Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
87c199c2 | 3 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 4 | * All Rights Reserved. |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
70a9883c | 8 | #include "xfs_shared.h" |
239880ef DC |
9 | #include "xfs_format.h" |
10 | #include "xfs_log_format.h" | |
11 | #include "xfs_trans_resv.h" | |
a844f451 | 12 | #include "xfs_bit.h" |
a844f451 | 13 | #include "xfs_sb.h" |
1da177e4 | 14 | #include "xfs_mount.h" |
50995582 | 15 | #include "xfs_defer.h" |
57062787 | 16 | #include "xfs_da_format.h" |
9a2cc41c | 17 | #include "xfs_da_btree.h" |
1da177e4 | 18 | #include "xfs_inode.h" |
239880ef | 19 | #include "xfs_trans.h" |
239880ef | 20 | #include "xfs_log.h" |
1da177e4 | 21 | #include "xfs_log_priv.h" |
1da177e4 | 22 | #include "xfs_log_recover.h" |
a4fbe6ab | 23 | #include "xfs_inode_item.h" |
1da177e4 LT |
24 | #include "xfs_extfree_item.h" |
25 | #include "xfs_trans_priv.h" | |
a4fbe6ab DC |
26 | #include "xfs_alloc.h" |
27 | #include "xfs_ialloc.h" | |
1da177e4 | 28 | #include "xfs_quota.h" |
0e446be4 | 29 | #include "xfs_cksum.h" |
0b1b213f | 30 | #include "xfs_trace.h" |
33479e05 | 31 | #include "xfs_icache.h" |
a4fbe6ab | 32 | #include "xfs_bmap_btree.h" |
a4fbe6ab | 33 | #include "xfs_error.h" |
2b9ab5ab | 34 | #include "xfs_dir2.h" |
9e88b5d8 | 35 | #include "xfs_rmap_item.h" |
60a4a222 | 36 | #include "xfs_buf_item.h" |
f997ee21 | 37 | #include "xfs_refcount_item.h" |
77d61fe4 | 38 | #include "xfs_bmap_item.h" |
1da177e4 | 39 | |
fc06c6d0 DC |
40 | #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1) |
41 | ||
9a8d2fdb MT |
42 | STATIC int |
43 | xlog_find_zeroed( | |
44 | struct xlog *, | |
45 | xfs_daddr_t *); | |
46 | STATIC int | |
47 | xlog_clear_stale_blocks( | |
48 | struct xlog *, | |
49 | xfs_lsn_t); | |
1da177e4 | 50 | #if defined(DEBUG) |
9a8d2fdb MT |
51 | STATIC void |
52 | xlog_recover_check_summary( | |
53 | struct xlog *); | |
1da177e4 LT |
54 | #else |
55 | #define xlog_recover_check_summary(log) | |
1da177e4 | 56 | #endif |
7088c413 BF |
57 | STATIC int |
58 | xlog_do_recovery_pass( | |
59 | struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *); | |
1da177e4 | 60 | |
d5689eaa CH |
61 | /* |
62 | * This structure is used during recovery to record the buf log items which | |
63 | * have been canceled and should not be replayed. | |
64 | */ | |
65 | struct xfs_buf_cancel { | |
66 | xfs_daddr_t bc_blkno; | |
67 | uint bc_len; | |
68 | int bc_refcount; | |
69 | struct list_head bc_list; | |
70 | }; | |
71 | ||
1da177e4 LT |
72 | /* |
73 | * Sector aligned buffer routines for buffer create/read/write/access | |
74 | */ | |
75 | ||
ff30a622 | 76 | /* |
99c26595 BF |
77 | * Verify the log-relative block number and length in basic blocks are valid for |
78 | * an operation involving the given XFS log buffer. Returns true if the fields | |
79 | * are valid, false otherwise. | |
ff30a622 | 80 | */ |
99c26595 BF |
81 | static inline bool |
82 | xlog_verify_bp( | |
9a8d2fdb | 83 | struct xlog *log, |
99c26595 | 84 | xfs_daddr_t blk_no, |
ff30a622 AE |
85 | int bbcount) |
86 | { | |
99c26595 BF |
87 | if (blk_no < 0 || blk_no >= log->l_logBBsize) |
88 | return false; | |
89 | if (bbcount <= 0 || (blk_no + bbcount) > log->l_logBBsize) | |
90 | return false; | |
91 | return true; | |
ff30a622 AE |
92 | } |
93 | ||
36adecff AE |
94 | /* |
95 | * Allocate a buffer to hold log data. The buffer needs to be able | |
96 | * to map to a range of nbblks basic blocks at any valid (basic | |
97 | * block) offset within the log. | |
98 | */ | |
5d77c0dc | 99 | STATIC xfs_buf_t * |
1da177e4 | 100 | xlog_get_bp( |
9a8d2fdb | 101 | struct xlog *log, |
3228149c | 102 | int nbblks) |
1da177e4 | 103 | { |
c8da0faf CH |
104 | struct xfs_buf *bp; |
105 | ||
99c26595 BF |
106 | /* |
107 | * Pass log block 0 since we don't have an addr yet, buffer will be | |
108 | * verified on read. | |
109 | */ | |
110 | if (!xlog_verify_bp(log, 0, nbblks)) { | |
a0fa2b67 | 111 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
ff30a622 AE |
112 | nbblks); |
113 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
3228149c DC |
114 | return NULL; |
115 | } | |
1da177e4 | 116 | |
36adecff AE |
117 | /* |
118 | * We do log I/O in units of log sectors (a power-of-2 | |
119 | * multiple of the basic block size), so we round up the | |
25985edc | 120 | * requested size to accommodate the basic blocks required |
36adecff AE |
121 | * for complete log sectors. |
122 | * | |
123 | * In addition, the buffer may be used for a non-sector- | |
124 | * aligned block offset, in which case an I/O of the | |
125 | * requested size could extend beyond the end of the | |
126 | * buffer. If the requested size is only 1 basic block it | |
127 | * will never straddle a sector boundary, so this won't be | |
128 | * an issue. Nor will this be a problem if the log I/O is | |
129 | * done in basic blocks (sector size 1). But otherwise we | |
130 | * extend the buffer by one extra log sector to ensure | |
25985edc | 131 | * there's space to accommodate this possibility. |
36adecff | 132 | */ |
69ce58f0 AE |
133 | if (nbblks > 1 && log->l_sectBBsize > 1) |
134 | nbblks += log->l_sectBBsize; | |
135 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
36adecff | 136 | |
e70b73f8 | 137 | bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0); |
c8da0faf CH |
138 | if (bp) |
139 | xfs_buf_unlock(bp); | |
140 | return bp; | |
1da177e4 LT |
141 | } |
142 | ||
5d77c0dc | 143 | STATIC void |
1da177e4 LT |
144 | xlog_put_bp( |
145 | xfs_buf_t *bp) | |
146 | { | |
147 | xfs_buf_free(bp); | |
148 | } | |
149 | ||
48389ef1 AE |
150 | /* |
151 | * Return the address of the start of the given block number's data | |
152 | * in a log buffer. The buffer covers a log sector-aligned region. | |
153 | */ | |
b2a922cd | 154 | STATIC char * |
076e6acb | 155 | xlog_align( |
9a8d2fdb | 156 | struct xlog *log, |
076e6acb CH |
157 | xfs_daddr_t blk_no, |
158 | int nbblks, | |
9a8d2fdb | 159 | struct xfs_buf *bp) |
076e6acb | 160 | { |
fdc07f44 | 161 | xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1); |
076e6acb | 162 | |
4e94b71b | 163 | ASSERT(offset + nbblks <= bp->b_length); |
62926044 | 164 | return bp->b_addr + BBTOB(offset); |
076e6acb CH |
165 | } |
166 | ||
1da177e4 LT |
167 | |
168 | /* | |
169 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. | |
170 | */ | |
076e6acb CH |
171 | STATIC int |
172 | xlog_bread_noalign( | |
9a8d2fdb | 173 | struct xlog *log, |
1da177e4 LT |
174 | xfs_daddr_t blk_no, |
175 | int nbblks, | |
9a8d2fdb | 176 | struct xfs_buf *bp) |
1da177e4 LT |
177 | { |
178 | int error; | |
179 | ||
99c26595 BF |
180 | if (!xlog_verify_bp(log, blk_no, nbblks)) { |
181 | xfs_warn(log->l_mp, | |
182 | "Invalid log block/length (0x%llx, 0x%x) for buffer", | |
183 | blk_no, nbblks); | |
ff30a622 | 184 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
2451337d | 185 | return -EFSCORRUPTED; |
3228149c DC |
186 | } |
187 | ||
69ce58f0 AE |
188 | blk_no = round_down(blk_no, log->l_sectBBsize); |
189 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
190 | |
191 | ASSERT(nbblks > 0); | |
4e94b71b | 192 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
193 | |
194 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
0cac682f | 195 | bp->b_flags |= XBF_READ; |
aa0e8833 | 196 | bp->b_io_length = nbblks; |
0e95f19a | 197 | bp->b_error = 0; |
1da177e4 | 198 | |
595bff75 DC |
199 | error = xfs_buf_submit_wait(bp); |
200 | if (error && !XFS_FORCED_SHUTDOWN(log->l_mp)) | |
901796af | 201 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
202 | return error; |
203 | } | |
204 | ||
076e6acb CH |
205 | STATIC int |
206 | xlog_bread( | |
9a8d2fdb | 207 | struct xlog *log, |
076e6acb CH |
208 | xfs_daddr_t blk_no, |
209 | int nbblks, | |
9a8d2fdb | 210 | struct xfs_buf *bp, |
b2a922cd | 211 | char **offset) |
076e6acb CH |
212 | { |
213 | int error; | |
214 | ||
215 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
216 | if (error) | |
217 | return error; | |
218 | ||
219 | *offset = xlog_align(log, blk_no, nbblks, bp); | |
220 | return 0; | |
221 | } | |
222 | ||
44396476 DC |
223 | /* |
224 | * Read at an offset into the buffer. Returns with the buffer in it's original | |
225 | * state regardless of the result of the read. | |
226 | */ | |
227 | STATIC int | |
228 | xlog_bread_offset( | |
9a8d2fdb | 229 | struct xlog *log, |
44396476 DC |
230 | xfs_daddr_t blk_no, /* block to read from */ |
231 | int nbblks, /* blocks to read */ | |
9a8d2fdb | 232 | struct xfs_buf *bp, |
b2a922cd | 233 | char *offset) |
44396476 | 234 | { |
b2a922cd | 235 | char *orig_offset = bp->b_addr; |
4e94b71b | 236 | int orig_len = BBTOB(bp->b_length); |
44396476 DC |
237 | int error, error2; |
238 | ||
02fe03d9 | 239 | error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks)); |
44396476 DC |
240 | if (error) |
241 | return error; | |
242 | ||
243 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
244 | ||
245 | /* must reset buffer pointer even on error */ | |
02fe03d9 | 246 | error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len); |
44396476 DC |
247 | if (error) |
248 | return error; | |
249 | return error2; | |
250 | } | |
251 | ||
1da177e4 LT |
252 | /* |
253 | * Write out the buffer at the given block for the given number of blocks. | |
254 | * The buffer is kept locked across the write and is returned locked. | |
255 | * This can only be used for synchronous log writes. | |
256 | */ | |
ba0f32d4 | 257 | STATIC int |
1da177e4 | 258 | xlog_bwrite( |
9a8d2fdb | 259 | struct xlog *log, |
1da177e4 LT |
260 | xfs_daddr_t blk_no, |
261 | int nbblks, | |
9a8d2fdb | 262 | struct xfs_buf *bp) |
1da177e4 LT |
263 | { |
264 | int error; | |
265 | ||
99c26595 BF |
266 | if (!xlog_verify_bp(log, blk_no, nbblks)) { |
267 | xfs_warn(log->l_mp, | |
268 | "Invalid log block/length (0x%llx, 0x%x) for buffer", | |
269 | blk_no, nbblks); | |
ff30a622 | 270 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
2451337d | 271 | return -EFSCORRUPTED; |
3228149c DC |
272 | } |
273 | ||
69ce58f0 AE |
274 | blk_no = round_down(blk_no, log->l_sectBBsize); |
275 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
1da177e4 LT |
276 | |
277 | ASSERT(nbblks > 0); | |
4e94b71b | 278 | ASSERT(nbblks <= bp->b_length); |
1da177e4 LT |
279 | |
280 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
72790aa1 | 281 | xfs_buf_hold(bp); |
0c842ad4 | 282 | xfs_buf_lock(bp); |
aa0e8833 | 283 | bp->b_io_length = nbblks; |
0e95f19a | 284 | bp->b_error = 0; |
1da177e4 | 285 | |
c2b006c1 | 286 | error = xfs_bwrite(bp); |
901796af CH |
287 | if (error) |
288 | xfs_buf_ioerror_alert(bp, __func__); | |
c2b006c1 | 289 | xfs_buf_relse(bp); |
1da177e4 LT |
290 | return error; |
291 | } | |
292 | ||
1da177e4 LT |
293 | #ifdef DEBUG |
294 | /* | |
295 | * dump debug superblock and log record information | |
296 | */ | |
297 | STATIC void | |
298 | xlog_header_check_dump( | |
299 | xfs_mount_t *mp, | |
300 | xlog_rec_header_t *head) | |
301 | { | |
08e96e1a | 302 | xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d", |
03daa57c | 303 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); |
08e96e1a | 304 | xfs_debug(mp, " log : uuid = %pU, fmt = %d", |
03daa57c | 305 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); |
1da177e4 LT |
306 | } |
307 | #else | |
308 | #define xlog_header_check_dump(mp, head) | |
309 | #endif | |
310 | ||
311 | /* | |
312 | * check log record header for recovery | |
313 | */ | |
314 | STATIC int | |
315 | xlog_header_check_recover( | |
316 | xfs_mount_t *mp, | |
317 | xlog_rec_header_t *head) | |
318 | { | |
69ef921b | 319 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 LT |
320 | |
321 | /* | |
322 | * IRIX doesn't write the h_fmt field and leaves it zeroed | |
323 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover | |
324 | * a dirty log created in IRIX. | |
325 | */ | |
69ef921b | 326 | if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) { |
a0fa2b67 DC |
327 | xfs_warn(mp, |
328 | "dirty log written in incompatible format - can't recover"); | |
1da177e4 LT |
329 | xlog_header_check_dump(mp, head); |
330 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", | |
331 | XFS_ERRLEVEL_HIGH, mp); | |
2451337d | 332 | return -EFSCORRUPTED; |
1da177e4 | 333 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
a0fa2b67 DC |
334 | xfs_warn(mp, |
335 | "dirty log entry has mismatched uuid - can't recover"); | |
1da177e4 LT |
336 | xlog_header_check_dump(mp, head); |
337 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", | |
338 | XFS_ERRLEVEL_HIGH, mp); | |
2451337d | 339 | return -EFSCORRUPTED; |
1da177e4 LT |
340 | } |
341 | return 0; | |
342 | } | |
343 | ||
344 | /* | |
345 | * read the head block of the log and check the header | |
346 | */ | |
347 | STATIC int | |
348 | xlog_header_check_mount( | |
349 | xfs_mount_t *mp, | |
350 | xlog_rec_header_t *head) | |
351 | { | |
69ef921b | 352 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
1da177e4 | 353 | |
d905fdaa | 354 | if (uuid_is_null(&head->h_fs_uuid)) { |
1da177e4 LT |
355 | /* |
356 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If | |
d905fdaa | 357 | * h_fs_uuid is null, we assume this log was last mounted |
1da177e4 LT |
358 | * by IRIX and continue. |
359 | */ | |
d905fdaa | 360 | xfs_warn(mp, "null uuid in log - IRIX style log"); |
1da177e4 | 361 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
a0fa2b67 | 362 | xfs_warn(mp, "log has mismatched uuid - can't recover"); |
1da177e4 LT |
363 | xlog_header_check_dump(mp, head); |
364 | XFS_ERROR_REPORT("xlog_header_check_mount", | |
365 | XFS_ERRLEVEL_HIGH, mp); | |
2451337d | 366 | return -EFSCORRUPTED; |
1da177e4 LT |
367 | } |
368 | return 0; | |
369 | } | |
370 | ||
371 | STATIC void | |
372 | xlog_recover_iodone( | |
373 | struct xfs_buf *bp) | |
374 | { | |
5a52c2a5 | 375 | if (bp->b_error) { |
1da177e4 LT |
376 | /* |
377 | * We're not going to bother about retrying | |
378 | * this during recovery. One strike! | |
379 | */ | |
595bff75 DC |
380 | if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
381 | xfs_buf_ioerror_alert(bp, __func__); | |
382 | xfs_force_shutdown(bp->b_target->bt_mount, | |
383 | SHUTDOWN_META_IO_ERROR); | |
384 | } | |
1da177e4 | 385 | } |
60a4a222 BF |
386 | |
387 | /* | |
388 | * On v5 supers, a bli could be attached to update the metadata LSN. | |
389 | * Clean it up. | |
390 | */ | |
fb1755a6 | 391 | if (bp->b_log_item) |
60a4a222 | 392 | xfs_buf_item_relse(bp); |
fb1755a6 | 393 | ASSERT(bp->b_log_item == NULL); |
60a4a222 | 394 | |
cb669ca5 | 395 | bp->b_iodone = NULL; |
e8aaba9a | 396 | xfs_buf_ioend(bp); |
1da177e4 LT |
397 | } |
398 | ||
399 | /* | |
400 | * This routine finds (to an approximation) the first block in the physical | |
401 | * log which contains the given cycle. It uses a binary search algorithm. | |
402 | * Note that the algorithm can not be perfect because the disk will not | |
403 | * necessarily be perfect. | |
404 | */ | |
a8272ce0 | 405 | STATIC int |
1da177e4 | 406 | xlog_find_cycle_start( |
9a8d2fdb MT |
407 | struct xlog *log, |
408 | struct xfs_buf *bp, | |
1da177e4 LT |
409 | xfs_daddr_t first_blk, |
410 | xfs_daddr_t *last_blk, | |
411 | uint cycle) | |
412 | { | |
b2a922cd | 413 | char *offset; |
1da177e4 | 414 | xfs_daddr_t mid_blk; |
e3bb2e30 | 415 | xfs_daddr_t end_blk; |
1da177e4 LT |
416 | uint mid_cycle; |
417 | int error; | |
418 | ||
e3bb2e30 AE |
419 | end_blk = *last_blk; |
420 | mid_blk = BLK_AVG(first_blk, end_blk); | |
421 | while (mid_blk != first_blk && mid_blk != end_blk) { | |
076e6acb CH |
422 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
423 | if (error) | |
1da177e4 | 424 | return error; |
03bea6fe | 425 | mid_cycle = xlog_get_cycle(offset); |
e3bb2e30 AE |
426 | if (mid_cycle == cycle) |
427 | end_blk = mid_blk; /* last_half_cycle == mid_cycle */ | |
428 | else | |
429 | first_blk = mid_blk; /* first_half_cycle == mid_cycle */ | |
430 | mid_blk = BLK_AVG(first_blk, end_blk); | |
1da177e4 | 431 | } |
e3bb2e30 AE |
432 | ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) || |
433 | (mid_blk == end_blk && mid_blk-1 == first_blk)); | |
434 | ||
435 | *last_blk = end_blk; | |
1da177e4 LT |
436 | |
437 | return 0; | |
438 | } | |
439 | ||
440 | /* | |
3f943d85 AE |
441 | * Check that a range of blocks does not contain stop_on_cycle_no. |
442 | * Fill in *new_blk with the block offset where such a block is | |
443 | * found, or with -1 (an invalid block number) if there is no such | |
444 | * block in the range. The scan needs to occur from front to back | |
445 | * and the pointer into the region must be updated since a later | |
446 | * routine will need to perform another test. | |
1da177e4 LT |
447 | */ |
448 | STATIC int | |
449 | xlog_find_verify_cycle( | |
9a8d2fdb | 450 | struct xlog *log, |
1da177e4 LT |
451 | xfs_daddr_t start_blk, |
452 | int nbblks, | |
453 | uint stop_on_cycle_no, | |
454 | xfs_daddr_t *new_blk) | |
455 | { | |
456 | xfs_daddr_t i, j; | |
457 | uint cycle; | |
458 | xfs_buf_t *bp; | |
459 | xfs_daddr_t bufblks; | |
b2a922cd | 460 | char *buf = NULL; |
1da177e4 LT |
461 | int error = 0; |
462 | ||
6881a229 AE |
463 | /* |
464 | * Greedily allocate a buffer big enough to handle the full | |
465 | * range of basic blocks we'll be examining. If that fails, | |
466 | * try a smaller size. We need to be able to read at least | |
467 | * a log sector, or we're out of luck. | |
468 | */ | |
1da177e4 | 469 | bufblks = 1 << ffs(nbblks); |
81158e0c DC |
470 | while (bufblks > log->l_logBBsize) |
471 | bufblks >>= 1; | |
1da177e4 | 472 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1da177e4 | 473 | bufblks >>= 1; |
69ce58f0 | 474 | if (bufblks < log->l_sectBBsize) |
2451337d | 475 | return -ENOMEM; |
1da177e4 LT |
476 | } |
477 | ||
478 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { | |
479 | int bcount; | |
480 | ||
481 | bcount = min(bufblks, (start_blk + nbblks - i)); | |
482 | ||
076e6acb CH |
483 | error = xlog_bread(log, i, bcount, bp, &buf); |
484 | if (error) | |
1da177e4 LT |
485 | goto out; |
486 | ||
1da177e4 | 487 | for (j = 0; j < bcount; j++) { |
03bea6fe | 488 | cycle = xlog_get_cycle(buf); |
1da177e4 LT |
489 | if (cycle == stop_on_cycle_no) { |
490 | *new_blk = i+j; | |
491 | goto out; | |
492 | } | |
493 | ||
494 | buf += BBSIZE; | |
495 | } | |
496 | } | |
497 | ||
498 | *new_blk = -1; | |
499 | ||
500 | out: | |
501 | xlog_put_bp(bp); | |
502 | return error; | |
503 | } | |
504 | ||
505 | /* | |
506 | * Potentially backup over partial log record write. | |
507 | * | |
508 | * In the typical case, last_blk is the number of the block directly after | |
509 | * a good log record. Therefore, we subtract one to get the block number | |
510 | * of the last block in the given buffer. extra_bblks contains the number | |
511 | * of blocks we would have read on a previous read. This happens when the | |
512 | * last log record is split over the end of the physical log. | |
513 | * | |
514 | * extra_bblks is the number of blocks potentially verified on a previous | |
515 | * call to this routine. | |
516 | */ | |
517 | STATIC int | |
518 | xlog_find_verify_log_record( | |
9a8d2fdb | 519 | struct xlog *log, |
1da177e4 LT |
520 | xfs_daddr_t start_blk, |
521 | xfs_daddr_t *last_blk, | |
522 | int extra_bblks) | |
523 | { | |
524 | xfs_daddr_t i; | |
525 | xfs_buf_t *bp; | |
b2a922cd | 526 | char *offset = NULL; |
1da177e4 LT |
527 | xlog_rec_header_t *head = NULL; |
528 | int error = 0; | |
529 | int smallmem = 0; | |
530 | int num_blks = *last_blk - start_blk; | |
531 | int xhdrs; | |
532 | ||
533 | ASSERT(start_blk != 0 || *last_blk != start_blk); | |
534 | ||
535 | if (!(bp = xlog_get_bp(log, num_blks))) { | |
536 | if (!(bp = xlog_get_bp(log, 1))) | |
2451337d | 537 | return -ENOMEM; |
1da177e4 LT |
538 | smallmem = 1; |
539 | } else { | |
076e6acb CH |
540 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
541 | if (error) | |
1da177e4 | 542 | goto out; |
1da177e4 LT |
543 | offset += ((num_blks - 1) << BBSHIFT); |
544 | } | |
545 | ||
546 | for (i = (*last_blk) - 1; i >= 0; i--) { | |
547 | if (i < start_blk) { | |
548 | /* valid log record not found */ | |
a0fa2b67 DC |
549 | xfs_warn(log->l_mp, |
550 | "Log inconsistent (didn't find previous header)"); | |
1da177e4 | 551 | ASSERT(0); |
2451337d | 552 | error = -EIO; |
1da177e4 LT |
553 | goto out; |
554 | } | |
555 | ||
556 | if (smallmem) { | |
076e6acb CH |
557 | error = xlog_bread(log, i, 1, bp, &offset); |
558 | if (error) | |
1da177e4 | 559 | goto out; |
1da177e4 LT |
560 | } |
561 | ||
562 | head = (xlog_rec_header_t *)offset; | |
563 | ||
69ef921b | 564 | if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
1da177e4 LT |
565 | break; |
566 | ||
567 | if (!smallmem) | |
568 | offset -= BBSIZE; | |
569 | } | |
570 | ||
571 | /* | |
572 | * We hit the beginning of the physical log & still no header. Return | |
573 | * to caller. If caller can handle a return of -1, then this routine | |
574 | * will be called again for the end of the physical log. | |
575 | */ | |
576 | if (i == -1) { | |
2451337d | 577 | error = 1; |
1da177e4 LT |
578 | goto out; |
579 | } | |
580 | ||
581 | /* | |
582 | * We have the final block of the good log (the first block | |
583 | * of the log record _before_ the head. So we check the uuid. | |
584 | */ | |
585 | if ((error = xlog_header_check_mount(log->l_mp, head))) | |
586 | goto out; | |
587 | ||
588 | /* | |
589 | * We may have found a log record header before we expected one. | |
590 | * last_blk will be the 1st block # with a given cycle #. We may end | |
591 | * up reading an entire log record. In this case, we don't want to | |
592 | * reset last_blk. Only when last_blk points in the middle of a log | |
593 | * record do we update last_blk. | |
594 | */ | |
62118709 | 595 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b53e675d | 596 | uint h_size = be32_to_cpu(head->h_size); |
1da177e4 LT |
597 | |
598 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; | |
599 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
600 | xhdrs++; | |
601 | } else { | |
602 | xhdrs = 1; | |
603 | } | |
604 | ||
b53e675d CH |
605 | if (*last_blk - i + extra_bblks != |
606 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) | |
1da177e4 LT |
607 | *last_blk = i; |
608 | ||
609 | out: | |
610 | xlog_put_bp(bp); | |
611 | return error; | |
612 | } | |
613 | ||
614 | /* | |
615 | * Head is defined to be the point of the log where the next log write | |
0a94da24 | 616 | * could go. This means that incomplete LR writes at the end are |
1da177e4 LT |
617 | * eliminated when calculating the head. We aren't guaranteed that previous |
618 | * LR have complete transactions. We only know that a cycle number of | |
619 | * current cycle number -1 won't be present in the log if we start writing | |
620 | * from our current block number. | |
621 | * | |
622 | * last_blk contains the block number of the first block with a given | |
623 | * cycle number. | |
624 | * | |
625 | * Return: zero if normal, non-zero if error. | |
626 | */ | |
ba0f32d4 | 627 | STATIC int |
1da177e4 | 628 | xlog_find_head( |
9a8d2fdb | 629 | struct xlog *log, |
1da177e4 LT |
630 | xfs_daddr_t *return_head_blk) |
631 | { | |
632 | xfs_buf_t *bp; | |
b2a922cd | 633 | char *offset; |
1da177e4 LT |
634 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; |
635 | int num_scan_bblks; | |
636 | uint first_half_cycle, last_half_cycle; | |
637 | uint stop_on_cycle; | |
638 | int error, log_bbnum = log->l_logBBsize; | |
639 | ||
640 | /* Is the end of the log device zeroed? */ | |
2451337d DC |
641 | error = xlog_find_zeroed(log, &first_blk); |
642 | if (error < 0) { | |
643 | xfs_warn(log->l_mp, "empty log check failed"); | |
644 | return error; | |
645 | } | |
646 | if (error == 1) { | |
1da177e4 LT |
647 | *return_head_blk = first_blk; |
648 | ||
649 | /* Is the whole lot zeroed? */ | |
650 | if (!first_blk) { | |
651 | /* Linux XFS shouldn't generate totally zeroed logs - | |
652 | * mkfs etc write a dummy unmount record to a fresh | |
653 | * log so we can store the uuid in there | |
654 | */ | |
a0fa2b67 | 655 | xfs_warn(log->l_mp, "totally zeroed log"); |
1da177e4 LT |
656 | } |
657 | ||
658 | return 0; | |
1da177e4 LT |
659 | } |
660 | ||
661 | first_blk = 0; /* get cycle # of 1st block */ | |
