
1407 | * headers if we have a filesystem using non-persistent counters. | |
1408 | */ | |
1409 | if (clean) | |
1410 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; | |
1da177e4 LT |
1411 | |
1412 | /* | |
1413 | * Make sure that there are no blocks in front of the head | |
1414 | * with the same cycle number as the head. This can happen | |
1415 | * because we allow multiple outstanding log writes concurrently, | |
1416 | * and the later writes might make it out before earlier ones. | |
1417 | * | |
1418 | * We use the lsn from before modifying it so that we'll never | |
1419 | * overwrite the unmount record after a clean unmount. | |
1420 | * | |
1421 | * Do this only if we are going to recover the filesystem | |
1422 | * | |
1423 | * NOTE: This used to say "if (!readonly)" | |
1424 | * However on Linux, we can & do recover a read-only filesystem. | |
1425 | * We only skip recovery if NORECOVERY is specified on mount, | |
1426 | * in which case we would not be here. | |
1427 | * | |
1428 | * But... if the -device- itself is readonly, just skip this. | |
1429 | * We can't recover this device anyway, so it won't matter. | |
1430 | */ | |
9db127ed | 1431 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) |
1da177e4 | 1432 | error = xlog_clear_stale_blocks(log, tail_lsn); |
1da177e4 | 1433 | |
9db127ed | 1434 | done: |
1da177e4 LT |
1435 | xlog_put_bp(bp); |
1436 | ||
1437 | if (error) | |
a0fa2b67 | 1438 | xfs_warn(log->l_mp, "failed to locate log tail"); |
1da177e4 LT |
1439 | return error; |
1440 | } | |
1441 | ||
1442 | /* | |
1443 | * Is the log zeroed at all? | |
1444 | * | |
1445 | * The last binary search should be changed to perform an X block read | |
1446 | * once X becomes small enough. You can then search linearly through | |
1447 | * the X blocks. This will cut down on the number of reads we need to do. | |
1448 | * | |
1449 | * If the log is partially zeroed, this routine will pass back the blkno | |
1450 | * of the first block with cycle number 0. It won't have a complete LR | |
1451 | * preceding it. | |
1452 | * | |
1453 | * Return: | |
1454 | * 0 => the log is completely written to | |
2451337d DC |
1455 | * 1 => use *blk_no as the first block of the log |
1456 | * <0 => error has occurred | |
1da177e4 | 1457 | */ |
a8272ce0 | 1458 | STATIC int |
1da177e4 | 1459 | xlog_find_zeroed( |
9a8d2fdb | 1460 | struct xlog *log, |
1da177e4 LT |
1461 | xfs_daddr_t *blk_no) |
1462 | { | |
1463 | xfs_buf_t *bp; | |
b2a922cd | 1464 | char *offset; |
1da177e4 LT |
1465 | uint first_cycle, last_cycle; |
1466 | xfs_daddr_t new_blk, last_blk, start_blk; | |
1467 | xfs_daddr_t num_scan_bblks; | |
1468 | int error, log_bbnum = log->l_logBBsize; | |
1469 | ||
6fdf8ccc NS |
1470 | *blk_no = 0; |
1471 | ||
1da177e4 LT |
1472 | /* check totally zeroed log */ |
1473 | bp = xlog_get_bp(log, 1); | |
1474 | if (!bp) | |
2451337d | 1475 | return -ENOMEM; |
076e6acb CH |
1476 | error = xlog_bread(log, 0, 1, bp, &offset); |
1477 | if (error) | |
1da177e4 | 1478 | goto bp_err; |
076e6acb | 1479 | |
03bea6fe | 1480 | first_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1481 | if (first_cycle == 0) { /* completely zeroed log */ |
1482 | *blk_no = 0; | |
1483 | xlog_put_bp(bp); | |
2451337d | 1484 | return 1; |
1da177e4 LT |
1485 | } |
1486 | ||
1487 | /* check partially zeroed log */ | |
076e6acb CH |
1488 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
1489 | if (error) | |
1da177e4 | 1490 | goto bp_err; |
076e6acb | 1491 | |
03bea6fe | 1492 | last_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1493 | if (last_cycle != 0) { /* log completely written to */ |
1494 | xlog_put_bp(bp); | |
1495 | return 0; | |
1496 | } else if (first_cycle != 1) { | |
1497 | /* | |
1498 | * If the cycle of the last block is zero, the cycle of | |
1499 | * the first block must be 1. If it's not, maybe we're | |
1500 | * not looking at a log... Bail out. | |
1501 | */ | |
a0fa2b67 DC |
1502 | xfs_warn(log->l_mp, |
1503 | "Log inconsistent or not a log (last==0, first!=1)"); | |
2451337d | 1504 | error = -EINVAL; |
5d0a6549 | 1505 | goto bp_err; |
1da177e4 LT |
1506 | } |
1507 | ||
1508 | /* we have a partially zeroed log */ | |
1509 | last_blk = log_bbnum-1; | |
1510 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) | |
1511 | goto bp_err; | |
1512 | ||
1513 | /* | |
1514 | * Validate the answer. Because there is no way to guarantee that | |
1515 | * the entire log is made up of log records which are the same size, | |
1516 | * we scan over the defined maximum blocks. At this point, the maximum | |
1517 | * is not chosen to mean anything special. XXXmiken | |
1518 | */ | |
1519 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
1520 | ASSERT(num_scan_bblks <= INT_MAX); | |
1521 | ||
1522 | if (last_blk < num_scan_bblks) | |
1523 | num_scan_bblks = last_blk; | |
1524 | start_blk = last_blk - num_scan_bblks; | |
1525 | ||
1526 | /* | |
1527 | * We search for any instances of cycle number 0 that occur before | |
1528 | * our current estimate of the head. What we're trying to detect is | |
1529 | * 1 ... | 0 | 1 | 0... | |
1530 | * ^ binary search ends here | |
1531 | */ | |
1532 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
1533 | (int)num_scan_bblks, 0, &new_blk))) | |
1534 | goto bp_err; | |
1535 | if (new_blk != -1) | |
1536 | last_blk = new_blk; | |
1537 | ||
1538 | /* | |
1539 | * Potentially backup over partial log record write. We don't need | |
1540 | * to search the end of the log because we know it is zero. | |
1541 | */ | |
2451337d DC |
1542 | error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0); |
1543 | if (error == 1) | |
1544 | error = -EIO; | |
1545 | if (error) | |
1546 | goto bp_err; | |
1da177e4 LT |
1547 | |
1548 | *blk_no = last_blk; | |
1549 | bp_err: | |
1550 | xlog_put_bp(bp); | |
1551 | if (error) | |
1552 | return error; | |
2451337d | 1553 | return 1; |
1da177e4 LT |
1554 | } |
1555 | ||
1556 | /* | |
1557 | * These are simple subroutines used by xlog_clear_stale_blocks() below | |
1558 | * to initialize a buffer full of empty log record headers and write | |
1559 | * them into the log. | |
1560 | */ | |
1561 | STATIC void | |
1562 | xlog_add_record( | |
9a8d2fdb | 1563 | struct xlog *log, |
b2a922cd | 1564 | char *buf, |
1da177e4 LT |
1565 | int cycle, |
1566 | int block, | |
1567 | int tail_cycle, | |
1568 | int tail_block) | |
1569 | { | |
1570 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; | |
1571 | ||
1572 | memset(buf, 0, BBSIZE); | |
b53e675d CH |
1573 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1574 | recp->h_cycle = cpu_to_be32(cycle); | |
1575 | recp->h_version = cpu_to_be32( | |
62118709 | 1576 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d CH |
1577 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
1578 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); | |
1579 | recp->h_fmt = cpu_to_be32(XLOG_FMT); | |
1da177e4 LT |
1580 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1581 | } | |
1582 | ||
1583 | STATIC int | |
1584 | xlog_write_log_records( | |
9a8d2fdb | 1585 | struct xlog *log, |
1da177e4 LT |
1586 | int cycle, |
1587 | int start_block, | |
1588 | int blocks, | |
1589 | int tail_cycle, | |
1590 | int tail_block) | |
1591 | { | |
b2a922cd | 1592 | char *offset; |
1da177e4 LT |
1593 | xfs_buf_t *bp; |
1594 | int balign, ealign; | |
69ce58f0 | 1595 | int sectbb = log->l_sectBBsize; |
1da177e4 LT |
1596 | int end_block = start_block + blocks; |
1597 | int bufblks; | |
1598 | int error = 0; | |
1599 | int i, j = 0; | |
1600 | ||
6881a229 AE |
1601 | /* |
1602 | * Greedily allocate a buffer big enough to handle the full | |
1603 | * range of basic blocks to be written. If that fails, try | |
1604 | * a smaller size. We need to be able to write at least a | |
1605 | * log sector, or we're out of luck. | |
1606 | */ | |
1da177e4 | 1607 | bufblks = 1 << ffs(blocks); |
81158e0c DC |
1608 | while (bufblks > log->l_logBBsize) |
1609 | bufblks >>= 1; | |
1da177e4 LT |
1610 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1611 | bufblks >>= 1; | |
69ce58f0 | 1612 | if (bufblks < sectbb) |
2451337d | 1613 | return -ENOMEM; |
1da177e4 LT |
1614 | } |
1615 | ||
1616 | /* We may need to do a read at the start to fill in part of | |
1617 | * the buffer in the starting sector not covered by the first | |
1618 | * write below. | |
1619 | */ | |
5c17f533 | 1620 | balign = round_down(start_block, sectbb); |
1da177e4 | 1621 | if (balign != start_block) { |
076e6acb CH |
1622 | error = xlog_bread_noalign(log, start_block, 1, bp); |
1623 | if (error) | |
1624 | goto out_put_bp; | |
1625 | ||
1da177e4 LT |
1626 | j = start_block - balign; |
1627 | } | |
1628 | ||
1629 | for (i = start_block; i < end_block; i += bufblks) { | |
1630 | int bcount, endcount; | |
1631 | ||
1632 | bcount = min(bufblks, end_block - start_block); | |
1633 | endcount = bcount - j; | |
1634 | ||
1635 | /* We may need to do a read at the end to fill in part of | |
1636 | * the buffer in the final sector not covered by the write. | |
1637 | * If this is the same sector as the above read, skip it. | |
1638 | */ | |
5c17f533 | 1639 | ealign = round_down(end_block, sectbb); |
1da177e4 | 1640 | if (j == 0 && (start_block + endcount > ealign)) { |
62926044 | 1641 | offset = bp->b_addr + BBTOB(ealign - start_block); |
44396476 DC |
1642 | error = xlog_bread_offset(log, ealign, sectbb, |
1643 | bp, offset); | |
076e6acb CH |
1644 | if (error) |
1645 | break; | |
1646 | ||
1da177e4 LT |
1647 | } |
1648 | ||
1649 | offset = xlog_align(log, start_block, endcount, bp); | |
1650 | for (; j < endcount; j++) { | |
1651 | xlog_add_record(log, offset, cycle, i+j, | |
1652 | tail_cycle, tail_block); | |
1653 | offset += BBSIZE; | |
1654 | } | |
1655 | error = xlog_bwrite(log, start_block, endcount, bp); | |
1656 | if (error) | |
1657 | break; | |
1658 | start_block += endcount; | |
1659 | j = 0; | |
1660 | } | |
076e6acb CH |
1661 | |
1662 | out_put_bp: | |
1da177e4 LT |
1663 | xlog_put_bp(bp); |
1664 | return error; | |
1665 | } | |
1666 | ||
1667 | /* | |
1668 | * This routine is called to blow away any incomplete log writes out | |
1669 | * in front of the log head. We do this so that we won't become confused | |
1670 | * if we come up, write only a little bit more, and then crash again. | |
1671 | * If we leave the partial log records out there, this situation could | |
1672 | * cause us to think those partial writes are valid blocks since they | |
1673 | * have the current cycle number. We get rid of them by overwriting them | |
1674 | * with empty log records with the old cycle number rather than the | |
1675 | * current one. | |
1676 | * | |
1677 | * The tail lsn is passed in rather than taken from | |
1678 | * the log so that we will not write over the unmount record after a | |
1679 | * clean unmount in a 512 block log. Doing so would leave the log without | |
1680 | * any valid log records in it until a new one was written. If we crashed | |
1681 | * during that time we would not be able to recover. | |
1682 | */ | |
1683 | STATIC int | |
1684 | xlog_clear_stale_blocks( | |
9a8d2fdb | 1685 | struct xlog *log, |
1da177e4 LT |
1686 | xfs_lsn_t tail_lsn) |
1687 | { | |
1688 | int tail_cycle, head_cycle; | |
1689 | int tail_block, head_block; | |
1690 | int tail_distance, max_distance; | |
1691 | int distance; | |
1692 | int error; | |
1693 | ||
1694 | tail_cycle = CYCLE_LSN(tail_lsn); | |
1695 | tail_block = BLOCK_LSN(tail_lsn); | |
1696 | head_cycle = log->l_curr_cycle; | |
1697 | head_block = log->l_curr_block; | |
1698 | ||
1699 | /* | |
1700 | * Figure out the distance between the new head of the log | |
1701 | * and the tail. We want to write over any blocks beyond the | |
1702 | * head that we may have written just before the crash, but | |
1703 | * we don't want to overwrite the tail of the log. | |
1704 | */ | |
1705 | if (head_cycle == tail_cycle) { | |
1706 | /* | |
1707 | * The tail is behind the head in the physical log, | |
1708 | * so the distance from the head to the tail is the | |
1709 | * distance from the head to the end of the log plus | |
1710 | * the distance from the beginning of the log to the | |
1711 | * tail. | |
1712 | */ | |
1713 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { | |
1714 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", | |
1715 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 1716 | return -EFSCORRUPTED; |
1da177e4 LT |
1717 | } |
1718 | tail_distance = tail_block + (log->l_logBBsize - head_block); | |
1719 | } else { | |
1720 | /* | |
1721 | * The head is behind the tail in the physical log, | |
1722 | * so the distance from the head to the tail is just | |
1723 | * the tail block minus the head block. | |
1724 | */ | |
1725 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ | |
1726 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", | |
1727 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 1728 | return -EFSCORRUPTED; |
1da177e4 LT |
1729 | } |
1730 | tail_distance = tail_block - head_block; | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * If the head is right up against the tail, we can't clear | |
1735 | * anything. | |
1736 | */ | |
1737 | if (tail_distance <= 0) { | |
1738 | ASSERT(tail_distance == 0); | |
1739 | return 0; | |
1740 | } | |
1741 | ||
1742 | max_distance = XLOG_TOTAL_REC_SHIFT(log); | |
1743 | /* | |
1744 | * Take the smaller of the maximum amount of outstanding I/O | |
1745 | * we could have and the distance to the tail to clear out. | |
1746 | * We take the smaller so that we don't overwrite the tail and | |
1747 | * we don't waste all day writing from the head to the tail | |
1748 | * for no reason. | |
1749 | */ | |
1750 | max_distance = MIN(max_distance, tail_distance); | |
1751 | ||
1752 | if ((head_block + max_distance) <= log->l_logBBsize) { | |
1753 | /* | |
1754 | * We can stomp all the blocks we need to without | |
1755 | * wrapping around the end of the log. Just do it | |
1756 | * in a single write. Use the cycle number of the | |
1757 | * current cycle minus one so that the log will look like: | |
1758 | * n ... | n - 1 ... | |
1759 | */ | |
1760 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1761 | head_block, max_distance, tail_cycle, | |
1762 | tail_block); | |
1763 | if (error) | |
1764 | return error; | |
1765 | } else { | |
1766 | /* | |
1767 | * We need to wrap around the end of the physical log in | |
1768 | * order to clear all the blocks. Do it in two separate | |
1769 | * I/Os. The first write should be from the head to the | |
1770 | * end of the physical log, and it should use the current | |
1771 | * cycle number minus one just like above. | |
1772 | */ | |
1773 | distance = log->l_logBBsize - head_block; | |
1774 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1775 | head_block, distance, tail_cycle, | |
1776 | tail_block); | |
1777 | ||
1778 | if (error) | |
1779 | return error; | |
1780 | ||
1781 | /* | |
1782 | * Now write the blocks at the start of the physical log. | |
1783 | * This writes the remainder of the blocks we want to clear. | |
1784 | * It uses the current cycle number since we're now on the | |
1785 | * same cycle as the head so that we get: | |
1786 | * n ... n ... | n - 1 ... | |
1787 | * ^^^^^ blocks we're writing | |
1788 | */ | |
1789 | distance = max_distance - (log->l_logBBsize - head_block); | |
1790 | error = xlog_write_log_records(log, head_cycle, 0, distance, | |
1791 | tail_cycle, tail_block); | |
1792 | if (error) | |
1793 | return error; | |
1794 | } | |
1795 | ||
1796 | return 0; | |
1797 | } | |
1798 | ||
1799 | /****************************************************************************** | |
1800 | * | |
1801 | * Log recover routines | |
1802 | * | |
1803 | ****************************************************************************** | |
1804 | */ | |
1805 | ||
f0a76953 | 1806 | /* |
a775ad77 DC |
1807 | * Sort the log items in the transaction. |
1808 | * | |
1809 | * The ordering constraints are defined by the inode allocation and unlink | |
1810 | * behaviour. The rules are: | |
1811 | * | |
1812 | * 1. Every item is only logged once in a given transaction. Hence it | |
1813 | * represents the last logged state of the item. Hence ordering is | |
1814 | * dependent on the order in which operations need to be performed so | |
1815 | * required initial conditions are always met. | |
1816 | * | |
1817 | * 2. Cancelled buffers are recorded in pass 1 in a separate table and | |
1818 | * there's nothing to replay from them so we can simply cull them | |
1819 | * from the transaction. However, we can't do that until after we've | |
1820 | * replayed all the other items because they may be dependent on the | |
1821 | * cancelled buffer and replaying the cancelled buffer can remove it | |
1822 | * form the cancelled buffer table. Hence they have tobe done last. | |
1823 | * | |
1824 | * 3. Inode allocation buffers must be replayed before inode items that | |
28c8e41a DC |
1825 | * read the buffer and replay changes into it. For filesystems using the |
1826 | * ICREATE transactions, this means XFS_LI_ICREATE objects need to get | |
1827 | * treated the same as inode allocation buffers as they create and | |
1828 | * initialise the buffers directly. | |
a775ad77 DC |
1829 | * |
1830 | * 4. Inode unlink buffers must be replayed after inode items are replayed. | |
1831 | * This ensures that inodes are completely flushed to the inode buffer | |
1832 | * in a "free" state before we remove the unlinked inode list pointer. | |
1833 | * | |
1834 | * Hence the ordering needs to be inode allocation buffers first, inode items | |
1835 | * second, inode unlink buffers third and cancelled buffers last. | |
1836 | * | |
1837 | * But there's a problem with that - we can't tell an inode allocation buffer | |
1838 | * apart from a regular buffer, so we can't separate them. We can, however, | |
1839 | * tell an inode unlink buffer from the others, and so we can separate them out | |
1840 | * from all the other buffers and move them to last. | |
1841 | * | |
1842 | * Hence, 4 lists, in order from head to tail: | |
28c8e41a DC |
1843 | * - buffer_list for all buffers except cancelled/inode unlink buffers |
1844 | * - item_list for all non-buffer items | |
1845 | * - inode_buffer_list for inode unlink buffers | |
1846 | * - cancel_list for the cancelled buffers | |
1847 | * | |
1848 | * Note that we add objects to the tail of the lists so that first-to-last | |
1849 | * ordering is preserved within the lists. Adding objects to the head of the | |
1850 | * list means when we traverse from the head we walk them in last-to-first | |
1851 | * order. For cancelled buffers and inode unlink buffers this doesn't matter, | |
1852 | * but for all other items there may be specific ordering that we need to | |
1853 | * preserve. | |
f0a76953 | 1854 | */ |
1da177e4 LT |
1855 | STATIC int |
1856 | xlog_recover_reorder_trans( | |
ad223e60 MT |
1857 | struct xlog *log, |
1858 | struct xlog_recover *trans, | |
9abbc539 | 1859 | int pass) |
1da177e4 | 1860 | { |
f0a76953 | 1861 | xlog_recover_item_t *item, *n; |
2a84108f | 1862 | int error = 0; |