662 | bp = xlog_get_bp(log, 1); | |
663 | if (!bp) | |
2451337d | 664 | return -ENOMEM; |
076e6acb CH |
665 | |
666 | error = xlog_bread(log, 0, 1, bp, &offset); | |
667 | if (error) | |
1da177e4 | 668 | goto bp_err; |
076e6acb | 669 | |
03bea6fe | 670 | first_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
671 | |
672 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ | |
076e6acb CH |
673 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
674 | if (error) | |
1da177e4 | 675 | goto bp_err; |
076e6acb | 676 | |
03bea6fe | 677 | last_half_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
678 | ASSERT(last_half_cycle != 0); |
679 | ||
680 | /* | |
681 | * If the 1st half cycle number is equal to the last half cycle number, | |
682 | * then the entire log is stamped with the same cycle number. In this | |
683 | * case, head_blk can't be set to zero (which makes sense). The below | |
684 | * math doesn't work out properly with head_blk equal to zero. Instead, | |
685 | * we set it to log_bbnum which is an invalid block number, but this | |
686 | * value makes the math correct. If head_blk doesn't changed through | |
687 | * all the tests below, *head_blk is set to zero at the very end rather | |
688 | * than log_bbnum. In a sense, log_bbnum and zero are the same block | |
689 | * in a circular file. | |
690 | */ | |
691 | if (first_half_cycle == last_half_cycle) { | |
692 | /* | |
693 | * In this case we believe that the entire log should have | |
694 | * cycle number last_half_cycle. We need to scan backwards | |
695 | * from the end verifying that there are no holes still | |
696 | * containing last_half_cycle - 1. If we find such a hole, | |
697 | * then the start of that hole will be the new head. The | |
698 | * simple case looks like | |
699 | * x | x ... | x - 1 | x | |
700 | * Another case that fits this picture would be | |
701 | * x | x + 1 | x ... | x | |
c41564b5 | 702 | * In this case the head really is somewhere at the end of the |
1da177e4 LT |
703 | * log, as one of the latest writes at the beginning was |
704 | * incomplete. | |
705 | * One more case is | |
706 | * x | x + 1 | x ... | x - 1 | x | |
707 | * This is really the combination of the above two cases, and | |
708 | * the head has to end up at the start of the x-1 hole at the | |
709 | * end of the log. | |
710 | * | |
711 | * In the 256k log case, we will read from the beginning to the | |
712 | * end of the log and search for cycle numbers equal to x-1. | |
713 | * We don't worry about the x+1 blocks that we encounter, | |
714 | * because we know that they cannot be the head since the log | |
715 | * started with x. | |
716 | */ | |
717 | head_blk = log_bbnum; | |
718 | stop_on_cycle = last_half_cycle - 1; | |
719 | } else { | |
720 | /* | |
721 | * In this case we want to find the first block with cycle | |
722 | * number matching last_half_cycle. We expect the log to be | |
723 | * some variation on | |
3f943d85 | 724 | * x + 1 ... | x ... | x |
1da177e4 LT |
725 | * The first block with cycle number x (last_half_cycle) will |
726 | * be where the new head belongs. First we do a binary search | |
727 | * for the first occurrence of last_half_cycle. The binary | |
728 | * search may not be totally accurate, so then we scan back | |
729 | * from there looking for occurrences of last_half_cycle before | |
730 | * us. If that backwards scan wraps around the beginning of | |
731 | * the log, then we look for occurrences of last_half_cycle - 1 | |
732 | * at the end of the log. The cases we're looking for look | |
733 | * like | |
3f943d85 AE |
734 | * v binary search stopped here |
735 | * x + 1 ... | x | x + 1 | x ... | x | |
736 | * ^ but we want to locate this spot | |
1da177e4 | 737 | * or |
1da177e4 | 738 | * <---------> less than scan distance |
3f943d85 AE |
739 | * x + 1 ... | x ... | x - 1 | x |
740 | * ^ we want to locate this spot | |
1da177e4 LT |
741 | */ |
742 | stop_on_cycle = last_half_cycle; | |
743 | if ((error = xlog_find_cycle_start(log, bp, first_blk, | |
744 | &head_blk, last_half_cycle))) | |
745 | goto bp_err; | |
746 | } | |
747 | ||
748 | /* | |
749 | * Now validate the answer. Scan back some number of maximum possible | |
750 | * blocks and make sure each one has the expected cycle number. The | |
751 | * maximum is determined by the total possible amount of buffering | |
752 | * in the in-core log. The following number can be made tighter if | |
753 | * we actually look at the block size of the filesystem. | |
754 | */ | |
9f2a4505 | 755 | num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log)); |
1da177e4 LT |
756 | if (head_blk >= num_scan_bblks) { |
757 | /* | |
758 | * We are guaranteed that the entire check can be performed | |
759 | * in one buffer. | |
760 | */ | |
761 | start_blk = head_blk - num_scan_bblks; | |
762 | if ((error = xlog_find_verify_cycle(log, | |
763 | start_blk, num_scan_bblks, | |
764 | stop_on_cycle, &new_blk))) | |
765 | goto bp_err; | |
766 | if (new_blk != -1) | |
767 | head_blk = new_blk; | |
768 | } else { /* need to read 2 parts of log */ | |
769 | /* | |
770 | * We are going to scan backwards in the log in two parts. | |
771 | * First we scan the physical end of the log. In this part | |
772 | * of the log, we are looking for blocks with cycle number | |
773 | * last_half_cycle - 1. | |
774 | * If we find one, then we know that the log starts there, as | |
775 | * we've found a hole that didn't get written in going around | |
776 | * the end of the physical log. The simple case for this is | |
777 | * x + 1 ... | x ... | x - 1 | x | |
778 | * <---------> less than scan distance | |
779 | * If all of the blocks at the end of the log have cycle number | |
780 | * last_half_cycle, then we check the blocks at the start of | |
781 | * the log looking for occurrences of last_half_cycle. If we | |
782 | * find one, then our current estimate for the location of the | |
783 | * first occurrence of last_half_cycle is wrong and we move | |
784 | * back to the hole we've found. This case looks like | |
785 | * x + 1 ... | x | x + 1 | x ... | |
786 | * ^ binary search stopped here | |
787 | * Another case we need to handle that only occurs in 256k | |
788 | * logs is | |
789 | * x + 1 ... | x ... | x+1 | x ... | |
790 | * ^ binary search stops here | |
791 | * In a 256k log, the scan at the end of the log will see the | |
792 | * x + 1 blocks. We need to skip past those since that is | |
793 | * certainly not the head of the log. By searching for | |
794 | * last_half_cycle-1 we accomplish that. | |
795 | */ | |
1da177e4 | 796 | ASSERT(head_blk <= INT_MAX && |
3f943d85 AE |
797 | (xfs_daddr_t) num_scan_bblks >= head_blk); |
798 | start_blk = log_bbnum - (num_scan_bblks - head_blk); | |
1da177e4 LT |
799 | if ((error = xlog_find_verify_cycle(log, start_blk, |
800 | num_scan_bblks - (int)head_blk, | |
801 | (stop_on_cycle - 1), &new_blk))) | |
802 | goto bp_err; | |
803 | if (new_blk != -1) { | |
804 | head_blk = new_blk; | |
9db127ed | 805 | goto validate_head; |
1da177e4 LT |
806 | } |
807 | ||
808 | /* | |
809 | * Scan beginning of log now. The last part of the physical | |
810 | * log is good. This scan needs to verify that it doesn't find | |
811 | * the last_half_cycle. | |
812 | */ | |
813 | start_blk = 0; | |
814 | ASSERT(head_blk <= INT_MAX); | |
815 | if ((error = xlog_find_verify_cycle(log, | |
816 | start_blk, (int)head_blk, | |
817 | stop_on_cycle, &new_blk))) | |
818 | goto bp_err; | |
819 | if (new_blk != -1) | |
820 | head_blk = new_blk; | |
821 | } | |
822 | ||
9db127ed | 823 | validate_head: |
1da177e4 LT |
824 | /* |
825 | * Now we need to make sure head_blk is not pointing to a block in | |
826 | * the middle of a log record. | |
827 | */ | |
828 | num_scan_bblks = XLOG_REC_SHIFT(log); | |
829 | if (head_blk >= num_scan_bblks) { | |
830 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ | |
831 | ||
832 | /* start ptr at last block ptr before head_blk */ | |
2451337d DC |
833 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
834 | if (error == 1) | |
835 | error = -EIO; | |
836 | if (error) | |
1da177e4 LT |
837 | goto bp_err; |
838 | } else { | |
839 | start_blk = 0; | |
840 | ASSERT(head_blk <= INT_MAX); | |
2451337d DC |
841 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
842 | if (error < 0) | |
843 | goto bp_err; | |
844 | if (error == 1) { | |
1da177e4 | 845 | /* We hit the beginning of the log during our search */ |
3f943d85 | 846 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
1da177e4 LT |
847 | new_blk = log_bbnum; |
848 | ASSERT(start_blk <= INT_MAX && | |
849 | (xfs_daddr_t) log_bbnum-start_blk >= 0); | |
850 | ASSERT(head_blk <= INT_MAX); | |
2451337d DC |
851 | error = xlog_find_verify_log_record(log, start_blk, |
852 | &new_blk, (int)head_blk); | |
853 | if (error == 1) | |
854 | error = -EIO; | |
855 | if (error) | |
1da177e4 LT |
856 | goto bp_err; |
857 | if (new_blk != log_bbnum) | |
858 | head_blk = new_blk; | |
859 | } else if (error) | |
860 | goto bp_err; | |
861 | } | |
862 | ||
863 | xlog_put_bp(bp); | |
864 | if (head_blk == log_bbnum) | |
865 | *return_head_blk = 0; | |
866 | else | |
867 | *return_head_blk = head_blk; | |
868 | /* | |
869 | * When returning here, we have a good block number. Bad block | |
870 | * means that during a previous crash, we didn't have a clean break | |
871 | * from cycle number N to cycle number N-1. In this case, we need | |
872 | * to find the first block with cycle number N-1. | |
873 | */ | |
874 | return 0; | |
875 | ||
876 | bp_err: | |
877 | xlog_put_bp(bp); | |
878 | ||
879 | if (error) | |
a0fa2b67 | 880 | xfs_warn(log->l_mp, "failed to find log head"); |
1da177e4 LT |
881 | return error; |
882 | } | |
883 | ||
eed6b462 BF |
884 | /* |
885 | * Seek backwards in the log for log record headers. | |
886 | * | |
887 | * Given a starting log block, walk backwards until we find the provided number | |
888 | * of records or hit the provided tail block. The return value is the number of | |
889 | * records encountered or a negative error code. The log block and buffer | |
890 | * pointer of the last record seen are returned in rblk and rhead respectively. | |
891 | */ | |
892 | STATIC int | |
893 | xlog_rseek_logrec_hdr( | |
894 | struct xlog *log, | |
895 | xfs_daddr_t head_blk, | |
896 | xfs_daddr_t tail_blk, | |
897 | int count, | |
898 | struct xfs_buf *bp, | |
899 | xfs_daddr_t *rblk, | |
900 | struct xlog_rec_header **rhead, | |
901 | bool *wrapped) | |
902 | { | |
903 | int i; | |
904 | int error; | |
905 | int found = 0; | |
906 | char *offset = NULL; | |
907 | xfs_daddr_t end_blk; | |
908 | ||
909 | *wrapped = false; | |
910 | ||
911 | /* | |
912 | * Walk backwards from the head block until we hit the tail or the first | |
913 | * block in the log. | |
914 | */ | |
915 | end_blk = head_blk > tail_blk ? tail_blk : 0; | |
916 | for (i = (int) head_blk - 1; i >= end_blk; i--) { | |
917 | error = xlog_bread(log, i, 1, bp, &offset); | |
918 | if (error) | |
919 | goto out_error; | |
920 | ||
921 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
922 | *rblk = i; | |
923 | *rhead = (struct xlog_rec_header *) offset; | |
924 | if (++found == count) | |
925 | break; | |
926 | } | |
927 | } | |
928 | ||
929 | /* | |
930 | * If we haven't hit the tail block or the log record header count, | |
931 | * start looking again from the end of the physical log. Note that | |
932 | * callers can pass head == tail if the tail is not yet known. | |
933 | */ | |
934 | if (tail_blk >= head_blk && found != count) { | |
935 | for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) { | |
936 | error = xlog_bread(log, i, 1, bp, &offset); | |
937 | if (error) | |
938 | goto out_error; | |
939 | ||
940 | if (*(__be32 *)offset == | |
941 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
942 | *wrapped = true; | |
943 | *rblk = i; | |
944 | *rhead = (struct xlog_rec_header *) offset; | |
945 | if (++found == count) | |
946 | break; | |
947 | } | |
948 | } | |
949 | } | |
950 | ||
951 | return found; | |
952 | ||
953 | out_error: | |
954 | return error; | |
955 | } | |
956 | ||
7088c413 BF |
957 | /* |
958 | * Seek forward in the log for log record headers. | |
959 | * | |
960 | * Given head and tail blocks, walk forward from the tail block until we find | |
961 | * the provided number of records or hit the head block. The return value is the | |
962 | * number of records encountered or a negative error code. The log block and | |
963 | * buffer pointer of the last record seen are returned in rblk and rhead | |
964 | * respectively. | |
965 | */ | |
966 | STATIC int | |
967 | xlog_seek_logrec_hdr( | |
968 | struct xlog *log, | |
969 | xfs_daddr_t head_blk, | |
970 | xfs_daddr_t tail_blk, | |
971 | int count, | |
972 | struct xfs_buf *bp, | |
973 | xfs_daddr_t *rblk, | |
974 | struct xlog_rec_header **rhead, | |
975 | bool *wrapped) | |
976 | { | |
977 | int i; | |
978 | int error; | |
979 | int found = 0; | |
980 | char *offset = NULL; | |
981 | xfs_daddr_t end_blk; | |
982 | ||
983 | *wrapped = false; | |
984 | ||
985 | /* | |
986 | * Walk forward from the tail block until we hit the head or the last | |
987 | * block in the log. | |
988 | */ | |
989 | end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1; | |
990 | for (i = (int) tail_blk; i <= end_blk; i++) { | |
991 | error = xlog_bread(log, i, 1, bp, &offset); | |
992 | if (error) | |
993 | goto out_error; | |
994 | ||
995 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
996 | *rblk = i; | |
997 | *rhead = (struct xlog_rec_header *) offset; | |
998 | if (++found == count) | |
999 | break; | |
1000 | } | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * If we haven't hit the head block or the log record header count, | |
1005 | * start looking again from the start of the physical log. | |
1006 | */ | |
1007 | if (tail_blk > head_blk && found != count) { | |
1008 | for (i = 0; i < (int) head_blk; i++) { | |
1009 | error = xlog_bread(log, i, 1, bp, &offset); | |
1010 | if (error) | |
1011 | goto out_error; | |
1012 | ||
1013 | if (*(__be32 *)offset == | |
1014 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
1015 | *wrapped = true; | |
1016 | *rblk = i; | |
1017 | *rhead = (struct xlog_rec_header *) offset; | |
1018 | if (++found == count) | |
1019 | break; | |
1020 | } | |
1021 | } | |
1022 | } | |
1023 | ||
1024 | return found; | |
1025 | ||
1026 | out_error: | |
1027 | return error; | |
1028 | } | |
1029 | ||
1030 | /* | |
4a4f66ea BF |
1031 | * Calculate distance from head to tail (i.e., unused space in the log). |
1032 | */ | |
1033 | static inline int | |
1034 | xlog_tail_distance( | |
1035 | struct xlog *log, | |
1036 | xfs_daddr_t head_blk, | |
1037 | xfs_daddr_t tail_blk) | |
1038 | { | |
1039 | if (head_blk < tail_blk) | |
1040 | return tail_blk - head_blk; | |
1041 | ||
1042 | return tail_blk + (log->l_logBBsize - head_blk); | |
1043 | } | |
1044 | ||
1045 | /* | |
1046 | * Verify the log tail. This is particularly important when torn or incomplete | |
1047 | * writes have been detected near the front of the log and the head has been | |
1048 | * walked back accordingly. | |
1049 | * | |
1050 | * We also have to handle the case where the tail was pinned and the head | |
1051 | * blocked behind the tail right before a crash. If the tail had been pushed | |
1052 | * immediately prior to the crash and the subsequent checkpoint was only | |
1053 | * partially written, it's possible it overwrote the last referenced tail in the | |
1054 | * log with garbage. This is not a coherency problem because the tail must have | |
1055 | * been pushed before it can be overwritten, but appears as log corruption to | |
1056 | * recovery because we have no way to know the tail was updated if the | |
1057 | * subsequent checkpoint didn't write successfully. | |
7088c413 | 1058 | * |
4a4f66ea BF |
1059 | * Therefore, CRC check the log from tail to head. If a failure occurs and the |
1060 | * offending record is within max iclog bufs from the head, walk the tail | |
1061 | * forward and retry until a valid tail is found or corruption is detected out | |
1062 | * of the range of a possible overwrite. | |
7088c413 BF |
1063 | */ |
1064 | STATIC int | |
1065 | xlog_verify_tail( | |
1066 | struct xlog *log, | |
1067 | xfs_daddr_t head_blk, | |
4a4f66ea BF |
1068 | xfs_daddr_t *tail_blk, |
1069 | int hsize) | |
7088c413 BF |
1070 | { |
1071 | struct xlog_rec_header *thead; | |
1072 | struct xfs_buf *bp; | |
1073 | xfs_daddr_t first_bad; | |
7088c413 BF |
1074 | int error = 0; |
1075 | bool wrapped; | |
4a4f66ea BF |
1076 | xfs_daddr_t tmp_tail; |
1077 | xfs_daddr_t orig_tail = *tail_blk; | |
7088c413 BF |
1078 | |
1079 | bp = xlog_get_bp(log, 1); | |
1080 | if (!bp) | |
1081 | return -ENOMEM; | |
1082 | ||
1083 | /* | |
4a4f66ea BF |
1084 | * Make sure the tail points to a record (returns positive count on |
1085 | * success). | |
7088c413 | 1086 | */ |
4a4f66ea BF |
1087 | error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, bp, |
1088 | &tmp_tail, &thead, &wrapped); | |
1089 | if (error < 0) | |
7088c413 | 1090 | goto out; |
4a4f66ea BF |
1091 | if (*tail_blk != tmp_tail) |
1092 | *tail_blk = tmp_tail; | |
7088c413 BF |
1093 | |
1094 | /* | |
4a4f66ea BF |
1095 | * Run a CRC check from the tail to the head. We can't just check |
1096 | * MAX_ICLOGS records past the tail because the tail may point to stale | |
1097 | * blocks cleared during the search for the head/tail. These blocks are | |
1098 | * overwritten with zero-length records and thus record count is not a | |
1099 | * reliable indicator of the iclog state before a crash. | |
7088c413 | 1100 | */ |
4a4f66ea BF |
1101 | first_bad = 0; |
1102 | error = xlog_do_recovery_pass(log, head_blk, *tail_blk, | |
7088c413 | 1103 | XLOG_RECOVER_CRCPASS, &first_bad); |
a4c9b34d | 1104 | while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) { |
4a4f66ea BF |
1105 | int tail_distance; |
1106 | ||
1107 | /* | |
1108 | * Is corruption within range of the head? If so, retry from | |
1109 | * the next record. Otherwise return an error. | |
1110 | */ | |
1111 | tail_distance = xlog_tail_distance(log, head_blk, first_bad); | |
1112 | if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize)) | |
1113 | break; | |
7088c413 | 1114 | |
4a4f66ea BF |
1115 | /* skip to the next record; returns positive count on success */ |
1116 | error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2, bp, | |
1117 | &tmp_tail, &thead, &wrapped); | |
1118 | if (error < 0) | |
1119 | goto out; | |
1120 | ||
1121 | *tail_blk = tmp_tail; | |
1122 | first_bad = 0; | |
1123 | error = xlog_do_recovery_pass(log, head_blk, *tail_blk, | |
1124 | XLOG_RECOVER_CRCPASS, &first_bad); | |
1125 | } | |
1126 | ||
1127 | if (!error && *tail_blk != orig_tail) | |
1128 | xfs_warn(log->l_mp, | |
1129 | "Tail block (0x%llx) overwrite detected. Updated to 0x%llx", | |
1130 | orig_tail, *tail_blk); | |
7088c413 BF |
1131 | out: |
1132 | xlog_put_bp(bp); | |
1133 | return error; | |
1134 | } | |
1135 | ||
1136 | /* | |
1137 | * Detect and trim torn writes from the head of the log. | |
1138 | * | |
1139 | * Storage without sector atomicity guarantees can result in torn writes in the | |
1140 | * log in the event of a crash. Our only means to detect this scenario is via | |
1141 | * CRC verification. While we can't always be certain that CRC verification | |
1142 | * failure is due to a torn write vs. an unrelated corruption, we do know that | |
1143 | * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at | |
1144 | * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of | |
1145 | * the log and treat failures in this range as torn writes as a matter of | |
1146 | * policy. In the event of CRC failure, the head is walked back to the last good | |
1147 | * record in the log and the tail is updated from that record and verified. | |
1148 | */ | |
1149 | STATIC int | |
1150 | xlog_verify_head( | |
1151 | struct xlog *log, | |
1152 | xfs_daddr_t *head_blk, /* in/out: unverified head */ | |
1153 | xfs_daddr_t *tail_blk, /* out: tail block */ | |
1154 | struct xfs_buf *bp, | |
1155 | xfs_daddr_t *rhead_blk, /* start blk of last record */ | |
1156 | struct xlog_rec_header **rhead, /* ptr to last record */ | |
1157 | bool *wrapped) /* last rec. wraps phys. log */ | |
1158 | { | |
1159 | struct xlog_rec_header *tmp_rhead; | |
1160 | struct xfs_buf *tmp_bp; | |
1161 | xfs_daddr_t first_bad; | |
1162 | xfs_daddr_t tmp_rhead_blk; | |
1163 | int found; | |
1164 | int error; | |
1165 | bool tmp_wrapped; | |
1166 | ||
1167 | /* | |
82ff6cc2 BF |
1168 | * Check the head of the log for torn writes. Search backwards from the |
1169 | * head until we hit the tail or the maximum number of log record I/Os | |
1170 | * that could have been in flight at one time. Use a temporary buffer so | |
1171 | * we don't trash the rhead/bp pointers from the caller. | |
7088c413 BF |
1172 | */ |
1173 | tmp_bp = xlog_get_bp(log, 1); | |
1174 | if (!tmp_bp) | |
1175 | return -ENOMEM; | |
1176 | error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk, | |
1177 | XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk, | |
1178 | &tmp_rhead, &tmp_wrapped); | |
1179 | xlog_put_bp(tmp_bp); | |
1180 | if (error < 0) | |
1181 | return error; | |
1182 | ||
1183 | /* | |
1184 | * Now run a CRC verification pass over the records starting at the | |
1185 | * block found above to the current head. If a CRC failure occurs, the | |
1186 | * log block of the first bad record is saved in first_bad. | |
1187 | */ | |
1188 | error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk, | |
1189 | XLOG_RECOVER_CRCPASS, &first_bad); | |
a4c9b34d | 1190 | if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) { |
7088c413 BF |
1191 | /* |
1192 | * We've hit a potential torn write. Reset the error and warn | |
1193 | * about it. | |
1194 | */ | |
1195 | error = 0; | |
1196 | xfs_warn(log->l_mp, | |
1197 | "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.", | |
1198 | first_bad, *head_blk); | |
1199 | ||
1200 | /* | |
1201 | * Get the header block and buffer pointer for the last good | |
1202 | * record before the bad record. | |
1203 | * | |
1204 | * Note that xlog_find_tail() clears the blocks at the new head | |
1205 | * (i.e., the records with invalid CRC) if the cycle number | |
1206 | * matches the the current cycle. | |
1207 | */ | |
1208 | found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp, | |
1209 | rhead_blk, rhead, wrapped); | |
1210 | if (found < 0) | |
1211 | return found; | |
1212 | if (found == 0) /* XXX: right thing to do here? */ | |
1213 | return -EIO; | |
1214 | ||
1215 | /* | |
1216 | * Reset the head block to the starting block of the first bad | |
1217 | * log record and set the tail block based on the last good | |
1218 | * record. | |
1219 | * | |
1220 | * Bail out if the updated head/tail match as this indicates | |
1221 | * possible corruption outside of the acceptable | |
1222 | * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair... | |
1223 | */ | |
1224 | *head_blk = first_bad; | |
1225 | *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn)); | |
1226 | if (*head_blk == *tail_blk) { | |
1227 | ASSERT(0); | |
1228 | return 0; | |
1229 | } | |
7088c413 | 1230 | } |
5297ac1f BF |
1231 | if (error) |
1232 | return error; | |
7088c413 | 1233 | |
4a4f66ea BF |
1234 | return xlog_verify_tail(log, *head_blk, tail_blk, |
1235 | be32_to_cpu((*rhead)->h_size)); | |
7088c413 BF |
1236 | } |
1237 | ||
0703a8e1 DC |
1238 | /* |
1239 | * We need to make sure we handle log wrapping properly, so we can't use the | |
1240 | * calculated logbno directly. Make sure it wraps to the correct bno inside the | |
1241 | * log. | |
1242 | * | |
1243 | * The log is limited to 32 bit sizes, so we use the appropriate modulus | |
1244 | * operation here and cast it back to a 64 bit daddr on return. | |
1245 | */ | |
1246 | static inline xfs_daddr_t | |
1247 | xlog_wrap_logbno( | |
1248 | struct xlog *log, | |
1249 | xfs_daddr_t bno) | |
1250 | { | |
1251 | int mod; | |
1252 | ||
1253 | div_s64_rem(bno, log->l_logBBsize, &mod); | |
1254 | return mod; | |
1255 | } | |
1256 | ||
65b99a08 BF |
1257 | /* |
1258 | * Check whether the head of the log points to an unmount record. In other | |
1259 | * words, determine whether the log is clean. If so, update the in-core state | |
1260 | * appropriately. | |
1261 | */ | |
1262 | static int | |
1263 | xlog_check_unmount_rec( | |
1264 | struct xlog *log, | |
1265 | xfs_daddr_t *head_blk, | |
1266 | xfs_daddr_t *tail_blk, | |
1267 | struct xlog_rec_header *rhead, | |
1268 | xfs_daddr_t rhead_blk, | |
1269 | struct xfs_buf *bp, | |
1270 | bool *clean) | |
1271 | { | |
1272 | struct xlog_op_header *op_head; | |
1273 | xfs_daddr_t umount_data_blk; | |
1274 | xfs_daddr_t after_umount_blk; | |
1275 | int hblks; | |
1276 | int error; | |
1277 | char *offset; | |
1278 | ||
1279 | *clean = false; | |
1280 | ||
1281 | /* | |
1282 | * Look for unmount record. If we find it, then we know there was a | |
1283 | * clean unmount. Since 'i' could be the last block in the physical | |
1284 | * log, we convert to a log block before comparing to the head_blk. | |
1285 | * | |
1286 | * Save the current tail lsn to use to pass to xlog_clear_stale_blocks() | |
1287 | * below. We won't want to clear the unmount record if there is one, so | |
1288 | * we pass the lsn of the unmount record rather than the block after it. | |
1289 | */ | |
1290 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1291 | int h_size = be32_to_cpu(rhead->h_size); | |
1292 | int h_version = be32_to_cpu(rhead->h_version); | |
1293 | ||
1294 | if ((h_version & XLOG_VERSION_2) && | |
1295 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
1296 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
1297 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
1298 | hblks++; | |
1299 | } else { | |
1300 | hblks = 1; | |
1301 | } | |
1302 | } else { | |
1303 | hblks = 1; | |
1304 | } | |
0703a8e1 DC |
1305 | |
1306 | after_umount_blk = xlog_wrap_logbno(log, | |
1307 | rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len))); | |
1308 | ||
65b99a08 BF |
1309 | if (*head_blk == after_umount_blk && |
1310 | be32_to_cpu(rhead->h_num_logops) == 1) { | |
0703a8e1 | 1311 | umount_data_blk = xlog_wrap_logbno(log, rhead_blk + hblks); |
65b99a08 BF |
1312 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); |
1313 | if (error) | |
1314 | return error; | |
1315 | ||
1316 | op_head = (struct xlog_op_header *)offset; | |
1317 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { | |
1318 | /* | |
1319 | * Set tail and last sync so that newly written log | |
1320 | * records will point recovery to after the current | |
1321 | * unmount record. | |
1322 | */ | |
1323 | xlog_assign_atomic_lsn(&log->l_tail_lsn, | |
1324 | log->l_curr_cycle, after_umount_blk); | |
1325 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, | |
1326 | log->l_curr_cycle, after_umount_blk); | |
1327 | *tail_blk = after_umount_blk; | |
1328 | ||
1329 | *clean = true; | |
1330 | } | |
1331 | } | |
1332 | ||
1333 | return 0; | |
1334 | } | |
1335 | ||
717bc0eb BF |
1336 | static void |
1337 | xlog_set_state( | |
1338 | struct xlog *log, | |
1339 | xfs_daddr_t head_blk, | |
1340 | struct xlog_rec_header *rhead, | |
1341 | xfs_daddr_t rhead_blk, | |
1342 | bool bump_cycle) | |
1343 | { | |
1344 | /* | |
1345 | * Reset log values according to the state of the log when we | |
1346 | * crashed. In the case where head_blk == 0, we bump curr_cycle | |
1347 | * one because the next write starts a new cycle rather than | |
1348 | * continuing the cycle of the last good log record. At this | |
1349 | * point we have guaranteed that all partial log records have been | |
1350 | * accounted for. Therefore, we know that the last good log record | |
1351 | * written was complete and ended exactly on the end boundary | |
1352 | * of the physical log. | |
1353 | */ | |
1354 | log->l_prev_block = rhead_blk; | |
1355 | log->l_curr_block = (int)head_blk; | |
1356 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); | |
1357 | if (bump_cycle) | |
1358 | log->l_curr_cycle++; | |
1359 | atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn)); | |
1360 | atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn)); | |
1361 | xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle, | |
1362 | BBTOB(log->l_curr_block)); | |
1363 | xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle, | |
1364 | BBTOB(log->l_curr_block)); | |
1365 | } | |
1366 | ||
1da177e4 LT |
1367 | /* |
1368 | * Find the sync block number or the tail of the log. | |
1369 | * | |
1370 | * This will be the block number of the last record to have its | |
1371 | * associated buffers synced to disk. Every log record header has | |
1372 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy | |
1373 | * to get a sync block number. The only concern is to figure out which | |
1374 | * log record header to believe. | |
1375 | * | |
1376 | * The following algorithm uses the log record header with the largest | |
1377 | * lsn. The entire log record does not need to be valid. We only care | |
1378 | * that the header is valid. | |
1379 | * | |
1380 | * We could speed up search by using current head_blk buffer, but it is not | |
1381 | * available. | |
1382 | */ | |
5d77c0dc | 1383 | STATIC int |
1da177e4 | 1384 | xlog_find_tail( |
9a8d2fdb | 1385 | struct xlog *log, |
1da177e4 | 1386 | xfs_daddr_t *head_blk, |
65be6054 | 1387 | xfs_daddr_t *tail_blk) |
1da177e4 LT |
1388 | { |
1389 | xlog_rec_header_t *rhead; | |
b2a922cd | 1390 | char *offset = NULL; |
1da177e4 | 1391 | xfs_buf_t *bp; |
7088c413 | 1392 | int error; |
7088c413 | 1393 | xfs_daddr_t rhead_blk; |
1da177e4 | 1394 | xfs_lsn_t tail_lsn; |
eed6b462 | 1395 | bool wrapped = false; |
65b99a08 | 1396 | bool clean = false; |
1da177e4 LT |
1397 | |
1398 | /* | |
1399 | * Find previous log record | |
1400 | */ | |
1401 | if ((error = xlog_find_head(log, head_blk))) | |
1402 | return error; | |
82ff6cc2 | 1403 | ASSERT(*head_blk < INT_MAX); |
1da177e4 LT |
1404 | |
1405 | bp = xlog_get_bp(log, 1); | |
1406 | if (!bp) | |
2451337d | 1407 | return -ENOMEM; |
1da177e4 | 1408 | if (*head_blk == 0) { /* special case */ |
076e6acb CH |
1409 | error = xlog_bread(log, 0, 1, bp, &offset); |
1410 | if (error) | |
9db127ed | 1411 | goto done; |
076e6acb | 1412 | |
03bea6fe | 1413 | if (xlog_get_cycle(offset) == 0) { |
1da177e4 LT |
1414 | *tail_blk = 0; |
1415 | /* leave all other log inited values alone */ | |
9db127ed | 1416 | goto done; |
1da177e4 LT |
1417 | } |
1418 | } | |
1419 | ||
1420 | /* | |
82ff6cc2 BF |
1421 | * Search backwards through the log looking for the log record header |
1422 | * block. This wraps all the way back around to the head so something is | |
1423 | * seriously wrong if we can't find it. | |
1da177e4 | 1424 | */ |
82ff6cc2 BF |
1425 | error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp, |
1426 | &rhead_blk, &rhead, &wrapped); | |
1427 | if (error < 0) | |
1428 | return error; | |
1429 | if (!error) { | |
1430 | xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__); | |
1431 | return -EIO; | |
1432 | } | |
1433 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); | |
1da177e4 LT |
1434 | |
1435 | /* | |
717bc0eb | 1436 | * Set the log state based on the current head record. |
1da177e4 | 1437 | */ |
717bc0eb | 1438 | xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped); |
65b99a08 | 1439 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
1da177e4 LT |
1440 | |
1441 | /* | |
65b99a08 BF |
1442 | * Look for an unmount record at the head of the log. This sets the log |
1443 | * state to determine whether recovery is necessary. | |
1da177e4 | 1444 | */ |
65b99a08 BF |
1445 | error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead, |
1446 | rhead_blk, bp, &clean); | |
1447 | if (error) | |
1448 | goto done; | |
1da177e4 LT |
1449 | |
1450 | /* | |
7f6aff3a BF |
1451 | * Verify the log head if the log is not clean (e.g., we have anything |
1452 | * but an unmount record at the head). This uses CRC verification to | |
1453 | * detect and trim torn writes. If discovered, CRC failures are | |
1454 | * considered torn writes and the log head is trimmed accordingly. | |
1da177e4 | 1455 | * |
7f6aff3a BF |
1456 | * Note that we can only run CRC verification when the log is dirty |
1457 | * because there's no guarantee that the log data behind an unmount | |
1458 | * record is compatible with the current architecture. | |
1da177e4 | 1459 | */ |
7f6aff3a BF |
1460 | if (!clean) { |
1461 | xfs_daddr_t orig_head = *head_blk; | |
1da177e4 | 1462 | |
7f6aff3a BF |
1463 | error = xlog_verify_head(log, head_blk, tail_blk, bp, |
1464 | &rhead_blk, &rhead, &wrapped); | |
076e6acb | 1465 | if (error) |
9db127ed | 1466 | goto done; |
076e6acb | 1467 | |
7f6aff3a BF |
1468 | /* update in-core state again if the head changed */ |
1469 | if (*head_blk != orig_head) { | |
1470 | xlog_set_state(log, *head_blk, rhead, rhead_blk, | |
1471 | wrapped); | |
1472 | tail_lsn = atomic64_read(&log->l_tail_lsn); | |
1473 | error = xlog_check_unmount_rec(log, head_blk, tail_blk, | |
1474 | rhead, rhead_blk, bp, | |
1475 | &clean); | |
1476 | if (error) | |
1477 | goto done; | |
1da177e4 LT |
1478 | } |
1479 | } | |
1480 | ||
65b99a08 BF |
1481 | /* |
1482 | * Note that the unmount was clean. If the unmount was not clean, we | |
1483 | * need to know this to rebuild the superblock counters from the perag | |
1484 | * headers if we have a filesystem using non-persistent counters. | |
1485 | */ | |
1486 | if (clean) | |
1487 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; | |
1da177e4 LT |
1488 | |
1489 | /* | |
1490 | * Make sure that there are no blocks in front of the head | |
1491 | * with the same cycle number as the head. This can happen | |
1492 | * because we allow multiple outstanding log writes concurrently, | |
1493 | * and the later writes might make it out before earlier ones. | |
1494 | * | |
1495 | * We use the lsn from before modifying it so that we'll never | |
1496 | * overwrite the unmount record after a clean unmount. | |
1497 | * | |
1498 | * Do this only if we are going to recover the filesystem | |
1499 | * | |
1500 | * NOTE: This used to say "if (!readonly)" | |
1501 | * However on Linux, we can & do recover a read-only filesystem. | |
1502 | * We only skip recovery if NORECOVERY is specified on mount, | |
1503 | * in which case we would not be here. | |
1504 | * | |
1505 | * But... if the -device- itself is readonly, just skip this. | |
1506 | * We can't recover this device anyway, so it won't matter. | |
1507 | */ | |
9db127ed | 1508 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) |
1da177e4 | 1509 | error = xlog_clear_stale_blocks(log, tail_lsn); |
1da177e4 | 1510 | |
9db127ed | 1511 | done: |
1da177e4 LT |
1512 | xlog_put_bp(bp); |
1513 | ||
1514 | if (error) | |
a0fa2b67 | 1515 | xfs_warn(log->l_mp, "failed to locate log tail"); |
1da177e4 LT |
1516 | return error; |
1517 | } | |
1518 | ||
1519 | /* | |
1520 | * Is the log zeroed at all? | |
1521 | * | |
1522 | * The last binary search should be changed to perform an X block read | |
1523 | * once X becomes small enough. You can then search linearly through | |
1524 | * the X blocks. This will cut down on the number of reads we need to do. | |
1525 | * | |
1526 | * If the log is partially zeroed, this routine will pass back the blkno | |
1527 | * of the first block with cycle number 0. It won't have a complete LR | |
1528 | * preceding it. | |
1529 | * | |
1530 | * Return: | |
1531 | * 0 => the log is completely written to | |
2451337d DC |
1532 | * 1 => use *blk_no as the first block of the log |
1533 | * <0 => error has occurred | |
1da177e4 | 1534 | */ |
a8272ce0 | 1535 | STATIC int |
1da177e4 | 1536 | xlog_find_zeroed( |
9a8d2fdb | 1537 | struct xlog *log, |
1da177e4 LT |
1538 | xfs_daddr_t *blk_no) |
1539 | { | |
1540 | xfs_buf_t *bp; | |
b2a922cd | 1541 | char *offset; |
1da177e4 LT |
1542 | uint first_cycle, last_cycle; |
1543 | xfs_daddr_t new_blk, last_blk, start_blk; | |
1544 | xfs_daddr_t num_scan_bblks; | |
1545 | int error, log_bbnum = log->l_logBBsize; | |
1546 | ||
6fdf8ccc NS |
1547 | *blk_no = 0; |
1548 | ||
1da177e4 LT |
1549 | /* check totally zeroed log */ |
1550 | bp = xlog_get_bp(log, 1); | |
1551 | if (!bp) | |
2451337d | 1552 | return -ENOMEM; |
076e6acb CH |
1553 | error = xlog_bread(log, 0, 1, bp, &offset); |
1554 | if (error) | |
1da177e4 | 1555 | goto bp_err; |
076e6acb | 1556 | |
03bea6fe | 1557 | first_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1558 | if (first_cycle == 0) { /* completely zeroed log */ |
1559 | *blk_no = 0; | |
1560 | xlog_put_bp(bp); | |
2451337d | 1561 | return 1; |
1da177e4 LT |
1562 | } |
1563 | ||
1564 | /* check partially zeroed log */ | |
076e6acb CH |
1565 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
1566 | if (error) | |
1da177e4 | 1567 | goto bp_err; |
076e6acb | 1568 | |
03bea6fe | 1569 | last_cycle = xlog_get_cycle(offset); |
1da177e4 LT |
1570 | if (last_cycle != 0) { /* log completely written to */ |
1571 | xlog_put_bp(bp); | |
1572 | return 0; | |
1573 | } else if (first_cycle != 1) { | |
1574 | /* | |
1575 | * If the cycle of the last block is zero, the cycle of | |
1576 | * the first block must be 1. If it's not, maybe we're | |
1577 | * not looking at a log... Bail out. | |
1578 | */ | |
a0fa2b67 DC |
1579 | xfs_warn(log->l_mp, |
1580 | "Log inconsistent or not a log (last==0, first!=1)"); | |
2451337d | 1581 | error = -EINVAL; |
5d0a6549 | 1582 | goto bp_err; |
1da177e4 LT |
1583 | } |
1584 | ||
1585 | /* we have a partially zeroed log */ | |
1586 | last_blk = log_bbnum-1; | |
1587 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) | |
1588 | goto bp_err; | |
1589 | ||
1590 | /* | |
1591 | * Validate the answer. Because there is no way to guarantee that | |
1592 | * the entire log is made up of log records which are the same size, | |
1593 | * we scan over the defined maximum blocks. At this point, the maximum | |
1594 | * is not chosen to mean anything special. XXXmiken | |
1595 | */ | |
1596 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
1597 | ASSERT(num_scan_bblks <= INT_MAX); | |
1598 | ||
1599 | if (last_blk < num_scan_bblks) | |
1600 | num_scan_bblks = last_blk; | |
1601 | start_blk = last_blk - num_scan_bblks; | |
1602 | ||
1603 | /* | |
1604 | * We search for any instances of cycle number 0 that occur before | |
1605 | * our current estimate of the head. What we're trying to detect is | |
1606 | * 1 ... | 0 | 1 | 0... | |
1607 | * ^ binary search ends here | |
1608 | */ | |
1609 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
1610 | (int)num_scan_bblks, 0, &new_blk))) | |
1611 | goto bp_err; | |
1612 | if (new_blk != -1) | |
1613 | last_blk = new_blk; | |
1614 | ||
1615 | /* | |
1616 | * Potentially backup over partial log record write. We don't need | |
1617 | * to search the end of the log because we know it is zero. | |
1618 | */ | |
2451337d DC |
1619 | error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0); |
1620 | if (error == 1) | |
1621 | error = -EIO; | |
1622 | if (error) | |
1623 | goto bp_err; | |
1da177e4 LT |
1624 | |
1625 | *blk_no = last_blk; | |
1626 | bp_err: | |
1627 | xlog_put_bp(bp); | |
1628 | if (error) | |
1629 | return error; | |
2451337d | 1630 | return 1; |
1da177e4 LT |
1631 | } |
1632 | ||
1633 | /* | |
1634 | * These are simple subroutines used by xlog_clear_stale_blocks() below | |
1635 | * to initialize a buffer full of empty log record headers and write | |
1636 | * them into the log. | |
1637 | */ | |
1638 | STATIC void | |
1639 | xlog_add_record( | |
9a8d2fdb | 1640 | struct xlog *log, |
b2a922cd | 1641 | char *buf, |
1da177e4 LT |
1642 | int cycle, |
1643 | int block, | |
1644 | int tail_cycle, | |
1645 | int tail_block) | |
1646 | { | |
1647 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; | |
1648 | ||
1649 | memset(buf, 0, BBSIZE); | |
b53e675d CH |
1650 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1651 | recp->h_cycle = cpu_to_be32(cycle); | |
1652 | recp->h_version = cpu_to_be32( | |
62118709 | 1653 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d CH |
1654 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
1655 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); | |
1656 | recp->h_fmt = cpu_to_be32(XLOG_FMT); | |
1da177e4 LT |
1657 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1658 | } | |
1659 | ||
1660 | STATIC int | |
1661 | xlog_write_log_records( | |
9a8d2fdb | 1662 | struct xlog *log, |
1da177e4 LT |
1663 | int cycle, |
1664 | int start_block, | |
1665 | int blocks, | |
1666 | int tail_cycle, | |
1667 | int tail_block) | |
1668 | { | |
b2a922cd | 1669 | char *offset; |
1da177e4 LT |
1670 | xfs_buf_t *bp; |
1671 | int balign, ealign; | |
69ce58f0 | 1672 | int sectbb = log->l_sectBBsize; |
1da177e4 LT |
1673 | int end_block = start_block + blocks; |
1674 | int bufblks; | |
1675 | int error = 0; | |
1676 | int i, j = 0; | |
1677 | ||
6881a229 AE |
1678 | /* |
1679 | * Greedily allocate a buffer big enough to handle the full | |
1680 | * range of basic blocks to be written. If that fails, try | |
1681 | * a smaller size. We need to be able to write at least a | |
1682 | * log sector, or we're out of luck. | |
1683 | */ | |
1da177e4 | 1684 | bufblks = 1 << ffs(blocks); |
81158e0c DC |
1685 | while (bufblks > log->l_logBBsize) |
1686 | bufblks >>= 1; | |
1da177e4 LT |
1687 | while (!(bp = xlog_get_bp(log, bufblks))) { |
1688 | bufblks >>= 1; | |
69ce58f0 | 1689 | if (bufblks < sectbb) |
2451337d | 1690 | return -ENOMEM; |
1da177e4 LT |
1691 | } |
1692 | ||
1693 | /* We may need to do a read at the start to fill in part of | |
1694 | * the buffer in the starting sector not covered by the first | |
1695 | * write below. | |
1696 | */ | |
5c17f533 | 1697 | balign = round_down(start_block, sectbb); |
1da177e4 | 1698 | if (balign != start_block) { |
076e6acb CH |
1699 | error = xlog_bread_noalign(log, start_block, 1, bp); |
1700 | if (error) | |
1701 | goto out_put_bp; | |
1702 | ||
1da177e4 LT |
1703 | j = start_block - balign; |
1704 | } | |
1705 | ||
1706 | for (i = start_block; i < end_block; i += bufblks) { | |
1707 | int bcount, endcount; | |
1708 | ||
1709 | bcount = min(bufblks, end_block - start_block); | |
1710 | endcount = bcount - j; | |
1711 | ||
1712 | /* We may need to do a read at the end to fill in part of | |
1713 | * the buffer in the final sector not covered by the write. | |
1714 | * If this is the same sector as the above read, skip it. | |
1715 | */ | |
5c17f533 | 1716 | ealign = round_down(end_block, sectbb); |
1da177e4 | 1717 | if (j == 0 && (start_block + endcount > ealign)) { |
62926044 | 1718 | offset = bp->b_addr + BBTOB(ealign - start_block); |
44396476 DC |
1719 | error = xlog_bread_offset(log, ealign, sectbb, |
1720 | bp, offset); | |
076e6acb CH |
1721 | if (error) |
1722 | break; | |
1723 | ||
1da177e4 LT |
1724 | } |
1725 | ||
1726 | offset = xlog_align(log, start_block, endcount, bp); | |
1727 | for (; j < endcount; j++) { | |
1728 | xlog_add_record(log, offset, cycle, i+j, | |
1729 | tail_cycle, tail_block); | |
1730 | offset += BBSIZE; | |
1731 | } | |
1732 | error = xlog_bwrite(log, start_block, endcount, bp); | |
1733 | if (error) | |
1734 | break; | |
1735 | start_block += endcount; | |
1736 | j = 0; | |
1737 | } | |
076e6acb CH |
1738 | |
1739 | out_put_bp: | |
1da177e4 LT |
1740 | xlog_put_bp(bp); |
1741 | return error; | |
1742 | } | |
1743 | ||
1744 | /* | |
1745 | * This routine is called to blow away any incomplete log writes out | |
1746 | * in front of the log head. We do this so that we won't become confused | |
1747 | * if we come up, write only a little bit more, and then crash again. | |
1748 | * If we leave the partial log records out there, this situation could | |
1749 | * cause us to think those partial writes are valid blocks since they | |
1750 | * have the current cycle number. We get rid of them by overwriting them | |
1751 | * with empty log records with the old cycle number rather than the | |
1752 | * current one. | |
1753 | * | |
1754 | * The tail lsn is passed in rather than taken from | |
1755 | * the log so that we will not write over the unmount record after a | |
1756 | * clean unmount in a 512 block log. Doing so would leave the log without | |
1757 | * any valid log records in it until a new one was written. If we crashed | |
1758 | * during that time we would not be able to recover. | |
1759 | */ | |
1760 | STATIC int | |
1761 | xlog_clear_stale_blocks( | |
9a8d2fdb | 1762 | struct xlog *log, |
1da177e4 LT |
1763 | xfs_lsn_t tail_lsn) |
1764 | { | |
1765 | int tail_cycle, head_cycle; | |
1766 | int tail_block, head_block; | |
1767 | int tail_distance, max_distance; | |
1768 | int distance; | |
1769 | int error; | |
1770 | ||
1771 | tail_cycle = CYCLE_LSN(tail_lsn); | |
1772 | tail_block = BLOCK_LSN(tail_lsn); | |
1773 | head_cycle = log->l_curr_cycle; | |
1774 | head_block = log->l_curr_block; | |
1775 | ||
1776 | /* | |
1777 | * Figure out the distance between the new head of the log | |
1778 | * and the tail. We want to write over any blocks beyond the | |
1779 | * head that we may have written just before the crash, but | |
1780 | * we don't want to overwrite the tail of the log. | |
1781 | */ | |
1782 | if (head_cycle == tail_cycle) { | |
1783 | /* | |
1784 | * The tail is behind the head in the physical log, | |
1785 | * so the distance from the head to the tail is the | |
1786 | * distance from the head to the end of the log plus | |
1787 | * the distance from the beginning of the log to the | |
1788 | * tail. | |
1789 | */ | |
1790 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { | |
1791 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", | |
1792 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 1793 | return -EFSCORRUPTED; |
1da177e4 LT |
1794 | } |
1795 | tail_distance = tail_block + (log->l_logBBsize - head_block); | |
1796 | } else { | |
1797 | /* | |
1798 | * The head is behind the tail in the physical log, | |
1799 | * so the distance from the head to the tail is just | |
1800 | * the tail block minus the head block. | |
1801 | */ | |
1802 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ | |
1803 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", | |
1804 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 1805 | return -EFSCORRUPTED; |
1da177e4 LT |
1806 | } |
1807 | tail_distance = tail_block - head_block; | |
1808 | } | |
1809 | ||
1810 | /* | |
1811 | * If the head is right up against the tail, we can't clear | |
1812 | * anything. | |
1813 | */ | |
1814 | if (tail_distance <= 0) { | |
1815 | ASSERT(tail_distance == 0); | |
1816 | return 0; | |
1817 | } | |
1818 | ||
1819 | max_distance = XLOG_TOTAL_REC_SHIFT(log); | |
1820 | /* | |
1821 | * Take the smaller of the maximum amount of outstanding I/O | |
1822 | * we could have and the distance to the tail to clear out. | |
1823 | * We take the smaller so that we don't overwrite the tail and | |
1824 | * we don't waste all day writing from the head to the tail | |
1825 | * for no reason. | |
1826 | */ | |
9bb54cb5 | 1827 | max_distance = min(max_distance, tail_distance); |
1da177e4 LT |
1828 | |
1829 | if ((head_block + max_distance) <= log->l_logBBsize) { | |
1830 | /* | |
1831 | * We can stomp all the blocks we need to without | |
1832 | * wrapping around the end of the log. Just do it | |
1833 | * in a single write. Use the cycle number of the | |
1834 | * current cycle minus one so that the log will look like: | |
1835 | * n ... | n - 1 ... | |
1836 | */ | |
1837 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1838 | head_block, max_distance, tail_cycle, | |
1839 | tail_block); | |
1840 | if (error) | |
1841 | return error; | |
1842 | } else { | |
1843 | /* | |
1844 | * We need to wrap around the end of the physical log in | |
1845 | * order to clear all the blocks. Do it in two separate | |
1846 | * I/Os. The first write should be from the head to the | |
1847 | * end of the physical log, and it should use the current | |
1848 | * cycle number minus one just like above. | |
1849 | */ | |
1850 | distance = log->l_logBBsize - head_block; | |
1851 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1852 | head_block, distance, tail_cycle, | |
1853 | tail_block); | |
1854 | ||
1855 | if (error) | |
1856 | return error; | |
1857 | ||
1858 | /* | |
1859 | * Now write the blocks at the start of the physical log. | |
1860 | * This writes the remainder of the blocks we want to clear. | |
1861 | * It uses the current cycle number since we're now on the | |
1862 | * same cycle as the head so that we get: | |
1863 | * n ... n ... | n - 1 ... | |