f0a76953 | 1863 | LIST_HEAD(sort_list); |
a775ad77 DC |
1864 | LIST_HEAD(cancel_list); |
1865 | LIST_HEAD(buffer_list); | |
1866 | LIST_HEAD(inode_buffer_list); | |
1867 | LIST_HEAD(inode_list); | |
f0a76953 DC |
1868 | |
1869 | list_splice_init(&trans->r_itemq, &sort_list); | |
1870 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { | |
4e0d5f92 | 1871 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
1da177e4 | 1872 | |
f0a76953 | 1873 | switch (ITEM_TYPE(item)) { |
28c8e41a DC |
1874 | case XFS_LI_ICREATE: |
1875 | list_move_tail(&item->ri_list, &buffer_list); | |
1876 | break; | |
1da177e4 | 1877 | case XFS_LI_BUF: |
a775ad77 | 1878 | if (buf_f->blf_flags & XFS_BLF_CANCEL) { |
9abbc539 DC |
1879 | trace_xfs_log_recover_item_reorder_head(log, |
1880 | trans, item, pass); | |
a775ad77 | 1881 | list_move(&item->ri_list, &cancel_list); |
1da177e4 LT |
1882 | break; |
1883 | } | |
a775ad77 DC |
1884 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
1885 | list_move(&item->ri_list, &inode_buffer_list); | |
1886 | break; | |
1887 | } | |
1888 | list_move_tail(&item->ri_list, &buffer_list); | |
1889 | break; | |
1da177e4 | 1890 | case XFS_LI_INODE: |
1da177e4 LT |
1891 | case XFS_LI_DQUOT: |
1892 | case XFS_LI_QUOTAOFF: | |
1893 | case XFS_LI_EFD: | |
1894 | case XFS_LI_EFI: | |
9abbc539 DC |
1895 | trace_xfs_log_recover_item_reorder_tail(log, |
1896 | trans, item, pass); | |
a775ad77 | 1897 | list_move_tail(&item->ri_list, &inode_list); |
1da177e4 LT |
1898 | break; |
1899 | default: | |
a0fa2b67 DC |
1900 | xfs_warn(log->l_mp, |
1901 | "%s: unrecognized type of log operation", | |
1902 | __func__); | |
1da177e4 | 1903 | ASSERT(0); |
2a84108f MT |
1904 | /* |
1905 | * return the remaining items back to the transaction | |
1906 | * item list so they can be freed in caller. | |
1907 | */ | |
1908 | if (!list_empty(&sort_list)) | |
1909 | list_splice_init(&sort_list, &trans->r_itemq); | |
2451337d | 1910 | error = -EIO; |
2a84108f | 1911 | goto out; |
1da177e4 | 1912 | } |
f0a76953 | 1913 | } |
2a84108f | 1914 | out: |
f0a76953 | 1915 | ASSERT(list_empty(&sort_list)); |
a775ad77 DC |
1916 | if (!list_empty(&buffer_list)) |
1917 | list_splice(&buffer_list, &trans->r_itemq); | |
1918 | if (!list_empty(&inode_list)) | |
1919 | list_splice_tail(&inode_list, &trans->r_itemq); | |
1920 | if (!list_empty(&inode_buffer_list)) | |
1921 | list_splice_tail(&inode_buffer_list, &trans->r_itemq); | |
1922 | if (!list_empty(&cancel_list)) | |
1923 | list_splice_tail(&cancel_list, &trans->r_itemq); | |
2a84108f | 1924 | return error; |
1da177e4 LT |
1925 | } |
1926 | ||
1927 | /* | |
1928 | * Build up the table of buf cancel records so that we don't replay | |
1929 | * cancelled data in the second pass. For buffer records that are | |
1930 | * not cancel records, there is nothing to do here so we just return. | |
1931 | * | |
1932 | * If we get a cancel record which is already in the table, this indicates | |
1933 | * that the buffer was cancelled multiple times. In order to ensure | |
1934 | * that during pass 2 we keep the record in the table until we reach its | |
1935 | * last occurrence in the log, we keep a reference count in the cancel | |
1936 | * record in the table to tell us how many times we expect to see this | |
1937 | * record during the second pass. | |
1938 | */ | |
c9f71f5f CH |
1939 | STATIC int |
1940 | xlog_recover_buffer_pass1( | |
ad223e60 MT |
1941 | struct xlog *log, |
1942 | struct xlog_recover_item *item) | |
1da177e4 | 1943 | { |
c9f71f5f | 1944 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
d5689eaa CH |
1945 | struct list_head *bucket; |
1946 | struct xfs_buf_cancel *bcp; | |
1da177e4 LT |
1947 | |
1948 | /* | |
1949 | * If this isn't a cancel buffer item, then just return. | |
1950 | */ | |
e2714bf8 | 1951 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
9abbc539 | 1952 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); |
c9f71f5f | 1953 | return 0; |
9abbc539 | 1954 | } |
1da177e4 LT |
1955 | |
1956 | /* | |
d5689eaa CH |
1957 | * Insert an xfs_buf_cancel record into the hash table of them. |
1958 | * If there is already an identical record, bump its reference count. | |
1da177e4 | 1959 | */ |
d5689eaa CH |
1960 | bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno); |
1961 | list_for_each_entry(bcp, bucket, bc_list) { | |
1962 | if (bcp->bc_blkno == buf_f->blf_blkno && | |
1963 | bcp->bc_len == buf_f->blf_len) { | |
1964 | bcp->bc_refcount++; | |
9abbc539 | 1965 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
c9f71f5f | 1966 | return 0; |
1da177e4 | 1967 | } |
d5689eaa CH |
1968 | } |
1969 | ||
1970 | bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP); | |
1971 | bcp->bc_blkno = buf_f->blf_blkno; | |
1972 | bcp->bc_len = buf_f->blf_len; | |
1da177e4 | 1973 | bcp->bc_refcount = 1; |
d5689eaa CH |
1974 | list_add_tail(&bcp->bc_list, bucket); |
1975 | ||
9abbc539 | 1976 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
c9f71f5f | 1977 | return 0; |
1da177e4 LT |
1978 | } |
1979 | ||
1980 | /* | |
1981 | * Check to see whether the buffer being recovered has a corresponding | |
84a5b730 DC |
1982 | * entry in the buffer cancel record table. If it is, return the cancel |
1983 | * buffer structure to the caller. | |
1da177e4 | 1984 | */ |
84a5b730 DC |
1985 | STATIC struct xfs_buf_cancel * |
1986 | xlog_peek_buffer_cancelled( | |
ad223e60 | 1987 | struct xlog *log, |
1da177e4 LT |
1988 | xfs_daddr_t blkno, |
1989 | uint len, | |
1990 | ushort flags) | |
1991 | { | |
d5689eaa CH |
1992 | struct list_head *bucket; |
1993 | struct xfs_buf_cancel *bcp; | |
1da177e4 | 1994 | |
84a5b730 DC |
1995 | if (!log->l_buf_cancel_table) { |
1996 | /* empty table means no cancelled buffers in the log */ | |
c1155410 | 1997 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
84a5b730 | 1998 | return NULL; |
1da177e4 LT |
1999 | } |
2000 | ||
d5689eaa CH |
2001 | bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno); |
2002 | list_for_each_entry(bcp, bucket, bc_list) { | |
2003 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) | |
84a5b730 | 2004 | return bcp; |
1da177e4 | 2005 | } |
d5689eaa | 2006 | |
1da177e4 | 2007 | /* |
d5689eaa CH |
2008 | * We didn't find a corresponding entry in the table, so return 0 so |
2009 | * that the buffer is NOT cancelled. | |
1da177e4 | 2010 | */ |
c1155410 | 2011 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
84a5b730 DC |
2012 | return NULL; |
2013 | } | |
2014 | ||
2015 | /* | |
2016 | * If the buffer is being cancelled then return 1 so that it will be cancelled, | |
2017 | * otherwise return 0. If the buffer is actually a buffer cancel item | |
2018 | * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the | |
2019 | * table and remove it from the table if this is the last reference. | |
2020 | * | |
2021 | * We remove the cancel record from the table when we encounter its last | |
2022 | * occurrence in the log so that if the same buffer is re-used again after its | |
2023 | * last cancellation we actually replay the changes made at that point. | |
2024 | */ | |
2025 | STATIC int | |
2026 | xlog_check_buffer_cancelled( | |
2027 | struct xlog *log, | |
2028 | xfs_daddr_t blkno, | |
2029 | uint len, | |
2030 | ushort flags) | |
2031 | { | |
2032 | struct xfs_buf_cancel *bcp; | |
2033 | ||
2034 | bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags); | |
2035 | if (!bcp) | |
2036 | return 0; | |
d5689eaa | 2037 | |
d5689eaa CH |
2038 | /* |
2039 | * We've go a match, so return 1 so that the recovery of this buffer | |
2040 | * is cancelled. If this buffer is actually a buffer cancel log | |
2041 | * item, then decrement the refcount on the one in the table and | |
2042 | * remove it if this is the last reference. | |
2043 | */ | |
2044 | if (flags & XFS_BLF_CANCEL) { | |
2045 | if (--bcp->bc_refcount == 0) { | |
2046 | list_del(&bcp->bc_list); | |
2047 | kmem_free(bcp); | |
2048 | } | |
2049 | } | |
2050 | return 1; | |
1da177e4 LT |
2051 | } |
2052 | ||
1da177e4 | 2053 | /* |
e2714bf8 CH |
2054 | * Perform recovery for a buffer full of inodes. In these buffers, the only |
2055 | * data which should be recovered is that which corresponds to the | |
2056 | * di_next_unlinked pointers in the on disk inode structures. The rest of the | |
2057 | * data for the inodes is always logged through the inodes themselves rather | |
2058 | * than the inode buffer and is recovered in xlog_recover_inode_pass2(). | |
1da177e4 | 2059 | * |
e2714bf8 CH |
2060 | * The only time when buffers full of inodes are fully recovered is when the |
2061 | * buffer is full of newly allocated inodes. In this case the buffer will | |
2062 | * not be marked as an inode buffer and so will be sent to | |
2063 | * xlog_recover_do_reg_buffer() below during recovery. | |
1da177e4 LT |
2064 | */ |
2065 | STATIC int | |
2066 | xlog_recover_do_inode_buffer( | |
e2714bf8 | 2067 | struct xfs_mount *mp, |
1da177e4 | 2068 | xlog_recover_item_t *item, |
e2714bf8 | 2069 | struct xfs_buf *bp, |
1da177e4 LT |
2070 | xfs_buf_log_format_t *buf_f) |
2071 | { | |
2072 | int i; | |
e2714bf8 CH |
2073 | int item_index = 0; |
2074 | int bit = 0; | |
2075 | int nbits = 0; | |
2076 | int reg_buf_offset = 0; | |
2077 | int reg_buf_bytes = 0; | |
1da177e4 LT |
2078 | int next_unlinked_offset; |
2079 | int inodes_per_buf; | |
2080 | xfs_agino_t *logged_nextp; | |
2081 | xfs_agino_t *buffer_nextp; | |
1da177e4 | 2082 | |
9abbc539 | 2083 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
9222a9cf DC |
2084 | |
2085 | /* | |
2086 | * Post recovery validation only works properly on CRC enabled | |
2087 | * filesystems. | |
2088 | */ | |
2089 | if (xfs_sb_version_hascrc(&mp->m_sb)) | |
2090 | bp->b_ops = &xfs_inode_buf_ops; | |
9abbc539 | 2091 | |
aa0e8833 | 2092 | inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog; |
1da177e4 LT |
2093 | for (i = 0; i < inodes_per_buf; i++) { |
2094 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + | |
2095 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2096 | ||
2097 | while (next_unlinked_offset >= | |
2098 | (reg_buf_offset + reg_buf_bytes)) { | |
2099 | /* | |
2100 | * The next di_next_unlinked field is beyond | |
2101 | * the current logged region. Find the next | |
2102 | * logged region that contains or is beyond | |
2103 | * the current di_next_unlinked field. | |
2104 | */ | |
2105 | bit += nbits; | |
e2714bf8 CH |
2106 | bit = xfs_next_bit(buf_f->blf_data_map, |
2107 | buf_f->blf_map_size, bit); | |
1da177e4 LT |
2108 | |
2109 | /* | |
2110 | * If there are no more logged regions in the | |
2111 | * buffer, then we're done. | |
2112 | */ | |
e2714bf8 | 2113 | if (bit == -1) |
1da177e4 | 2114 | return 0; |
1da177e4 | 2115 | |
e2714bf8 CH |
2116 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
2117 | buf_f->blf_map_size, bit); | |
1da177e4 | 2118 | ASSERT(nbits > 0); |
c1155410 DC |
2119 | reg_buf_offset = bit << XFS_BLF_SHIFT; |
2120 | reg_buf_bytes = nbits << XFS_BLF_SHIFT; | |
1da177e4 LT |
2121 | item_index++; |
2122 | } | |
2123 | ||
2124 | /* | |
2125 | * If the current logged region starts after the current | |
2126 | * di_next_unlinked field, then move on to the next | |
2127 | * di_next_unlinked field. | |
2128 | */ | |
e2714bf8 | 2129 | if (next_unlinked_offset < reg_buf_offset) |
1da177e4 | 2130 | continue; |
1da177e4 LT |
2131 | |
2132 | ASSERT(item->ri_buf[item_index].i_addr != NULL); | |
c1155410 | 2133 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0); |
aa0e8833 DC |
2134 | ASSERT((reg_buf_offset + reg_buf_bytes) <= |
2135 | BBTOB(bp->b_io_length)); | |
1da177e4 LT |
2136 | |
2137 | /* | |
2138 | * The current logged region contains a copy of the | |
2139 | * current di_next_unlinked field. Extract its value | |
2140 | * and copy it to the buffer copy. | |
2141 | */ | |
4e0d5f92 CH |
2142 | logged_nextp = item->ri_buf[item_index].i_addr + |
2143 | next_unlinked_offset - reg_buf_offset; | |
1da177e4 | 2144 | if (unlikely(*logged_nextp == 0)) { |
a0fa2b67 DC |
2145 | xfs_alert(mp, |
2146 | "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). " | |
2147 | "Trying to replay bad (0) inode di_next_unlinked field.", | |
1da177e4 LT |
2148 | item, bp); |
2149 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", | |
2150 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 2151 | return -EFSCORRUPTED; |
1da177e4 LT |
2152 | } |
2153 | ||
88ee2df7 | 2154 | buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset); |
87c199c2 | 2155 | *buffer_nextp = *logged_nextp; |
0a32c26e DC |
2156 | |
2157 | /* | |
2158 | * If necessary, recalculate the CRC in the on-disk inode. We | |
2159 | * have to leave the inode in a consistent state for whoever | |
2160 | * reads it next.... | |
2161 | */ | |
88ee2df7 | 2162 | xfs_dinode_calc_crc(mp, |
0a32c26e DC |
2163 | xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize)); |
2164 | ||
1da177e4 LT |
2165 | } |
2166 | ||
2167 | return 0; | |
2168 | } | |
2169 | ||
50d5c8d8 DC |
2170 | /* |
2171 | * V5 filesystems know the age of the buffer on disk being recovered. We can | |
2172 | * have newer objects on disk than we are replaying, and so for these cases we | |
2173 | * don't want to replay the current change as that will make the buffer contents | |
2174 | * temporarily invalid on disk. | |
2175 | * | |
2176 | * The magic number might not match the buffer type we are going to recover | |
2177 | * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence | |
2178 | * extract the LSN of the existing object in the buffer based on it's current | |
2179 | * magic number. If we don't recognise the magic number in the buffer, then | |
2180 | * return a LSN of -1 so that the caller knows it was an unrecognised block and | |
2181 | * so can recover the buffer. | |
566055d3 DC |
2182 | * |
2183 | * Note: we cannot rely solely on magic number matches to determine that the | |
2184 | * buffer has a valid LSN - we also need to verify that it belongs to this | |
2185 | * filesystem, so we need to extract the object's LSN and compare it to that | |
2186 | * which we read from the superblock. If the UUIDs don't match, then we've got a | |
2187 | * stale metadata block from an old filesystem instance that we need to recover | |
2188 | * over the top of. | |
50d5c8d8 DC |
2189 | */ |
2190 | static xfs_lsn_t | |
2191 | xlog_recover_get_buf_lsn( | |
2192 | struct xfs_mount *mp, | |
2193 | struct xfs_buf *bp) | |
2194 | { | |
2195 | __uint32_t magic32; | |
2196 | __uint16_t magic16; | |
2197 | __uint16_t magicda; | |
2198 | void *blk = bp->b_addr; | |
566055d3 DC |
2199 | uuid_t *uuid; |
2200 | xfs_lsn_t lsn = -1; | |
50d5c8d8 DC |
2201 | |
2202 | /* v4 filesystems always recover immediately */ | |
2203 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2204 | goto recover_immediately; | |
2205 | ||
2206 | magic32 = be32_to_cpu(*(__be32 *)blk); | |
2207 | switch (magic32) { | |
2208 | case XFS_ABTB_CRC_MAGIC: | |
2209 | case XFS_ABTC_CRC_MAGIC: | |
2210 | case XFS_ABTB_MAGIC: | |
2211 | case XFS_ABTC_MAGIC: | |
2212 | case XFS_IBT_CRC_MAGIC: | |
566055d3 DC |
2213 | case XFS_IBT_MAGIC: { |
2214 | struct xfs_btree_block *btb = blk; | |
2215 | ||
2216 | lsn = be64_to_cpu(btb->bb_u.s.bb_lsn); | |
2217 | uuid = &btb->bb_u.s.bb_uuid; | |
2218 | break; | |
2219 | } | |
50d5c8d8 | 2220 | case XFS_BMAP_CRC_MAGIC: |
566055d3 DC |
2221 | case XFS_BMAP_MAGIC: { |
2222 | struct xfs_btree_block *btb = blk; | |
2223 | ||
2224 | lsn = be64_to_cpu(btb->bb_u.l.bb_lsn); | |
2225 | uuid = &btb->bb_u.l.bb_uuid; | |
2226 | break; | |
2227 | } | |
50d5c8d8 | 2228 | case XFS_AGF_MAGIC: |
566055d3 DC |
2229 | lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn); |
2230 | uuid = &((struct xfs_agf *)blk)->agf_uuid; | |
2231 | break; | |
50d5c8d8 | 2232 | case XFS_AGFL_MAGIC: |
566055d3 DC |
2233 | lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn); |
2234 | uuid = &((struct xfs_agfl *)blk)->agfl_uuid; | |
2235 | break; | |
50d5c8d8 | 2236 | case XFS_AGI_MAGIC: |
566055d3 DC |
2237 | lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn); |
2238 | uuid = &((struct xfs_agi *)blk)->agi_uuid; | |
2239 | break; | |
50d5c8d8 | 2240 | case XFS_SYMLINK_MAGIC: |
566055d3 DC |
2241 | lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn); |
2242 | uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid; | |
2243 | break; | |
50d5c8d8 DC |
2244 | case XFS_DIR3_BLOCK_MAGIC: |
2245 | case XFS_DIR3_DATA_MAGIC: | |
2246 | case XFS_DIR3_FREE_MAGIC: | |
566055d3 DC |
2247 | lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn); |
2248 | uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid; | |
2249 | break; | |
50d5c8d8 | 2250 | case XFS_ATTR3_RMT_MAGIC: |
e3c32ee9 DC |
2251 | /* |
2252 | * Remote attr blocks are written synchronously, rather than | |
2253 | * being logged. That means they do not contain a valid LSN | |
2254 | * (i.e. transactionally ordered) in them, and hence any time we | |
2255 | * see a buffer to replay over the top of a remote attribute | |
2256 | * block we should simply do so. | |
2257 | */ | |
2258 | goto recover_immediately; | |
50d5c8d8 | 2259 | case XFS_SB_MAGIC: |
fcfbe2c4 DC |
2260 | /* |
2261 | * superblock uuids are magic. We may or may not have a | |
2262 | * sb_meta_uuid on disk, but it will be set in the in-core | |
2263 | * superblock. We set the uuid pointer for verification | |
2264 | * according to the superblock feature mask to ensure we check | |
2265 | * the relevant UUID in the superblock. | |
2266 | */ | |
566055d3 | 2267 | lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn); |
fcfbe2c4 DC |
2268 | if (xfs_sb_version_hasmetauuid(&mp->m_sb)) |
2269 | uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid; | |
2270 | else | |
2271 | uuid = &((struct xfs_dsb *)blk)->sb_uuid; | |
566055d3 | 2272 | break; |
50d5c8d8 DC |
2273 | default: |
2274 | break; | |
2275 | } | |
2276 | ||
566055d3 | 2277 | if (lsn != (xfs_lsn_t)-1) { |
fcfbe2c4 | 2278 | if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid)) |
566055d3 DC |
2279 | goto recover_immediately; |
2280 | return lsn; | |
2281 | } | |
2282 | ||
50d5c8d8 DC |
2283 | magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic); |
2284 | switch (magicda) { | |
2285 | case XFS_DIR3_LEAF1_MAGIC: | |
2286 | case XFS_DIR3_LEAFN_MAGIC: | |
2287 | case XFS_DA3_NODE_MAGIC: | |
566055d3 DC |
2288 | lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn); |
2289 | uuid = &((struct xfs_da3_blkinfo *)blk)->uuid; | |
2290 | break; | |
50d5c8d8 DC |
2291 | default: |
2292 | break; | |
2293 | } | |
2294 | ||
566055d3 DC |
2295 | if (lsn != (xfs_lsn_t)-1) { |
2296 | if (!uuid_equal(&mp->m_sb.sb_uuid, uuid)) | |
2297 | goto recover_immediately; | |
2298 | return lsn; | |
2299 | } | |
2300 | ||
50d5c8d8 DC |
2301 | /* |
2302 | * We do individual object checks on dquot and inode buffers as they | |
2303 | * have their own individual LSN records. Also, we could have a stale | |
2304 | * buffer here, so we have to at least recognise these buffer types. | |
2305 | * | |
2306 | * A notd complexity here is inode unlinked list processing - it logs | |
2307 | * the inode directly in the buffer, but we don't know which inodes have | |