1864 | * ^^^^^ blocks we're writing | |
1865 | */ | |
1866 | distance = max_distance - (log->l_logBBsize - head_block); | |
1867 | error = xlog_write_log_records(log, head_cycle, 0, distance, | |
1868 | tail_cycle, tail_block); | |
1869 | if (error) | |
1870 | return error; | |
1871 | } | |
1872 | ||
1873 | return 0; | |
1874 | } | |
1875 | ||
1876 | /****************************************************************************** | |
1877 | * | |
1878 | * Log recover routines | |
1879 | * | |
1880 | ****************************************************************************** | |
1881 | */ | |
1882 | ||
f0a76953 | 1883 | /* |
a775ad77 DC |
1884 | * Sort the log items in the transaction. |
1885 | * | |
1886 | * The ordering constraints are defined by the inode allocation and unlink | |
1887 | * behaviour. The rules are: | |
1888 | * | |
1889 | * 1. Every item is only logged once in a given transaction. Hence it | |
1890 | * represents the last logged state of the item. Hence ordering is | |
1891 | * dependent on the order in which operations need to be performed so | |
1892 | * required initial conditions are always met. | |
1893 | * | |
1894 | * 2. Cancelled buffers are recorded in pass 1 in a separate table and | |
1895 | * there's nothing to replay from them so we can simply cull them | |
1896 | * from the transaction. However, we can't do that until after we've | |
1897 | * replayed all the other items because they may be dependent on the | |
1898 | * cancelled buffer and replaying the cancelled buffer can remove it | |
1899 | * form the cancelled buffer table. Hence they have tobe done last. | |
1900 | * | |
1901 | * 3. Inode allocation buffers must be replayed before inode items that | |
28c8e41a DC |
1902 | * read the buffer and replay changes into it. For filesystems using the |
1903 | * ICREATE transactions, this means XFS_LI_ICREATE objects need to get | |
1904 | * treated the same as inode allocation buffers as they create and | |
1905 | * initialise the buffers directly. | |
a775ad77 DC |
1906 | * |
1907 | * 4. Inode unlink buffers must be replayed after inode items are replayed. | |
1908 | * This ensures that inodes are completely flushed to the inode buffer | |
1909 | * in a "free" state before we remove the unlinked inode list pointer. | |
1910 | * | |
1911 | * Hence the ordering needs to be inode allocation buffers first, inode items | |
1912 | * second, inode unlink buffers third and cancelled buffers last. | |
1913 | * | |
1914 | * But there's a problem with that - we can't tell an inode allocation buffer | |
1915 | * apart from a regular buffer, so we can't separate them. We can, however, | |
1916 | * tell an inode unlink buffer from the others, and so we can separate them out | |
1917 | * from all the other buffers and move them to last. | |
1918 | * | |
1919 | * Hence, 4 lists, in order from head to tail: | |
28c8e41a DC |
1920 | * - buffer_list for all buffers except cancelled/inode unlink buffers |
1921 | * - item_list for all non-buffer items | |
1922 | * - inode_buffer_list for inode unlink buffers | |
1923 | * - cancel_list for the cancelled buffers | |
1924 | * | |
1925 | * Note that we add objects to the tail of the lists so that first-to-last | |
1926 | * ordering is preserved within the lists. Adding objects to the head of the | |
1927 | * list means when we traverse from the head we walk them in last-to-first | |
1928 | * order. For cancelled buffers and inode unlink buffers this doesn't matter, | |
1929 | * but for all other items there may be specific ordering that we need to | |
1930 | * preserve. | |
f0a76953 | 1931 | */ |
1da177e4 LT |
1932 | STATIC int |
1933 | xlog_recover_reorder_trans( | |
ad223e60 MT |
1934 | struct xlog *log, |
1935 | struct xlog_recover *trans, | |
9abbc539 | 1936 | int pass) |
1da177e4 | 1937 | { |
f0a76953 | 1938 | xlog_recover_item_t *item, *n; |
2a84108f | 1939 | int error = 0; |
f0a76953 | 1940 | LIST_HEAD(sort_list); |
a775ad77 DC |
1941 | LIST_HEAD(cancel_list); |
1942 | LIST_HEAD(buffer_list); | |
1943 | LIST_HEAD(inode_buffer_list); | |
1944 | LIST_HEAD(inode_list); | |
f0a76953 DC |
1945 | |
1946 | list_splice_init(&trans->r_itemq, &sort_list); | |
1947 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { | |
4e0d5f92 | 1948 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
1da177e4 | 1949 | |
f0a76953 | 1950 | switch (ITEM_TYPE(item)) { |
28c8e41a DC |
1951 | case XFS_LI_ICREATE: |
1952 | list_move_tail(&item->ri_list, &buffer_list); | |
1953 | break; | |
1da177e4 | 1954 | case XFS_LI_BUF: |
a775ad77 | 1955 | if (buf_f->blf_flags & XFS_BLF_CANCEL) { |
9abbc539 DC |
1956 | trace_xfs_log_recover_item_reorder_head(log, |
1957 | trans, item, pass); | |
a775ad77 | 1958 | list_move(&item->ri_list, &cancel_list); |
1da177e4 LT |
1959 | break; |
1960 | } | |
a775ad77 DC |
1961 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
1962 | list_move(&item->ri_list, &inode_buffer_list); | |
1963 | break; | |
1964 | } | |
1965 | list_move_tail(&item->ri_list, &buffer_list); | |
1966 | break; | |
1da177e4 | 1967 | case XFS_LI_INODE: |
1da177e4 LT |
1968 | case XFS_LI_DQUOT: |
1969 | case XFS_LI_QUOTAOFF: | |
1970 | case XFS_LI_EFD: | |
1971 | case XFS_LI_EFI: | |
9e88b5d8 DW |
1972 | case XFS_LI_RUI: |
1973 | case XFS_LI_RUD: | |
f997ee21 DW |
1974 | case XFS_LI_CUI: |
1975 | case XFS_LI_CUD: | |
77d61fe4 DW |
1976 | case XFS_LI_BUI: |
1977 | case XFS_LI_BUD: | |
9abbc539 DC |
1978 | trace_xfs_log_recover_item_reorder_tail(log, |
1979 | trans, item, pass); | |
a775ad77 | 1980 | list_move_tail(&item->ri_list, &inode_list); |
1da177e4 LT |
1981 | break; |
1982 | default: | |
a0fa2b67 DC |
1983 | xfs_warn(log->l_mp, |
1984 | "%s: unrecognized type of log operation", | |
1985 | __func__); | |
1da177e4 | 1986 | ASSERT(0); |
2a84108f MT |
1987 | /* |
1988 | * return the remaining items back to the transaction | |
1989 | * item list so they can be freed in caller. | |
1990 | */ | |
1991 | if (!list_empty(&sort_list)) | |
1992 | list_splice_init(&sort_list, &trans->r_itemq); | |
2451337d | 1993 | error = -EIO; |
2a84108f | 1994 | goto out; |
1da177e4 | 1995 | } |
f0a76953 | 1996 | } |
2a84108f | 1997 | out: |
f0a76953 | 1998 | ASSERT(list_empty(&sort_list)); |
a775ad77 DC |
1999 | if (!list_empty(&buffer_list)) |
2000 | list_splice(&buffer_list, &trans->r_itemq); | |
2001 | if (!list_empty(&inode_list)) | |
2002 | list_splice_tail(&inode_list, &trans->r_itemq); | |
2003 | if (!list_empty(&inode_buffer_list)) | |
2004 | list_splice_tail(&inode_buffer_list, &trans->r_itemq); | |
2005 | if (!list_empty(&cancel_list)) | |
2006 | list_splice_tail(&cancel_list, &trans->r_itemq); | |
2a84108f | 2007 | return error; |
1da177e4 LT |
2008 | } |
2009 | ||
2010 | /* | |
2011 | * Build up the table of buf cancel records so that we don't replay | |
2012 | * cancelled data in the second pass. For buffer records that are | |
2013 | * not cancel records, there is nothing to do here so we just return. | |
2014 | * | |
2015 | * If we get a cancel record which is already in the table, this indicates | |
2016 | * that the buffer was cancelled multiple times. In order to ensure | |
2017 | * that during pass 2 we keep the record in the table until we reach its | |
2018 | * last occurrence in the log, we keep a reference count in the cancel | |
2019 | * record in the table to tell us how many times we expect to see this | |
2020 | * record during the second pass. | |
2021 | */ | |
c9f71f5f CH |
2022 | STATIC int |
2023 | xlog_recover_buffer_pass1( | |
ad223e60 MT |
2024 | struct xlog *log, |
2025 | struct xlog_recover_item *item) | |
1da177e4 | 2026 | { |
c9f71f5f | 2027 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
d5689eaa CH |
2028 | struct list_head *bucket; |
2029 | struct xfs_buf_cancel *bcp; | |
1da177e4 LT |
2030 | |
2031 | /* | |
2032 | * If this isn't a cancel buffer item, then just return. | |
2033 | */ | |
e2714bf8 | 2034 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
9abbc539 | 2035 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); |
c9f71f5f | 2036 | return 0; |
9abbc539 | 2037 | } |
1da177e4 LT |
2038 | |
2039 | /* | |
d5689eaa CH |
2040 | * Insert an xfs_buf_cancel record into the hash table of them. |
2041 | * If there is already an identical record, bump its reference count. | |
1da177e4 | 2042 | */ |
d5689eaa CH |
2043 | bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno); |
2044 | list_for_each_entry(bcp, bucket, bc_list) { | |
2045 | if (bcp->bc_blkno == buf_f->blf_blkno && | |
2046 | bcp->bc_len == buf_f->blf_len) { | |
2047 | bcp->bc_refcount++; | |
9abbc539 | 2048 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
c9f71f5f | 2049 | return 0; |
1da177e4 | 2050 | } |
d5689eaa CH |
2051 | } |
2052 | ||
2053 | bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP); | |
2054 | bcp->bc_blkno = buf_f->blf_blkno; | |
2055 | bcp->bc_len = buf_f->blf_len; | |
1da177e4 | 2056 | bcp->bc_refcount = 1; |
d5689eaa CH |
2057 | list_add_tail(&bcp->bc_list, bucket); |
2058 | ||
9abbc539 | 2059 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
c9f71f5f | 2060 | return 0; |
1da177e4 LT |
2061 | } |
2062 | ||
2063 | /* | |
2064 | * Check to see whether the buffer being recovered has a corresponding | |
84a5b730 DC |
2065 | * entry in the buffer cancel record table. If it is, return the cancel |
2066 | * buffer structure to the caller. | |
1da177e4 | 2067 | */ |
84a5b730 DC |
2068 | STATIC struct xfs_buf_cancel * |
2069 | xlog_peek_buffer_cancelled( | |
ad223e60 | 2070 | struct xlog *log, |
1da177e4 LT |
2071 | xfs_daddr_t blkno, |
2072 | uint len, | |
755c7bf5 | 2073 | unsigned short flags) |
1da177e4 | 2074 | { |
d5689eaa CH |
2075 | struct list_head *bucket; |
2076 | struct xfs_buf_cancel *bcp; | |
1da177e4 | 2077 | |
84a5b730 DC |
2078 | if (!log->l_buf_cancel_table) { |
2079 | /* empty table means no cancelled buffers in the log */ | |
c1155410 | 2080 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
84a5b730 | 2081 | return NULL; |
1da177e4 LT |
2082 | } |
2083 | ||
d5689eaa CH |
2084 | bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno); |
2085 | list_for_each_entry(bcp, bucket, bc_list) { | |
2086 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) | |
84a5b730 | 2087 | return bcp; |
1da177e4 | 2088 | } |
d5689eaa | 2089 | |
1da177e4 | 2090 | /* |
d5689eaa CH |
2091 | * We didn't find a corresponding entry in the table, so return 0 so |
2092 | * that the buffer is NOT cancelled. | |
1da177e4 | 2093 | */ |
c1155410 | 2094 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
84a5b730 DC |
2095 | return NULL; |
2096 | } | |
2097 | ||
2098 | /* | |
2099 | * If the buffer is being cancelled then return 1 so that it will be cancelled, | |
2100 | * otherwise return 0. If the buffer is actually a buffer cancel item | |
2101 | * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the | |
2102 | * table and remove it from the table if this is the last reference. | |
2103 | * | |
2104 | * We remove the cancel record from the table when we encounter its last | |
2105 | * occurrence in the log so that if the same buffer is re-used again after its | |
2106 | * last cancellation we actually replay the changes made at that point. | |
2107 | */ | |
2108 | STATIC int | |
2109 | xlog_check_buffer_cancelled( | |
2110 | struct xlog *log, | |
2111 | xfs_daddr_t blkno, | |
2112 | uint len, | |
755c7bf5 | 2113 | unsigned short flags) |
84a5b730 DC |
2114 | { |
2115 | struct xfs_buf_cancel *bcp; | |
2116 | ||
2117 | bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags); | |
2118 | if (!bcp) | |
2119 | return 0; | |
d5689eaa | 2120 | |
d5689eaa CH |
2121 | /* |
2122 | * We've go a match, so return 1 so that the recovery of this buffer | |
2123 | * is cancelled. If this buffer is actually a buffer cancel log | |
2124 | * item, then decrement the refcount on the one in the table and | |
2125 | * remove it if this is the last reference. | |
2126 | */ | |
2127 | if (flags & XFS_BLF_CANCEL) { | |
2128 | if (--bcp->bc_refcount == 0) { | |
2129 | list_del(&bcp->bc_list); | |
2130 | kmem_free(bcp); | |
2131 | } | |
2132 | } | |
2133 | return 1; | |
1da177e4 LT |
2134 | } |
2135 | ||
1da177e4 | 2136 | /* |
e2714bf8 CH |
2137 | * Perform recovery for a buffer full of inodes. In these buffers, the only |
2138 | * data which should be recovered is that which corresponds to the | |
2139 | * di_next_unlinked pointers in the on disk inode structures. The rest of the | |
2140 | * data for the inodes is always logged through the inodes themselves rather | |
2141 | * than the inode buffer and is recovered in xlog_recover_inode_pass2(). | |
1da177e4 | 2142 | * |
e2714bf8 CH |
2143 | * The only time when buffers full of inodes are fully recovered is when the |
2144 | * buffer is full of newly allocated inodes. In this case the buffer will | |
2145 | * not be marked as an inode buffer and so will be sent to | |
2146 | * xlog_recover_do_reg_buffer() below during recovery. | |
1da177e4 LT |
2147 | */ |
2148 | STATIC int | |
2149 | xlog_recover_do_inode_buffer( | |
e2714bf8 | 2150 | struct xfs_mount *mp, |
1da177e4 | 2151 | xlog_recover_item_t *item, |
e2714bf8 | 2152 | struct xfs_buf *bp, |
1da177e4 LT |
2153 | xfs_buf_log_format_t *buf_f) |
2154 | { | |
2155 | int i; | |
e2714bf8 CH |
2156 | int item_index = 0; |
2157 | int bit = 0; | |
2158 | int nbits = 0; | |
2159 | int reg_buf_offset = 0; | |
2160 | int reg_buf_bytes = 0; | |
1da177e4 LT |
2161 | int next_unlinked_offset; |
2162 | int inodes_per_buf; | |
2163 | xfs_agino_t *logged_nextp; | |
2164 | xfs_agino_t *buffer_nextp; | |
1da177e4 | 2165 | |
9abbc539 | 2166 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
9222a9cf DC |
2167 | |
2168 | /* | |
2169 | * Post recovery validation only works properly on CRC enabled | |
2170 | * filesystems. | |
2171 | */ | |
2172 | if (xfs_sb_version_hascrc(&mp->m_sb)) | |
2173 | bp->b_ops = &xfs_inode_buf_ops; | |
9abbc539 | 2174 | |
aa0e8833 | 2175 | inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog; |
1da177e4 LT |
2176 | for (i = 0; i < inodes_per_buf; i++) { |
2177 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + | |
2178 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2179 | ||
2180 | while (next_unlinked_offset >= | |
2181 | (reg_buf_offset + reg_buf_bytes)) { | |
2182 | /* | |
2183 | * The next di_next_unlinked field is beyond | |
2184 | * the current logged region. Find the next | |
2185 | * logged region that contains or is beyond | |
2186 | * the current di_next_unlinked field. | |
2187 | */ | |
2188 | bit += nbits; | |
e2714bf8 CH |
2189 | bit = xfs_next_bit(buf_f->blf_data_map, |
2190 | buf_f->blf_map_size, bit); | |
1da177e4 LT |
2191 | |
2192 | /* | |
2193 | * If there are no more logged regions in the | |
2194 | * buffer, then we're done. | |
2195 | */ | |
e2714bf8 | 2196 | if (bit == -1) |
1da177e4 | 2197 | return 0; |
1da177e4 | 2198 | |
e2714bf8 CH |
2199 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
2200 | buf_f->blf_map_size, bit); | |
1da177e4 | 2201 | ASSERT(nbits > 0); |
c1155410 DC |
2202 | reg_buf_offset = bit << XFS_BLF_SHIFT; |
2203 | reg_buf_bytes = nbits << XFS_BLF_SHIFT; | |
1da177e4 LT |
2204 | item_index++; |
2205 | } | |
2206 | ||
2207 | /* | |
2208 | * If the current logged region starts after the current | |
2209 | * di_next_unlinked field, then move on to the next | |
2210 | * di_next_unlinked field. | |
2211 | */ | |
e2714bf8 | 2212 | if (next_unlinked_offset < reg_buf_offset) |
1da177e4 | 2213 | continue; |
1da177e4 LT |
2214 | |
2215 | ASSERT(item->ri_buf[item_index].i_addr != NULL); | |
c1155410 | 2216 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0); |
aa0e8833 DC |
2217 | ASSERT((reg_buf_offset + reg_buf_bytes) <= |
2218 | BBTOB(bp->b_io_length)); | |
1da177e4 LT |
2219 | |
2220 | /* | |
2221 | * The current logged region contains a copy of the | |
2222 | * current di_next_unlinked field. Extract its value | |
2223 | * and copy it to the buffer copy. | |
2224 | */ | |
4e0d5f92 CH |
2225 | logged_nextp = item->ri_buf[item_index].i_addr + |
2226 | next_unlinked_offset - reg_buf_offset; | |
1da177e4 | 2227 | if (unlikely(*logged_nextp == 0)) { |
a0fa2b67 | 2228 | xfs_alert(mp, |
c9690043 | 2229 | "Bad inode buffer log record (ptr = "PTR_FMT", bp = "PTR_FMT"). " |
a0fa2b67 | 2230 | "Trying to replay bad (0) inode di_next_unlinked field.", |
1da177e4 LT |
2231 | item, bp); |
2232 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", | |
2233 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 2234 | return -EFSCORRUPTED; |
1da177e4 LT |
2235 | } |
2236 | ||
88ee2df7 | 2237 | buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset); |
87c199c2 | 2238 | *buffer_nextp = *logged_nextp; |
0a32c26e DC |
2239 | |
2240 | /* | |
2241 | * If necessary, recalculate the CRC in the on-disk inode. We | |
2242 | * have to leave the inode in a consistent state for whoever | |
2243 | * reads it next.... | |
2244 | */ | |
88ee2df7 | 2245 | xfs_dinode_calc_crc(mp, |
0a32c26e DC |
2246 | xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize)); |
2247 | ||
1da177e4 LT |
2248 | } |
2249 | ||
2250 | return 0; | |
2251 | } | |
2252 | ||
50d5c8d8 DC |
2253 | /* |
2254 | * V5 filesystems know the age of the buffer on disk being recovered. We can | |
2255 | * have newer objects on disk than we are replaying, and so for these cases we | |
2256 | * don't want to replay the current change as that will make the buffer contents | |
2257 | * temporarily invalid on disk. | |
2258 | * | |
2259 | * The magic number might not match the buffer type we are going to recover | |
2260 | * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence | |
2261 | * extract the LSN of the existing object in the buffer based on it's current | |
2262 | * magic number. If we don't recognise the magic number in the buffer, then | |
2263 | * return a LSN of -1 so that the caller knows it was an unrecognised block and | |
2264 | * so can recover the buffer. | |
566055d3 DC |
2265 | * |
2266 | * Note: we cannot rely solely on magic number matches to determine that the | |
2267 | * buffer has a valid LSN - we also need to verify that it belongs to this | |
2268 | * filesystem, so we need to extract the object's LSN and compare it to that | |
2269 | * which we read from the superblock. If the UUIDs don't match, then we've got a | |
2270 | * stale metadata block from an old filesystem instance that we need to recover | |
2271 | * over the top of. | |
50d5c8d8 DC |
2272 | */ |
2273 | static xfs_lsn_t | |
2274 | xlog_recover_get_buf_lsn( | |
2275 | struct xfs_mount *mp, | |
2276 | struct xfs_buf *bp) | |
2277 | { | |
c8ce540d DW |
2278 | uint32_t magic32; |
2279 | uint16_t magic16; | |
2280 | uint16_t magicda; | |
50d5c8d8 | 2281 | void *blk = bp->b_addr; |
566055d3 DC |
2282 | uuid_t *uuid; |
2283 | xfs_lsn_t lsn = -1; | |
50d5c8d8 DC |
2284 | |
2285 | /* v4 filesystems always recover immediately */ | |
2286 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2287 | goto recover_immediately; | |
2288 | ||
2289 | magic32 = be32_to_cpu(*(__be32 *)blk); | |
2290 | switch (magic32) { | |
2291 | case XFS_ABTB_CRC_MAGIC: | |
2292 | case XFS_ABTC_CRC_MAGIC: | |
2293 | case XFS_ABTB_MAGIC: | |
2294 | case XFS_ABTC_MAGIC: | |
a650e8f9 | 2295 | case XFS_RMAP_CRC_MAGIC: |
a90c00f0 | 2296 | case XFS_REFC_CRC_MAGIC: |
50d5c8d8 | 2297 | case XFS_IBT_CRC_MAGIC: |
566055d3 DC |
2298 | case XFS_IBT_MAGIC: { |
2299 | struct xfs_btree_block *btb = blk; | |
2300 | ||
2301 | lsn = be64_to_cpu(btb->bb_u.s.bb_lsn); | |
2302 | uuid = &btb->bb_u.s.bb_uuid; | |
2303 | break; | |
2304 | } | |
50d5c8d8 | 2305 | case XFS_BMAP_CRC_MAGIC: |
566055d3 DC |
2306 | case XFS_BMAP_MAGIC: { |
2307 | struct xfs_btree_block *btb = blk; | |
2308 | ||
2309 | lsn = be64_to_cpu(btb->bb_u.l.bb_lsn); | |
2310 | uuid = &btb->bb_u.l.bb_uuid; | |
2311 | break; | |
2312 | } | |
50d5c8d8 | 2313 | case XFS_AGF_MAGIC: |
566055d3 DC |
2314 | lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn); |
2315 | uuid = &((struct xfs_agf *)blk)->agf_uuid; | |
2316 | break; | |
50d5c8d8 | 2317 | case XFS_AGFL_MAGIC: |
566055d3 DC |
2318 | lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn); |
2319 | uuid = &((struct xfs_agfl *)blk)->agfl_uuid; | |
2320 | break; | |
50d5c8d8 | 2321 | case XFS_AGI_MAGIC: |
566055d3 DC |
2322 | lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn); |
2323 | uuid = &((struct xfs_agi *)blk)->agi_uuid; | |
2324 | break; | |
50d5c8d8 | 2325 | case XFS_SYMLINK_MAGIC: |
566055d3 DC |
2326 | lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn); |
2327 | uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid; | |
2328 | break; | |
50d5c8d8 DC |
2329 | case XFS_DIR3_BLOCK_MAGIC: |
2330 | case XFS_DIR3_DATA_MAGIC: | |
2331 | case XFS_DIR3_FREE_MAGIC: | |
566055d3 DC |
2332 | lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn); |
2333 | uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid; | |
2334 | break; | |
50d5c8d8 | 2335 | case XFS_ATTR3_RMT_MAGIC: |
e3c32ee9 DC |
2336 | /* |
2337 | * Remote attr blocks are written synchronously, rather than | |
2338 | * being logged. That means they do not contain a valid LSN | |
2339 | * (i.e. transactionally ordered) in them, and hence any time we | |
2340 | * see a buffer to replay over the top of a remote attribute | |
2341 | * block we should simply do so. | |
2342 | */ | |
2343 | goto recover_immediately; | |
50d5c8d8 | 2344 | case XFS_SB_MAGIC: |
fcfbe2c4 DC |
2345 | /* |
2346 | * superblock uuids are magic. We may or may not have a | |
2347 | * sb_meta_uuid on disk, but it will be set in the in-core | |
2348 | * superblock. We set the uuid pointer for verification | |
2349 | * according to the superblock feature mask to ensure we check | |
2350 | * the relevant UUID in the superblock. | |
2351 | */ | |
566055d3 | 2352 | lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn); |
fcfbe2c4 DC |
2353 | if (xfs_sb_version_hasmetauuid(&mp->m_sb)) |
2354 | uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid; | |
2355 | else | |
2356 | uuid = &((struct xfs_dsb *)blk)->sb_uuid; | |
566055d3 | 2357 | break; |
50d5c8d8 DC |
2358 | default: |
2359 | break; | |
2360 | } | |
2361 | ||
566055d3 | 2362 | if (lsn != (xfs_lsn_t)-1) { |
fcfbe2c4 | 2363 | if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid)) |
566055d3 DC |
2364 | goto recover_immediately; |
2365 | return lsn; | |
2366 | } | |
2367 | ||
50d5c8d8 DC |
2368 | magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic); |
2369 | switch (magicda) { | |
2370 | case XFS_DIR3_LEAF1_MAGIC: | |
2371 | case XFS_DIR3_LEAFN_MAGIC: | |
2372 | case XFS_DA3_NODE_MAGIC: | |
566055d3 DC |
2373 | lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn); |
2374 | uuid = &((struct xfs_da3_blkinfo *)blk)->uuid; | |
2375 | break; | |
50d5c8d8 DC |
2376 | default: |
2377 | break; | |
2378 | } | |
2379 | ||
566055d3 DC |
2380 | if (lsn != (xfs_lsn_t)-1) { |
2381 | if (!uuid_equal(&mp->m_sb.sb_uuid, uuid)) | |
2382 | goto recover_immediately; | |
2383 | return lsn; | |
2384 | } | |
2385 | ||
50d5c8d8 DC |
2386 | /* |
2387 | * We do individual object checks on dquot and inode buffers as they | |
2388 | * have their own individual LSN records. Also, we could have a stale | |
2389 | * buffer here, so we have to at least recognise these buffer types. | |
2390 | * | |
2391 | * A notd complexity here is inode unlinked list processing - it logs | |
2392 | * the inode directly in the buffer, but we don't know which inodes have | |
2393 | * been modified, and there is no global buffer LSN. Hence we need to | |
2394 | * recover all inode buffer types immediately. This problem will be | |
2395 | * fixed by logical logging of the unlinked list modifications. | |
2396 | */ | |
2397 | magic16 = be16_to_cpu(*(__be16 *)blk); | |
2398 | switch (magic16) { | |
2399 | case XFS_DQUOT_MAGIC: | |
2400 | case XFS_DINODE_MAGIC: | |
2401 | goto recover_immediately; | |
2402 | default: | |
2403 | break; | |
2404 | } | |
2405 | ||
2406 | /* unknown buffer contents, recover immediately */ | |
2407 | ||
2408 | recover_immediately: | |
2409 | return (xfs_lsn_t)-1; | |
2410 | ||
2411 | } | |
2412 | ||
1da177e4 | 2413 | /* |
d75afeb3 DC |
2414 | * Validate the recovered buffer is of the correct type and attach the |
2415 | * appropriate buffer operations to them for writeback. Magic numbers are in a | |
2416 | * few places: | |
2417 | * the first 16 bits of the buffer (inode buffer, dquot buffer), | |
2418 | * the first 32 bits of the buffer (most blocks), | |
2419 | * inside a struct xfs_da_blkinfo at the start of the buffer. | |
1da177e4 | 2420 | */ |
d75afeb3 | 2421 | static void |
50d5c8d8 | 2422 | xlog_recover_validate_buf_type( |
9abbc539 | 2423 | struct xfs_mount *mp, |
e2714bf8 | 2424 | struct xfs_buf *bp, |
22db9af2 BF |
2425 | xfs_buf_log_format_t *buf_f, |
2426 | xfs_lsn_t current_lsn) | |
1da177e4 | 2427 | { |
d75afeb3 | 2428 | struct xfs_da_blkinfo *info = bp->b_addr; |
c8ce540d DW |
2429 | uint32_t magic32; |
2430 | uint16_t magic16; | |
2431 | uint16_t magicda; | |
040c52c0 | 2432 | char *warnmsg = NULL; |
d75afeb3 | 2433 | |
67dc288c DC |
2434 | /* |
2435 | * We can only do post recovery validation on items on CRC enabled | |
2436 | * fielsystems as we need to know when the buffer was written to be able | |
2437 | * to determine if we should have replayed the item. If we replay old | |
2438 | * metadata over a newer buffer, then it will enter a temporarily | |
2439 | * inconsistent state resulting in verification failures. Hence for now | |
2440 | * just avoid the verification stage for non-crc filesystems | |
2441 | */ | |
2442 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2443 | return; | |
2444 | ||
d75afeb3 DC |
2445 | magic32 = be32_to_cpu(*(__be32 *)bp->b_addr); |
2446 | magic16 = be16_to_cpu(*(__be16*)bp->b_addr); | |
2447 | magicda = be16_to_cpu(info->magic); | |
61fe135c DC |
2448 | switch (xfs_blft_from_flags(buf_f)) { |
2449 | case XFS_BLFT_BTREE_BUF: | |
d75afeb3 | 2450 | switch (magic32) { |
ee1a47ab CH |
2451 | case XFS_ABTB_CRC_MAGIC: |
2452 | case XFS_ABTC_CRC_MAGIC: | |
2453 | case XFS_ABTB_MAGIC: | |
2454 | case XFS_ABTC_MAGIC: | |
2455 | bp->b_ops = &xfs_allocbt_buf_ops; | |
2456 | break; | |
2457 | case XFS_IBT_CRC_MAGIC: | |
aafc3c24 | 2458 | case XFS_FIBT_CRC_MAGIC: |
ee1a47ab | 2459 | case XFS_IBT_MAGIC: |
aafc3c24 | 2460 | case XFS_FIBT_MAGIC: |
ee1a47ab CH |
2461 | bp->b_ops = &xfs_inobt_buf_ops; |
2462 | break; | |
2463 | case XFS_BMAP_CRC_MAGIC: | |
2464 | case XFS_BMAP_MAGIC: | |
2465 | bp->b_ops = &xfs_bmbt_buf_ops; | |
2466 | break; | |
a650e8f9 DW |
2467 | case XFS_RMAP_CRC_MAGIC: |
2468 | bp->b_ops = &xfs_rmapbt_buf_ops; | |
2469 | break; | |
a90c00f0 DW |
2470 | case XFS_REFC_CRC_MAGIC: |
2471 | bp->b_ops = &xfs_refcountbt_buf_ops; | |
2472 | break; | |
ee1a47ab | 2473 | default: |
040c52c0 | 2474 | warnmsg = "Bad btree block magic!"; |
ee1a47ab CH |
2475 | break; |
2476 | } | |
2477 | break; | |
61fe135c | 2478 | case XFS_BLFT_AGF_BUF: |
d75afeb3 | 2479 | if (magic32 != XFS_AGF_MAGIC) { |
040c52c0 | 2480 | warnmsg = "Bad AGF block magic!"; |
4e0e6040 DC |
2481 | break; |
2482 | } | |
2483 | bp->b_ops = &xfs_agf_buf_ops; | |
2484 | break; | |
61fe135c | 2485 | case XFS_BLFT_AGFL_BUF: |
d75afeb3 | 2486 | if (magic32 != XFS_AGFL_MAGIC) { |
040c52c0 | 2487 | warnmsg = "Bad AGFL block magic!"; |
77c95bba CH |
2488 | break; |
2489 | } | |
2490 | bp->b_ops = &xfs_agfl_buf_ops; | |
2491 | break; | |
61fe135c | 2492 | case XFS_BLFT_AGI_BUF: |
d75afeb3 | 2493 | if (magic32 != XFS_AGI_MAGIC) { |
040c52c0 | 2494 | warnmsg = "Bad AGI block magic!"; |
983d09ff DC |
2495 | break; |
2496 | } | |
2497 | bp->b_ops = &xfs_agi_buf_ops; | |
2498 | break; | |
61fe135c DC |
2499 | case XFS_BLFT_UDQUOT_BUF: |
2500 | case XFS_BLFT_PDQUOT_BUF: | |
2501 | case XFS_BLFT_GDQUOT_BUF: | |
123887e8 | 2502 | #ifdef CONFIG_XFS_QUOTA |
d75afeb3 | 2503 | if (magic16 != XFS_DQUOT_MAGIC) { |
040c52c0 | 2504 | warnmsg = "Bad DQUOT block magic!"; |
3fe58f30 CH |
2505 | break; |
2506 | } | |
2507 | bp->b_ops = &xfs_dquot_buf_ops; | |
123887e8 DC |
2508 | #else |
2509 | xfs_alert(mp, | |
2510 | "Trying to recover dquots without QUOTA support built in!"); | |
2511 | ASSERT(0); | |
2512 | #endif | |
3fe58f30 | 2513 | break; |
61fe135c | 2514 | case XFS_BLFT_DINO_BUF: |
d75afeb3 | 2515 | if (magic16 != XFS_DINODE_MAGIC) { |
040c52c0 | 2516 | warnmsg = "Bad INODE block magic!"; |
93848a99 CH |
2517 | break; |
2518 | } | |
2519 | bp->b_ops = &xfs_inode_buf_ops; | |
2520 | break; | |
61fe135c | 2521 | case XFS_BLFT_SYMLINK_BUF: |
d75afeb3 | 2522 | if (magic32 != XFS_SYMLINK_MAGIC) { |
040c52c0 | 2523 | warnmsg = "Bad symlink block magic!"; |
f948dd76 DC |
2524 | break; |
2525 | } | |
2526 | bp->b_ops = &xfs_symlink_buf_ops; | |
2527 | break; | |
61fe135c | 2528 | case XFS_BLFT_DIR_BLOCK_BUF: |
d75afeb3 DC |
2529 | if (magic32 != XFS_DIR2_BLOCK_MAGIC && |
2530 | magic32 != XFS_DIR3_BLOCK_MAGIC) { | |
040c52c0 | 2531 | warnmsg = "Bad dir block magic!"; |
d75afeb3 DC |
2532 | break; |
2533 | } | |
2534 | bp->b_ops = &xfs_dir3_block_buf_ops; | |
2535 | break; | |
61fe135c | 2536 | case XFS_BLFT_DIR_DATA_BUF: |
d75afeb3 DC |
2537 | if (magic32 != XFS_DIR2_DATA_MAGIC && |
2538 | magic32 != XFS_DIR3_DATA_MAGIC) { | |
040c52c0 | 2539 | warnmsg = "Bad dir data magic!"; |
d75afeb3 DC |
2540 | break; |
2541 | } | |
2542 | bp->b_ops = &xfs_dir3_data_buf_ops; | |
2543 | break; | |
61fe135c | 2544 | case XFS_BLFT_DIR_FREE_BUF: |
d75afeb3 DC |
2545 | if (magic32 != XFS_DIR2_FREE_MAGIC && |
2546 | magic32 != XFS_DIR3_FREE_MAGIC) { | |
040c52c0 | 2547 | warnmsg = "Bad dir3 free magic!"; |
d75afeb3 DC |
2548 | break; |
2549 | } | |
2550 | bp->b_ops = &xfs_dir3_free_buf_ops; | |
2551 | break; | |
61fe135c | 2552 | case XFS_BLFT_DIR_LEAF1_BUF: |
d75afeb3 DC |
2553 | if (magicda != XFS_DIR2_LEAF1_MAGIC && |
2554 | magicda != XFS_DIR3_LEAF1_MAGIC) { | |
040c52c0 | 2555 | warnmsg = "Bad dir leaf1 magic!"; |
d75afeb3 DC |
2556 | break; |
2557 | } | |
2558 | bp->b_ops = &xfs_dir3_leaf1_buf_ops; | |
2559 | break; | |
61fe135c | 2560 | case XFS_BLFT_DIR_LEAFN_BUF: |
d75afeb3 DC |
2561 | if (magicda != XFS_DIR2_LEAFN_MAGIC && |
2562 | magicda != XFS_DIR3_LEAFN_MAGIC) { | |
040c52c0 | 2563 | warnmsg = "Bad dir leafn magic!"; |
d75afeb3 DC |
2564 | break; |
2565 | } | |
2566 | bp->b_ops = &xfs_dir3_leafn_buf_ops; | |
2567 | break; | |
61fe135c | 2568 | case XFS_BLFT_DA_NODE_BUF: |
d75afeb3 DC |
2569 | if (magicda != XFS_DA_NODE_MAGIC && |
2570 | magicda != XFS_DA3_NODE_MAGIC) { | |
040c52c0 | 2571 | warnmsg = "Bad da node magic!"; |
d75afeb3 DC |
2572 | break; |
2573 | } | |
2574 | bp->b_ops = &xfs_da3_node_buf_ops; | |
2575 | break; | |
61fe135c | 2576 | case XFS_BLFT_ATTR_LEAF_BUF: |
d75afeb3 DC |
2577 | if (magicda != XFS_ATTR_LEAF_MAGIC && |
2578 | magicda != XFS_ATTR3_LEAF_MAGIC) { | |
040c52c0 | 2579 | warnmsg = "Bad attr leaf magic!"; |
d75afeb3 DC |
2580 | break; |
2581 | } | |
2582 | bp->b_ops = &xfs_attr3_leaf_buf_ops; | |
2583 | break; | |
61fe135c | 2584 | case XFS_BLFT_ATTR_RMT_BUF: |
cab09a81 | 2585 | if (magic32 != XFS_ATTR3_RMT_MAGIC) { |
040c52c0 | 2586 | warnmsg = "Bad attr remote magic!"; |
d75afeb3 DC |
2587 | break; |
2588 | } | |
2589 | bp->b_ops = &xfs_attr3_rmt_buf_ops; | |
2590 | break; | |
04a1e6c5 DC |
2591 | case XFS_BLFT_SB_BUF: |
2592 | if (magic32 != XFS_SB_MAGIC) { | |
040c52c0 | 2593 | warnmsg = "Bad SB block magic!"; |
04a1e6c5 DC |
2594 | break; |
2595 | } | |
2596 | bp->b_ops = &xfs_sb_buf_ops; | |
2597 | break; | |
f67ca6ec DC |
2598 | #ifdef CONFIG_XFS_RT |
2599 | case XFS_BLFT_RTBITMAP_BUF: | |
2600 | case XFS_BLFT_RTSUMMARY_BUF: | |
bf85e099 DC |
2601 | /* no magic numbers for verification of RT buffers */ |
2602 | bp->b_ops = &xfs_rtbuf_ops; | |
f67ca6ec DC |
2603 | break; |
2604 | #endif /* CONFIG_XFS_RT */ | |
ee1a47ab | 2605 | default: |
61fe135c DC |
2606 | xfs_warn(mp, "Unknown buffer type %d!", |
2607 | xfs_blft_from_flags(buf_f)); | |
ee1a47ab CH |
2608 | break; |
2609 | } | |
040c52c0 BF |
2610 | |
2611 | /* | |
60a4a222 BF |
2612 | * Nothing else to do in the case of a NULL current LSN as this means |
2613 | * the buffer is more recent than the change in the log and will be | |
2614 | * skipped. | |
040c52c0 | 2615 | */ |
60a4a222 BF |
2616 | if (current_lsn == NULLCOMMITLSN) |
2617 | return; | |
2618 | ||
2619 | if (warnmsg) { | |
040c52c0 BF |
2620 | xfs_warn(mp, warnmsg); |
2621 | ASSERT(0); | |
2622 | } | |
60a4a222 BF |
2623 | |
2624 | /* | |
2625 | * We must update the metadata LSN of the buffer as it is written out to | |
2626 | * ensure that older transactions never replay over this one and corrupt | |
2627 | * the buffer. This can occur if log recovery is interrupted at some | |
2628 | * point after the current transaction completes, at which point a | |
2629 | * subsequent mount starts recovery from the beginning. | |
2630 | * | |
2631 | * Write verifiers update the metadata LSN from log items attached to | |
2632 | * the buffer. Therefore, initialize a bli purely to carry the LSN to | |
2633 | * the verifier. We'll clean it up in our ->iodone() callback. | |
2634 | */ | |
2635 | if (bp->b_ops) { | |
2636 | struct xfs_buf_log_item *bip; | |
2637 | ||
2638 | ASSERT(!bp->b_iodone || bp->b_iodone == xlog_recover_iodone); | |
2639 | bp->b_iodone = xlog_recover_iodone; | |
2640 | xfs_buf_item_init(bp, mp); | |
fb1755a6 | 2641 | bip = bp->b_log_item; |
60a4a222 BF |
2642 | bip->bli_item.li_lsn = current_lsn; |
2643 | } | |
1da177e4 LT |
2644 | } |
2645 | ||
d75afeb3 DC |
2646 | /* |
2647 | * Perform a 'normal' buffer recovery. Each logged region of the | |
2648 | * buffer should be copied over the corresponding region in the | |
2649 | * given buffer. The bitmap in the buf log format structure indicates | |
2650 | * where to place the logged data. | |
2651 | */ | |
2652 | STATIC void | |
2653 | xlog_recover_do_reg_buffer( | |
2654 | struct xfs_mount *mp, | |
2655 | xlog_recover_item_t *item, | |
2656 | struct xfs_buf *bp, | |
22db9af2 BF |
2657 | xfs_buf_log_format_t *buf_f, |
2658 | xfs_lsn_t current_lsn) | |
d75afeb3 DC |
2659 | { |
2660 | int i; | |
2661 | int bit; | |
2662 | int nbits; | |
eebf3cab | 2663 | xfs_failaddr_t fa; |
d75afeb3 DC |
2664 | |
2665 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); | |
2666 | ||
2667 | bit = 0; | |
2668 | i = 1; /* 0 is the buf format structure */ | |
2669 | while (1) { | |
2670 | bit = xfs_next_bit(buf_f->blf_data_map, | |
2671 | buf_f->blf_map_size, bit); | |
2672 | if (bit == -1) | |
2673 | break; | |
2674 | nbits = xfs_contig_bits(buf_f->blf_data_map, | |
2675 | buf_f->blf_map_size, bit); | |
2676 | ASSERT(nbits > 0); | |
2677 | ASSERT(item->ri_buf[i].i_addr != NULL); | |
2678 | ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0); | |
2679 | ASSERT(BBTOB(bp->b_io_length) >= | |
2680 | ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT)); | |
2681 | ||
709da6a6 DC |
2682 | /* |
2683 | * The dirty regions logged in the buffer, even though | |
2684 | * contiguous, may span multiple chunks. This is because the | |
2685 | * dirty region may span a physical page boundary in a buffer | |
2686 | * and hence be split into two separate vectors for writing into | |
2687 | * the log. Hence we need to trim nbits back to the length of | |
2688 | * the current region being copied out of the log. | |
2689 | */ | |
2690 | if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT)) | |
2691 | nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT; | |
2692 | ||
d75afeb3 DC |
2693 | /* |
2694 | * Do a sanity check if this is a dquot buffer. Just checking | |
2695 | * the first dquot in the buffer should do. XXXThis is | |
2696 | * probably a good thing to do for other buf types also. | |
2697 | */ | |
eebf3cab | 2698 | fa = NULL; |
d75afeb3 DC |
2699 | if (buf_f->blf_flags & |
2700 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { | |
2701 | if (item->ri_buf[i].i_addr == NULL) { | |
2702 | xfs_alert(mp, | |
2703 | "XFS: NULL dquot in %s.", __func__); | |
2704 | goto next; | |
2705 | } | |
2706 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { | |
2707 | xfs_alert(mp, | |
2708 | "XFS: dquot too small (%d) in %s.", | |
2709 | item->ri_buf[i].i_len, __func__); | |
2710 | goto next; | |
2711 | } | |
eebf3cab | 2712 | fa = xfs_dquot_verify(mp, item->ri_buf[i].i_addr, |
e381a0f6 | 2713 | -1, 0); |
eebf3cab DW |
2714 | if (fa) { |
2715 | xfs_alert(mp, | |
2716 | "dquot corrupt at %pS trying to replay into block 0x%llx", | |
2717 | fa, bp->b_bn); | |
d75afeb3 | 2718 | goto next; |
eebf3cab | 2719 | } |
d75afeb3 DC |
2720 | } |
2721 | ||
2722 | memcpy(xfs_buf_offset(bp, | |
2723 | (uint)bit << XFS_BLF_SHIFT), /* dest */ | |
2724 | item->ri_buf[i].i_addr, /* source */ | |
2725 | nbits<<XFS_BLF_SHIFT); /* length */ | |
2726 | next: | |
2727 | i++; | |
2728 | bit += nbits; | |
2729 | } | |
2730 | ||
2731 | /* Shouldn't be any more regions */ | |
2732 | ASSERT(i == item->ri_total); | |
2733 | ||
22db9af2 | 2734 | xlog_recover_validate_buf_type(mp, bp, buf_f, current_lsn); |
d75afeb3 DC |
2735 | } |
2736 | ||
1da177e4 LT |
2737 | /* |
2738 | * Perform a dquot buffer recovery. | |
8ba701ee | 2739 | * Simple algorithm: if we have found a QUOTAOFF log item of the same type |
1da177e4 LT |
2740 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. |
2741 | * Else, treat it as a regular buffer and do recovery. | |
ad3714b8 DC |
2742 | * |
2743 | * Return false if the buffer was tossed and true if we recovered the buffer to | |
2744 | * indicate to the caller if the buffer needs writing. | |
1da177e4 | 2745 | */ |
ad3714b8 | 2746 | STATIC bool |
1da177e4 | 2747 | xlog_recover_do_dquot_buffer( |
9a8d2fdb MT |
2748 | struct xfs_mount *mp, |
2749 | struct xlog *log, | |
2750 | struct xlog_recover_item *item, | |
2751 | struct xfs_buf *bp, | |
2752 | struct xfs_buf_log_format *buf_f) | |
1da177e4 LT |
2753 | { |
2754 | uint type; | |
2755 | ||
9abbc539 DC |
2756 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
2757 | ||
1da177e4 LT |
2758 | /* |
2759 | * Filesystems are required to send in quota flags at mount time. | |
2760 | */ | |
ad3714b8 DC |
2761 | if (!mp->m_qflags) |
2762 | return false; | |
1da177e4 LT |
2763 | |
2764 | type = 0; | |
c1155410 | 2765 | if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF) |
1da177e4 | 2766 | type |= XFS_DQ_USER; |
c1155410 | 2767 | if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF) |
c8ad20ff | 2768 | type |= XFS_DQ_PROJ; |
c1155410 | 2769 | if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF) |
1da177e4 LT |
2770 | type |= XFS_DQ_GROUP; |
2771 | /* | |
2772 | * This type of quotas was turned off, so ignore this buffer | |
2773 | */ | |
2774 | if (log->l_quotaoffs_flag & type) | |
ad3714b8 | 2775 | return false; |
1da177e4 | 2776 | |
22db9af2 | 2777 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f, NULLCOMMITLSN); |
ad3714b8 | 2778 | return true; |
1da177e4 LT |
2779 | } |
2780 | ||
2781 | /* | |
2782 | * This routine replays a modification made to a buffer at runtime. | |
2783 | * There are actually two types of buffer, regular and inode, which | |
2784 | * are handled differently. Inode buffers are handled differently | |
2785 | * in that we only recover a specific set of data from them, namely | |
2786 | * the inode di_next_unlinked fields. This is because all other inode | |
2787 | * data is actually logged via inode records and any data we replay | |
2788 | * here which overlaps that may be stale. | |
2789 | * | |
2790 | * When meta-data buffers are freed at run time we log a buffer item | |
c1155410 | 2791 | * with the XFS_BLF_CANCEL bit set to indicate that previous copies |
1da177e4 LT |
2792 | * of the buffer in the log should not be replayed at recovery time. |
2793 | * This is so that if the blocks covered by the buffer are reused for | |
2794 | * file data before we crash we don't end up replaying old, freed | |
2795 | * meta-data into a user's file. | |
2796 | * | |
2797 | * To handle the cancellation of buffer log items, we make two passes | |
2798 | * over the log during recovery. During the first we build a table of | |
2799 | * those buffers which have been cancelled, and during the second we | |
2800 | * only replay those buffers which do not have corresponding cancel | |
34be5ff3 | 2801 | * records in the table. See xlog_recover_buffer_pass[1,2] above |
1da177e4 LT |
2802 | * for more details on the implementation of the table of cancel records. |
2803 | */ | |
2804 | STATIC int | |
c9f71f5f | 2805 | xlog_recover_buffer_pass2( |
9a8d2fdb MT |
2806 | struct xlog *log, |
2807 | struct list_head *buffer_list, | |
50d5c8d8 DC |
2808 | struct xlog_recover_item *item, |
2809 | xfs_lsn_t current_lsn) | |
1da177e4 | 2810 | { |
4e0d5f92 | 2811 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
e2714bf8 | 2812 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
2813 | xfs_buf_t *bp; |
2814 | int error; | |
6ad112bf | 2815 | uint buf_flags; |
50d5c8d8 | 2816 | xfs_lsn_t lsn; |
1da177e4 | 2817 | |
c9f71f5f CH |
2818 | /* |
2819 | * In this pass we only want to recover all the buffers which have | |
2820 | * not been cancelled and are not cancellation buffers themselves. | |
2821 | */ | |
2822 | if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno, | |
2823 | buf_f->blf_len, buf_f->blf_flags)) { | |
2824 | trace_xfs_log_recover_buf_cancel(log, buf_f); | |
1da177e4 | 2825 | return 0; |
1da177e4 | 2826 | } |
c9f71f5f | 2827 | |
9abbc539 | 2828 | trace_xfs_log_recover_buf_recover(log, buf_f); |
1da177e4 | 2829 | |
a8acad70 | 2830 | buf_flags = 0; |
611c9946 DC |
2831 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) |
2832 | buf_flags |= XBF_UNMAPPED; | |
6ad112bf | 2833 | |
e2714bf8 | 2834 | bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len, |
c3f8fc73 | 2835 | buf_flags, NULL); |
ac4d6888 | 2836 | if (!bp) |
2451337d | 2837 | return -ENOMEM; |
e5702805 | 2838 | error = bp->b_error; |
5a52c2a5 | 2839 | if (error) { |
901796af | 2840 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)"); |
50d5c8d8 | 2841 | goto out_release; |
1da177e4 LT |
2842 | } |
2843 | ||
50d5c8d8 | 2844 | /* |
67dc288c | 2845 | * Recover the buffer only if we get an LSN from it and it's less than |
50d5c8d8 | 2846 | * the lsn of the transaction we are replaying. |
67dc288c DC |
2847 | * |
2848 | * Note that we have to be extremely careful of readahead here. | |
2849 | * Readahead does not attach verfiers to the buffers so if we don't | |
2850 | * actually do any replay after readahead because of the LSN we found | |
2851 | * in the buffer if more recent than that current transaction then we | |
2852 | * need to attach the verifier directly. Failure to do so can lead to | |
2853 | * future recovery actions (e.g. EFI and unlinked list recovery) can | |
2854 | * operate on the buffers and they won't get the verifier attached. This | |
2855 | * can lead to blocks on disk having the correct content but a stale | |
2856 | * CRC. | |
2857 | * | |
2858 | * It is safe to assume these clean buffers are currently up to date. | |
2859 | * If the buffer is dirtied by a later transaction being replayed, then | |
2860 | * the verifier will be reset to match whatever recover turns that | |
2861 | * buffer into. | |
50d5c8d8 DC |
2862 | */ |
2863 | lsn = xlog_recover_get_buf_lsn(mp, bp); | |
67dc288c | 2864 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { |
5cd9cee9 | 2865 | trace_xfs_log_recover_buf_skip(log, buf_f); |
22db9af2 | 2866 | xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN); |
50d5c8d8 | 2867 | goto out_release; |
67dc288c | 2868 | } |
50d5c8d8 | 2869 | |
e2714bf8 | 2870 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
1da177e4 | 2871 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); |
ad3714b8 DC |
2872 | if (error) |
2873 | goto out_release; | |
e2714bf8 | 2874 | } else if (buf_f->blf_flags & |
c1155410 | 2875 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
ad3714b8 DC |
2876 | bool dirty; |
2877 | ||
2878 | dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); | |
2879 | if (!dirty) | |
2880 | goto out_release; | |
1da177e4 | 2881 | } else { |
22db9af2 | 2882 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f, current_lsn); |
1da177e4 | 2883 | } |
1da177e4 LT |
2884 | |
2885 | /* | |
2886 | * Perform delayed write on the buffer. Asynchronous writes will be | |
2887 | * slower when taking into account all the buffers to be flushed. | |
2888 | * | |
2889 | * Also make sure that only inode buffers with good sizes stay in | |
2890 | * the buffer cache. The kernel moves inodes in buffers of 1 block | |
0f49efd8 | 2891 | * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode |
1da177e4 LT |
2892 | * buffers in the log can be a different size if the log was generated |
2893 | * by an older kernel using unclustered inode buffers or a newer kernel | |
2894 | * running with a different inode cluster size. Regardless, if the | |
9bb54cb5 | 2895 | * the inode buffer size isn't max(blocksize, mp->m_inode_cluster_size) |
0f49efd8 | 2896 | * for *our* value of mp->m_inode_cluster_size, then we need to keep |
1da177e4 LT |
2897 | * the buffer out of the buffer cache so that the buffer won't |
2898 | * overlap with future reads of those inodes. | |
2899 | */ | |
2900 | if (XFS_DINODE_MAGIC == | |
b53e675d | 2901 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
9bb54cb5 | 2902 | (BBTOB(bp->b_io_length) != max(log->l_mp->m_sb.sb_blocksize, |
c8ce540d | 2903 | (uint32_t)log->l_mp->m_inode_cluster_size))) { |
c867cb61 | 2904 | xfs_buf_stale(bp); |
c2b006c1 | 2905 | error = xfs_bwrite(bp); |
1da177e4 | 2906 | } else { |
ebad861b | 2907 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 2908 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 2909 | xfs_buf_delwri_queue(bp, buffer_list); |
1da177e4 LT |
2910 | } |
2911 | ||
50d5c8d8 | 2912 | out_release: |
c2b006c1 CH |
2913 | xfs_buf_relse(bp); |
2914 | return error; | |
1da177e4 LT |
2915 | } |
2916 | ||
638f4416 DC |
2917 | /* |
2918 | * Inode fork owner changes | |
2919 | * | |
2920 | * If we have been told that we have to reparent the inode fork, it's because an | |
2921 | * extent swap operation on a CRC enabled filesystem has been done and we are | |
2922 | * replaying it. We need to walk the BMBT of the appropriate fork and change the | |
2923 | * owners of it. | |
2924 | * | |
2925 | * The complexity here is that we don't have an inode context to work with, so | |
2926 | * after we've replayed the inode we need to instantiate one. This is where the | |
2927 | * fun begins. | |
2928 | * | |
2929 | * We are in the middle of log recovery, so we can't run transactions. That | |
2930 | * means we cannot use cache coherent inode instantiation via xfs_iget(), as | |
2931 | * that will result in the corresponding iput() running the inode through | |
2932 | * xfs_inactive(). If we've just replayed an inode core that changes the link | |
2933 | * count to zero (i.e. it's been unlinked), then xfs_inactive() will run | |
2934 | * transactions (bad!). | |
2935 | * | |
2936 | * So, to avoid this, we instantiate an inode directly from the inode core we've | |
2937 | * just recovered. We have the buffer still locked, and all we really need to | |
2938 | * instantiate is the inode core and the forks being modified. We can do this | |
2939 | * manually, then run the inode btree owner change, and then tear down the | |
2940 | * xfs_inode without having to run any transactions at all. | |
2941 | * | |
2942 | * Also, because we don't have a transaction context available here but need to | |
2943 | * gather all the buffers we modify for writeback so we pass the buffer_list | |
2944 | * instead for the operation to use. | |
2945 | */ | |
2946 | ||
2947 | STATIC int | |
2948 | xfs_recover_inode_owner_change( | |
2949 | struct xfs_mount *mp, | |
2950 | struct xfs_dinode *dip, | |
2951 | struct xfs_inode_log_format *in_f, | |
2952 | struct list_head *buffer_list) | |
2953 | { | |
2954 | struct xfs_inode *ip; | |
2955 | int error; | |
2956 | ||
2957 | ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)); | |
2958 | ||
2959 | ip = xfs_inode_alloc(mp, in_f->ilf_ino); | |
2960 | if (!ip) | |
2451337d | 2961 | return -ENOMEM; |
638f4416 DC |
2962 | |
2963 | /* instantiate the inode */ | |
3987848c | 2964 | xfs_inode_from_disk(ip, dip); |
638f4416 DC |
2965 | ASSERT(ip->i_d.di_version >= 3); |
2966 | ||
2967 | error = xfs_iformat_fork(ip, dip); | |
2968 | if (error) | |
2969 | goto out_free_ip; | |
2970 | ||
9cfb9b47 DW |
2971 | if (!xfs_inode_verify_forks(ip)) { |
2972 | error = -EFSCORRUPTED; | |
2973 | goto out_free_ip; | |
2974 | } | |
638f4416 DC |
2975 | |
2976 | if (in_f->ilf_fields & XFS_ILOG_DOWNER) { | |
2977 | ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT); | |