2308 | * been modified, and there is no global buffer LSN. Hence we need to | |
2309 | * recover all inode buffer types immediately. This problem will be | |
2310 | * fixed by logical logging of the unlinked list modifications. | |
2311 | */ | |
2312 | magic16 = be16_to_cpu(*(__be16 *)blk); | |
2313 | switch (magic16) { | |
2314 | case XFS_DQUOT_MAGIC: | |
2315 | case XFS_DINODE_MAGIC: | |
2316 | goto recover_immediately; | |
2317 | default: | |
2318 | break; | |
2319 | } | |
2320 | ||
2321 | /* unknown buffer contents, recover immediately */ | |
2322 | ||
2323 | recover_immediately: | |
2324 | return (xfs_lsn_t)-1; | |
2325 | ||
2326 | } | |
2327 | ||
1da177e4 | 2328 | /* |
d75afeb3 DC |
2329 | * Validate the recovered buffer is of the correct type and attach the |
2330 | * appropriate buffer operations to them for writeback. Magic numbers are in a | |
2331 | * few places: | |
2332 | * the first 16 bits of the buffer (inode buffer, dquot buffer), | |
2333 | * the first 32 bits of the buffer (most blocks), | |
2334 | * inside a struct xfs_da_blkinfo at the start of the buffer. | |
1da177e4 | 2335 | */ |
d75afeb3 | 2336 | static void |
50d5c8d8 | 2337 | xlog_recover_validate_buf_type( |
9abbc539 | 2338 | struct xfs_mount *mp, |
e2714bf8 | 2339 | struct xfs_buf *bp, |
1da177e4 LT |
2340 | xfs_buf_log_format_t *buf_f) |
2341 | { | |
d75afeb3 DC |
2342 | struct xfs_da_blkinfo *info = bp->b_addr; |
2343 | __uint32_t magic32; | |
2344 | __uint16_t magic16; | |
2345 | __uint16_t magicda; | |
2346 | ||
67dc288c DC |
2347 | /* |
2348 | * We can only do post recovery validation on items on CRC enabled | |
2349 | * fielsystems as we need to know when the buffer was written to be able | |
2350 | * to determine if we should have replayed the item. If we replay old | |
2351 | * metadata over a newer buffer, then it will enter a temporarily | |
2352 | * inconsistent state resulting in verification failures. Hence for now | |
2353 | * just avoid the verification stage for non-crc filesystems | |
2354 | */ | |
2355 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2356 | return; | |
2357 | ||
d75afeb3 DC |
2358 | magic32 = be32_to_cpu(*(__be32 *)bp->b_addr); |
2359 | magic16 = be16_to_cpu(*(__be16*)bp->b_addr); | |
2360 | magicda = be16_to_cpu(info->magic); | |
61fe135c DC |
2361 | switch (xfs_blft_from_flags(buf_f)) { |
2362 | case XFS_BLFT_BTREE_BUF: | |
d75afeb3 | 2363 | switch (magic32) { |
ee1a47ab CH |
2364 | case XFS_ABTB_CRC_MAGIC: |
2365 | case XFS_ABTC_CRC_MAGIC: | |
2366 | case XFS_ABTB_MAGIC: | |
2367 | case XFS_ABTC_MAGIC: | |
2368 | bp->b_ops = &xfs_allocbt_buf_ops; | |
2369 | break; | |
2370 | case XFS_IBT_CRC_MAGIC: | |
aafc3c24 | 2371 | case XFS_FIBT_CRC_MAGIC: |
ee1a47ab | 2372 | case XFS_IBT_MAGIC: |
aafc3c24 | 2373 | case XFS_FIBT_MAGIC: |
ee1a47ab CH |
2374 | bp->b_ops = &xfs_inobt_buf_ops; |
2375 | break; | |
2376 | case XFS_BMAP_CRC_MAGIC: | |
2377 | case XFS_BMAP_MAGIC: | |
2378 | bp->b_ops = &xfs_bmbt_buf_ops; | |
2379 | break; | |
2380 | default: | |
2381 | xfs_warn(mp, "Bad btree block magic!"); | |
2382 | ASSERT(0); | |
2383 | break; | |
2384 | } | |
2385 | break; | |
61fe135c | 2386 | case XFS_BLFT_AGF_BUF: |
d75afeb3 | 2387 | if (magic32 != XFS_AGF_MAGIC) { |
4e0e6040 DC |
2388 | xfs_warn(mp, "Bad AGF block magic!"); |
2389 | ASSERT(0); | |
2390 | break; | |
2391 | } | |
2392 | bp->b_ops = &xfs_agf_buf_ops; | |
2393 | break; | |
61fe135c | 2394 | case XFS_BLFT_AGFL_BUF: |
d75afeb3 | 2395 | if (magic32 != XFS_AGFL_MAGIC) { |
77c95bba CH |
2396 | xfs_warn(mp, "Bad AGFL block magic!"); |
2397 | ASSERT(0); | |
2398 | break; | |
2399 | } | |
2400 | bp->b_ops = &xfs_agfl_buf_ops; | |
2401 | break; | |
61fe135c | 2402 | case XFS_BLFT_AGI_BUF: |
d75afeb3 | 2403 | if (magic32 != XFS_AGI_MAGIC) { |
983d09ff DC |
2404 | xfs_warn(mp, "Bad AGI block magic!"); |
2405 | ASSERT(0); | |
2406 | break; | |
2407 | } | |
2408 | bp->b_ops = &xfs_agi_buf_ops; | |
2409 | break; | |
61fe135c DC |
2410 | case XFS_BLFT_UDQUOT_BUF: |
2411 | case XFS_BLFT_PDQUOT_BUF: | |
2412 | case XFS_BLFT_GDQUOT_BUF: | |
123887e8 | 2413 | #ifdef CONFIG_XFS_QUOTA |
d75afeb3 | 2414 | if (magic16 != XFS_DQUOT_MAGIC) { |
3fe58f30 CH |
2415 | xfs_warn(mp, "Bad DQUOT block magic!"); |
2416 | ASSERT(0); | |
2417 | break; | |
2418 | } | |
2419 | bp->b_ops = &xfs_dquot_buf_ops; | |
123887e8 DC |
2420 | #else |
2421 | xfs_alert(mp, | |
2422 | "Trying to recover dquots without QUOTA support built in!"); | |
2423 | ASSERT(0); | |
2424 | #endif | |
3fe58f30 | 2425 | break; |
61fe135c | 2426 | case XFS_BLFT_DINO_BUF: |
d75afeb3 | 2427 | if (magic16 != XFS_DINODE_MAGIC) { |
93848a99 CH |
2428 | xfs_warn(mp, "Bad INODE block magic!"); |
2429 | ASSERT(0); | |
2430 | break; | |
2431 | } | |
2432 | bp->b_ops = &xfs_inode_buf_ops; | |
2433 | break; | |
61fe135c | 2434 | case XFS_BLFT_SYMLINK_BUF: |
d75afeb3 | 2435 | if (magic32 != XFS_SYMLINK_MAGIC) { |
f948dd76 DC |
2436 | xfs_warn(mp, "Bad symlink block magic!"); |
2437 | ASSERT(0); | |
2438 | break; | |
2439 | } | |
2440 | bp->b_ops = &xfs_symlink_buf_ops; | |
2441 | break; | |
61fe135c | 2442 | case XFS_BLFT_DIR_BLOCK_BUF: |
d75afeb3 DC |
2443 | if (magic32 != XFS_DIR2_BLOCK_MAGIC && |
2444 | magic32 != XFS_DIR3_BLOCK_MAGIC) { | |
2445 | xfs_warn(mp, "Bad dir block magic!"); | |
2446 | ASSERT(0); | |
2447 | break; | |
2448 | } | |
2449 | bp->b_ops = &xfs_dir3_block_buf_ops; | |
2450 | break; | |
61fe135c | 2451 | case XFS_BLFT_DIR_DATA_BUF: |
d75afeb3 DC |
2452 | if (magic32 != XFS_DIR2_DATA_MAGIC && |
2453 | magic32 != XFS_DIR3_DATA_MAGIC) { | |
2454 | xfs_warn(mp, "Bad dir data magic!"); | |
2455 | ASSERT(0); | |
2456 | break; | |
2457 | } | |
2458 | bp->b_ops = &xfs_dir3_data_buf_ops; | |
2459 | break; | |
61fe135c | 2460 | case XFS_BLFT_DIR_FREE_BUF: |
d75afeb3 DC |
2461 | if (magic32 != XFS_DIR2_FREE_MAGIC && |
2462 | magic32 != XFS_DIR3_FREE_MAGIC) { | |
2463 | xfs_warn(mp, "Bad dir3 free magic!"); | |
2464 | ASSERT(0); | |
2465 | break; | |
2466 | } | |
2467 | bp->b_ops = &xfs_dir3_free_buf_ops; | |
2468 | break; | |
61fe135c | 2469 | case XFS_BLFT_DIR_LEAF1_BUF: |
d75afeb3 DC |
2470 | if (magicda != XFS_DIR2_LEAF1_MAGIC && |
2471 | magicda != XFS_DIR3_LEAF1_MAGIC) { | |
2472 | xfs_warn(mp, "Bad dir leaf1 magic!"); | |
2473 | ASSERT(0); | |
2474 | break; | |
2475 | } | |
2476 | bp->b_ops = &xfs_dir3_leaf1_buf_ops; | |
2477 | break; | |
61fe135c | 2478 | case XFS_BLFT_DIR_LEAFN_BUF: |
d75afeb3 DC |
2479 | if (magicda != XFS_DIR2_LEAFN_MAGIC && |
2480 | magicda != XFS_DIR3_LEAFN_MAGIC) { | |
2481 | xfs_warn(mp, "Bad dir leafn magic!"); | |
2482 | ASSERT(0); | |
2483 | break; | |
2484 | } | |
2485 | bp->b_ops = &xfs_dir3_leafn_buf_ops; | |
2486 | break; | |
61fe135c | 2487 | case XFS_BLFT_DA_NODE_BUF: |
d75afeb3 DC |
2488 | if (magicda != XFS_DA_NODE_MAGIC && |
2489 | magicda != XFS_DA3_NODE_MAGIC) { | |
2490 | xfs_warn(mp, "Bad da node magic!"); | |
2491 | ASSERT(0); | |
2492 | break; | |
2493 | } | |
2494 | bp->b_ops = &xfs_da3_node_buf_ops; | |
2495 | break; | |
61fe135c | 2496 | case XFS_BLFT_ATTR_LEAF_BUF: |
d75afeb3 DC |
2497 | if (magicda != XFS_ATTR_LEAF_MAGIC && |
2498 | magicda != XFS_ATTR3_LEAF_MAGIC) { | |
2499 | xfs_warn(mp, "Bad attr leaf magic!"); | |
2500 | ASSERT(0); | |
2501 | break; | |
2502 | } | |
2503 | bp->b_ops = &xfs_attr3_leaf_buf_ops; | |
2504 | break; | |
61fe135c | 2505 | case XFS_BLFT_ATTR_RMT_BUF: |
cab09a81 | 2506 | if (magic32 != XFS_ATTR3_RMT_MAGIC) { |
d75afeb3 DC |
2507 | xfs_warn(mp, "Bad attr remote magic!"); |
2508 | ASSERT(0); | |
2509 | break; | |
2510 | } | |
2511 | bp->b_ops = &xfs_attr3_rmt_buf_ops; | |
2512 | break; | |
04a1e6c5 DC |
2513 | case XFS_BLFT_SB_BUF: |
2514 | if (magic32 != XFS_SB_MAGIC) { | |
2515 | xfs_warn(mp, "Bad SB block magic!"); | |
2516 | ASSERT(0); | |
2517 | break; | |
2518 | } | |
2519 | bp->b_ops = &xfs_sb_buf_ops; | |
2520 | break; | |
ee1a47ab | 2521 | default: |
61fe135c DC |
2522 | xfs_warn(mp, "Unknown buffer type %d!", |
2523 | xfs_blft_from_flags(buf_f)); | |
ee1a47ab CH |
2524 | break; |
2525 | } | |
1da177e4 LT |
2526 | } |
2527 | ||
d75afeb3 DC |
2528 | /* |
2529 | * Perform a 'normal' buffer recovery. Each logged region of the | |
2530 | * buffer should be copied over the corresponding region in the | |
2531 | * given buffer. The bitmap in the buf log format structure indicates | |
2532 | * where to place the logged data. | |
2533 | */ | |
2534 | STATIC void | |
2535 | xlog_recover_do_reg_buffer( | |
2536 | struct xfs_mount *mp, | |
2537 | xlog_recover_item_t *item, | |
2538 | struct xfs_buf *bp, | |
2539 | xfs_buf_log_format_t *buf_f) | |
2540 | { | |
2541 | int i; | |
2542 | int bit; | |
2543 | int nbits; | |
2544 | int error; | |
2545 | ||
2546 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); | |
2547 | ||
2548 | bit = 0; | |
2549 | i = 1; /* 0 is the buf format structure */ | |
2550 | while (1) { | |
2551 | bit = xfs_next_bit(buf_f->blf_data_map, | |
2552 | buf_f->blf_map_size, bit); | |
2553 | if (bit == -1) | |
2554 | break; | |
2555 | nbits = xfs_contig_bits(buf_f->blf_data_map, | |
2556 | buf_f->blf_map_size, bit); | |
2557 | ASSERT(nbits > 0); | |
2558 | ASSERT(item->ri_buf[i].i_addr != NULL); | |
2559 | ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0); | |
2560 | ASSERT(BBTOB(bp->b_io_length) >= | |
2561 | ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT)); | |
2562 | ||
709da6a6 DC |
2563 | /* |
2564 | * The dirty regions logged in the buffer, even though | |
2565 | * contiguous, may span multiple chunks. This is because the | |
2566 | * dirty region may span a physical page boundary in a buffer | |
2567 | * and hence be split into two separate vectors for writing into | |
2568 | * the log. Hence we need to trim nbits back to the length of | |
2569 | * the current region being copied out of the log. | |
2570 | */ | |
2571 | if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT)) | |
2572 | nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT; | |
2573 | ||
d75afeb3 DC |
2574 | /* |
2575 | * Do a sanity check if this is a dquot buffer. Just checking | |
2576 | * the first dquot in the buffer should do. XXXThis is | |
2577 | * probably a good thing to do for other buf types also. | |
2578 | */ | |
2579 | error = 0; | |
2580 | if (buf_f->blf_flags & | |
2581 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { | |
2582 | if (item->ri_buf[i].i_addr == NULL) { | |
2583 | xfs_alert(mp, | |
2584 | "XFS: NULL dquot in %s.", __func__); | |
2585 | goto next; | |
2586 | } | |
2587 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { | |
2588 | xfs_alert(mp, | |
2589 | "XFS: dquot too small (%d) in %s.", | |
2590 | item->ri_buf[i].i_len, __func__); | |
2591 | goto next; | |
2592 | } | |
9aede1d8 | 2593 | error = xfs_dqcheck(mp, item->ri_buf[i].i_addr, |
d75afeb3 DC |
2594 | -1, 0, XFS_QMOPT_DOWARN, |
2595 | "dquot_buf_recover"); | |
2596 | if (error) | |
2597 | goto next; | |
2598 | } | |
2599 | ||
2600 | memcpy(xfs_buf_offset(bp, | |
2601 | (uint)bit << XFS_BLF_SHIFT), /* dest */ | |
2602 | item->ri_buf[i].i_addr, /* source */ | |
2603 | nbits<<XFS_BLF_SHIFT); /* length */ | |
2604 | next: | |
2605 | i++; | |
2606 | bit += nbits; | |
2607 | } | |
2608 | ||
2609 | /* Shouldn't be any more regions */ | |
2610 | ASSERT(i == item->ri_total); | |
2611 | ||
67dc288c | 2612 | xlog_recover_validate_buf_type(mp, bp, buf_f); |
d75afeb3 DC |
2613 | } |
2614 | ||
1da177e4 LT |
2615 | /* |
2616 | * Perform a dquot buffer recovery. | |
8ba701ee | 2617 | * Simple algorithm: if we have found a QUOTAOFF log item of the same type |
1da177e4 LT |
2618 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. |
2619 | * Else, treat it as a regular buffer and do recovery. | |
ad3714b8 DC |
2620 | * |
2621 | * Return false if the buffer was tossed and true if we recovered the buffer to | |
2622 | * indicate to the caller if the buffer needs writing. | |
1da177e4 | 2623 | */ |
ad3714b8 | 2624 | STATIC bool |
1da177e4 | 2625 | xlog_recover_do_dquot_buffer( |
9a8d2fdb MT |
2626 | struct xfs_mount *mp, |
2627 | struct xlog *log, | |
2628 | struct xlog_recover_item *item, | |
2629 | struct xfs_buf *bp, | |
2630 | struct xfs_buf_log_format *buf_f) | |
1da177e4 LT |
2631 | { |
2632 | uint type; | |
2633 | ||
9abbc539 DC |
2634 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
2635 | ||
1da177e4 LT |
2636 | /* |
2637 | * Filesystems are required to send in quota flags at mount time. | |
2638 | */ | |
ad3714b8 DC |
2639 | if (!mp->m_qflags) |
2640 | return false; | |
1da177e4 LT |
2641 | |
2642 | type = 0; | |
c1155410 | 2643 | if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF) |
1da177e4 | 2644 | type |= XFS_DQ_USER; |
c1155410 | 2645 | if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF) |
c8ad20ff | 2646 | type |= XFS_DQ_PROJ; |
c1155410 | 2647 | if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF) |
1da177e4 LT |
2648 | type |= XFS_DQ_GROUP; |
2649 | /* | |
2650 | * This type of quotas was turned off, so ignore this buffer | |
2651 | */ | |
2652 | if (log->l_quotaoffs_flag & type) | |
ad3714b8 | 2653 | return false; |
1da177e4 | 2654 | |
9abbc539 | 2655 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
ad3714b8 | 2656 | return true; |
1da177e4 LT |
2657 | } |
2658 | ||
2659 | /* | |
2660 | * This routine replays a modification made to a buffer at runtime. | |
2661 | * There are actually two types of buffer, regular and inode, which | |
2662 | * are handled differently. Inode buffers are handled differently | |
2663 | * in that we only recover a specific set of data from them, namely | |
2664 | * the inode di_next_unlinked fields. This is because all other inode | |
2665 | * data is actually logged via inode records and any data we replay | |
2666 | * here which overlaps that may be stale. | |
2667 | * | |
2668 | * When meta-data buffers are freed at run time we log a buffer item | |
c1155410 | 2669 | * with the XFS_BLF_CANCEL bit set to indicate that previous copies |
1da177e4 LT |
2670 | * of the buffer in the log should not be replayed at recovery time. |
2671 | * This is so that if the blocks covered by the buffer are reused for | |
2672 | * file data before we crash we don't end up replaying old, freed | |
2673 | * meta-data into a user's file. | |
2674 | * | |
2675 | * To handle the cancellation of buffer log items, we make two passes | |
2676 | * over the log during recovery. During the first we build a table of | |
2677 | * those buffers which have been cancelled, and during the second we | |
2678 | * only replay those buffers which do not have corresponding cancel | |
34be5ff3 | 2679 | * records in the table. See xlog_recover_buffer_pass[1,2] above |
1da177e4 LT |
2680 | * for more details on the implementation of the table of cancel records. |
2681 | */ | |
2682 | STATIC int | |
c9f71f5f | 2683 | xlog_recover_buffer_pass2( |
9a8d2fdb MT |
2684 | struct xlog *log, |
2685 | struct list_head *buffer_list, | |
50d5c8d8 DC |
2686 | struct xlog_recover_item *item, |
2687 | xfs_lsn_t current_lsn) | |
1da177e4 | 2688 | { |
4e0d5f92 | 2689 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
e2714bf8 | 2690 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2691 | xfs_buf_t *bp; |
2692 | int error; | |
6ad112bf | 2693 | uint buf_flags; |
50d5c8d8 | 2694 | xfs_lsn_t lsn; |
1da177e4 | 2695 | |
c9f71f5f CH |
2696 | /* |
2697 | * In this pass we only want to recover all the buffers which have | |
2698 | * not been cancelled and are not cancellation buffers themselves. | |
2699 | */ | |
2700 | if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno, | |
2701 | buf_f->blf_len, buf_f->blf_flags)) { | |
2702 | trace_xfs_log_recover_buf_cancel(log, buf_f); | |
1da177e4 | 2703 | return 0; |
1da177e4 | 2704 | } |
c9f71f5f | 2705 | |
9abbc539 | 2706 | trace_xfs_log_recover_buf_recover(log, buf_f); |
1da177e4 | 2707 | |
a8acad70 | 2708 | buf_flags = 0; |
611c9946 DC |
2709 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) |
2710 | buf_flags |= XBF_UNMAPPED; | |
6ad112bf | 2711 | |
e2714bf8 | 2712 | bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len, |
c3f8fc73 | 2713 | buf_flags, NULL); |
ac4d6888 | 2714 | if (!bp) |
2451337d | 2715 | return -ENOMEM; |
e5702805 | 2716 | error = bp->b_error; |
5a52c2a5 | 2717 | if (error) { |
901796af | 2718 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)"); |
50d5c8d8 | 2719 | goto out_release; |
1da177e4 LT |
2720 | } |
2721 | ||
50d5c8d8 | 2722 | /* |
67dc288c | 2723 | * Recover the buffer only if we get an LSN from it and it's less than |
50d5c8d8 | 2724 | * the lsn of the transaction we are replaying. |
67dc288c DC |
2725 | * |
2726 | * Note that we have to be extremely careful of readahead here. | |
2727 | * Readahead does not attach verfiers to the buffers so if we don't | |
2728 | * actually do any replay after readahead because of the LSN we found | |
2729 | * in the buffer if more recent than that current transaction then we | |
2730 | * need to attach the verifier directly. Failure to do so can lead to | |
2731 | * future recovery actions (e.g. EFI and unlinked list recovery) can | |
2732 | * operate on the buffers and they won't get the verifier attached. This | |
2733 | * can lead to blocks on disk having the correct content but a stale | |
2734 | * CRC. | |
2735 | * | |
2736 | * It is safe to assume these clean buffers are currently up to date. | |
2737 | * If the buffer is dirtied by a later transaction being replayed, then | |
2738 | * the verifier will be reset to match whatever recover turns that | |
2739 | * buffer into. | |
50d5c8d8 DC |
2740 | */ |
2741 | lsn = xlog_recover_get_buf_lsn(mp, bp); | |
67dc288c DC |
2742 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { |
2743 | xlog_recover_validate_buf_type(mp, bp, buf_f); | |
50d5c8d8 | 2744 | goto out_release; |
67dc288c | 2745 | } |
50d5c8d8 | 2746 | |
e2714bf8 | 2747 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
1da177e4 | 2748 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); |
ad3714b8 DC |
2749 | if (error) |
2750 | goto out_release; | |
e2714bf8 | 2751 | } else if (buf_f->blf_flags & |
c1155410 | 2752 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
ad3714b8 DC |
2753 | bool dirty; |
2754 | ||
2755 | dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); | |
2756 | if (!dirty) | |
2757 | goto out_release; | |
1da177e4 | 2758 | } else { |
9abbc539 | 2759 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
1da177e4 | 2760 | } |
1da177e4 LT |
2761 | |
2762 | /* | |
2763 | * Perform delayed write on the buffer. Asynchronous writes will be | |
2764 | * slower when taking into account all the buffers to be flushed. | |
2765 | * | |
2766 | * Also make sure that only inode buffers with good sizes stay in | |
2767 | * the buffer cache. The kernel moves inodes in buffers of 1 block | |
0f49efd8 | 2768 | * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode |
1da177e4 LT |
2769 | * buffers in the log can be a different size if the log was generated |
2770 | * by an older kernel using unclustered inode buffers or a newer kernel | |
2771 | * running with a different inode cluster size. Regardless, if the | |
0f49efd8 JL |
2772 | * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size) |