2978 | error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK, | |
2979 | ip->i_ino, buffer_list); | |
2980 | if (error) | |
2981 | goto out_free_ip; | |
2982 | } | |
2983 | ||
2984 | if (in_f->ilf_fields & XFS_ILOG_AOWNER) { | |
2985 | ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT); | |
2986 | error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK, | |
2987 | ip->i_ino, buffer_list); | |
2988 | if (error) | |
2989 | goto out_free_ip; | |
2990 | } | |
2991 | ||
2992 | out_free_ip: | |
2993 | xfs_inode_free(ip); | |
2994 | return error; | |
2995 | } | |
2996 | ||
1da177e4 | 2997 | STATIC int |
c9f71f5f | 2998 | xlog_recover_inode_pass2( |
9a8d2fdb MT |
2999 | struct xlog *log, |
3000 | struct list_head *buffer_list, | |
50d5c8d8 DC |
3001 | struct xlog_recover_item *item, |
3002 | xfs_lsn_t current_lsn) | |
1da177e4 | 3003 | { |
06b11321 | 3004 | struct xfs_inode_log_format *in_f; |
c9f71f5f | 3005 | xfs_mount_t *mp = log->l_mp; |
1da177e4 | 3006 | xfs_buf_t *bp; |
1da177e4 | 3007 | xfs_dinode_t *dip; |
1da177e4 | 3008 | int len; |
b2a922cd CH |
3009 | char *src; |
3010 | char *dest; | |
1da177e4 LT |
3011 | int error; |
3012 | int attr_index; | |
3013 | uint fields; | |
f8d55aa0 | 3014 | struct xfs_log_dinode *ldip; |
93848a99 | 3015 | uint isize; |
6d192a9b | 3016 | int need_free = 0; |
1da177e4 | 3017 | |
06b11321 | 3018 | if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) { |
4e0d5f92 | 3019 | in_f = item->ri_buf[0].i_addr; |
6d192a9b | 3020 | } else { |
06b11321 | 3021 | in_f = kmem_alloc(sizeof(struct xfs_inode_log_format), KM_SLEEP); |
6d192a9b TS |
3022 | need_free = 1; |
3023 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); | |
3024 | if (error) | |
3025 | goto error; | |
3026 | } | |
1da177e4 LT |
3027 | |
3028 | /* | |
3029 | * Inode buffers can be freed, look out for it, | |
3030 | * and do not replay the inode. | |
3031 | */ | |
a1941895 CH |
3032 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
3033 | in_f->ilf_len, 0)) { | |
6d192a9b | 3034 | error = 0; |
9abbc539 | 3035 | trace_xfs_log_recover_inode_cancel(log, in_f); |
6d192a9b TS |
3036 | goto error; |
3037 | } | |
9abbc539 | 3038 | trace_xfs_log_recover_inode_recover(log, in_f); |
1da177e4 | 3039 | |
c3f8fc73 | 3040 | bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0, |
93848a99 | 3041 | &xfs_inode_buf_ops); |
ac4d6888 | 3042 | if (!bp) { |
2451337d | 3043 | error = -ENOMEM; |
ac4d6888 CS |
3044 | goto error; |
3045 | } | |
e5702805 | 3046 | error = bp->b_error; |
5a52c2a5 | 3047 | if (error) { |
901796af | 3048 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)"); |
638f4416 | 3049 | goto out_release; |
1da177e4 | 3050 | } |
1da177e4 | 3051 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); |
88ee2df7 | 3052 | dip = xfs_buf_offset(bp, in_f->ilf_boffset); |
1da177e4 LT |
3053 | |
3054 | /* | |
3055 | * Make sure the place we're flushing out to really looks | |
3056 | * like an inode! | |
3057 | */ | |
69ef921b | 3058 | if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) { |
a0fa2b67 | 3059 | xfs_alert(mp, |
c9690043 | 3060 | "%s: Bad inode magic number, dip = "PTR_FMT", dino bp = "PTR_FMT", ino = %Ld", |
a0fa2b67 | 3061 | __func__, dip, bp, in_f->ilf_ino); |
c9f71f5f | 3062 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)", |
1da177e4 | 3063 | XFS_ERRLEVEL_LOW, mp); |
2451337d | 3064 | error = -EFSCORRUPTED; |
638f4416 | 3065 | goto out_release; |
1da177e4 | 3066 | } |
f8d55aa0 DC |
3067 | ldip = item->ri_buf[1].i_addr; |
3068 | if (unlikely(ldip->di_magic != XFS_DINODE_MAGIC)) { | |
a0fa2b67 | 3069 | xfs_alert(mp, |
c9690043 | 3070 | "%s: Bad inode log record, rec ptr "PTR_FMT", ino %Ld", |
a0fa2b67 | 3071 | __func__, item, in_f->ilf_ino); |
c9f71f5f | 3072 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)", |
1da177e4 | 3073 | XFS_ERRLEVEL_LOW, mp); |
2451337d | 3074 | error = -EFSCORRUPTED; |
638f4416 | 3075 | goto out_release; |
1da177e4 LT |
3076 | } |
3077 | ||
50d5c8d8 DC |
3078 | /* |
3079 | * If the inode has an LSN in it, recover the inode only if it's less | |
638f4416 DC |
3080 | * than the lsn of the transaction we are replaying. Note: we still |
3081 | * need to replay an owner change even though the inode is more recent | |
3082 | * than the transaction as there is no guarantee that all the btree | |
3083 | * blocks are more recent than this transaction, too. | |
50d5c8d8 DC |
3084 | */ |
3085 | if (dip->di_version >= 3) { | |
3086 | xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn); | |
3087 | ||
3088 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { | |
3089 | trace_xfs_log_recover_inode_skip(log, in_f); | |
3090 | error = 0; | |
638f4416 | 3091 | goto out_owner_change; |
50d5c8d8 DC |
3092 | } |
3093 | } | |
3094 | ||
e60896d8 DC |
3095 | /* |
3096 | * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes | |
3097 | * are transactional and if ordering is necessary we can determine that | |
3098 | * more accurately by the LSN field in the V3 inode core. Don't trust | |
3099 | * the inode versions we might be changing them here - use the | |
3100 | * superblock flag to determine whether we need to look at di_flushiter | |
3101 | * to skip replay when the on disk inode is newer than the log one | |
3102 | */ | |
3103 | if (!xfs_sb_version_hascrc(&mp->m_sb) && | |
f8d55aa0 | 3104 | ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) { |
1da177e4 LT |
3105 | /* |
3106 | * Deal with the wrap case, DI_MAX_FLUSH is less | |
3107 | * than smaller numbers | |
3108 | */ | |
81591fe2 | 3109 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
f8d55aa0 | 3110 | ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
1da177e4 LT |
3111 | /* do nothing */ |
3112 | } else { | |
9abbc539 | 3113 | trace_xfs_log_recover_inode_skip(log, in_f); |
6d192a9b | 3114 | error = 0; |
638f4416 | 3115 | goto out_release; |
1da177e4 LT |
3116 | } |
3117 | } | |
e60896d8 | 3118 | |
1da177e4 | 3119 | /* Take the opportunity to reset the flush iteration count */ |
f8d55aa0 | 3120 | ldip->di_flushiter = 0; |
1da177e4 | 3121 | |
f8d55aa0 DC |
3122 | if (unlikely(S_ISREG(ldip->di_mode))) { |
3123 | if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) && | |
3124 | (ldip->di_format != XFS_DINODE_FMT_BTREE)) { | |
c9f71f5f | 3125 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)", |
2551a530 DW |
3126 | XFS_ERRLEVEL_LOW, mp, ldip, |
3127 | sizeof(*ldip)); | |
a0fa2b67 | 3128 | xfs_alert(mp, |
c9690043 DW |
3129 | "%s: Bad regular inode log record, rec ptr "PTR_FMT", " |
3130 | "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld", | |
a0fa2b67 | 3131 | __func__, item, dip, bp, in_f->ilf_ino); |
2451337d | 3132 | error = -EFSCORRUPTED; |
638f4416 | 3133 | goto out_release; |
1da177e4 | 3134 | } |
f8d55aa0 DC |
3135 | } else if (unlikely(S_ISDIR(ldip->di_mode))) { |
3136 | if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) && | |
3137 | (ldip->di_format != XFS_DINODE_FMT_BTREE) && | |
3138 | (ldip->di_format != XFS_DINODE_FMT_LOCAL)) { | |
c9f71f5f | 3139 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)", |
2551a530 DW |
3140 | XFS_ERRLEVEL_LOW, mp, ldip, |
3141 | sizeof(*ldip)); | |
a0fa2b67 | 3142 | xfs_alert(mp, |
c9690043 DW |
3143 | "%s: Bad dir inode log record, rec ptr "PTR_FMT", " |
3144 | "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld", | |
a0fa2b67 | 3145 | __func__, item, dip, bp, in_f->ilf_ino); |
2451337d | 3146 | error = -EFSCORRUPTED; |
638f4416 | 3147 | goto out_release; |
1da177e4 LT |
3148 | } |
3149 | } | |
f8d55aa0 | 3150 | if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){ |
c9f71f5f | 3151 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)", |
2551a530 DW |
3152 | XFS_ERRLEVEL_LOW, mp, ldip, |
3153 | sizeof(*ldip)); | |
a0fa2b67 | 3154 | xfs_alert(mp, |
c9690043 DW |
3155 | "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", " |
3156 | "dino bp "PTR_FMT", ino %Ld, total extents = %d, nblocks = %Ld", | |
a0fa2b67 | 3157 | __func__, item, dip, bp, in_f->ilf_ino, |
f8d55aa0 DC |
3158 | ldip->di_nextents + ldip->di_anextents, |
3159 | ldip->di_nblocks); | |
2451337d | 3160 | error = -EFSCORRUPTED; |
638f4416 | 3161 | goto out_release; |
1da177e4 | 3162 | } |
f8d55aa0 | 3163 | if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) { |
c9f71f5f | 3164 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)", |
2551a530 DW |
3165 | XFS_ERRLEVEL_LOW, mp, ldip, |
3166 | sizeof(*ldip)); | |
a0fa2b67 | 3167 | xfs_alert(mp, |
c9690043 DW |
3168 | "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", " |
3169 | "dino bp "PTR_FMT", ino %Ld, forkoff 0x%x", __func__, | |
f8d55aa0 | 3170 | item, dip, bp, in_f->ilf_ino, ldip->di_forkoff); |
2451337d | 3171 | error = -EFSCORRUPTED; |
638f4416 | 3172 | goto out_release; |
1da177e4 | 3173 | } |
f8d55aa0 | 3174 | isize = xfs_log_dinode_size(ldip->di_version); |
93848a99 | 3175 | if (unlikely(item->ri_buf[1].i_len > isize)) { |
c9f71f5f | 3176 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)", |
2551a530 DW |
3177 | XFS_ERRLEVEL_LOW, mp, ldip, |
3178 | sizeof(*ldip)); | |
a0fa2b67 | 3179 | xfs_alert(mp, |
c9690043 | 3180 | "%s: Bad inode log record length %d, rec ptr "PTR_FMT, |
a0fa2b67 | 3181 | __func__, item->ri_buf[1].i_len, item); |
2451337d | 3182 | error = -EFSCORRUPTED; |
638f4416 | 3183 | goto out_release; |
1da177e4 LT |
3184 | } |
3185 | ||
3987848c DC |
3186 | /* recover the log dinode inode into the on disk inode */ |
3187 | xfs_log_dinode_to_disk(ldip, dip); | |
1da177e4 | 3188 | |
1da177e4 | 3189 | fields = in_f->ilf_fields; |
42b67dc6 | 3190 | if (fields & XFS_ILOG_DEV) |
81591fe2 | 3191 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
1da177e4 LT |
3192 | |
3193 | if (in_f->ilf_size == 2) | |
638f4416 | 3194 | goto out_owner_change; |
1da177e4 LT |
3195 | len = item->ri_buf[2].i_len; |
3196 | src = item->ri_buf[2].i_addr; | |
3197 | ASSERT(in_f->ilf_size <= 4); | |
3198 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); | |
3199 | ASSERT(!(fields & XFS_ILOG_DFORK) || | |
3200 | (len == in_f->ilf_dsize)); | |
3201 | ||
3202 | switch (fields & XFS_ILOG_DFORK) { | |
3203 | case XFS_ILOG_DDATA: | |
3204 | case XFS_ILOG_DEXT: | |
81591fe2 | 3205 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
1da177e4 LT |
3206 | break; |
3207 | ||
3208 | case XFS_ILOG_DBROOT: | |
7cc95a82 | 3209 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
81591fe2 | 3210 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
1da177e4 LT |
3211 | XFS_DFORK_DSIZE(dip, mp)); |
3212 | break; | |
3213 | ||
3214 | default: | |
3215 | /* | |
3216 | * There are no data fork flags set. | |
3217 | */ | |
3218 | ASSERT((fields & XFS_ILOG_DFORK) == 0); | |
3219 | break; | |
3220 | } | |
3221 | ||
3222 | /* | |
3223 | * If we logged any attribute data, recover it. There may or | |
3224 | * may not have been any other non-core data logged in this | |
3225 | * transaction. | |
3226 | */ | |
3227 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
3228 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { | |
3229 | attr_index = 3; | |
3230 | } else { | |
3231 | attr_index = 2; | |
3232 | } | |
3233 | len = item->ri_buf[attr_index].i_len; | |
3234 | src = item->ri_buf[attr_index].i_addr; | |
3235 | ASSERT(len == in_f->ilf_asize); | |
3236 | ||
3237 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
3238 | case XFS_ILOG_ADATA: | |
3239 | case XFS_ILOG_AEXT: | |
3240 | dest = XFS_DFORK_APTR(dip); | |
3241 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); | |
3242 | memcpy(dest, src, len); | |
3243 | break; | |
3244 | ||
3245 | case XFS_ILOG_ABROOT: | |
3246 | dest = XFS_DFORK_APTR(dip); | |
7cc95a82 CH |
3247 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
3248 | len, (xfs_bmdr_block_t*)dest, | |
1da177e4 LT |
3249 | XFS_DFORK_ASIZE(dip, mp)); |
3250 | break; | |
3251 | ||
3252 | default: | |
a0fa2b67 | 3253 | xfs_warn(log->l_mp, "%s: Invalid flag", __func__); |
1da177e4 | 3254 | ASSERT(0); |
2451337d | 3255 | error = -EIO; |
638f4416 | 3256 | goto out_release; |
1da177e4 LT |
3257 | } |
3258 | } | |
3259 | ||
638f4416 | 3260 | out_owner_change: |
dc1baa71 ES |
3261 | /* Recover the swapext owner change unless inode has been deleted */ |
3262 | if ((in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) && | |
3263 | (dip->di_mode != 0)) | |
638f4416 DC |
3264 | error = xfs_recover_inode_owner_change(mp, dip, in_f, |
3265 | buffer_list); | |
93848a99 CH |
3266 | /* re-generate the checksum. */ |
3267 | xfs_dinode_calc_crc(log->l_mp, dip); | |
3268 | ||
ebad861b | 3269 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 3270 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 3271 | xfs_buf_delwri_queue(bp, buffer_list); |
50d5c8d8 DC |
3272 | |
3273 | out_release: | |
61551f1e | 3274 | xfs_buf_relse(bp); |
6d192a9b TS |
3275 | error: |
3276 | if (need_free) | |
f0e2d93c | 3277 | kmem_free(in_f); |
b474c7ae | 3278 | return error; |
1da177e4 LT |
3279 | } |
3280 | ||
3281 | /* | |
9a8d2fdb | 3282 | * Recover QUOTAOFF records. We simply make a note of it in the xlog |
1da177e4 LT |
3283 | * structure, so that we know not to do any dquot item or dquot buffer recovery, |
3284 | * of that type. | |
3285 | */ | |
3286 | STATIC int | |
c9f71f5f | 3287 | xlog_recover_quotaoff_pass1( |
9a8d2fdb MT |
3288 | struct xlog *log, |
3289 | struct xlog_recover_item *item) | |
1da177e4 | 3290 | { |
c9f71f5f | 3291 | xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr; |
1da177e4 LT |
3292 | ASSERT(qoff_f); |
3293 | ||
3294 | /* | |
3295 | * The logitem format's flag tells us if this was user quotaoff, | |
77a7cce4 | 3296 | * group/project quotaoff or both. |
1da177e4 LT |
3297 | */ |
3298 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) | |
3299 | log->l_quotaoffs_flag |= XFS_DQ_USER; | |
77a7cce4 NS |
3300 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
3301 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; | |
1da177e4 LT |
3302 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
3303 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; | |
3304 | ||
d99831ff | 3305 | return 0; |
1da177e4 LT |
3306 | } |
3307 | ||
3308 | /* | |
3309 | * Recover a dquot record | |
3310 | */ | |
3311 | STATIC int | |
c9f71f5f | 3312 | xlog_recover_dquot_pass2( |
9a8d2fdb MT |
3313 | struct xlog *log, |
3314 | struct list_head *buffer_list, | |
50d5c8d8 DC |
3315 | struct xlog_recover_item *item, |
3316 | xfs_lsn_t current_lsn) | |
1da177e4 | 3317 | { |
c9f71f5f | 3318 | xfs_mount_t *mp = log->l_mp; |
1da177e4 LT |
3319 | xfs_buf_t *bp; |
3320 | struct xfs_disk_dquot *ddq, *recddq; | |
eebf3cab | 3321 | xfs_failaddr_t fa; |
1da177e4 LT |
3322 | int error; |
3323 | xfs_dq_logformat_t *dq_f; | |
3324 | uint type; | |
3325 | ||
1da177e4 LT |
3326 | |
3327 | /* | |
3328 | * Filesystems are required to send in quota flags at mount time. | |
3329 | */ | |
3330 | if (mp->m_qflags == 0) | |
d99831ff | 3331 | return 0; |
1da177e4 | 3332 | |
4e0d5f92 CH |
3333 | recddq = item->ri_buf[1].i_addr; |
3334 | if (recddq == NULL) { | |
a0fa2b67 | 3335 | xfs_alert(log->l_mp, "NULL dquot in %s.", __func__); |
2451337d | 3336 | return -EIO; |
0c5e1ce8 | 3337 | } |
8ec6dba2 | 3338 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
a0fa2b67 | 3339 | xfs_alert(log->l_mp, "dquot too small (%d) in %s.", |
0c5e1ce8 | 3340 | item->ri_buf[1].i_len, __func__); |
2451337d | 3341 | return -EIO; |
0c5e1ce8 CH |
3342 | } |
3343 | ||
1da177e4 LT |
3344 | /* |
3345 | * This type of quotas was turned off, so ignore this record. | |
3346 | */ | |
b53e675d | 3347 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
1da177e4 LT |
3348 | ASSERT(type); |
3349 | if (log->l_quotaoffs_flag & type) | |
d99831ff | 3350 | return 0; |
1da177e4 LT |
3351 | |
3352 | /* | |
3353 | * At this point we know that quota was _not_ turned off. | |
3354 | * Since the mount flags are not indicating to us otherwise, this | |
3355 | * must mean that quota is on, and the dquot needs to be replayed. | |
3356 | * Remember that we may not have fully recovered the superblock yet, | |
3357 | * so we can't do the usual trick of looking at the SB quota bits. | |
3358 | * | |
3359 | * The other possibility, of course, is that the quota subsystem was | |
3360 | * removed since the last mount - ENOSYS. | |
3361 | */ | |
4e0d5f92 | 3362 | dq_f = item->ri_buf[0].i_addr; |
1da177e4 | 3363 | ASSERT(dq_f); |
e381a0f6 | 3364 | fa = xfs_dquot_verify(mp, recddq, dq_f->qlf_id, 0); |
eebf3cab DW |
3365 | if (fa) { |
3366 | xfs_alert(mp, "corrupt dquot ID 0x%x in log at %pS", | |
3367 | dq_f->qlf_id, fa); | |
2451337d | 3368 | return -EIO; |
eebf3cab | 3369 | } |
1da177e4 LT |
3370 | ASSERT(dq_f->qlf_len == 1); |
3371 | ||
ad3714b8 DC |
3372 | /* |
3373 | * At this point we are assuming that the dquots have been allocated | |
3374 | * and hence the buffer has valid dquots stamped in it. It should, | |
3375 | * therefore, pass verifier validation. If the dquot is bad, then the | |
3376 | * we'll return an error here, so we don't need to specifically check | |
3377 | * the dquot in the buffer after the verifier has run. | |
3378 | */ | |
7ca790a5 | 3379 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno, |
c3f8fc73 | 3380 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp, |
ad3714b8 | 3381 | &xfs_dquot_buf_ops); |
7ca790a5 | 3382 | if (error) |
1da177e4 | 3383 | return error; |
7ca790a5 | 3384 | |
1da177e4 | 3385 | ASSERT(bp); |
88ee2df7 | 3386 | ddq = xfs_buf_offset(bp, dq_f->qlf_boffset); |
1da177e4 | 3387 | |
50d5c8d8 DC |
3388 | /* |
3389 | * If the dquot has an LSN in it, recover the dquot only if it's less | |
3390 | * than the lsn of the transaction we are replaying. | |
3391 | */ | |
3392 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
3393 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq; | |
3394 | xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn); | |
3395 | ||
3396 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { | |
3397 | goto out_release; | |
3398 | } | |
3399 | } | |
3400 | ||
1da177e4 | 3401 | memcpy(ddq, recddq, item->ri_buf[1].i_len); |
6fcdc59d DC |
3402 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
3403 | xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk), | |
3404 | XFS_DQUOT_CRC_OFF); | |
3405 | } | |
1da177e4 LT |
3406 | |
3407 | ASSERT(dq_f->qlf_size == 2); | |
ebad861b | 3408 | ASSERT(bp->b_target->bt_mount == mp); |
cb669ca5 | 3409 | bp->b_iodone = xlog_recover_iodone; |
43ff2122 | 3410 | xfs_buf_delwri_queue(bp, buffer_list); |
1da177e4 | 3411 | |
50d5c8d8 DC |
3412 | out_release: |
3413 | xfs_buf_relse(bp); | |
3414 | return 0; | |
1da177e4 LT |
3415 | } |
3416 | ||
3417 | /* | |
3418 | * This routine is called to create an in-core extent free intent | |
3419 | * item from the efi format structure which was logged on disk. | |
3420 | * It allocates an in-core efi, copies the extents from the format | |
3421 | * structure into it, and adds the efi to the AIL with the given | |
3422 | * LSN. | |
3423 | */ | |
6d192a9b | 3424 | STATIC int |
c9f71f5f | 3425 | xlog_recover_efi_pass2( |
9a8d2fdb MT |
3426 | struct xlog *log, |
3427 | struct xlog_recover_item *item, | |
3428 | xfs_lsn_t lsn) | |
1da177e4 | 3429 | { |
e32a1d1f BF |
3430 | int error; |
3431 | struct xfs_mount *mp = log->l_mp; | |
3432 | struct xfs_efi_log_item *efip; | |
3433 | struct xfs_efi_log_format *efi_formatp; | |
1da177e4 | 3434 | |
4e0d5f92 | 3435 | efi_formatp = item->ri_buf[0].i_addr; |
1da177e4 | 3436 | |
1da177e4 | 3437 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); |
e32a1d1f BF |
3438 | error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format); |
3439 | if (error) { | |
6d192a9b TS |
3440 | xfs_efi_item_free(efip); |
3441 | return error; | |
3442 | } | |
b199c8a4 | 3443 | atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents); |
1da177e4 | 3444 | |
57e80956 | 3445 | spin_lock(&log->l_ailp->ail_lock); |
1da177e4 | 3446 | /* |
e32a1d1f BF |
3447 | * The EFI has two references. One for the EFD and one for EFI to ensure |
3448 | * it makes it into the AIL. Insert the EFI into the AIL directly and | |
3449 | * drop the EFI reference. Note that xfs_trans_ail_update() drops the | |
3450 | * AIL lock. | |
1da177e4 | 3451 | */ |
e6059949 | 3452 | xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn); |
e32a1d1f | 3453 | xfs_efi_release(efip); |
6d192a9b | 3454 | return 0; |
1da177e4 LT |
3455 | } |
3456 | ||
3457 | ||
3458 | /* | |
e32a1d1f BF |
3459 | * This routine is called when an EFD format structure is found in a committed |
3460 | * transaction in the log. Its purpose is to cancel the corresponding EFI if it | |
3461 | * was still in the log. To do this it searches the AIL for the EFI with an id | |
3462 | * equal to that in the EFD format structure. If we find it we drop the EFD | |
3463 | * reference, which removes the EFI from the AIL and frees it. | |
1da177e4 | 3464 | */ |
c9f71f5f CH |
3465 | STATIC int |
3466 | xlog_recover_efd_pass2( | |
9a8d2fdb MT |
3467 | struct xlog *log, |
3468 | struct xlog_recover_item *item) | |
1da177e4 | 3469 | { |
1da177e4 LT |
3470 | xfs_efd_log_format_t *efd_formatp; |
3471 | xfs_efi_log_item_t *efip = NULL; | |
3472 | xfs_log_item_t *lip; | |
c8ce540d | 3473 | uint64_t efi_id; |
27d8d5fe | 3474 | struct xfs_ail_cursor cur; |
783a2f65 | 3475 | struct xfs_ail *ailp = log->l_ailp; |
1da177e4 | 3476 | |
4e0d5f92 | 3477 | efd_formatp = item->ri_buf[0].i_addr; |
6d192a9b TS |
3478 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
3479 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || | |
3480 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + | |
3481 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); | |
1da177e4 LT |
3482 | efi_id = efd_formatp->efd_efi_id; |
3483 | ||
3484 | /* | |
e32a1d1f BF |
3485 | * Search for the EFI with the id in the EFD format structure in the |
3486 | * AIL. | |
1da177e4 | 3487 | */ |
57e80956 | 3488 | spin_lock(&ailp->ail_lock); |
a9c21c1b | 3489 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
1da177e4 LT |
3490 | while (lip != NULL) { |
3491 | if (lip->li_type == XFS_LI_EFI) { | |
3492 | efip = (xfs_efi_log_item_t *)lip; | |
3493 | if (efip->efi_format.efi_id == efi_id) { | |
3494 | /* | |
e32a1d1f BF |
3495 | * Drop the EFD reference to the EFI. This |
3496 | * removes the EFI from the AIL and frees it. | |
1da177e4 | 3497 | */ |
57e80956 | 3498 | spin_unlock(&ailp->ail_lock); |
e32a1d1f | 3499 | xfs_efi_release(efip); |
57e80956 | 3500 | spin_lock(&ailp->ail_lock); |
27d8d5fe | 3501 | break; |
1da177e4 LT |
3502 | } |
3503 | } | |
a9c21c1b | 3504 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 3505 | } |
e32a1d1f | 3506 | |
e4a1e29c | 3507 | xfs_trans_ail_cursor_done(&cur); |
57e80956 | 3508 | spin_unlock(&ailp->ail_lock); |
c9f71f5f CH |
3509 | |
3510 | return 0; | |
1da177e4 LT |
3511 | } |
3512 | ||
9e88b5d8 DW |
3513 | /* |
3514 | * This routine is called to create an in-core extent rmap update | |
3515 | * item from the rui format structure which was logged on disk. | |
3516 | * It allocates an in-core rui, copies the extents from the format | |
3517 | * structure into it, and adds the rui to the AIL with the given | |
3518 | * LSN. | |
3519 | */ | |
3520 | STATIC int | |
3521 | xlog_recover_rui_pass2( | |
3522 | struct xlog *log, | |
3523 | struct xlog_recover_item *item, | |
3524 | xfs_lsn_t lsn) | |
3525 | { | |
3526 | int error; | |
3527 | struct xfs_mount *mp = log->l_mp; | |
3528 | struct xfs_rui_log_item *ruip; | |
3529 | struct xfs_rui_log_format *rui_formatp; | |
3530 | ||
3531 | rui_formatp = item->ri_buf[0].i_addr; | |
3532 | ||
3533 | ruip = xfs_rui_init(mp, rui_formatp->rui_nextents); | |
3534 | error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format); | |
3535 | if (error) { | |
3536 | xfs_rui_item_free(ruip); | |
3537 | return error; | |
3538 | } | |
3539 | atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents); | |
3540 | ||
57e80956 | 3541 | spin_lock(&log->l_ailp->ail_lock); |
9e88b5d8 DW |
3542 | /* |
3543 | * The RUI has two references. One for the RUD and one for RUI to ensure | |
3544 | * it makes it into the AIL. Insert the RUI into the AIL directly and | |
3545 | * drop the RUI reference. Note that xfs_trans_ail_update() drops the | |
3546 | * AIL lock. | |
3547 | */ | |
3548 | xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn); | |
3549 | xfs_rui_release(ruip); | |
3550 | return 0; | |
3551 | } | |
3552 | ||
3553 | ||
3554 | /* | |
3555 | * This routine is called when an RUD format structure is found in a committed | |
3556 | * transaction in the log. Its purpose is to cancel the corresponding RUI if it | |
3557 | * was still in the log. To do this it searches the AIL for the RUI with an id | |
3558 | * equal to that in the RUD format structure. If we find it we drop the RUD | |
3559 | * reference, which removes the RUI from the AIL and frees it. | |
3560 | */ | |
3561 | STATIC int | |
3562 | xlog_recover_rud_pass2( | |
3563 | struct xlog *log, | |
3564 | struct xlog_recover_item *item) | |
3565 | { | |
3566 | struct xfs_rud_log_format *rud_formatp; | |
3567 | struct xfs_rui_log_item *ruip = NULL; | |
3568 | struct xfs_log_item *lip; | |
c8ce540d | 3569 | uint64_t rui_id; |
9e88b5d8 DW |
3570 | struct xfs_ail_cursor cur; |
3571 | struct xfs_ail *ailp = log->l_ailp; | |
3572 | ||
3573 | rud_formatp = item->ri_buf[0].i_addr; | |
722e2517 | 3574 | ASSERT(item->ri_buf[0].i_len == sizeof(struct xfs_rud_log_format)); |
9e88b5d8 DW |
3575 | rui_id = rud_formatp->rud_rui_id; |
3576 | ||
3577 | /* | |
3578 | * Search for the RUI with the id in the RUD format structure in the | |
3579 | * AIL. | |
3580 | */ | |
57e80956 | 3581 | spin_lock(&ailp->ail_lock); |
9e88b5d8 DW |
3582 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
3583 | while (lip != NULL) { | |
3584 | if (lip->li_type == XFS_LI_RUI) { | |
3585 | ruip = (struct xfs_rui_log_item *)lip; | |
3586 | if (ruip->rui_format.rui_id == rui_id) { | |
3587 | /* | |
3588 | * Drop the RUD reference to the RUI. This | |
3589 | * removes the RUI from the AIL and frees it. | |
3590 | */ | |
57e80956 | 3591 | spin_unlock(&ailp->ail_lock); |
9e88b5d8 | 3592 | xfs_rui_release(ruip); |
57e80956 | 3593 | spin_lock(&ailp->ail_lock); |
9e88b5d8 DW |
3594 | break; |
3595 | } | |
3596 | } | |
3597 | lip = xfs_trans_ail_cursor_next(ailp, &cur); | |
3598 | } | |
3599 | ||
3600 | xfs_trans_ail_cursor_done(&cur); | |
57e80956 | 3601 | spin_unlock(&ailp->ail_lock); |
9e88b5d8 DW |
3602 | |
3603 | return 0; | |
3604 | } | |
3605 | ||
f997ee21 DW |
3606 | /* |
3607 | * Copy an CUI format buffer from the given buf, and into the destination | |
3608 | * CUI format structure. The CUI/CUD items were designed not to need any | |
3609 | * special alignment handling. | |
3610 | */ | |
3611 | static int | |
3612 | xfs_cui_copy_format( | |
3613 | struct xfs_log_iovec *buf, | |
3614 | struct xfs_cui_log_format *dst_cui_fmt) | |
3615 | { | |
3616 | struct xfs_cui_log_format *src_cui_fmt; | |
3617 | uint len; | |
3618 | ||
3619 | src_cui_fmt = buf->i_addr; | |
3620 | len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents); | |
3621 | ||
3622 | if (buf->i_len == len) { | |
3623 | memcpy(dst_cui_fmt, src_cui_fmt, len); | |
3624 | return 0; | |
3625 | } | |
3626 | return -EFSCORRUPTED; | |
3627 | } | |
3628 | ||
3629 | /* | |
3630 | * This routine is called to create an in-core extent refcount update | |
3631 | * item from the cui format structure which was logged on disk. | |
3632 | * It allocates an in-core cui, copies the extents from the format | |
3633 | * structure into it, and adds the cui to the AIL with the given | |
3634 | * LSN. | |
3635 | */ | |
3636 | STATIC int | |
3637 | xlog_recover_cui_pass2( | |
3638 | struct xlog *log, | |
3639 | struct xlog_recover_item *item, | |
3640 | xfs_lsn_t lsn) | |
3641 | { | |
3642 | int error; | |
3643 | struct xfs_mount *mp = log->l_mp; | |
3644 | struct xfs_cui_log_item *cuip; | |
3645 | struct xfs_cui_log_format *cui_formatp; | |
3646 | ||
3647 | cui_formatp = item->ri_buf[0].i_addr; | |
3648 | ||
3649 | cuip = xfs_cui_init(mp, cui_formatp->cui_nextents); | |
3650 | error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format); | |
3651 | if (error) { | |
3652 | xfs_cui_item_free(cuip); | |
3653 | return error; | |
3654 | } | |
3655 | atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents); | |
3656 | ||
57e80956 | 3657 | spin_lock(&log->l_ailp->ail_lock); |
f997ee21 DW |
3658 | /* |
3659 | * The CUI has two references. One for the CUD and one for CUI to ensure | |
3660 | * it makes it into the AIL. Insert the CUI into the AIL directly and | |
3661 | * drop the CUI reference. Note that xfs_trans_ail_update() drops the | |
3662 | * AIL lock. | |
3663 | */ | |
3664 | xfs_trans_ail_update(log->l_ailp, &cuip->cui_item, lsn); | |
3665 | xfs_cui_release(cuip); | |
3666 | return 0; | |
3667 | } | |
3668 | ||
3669 | ||
3670 | /* | |
3671 | * This routine is called when an CUD format structure is found in a committed | |
3672 | * transaction in the log. Its purpose is to cancel the corresponding CUI if it | |
3673 | * was still in the log. To do this it searches the AIL for the CUI with an id | |
3674 | * equal to that in the CUD format structure. If we find it we drop the CUD | |
3675 | * reference, which removes the CUI from the AIL and frees it. | |
3676 | */ | |
3677 | STATIC int | |
3678 | xlog_recover_cud_pass2( | |
3679 | struct xlog *log, | |
3680 | struct xlog_recover_item *item) | |
3681 | { | |
3682 | struct xfs_cud_log_format *cud_formatp; | |
3683 | struct xfs_cui_log_item *cuip = NULL; | |
3684 | struct xfs_log_item *lip; | |
c8ce540d | 3685 | uint64_t cui_id; |
f997ee21 DW |
3686 | struct xfs_ail_cursor cur; |
3687 | struct xfs_ail *ailp = log->l_ailp; | |
3688 | ||
3689 | cud_formatp = item->ri_buf[0].i_addr; | |
3690 | if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format)) | |
3691 | return -EFSCORRUPTED; | |
3692 | cui_id = cud_formatp->cud_cui_id; | |
3693 | ||
3694 | /* | |
3695 | * Search for the CUI with the id in the CUD format structure in the | |
3696 | * AIL. | |
3697 | */ | |
57e80956 | 3698 | spin_lock(&ailp->ail_lock); |
f997ee21 DW |
3699 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
3700 | while (lip != NULL) { | |
3701 | if (lip->li_type == XFS_LI_CUI) { | |
3702 | cuip = (struct xfs_cui_log_item *)lip; | |
3703 | if (cuip->cui_format.cui_id == cui_id) { | |
3704 | /* | |
3705 | * Drop the CUD reference to the CUI. This | |
3706 | * removes the CUI from the AIL and frees it. | |
3707 | */ | |
57e80956 | 3708 | spin_unlock(&ailp->ail_lock); |
f997ee21 | 3709 | xfs_cui_release(cuip); |
57e80956 | 3710 | spin_lock(&ailp->ail_lock); |
f997ee21 DW |
3711 | break; |
3712 | } | |
3713 | } | |
3714 | lip = xfs_trans_ail_cursor_next(ailp, &cur); | |
3715 | } | |
3716 | ||
3717 | xfs_trans_ail_cursor_done(&cur); | |
57e80956 | 3718 | spin_unlock(&ailp->ail_lock); |
f997ee21 DW |
3719 | |
3720 | return 0; | |
3721 | } | |
3722 | ||
77d61fe4 DW |
3723 | /* |
3724 | * Copy an BUI format buffer from the given buf, and into the destination | |
3725 | * BUI format structure. The BUI/BUD items were designed not to need any | |
3726 | * special alignment handling. | |
3727 | */ | |
3728 | static int | |
3729 | xfs_bui_copy_format( | |
3730 | struct xfs_log_iovec *buf, | |
3731 | struct xfs_bui_log_format *dst_bui_fmt) | |
3732 | { | |
3733 | struct xfs_bui_log_format *src_bui_fmt; | |
3734 | uint len; | |
3735 | ||
3736 | src_bui_fmt = buf->i_addr; | |
3737 | len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents); | |
3738 | ||
3739 | if (buf->i_len == len) { | |
3740 | memcpy(dst_bui_fmt, src_bui_fmt, len); | |
3741 | return 0; | |
3742 | } | |
3743 | return -EFSCORRUPTED; | |
3744 | } | |
3745 | ||
3746 | /* | |
3747 | * This routine is called to create an in-core extent bmap update | |
3748 | * item from the bui format structure which was logged on disk. | |
3749 | * It allocates an in-core bui, copies the extents from the format | |
3750 | * structure into it, and adds the bui to the AIL with the given | |
3751 | * LSN. | |
3752 | */ | |
3753 | STATIC int | |
3754 | xlog_recover_bui_pass2( | |
3755 | struct xlog *log, | |
3756 | struct xlog_recover_item *item, | |
3757 | xfs_lsn_t lsn) | |
3758 | { | |
3759 | int error; | |
3760 | struct xfs_mount *mp = log->l_mp; | |
3761 | struct xfs_bui_log_item *buip; | |
3762 | struct xfs_bui_log_format *bui_formatp; | |
3763 | ||
3764 | bui_formatp = item->ri_buf[0].i_addr; | |
3765 | ||
3766 | if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) | |
3767 | return -EFSCORRUPTED; | |
3768 | buip = xfs_bui_init(mp); | |
3769 | error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format); | |
3770 | if (error) { | |
3771 | xfs_bui_item_free(buip); | |
3772 | return error; | |
3773 | } | |
3774 | atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents); | |
3775 | ||
57e80956 | 3776 | spin_lock(&log->l_ailp->ail_lock); |
77d61fe4 DW |
3777 | /* |
3778 | * The RUI has two references. One for the RUD and one for RUI to ensure | |
3779 | * it makes it into the AIL. Insert the RUI into the AIL directly and | |
3780 | * drop the RUI reference. Note that xfs_trans_ail_update() drops the | |
3781 | * AIL lock. | |
3782 | */ | |
3783 | xfs_trans_ail_update(log->l_ailp, &buip->bui_item, lsn); | |
3784 | xfs_bui_release(buip); | |
3785 | return 0; | |
3786 | } | |
3787 | ||
3788 | ||
3789 | /* | |
3790 | * This routine is called when an BUD format structure is found in a committed | |
3791 | * transaction in the log. Its purpose is to cancel the corresponding BUI if it | |
3792 | * was still in the log. To do this it searches the AIL for the BUI with an id | |
3793 | * equal to that in the BUD format structure. If we find it we drop the BUD | |
3794 | * reference, which removes the BUI from the AIL and frees it. | |
3795 | */ | |
3796 | STATIC int | |
3797 | xlog_recover_bud_pass2( | |
3798 | struct xlog *log, | |
3799 | struct xlog_recover_item *item) | |
3800 | { | |
3801 | struct xfs_bud_log_format *bud_formatp; | |
3802 | struct xfs_bui_log_item *buip = NULL; | |
3803 | struct xfs_log_item *lip; | |
c8ce540d | 3804 | uint64_t bui_id; |
77d61fe4 DW |
3805 | struct xfs_ail_cursor cur; |
3806 | struct xfs_ail *ailp = log->l_ailp; | |
3807 | ||
3808 | bud_formatp = item->ri_buf[0].i_addr; | |
3809 | if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) | |
3810 | return -EFSCORRUPTED; | |
3811 | bui_id = bud_formatp->bud_bui_id; | |
3812 | ||
3813 | /* | |
3814 | * Search for the BUI with the id in the BUD format structure in the | |
3815 | * AIL. | |
3816 | */ | |
57e80956 | 3817 | spin_lock(&ailp->ail_lock); |
77d61fe4 DW |
3818 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
3819 | while (lip != NULL) { | |
3820 | if (lip->li_type == XFS_LI_BUI) { | |
3821 | buip = (struct xfs_bui_log_item *)lip; | |
3822 | if (buip->bui_format.bui_id == bui_id) { | |
3823 | /* | |
3824 | * Drop the BUD reference to the BUI. This | |
3825 | * removes the BUI from the AIL and frees it. | |
3826 | */ | |
57e80956 | 3827 | spin_unlock(&ailp->ail_lock); |
77d61fe4 | 3828 | xfs_bui_release(buip); |
57e80956 | 3829 | spin_lock(&ailp->ail_lock); |
77d61fe4 DW |
3830 | break; |
3831 | } | |
3832 | } | |
3833 | lip = xfs_trans_ail_cursor_next(ailp, &cur); | |
3834 | } | |
3835 | ||
3836 | xfs_trans_ail_cursor_done(&cur); | |
57e80956 | 3837 | spin_unlock(&ailp->ail_lock); |
77d61fe4 DW |
3838 | |
3839 | return 0; | |
3840 | } | |
3841 | ||
28c8e41a DC |
3842 | /* |
3843 | * This routine is called when an inode create format structure is found in a | |
3844 | * committed transaction in the log. It's purpose is to initialise the inodes | |
3845 | * being allocated on disk. This requires us to get inode cluster buffers that | |
6e7c2b4d | 3846 | * match the range to be initialised, stamped with inode templates and written |
28c8e41a DC |
3847 | * by delayed write so that subsequent modifications will hit the cached buffer |
3848 | * and only need writing out at the end of recovery. | |
3849 | */ | |
3850 | STATIC int | |
3851 | xlog_recover_do_icreate_pass2( | |
3852 | struct xlog *log, | |
3853 | struct list_head *buffer_list, | |
3854 | xlog_recover_item_t *item) | |
3855 | { | |
3856 | struct xfs_mount *mp = log->l_mp; | |
3857 | struct xfs_icreate_log *icl; | |
3858 | xfs_agnumber_t agno; | |
3859 | xfs_agblock_t agbno; | |
3860 | unsigned int count; | |
3861 | unsigned int isize; | |
3862 | xfs_agblock_t length; | |
fc0d1656 BF |
3863 | int blks_per_cluster; |
3864 | int bb_per_cluster; | |
3865 | int cancel_count; | |
3866 | int nbufs; | |
3867 | int i; | |
28c8e41a DC |
3868 | |
3869 | icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr; | |
3870 | if (icl->icl_type != XFS_LI_ICREATE) { | |
3871 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type"); | |
2451337d | 3872 | return -EINVAL; |
28c8e41a DC |
3873 | } |
3874 | ||
3875 | if (icl->icl_size != 1) { | |
3876 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size"); | |
2451337d | 3877 | return -EINVAL; |
28c8e41a DC |
3878 | } |
3879 | ||
3880 | agno = be32_to_cpu(icl->icl_ag); | |
3881 | if (agno >= mp->m_sb.sb_agcount) { | |
3882 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno"); | |
2451337d | 3883 | return -EINVAL; |
28c8e41a DC |
3884 | } |
3885 | agbno = be32_to_cpu(icl->icl_agbno); | |
3886 | if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) { | |
3887 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno"); | |
2451337d | 3888 | return -EINVAL; |
28c8e41a DC |
3889 | } |
3890 | isize = be32_to_cpu(icl->icl_isize); | |
3891 | if (isize != mp->m_sb.sb_inodesize) { | |
3892 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize"); | |
2451337d | 3893 | return -EINVAL; |
28c8e41a DC |
3894 | } |
3895 | count = be32_to_cpu(icl->icl_count); | |
3896 | if (!count) { | |
3897 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count"); | |
2451337d | 3898 | return -EINVAL; |
28c8e41a DC |
3899 | } |
3900 | length = be32_to_cpu(icl->icl_length); | |
3901 | if (!length || length >= mp->m_sb.sb_agblocks) { | |
3902 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length"); | |
2451337d | 3903 | return -EINVAL; |
28c8e41a DC |
3904 | } |
3905 | ||
7f43c907 BF |
3906 | /* |
3907 | * The inode chunk is either full or sparse and we only support | |
3908 | * m_ialloc_min_blks sized sparse allocations at this time. | |
3909 | */ | |
3910 | if (length != mp->m_ialloc_blks && | |
3911 | length != mp->m_ialloc_min_blks) { | |
3912 | xfs_warn(log->l_mp, | |
3913 | "%s: unsupported chunk length", __FUNCTION__); | |
3914 | return -EINVAL; | |
3915 | } | |
3916 | ||
3917 | /* verify inode count is consistent with extent length */ | |
3918 | if ((count >> mp->m_sb.sb_inopblog) != length) { | |
3919 | xfs_warn(log->l_mp, | |
3920 | "%s: inconsistent inode count and chunk length", | |
3921 | __FUNCTION__); | |
2451337d | 3922 | return -EINVAL; |
28c8e41a DC |
3923 | } |
3924 | ||
3925 | /* | |
fc0d1656 BF |
3926 | * The icreate transaction can cover multiple cluster buffers and these |
3927 | * buffers could have been freed and reused. Check the individual | |
3928 | * buffers for cancellation so we don't overwrite anything written after | |
3929 | * a cancellation. | |
3930 | */ | |
3931 | blks_per_cluster = xfs_icluster_size_fsb(mp); | |
3932 | bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
3933 | nbufs = length / blks_per_cluster; | |
3934 | for (i = 0, cancel_count = 0; i < nbufs; i++) { | |
3935 | xfs_daddr_t daddr; | |
3936 | ||
3937 | daddr = XFS_AGB_TO_DADDR(mp, agno, | |
3938 | agbno + i * blks_per_cluster); | |
3939 | if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0)) | |
3940 | cancel_count++; | |
3941 | } | |
3942 | ||
3943 | /* | |
3944 | * We currently only use icreate for a single allocation at a time. This | |
3945 | * means we should expect either all or none of the buffers to be | |
3946 | * cancelled. Be conservative and skip replay if at least one buffer is | |
3947 | * cancelled, but warn the user that something is awry if the buffers | |
3948 | * are not consistent. | |
28c8e41a | 3949 | * |
fc0d1656 BF |
3950 | * XXX: This must be refined to only skip cancelled clusters once we use |
3951 | * icreate for multiple chunk allocations. | |
28c8e41a | 3952 | */ |
fc0d1656 BF |
3953 | ASSERT(!cancel_count || cancel_count == nbufs); |
3954 | if (cancel_count) { | |
3955 | if (cancel_count != nbufs) | |
3956 | xfs_warn(mp, | |
3957 | "WARNING: partial inode chunk cancellation, skipped icreate."); | |
78d57e45 | 3958 | trace_xfs_log_recover_icreate_cancel(log, icl); |
28c8e41a | 3959 | return 0; |
78d57e45 | 3960 | } |
28c8e41a | 3961 | |
78d57e45 | 3962 | trace_xfs_log_recover_icreate_recover(log, icl); |
fc0d1656 BF |
3963 | return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno, |
3964 | length, be32_to_cpu(icl->icl_gen)); | |
28c8e41a DC |
3965 | } |
3966 | ||
00574da1 ZYW |
3967 | STATIC void |
3968 | xlog_recover_buffer_ra_pass2( | |
3969 | struct xlog *log, | |
3970 | struct xlog_recover_item *item) | |
3971 | { | |
3972 | struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr; | |
3973 | struct xfs_mount *mp = log->l_mp; | |
3974 | ||
84a5b730 | 3975 | if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno, |
00574da1 ZYW |
3976 | buf_f->blf_len, buf_f->blf_flags)) { |
3977 | return; | |
3978 | } | |
3979 | ||
3980 | xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno, | |
3981 | buf_f->blf_len, NULL); | |
3982 | } | |
3983 | ||
3984 | STATIC void | |
3985 | xlog_recover_inode_ra_pass2( | |
3986 | struct xlog *log, | |
3987 | struct xlog_recover_item *item) | |
3988 | { | |
3989 | struct xfs_inode_log_format ilf_buf; | |
3990 | struct xfs_inode_log_format *ilfp; | |
3991 | struct xfs_mount *mp = log->l_mp; | |
3992 | int error; | |
3993 | ||
3994 | if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) { | |
3995 | ilfp = item->ri_buf[0].i_addr; | |
3996 | } else { | |
3997 | ilfp = &ilf_buf; | |
3998 | memset(ilfp, 0, sizeof(*ilfp)); | |
3999 | error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp); | |
4000 | if (error) | |
4001 | return; | |
4002 | } | |
4003 | ||
84a5b730 | 4004 | if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0)) |
00574da1 ZYW |
4005 | return; |
4006 | ||
4007 | xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno, | |
d8914002 | 4008 | ilfp->ilf_len, &xfs_inode_buf_ra_ops); |
00574da1 ZYW |
4009 | } |
4010 | ||
4011 | STATIC void | |
4012 | xlog_recover_dquot_ra_pass2( | |
4013 | struct xlog *log, | |
4014 | struct xlog_recover_item *item) | |
4015 | { | |
4016 | struct xfs_mount *mp = log->l_mp; | |
4017 | struct xfs_disk_dquot *recddq; | |
4018 | struct xfs_dq_logformat *dq_f; | |
4019 | uint type; | |
7d6a13f0 | 4020 | int len; |
00574da1 ZYW |
4021 | |
4022 | ||
4023 | if (mp->m_qflags == 0) | |
4024 | return; | |
4025 | ||
4026 | recddq = item->ri_buf[1].i_addr; | |
4027 | if (recddq == NULL) | |
4028 | return; | |
4029 | if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot)) | |
4030 | return; | |
4031 | ||
4032 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); | |
4033 | ASSERT(type); | |
4034 | if (log->l_quotaoffs_flag & type) | |
4035 | return; | |
4036 | ||
4037 | dq_f = item->ri_buf[0].i_addr; | |
4038 | ASSERT(dq_f); | |
4039 | ASSERT(dq_f->qlf_len == 1); | |
4040 | ||
7d6a13f0 DC |
4041 | len = XFS_FSB_TO_BB(mp, dq_f->qlf_len); |
4042 | if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0)) | |
4043 | return; | |
4044 | ||
4045 | xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len, | |
4046 | &xfs_dquot_buf_ra_ops); | |
00574da1 ZYW |
4047 | } |
4048 | ||
4049 | STATIC void | |
4050 | xlog_recover_ra_pass2( | |
4051 | struct xlog *log, | |
4052 | struct xlog_recover_item *item) | |
4053 | { | |
4054 | switch (ITEM_TYPE(item)) { | |
4055 | case XFS_LI_BUF: | |
4056 | xlog_recover_buffer_ra_pass2(log, item); | |
4057 | break; | |
4058 | case XFS_LI_INODE: | |
4059 | xlog_recover_inode_ra_pass2(log, item); | |
4060 | break; | |
4061 | case XFS_LI_DQUOT: | |
4062 | xlog_recover_dquot_ra_pass2(log, item); | |
4063 | break; | |
4064 | case XFS_LI_EFI: | |
4065 | case XFS_LI_EFD: | |
4066 | case XFS_LI_QUOTAOFF: | |
9e88b5d8 DW |
4067 | case XFS_LI_RUI: |
4068 | case XFS_LI_RUD: | |
f997ee21 DW |
4069 | case XFS_LI_CUI: |
4070 | case XFS_LI_CUD: | |
77d61fe4 DW |
4071 | case XFS_LI_BUI: |
4072 | case XFS_LI_BUD: | |
00574da1 ZYW |
4073 | default: |
4074 | break; | |
4075 | } | |
4076 | } | |
4077 | ||
d0450948 | 4078 | STATIC int |
c9f71f5f | 4079 | xlog_recover_commit_pass1( |
ad223e60 MT |
4080 | struct xlog *log, |
4081 | struct xlog_recover *trans, | |
4082 | struct xlog_recover_item *item) | |
d0450948 | 4083 | { |
c9f71f5f | 4084 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1); |
d0450948 CH |
4085 | |
4086 | switch (ITEM_TYPE(item)) { | |
4087 | case XFS_LI_BUF: | |
c9f71f5f CH |
4088 | return xlog_recover_buffer_pass1(log, item); |
4089 | case XFS_LI_QUOTAOFF: | |
4090 | return xlog_recover_quotaoff_pass1(log, item); | |
d0450948 | 4091 | case XFS_LI_INODE: |
d0450948 | 4092 | case XFS_LI_EFI: |
d0450948 | 4093 | case XFS_LI_EFD: |
c9f71f5f | 4094 | case XFS_LI_DQUOT: |
28c8e41a | 4095 | case XFS_LI_ICREATE: |
9e88b5d8 DW |
4096 | case XFS_LI_RUI: |
4097 | case XFS_LI_RUD: | |
f997ee21 DW |
4098 | case XFS_LI_CUI: |
4099 | case XFS_LI_CUD: | |
77d61fe4 DW |
4100 | case XFS_LI_BUI: |
4101 | case XFS_LI_BUD: | |
c9f71f5f | 4102 | /* nothing to do in pass 1 */ |
d0450948 | 4103 | return 0; |
c9f71f5f | 4104 | default: |
a0fa2b67 DC |
4105 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
4106 | __func__, ITEM_TYPE(item)); | |
c9f71f5f | 4107 | ASSERT(0); |
2451337d | 4108 | return -EIO; |
c9f71f5f CH |
4109 | } |
4110 | } | |
4111 | ||
4112 | STATIC int | |
4113 | xlog_recover_commit_pass2( | |
ad223e60 MT |
4114 | struct xlog *log, |
4115 | struct xlog_recover *trans, | |
4116 | struct list_head *buffer_list, | |
4117 | struct xlog_recover_item *item) | |
c9f71f5f CH |
4118 | { |
4119 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2); | |
4120 | ||
4121 | switch (ITEM_TYPE(item)) { | |
4122 | case XFS_LI_BUF: | |
50d5c8d8 DC |
4123 | return xlog_recover_buffer_pass2(log, buffer_list, item, |
4124 | trans->r_lsn); | |
c9f71f5f | 4125 | case XFS_LI_INODE: |
50d5c8d8 DC |
4126 | return xlog_recover_inode_pass2(log, buffer_list, item, |
4127 | trans->r_lsn); | |
c9f71f5f CH |
4128 | case XFS_LI_EFI: |
4129 | return xlog_recover_efi_pass2(log, item, trans->r_lsn); | |
4130 | case XFS_LI_EFD: | |
4131 | return xlog_recover_efd_pass2(log, item); | |
9e88b5d8 DW |
4132 | case XFS_LI_RUI: |
4133 | return xlog_recover_rui_pass2(log, item, trans->r_lsn); | |
4134 | case XFS_LI_RUD: | |
4135 | return xlog_recover_rud_pass2(log, item); | |
f997ee21 DW |
4136 | case XFS_LI_CUI: |
4137 | return xlog_recover_cui_pass2(log, item, trans->r_lsn); | |
4138 | case XFS_LI_CUD: | |
4139 | return xlog_recover_cud_pass2(log, item); | |
77d61fe4 DW |
4140 | case XFS_LI_BUI: |
4141 | return xlog_recover_bui_pass2(log, item, trans->r_lsn); | |
4142 | case XFS_LI_BUD: | |
4143 | return xlog_recover_bud_pass2(log, item); | |
d0450948 | 4144 | case XFS_LI_DQUOT: |
50d5c8d8 DC |
4145 | return xlog_recover_dquot_pass2(log, buffer_list, item, |
4146 | trans->r_lsn); | |
28c8e41a DC |
4147 | case XFS_LI_ICREATE: |
4148 | return xlog_recover_do_icreate_pass2(log, buffer_list, item); | |
d0450948 | 4149 | case XFS_LI_QUOTAOFF: |
c9f71f5f CH |
4150 | /* nothing to do in pass2 */ |
4151 | return 0; | |
d0450948 | 4152 | default: |
a0fa2b67 DC |
4153 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
4154 | __func__, ITEM_TYPE(item)); | |
d0450948 | 4155 | ASSERT(0); |
2451337d | 4156 | return -EIO; |
d0450948 CH |
4157 | } |
4158 | } | |
4159 | ||
00574da1 ZYW |
4160 | STATIC int |
4161 | xlog_recover_items_pass2( | |
4162 | struct xlog *log, | |
4163 | struct xlog_recover *trans, | |
4164 | struct list_head *buffer_list, | |
4165 | struct list_head *item_list) | |
4166 | { | |
4167 | struct xlog_recover_item *item; | |
4168 | int error = 0; | |
4169 | ||
4170 | list_for_each_entry(item, item_list, ri_list) { | |
4171 | error = xlog_recover_commit_pass2(log, trans, | |
4172 | buffer_list, item); | |
4173 | if (error) | |
4174 | return error; | |
4175 | } | |
4176 | ||
4177 | return error; | |
4178 | } | |
4179 | ||
d0450948 CH |
4180 | /* |
4181 | * Perform the transaction. | |
4182 | * | |
4183 | * If the transaction modifies a buffer or inode, do it now. Otherwise, | |
4184 | * EFIs and EFDs get queued up by adding entries into the AIL for them. | |
4185 | */ | |
1da177e4 LT |
4186 | STATIC int |
4187 | xlog_recover_commit_trans( | |
ad223e60 | 4188 | struct xlog *log, |