2773 | * for *our* value of mp->m_inode_cluster_size, then we need to keep | |
1da177e4 LT |
2774 | * the buffer out of the buffer cache so that the buffer won't |
2775 | * overlap with future reads of those inodes. | |
2776 | */ | |
2777 | if (XFS_DINODE_MAGIC == | |
b53e675d | 2778 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
aa0e8833 | 2779 | (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize, |
0f49efd8 | 2780 | (__uint32_t)log->l_mp->m_inode_cluster_size))) { |
c867cb61 | 2781 | xfs_buf_stale(bp); |
c2b006c1 | 2782 | error = xfs_bwrite(bp); |
1da177e4 | 2783 | } else { |
ebad861b | 2784 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2785 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2786 | xfs_buf_delwri_queue(bp, buffer_list); |
1da177e4 LT |
2787 | } |
2788 | ||
50d5c8d8 | 2789 | out_release: |
c2b006c1 CH |
2790 | xfs_buf_relse(bp); |
2791 | return error; | |
1da177e4 LT |
2792 | } |
2793 | ||
638f4416 DC |
2794 | /* |
2795 | * Inode fork owner changes | |
2796 | * | |
2797 | * If we have been told that we have to reparent the inode fork, it's because an | |
2798 | * extent swap operation on a CRC enabled filesystem has been done and we are | |
2799 | * replaying it. We need to walk the BMBT of the appropriate fork and change the | |
2800 | * owners of it. | |
2801 | * | |
2802 | * The complexity here is that we don't have an inode context to work with, so | |
2803 | * after we've replayed the inode we need to instantiate one. This is where the | |
2804 | * fun begins. | |
2805 | * | |
2806 | * We are in the middle of log recovery, so we can't run transactions. That | |
2807 | * means we cannot use cache coherent inode instantiation via xfs_iget(), as | |
2808 | * that will result in the corresponding iput() running the inode through | |
2809 | * xfs_inactive(). If we've just replayed an inode core that changes the link | |
2810 | * count to zero (i.e. it's been unlinked), then xfs_inactive() will run | |
2811 | * transactions (bad!). | |
2812 | * | |
2813 | * So, to avoid this, we instantiate an inode directly from the inode core we've | |
2814 | * just recovered. We have the buffer still locked, and all we really need to | |
2815 | * instantiate is the inode core and the forks being modified. We can do this | |
2816 | * manually, then run the inode btree owner change, and then tear down the | |
2817 | * xfs_inode without having to run any transactions at all. | |
2818 | * | |
2819 | * Also, because we don't have a transaction context available here but need to | |
2820 | * gather all the buffers we modify for writeback so we pass the buffer_list | |
2821 | * instead for the operation to use. | |
2822 | */ | |
2823 | ||
2824 | STATIC int | |
2825 | xfs_recover_inode_owner_change( | |
2826 | struct xfs_mount *mp, | |
2827 | struct xfs_dinode *dip, | |
2828 | struct xfs_inode_log_format *in_f, | |
2829 | struct list_head *buffer_list) | |
2830 | { | |
2831 | struct xfs_inode *ip; | |
2832 | int error; | |
2833 | ||
2834 | ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)); | |
2835 | ||
2836 | ip = xfs_inode_alloc(mp, in_f->ilf_ino); | |
2837 | if (!ip) | |
2451337d | 2838 | return -ENOMEM; |
638f4416 DC |
2839 | |
2840 | /* instantiate the inode */ | |
2841 | xfs_dinode_from_disk(&ip->i_d, dip); | |
2842 | ASSERT(ip->i_d.di_version >= 3); | |
2843 | ||
2844 | error = xfs_iformat_fork(ip, dip); | |
2845 | if (error) | |
2846 | goto out_free_ip; | |
2847 | ||
2848 | ||
2849 | if (in_f->ilf_fields & XFS_ILOG_DOWNER) { | |
2850 | ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT); | |
2851 | error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK, | |
2852 | ip->i_ino, buffer_list); | |
2853 | if (error) | |
2854 | goto out_free_ip; | |
2855 | } | |
2856 | ||
2857 | if (in_f->ilf_fields & XFS_ILOG_AOWNER) { | |
2858 | ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT); | |
2859 | error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK, | |
2860 | ip->i_ino, buffer_list); | |
2861 | if (error) | |
2862 | goto out_free_ip; | |
2863 | } | |
2864 | ||
2865 | out_free_ip: | |
2866 | xfs_inode_free(ip); | |
2867 | return error; | |
2868 | } | |
2869 | ||
1da177e4 | 2870 | STATIC int |
c9f71f5f | 2871 | xlog_recover_inode_pass2( |
9a8d2fdb MT |
2872 | struct xlog *log, |
2873 | struct list_head *buffer_list, | |
50d5c8d8 DC |
2874 | struct xlog_recover_item *item, |
2875 | xfs_lsn_t current_lsn) | |
1da177e4 LT |
2876 | { |
2877 | xfs_inode_log_format_t *in_f; | |
c9f71f5f | 2878 | xfs_mount_t *mp = log->l_mp; |
1da177e4 | 2879 | xfs_buf_t *bp; |
1da177e4 | 2880 | xfs_dinode_t *dip; |
1da177e4 | 2881 | int len; |
b2a922cd CH |
2882 | char *src; |
2883 | char *dest; | |
1da177e4 LT |
2884 | int error; |
2885 | int attr_index; | |
2886 | uint fields; | |
347d1c01 | 2887 | xfs_icdinode_t *dicp; |
93848a99 | 2888 | uint isize; |
6d192a9b | 2889 | int need_free = 0; |
1da177e4 | 2890 | |
6d192a9b | 2891 | if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) { |
4e0d5f92 | 2892 | in_f = item->ri_buf[0].i_addr; |
6d192a9b | 2893 | } else { |
4e0d5f92 | 2894 | in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP); |
6d192a9b TS |
2895 | need_free = 1; |
2896 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); | |
2897 | if (error) | |
2898 | goto error; | |
2899 | } | |
1da177e4 LT |
2900 | |
2901 | /* | |
2902 | * Inode buffers can be freed, look out for it, | |
2903 | * and do not replay the inode. | |
2904 | */ | |
a1941895 CH |
2905 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
2906 | in_f->ilf_len, 0)) { | |
6d192a9b | 2907 | error = 0; |
9abbc539 | 2908 | trace_xfs_log_recover_inode_cancel(log, in_f); |
6d192a9b TS |
2909 | goto error; |
2910 | } | |
9abbc539 | 2911 | trace_xfs_log_recover_inode_recover(log, in_f); |
1da177e4 | 2912 | |
c3f8fc73 | 2913 | bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0, |
93848a99 | 2914 | &xfs_inode_buf_ops); |
ac4d6888 | 2915 | if (!bp) { |
2451337d | 2916 | error = -ENOMEM; |
ac4d6888 CS |
2917 | goto error; |
2918 | } | |
e5702805 | 2919 | error = bp->b_error; |
5a52c2a5 | 2920 | if (error) { |
901796af | 2921 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)"); |
638f4416 | 2922 | goto out_release; |
1da177e4 | 2923 | } |
1da177e4 | 2924 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); |
88ee2df7 | 2925 | dip = xfs_buf_offset(bp, in_f->ilf_boffset); |
1da177e4 LT |
2926 | |
2927 | /* | |
2928 | * Make sure the place we're flushing out to really looks | |
2929 | * like an inode! | |
2930 | */ | |
69ef921b | 2931 | if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) { |
a0fa2b67 DC |
2932 | xfs_alert(mp, |
2933 | "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld", | |
2934 | __func__, dip, bp, in_f->ilf_ino); | |
c9f71f5f | 2935 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)", |
1da177e4 | 2936 | XFS_ERRLEVEL_LOW, mp); |
2451337d | 2937 | error = -EFSCORRUPTED; |
638f4416 | 2938 | goto out_release; |
1da177e4 | 2939 | } |
4e0d5f92 | 2940 | dicp = item->ri_buf[1].i_addr; |
1da177e4 | 2941 | if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) { |
a0fa2b67 DC |
2942 | xfs_alert(mp, |
2943 | "%s: Bad inode log record, rec ptr 0x%p, ino %Ld", | |
2944 | __func__, item, in_f->ilf_ino); | |
c9f71f5f | 2945 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)", |
1da177e4 | 2946 | XFS_ERRLEVEL_LOW, mp); |
2451337d | 2947 | error = -EFSCORRUPTED; |
638f4416 | 2948 | goto out_release; |
1da177e4 LT |
2949 | } |
2950 | ||
50d5c8d8 DC |
2951 | /* |
2952 | * If the inode has an LSN in it, recover the inode only if it's less | |
638f4416 DC |
2953 | * than the lsn of the transaction we are replaying. Note: we still |
2954 | * need to replay an owner change even though the inode is more recent | |
2955 | * than the transaction as there is no guarantee that all the btree | |
2956 | * blocks are more recent than this transaction, too. | |
50d5c8d8 DC |
2957 | */ |
2958 | if (dip->di_version >= 3) { | |
2959 | xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn); | |
2960 | ||
2961 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { | |
2962 | trace_xfs_log_recover_inode_skip(log, in_f); | |
2963 | error = 0; | |
638f4416 | 2964 | goto out_owner_change; |
50d5c8d8 DC |
2965 | } |
2966 | } | |
2967 | ||
e60896d8 DC |
2968 | /* |
2969 | * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes | |
2970 | * are transactional and if ordering is necessary we can determine that | |
2971 | * more accurately by the LSN field in the V3 inode core. Don't trust | |
2972 | * the inode versions we might be changing them here - use the | |
2973 | * superblock flag to determine whether we need to look at di_flushiter | |
2974 | * to skip replay when the on disk inode is newer than the log one | |
2975 | */ | |
2976 | if (!xfs_sb_version_hascrc(&mp->m_sb) && | |
2977 | dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) { | |
1da177e4 LT |
2978 | /* |
2979 | * Deal with the wrap case, DI_MAX_FLUSH is less | |
2980 | * than smaller numbers | |
2981 | */ | |
81591fe2 | 2982 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
347d1c01 | 2983 | dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
1da177e4 LT |
2984 | /* do nothing */ |
2985 | } else { | |
9abbc539 | 2986 | trace_xfs_log_recover_inode_skip(log, in_f); |
6d192a9b | 2987 | error = 0; |
638f4416 | 2988 | goto out_release; |
1da177e4 LT |
2989 | } |
2990 | } | |
e60896d8 | 2991 | |
1da177e4 LT |
2992 | /* Take the opportunity to reset the flush iteration count */ |
2993 | dicp->di_flushiter = 0; | |
2994 | ||
abbede1b | 2995 | if (unlikely(S_ISREG(dicp->di_mode))) { |
1da177e4 LT |
2996 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
2997 | (dicp->di_format != XFS_DINODE_FMT_BTREE)) { | |
c9f71f5f | 2998 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)", |
1da177e4 | 2999 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
3000 | xfs_alert(mp, |
3001 | "%s: Bad regular inode log record, rec ptr 0x%p, " | |
3002 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
3003 | __func__, item, dip, bp, in_f->ilf_ino); | |
2451337d | 3004 | error = -EFSCORRUPTED; |
638f4416 | 3005 | goto out_release; |
1da177e4 | 3006 | } |
abbede1b | 3007 | } else if (unlikely(S_ISDIR(dicp->di_mode))) { |
1da177e4 LT |
3008 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
3009 | (dicp->di_format != XFS_DINODE_FMT_BTREE) && | |
3010 | (dicp->di_format != XFS_DINODE_FMT_LOCAL)) { | |
c9f71f5f | 3011 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)", |
1da177e4 | 3012 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
3013 | xfs_alert(mp, |
3014 | "%s: Bad dir inode log record, rec ptr 0x%p, " | |
3015 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
3016 | __func__, item, dip, bp, in_f->ilf_ino); | |
2451337d | 3017 | error = -EFSCORRUPTED; |
638f4416 | 3018 | goto out_release; |
1da177e4 LT |
3019 | } |
3020 | } | |
3021 | if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){ | |
c9f71f5f | 3022 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)", |
1da177e4 | 3023 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
3024 | xfs_alert(mp, |
3025 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
3026 | "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld", | |
3027 | __func__, item, dip, bp, in_f->ilf_ino, | |
1da177e4 LT |
3028 | dicp->di_nextents + dicp->di_anextents, |
3029 | dicp->di_nblocks); | |
2451337d | 3030 | error = -EFSCORRUPTED; |
638f4416 | 3031 | goto out_release; |
1da177e4 LT |
3032 | } |
3033 | if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) { | |
c9f71f5f | 3034 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)", |
1da177e4 | 3035 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
3036 | xfs_alert(mp, |
3037 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
3038 | "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__, | |
c9f71f5f | 3039 | item, dip, bp, in_f->ilf_ino, dicp->di_forkoff); |
2451337d | 3040 | error = -EFSCORRUPTED; |
638f4416 | 3041 | goto out_release; |
1da177e4 | 3042 | } |
93848a99 CH |
3043 | isize = xfs_icdinode_size(dicp->di_version); |
3044 | if (unlikely(item->ri_buf[1].i_len > isize)) { | |
c9f71f5f | 3045 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)", |
1da177e4 | 3046 | XFS_ERRLEVEL_LOW, mp, dicp); |
a0fa2b67 DC |
3047 | xfs_alert(mp, |
3048 | "%s: Bad inode log record length %d, rec ptr 0x%p", | |
3049 | __func__, item->ri_buf[1].i_len, item); | |
2451337d | 3050 | error = -EFSCORRUPTED; |
638f4416 | 3051 | goto out_release; |
1da177e4 LT |
3052 | } |
3053 | ||
3054 | /* The core is in in-core format */ | |
93848a99 | 3055 | xfs_dinode_to_disk(dip, dicp); |
1da177e4 LT |
3056 | |
3057 | /* the rest is in on-disk format */ | |
93848a99 CH |
3058 | if (item->ri_buf[1].i_len > isize) { |
3059 | memcpy((char *)dip + isize, | |
3060 | item->ri_buf[1].i_addr + isize, | |
3061 | item->ri_buf[1].i_len - isize); | |
1da177e4 LT |
3062 | } |
3063 | ||
3064 | fields = in_f->ilf_fields; | |
3065 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { | |
3066 | case XFS_ILOG_DEV: | |
81591fe2 | 3067 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
1da177e4 LT |
3068 | break; |
3069 | case XFS_ILOG_UUID: | |
81591fe2 CH |
3070 | memcpy(XFS_DFORK_DPTR(dip), |
3071 | &in_f->ilf_u.ilfu_uuid, | |
3072 | sizeof(uuid_t)); | |
1da177e4 LT |
3073 | break; |
3074 | } | |
3075 | ||
3076 | if (in_f->ilf_size == 2) | |
638f4416 | 3077 | goto out_owner_change; |
1da177e4 LT |
3078 | len = item->ri_buf[2].i_len; |
3079 | src = item->ri_buf[2].i_addr; | |
3080 | ASSERT(in_f->ilf_size <= 4); | |
3081 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); | |
3082 | ASSERT(!(fields & XFS_ILOG_DFORK) || | |
3083 | (len == in_f->ilf_dsize)); | |
3084 | ||
3085 | switch (fields & XFS_ILOG_DFORK) { | |
3086 | case XFS_ILOG_DDATA: | |
3087 | case XFS_ILOG_DEXT: | |
81591fe2 | 3088 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
1da177e4 LT |
3089 | break; |
3090 | ||
3091 | case XFS_ILOG_DBROOT: | |
7cc95a82 | 3092 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
81591fe2 | 3093 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
1da177e4 LT |
3094 | XFS_DFORK_DSIZE(dip, mp)); |
3095 | break; | |
3096 | ||
3097 | default: | |
3098 | /* | |
3099 | * There are no data fork flags set. | |
3100 | */ | |
3101 | ASSERT((fields & XFS_ILOG_DFORK) == 0); | |
3102 | break; | |
3103 | } | |
3104 | ||
3105 | /* | |
3106 | * If we logged any attribute data, recover it. There may or | |
3107 | * may not have been any other non-core data logged in this | |
3108 | * transaction. | |
3109 | */ | |
3110 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
3111 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { | |
3112 | attr_index = 3; | |
3113 | } else { | |
3114 | attr_index = 2; | |
3115 | } | |
3116 | len = item->ri_buf[attr_index].i_len; | |
3117 | src = item->ri_buf[attr_index].i_addr; | |
3118 | ASSERT(len == in_f->ilf_asize); | |
3119 | ||
3120 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
3121 | case XFS_ILOG_ADATA: | |
3122 | case XFS_ILOG_AEXT: | |
3123 | dest = XFS_DFORK_APTR(dip); | |
3124 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); | |
3125 | memcpy(dest, src, len); | |
3126 | break; | |
3127 | ||
3128 | case XFS_ILOG_ABROOT: | |
3129 | dest = XFS_DFORK_APTR(dip); | |
7cc95a82 CH |
3130 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
3131 | len, (xfs_bmdr_block_t*)dest, | |
1da177e4 LT |
3132 | XFS_DFORK_ASIZE(dip, mp)); |
3133 | break; | |
3134 | ||
3135 | default: | |
a0fa2b67 | 3136 | xfs_warn(log->l_mp, "%s: Invalid flag", __func__); |
1da177e4 | 3137 | ASSERT(0); |
2451337d | 3138 | error = -EIO; |
638f4416 | 3139 | goto out_release; |
1da177e4 LT |
3140 | } |
3141 | } | |
3142 | ||
638f4416 DC |
3143 | out_owner_change: |
3144 | if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) | |
3145 | error = xfs_recover_inode_owner_change(mp, dip, in_f, | |
3146 | buffer_list); | |
93848a99 CH |
3147 | /* re-generate the checksum. */ |
3148 | xfs_dinode_calc_crc(log->l_mp, dip); | |
3149 | ||
ebad861b | 3150 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 3151 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 3152 | xfs_buf_delwri_queue(bp, buffer_list); |
50d5c8d8 DC |
3153 | |
3154 | out_release: | |
61551f1e | 3155 | xfs_buf_relse(bp); |
6d192a9b TS |
3156 | error: |
3157 | if (need_free) | |
f0e2d93c | 3158 | kmem_free(in_f); |
b474c7ae | 3159 | return error; |
1da177e4 LT |
3160 | } |
3161 | ||
3162 | /* | |
9a8d2fdb | 3163 | * Recover QUOTAOFF records. We simply make a note of it in the xlog |
1da177e4 LT |
3164 | * structure, so that we know not to do any dquot item or dquot buffer recovery, |
3165 | * of that type. | |
3166 | */ | |
3167 | STATIC int | |
c9f71f5f | 3168 | xlog_recover_quotaoff_pass1( |
9a8d2fdb MT |
3169 | struct xlog *log, |
3170 | struct xlog_recover_item *item) | |
1da177e4 | 3171 | { |
c9f71f5f | 3172 | xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr; |
1da177e4 LT |
3173 | ASSERT(qoff_f); |
3174 | ||
3175 | /* | |
3176 | * The logitem format's flag tells us if this was user quotaoff, | |
77a7cce4 | 3177 | * group/project quotaoff or both. |
1da177e4 LT |
3178 | */ |
3179 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) | |
3180 | log->l_quotaoffs_flag |= XFS_DQ_USER; | |
77a7cce4 NS |
3181 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
3182 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; | |
1da177e4 LT |
3183 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
3184 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; | |
3185 | ||
d99831ff | 3186 | return 0; |
1da177e4 LT |
3187 | } |
3188 | ||
3189 | /* | |
3190 | * Recover a dquot record | |
3191 | */ | |
3192 | STATIC int | |
c9f71f5f | 3193 | xlog_recover_dquot_pass2( |
9a8d2fdb MT |
3194 | struct xlog *log, |
3195 | struct list_head *buffer_list, | |
50d5c8d8 DC |
3196 | struct xlog_recover_item *item, |
3197 | xfs_lsn_t current_lsn) | |
1da177e4 | 3198 | { |
c9f71f5f | 3199 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
3200 | xfs_buf_t *bp; |
3201 | struct xfs_disk_dquot *ddq, *recddq; | |
3202 | int error; | |
3203 | xfs_dq_logformat_t *dq_f; | |
3204 | uint type; | |
3205 | ||
1da177e4 LT |
3206 | |
3207 | /* | |
3208 | * Filesystems are required to send in quota flags at mount time. | |
3209 | */ | |
3210 | if (mp->m_qflags == 0) | |
d99831ff | 3211 | return 0; |
1da177e4 | 3212 | |
4e0d5f92 CH |
3213 | recddq = item->ri_buf[1].i_addr; |
3214 | if (recddq == NULL) { | |
a0fa2b67 | 3215 | xfs_alert(log->l_mp, "NULL dquot in %s.", __func__); |
2451337d | 3216 | return -EIO; |
0c5e1ce8 | 3217 | } |
8ec6dba2 | 3218 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
a0fa2b67 | 3219 | xfs_alert(log->l_mp, "dquot too small (%d) in %s.", |
0c5e1ce8 | 3220 | item->ri_buf[1].i_len, __func__); |
2451337d | 3221 | return -EIO; |
0c5e1ce8 CH |
3222 | } |
3223 | ||
1da177e4 LT |
3224 | /* |
3225 | * This type of quotas was turned off, so ignore this record. | |
3226 | */ | |
b53e675d | 3227 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
1da177e4 LT |
3228 | ASSERT(type); |
3229 | if (log->l_quotaoffs_flag & type) | |
d99831ff | 3230 | return 0; |
1da177e4 LT |
3231 | |
3232 | /* | |
3233 | * At this point we know that quota was _not_ turned off. | |
3234 | * Since the mount flags are not indicating to us otherwise, this | |
3235 | * must mean that quota is on, and the dquot needs to be replayed. | |
3236 | * Remember that we may not have fully recovered the superblock yet, | |
3237 | * so we can't do the usual trick of looking at the SB quota bits. | |
3238 | * | |
3239 | * The other possibility, of course, is that the quota subsystem was | |
3240 | * removed since the last mount - ENOSYS. | |
3241 | */ | |
4e0d5f92 | 3242 | dq_f = item->ri_buf[0].i_addr; |
1da177e4 | 3243 | ASSERT(dq_f); |
9aede1d8 | 3244 | error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, |
a0fa2b67 DC |
3245 | "xlog_recover_dquot_pass2 (log copy)"); |
3246 | if (error) | |
2451337d | 3247 | return -EIO; |
1da177e4 LT |
3248 | ASSERT(dq_f->qlf_len == 1); |
3249 | ||
ad3714b8 DC |
3250 | /* |
3251 | * At this point we are assuming that the dquots have been allocated | |
3252 | * and hence the buffer has valid dquots stamped in it. It should, | |
3253 | * therefore, pass verifier validation. If the dquot is bad, then the | |
3254 | * we'll return an error here, so we don't need to specifically check | |
3255 | * the dquot in the buffer after the verifier has run. | |
3256 | */ | |
7ca790a5 | 3257 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno, |
c3f8fc73 | 3258 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp, |
ad3714b8 | 3259 | &xfs_dquot_buf_ops); |
7ca790a5 | 3260 | if (error) |
1da177e4 | 3261 | return error; |
7ca790a5 | 3262 | |
1da177e4 | 3263 | ASSERT(bp); |
88ee2df7 | 3264 | ddq = xfs_buf_offset(bp, dq_f->qlf_boffset); |
1da177e4 | 3265 | |
50d5c8d8 DC |
3266 | /* |
3267 | * If the dquot has an LSN in it, recover the dquot only if it's less | |
3268 | * than the lsn of the transaction we are replaying. | |
3269 | */ | |
3270 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
3271 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq; | |
3272 | xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn); | |
3273 | ||
3274 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { | |
3275 | goto out_release; | |
3276 | } | |
3277 | } | |
3278 | ||
1da177e4 | 3279 | memcpy(ddq, recddq, item->ri_buf[1].i_len); |
6fcdc59d DC |
3280 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
3281 | xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk), | |
3282 | XFS_DQUOT_CRC_OFF); | |
3283 | } | |
1da177e4 LT |
3284 | |
3285 | ASSERT(dq_f->qlf_size == 2); | |
ebad861b | 3286 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 3287 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 3288 | xfs_buf_delwri_queue(bp, buffer_list); |
1da177e4 | 3289 | |
50d5c8d8 DC |
3290 | out_release: |
3291 | xfs_buf_relse(bp); | |
3292 | return 0; | |
1da177e4 LT |
3293 | } |
3294 | ||
3295 | /* | |
3296 | * This routine is called to create an in-core extent free intent | |
3297 | * item from the efi format structure which was logged on disk. | |
3298 | * It allocates an in-core efi, copies the extents from the format | |
3299 | * structure into it, and adds the efi to the AIL with the given | |
3300 | * LSN. | |
3301 | */ | |
6d192a9b | 3302 | STATIC int |
c9f71f5f | 3303 | xlog_recover_efi_pass2( |
9a8d2fdb MT |
3304 | struct xlog *log, |
3305 | struct xlog_recover_item *item, | |
3306 | xfs_lsn_t lsn) | |
1da177e4 | 3307 | { |
e32a1d1f BF |
3308 | int error; |
3309 | struct xfs_mount *mp = log->l_mp; | |
3310 | struct xfs_efi_log_item *efip; | |
3311 | struct xfs_efi_log_format *efi_formatp; | |
1da177e4 | 3312 | |
4e0d5f92 | 3313 | efi_formatp = item->ri_buf[0].i_addr; |
1da177e4 | 3314 | |
1da177e4 | 3315 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); |
e32a1d1f BF |
3316 | error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format); |
3317 | if (error) { | |
6d192a9b TS |
3318 | xfs_efi_item_free(efip); |
3319 | return error; | |
3320 | } | |
b199c8a4 | 3321 | atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents); |
1da177e4 | 3322 | |
a9c21c1b | 3323 | spin_lock(&log->l_ailp->xa_lock); |
1da177e4 | 3324 | /* |
e32a1d1f BF |
3325 | * The EFI has two references. One for the EFD and one for EFI to ensure |
3326 | * it makes it into the AIL. Insert the EFI into the AIL directly and | |
3327 | * drop the EFI reference. Note that xfs_trans_ail_update() drops the | |
3328 | * AIL lock. | |
1da177e4 | 3329 | */ |
e6059949 | 3330 | xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn); |
e32a1d1f | 3331 | xfs_efi_release(efip); |
6d192a9b | 3332 | return 0; |
1da177e4 LT |
3333 | } |
3334 | ||
3335 | ||
3336 | /* | |
e32a1d1f BF |
3337 | * This routine is called when an EFD format structure is found in a committed |
3338 | * transaction in the log. Its purpose is to cancel the corresponding EFI if it | |
3339 | * was still in the log. To do this it searches the AIL for the EFI with an id | |
3340 | * equal to that in the EFD format structure. If we find it we drop the EFD | |
3341 | * reference, which removes the EFI from the AIL and frees it. | |
1da177e4 | 3342 | */ |
c9f71f5f CH |
3343 | STATIC int |
3344 | xlog_recover_efd_pass2( | |
9a8d2fdb MT |
3345 | struct xlog *log, |
3346 | struct xlog_recover_item *item) | |
1da177e4 | 3347 | { |
1da177e4 LT |
3348 | xfs_efd_log_format_t *efd_formatp; |
3349 | xfs_efi_log_item_t *efip = NULL; | |
3350 | xfs_log_item_t *lip; | |
1da177e4 | 3351 | __uint64_t efi_id; |
27d8d5fe | 3352 | struct xfs_ail_cursor cur; |
783a2f65 | 3353 | struct xfs_ail *ailp = log->l_ailp; |
1da177e4 | 3354 | |
4e0d5f92 | 3355 | efd_formatp = item->ri_buf[0].i_addr; |
6d192a9b TS |
3356 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
3357 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || | |
3358 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + | |
3359 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); | |
1da177e4 LT |
3360 | efi_id = efd_formatp->efd_efi_id; |
3361 | ||
3362 | /* | |
e32a1d1f BF |
3363 | * Search for the EFI with the id in the EFD format structure in the |
3364 | * AIL. | |
1da177e4 | 3365 | */ |
a9c21c1b DC |
3366 | spin_lock(&ailp->xa_lock); |
3367 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
3368 | while (lip != NULL) { |
3369 | if (lip->li_type == XFS_LI_EFI) { | |
3370 | efip = (xfs_efi_log_item_t *)lip; | |
3371 | if (efip->efi_format.efi_id == efi_id) { | |
3372 | /* | |
e32a1d1f BF |
3373 | * Drop the EFD reference to the EFI. This |
3374 | * removes the EFI from the AIL and frees it. | |
1da177e4 | 3375 | */ |
e32a1d1f BF |
3376 | spin_unlock(&ailp->xa_lock); |
3377 | xfs_efi_release(efip); | |
a9c21c1b | 3378 | spin_lock(&ailp->xa_lock); |
27d8d5fe | 3379 | break; |
1da177e4 LT |
3380 | } |
3381 | } | |
a9c21c1b | 3382 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 3383 | } |
e32a1d1f | 3384 | |
e4a1e29c | 3385 | xfs_trans_ail_cursor_done(&cur); |
a9c21c1b | 3386 | spin_unlock(&ailp->xa_lock); |
c9f71f5f CH |
3387 | |
3388 | return 0; | |
1da177e4 LT |
3389 | } |
3390 | ||
28c8e41a DC |
3391 | /* |
3392 | * This routine is called when an inode create format structure is found in a | |
3393 | * committed transaction in the log. It's purpose is to initialise the inodes | |
3394 | * being allocated on disk. This requires us to get inode cluster buffers that | |
3395 | * match the range to be intialised, stamped with inode templates and written | |
3396 | * by delayed write so that subsequent modifications will hit the cached buffer | |
3397 | * and only need writing out at the end of recovery. | |
3398 | */ | |
3399 | STATIC int | |
3400 | xlog_recover_do_icreate_pass2( | |
3401 | struct xlog *log, | |
3402 | struct list_head *buffer_list, | |
3403 | xlog_recover_item_t *item) | |
3404 | { | |
3405 | struct xfs_mount *mp = log->l_mp; | |
3406 | struct xfs_icreate_log *icl; | |
3407 | xfs_agnumber_t agno; | |
3408 | xfs_agblock_t agbno; | |
3409 | unsigned int count; | |
3410 | unsigned int isize; | |
3411 | xfs_agblock_t length; | |
fc0d1656 BF |
3412 | int blks_per_cluster; |
3413 | int bb_per_cluster; | |
3414 | int cancel_count; | |
3415 | int nbufs; | |
3416 | int i; | |
28c8e41a DC |
3417 | |
3418 | icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr; | |
3419 | if (icl->icl_type != XFS_LI_ICREATE) { | |
3420 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type"); | |
2451337d | 3421 | return -EINVAL; |
28c8e41a DC |
3422 | } |
3423 | ||
3424 | if (icl->icl_size != 1) { | |
3425 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size"); | |
2451337d | 3426 | return -EINVAL; |
28c8e41a DC |
3427 | } |
3428 | ||
3429 | agno = be32_to_cpu(icl->icl_ag); | |
3430 | if (agno >= mp->m_sb.sb_agcount) { | |
3431 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno"); | |
2451337d | 3432 | return -EINVAL; |
28c8e41a DC |
3433 | } |
3434 | agbno = be32_to_cpu(icl->icl_agbno); | |
3435 | if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) { | |
3436 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno"); | |
2451337d | 3437 | return -EINVAL; |
28c8e41a DC |
3438 | } |
3439 | isize = be32_to_cpu(icl->icl_isize); | |
3440 | if (isize != mp->m_sb.sb_inodesize) { | |
3441 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize"); | |
2451337d | 3442 | return -EINVAL; |
28c8e41a DC |
3443 | } |
3444 | count = be32_to_cpu(icl->icl_count); | |
3445 | if (!count) { | |
3446 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count"); | |
2451337d | 3447 | return -EINVAL; |
28c8e41a DC |
3448 | } |
3449 | length = be32_to_cpu(icl->icl_length); | |
3450 | if (!length || length >= mp->m_sb.sb_agblocks) { | |
3451 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length"); | |
2451337d | 3452 | return -EINVAL; |
28c8e41a DC |
3453 | } |
3454 | ||
7f43c907 BF |
3455 | /* |
3456 | * The inode chunk is either full or sparse and we only support | |
3457 | * m_ialloc_min_blks sized sparse allocations at this time. | |
3458 | */ | |
3459 | if (length != mp->m_ialloc_blks && | |
3460 | length != mp->m_ialloc_min_blks) { | |
3461 | xfs_warn(log->l_mp, | |
3462 | "%s: unsupported chunk length", __FUNCTION__); | |
3463 | return -EINVAL; | |
3464 | } | |
3465 | ||
3466 | /* verify inode count is consistent with extent length */ | |
3467 | if ((count >> mp->m_sb.sb_inopblog) != length) { | |
3468 | xfs_warn(log->l_mp, | |
3469 | "%s: inconsistent inode count and chunk length", | |
3470 | __FUNCTION__); | |
2451337d | 3471 | return -EINVAL; |
28c8e41a DC |
3472 | } |
3473 | ||
3474 | /* | |
fc0d1656 BF |
3475 | * The icreate transaction can cover multiple cluster buffers and these |
3476 | * buffers could have been freed and reused. Check the individual | |
3477 | * buffers for cancellation so we don't overwrite anything written after | |
3478 | * a cancellation. | |
3479 | */ | |
3480 | blks_per_cluster = xfs_icluster_size_fsb(mp); | |
3481 | bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
3482 | nbufs = length / blks_per_cluster; | |
3483 | for (i = 0, cancel_count = 0; i < nbufs; i++) { | |
3484 | xfs_daddr_t daddr; | |
3485 | ||
3486 | daddr = XFS_AGB_TO_DADDR(mp, agno, | |
3487 | agbno + i * blks_per_cluster); | |
3488 | if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0)) | |
3489 | cancel_count++; | |
3490 | } | |
3491 | ||
3492 | /* | |
3493 | * We currently only use icreate for a single allocation at a time. This | |
3494 | * means we should expect either all or none of the buffers to be | |
3495 | * cancelled. Be conservative and skip replay if at least one buffer is | |
3496 | * cancelled, but warn the user that something is awry if the buffers | |
3497 | * are not consistent. | |
28c8e41a | 3498 | * |
fc0d1656 BF |
3499 | * XXX: This must be refined to only skip cancelled clusters once we use |
3500 | * icreate for multiple chunk allocations. | |
28c8e41a | 3501 | */ |
fc0d1656 BF |
3502 | ASSERT(!cancel_count || cancel_count == nbufs); |
3503 | if (cancel_count) { | |
3504 | if (cancel_count != nbufs) | |
3505 | xfs_warn(mp, | |
3506 | "WARNING: partial inode chunk cancellation, skipped icreate."); | |
78d57e45 | 3507 | trace_xfs_log_recover_icreate_cancel(log, icl); |
28c8e41a | 3508 | return 0; |
78d57e45 | 3509 | } |
28c8e41a | 3510 | |
78d57e45 | 3511 | trace_xfs_log_recover_icreate_recover(log, icl); |
fc0d1656 BF |
3512 | return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno, |
3513 | length, be32_to_cpu(icl->icl_gen)); | |
28c8e41a DC |
3514 | } |
3515 | ||
00574da1 ZYW |
3516 | STATIC void |
3517 | xlog_recover_buffer_ra_pass2( | |
3518 | struct xlog *log, | |
3519 | struct xlog_recover_item *item) | |
3520 | { | |
3521 | struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr; | |
3522 | struct xfs_mount *mp = log->l_mp; | |
3523 | ||
84a5b730 | 3524 | if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno, |
00574da1 ZYW |
3525 | buf_f->blf_len, buf_f->blf_flags)) { |
3526 | return; | |
3527 | } | |
3528 | ||
3529 | xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno, | |
3530 | buf_f->blf_len, NULL); | |
3531 | } | |
3532 | ||
3533 | STATIC void | |
3534 | xlog_recover_inode_ra_pass2( | |
3535 | struct xlog *log, | |
3536 | struct xlog_recover_item *item) | |
3537 | { | |
3538 | struct xfs_inode_log_format ilf_buf; | |
3539 | struct xfs_inode_log_format *ilfp; | |
3540 | struct xfs_mount *mp = log->l_mp; | |
3541 | int error; | |
3542 | ||
3543 | if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) { | |
3544 | ilfp = item->ri_buf[0].i_addr; | |
3545 | } else { | |
3546 | ilfp = &ilf_buf; | |
3547 | memset(ilfp, 0, sizeof(*ilfp)); | |
3548 | error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp); | |
3549 | if (error) | |
3550 | return; | |
3551 | } | |
3552 | ||
84a5b730 | 3553 | if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0)) |
00574da1 ZYW |
3554 | return; |
3555 | ||
3556 | xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno, | |
d8914002 | 3557 | ilfp->ilf_len, &xfs_inode_buf_ra_ops); |
00574da1 ZYW |
3558 | } |
3559 | ||
3560 | STATIC void | |
3561 | xlog_recover_dquot_ra_pass2( | |
3562 | struct xlog *log, | |
3563 | struct xlog_recover_item *item) | |
3564 | { | |
3565 | struct xfs_mount *mp = log->l_mp; | |
3566 | struct xfs_disk_dquot *recddq; | |
3567 | struct xfs_dq_logformat *dq_f; | |
3568 | uint type; | |
7d6a13f0 | 3569 | int len; |
00574da1 ZYW |
3570 | |
3571 | ||
3572 | if (mp->m_qflags == 0) | |
3573 | return; | |
3574 | ||
3575 | recddq = item->ri_buf[1].i_addr; | |
3576 | if (recddq == NULL) | |
3577 | return; | |
3578 | if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot)) | |
3579 | return; | |
3580 | ||
3581 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); | |
3582 | ASSERT(type); | |
3583 | if (log->l_quotaoffs_flag & type) | |
3584 | return; | |
3585 | ||
3586 | dq_f = item->ri_buf[0].i_addr; | |
3587 | ASSERT(dq_f); | |
3588 | ASSERT(dq_f->qlf_len == 1); | |
3589 | ||
7d6a13f0 DC |
3590 | len = XFS_FSB_TO_BB(mp, dq_f->qlf_len); |
3591 | if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0)) | |
3592 | return; | |
3593 | ||
3594 | xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len, | |
3595 | &xfs_dquot_buf_ra_ops); | |
00574da1 ZYW |
3596 | } |
3597 | ||
3598 | STATIC void | |
3599 | xlog_recover_ra_pass2( | |
3600 | struct xlog *log, | |
3601 | struct xlog_recover_item *item) | |
3602 | { | |
3603 | switch (ITEM_TYPE(item)) { | |
3604 | case XFS_LI_BUF: | |
3605 | xlog_recover_buffer_ra_pass2(log, item); | |
3606 | break; | |
3607 | case XFS_LI_INODE: | |
3608 | xlog_recover_inode_ra_pass2(log, item); | |
3609 | break; | |
3610 | case XFS_LI_DQUOT: | |
3611 | xlog_recover_dquot_ra_pass2(log, item); | |
3612 | break; | |
3613 | case XFS_LI_EFI: | |
3614 | case XFS_LI_EFD: | |
3615 | case XFS_LI_QUOTAOFF: | |
3616 | default: | |
3617 | break; | |
3618 | } | |
3619 | } | |
3620 | ||
d0450948 | 3621 | STATIC int |
c9f71f5f | 3622 | xlog_recover_commit_pass1( |
ad223e60 MT |
3623 | struct xlog *log, |
3624 | struct xlog_recover *trans, | |
3625 | struct xlog_recover_item *item) | |
d0450948 | 3626 | { |
c9f71f5f | 3627 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1); |
d0450948 CH |
3628 | |
3629 | switch (ITEM_TYPE(item)) { | |
3630 | case XFS_LI_BUF: | |
c9f71f5f CH |
3631 | return xlog_recover_buffer_pass1(log, item); |
3632 | case XFS_LI_QUOTAOFF: | |
3633 | return xlog_recover_quotaoff_pass1(log, item); | |
d0450948 | 3634 | case XFS_LI_INODE: |
d0450948 | 3635 | case XFS_LI_EFI: |
d0450948 | 3636 | case XFS_LI_EFD: |
c9f71f5f | 3637 | case XFS_LI_DQUOT: |
28c8e41a | 3638 | case XFS_LI_ICREATE: |
c9f71f5f | 3639 | /* nothing to do in pass 1 */ |
d0450948 | 3640 | return 0; |
c9f71f5f | 3641 | default: |
a0fa2b67 DC |
3642 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
3643 | __func__, ITEM_TYPE(item)); | |
c9f71f5f | 3644 | ASSERT(0); |
2451337d | 3645 | return -EIO; |
c9f71f5f CH |
3646 | } |
3647 | } | |
3648 | ||
3649 | STATIC int | |
3650 | xlog_recover_commit_pass2( | |
ad223e60 MT |
3651 | struct xlog *log, |
3652 | struct xlog_recover *trans, | |
3653 | struct list_head *buffer_list, | |
3654 | struct xlog_recover_item *item) | |
c9f71f5f CH |
3655 | { |
3656 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2); | |
3657 | ||
3658 | switch (ITEM_TYPE(item)) { | |
3659 | case XFS_LI_BUF: | |
50d5c8d8 DC |
3660 | return xlog_recover_buffer_pass2(log, buffer_list, item, |
3661 | trans->r_lsn); | |
c9f71f5f | 3662 | case XFS_LI_INODE: |
50d5c8d8 DC |
3663 | return xlog_recover_inode_pass2(log, buffer_list, item, |
3664 | trans->r_lsn); | |
c9f71f5f CH |
3665 | case XFS_LI_EFI: |
3666 | return xlog_recover_efi_pass2(log, item, trans->r_lsn); | |
3667 | case XFS_LI_EFD: | |
3668 | return xlog_recover_efd_pass2(log, item); | |
d0450948 | 3669 | case XFS_LI_DQUOT: |
50d5c8d8 DC |
3670 | return xlog_recover_dquot_pass2(log, buffer_list, item, |
3671 | trans->r_lsn); | |
28c8e41a DC |
3672 | case XFS_LI_ICREATE: |
3673 | return xlog_recover_do_icreate_pass2(log, buffer_list, item); | |
d0450948 | 3674 | case XFS_LI_QUOTAOFF: |
c9f71f5f CH |
3675 | /* nothing to do in pass2 */ |
3676 | return 0; | |
d0450948 | 3677 | default: |
a0fa2b67 DC |
3678 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
3679 | __func__, ITEM_TYPE(item)); | |
d0450948 | 3680 | ASSERT(0); |
2451337d | 3681 | return -EIO; |
d0450948 CH |
3682 | } |
3683 | } | |
3684 | ||
00574da1 ZYW |
3685 | STATIC int |
3686 | xlog_recover_items_pass2( | |
3687 | struct xlog *log, | |
3688 | struct xlog_recover *trans, | |
3689 | struct list_head *buffer_list, | |
3690 | struct list_head *item_list) | |
3691 | { | |
3692 | struct xlog_recover_item *item; | |
3693 | int error = 0; | |
3694 | ||
3695 | list_for_each_entry(item, item_list, ri_list) { | |
3696 | error = xlog_recover_commit_pass2(log, trans, | |
3697 | buffer_list, item); | |
3698 | if (error) | |
3699 | return error; | |
3700 | } | |
3701 | ||
3702 | return error; | |
3703 | } | |
3704 | ||
d0450948 CH |
3705 | /* |
3706 | * Perform the transaction. | |
3707 | * | |
3708 | * If the transaction modifies a buffer or inode, do it now. Otherwise, | |
3709 | * EFIs and EFDs get queued up by adding entries into the AIL for them. | |
3710 | */ | |
1da177e4 LT |
3711 | STATIC int |
3712 | xlog_recover_commit_trans( | |
ad223e60 | 3713 | struct xlog *log, |
d0450948 | 3714 | struct xlog_recover *trans, |
1da177e4 LT |
3715 | int pass) |
3716 | { | |
00574da1 ZYW |
3717 | int error = 0; |
3718 | int error2; | |
3719 | int items_queued = 0; | |
3720 | struct xlog_recover_item *item; | |
3721 | struct xlog_recover_item *next; | |
3722 | LIST_HEAD (buffer_list); | |
3723 | LIST_HEAD (ra_list); | |
3724 | LIST_HEAD (done_list); | |
3725 | ||
3726 | #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100 | |
1da177e4 | 3727 | |
f0a76953 | 3728 | hlist_del(&trans->r_list); |
d0450948 CH |
3729 | |
3730 | error = xlog_recover_reorder_trans(log, trans, pass); | |
3731 | if (error) | |
1da177e4 | 3732 | return error; |
d0450948 | 3733 | |
00574da1 | 3734 | list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) { |
43ff2122 CH |
3735 | switch (pass) { |
3736 | case XLOG_RECOVER_PASS1: | |
c9f71f5f | 3737 | error = xlog_recover_commit_pass1(log, trans, item); |
43ff2122 CH |
3738 | break; |
3739 | case XLOG_RECOVER_PASS2: | |
00574da1 ZYW |
3740 | xlog_recover_ra_pass2(log, item); |
3741 | list_move_tail(&item->ri_list, &ra_list); | |
3742 | items_queued++; | |
3743 | if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) { | |
3744 | error = xlog_recover_items_pass2(log, trans, | |
3745 | &buffer_list, &ra_list); | |
3746 | list_splice_tail_init(&ra_list, &done_list); | |
3747 | items_queued = 0; | |
3748 | } | |
3749 | ||
43ff2122 CH |
3750 | break; |
3751 | default: | |
3752 | ASSERT(0); | |
3753 | } | |
3754 | ||
d0450948 | 3755 | if (error) |
43ff2122 | 3756 | goto out; |
d0450948 CH |
3757 | } |
3758 | ||
00574da1 ZYW |
3759 | out: |
3760 | if (!list_empty(&ra_list)) { | |
3761 | if (!error) | |
3762 | error = xlog_recover_items_pass2(log, trans, | |
3763 | &buffer_list, &ra_list); | |
3764 | list_splice_tail_init(&ra_list, &done_list); | |
3765 | } | |
3766 | ||
3767 | if (!list_empty(&done_list)) | |
3768 | list_splice_init(&done_list, &trans->r_itemq); | |
3769 | ||
43ff2122 CH |
3770 | error2 = xfs_buf_delwri_submit(&buffer_list); |
3771 | return error ? error : error2; | |
1da177e4 LT |
3772 | } |
3773 | ||
76560669 DC |
3774 | STATIC void |
3775 | xlog_recover_add_item( | |
3776 | struct list_head *head) | |
3777 | { | |
3778 | xlog_recover_item_t *item; | |
3779 | ||
3780 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); | |
3781 | INIT_LIST_HEAD(&item->ri_list); | |
3782 | list_add_tail(&item->ri_list, head); | |
3783 | } | |
3784 | ||
1da177e4 | 3785 | STATIC int |
76560669 DC |
3786 | xlog_recover_add_to_cont_trans( |
3787 | struct xlog *log, | |
3788 | struct xlog_recover *trans, | |
b2a922cd | 3789 | char *dp, |
76560669 | 3790 | int len) |
1da177e4 | 3791 | { |
76560669 | 3792 | xlog_recover_item_t *item; |
b2a922cd | 3793 | char *ptr, *old_ptr; |
76560669 DC |
3794 | int old_len; |
3795 | ||
89cebc84 BF |
3796 | /* |
3797 | * If the transaction is empty, the header was split across this and the | |
3798 | * previous record. Copy the rest of the header. | |
3799 | */ | |
76560669 | 3800 | if (list_empty(&trans->r_itemq)) { |
848ccfc8 | 3801 | ASSERT(len <= sizeof(struct xfs_trans_header)); |
89cebc84 BF |
3802 | if (len > sizeof(struct xfs_trans_header)) { |
3803 | xfs_warn(log->l_mp, "%s: bad header length", __func__); | |
3804 | return -EIO; | |
3805 | } | |
3806 | ||
76560669 | 3807 | xlog_recover_add_item(&trans->r_itemq); |
b2a922cd | 3808 | ptr = (char *)&trans->r_theader + |
89cebc84 | 3809 | sizeof(struct xfs_trans_header) - len; |
76560669 DC |
3810 | memcpy(ptr, dp, len); |
3811 | return 0; | |
3812 | } | |
89cebc84 | 3813 | |
76560669 DC |
3814 | /* take the tail entry */ |
3815 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
3816 | ||
3817 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; | |
3818 | old_len = item->ri_buf[item->ri_cnt-1].i_len; | |
3819 | ||
3820 | ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP); | |
3821 | memcpy(&ptr[old_len], dp, len); | |
3822 | item->ri_buf[item->ri_cnt-1].i_len += len; | |
3823 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; | |
3824 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); | |
1da177e4 LT |
3825 | return 0; |
3826 | } | |
3827 | ||
76560669 DC |
3828 | /* |
3829 | * The next region to add is the start of a new region. It could be | |
3830 | * a whole region or it could be the first part of a new region. Because | |
3831 | * of this, the assumption here is that the type and size fields of all | |
3832 | * format structures fit into the first 32 bits of the structure. | |
3833 | * | |
3834 | * This works because all regions must be 32 bit aligned. Therefore, we | |
3835 | * either have both fields or we have neither field. In the case we have | |
3836 | * neither field, the data part of the region is zero length. We only have | |
3837 | * a log_op_header and can throw away the header since a new one will appear | |
3838 | * later. If we have at least 4 bytes, then we can determine how many regions | |
3839 | * will appear in the current log item. | |
3840 | */ | |
3841 | STATIC int | |
3842 | xlog_recover_add_to_trans( | |
3843 | struct xlog *log, | |
3844 | struct xlog_recover *trans, | |
b2a922cd | 3845 | char *dp, |
76560669 DC |
3846 | int len) |
3847 | { | |
3848 | xfs_inode_log_format_t *in_f; /* any will do */ | |
3849 | xlog_recover_item_t *item; | |
b2a922cd | 3850 | char *ptr; |
76560669 DC |
3851 | |
3852 | if (!len) | |
3853 | return 0; | |
3854 | if (list_empty(&trans->r_itemq)) { | |
3855 | /* we need to catch log corruptions here */ | |
3856 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { | |
3857 | xfs_warn(log->l_mp, "%s: bad header magic number", | |
3858 | __func__); | |
3859 | ASSERT(0); | |
3860 | return -EIO; | |
3861 | } | |
89cebc84 BF |
3862 | |
3863 | if (len > sizeof(struct xfs_trans_header)) { | |
3864 | xfs_warn(log->l_mp, "%s: bad header length", __func__); | |
3865 | ASSERT(0); | |
3866 | return -EIO; | |
3867 | } | |
3868 | ||
3869 | /* | |
3870 | * The transaction header can be arbitrarily split across op | |
3871 | * records. If we don't have the whole thing here, copy what we | |
3872 | * do have and handle the rest in the next record. | |
3873 | */ | |
3874 | if (len == sizeof(struct xfs_trans_header)) | |
76560669 DC |
3875 | xlog_recover_add_item(&trans->r_itemq); |
3876 | memcpy(&trans->r_theader, dp, len); | |
3877 | return 0; | |
3878 | } | |
3879 | ||
3880 | ptr = kmem_alloc(len, KM_SLEEP); | |
3881 | memcpy(ptr, dp, len); | |
3882 | in_f = (xfs_inode_log_format_t *)ptr; | |
3883 | ||
3884 | /* take the tail entry */ | |
3885 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
3886 | if (item->ri_total != 0 && | |
3887 | item->ri_total == item->ri_cnt) { | |
3888 | /* tail item is in use, get a new one */ | |
3889 | xlog_recover_add_item(&trans->r_itemq); | |
3890 | item = list_entry(trans->r_itemq.prev, | |
3891 | xlog_recover_item_t, ri_list); | |
3892 | } | |
3893 | ||
3894 | if (item->ri_total == 0) { /* first region to be added */ | |
3895 | if (in_f->ilf_size == 0 || | |
3896 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { | |
3897 | xfs_warn(log->l_mp, | |
3898 | "bad number of regions (%d) in inode log format", | |
3899 | in_f->ilf_size); | |
3900 | ASSERT(0); | |
3901 | kmem_free(ptr); | |
3902 | return -EIO; | |
3903 | } | |
3904 | ||
3905 | item->ri_total = in_f->ilf_size; | |
3906 | item->ri_buf = | |
3907 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), | |
3908 | KM_SLEEP); | |
3909 | } | |
3910 | ASSERT(item->ri_total > item->ri_cnt); | |
3911 | /* Description region is ri_buf[0] */ | |
3912 | item->ri_buf[item->ri_cnt].i_addr = ptr; | |
3913 | item->ri_buf[item->ri_cnt].i_len = len; | |
3914 | item->ri_cnt++; | |
3915 | trace_xfs_log_recover_item_add(log, trans, item, 0); | |
3916 | return 0; | |
3917 | } | |
b818cca1 | 3918 | |
76560669 DC |
3919 | /* |
3920 | * Free up any resources allocated by the transaction | |
3921 | * | |
3922 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. | |
3923 | */ | |
3924 | STATIC void | |
3925 | xlog_recover_free_trans( | |
3926 | struct xlog_recover *trans) | |
3927 | { | |
3928 | xlog_recover_item_t *item, *n; | |
3929 | int i; | |
3930 | ||
3931 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { | |
3932 | /* Free the regions in the item. */ | |
3933 | list_del(&item->ri_list); | |
3934 | for (i = 0; i < item->ri_cnt; i++) | |
3935 | kmem_free(item->ri_buf[i].i_addr); | |
3936 | /* Free the item itself */ | |
3937 | kmem_free(item->ri_buf); | |
3938 | kmem_free(item); | |
3939 | } | |
3940 | /* Free the transaction recover structure */ | |
3941 | kmem_free(trans); | |
3942 | } | |
3943 | ||
e9131e50 DC |
3944 | /* |
3945 | * On error or completion, trans is freed. | |
3946 | */ | |
1da177e4 | 3947 | STATIC int |
eeb11688 DC |
3948 | xlog_recovery_process_trans( |
3949 | struct xlog *log, | |
3950 | struct xlog_recover *trans, | |
b2a922cd | 3951 | char *dp, |
eeb11688 DC |
3952 | unsigned int len, |
3953 | unsigned int flags, | |
3954 | int pass) | |
1da177e4 | 3955 | { |
e9131e50 DC |
3956 | int error = 0; |
3957 | bool freeit = false; | |
eeb11688 DC |
3958 | |
3959 | /* mask off ophdr transaction container flags */ | |
3960 | flags &= ~XLOG_END_TRANS; | |
3961 | if (flags & XLOG_WAS_CONT_TRANS) | |
3962 | flags &= ~XLOG_CONTINUE_TRANS; | |
3963 | ||
88b863db DC |
3964 | /* |
3965 | * Callees must not free the trans structure. We'll decide if we need to | |
3966 | * free it or not based on the operation being done and it's result. | |
3967 | */ | |
eeb11688 DC |
3968 | switch (flags) { |
3969 | /* expected flag values */ | |
3970 | case 0: | |
3971 | case XLOG_CONTINUE_TRANS: | |
3972 | error = xlog_recover_add_to_trans(log, trans, dp, len); | |
3973 | break; | |
3974 | case XLOG_WAS_CONT_TRANS: | |
3975 | error = xlog_recover_add_to_cont_trans(log, trans, dp, len); | |
3976 | break; | |
3977 | case XLOG_COMMIT_TRANS: | |
3978 | error = xlog_recover_commit_trans(log, trans, pass); | |
88b863db DC |
3979 | /* success or fail, we are now done with this transaction. */ |
3980 | freeit = true; | |
eeb11688 DC |
3981 | break; |
3982 | ||
3983 | /* unexpected flag values */ | |
3984 | case XLOG_UNMOUNT_TRANS: | |
e9131e50 | 3985 | /* just skip trans */ |
eeb11688 | 3986 | xfs_warn(log->l_mp, "%s: Unmount LR", __func__); |
e9131e50 | 3987 | freeit = true; |
eeb11688 DC |
3988 | break; |
3989 | case XLOG_START_TRANS: | |
eeb11688 DC |
3990 | default: |
3991 | xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags); | |
3992 | ASSERT(0); | |
e9131e50 | 3993 | error = -EIO; |
eeb11688 DC |
3994 | break; |
3995 | } | |
e9131e50 DC |
3996 | if (error || freeit) |
3997 | xlog_recover_free_trans(trans); | |
eeb11688 DC |
3998 | return error; |
3999 | } | |
4000 | ||
b818cca1 DC |
4001 | /* |
4002 | * Lookup the transaction recovery structure associated with the ID in the | |
4003 | * current ophdr. If the transaction doesn't exist and the start flag is set in | |
4004 | * the ophdr, then allocate a new transaction for future ID matches to find. | |
4005 | * Either way, return what we found during the lookup - an existing transaction | |
4006 | * or nothing. | |
4007 | */ | |
eeb11688 DC |
4008 | STATIC struct xlog_recover * |
4009 | xlog_recover_ophdr_to_trans( | |
4010 | struct hlist_head rhash[], | |
4011 | struct xlog_rec_header *rhead, | |
4012 | struct xlog_op_header *ohead) | |
4013 | { | |
4014 | struct xlog_recover *trans; | |
4015 | xlog_tid_t tid; | |
4016 | struct hlist_head *rhp; | |
4017 | ||
4018 | tid = be32_to_cpu(ohead->oh_tid); | |
4019 | rhp = &rhash[XLOG_RHASH(tid)]; | |
b818cca1 DC |
4020 | hlist_for_each_entry(trans, rhp, r_list) { |
4021 | if (trans->r_log_tid == tid) | |
4022 | return trans; | |
4023 | } | |
eeb11688 DC |
4024 | |
4025 | /* | |
b818cca1 DC |
4026 | * skip over non-start transaction headers - we could be |
4027 | * processing slack space before the next transaction starts | |
4028 | */ | |
4029 | if (!(ohead->oh_flags & XLOG_START_TRANS)) | |
4030 | return NULL; | |
4031 | ||
4032 | ASSERT(be32_to_cpu(ohead->oh_len) == 0); | |
4033 | ||
4034 | /* | |
4035 | * This is a new transaction so allocate a new recovery container to | |
4036 | * hold the recovery ops that will follow. | |
4037 | */ | |
4038 | trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP); | |
4039 | trans->r_log_tid = tid; | |
4040 | trans->r_lsn = be64_to_cpu(rhead->h_lsn); | |
4041 | INIT_LIST_HEAD(&trans->r_itemq); | |
4042 | INIT_HLIST_NODE(&trans->r_list); | |
4043 | hlist_add_head(&trans->r_list, rhp); | |
4044 | ||
4045 | /* | |
4046 | * Nothing more to do for this ophdr. Items to be added to this new | |
4047 | * transaction will be in subsequent ophdr containers. | |
eeb11688 | 4048 | */ |
eeb11688 DC |
4049 | return NULL; |
4050 | } | |
4051 | ||
4052 | STATIC int | |
4053 | xlog_recover_process_ophdr( | |
4054 | struct xlog *log, | |
4055 | struct hlist_head rhash[], | |
4056 | struct xlog_rec_header *rhead, | |
4057 | struct xlog_op_header *ohead, | |
b2a922cd CH |
4058 | char *dp, |
4059 | char *end, | |
eeb11688 DC |
4060 | int pass) |
4061 | { | |
4062 | struct xlog_recover *trans; | |
eeb11688 DC |
4063 | unsigned int len; |
4064 | ||
4065 | /* Do we understand who wrote this op? */ | |
4066 | if (ohead->oh_clientid != XFS_TRANSACTION && | |
4067 | ohead->oh_clientid != XFS_LOG) { | |
4068 | xfs_warn(log->l_mp, "%s: bad clientid 0x%x", | |
4069 | __func__, ohead->oh_clientid); | |
4070 | ASSERT(0); | |
4071 | return -EIO; | |
4072 | } | |
4073 | ||
4074 | /* | |
4075 | * Check the ophdr contains all the data it is supposed to contain. | |
4076 | */ | |
4077 | len = be32_to_cpu(ohead->oh_len); | |
4078 | if (dp + len > end) { | |
4079 | xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len); | |
4080 | WARN_ON(1); | |
4081 | return -EIO; | |
4082 | } | |
4083 | ||
4084 | trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead); | |
4085 | if (!trans) { | |
4086 | /* nothing to do, so skip over this ophdr */ | |
4087 | return 0; | |
4088 | } | |
4089 | ||
e9131e50 DC |
4090 | return xlog_recovery_process_trans(log, trans, dp, len, |
4091 | ohead->oh_flags, pass); | |
1da177e4 LT |
4092 | } |
4093 | ||
4094 | /* | |
4095 | * There are two valid states of the r_state field. 0 indicates that the | |
4096 | * transaction structure is in a normal state. We have either seen the | |
4097 | * start of the transaction or the last operation we added was not a partial | |
4098 | * operation. If the last operation we added to the transaction was a | |
4099 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. | |
4100 | * | |
4101 | * NOTE: skip LRs with 0 data length. | |
4102 | */ | |
4103 | STATIC int | |
4104 | xlog_recover_process_data( | |
9a8d2fdb | 4105 | struct xlog *log, |
f0a76953 | 4106 | struct hlist_head rhash[], |
9a8d2fdb | 4107 | struct xlog_rec_header *rhead, |
b2a922cd | 4108 | char *dp, |
1da177e4 LT |
4109 | int pass) |
4110 | { | |
eeb11688 | 4111 | struct xlog_op_header *ohead; |
b2a922cd | 4112 | char *end; |
1da177e4 | 4113 | int num_logops; |
1da177e4 | 4114 | int error; |
1da177e4 | 4115 | |
eeb11688 | 4116 | end = dp + be32_to_cpu(rhead->h_len); |
b53e675d | 4117 | num_logops = be32_to_cpu(rhead->h_num_logops); |
1da177e4 LT |
4118 | |
4119 | /* check the log format matches our own - else we can't recover */ | |
4120 | if (xlog_header_check_recover(log->l_mp, rhead)) | |
2451337d | 4121 | return -EIO; |
1da177e4 | 4122 | |
eeb11688 DC |
4123 | while ((dp < end) && num_logops) { |
4124 | ||
4125 | ohead = (struct xlog_op_header *)dp; | |
4126 | dp += sizeof(*ohead); | |
4127 | ASSERT(dp <= end); | |
4128 | ||
4129 | /* errors will abort recovery */ | |
4130 | error = xlog_recover_process_ophdr(log, rhash, rhead, ohead, | |
4131 | dp, end, pass); | |
4132 | if (error) | |
4133 | return error; | |
4134 | ||
67fcb7bf | 4135 | dp += be32_to_cpu(ohead->oh_len); |
1da177e4 LT |
4136 | num_logops--; |
4137 | } | |
4138 | return 0; | |
4139 | } | |
4140 | ||
4141 | /* | |
4142 | * Process an extent free intent item that was recovered from | |
4143 | * the log. We need to free the extents that it describes. | |
4144 | */ | |
3c1e2bbe | 4145 | STATIC int |
1da177e4 LT |
4146 | xlog_recover_process_efi( |
4147 | xfs_mount_t *mp, | |
4148 | xfs_efi_log_item_t *efip) | |
4149 | { | |
4150 | xfs_efd_log_item_t *efdp; | |
4151 | xfs_trans_t *tp; | |
4152 | int i; | |
3c1e2bbe | 4153 | int error = 0; |
1da177e4 LT |
4154 | xfs_extent_t *extp; |
4155 | xfs_fsblock_t startblock_fsb; | |
4156 | ||
b199c8a4 | 4157 | ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)); |
1da177e4 LT |
4158 | |
4159 | /* | |
4160 | * First check the validity of the extents described by the | |
4161 | * EFI. If any are bad, then assume that all are bad and | |
4162 | * just toss the EFI. | |
4163 | */ | |
4164 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
4165 | extp = &(efip->efi_format.efi_extents[i]); | |
4166 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
4167 | XFS_FSB_TO_DADDR(mp, extp->ext_start)); | |
4168 | if ((startblock_fsb == 0) || | |
4169 | (extp->ext_len == 0) || | |
4170 | (startblock_fsb >= mp->m_sb.sb_dblocks) || | |
4171 | (extp->ext_len >= mp->m_sb.sb_agblocks)) { | |
4172 | /* | |
4173 | * This will pull the EFI from the AIL and | |
4174 | * free the memory associated with it. | |
4175 | */ | |
666d644c | 4176 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
5e4b5386 | 4177 | xfs_efi_release(efip); |
2451337d | 4178 | return -EIO; |
1da177e4 LT |
4179 | } |
4180 | } | |
4181 | ||
4182 | tp = xfs_trans_alloc(mp, 0); | |
3d3c8b52 | 4183 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); |
fc6149d8 DC |
4184 | if (error) |
4185 | goto abort_error; | |
1da177e4 LT |
4186 | efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents); |
4187 | ||
4188 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
4189 | extp = &(efip->efi_format.efi_extents[i]); | |
6bc43af3 BF |
4190 | error = xfs_trans_free_extent(tp, efdp, extp->ext_start, |
4191 | extp->ext_len); | |
fc6149d8 DC |
4192 | if (error) |
4193 | goto abort_error; | |
6bc43af3 | 4194 | |
1da177e4 LT |
4195 | } |
4196 | ||
b199c8a4 | 4197 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
70393313 | 4198 | error = xfs_trans_commit(tp); |
3c1e2bbe | 4199 | return error; |
fc6149d8 DC |
4200 | |
4201 | abort_error: | |
4906e215 | 4202 | xfs_trans_cancel(tp); |
fc6149d8 | 4203 | return error; |
1da177e4 LT |
4204 | } |
4205 | ||
1da177e4 LT |
4206 | /* |
4207 | * When this is called, all of the EFIs which did not have | |
4208 | * corresponding EFDs should be in the AIL. What we do now | |
4209 | * is free the extents associated with each one. | |
4210 | * | |
4211 | * Since we process the EFIs in normal transactions, they | |
4212 | * will be removed at some point after the commit. This prevents | |
4213 | * us from just walking down the list processing each one. | |
4214 | * We'll use a flag in the EFI to skip those that we've already | |
4215 | * processed and use the AIL iteration mechanism's generation | |
4216 | * count to try to speed this up at least a bit. | |
4217 | * | |
4218 | * When we start, we know that the EFIs are the only things in | |
4219 | * the AIL. As we process them, however, other items are added | |
4220 | * to the AIL. Since everything added to the AIL must come after | |
4221 | * everything already in the AIL, we stop processing as soon as | |
4222 | * we see something other than an EFI in the AIL. | |
4223 | */ | |
3c1e2bbe | 4224 | STATIC int |
1da177e4 | 4225 | xlog_recover_process_efis( |
f0b2efad | 4226 | struct xlog *log) |
1da177e4 | 4227 | { |
f0b2efad BF |
4228 | struct xfs_log_item *lip; |
4229 | struct xfs_efi_log_item *efip; | |
3c1e2bbe | 4230 | int error = 0; |
27d8d5fe | 4231 | struct xfs_ail_cursor cur; |
a9c21c1b | 4232 | struct xfs_ail *ailp; |
1da177e4 | 4233 | |
a9c21c1b DC |
4234 | ailp = log->l_ailp; |
4235 | spin_lock(&ailp->xa_lock); | |
4236 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
1da177e4 LT |
4237 | while (lip != NULL) { |
4238 | /* | |
4239 | * We're done when we see something other than an EFI. | |
27d8d5fe | 4240 | * There should be no EFIs left in the AIL now. |
1da177e4 LT |
4241 | */ |
4242 | if (lip->li_type != XFS_LI_EFI) { | |
27d8d5fe | 4243 | #ifdef DEBUG |
a9c21c1b | 4244 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
27d8d5fe DC |
4245 | ASSERT(lip->li_type != XFS_LI_EFI); |
4246 | #endif | |
1da177e4 LT |
4247 | break; |
4248 | } | |
4249 | ||
4250 | /* | |
4251 | * Skip EFIs that we've already processed. | |
4252 | */ | |
f0b2efad | 4253 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); |
b199c8a4 | 4254 | if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) { |
a9c21c1b | 4255 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 LT |
4256 | continue; |
4257 | } | |
4258 | ||
a9c21c1b DC |
4259 | spin_unlock(&ailp->xa_lock); |
4260 | error = xlog_recover_process_efi(log->l_mp, efip); | |
4261 | spin_lock(&ailp->xa_lock); | |
27d8d5fe DC |
4262 | if (error) |
4263 | goto out; | |
a9c21c1b | 4264 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 4265 | } |
27d8d5fe | 4266 | out: |
e4a1e29c | 4267 | xfs_trans_ail_cursor_done(&cur); |
a9c21c1b | 4268 | spin_unlock(&ailp->xa_lock); |
3c1e2bbe | 4269 | return error; |
1da177e4 LT |
4270 | } |
4271 | ||
f0b2efad BF |
4272 | /* |
4273 | * A cancel occurs when the mount has failed and we're bailing out. Release all | |
4274 | * pending EFIs so they don't pin the AIL. | |
4275 | */ | |
4276 | STATIC int | |
4277 | xlog_recover_cancel_efis( | |
4278 | struct xlog *log) | |
4279 | { | |
4280 | struct xfs_log_item *lip; | |
4281 | struct xfs_efi_log_item *efip; | |
4282 | int error = 0; | |
4283 | struct xfs_ail_cursor cur; | |
4284 | struct xfs_ail *ailp; | |
4285 | ||
4286 | ailp = log->l_ailp; | |
4287 | spin_lock(&ailp->xa_lock); | |
4288 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
4289 | while (lip != NULL) { | |
4290 | /* | |
4291 | * We're done when we see something other than an EFI. | |
4292 | * There should be no EFIs left in the AIL now. | |
4293 | */ | |
4294 | if (lip->li_type != XFS_LI_EFI) { | |
4295 | #ifdef DEBUG | |
4296 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) | |
4297 | ASSERT(lip->li_type != XFS_LI_EFI); | |
4298 | #endif | |
4299 | break; | |
4300 | } | |
4301 | ||
4302 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); | |
4303 | ||
4304 | spin_unlock(&ailp->xa_lock); | |
4305 | xfs_efi_release(efip); | |
4306 | spin_lock(&ailp->xa_lock); | |
4307 | ||
4308 | lip = xfs_trans_ail_cursor_next(ailp, &cur); | |
4309 | } | |
4310 | ||
4311 | xfs_trans_ail_cursor_done(&cur); | |
4312 | spin_unlock(&ailp->xa_lock); | |
4313 | return error; | |
4314 | } | |
4315 | ||
1da177e4 LT |
4316 | /* |
4317 | * This routine performs a transaction to null out a bad inode pointer | |
4318 | * in an agi unlinked inode hash bucket. | |
4319 | */ | |
4320 | STATIC void | |
4321 | xlog_recover_clear_agi_bucket( | |
4322 | xfs_mount_t *mp, | |
4323 | xfs_agnumber_t agno, | |
4324 | int bucket) | |
4325 | { | |
4326 | xfs_trans_t *tp; | |
4327 | xfs_agi_t *agi; | |
4328 | xfs_buf_t *agibp; | |
4329 | int offset; | |
4330 | int error; | |
4331 | ||
4332 | tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET); | |
3d3c8b52 | 4333 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0); |
e5720eec DC |
4334 | if (error) |
4335 | goto out_abort; | |
1da177e4 | 4336 | |
5e1be0fb CH |
4337 | error = xfs_read_agi(mp, tp, agno, &agibp); |
4338 | if (error) | |
e5720eec | 4339 | goto out_abort; |
1da177e4 | 4340 | |
5e1be0fb | 4341 | agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 4342 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
4343 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
4344 | (sizeof(xfs_agino_t) * bucket); | |
4345 | xfs_trans_log_buf(tp, agibp, offset, | |
4346 | (offset + sizeof(xfs_agino_t) - 1)); | |
4347 | ||
70393313 | 4348 | error = xfs_trans_commit(tp); |
e5720eec DC |
4349 | if (error) |
4350 | goto out_error; | |
4351 | return; | |
4352 | ||
4353 | out_abort: | |
4906e215 | 4354 | xfs_trans_cancel(tp); |
e5720eec | 4355 | out_error: |
a0fa2b67 | 4356 | xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno); |
e5720eec | 4357 | return; |
1da177e4 LT |
4358 | } |
4359 | ||
23fac50f CH |
4360 | STATIC xfs_agino_t |
4361 | xlog_recover_process_one_iunlink( | |
4362 | struct xfs_mount *mp, | |
4363 | xfs_agnumber_t agno, | |
4364 | xfs_agino_t agino, | |
4365 | int bucket) | |
4366 | { | |
4367 | struct xfs_buf *ibp; | |
4368 | struct xfs_dinode *dip; | |
4369 | struct xfs_inode *ip; | |
4370 | xfs_ino_t ino; | |
4371 | int error; | |
4372 | ||
4373 | ino = XFS_AGINO_TO_INO(mp, agno, agino); | |
7b6259e7 | 4374 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip); |
23fac50f CH |
4375 | if (error) |
4376 | goto fail; | |
4377 | ||
4378 | /* | |
4379 | * Get the on disk inode to find the next inode in the bucket. | |
4380 | */ | |
475ee413 | 4381 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0); |
23fac50f | 4382 | if (error) |
0e446673 | 4383 | goto fail_iput; |
23fac50f | 4384 | |
23fac50f | 4385 | ASSERT(ip->i_d.di_nlink == 0); |
0e446673 | 4386 | ASSERT(ip->i_d.di_mode != 0); |
23fac50f CH |
4387 | |
4388 | /* setup for the next pass */ | |
4389 | agino = be32_to_cpu(dip->di_next_unlinked); | |
4390 | xfs_buf_relse(ibp); | |
4391 | ||
4392 | /* | |
4393 | * Prevent any DMAPI event from being sent when the reference on | |
4394 | * the inode is dropped. | |
4395 | */ | |
4396 | ip->i_d.di_dmevmask = 0; | |
4397 | ||
0e446673 | 4398 | IRELE(ip); |
23fac50f CH |
4399 | return agino; |
4400 | ||
0e446673 CH |
4401 | fail_iput: |
4402 | IRELE(ip); | |
23fac50f CH |
4403 | fail: |
4404 | /* | |
4405 | * We can't read in the inode this bucket points to, or this inode | |
4406 | * is messed up. Just ditch this bucket of inodes. We will lose | |
4407 | * some inodes and space, but at least we won't hang. | |
4408 | * | |
4409 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to | |
4410 | * clear the inode pointer in the bucket. | |
4411 | */ | |
4412 | xlog_recover_clear_agi_bucket(mp, agno, bucket); | |
4413 | return NULLAGINO; | |
4414 | } | |
4415 | ||
1da177e4 LT |
4416 | /* |
4417 | * xlog_iunlink_recover | |
4418 | * | |
4419 | * This is called during recovery to process any inodes which | |
4420 | * we unlinked but not freed when the system crashed. These | |
4421 | * inodes will be on the lists in the AGI blocks. What we do | |
4422 | * here is scan all the AGIs and fully truncate and free any | |
4423 | * inodes found on the lists. Each inode is removed from the | |
4424 | * lists when it has been fully truncated and is freed. The | |
4425 | * freeing of the inode and its removal from the list must be | |
4426 | * atomic. | |
4427 | */ | |
d96f8f89 | 4428 | STATIC void |
1da177e4 | 4429 | xlog_recover_process_iunlinks( |
9a8d2fdb | 4430 | struct xlog *log) |
1da177e4 LT |
4431 | { |
4432 | xfs_mount_t *mp; | |
4433 | xfs_agnumber_t agno; | |
4434 | xfs_agi_t *agi; | |
4435 | xfs_buf_t *agibp; | |
1da177e4 | 4436 | xfs_agino_t agino; |
1da177e4 LT |
4437 | int bucket; |
4438 | int error; | |
4439 | uint mp_dmevmask; | |
4440 | ||
4441 | mp = log->l_mp; | |
4442 | ||
4443 | /* | |
4444 | * Prevent any DMAPI event from being sent while in this function. | |
4445 | */ | |
4446 | mp_dmevmask = mp->m_dmevmask; | |
4447 | mp->m_dmevmask = 0; | |
4448 | ||
4449 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
4450 | /* | |
4451 | * Find the agi for this ag. | |
4452 | */ | |
5e1be0fb CH |
4453 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
4454 | if (error) { | |
4455 | /* | |
4456 | * AGI is b0rked. Don't process it. | |
4457 | * | |
4458 | * We should probably mark the filesystem as corrupt | |
4459 | * after we've recovered all the ag's we can.... | |
4460 | */ | |
4461 | continue; | |
1da177e4 | 4462 | } |
d97d32ed JK |
4463 | /* |
4464 | * Unlock the buffer so that it can be acquired in the normal | |
4465 | * course of the transaction to truncate and free each inode. | |
4466 | * Because we are not racing with anyone else here for the AGI | |
4467 | * buffer, we don't even need to hold it locked to read the | |
4468 | * initial unlinked bucket entries out of the buffer. We keep | |
4469 | * buffer reference though, so that it stays pinned in memory | |
4470 | * while we need the buffer. | |
4471 | */ | |
1da177e4 | 4472 | agi = XFS_BUF_TO_AGI(agibp); |
d97d32ed | 4473 | xfs_buf_unlock(agibp); |
1da177e4 LT |
4474 | |
4475 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { | |
16259e7d | 4476 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
1da177e4 | 4477 | while (agino != NULLAGINO) { |
23fac50f CH |
4478 | agino = xlog_recover_process_one_iunlink(mp, |
4479 | agno, agino, bucket); | |
1da177e4 LT |
4480 | } |
4481 | } | |
d97d32ed | 4482 | xfs_buf_rele(agibp); |
1da177e4 LT |
4483 | } |
4484 | ||
4485 | mp->m_dmevmask = mp_dmevmask; | |
4486 | } | |
4487 | ||
0e446be4 | 4488 | STATIC int |
1da177e4 | 4489 | xlog_unpack_data( |
9a8d2fdb | 4490 | struct xlog_rec_header *rhead, |
b2a922cd | 4491 | char *dp, |
9a8d2fdb | 4492 | struct xlog *log) |
1da177e4 LT |
4493 | { |
4494 | int i, j, k; | |
1da177e4 | 4495 | |
b53e675d | 4496 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
1da177e4 | 4497 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
b53e675d | 4498 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
1da177e4 LT |
4499 | dp += BBSIZE; |
4500 | } | |
4501 | ||
62118709 | 4502 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b28708d6 | 4503 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
b53e675d | 4504 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
1da177e4 LT |
4505 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
4506 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 4507 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
1da177e4 LT |
4508 | dp += BBSIZE; |
4509 | } | |
4510 | } | |
0e446be4 CH |
4511 | |
4512 | return 0; | |
1da177e4 LT |
4513 | } |
4514 | ||
9d94901f | 4515 | /* |
b94fb2d1 | 4516 | * CRC check, unpack and process a log record. |
9d94901f BF |
4517 | */ |
4518 | STATIC int | |
4519 | xlog_recover_process( | |
4520 | struct xlog *log, | |
4521 | struct hlist_head rhash[], | |
4522 | struct xlog_rec_header *rhead, | |
4523 | char *dp, | |
4524 | int pass) | |
4525 | { | |
4526 | int error; | |
b94fb2d1 BF |
4527 | __le32 crc; |
4528 | ||
6528250b BF |
4529 | crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len)); |
4530 | ||
b94fb2d1 | 4531 | /* |
6528250b BF |
4532 | * Nothing else to do if this is a CRC verification pass. Just return |
4533 | * if this a record with a non-zero crc. Unfortunately, mkfs always | |
4534 | * sets h_crc to 0 so we must consider this valid even on v5 supers. | |
4535 | * Otherwise, return EFSBADCRC on failure so the callers up the stack | |
4536 | * know precisely what failed. | |
4537 | */ | |
4538 | if (pass == XLOG_RECOVER_CRCPASS) { | |
4539 | if (rhead->h_crc && crc != le32_to_cpu(rhead->h_crc)) | |
4540 | return -EFSBADCRC; | |
4541 | return 0; | |
4542 | } | |
4543 | ||
4544 | /* | |
4545 | * We're in the normal recovery path. Issue a warning if and only if the | |
4546 | * CRC in the header is non-zero. This is an advisory warning and the | |
4547 | * zero CRC check prevents warnings from being emitted when upgrading | |
4548 | * the kernel from one that does not add CRCs by default. | |
b94fb2d1 | 4549 | */ |
b94fb2d1 BF |
4550 | if (crc != le32_to_cpu(rhead->h_crc)) { |
4551 | if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) { | |
4552 | xfs_alert(log->l_mp, | |
4553 | "log record CRC mismatch: found 0x%x, expected 0x%x.", | |
4554 | le32_to_cpu(rhead->h_crc), | |
4555 | le32_to_cpu(crc)); | |
4556 | xfs_hex_dump(dp, 32); | |
4557 | } | |
4558 | ||
4559 | /* | |
4560 | * If the filesystem is CRC enabled, this mismatch becomes a | |
4561 | * fatal log corruption failure. | |
4562 | */ | |
4563 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) | |
4564 | return -EFSCORRUPTED; | |
4565 | } | |
9d94901f BF |
4566 | |
4567 | error = xlog_unpack_data(rhead, dp, log); | |
4568 | if (error) | |
4569 | return error; | |
4570 | ||
4571 | return xlog_recover_process_data(log, rhash, rhead, dp, pass); | |
4572 | } | |
4573 | ||
1da177e4 LT |
4574 | STATIC int |
4575 | xlog_valid_rec_header( | |
9a8d2fdb MT |
4576 | struct xlog *log, |
4577 | struct xlog_rec_header *rhead, | |
1da177e4 LT |
4578 | xfs_daddr_t blkno) |
4579 | { | |
4580 | int hlen; | |
4581 | ||
69ef921b | 4582 | if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) { |
1da177e4 LT |
4583 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
4584 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 4585 | return -EFSCORRUPTED; |
1da177e4 LT |
4586 | } |
4587 | if (unlikely( | |
4588 | (!rhead->h_version || | |
b53e675d | 4589 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
a0fa2b67 | 4590 | xfs_warn(log->l_mp, "%s: unrecognised log version (%d).", |
34a622b2 | 4591 | __func__, be32_to_cpu(rhead->h_version)); |
2451337d | 4592 | return -EIO; |
1da177e4 LT |
4593 | } |
4594 | ||
4595 | /* LR body must have data or it wouldn't have been written */ | |
b53e675d | 4596 | hlen = be32_to_cpu(rhead->h_len); |
1da177e4 LT |
4597 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
4598 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", | |
4599 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 4600 | return -EFSCORRUPTED; |
1da177e4 LT |
4601 | } |
4602 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { | |
4603 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", | |
4604 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 4605 | return -EFSCORRUPTED; |
1da177e4 LT |
4606 | } |
4607 | return 0; | |
4608 | } | |
4609 | ||
4610 | /* | |
4611 | * Read the log from tail to head and process the log records found. | |
4612 | * Handle the two cases where the tail and head are in the same cycle | |
4613 | * and where the active portion of the log wraps around the end of | |
4614 | * the physical log separately. The pass parameter is passed through | |
4615 | * to the routines called to process the data and is not looked at | |
4616 | * here. | |
4617 | */ | |
4618 | STATIC int | |
4619 | xlog_do_recovery_pass( | |
9a8d2fdb | 4620 | struct xlog *log, |
1da177e4 LT |
4621 | xfs_daddr_t head_blk, |
4622 | xfs_daddr_t tail_blk, | |
d7f37692 BF |
4623 | int pass, |
4624 | xfs_daddr_t *first_bad) /* out: first bad log rec */ | |
1da177e4 LT |
4625 | { |
4626 | xlog_rec_header_t *rhead; | |
4627 | xfs_daddr_t blk_no; | |
d7f37692 | 4628 | xfs_daddr_t rhead_blk; |
b2a922cd | 4629 | char *offset; |
1da177e4 | 4630 | xfs_buf_t *hbp, *dbp; |
a70f9fe5 | 4631 | int error = 0, h_size, h_len; |
1da177e4 LT |
4632 | int bblks, split_bblks; |
4633 | int hblks, split_hblks, wrapped_hblks; | |
f0a76953 | 4634 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
1da177e4 LT |
4635 | |
4636 | ASSERT(head_blk != tail_blk); | |
d7f37692 | 4637 | rhead_blk = 0; |
1da177e4 LT |
4638 | |
4639 | /* | |
4640 | * Read the header of the tail block and get the iclog buffer size from | |
4641 | * h_size. Use this to tell how many sectors make up the log header. | |
4642 | */ | |
62118709 | 4643 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
1da177e4 LT |
4644 | /* |
4645 | * When using variable length iclogs, read first sector of | |
4646 | * iclog header and extract the header size from it. Get a | |
4647 | * new hbp that is the correct size. | |
4648 | */ | |
4649 | hbp = xlog_get_bp(log, 1); | |
4650 | if (!hbp) | |
2451337d | 4651 | return -ENOMEM; |
076e6acb CH |
4652 | |
4653 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); | |
4654 | if (error) | |
1da177e4 | 4655 | goto bread_err1; |
076e6acb | 4656 | |
1da177e4 LT |
4657 | rhead = (xlog_rec_header_t *)offset; |
4658 | error = xlog_valid_rec_header(log, rhead, tail_blk); | |
4659 | if (error) | |
4660 | goto bread_err1; | |
a70f9fe5 BF |
4661 | |
4662 | /* | |
4663 | * xfsprogs has a bug where record length is based on lsunit but | |
4664 | * h_size (iclog size) is hardcoded to 32k. Now that we | |
4665 | * unconditionally CRC verify the unmount record, this means the | |
4666 | * log buffer can be too small for the record and cause an | |
4667 | * overrun. | |
4668 | * | |
4669 | * Detect this condition here. Use lsunit for the buffer size as | |
4670 | * long as this looks like the mkfs case. Otherwise, return an | |
4671 | * error to avoid a buffer overrun. | |
4672 | */ | |
b53e675d | 4673 | h_size = be32_to_cpu(rhead->h_size); |
a70f9fe5 BF |
4674 | h_len = be32_to_cpu(rhead->h_len); |
4675 | if (h_len > h_size) { | |
4676 | if (h_len <= log->l_mp->m_logbsize && | |
4677 | be32_to_cpu(rhead->h_num_logops) == 1) { | |
4678 | xfs_warn(log->l_mp, | |
4679 | "invalid iclog size (%d bytes), using lsunit (%d bytes)", | |
4680 | h_size, log->l_mp->m_logbsize); | |
4681 | h_size = log->l_mp->m_logbsize; | |
4682 | } else | |
4683 | return -EFSCORRUPTED; | |
4684 | } | |
4685 | ||
b53e675d | 4686 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && |
1da177e4 LT |
4687 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
4688 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
4689 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
4690 | hblks++; | |
4691 | xlog_put_bp(hbp); | |
4692 | hbp = xlog_get_bp(log, hblks); | |
4693 | } else { | |
4694 | hblks = 1; | |
4695 | } | |
4696 | } else { | |
69ce58f0 | 4697 | ASSERT(log->l_sectBBsize == 1); |
1da177e4 LT |
4698 | hblks = 1; |
4699 | hbp = xlog_get_bp(log, 1); | |
4700 | h_size = XLOG_BIG_RECORD_BSIZE; | |
4701 | } | |
4702 | ||
4703 | if (!hbp) | |
2451337d | 4704 | return -ENOMEM; |
1da177e4 LT |
4705 | dbp = xlog_get_bp(log, BTOBB(h_size)); |
4706 | if (!dbp) { | |
4707 | xlog_put_bp(hbp); | |
2451337d | 4708 | return -ENOMEM; |
1da177e4 LT |
4709 | } |
4710 | ||
4711 | memset(rhash, 0, sizeof(rhash)); | |
d7f37692 | 4712 | blk_no = rhead_blk = tail_blk; |
970fd3f0 | 4713 | if (tail_blk > head_blk) { |
1da177e4 LT |
4714 | /* |
4715 | * Perform recovery around the end of the physical log. | |
4716 | * When the head is not on the same cycle number as the tail, | |
970fd3f0 | 4717 | * we can't do a sequential recovery. |
1da177e4 | 4718 | */ |
1da177e4 LT |
4719 | while (blk_no < log->l_logBBsize) { |
4720 | /* | |
4721 | * Check for header wrapping around physical end-of-log | |
4722 | */ | |
62926044 | 4723 | offset = hbp->b_addr; |
1da177e4 LT |
4724 | split_hblks = 0; |
4725 | wrapped_hblks = 0; | |
4726 | if (blk_no + hblks <= log->l_logBBsize) { | |
4727 | /* Read header in one read */ | |
076e6acb CH |
4728 | error = xlog_bread(log, blk_no, hblks, hbp, |
4729 | &offset); | |
1da177e4 LT |
4730 | if (error) |
4731 | goto bread_err2; | |
1da177e4 LT |
4732 | } else { |
4733 | /* This LR is split across physical log end */ | |
4734 | if (blk_no != log->l_logBBsize) { | |
4735 | /* some data before physical log end */ | |
4736 | ASSERT(blk_no <= INT_MAX); | |
4737 | split_hblks = log->l_logBBsize - (int)blk_no; | |
4738 | ASSERT(split_hblks > 0); | |
076e6acb CH |
4739 | error = xlog_bread(log, blk_no, |
4740 | split_hblks, hbp, | |
4741 | &offset); | |
4742 | if (error) | |
1da177e4 | 4743 | goto bread_err2; |
1da177e4 | 4744 | } |
076e6acb | 4745 | |
1da177e4 LT |
4746 | /* |
4747 | * Note: this black magic still works with | |
4748 | * large sector sizes (non-512) only because: | |
4749 | * - we increased the buffer size originally | |
4750 | * by 1 sector giving us enough extra space | |
4751 | * for the second read; | |
4752 | * - the log start is guaranteed to be sector | |
4753 | * aligned; | |
4754 | * - we read the log end (LR header start) | |
4755 | * _first_, then the log start (LR header end) | |
4756 | * - order is important. | |
4757 | */ | |
234f56ac | 4758 | wrapped_hblks = hblks - split_hblks; |
44396476 DC |
4759 | error = xlog_bread_offset(log, 0, |
4760 | wrapped_hblks, hbp, | |
4761 | offset + BBTOB(split_hblks)); | |
1da177e4 LT |
4762 | if (error) |
4763 | goto bread_err2; | |
1da177e4 LT |
4764 | } |
4765 | rhead = (xlog_rec_header_t *)offset; | |
4766 | error = xlog_valid_rec_header(log, rhead, | |
4767 | split_hblks ? blk_no : 0); | |
4768 | if (error) | |
4769 | goto bread_err2; | |
4770 | ||
b53e675d | 4771 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
1da177e4 LT |
4772 | blk_no += hblks; |
4773 | ||
4774 | /* Read in data for log record */ | |
4775 | if (blk_no + bblks <= log->l_logBBsize) { | |
076e6acb CH |
4776 | error = xlog_bread(log, blk_no, bblks, dbp, |
4777 | &offset); | |
1da177e4 LT |
4778 | if (error) |
4779 | goto bread_err2; | |
1da177e4 LT |
4780 | } else { |
4781 | /* This log record is split across the | |
4782 | * physical end of log */ | |
62926044 | 4783 | offset = dbp->b_addr; |
1da177e4 LT |
4784 | split_bblks = 0; |
4785 | if (blk_no != log->l_logBBsize) { | |
4786 | /* some data is before the physical | |
4787 | * end of log */ | |
4788 | ASSERT(!wrapped_hblks); | |
4789 | ASSERT(blk_no <= INT_MAX); | |
4790 | split_bblks = | |
4791 | log->l_logBBsize - (int)blk_no; | |
4792 | ASSERT(split_bblks > 0); | |
076e6acb CH |
4793 | error = xlog_bread(log, blk_no, |
4794 | split_bblks, dbp, | |
4795 | &offset); | |
4796 | if (error) | |
1da177e4 | 4797 | goto bread_err2; |
1da177e4 | 4798 | } |
076e6acb | 4799 | |
1da177e4 LT |
4800 | /* |
4801 | * Note: this black magic still works with | |
4802 | * large sector sizes (non-512) only because: | |
4803 | * - we increased the buffer size originally | |
4804 | * by 1 sector giving us enough extra space | |
4805 | * for the second read; | |
4806 | * - the log start is guaranteed to be sector | |
4807 | * aligned; | |
4808 | * - we read the log end (LR header start) | |
4809 | * _first_, then the log start (LR header end) | |
4810 | * - order is important. | |
4811 | */ | |
44396476 | 4812 | error = xlog_bread_offset(log, 0, |
009507b0 | 4813 | bblks - split_bblks, dbp, |
44396476 | 4814 | offset + BBTOB(split_bblks)); |
076e6acb CH |
4815 | if (error) |
4816 | goto bread_err2; | |
1da177e4 | 4817 | } |
0e446be4 | 4818 | |
9d94901f BF |
4819 | error = xlog_recover_process(log, rhash, rhead, offset, |
4820 | pass); | |
0e446be4 | 4821 | if (error) |
1da177e4 | 4822 | goto bread_err2; |
d7f37692 | 4823 | |
1da177e4 | 4824 | blk_no += bblks; |
d7f37692 | 4825 | rhead_blk = blk_no; |
1da177e4 LT |
4826 | } |
4827 | ||
4828 | ASSERT(blk_no >= log->l_logBBsize); | |
4829 | blk_no -= log->l_logBBsize; | |
d7f37692 | 4830 | rhead_blk = blk_no; |
970fd3f0 | 4831 | } |
1da177e4 | 4832 | |
970fd3f0 ES |
4833 | /* read first part of physical log */ |
4834 | while (blk_no < head_blk) { | |
4835 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); | |
4836 | if (error) | |
4837 | goto bread_err2; | |
076e6acb | 4838 | |
970fd3f0 ES |
4839 | rhead = (xlog_rec_header_t *)offset; |
4840 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
4841 | if (error) | |
4842 | goto bread_err2; | |
076e6acb | 4843 | |
970fd3f0 ES |
4844 | /* blocks in data section */ |
4845 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); | |
4846 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, | |
4847 | &offset); | |
4848 | if (error) | |
4849 | goto bread_err2; | |
076e6acb | 4850 | |
9d94901f | 4851 | error = xlog_recover_process(log, rhash, rhead, offset, pass); |
970fd3f0 ES |
4852 | if (error) |
4853 | goto bread_err2; | |
d7f37692 | 4854 | |
970fd3f0 | 4855 | blk_no += bblks + hblks; |
d7f37692 | 4856 | rhead_blk = blk_no; |
1da177e4 LT |
4857 | } |
4858 | ||
4859 | bread_err2: | |
4860 | xlog_put_bp(dbp); | |
4861 | bread_err1: | |
4862 | xlog_put_bp(hbp); | |
d7f37692 BF |
4863 | |
4864 | if (error && first_bad) | |
4865 | *first_bad = rhead_blk; | |
4866 | ||
1da177e4 LT |
4867 | return error; |
4868 | } | |
4869 | ||
4870 | /* | |
4871 | * Do the recovery of the log. We actually do this in two phases. | |
4872 | * The two passes are necessary in order to implement the function | |
4873 | * of cancelling a record written into the log. The first pass | |
4874 | * determines those things which have been cancelled, and the | |
4875 | * second pass replays log items normally except for those which | |
4876 | * have been cancelled. The handling of the replay and cancellations | |
4877 | * takes place in the log item type specific routines. | |
4878 | * | |
4879 | * The table of items which have cancel records in the log is allocated | |
4880 | * and freed at this level, since only here do we know when all of | |
4881 | * the log recovery has been completed. | |
4882 | */ | |
4883 | STATIC int | |
4884 | xlog_do_log_recovery( | |
9a8d2fdb | 4885 | struct xlog *log, |
1da177e4 LT |
4886 | xfs_daddr_t head_blk, |
4887 | xfs_daddr_t tail_blk) | |
4888 | { | |
d5689eaa | 4889 | int error, i; |
1da177e4 LT |
4890 | |
4891 | ASSERT(head_blk != tail_blk); | |
4892 | ||
4893 | /* | |
4894 | * First do a pass to find all of the cancelled buf log items. | |
4895 | * Store them in the buf_cancel_table for use in the second pass. | |
4896 | */ | |
d5689eaa CH |
4897 | log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE * |
4898 | sizeof(struct list_head), | |
1da177e4 | 4899 | KM_SLEEP); |
d5689eaa CH |
4900 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
4901 | INIT_LIST_HEAD(&log->l_buf_cancel_table[i]); | |
4902 | ||
1da177e4 | 4903 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
d7f37692 | 4904 | XLOG_RECOVER_PASS1, NULL); |
1da177e4 | 4905 | if (error != 0) { |
f0e2d93c | 4906 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
4907 | log->l_buf_cancel_table = NULL; |
4908 | return error; | |
4909 | } | |
4910 | /* | |
4911 | * Then do a second pass to actually recover the items in the log. | |
4912 | * When it is complete free the table of buf cancel items. | |
4913 | */ | |
4914 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
d7f37692 | 4915 | XLOG_RECOVER_PASS2, NULL); |
1da177e4 | 4916 | #ifdef DEBUG |
6d192a9b | 4917 | if (!error) { |
1da177e4 LT |
4918 | int i; |
4919 | ||
4920 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
d5689eaa | 4921 | ASSERT(list_empty(&log->l_buf_cancel_table[i])); |
1da177e4 LT |
4922 | } |
4923 | #endif /* DEBUG */ | |
4924 | ||
f0e2d93c | 4925 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
4926 | log->l_buf_cancel_table = NULL; |
4927 | ||
4928 | return error; | |
4929 | } | |
4930 | ||
4931 | /* | |
4932 | * Do the actual recovery | |
4933 | */ | |
4934 | STATIC int | |
4935 | xlog_do_recover( | |
9a8d2fdb | 4936 | struct xlog *log, |
1da177e4 LT |
4937 | xfs_daddr_t head_blk, |
4938 | xfs_daddr_t tail_blk) | |
4939 | { | |
4940 | int error; | |
4941 | xfs_buf_t *bp; | |
4942 | xfs_sb_t *sbp; | |
4943 | ||
4944 | /* | |
4945 | * First replay the images in the log. | |
4946 | */ | |
4947 | error = xlog_do_log_recovery(log, head_blk, tail_blk); | |
43ff2122 | 4948 | if (error) |
1da177e4 | 4949 | return error; |
1da177e4 LT |
4950 | |
4951 | /* | |
4952 | * If IO errors happened during recovery, bail out. | |
4953 | */ | |
4954 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) { | |
2451337d | 4955 | return -EIO; |
1da177e4 LT |
4956 | } |
4957 | ||
4958 | /* | |
4959 | * We now update the tail_lsn since much of the recovery has completed | |
4960 | * and there may be space available to use. If there were no extent | |
4961 | * or iunlinks, we can free up the entire log and set the tail_lsn to | |
4962 | * be the last_sync_lsn. This was set in xlog_find_tail to be the | |
4963 | * lsn of the last known good LR on disk. If there are extent frees | |
4964 | * or iunlinks they will have some entries in the AIL; so we look at | |
4965 | * the AIL to determine how to set the tail_lsn. | |
4966 | */ | |
4967 | xlog_assign_tail_lsn(log->l_mp); | |
4968 | ||
4969 | /* | |
4970 | * Now that we've finished replaying all buffer and inode | |
98021821 | 4971 | * updates, re-read in the superblock and reverify it. |
1da177e4 LT |
4972 | */ |
4973 | bp = xfs_getsb(log->l_mp, 0); | |
4974 | XFS_BUF_UNDONE(bp); | |
bebf963f | 4975 | ASSERT(!(XFS_BUF_ISWRITE(bp))); |
1da177e4 | 4976 | XFS_BUF_READ(bp); |
bebf963f | 4977 | XFS_BUF_UNASYNC(bp); |
1813dd64 | 4978 | bp->b_ops = &xfs_sb_buf_ops; |
83a0adc3 | 4979 | |
595bff75 | 4980 | error = xfs_buf_submit_wait(bp); |
d64e31a2 | 4981 | if (error) { |
595bff75 DC |
4982 | if (!XFS_FORCED_SHUTDOWN(log->l_mp)) { |
4983 | xfs_buf_ioerror_alert(bp, __func__); | |
4984 | ASSERT(0); | |
4985 | } | |
1da177e4 LT |
4986 | xfs_buf_relse(bp); |
4987 | return error; | |
4988 | } | |
4989 | ||
4990 | /* Convert superblock from on-disk format */ | |
4991 | sbp = &log->l_mp->m_sb; | |
98021821 | 4992 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 4993 | ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC); |
62118709 | 4994 | ASSERT(xfs_sb_good_version(sbp)); |
5681ca40 DC |
4995 | xfs_reinit_percpu_counters(log->l_mp); |
4996 | ||
1da177e4 LT |
4997 | xfs_buf_relse(bp); |
4998 | ||
5478eead | 4999 | |
1da177e4 LT |
5000 | xlog_recover_check_summary(log); |
5001 | ||
5002 | /* Normal transactions can now occur */ | |
5003 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
5004 | return 0; | |
5005 | } | |
5006 | ||
5007 | /* | |
5008 | * Perform recovery and re-initialize some log variables in xlog_find_tail. | |
5009 | * | |
5010 | * Return error or zero. | |
5011 | */ | |
5012 | int | |
5013 | xlog_recover( | |
9a8d2fdb | 5014 | struct xlog *log) |
1da177e4 LT |
5015 | { |
5016 | xfs_daddr_t head_blk, tail_blk; | |
5017 | int error; | |
5018 | ||
5019 | /* find the tail of the log */ | |
a45086e2 BF |
5020 | error = xlog_find_tail(log, &head_blk, &tail_blk); |
5021 | if (error) | |
1da177e4 LT |
5022 | return error; |
5023 | ||
a45086e2 BF |
5024 | /* |
5025 | * The superblock was read before the log was available and thus the LSN | |
5026 | * could not be verified. Check the superblock LSN against the current | |
5027 | * LSN now that it's known. | |
5028 | */ | |
5029 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb) && | |
5030 | !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn)) | |
5031 | return -EINVAL; | |
5032 | ||
1da177e4 LT |
5033 | if (tail_blk != head_blk) { |
5034 | /* There used to be a comment here: | |
5035 | * | |
5036 | * disallow recovery on read-only mounts. note -- mount | |
5037 | * checks for ENOSPC and turns it into an intelligent | |
5038 | * error message. | |
5039 | * ...but this is no longer true. Now, unless you specify | |
5040 | * NORECOVERY (in which case this function would never be | |
5041 | * called), we just go ahead and recover. We do this all | |
5042 | * under the vfs layer, so we can get away with it unless | |
5043 | * the device itself is read-only, in which case we fail. | |
5044 | */ | |
3a02ee18 | 5045 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
1da177e4 LT |
5046 | return error; |
5047 | } | |
5048 | ||
e721f504 DC |
5049 | /* |
5050 | * Version 5 superblock log feature mask validation. We know the | |
5051 | * log is dirty so check if there are any unknown log features | |
5052 | * in what we need to recover. If there are unknown features | |
5053 | * (e.g. unsupported transactions, then simply reject the | |
5054 | * attempt at recovery before touching anything. | |
5055 | */ | |
5056 | if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 && | |
5057 | xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb, | |
5058 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) { | |
5059 | xfs_warn(log->l_mp, | |
f41febd2 | 5060 | "Superblock has unknown incompatible log features (0x%x) enabled.", |
e721f504 DC |
5061 | (log->l_mp->m_sb.sb_features_log_incompat & |
5062 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); | |
f41febd2 JP |
5063 | xfs_warn(log->l_mp, |
5064 | "The log can not be fully and/or safely recovered by this kernel."); | |
5065 | xfs_warn(log->l_mp, | |
5066 | "Please recover the log on a kernel that supports the unknown features."); | |
2451337d | 5067 | return -EINVAL; |
e721f504 DC |
5068 | } |
5069 | ||
2e227178 BF |
5070 | /* |
5071 | * Delay log recovery if the debug hook is set. This is debug | |
5072 | * instrumention to coordinate simulation of I/O failures with | |
5073 | * log recovery. | |
5074 | */ | |
5075 | if (xfs_globals.log_recovery_delay) { | |
5076 | xfs_notice(log->l_mp, | |
5077 | "Delaying log recovery for %d seconds.", | |
5078 | xfs_globals.log_recovery_delay); | |
5079 | msleep(xfs_globals.log_recovery_delay * 1000); | |
5080 | } | |
5081 | ||
a0fa2b67 DC |
5082 | xfs_notice(log->l_mp, "Starting recovery (logdev: %s)", |
5083 | log->l_mp->m_logname ? log->l_mp->m_logname | |
5084 | : "internal"); | |
1da177e4 LT |
5085 | |
5086 | error = xlog_do_recover(log, head_blk, tail_blk); | |
5087 | log->l_flags |= XLOG_RECOVERY_NEEDED; | |
5088 | } | |
5089 | return error; | |
5090 | } | |
5091 | ||
5092 | /* | |
5093 | * In the first part of recovery we replay inodes and buffers and build | |
5094 | * up the list of extent free items which need to be processed. Here | |
5095 | * we process the extent free items and clean up the on disk unlinked | |
5096 | * inode lists. This is separated from the first part of recovery so | |
5097 | * that the root and real-time bitmap inodes can be read in from disk in | |
5098 | * between the two stages. This is necessary so that we can free space | |
5099 | * in the real-time portion of the file system. | |
5100 | */ | |
5101 | int | |
5102 | xlog_recover_finish( | |
9a8d2fdb | 5103 | struct xlog *log) |
1da177e4 LT |
5104 | { |
5105 | /* | |
5106 | * Now we're ready to do the transactions needed for the | |
5107 | * rest of recovery. Start with completing all the extent | |
5108 | * free intent records and then process the unlinked inode | |
5109 | * lists. At this point, we essentially run in normal mode | |
5110 | * except that we're still performing recovery actions | |
5111 | * rather than accepting new requests. | |
5112 | */ | |
5113 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { | |
3c1e2bbe DC |
5114 | int error; |
5115 | error = xlog_recover_process_efis(log); | |
5116 | if (error) { | |
a0fa2b67 | 5117 | xfs_alert(log->l_mp, "Failed to recover EFIs"); |
3c1e2bbe DC |
5118 | return error; |
5119 | } | |
1da177e4 LT |
5120 | /* |
5121 | * Sync the log to get all the EFIs out of the AIL. | |
5122 | * This isn't absolutely necessary, but it helps in | |
5123 | * case the unlink transactions would have problems | |
5124 | * pushing the EFIs out of the way. | |
5125 | */ | |
a14a348b | 5126 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
1da177e4 | 5127 | |
4249023a | 5128 | xlog_recover_process_iunlinks(log); |
1da177e4 LT |
5129 | |
5130 | xlog_recover_check_summary(log); | |
5131 | ||
a0fa2b67 DC |
5132 | xfs_notice(log->l_mp, "Ending recovery (logdev: %s)", |
5133 | log->l_mp->m_logname ? log->l_mp->m_logname | |
5134 | : "internal"); | |
1da177e4 LT |
5135 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
5136 | } else { | |
a0fa2b67 | 5137 | xfs_info(log->l_mp, "Ending clean mount"); |
1da177e4 LT |
5138 | } |
5139 | return 0; | |
5140 | } | |
5141 | ||
f0b2efad BF |
5142 | int |
5143 | xlog_recover_cancel( | |
5144 | struct xlog *log) | |
5145 | { | |
5146 | int error = 0; | |
5147 | ||
5148 | if (log->l_flags & XLOG_RECOVERY_NEEDED) | |
5149 | error = xlog_recover_cancel_efis(log); | |
5150 | ||
5151 | return error; | |
5152 | } | |
1da177e4 LT |
5153 | |
5154 | #if defined(DEBUG) | |
5155 | /* | |
5156 | * Read all of the agf and agi counters and check that they | |
5157 | * are consistent with the superblock counters. | |
5158 | */ | |
5159 | void | |
5160 | xlog_recover_check_summary( | |
9a8d2fdb | 5161 | struct xlog *log) |
1da177e4 LT |
5162 | { |
5163 | xfs_mount_t *mp; | |
5164 | xfs_agf_t *agfp; | |
1da177e4 LT |
5165 | xfs_buf_t *agfbp; |
5166 | xfs_buf_t *agibp; | |
1da177e4 LT |
5167 | xfs_agnumber_t agno; |
5168 | __uint64_t freeblks; | |
5169 | __uint64_t itotal; | |
5170 | __uint64_t ifree; | |
5e1be0fb | 5171 | int error; |
1da177e4 LT |
5172 | |
5173 | mp = log->l_mp; | |
5174 | ||
5175 | freeblks = 0LL; | |
5176 | itotal = 0LL; | |
5177 | ifree = 0LL; | |
5178 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
4805621a CH |
5179 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
5180 | if (error) { | |
a0fa2b67 DC |
5181 | xfs_alert(mp, "%s agf read failed agno %d error %d", |
5182 | __func__, agno, error); | |
4805621a CH |
5183 | } else { |
5184 | agfp = XFS_BUF_TO_AGF(agfbp); | |
5185 | freeblks += be32_to_cpu(agfp->agf_freeblks) + | |
5186 | be32_to_cpu(agfp->agf_flcount); | |
5187 | xfs_buf_relse(agfbp); | |
1da177e4 | 5188 | } |
1da177e4 | 5189 | |
5e1be0fb | 5190 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
a0fa2b67 DC |
5191 | if (error) { |
5192 | xfs_alert(mp, "%s agi read failed agno %d error %d", | |
5193 | __func__, agno, error); | |
5194 | } else { | |
5e1be0fb | 5195 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 5196 | |
5e1be0fb CH |
5197 | itotal += be32_to_cpu(agi->agi_count); |
5198 | ifree += be32_to_cpu(agi->agi_freecount); | |
5199 | xfs_buf_relse(agibp); | |
5200 | } | |
1da177e4 | 5201 | } |
1da177e4 LT |
5202 | } |
5203 | #endif /* DEBUG */ |