d0450948 | 4189 | struct xlog_recover *trans, |
12818d24 BF |
4190 | int pass, |
4191 | struct list_head *buffer_list) | |
1da177e4 | 4192 | { |
00574da1 | 4193 | int error = 0; |
00574da1 ZYW |
4194 | int items_queued = 0; |
4195 | struct xlog_recover_item *item; | |
4196 | struct xlog_recover_item *next; | |
00574da1 ZYW |
4197 | LIST_HEAD (ra_list); |
4198 | LIST_HEAD (done_list); | |
4199 | ||
4200 | #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100 | |
1da177e4 | 4201 | |
39775431 | 4202 | hlist_del_init(&trans->r_list); |
d0450948 CH |
4203 | |
4204 | error = xlog_recover_reorder_trans(log, trans, pass); | |
4205 | if (error) | |
1da177e4 | 4206 | return error; |
d0450948 | 4207 | |
00574da1 | 4208 | list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) { |
43ff2122 CH |
4209 | switch (pass) { |
4210 | case XLOG_RECOVER_PASS1: | |
c9f71f5f | 4211 | error = xlog_recover_commit_pass1(log, trans, item); |
43ff2122 CH |
4212 | break; |
4213 | case XLOG_RECOVER_PASS2: | |
00574da1 ZYW |
4214 | xlog_recover_ra_pass2(log, item); |
4215 | list_move_tail(&item->ri_list, &ra_list); | |
4216 | items_queued++; | |
4217 | if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) { | |
4218 | error = xlog_recover_items_pass2(log, trans, | |
12818d24 | 4219 | buffer_list, &ra_list); |
00574da1 ZYW |
4220 | list_splice_tail_init(&ra_list, &done_list); |
4221 | items_queued = 0; | |
4222 | } | |
4223 | ||
43ff2122 CH |
4224 | break; |
4225 | default: | |
4226 | ASSERT(0); | |
4227 | } | |
4228 | ||
d0450948 | 4229 | if (error) |
43ff2122 | 4230 | goto out; |
d0450948 CH |
4231 | } |
4232 | ||
00574da1 ZYW |
4233 | out: |
4234 | if (!list_empty(&ra_list)) { | |
4235 | if (!error) | |
4236 | error = xlog_recover_items_pass2(log, trans, | |
12818d24 | 4237 | buffer_list, &ra_list); |
00574da1 ZYW |
4238 | list_splice_tail_init(&ra_list, &done_list); |
4239 | } | |
4240 | ||
4241 | if (!list_empty(&done_list)) | |
4242 | list_splice_init(&done_list, &trans->r_itemq); | |
4243 | ||
12818d24 | 4244 | return error; |
1da177e4 LT |
4245 | } |
4246 | ||
76560669 DC |
4247 | STATIC void |
4248 | xlog_recover_add_item( | |
4249 | struct list_head *head) | |
4250 | { | |
4251 | xlog_recover_item_t *item; | |
4252 | ||
4253 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); | |
4254 | INIT_LIST_HEAD(&item->ri_list); | |
4255 | list_add_tail(&item->ri_list, head); | |
4256 | } | |
4257 | ||
1da177e4 | 4258 | STATIC int |
76560669 DC |
4259 | xlog_recover_add_to_cont_trans( |
4260 | struct xlog *log, | |
4261 | struct xlog_recover *trans, | |
b2a922cd | 4262 | char *dp, |
76560669 | 4263 | int len) |
1da177e4 | 4264 | { |
76560669 | 4265 | xlog_recover_item_t *item; |
b2a922cd | 4266 | char *ptr, *old_ptr; |
76560669 DC |
4267 | int old_len; |
4268 | ||
89cebc84 BF |
4269 | /* |
4270 | * If the transaction is empty, the header was split across this and the | |
4271 | * previous record. Copy the rest of the header. | |
4272 | */ | |
76560669 | 4273 | if (list_empty(&trans->r_itemq)) { |
848ccfc8 | 4274 | ASSERT(len <= sizeof(struct xfs_trans_header)); |
89cebc84 BF |
4275 | if (len > sizeof(struct xfs_trans_header)) { |
4276 | xfs_warn(log->l_mp, "%s: bad header length", __func__); | |
4277 | return -EIO; | |
4278 | } | |
4279 | ||
76560669 | 4280 | xlog_recover_add_item(&trans->r_itemq); |
b2a922cd | 4281 | ptr = (char *)&trans->r_theader + |
89cebc84 | 4282 | sizeof(struct xfs_trans_header) - len; |
76560669 DC |
4283 | memcpy(ptr, dp, len); |
4284 | return 0; | |
4285 | } | |
89cebc84 | 4286 | |
76560669 DC |
4287 | /* take the tail entry */ |
4288 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
4289 | ||
4290 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; | |
4291 | old_len = item->ri_buf[item->ri_cnt-1].i_len; | |
4292 | ||
664b60f6 | 4293 | ptr = kmem_realloc(old_ptr, len + old_len, KM_SLEEP); |
76560669 DC |
4294 | memcpy(&ptr[old_len], dp, len); |
4295 | item->ri_buf[item->ri_cnt-1].i_len += len; | |
4296 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; | |
4297 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); | |
1da177e4 LT |
4298 | return 0; |
4299 | } | |
4300 | ||
76560669 DC |
4301 | /* |
4302 | * The next region to add is the start of a new region. It could be | |
4303 | * a whole region or it could be the first part of a new region. Because | |
4304 | * of this, the assumption here is that the type and size fields of all | |
4305 | * format structures fit into the first 32 bits of the structure. | |
4306 | * | |
4307 | * This works because all regions must be 32 bit aligned. Therefore, we | |
4308 | * either have both fields or we have neither field. In the case we have | |
4309 | * neither field, the data part of the region is zero length. We only have | |
4310 | * a log_op_header and can throw away the header since a new one will appear | |
4311 | * later. If we have at least 4 bytes, then we can determine how many regions | |
4312 | * will appear in the current log item. | |
4313 | */ | |
4314 | STATIC int | |
4315 | xlog_recover_add_to_trans( | |
4316 | struct xlog *log, | |
4317 | struct xlog_recover *trans, | |
b2a922cd | 4318 | char *dp, |
76560669 DC |
4319 | int len) |
4320 | { | |
06b11321 | 4321 | struct xfs_inode_log_format *in_f; /* any will do */ |
76560669 | 4322 | xlog_recover_item_t *item; |
b2a922cd | 4323 | char *ptr; |
76560669 DC |
4324 | |
4325 | if (!len) | |
4326 | return 0; | |
4327 | if (list_empty(&trans->r_itemq)) { | |
4328 | /* we need to catch log corruptions here */ | |
4329 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { | |
4330 | xfs_warn(log->l_mp, "%s: bad header magic number", | |
4331 | __func__); | |
4332 | ASSERT(0); | |
4333 | return -EIO; | |
4334 | } | |
89cebc84 BF |
4335 | |
4336 | if (len > sizeof(struct xfs_trans_header)) { | |
4337 | xfs_warn(log->l_mp, "%s: bad header length", __func__); | |
4338 | ASSERT(0); | |
4339 | return -EIO; | |
4340 | } | |
4341 | ||
4342 | /* | |
4343 | * The transaction header can be arbitrarily split across op | |
4344 | * records. If we don't have the whole thing here, copy what we | |
4345 | * do have and handle the rest in the next record. | |
4346 | */ | |
4347 | if (len == sizeof(struct xfs_trans_header)) | |
76560669 DC |
4348 | xlog_recover_add_item(&trans->r_itemq); |
4349 | memcpy(&trans->r_theader, dp, len); | |
4350 | return 0; | |
4351 | } | |
4352 | ||
4353 | ptr = kmem_alloc(len, KM_SLEEP); | |
4354 | memcpy(ptr, dp, len); | |
06b11321 | 4355 | in_f = (struct xfs_inode_log_format *)ptr; |
76560669 DC |
4356 | |
4357 | /* take the tail entry */ | |
4358 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
4359 | if (item->ri_total != 0 && | |
4360 | item->ri_total == item->ri_cnt) { | |
4361 | /* tail item is in use, get a new one */ | |
4362 | xlog_recover_add_item(&trans->r_itemq); | |
4363 | item = list_entry(trans->r_itemq.prev, | |
4364 | xlog_recover_item_t, ri_list); | |
4365 | } | |
4366 | ||
4367 | if (item->ri_total == 0) { /* first region to be added */ | |
4368 | if (in_f->ilf_size == 0 || | |
4369 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { | |
4370 | xfs_warn(log->l_mp, | |
4371 | "bad number of regions (%d) in inode log format", | |
4372 | in_f->ilf_size); | |
4373 | ASSERT(0); | |
4374 | kmem_free(ptr); | |
4375 | return -EIO; | |
4376 | } | |
4377 | ||
4378 | item->ri_total = in_f->ilf_size; | |
4379 | item->ri_buf = | |
4380 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), | |
4381 | KM_SLEEP); | |
4382 | } | |
4383 | ASSERT(item->ri_total > item->ri_cnt); | |
4384 | /* Description region is ri_buf[0] */ | |
4385 | item->ri_buf[item->ri_cnt].i_addr = ptr; | |
4386 | item->ri_buf[item->ri_cnt].i_len = len; | |
4387 | item->ri_cnt++; | |
4388 | trace_xfs_log_recover_item_add(log, trans, item, 0); | |
4389 | return 0; | |
4390 | } | |
b818cca1 | 4391 | |
76560669 DC |
4392 | /* |
4393 | * Free up any resources allocated by the transaction | |
4394 | * | |
4395 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. | |
4396 | */ | |
4397 | STATIC void | |
4398 | xlog_recover_free_trans( | |
4399 | struct xlog_recover *trans) | |
4400 | { | |
4401 | xlog_recover_item_t *item, *n; | |
4402 | int i; | |
4403 | ||
39775431 BF |
4404 | hlist_del_init(&trans->r_list); |
4405 | ||
76560669 DC |
4406 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { |
4407 | /* Free the regions in the item. */ | |
4408 | list_del(&item->ri_list); | |
4409 | for (i = 0; i < item->ri_cnt; i++) | |
4410 | kmem_free(item->ri_buf[i].i_addr); | |
4411 | /* Free the item itself */ | |
4412 | kmem_free(item->ri_buf); | |
4413 | kmem_free(item); | |
4414 | } | |
4415 | /* Free the transaction recover structure */ | |
4416 | kmem_free(trans); | |
4417 | } | |
4418 | ||
e9131e50 DC |
4419 | /* |
4420 | * On error or completion, trans is freed. | |
4421 | */ | |
1da177e4 | 4422 | STATIC int |
eeb11688 DC |
4423 | xlog_recovery_process_trans( |
4424 | struct xlog *log, | |
4425 | struct xlog_recover *trans, | |
b2a922cd | 4426 | char *dp, |
eeb11688 DC |
4427 | unsigned int len, |
4428 | unsigned int flags, | |
12818d24 BF |
4429 | int pass, |
4430 | struct list_head *buffer_list) | |
1da177e4 | 4431 | { |
e9131e50 DC |
4432 | int error = 0; |
4433 | bool freeit = false; | |
eeb11688 DC |
4434 | |
4435 | /* mask off ophdr transaction container flags */ | |
4436 | flags &= ~XLOG_END_TRANS; | |
4437 | if (flags & XLOG_WAS_CONT_TRANS) | |
4438 | flags &= ~XLOG_CONTINUE_TRANS; | |
4439 | ||
88b863db DC |
4440 | /* |
4441 | * Callees must not free the trans structure. We'll decide if we need to | |
4442 | * free it or not based on the operation being done and it's result. | |
4443 | */ | |
eeb11688 DC |
4444 | switch (flags) { |
4445 | /* expected flag values */ | |
4446 | case 0: | |
4447 | case XLOG_CONTINUE_TRANS: | |
4448 | error = xlog_recover_add_to_trans(log, trans, dp, len); | |
4449 | break; | |
4450 | case XLOG_WAS_CONT_TRANS: | |
4451 | error = xlog_recover_add_to_cont_trans(log, trans, dp, len); | |
4452 | break; | |
4453 | case XLOG_COMMIT_TRANS: | |
12818d24 BF |
4454 | error = xlog_recover_commit_trans(log, trans, pass, |
4455 | buffer_list); | |
88b863db DC |
4456 | /* success or fail, we are now done with this transaction. */ |
4457 | freeit = true; | |
eeb11688 DC |
4458 | break; |
4459 | ||
4460 | /* unexpected flag values */ | |
4461 | case XLOG_UNMOUNT_TRANS: | |
e9131e50 | 4462 | /* just skip trans */ |
eeb11688 | 4463 | xfs_warn(log->l_mp, "%s: Unmount LR", __func__); |
e9131e50 | 4464 | freeit = true; |
eeb11688 DC |
4465 | break; |
4466 | case XLOG_START_TRANS: | |
eeb11688 DC |
4467 | default: |
4468 | xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags); | |
4469 | ASSERT(0); | |
e9131e50 | 4470 | error = -EIO; |
eeb11688 DC |
4471 | break; |
4472 | } | |
e9131e50 DC |
4473 | if (error || freeit) |
4474 | xlog_recover_free_trans(trans); | |
eeb11688 DC |
4475 | return error; |
4476 | } | |
4477 | ||
b818cca1 DC |
4478 | /* |
4479 | * Lookup the transaction recovery structure associated with the ID in the | |
4480 | * current ophdr. If the transaction doesn't exist and the start flag is set in | |
4481 | * the ophdr, then allocate a new transaction for future ID matches to find. | |
4482 | * Either way, return what we found during the lookup - an existing transaction | |
4483 | * or nothing. | |
4484 | */ | |
eeb11688 DC |
4485 | STATIC struct xlog_recover * |
4486 | xlog_recover_ophdr_to_trans( | |
4487 | struct hlist_head rhash[], | |
4488 | struct xlog_rec_header *rhead, | |
4489 | struct xlog_op_header *ohead) | |
4490 | { | |
4491 | struct xlog_recover *trans; | |
4492 | xlog_tid_t tid; | |
4493 | struct hlist_head *rhp; | |
4494 | ||
4495 | tid = be32_to_cpu(ohead->oh_tid); | |
4496 | rhp = &rhash[XLOG_RHASH(tid)]; | |
b818cca1 DC |
4497 | hlist_for_each_entry(trans, rhp, r_list) { |
4498 | if (trans->r_log_tid == tid) | |
4499 | return trans; | |
4500 | } | |
eeb11688 DC |
4501 | |
4502 | /* | |
b818cca1 DC |
4503 | * skip over non-start transaction headers - we could be |
4504 | * processing slack space before the next transaction starts | |
4505 | */ | |
4506 | if (!(ohead->oh_flags & XLOG_START_TRANS)) | |
4507 | return NULL; | |
4508 | ||
4509 | ASSERT(be32_to_cpu(ohead->oh_len) == 0); | |
4510 | ||
4511 | /* | |
4512 | * This is a new transaction so allocate a new recovery container to | |
4513 | * hold the recovery ops that will follow. | |
4514 | */ | |
4515 | trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP); | |
4516 | trans->r_log_tid = tid; | |
4517 | trans->r_lsn = be64_to_cpu(rhead->h_lsn); | |
4518 | INIT_LIST_HEAD(&trans->r_itemq); | |
4519 | INIT_HLIST_NODE(&trans->r_list); | |
4520 | hlist_add_head(&trans->r_list, rhp); | |
4521 | ||
4522 | /* | |
4523 | * Nothing more to do for this ophdr. Items to be added to this new | |
4524 | * transaction will be in subsequent ophdr containers. | |
eeb11688 | 4525 | */ |
eeb11688 DC |
4526 | return NULL; |
4527 | } | |
4528 | ||
4529 | STATIC int | |
4530 | xlog_recover_process_ophdr( | |
4531 | struct xlog *log, | |
4532 | struct hlist_head rhash[], | |
4533 | struct xlog_rec_header *rhead, | |
4534 | struct xlog_op_header *ohead, | |
b2a922cd CH |
4535 | char *dp, |
4536 | char *end, | |
12818d24 BF |
4537 | int pass, |
4538 | struct list_head *buffer_list) | |
eeb11688 DC |
4539 | { |
4540 | struct xlog_recover *trans; | |
eeb11688 | 4541 | unsigned int len; |
12818d24 | 4542 | int error; |
eeb11688 DC |
4543 | |
4544 | /* Do we understand who wrote this op? */ | |
4545 | if (ohead->oh_clientid != XFS_TRANSACTION && | |
4546 | ohead->oh_clientid != XFS_LOG) { | |
4547 | xfs_warn(log->l_mp, "%s: bad clientid 0x%x", | |
4548 | __func__, ohead->oh_clientid); | |
4549 | ASSERT(0); | |
4550 | return -EIO; | |
4551 | } | |
4552 | ||
4553 | /* | |
4554 | * Check the ophdr contains all the data it is supposed to contain. | |
4555 | */ | |
4556 | len = be32_to_cpu(ohead->oh_len); | |
4557 | if (dp + len > end) { | |
4558 | xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len); | |
4559 | WARN_ON(1); | |
4560 | return -EIO; | |
4561 | } | |
4562 | ||
4563 | trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead); | |
4564 | if (!trans) { | |
4565 | /* nothing to do, so skip over this ophdr */ | |
4566 | return 0; | |
4567 | } | |
4568 | ||
12818d24 BF |
4569 | /* |
4570 | * The recovered buffer queue is drained only once we know that all | |
4571 | * recovery items for the current LSN have been processed. This is | |
4572 | * required because: | |
4573 | * | |
4574 | * - Buffer write submission updates the metadata LSN of the buffer. | |
4575 | * - Log recovery skips items with a metadata LSN >= the current LSN of | |
4576 | * the recovery item. | |
4577 | * - Separate recovery items against the same metadata buffer can share | |
4578 | * a current LSN. I.e., consider that the LSN of a recovery item is | |
4579 | * defined as the starting LSN of the first record in which its | |
4580 | * transaction appears, that a record can hold multiple transactions, | |
4581 | * and/or that a transaction can span multiple records. | |
4582 | * | |
4583 | * In other words, we are allowed to submit a buffer from log recovery | |
4584 | * once per current LSN. Otherwise, we may incorrectly skip recovery | |
4585 | * items and cause corruption. | |
4586 | * | |
4587 | * We don't know up front whether buffers are updated multiple times per | |
4588 | * LSN. Therefore, track the current LSN of each commit log record as it | |
4589 | * is processed and drain the queue when it changes. Use commit records | |
4590 | * because they are ordered correctly by the logging code. | |
4591 | */ | |
4592 | if (log->l_recovery_lsn != trans->r_lsn && | |
4593 | ohead->oh_flags & XLOG_COMMIT_TRANS) { | |
4594 | error = xfs_buf_delwri_submit(buffer_list); | |
4595 | if (error) | |
4596 | return error; | |
4597 | log->l_recovery_lsn = trans->r_lsn; | |
4598 | } | |
4599 | ||
e9131e50 | 4600 | return xlog_recovery_process_trans(log, trans, dp, len, |
12818d24 | 4601 | ohead->oh_flags, pass, buffer_list); |
1da177e4 LT |
4602 | } |
4603 | ||
4604 | /* | |
4605 | * There are two valid states of the r_state field. 0 indicates that the | |
4606 | * transaction structure is in a normal state. We have either seen the | |
4607 | * start of the transaction or the last operation we added was not a partial | |
4608 | * operation. If the last operation we added to the transaction was a | |
4609 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. | |
4610 | * | |
4611 | * NOTE: skip LRs with 0 data length. | |
4612 | */ | |
4613 | STATIC int | |
4614 | xlog_recover_process_data( | |
9a8d2fdb | 4615 | struct xlog *log, |
f0a76953 | 4616 | struct hlist_head rhash[], |
9a8d2fdb | 4617 | struct xlog_rec_header *rhead, |
b2a922cd | 4618 | char *dp, |
12818d24 BF |
4619 | int pass, |
4620 | struct list_head *buffer_list) | |
1da177e4 | 4621 | { |
eeb11688 | 4622 | struct xlog_op_header *ohead; |
b2a922cd | 4623 | char *end; |
1da177e4 | 4624 | int num_logops; |
1da177e4 | 4625 | int error; |
1da177e4 | 4626 | |
eeb11688 | 4627 | end = dp + be32_to_cpu(rhead->h_len); |
b53e675d | 4628 | num_logops = be32_to_cpu(rhead->h_num_logops); |
1da177e4 LT |
4629 | |
4630 | /* check the log format matches our own - else we can't recover */ | |
4631 | if (xlog_header_check_recover(log->l_mp, rhead)) | |
2451337d | 4632 | return -EIO; |
1da177e4 | 4633 | |
5cd9cee9 | 4634 | trace_xfs_log_recover_record(log, rhead, pass); |
eeb11688 DC |
4635 | while ((dp < end) && num_logops) { |
4636 | ||
4637 | ohead = (struct xlog_op_header *)dp; | |
4638 | dp += sizeof(*ohead); | |
4639 | ASSERT(dp <= end); | |
4640 | ||
4641 | /* errors will abort recovery */ | |
4642 | error = xlog_recover_process_ophdr(log, rhash, rhead, ohead, | |
12818d24 | 4643 | dp, end, pass, buffer_list); |
eeb11688 DC |
4644 | if (error) |
4645 | return error; | |
4646 | ||
67fcb7bf | 4647 | dp += be32_to_cpu(ohead->oh_len); |
1da177e4 LT |
4648 | num_logops--; |
4649 | } | |
4650 | return 0; | |
4651 | } | |
4652 | ||
dc42375d | 4653 | /* Recover the EFI if necessary. */ |
3c1e2bbe | 4654 | STATIC int |
1da177e4 | 4655 | xlog_recover_process_efi( |
dc42375d DW |
4656 | struct xfs_mount *mp, |
4657 | struct xfs_ail *ailp, | |
4658 | struct xfs_log_item *lip) | |
1da177e4 | 4659 | { |
dc42375d DW |
4660 | struct xfs_efi_log_item *efip; |
4661 | int error; | |
1da177e4 LT |
4662 | |
4663 | /* | |
dc42375d | 4664 | * Skip EFIs that we've already processed. |
1da177e4 | 4665 | */ |
dc42375d DW |
4666 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); |
4667 | if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) | |
4668 | return 0; | |
1da177e4 | 4669 | |
57e80956 | 4670 | spin_unlock(&ailp->ail_lock); |
dc42375d | 4671 | error = xfs_efi_recover(mp, efip); |
57e80956 | 4672 | spin_lock(&ailp->ail_lock); |
1da177e4 | 4673 | |
dc42375d DW |
4674 | return error; |
4675 | } | |
6bc43af3 | 4676 | |
dc42375d DW |
4677 | /* Release the EFI since we're cancelling everything. */ |
4678 | STATIC void | |
4679 | xlog_recover_cancel_efi( | |
4680 | struct xfs_mount *mp, | |
4681 | struct xfs_ail *ailp, | |
4682 | struct xfs_log_item *lip) | |
4683 | { | |
4684 | struct xfs_efi_log_item *efip; | |
1da177e4 | 4685 | |
dc42375d | 4686 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); |
fc6149d8 | 4687 | |
57e80956 | 4688 | spin_unlock(&ailp->ail_lock); |
dc42375d | 4689 | xfs_efi_release(efip); |
57e80956 | 4690 | spin_lock(&ailp->ail_lock); |
dc42375d DW |
4691 | } |
4692 | ||
9e88b5d8 DW |
4693 | /* Recover the RUI if necessary. */ |
4694 | STATIC int | |
4695 | xlog_recover_process_rui( | |
4696 | struct xfs_mount *mp, | |
4697 | struct xfs_ail *ailp, | |
4698 | struct xfs_log_item *lip) | |
4699 | { | |
4700 | struct xfs_rui_log_item *ruip; | |
4701 | int error; | |
4702 | ||
4703 | /* | |
4704 | * Skip RUIs that we've already processed. | |
4705 | */ | |
4706 | ruip = container_of(lip, struct xfs_rui_log_item, rui_item); | |
4707 | if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags)) | |
4708 | return 0; | |
4709 | ||
57e80956 | 4710 | spin_unlock(&ailp->ail_lock); |
9e88b5d8 | 4711 | error = xfs_rui_recover(mp, ruip); |
57e80956 | 4712 | spin_lock(&ailp->ail_lock); |
9e88b5d8 DW |
4713 | |
4714 | return error; | |
4715 | } | |
4716 | ||
4717 | /* Release the RUI since we're cancelling everything. */ | |
4718 | STATIC void | |
4719 | xlog_recover_cancel_rui( | |
4720 | struct xfs_mount *mp, | |
4721 | struct xfs_ail *ailp, | |
4722 | struct xfs_log_item *lip) | |
4723 | { | |
4724 | struct xfs_rui_log_item *ruip; | |
4725 | ||
4726 | ruip = container_of(lip, struct xfs_rui_log_item, rui_item); | |
4727 | ||
57e80956 | 4728 | spin_unlock(&ailp->ail_lock); |
9e88b5d8 | 4729 | xfs_rui_release(ruip); |
57e80956 | 4730 | spin_lock(&ailp->ail_lock); |
9e88b5d8 DW |
4731 | } |
4732 | ||
f997ee21 DW |
4733 | /* Recover the CUI if necessary. */ |
4734 | STATIC int | |
4735 | xlog_recover_process_cui( | |
4736 | struct xfs_mount *mp, | |
4737 | struct xfs_ail *ailp, | |
50995582 DW |
4738 | struct xfs_log_item *lip, |
4739 | struct xfs_defer_ops *dfops) | |
f997ee21 DW |
4740 | { |
4741 | struct xfs_cui_log_item *cuip; | |
4742 | int error; | |
4743 | ||
4744 | /* | |
4745 | * Skip CUIs that we've already processed. | |
4746 | */ | |
4747 | cuip = container_of(lip, struct xfs_cui_log_item, cui_item); | |
4748 | if (test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags)) | |
4749 | return 0; | |
4750 | ||
57e80956 | 4751 | spin_unlock(&ailp->ail_lock); |
50995582 | 4752 | error = xfs_cui_recover(mp, cuip, dfops); |
57e80956 | 4753 | spin_lock(&ailp->ail_lock); |
f997ee21 DW |
4754 | |
4755 | return error; | |
4756 | } | |
4757 | ||
4758 | /* Release the CUI since we're cancelling everything. */ | |
4759 | STATIC void | |
4760 | xlog_recover_cancel_cui( | |
4761 | struct xfs_mount *mp, | |
4762 | struct xfs_ail *ailp, | |
4763 | struct xfs_log_item *lip) | |
4764 | { | |
4765 | struct xfs_cui_log_item *cuip; | |
4766 | ||
4767 | cuip = container_of(lip, struct xfs_cui_log_item, cui_item); | |
4768 | ||
57e80956 | 4769 | spin_unlock(&ailp->ail_lock); |
f997ee21 | 4770 | xfs_cui_release(cuip); |
57e80956 | 4771 | spin_lock(&ailp->ail_lock); |
f997ee21 DW |
4772 | } |
4773 | ||
77d61fe4 DW |
4774 | /* Recover the BUI if necessary. */ |
4775 | STATIC int | |
4776 | xlog_recover_process_bui( | |
4777 | struct xfs_mount *mp, | |
4778 | struct xfs_ail *ailp, | |
50995582 DW |
4779 | struct xfs_log_item *lip, |
4780 | struct xfs_defer_ops *dfops) | |
77d61fe4 DW |
4781 | { |
4782 | struct xfs_bui_log_item *buip; | |
4783 | int error; | |
4784 | ||
4785 | /* | |
4786 | * Skip BUIs that we've already processed. | |
4787 | */ | |
4788 | buip = container_of(lip, struct xfs_bui_log_item, bui_item); | |
4789 | if (test_bit(XFS_BUI_RECOVERED, &buip->bui_flags)) | |
4790 | return 0; | |
4791 | ||
57e80956 | 4792 | spin_unlock(&ailp->ail_lock); |
50995582 | 4793 | error = xfs_bui_recover(mp, buip, dfops); |
57e80956 | 4794 | spin_lock(&ailp->ail_lock); |
77d61fe4 DW |
4795 | |
4796 | return error; | |
4797 | } | |
4798 | ||
4799 | /* Release the BUI since we're cancelling everything. */ | |
4800 | STATIC void | |
4801 | xlog_recover_cancel_bui( | |
4802 | struct xfs_mount *mp, | |
4803 | struct xfs_ail *ailp, | |
4804 | struct xfs_log_item *lip) | |
4805 | { | |
4806 | struct xfs_bui_log_item *buip; | |
4807 | ||
4808 | buip = container_of(lip, struct xfs_bui_log_item, bui_item); | |
4809 | ||
57e80956 | 4810 | spin_unlock(&ailp->ail_lock); |
77d61fe4 | 4811 | xfs_bui_release(buip); |
57e80956 | 4812 | spin_lock(&ailp->ail_lock); |
77d61fe4 DW |
4813 | } |
4814 | ||
dc42375d DW |
4815 | /* Is this log item a deferred action intent? */ |
4816 | static inline bool xlog_item_is_intent(struct xfs_log_item *lip) | |
4817 | { | |
4818 | switch (lip->li_type) { | |
4819 | case XFS_LI_EFI: | |
9e88b5d8 | 4820 | case XFS_LI_RUI: |
f997ee21 | 4821 | case XFS_LI_CUI: |
77d61fe4 | 4822 | case XFS_LI_BUI: |
dc42375d DW |
4823 | return true; |
4824 | default: | |
4825 | return false; | |
4826 | } | |
1da177e4 LT |
4827 | } |
4828 | ||
50995582 DW |
4829 | /* Take all the collected deferred ops and finish them in order. */ |
4830 | static int | |
4831 | xlog_finish_defer_ops( | |
4832 | struct xfs_mount *mp, | |
4833 | struct xfs_defer_ops *dfops) | |
4834 | { | |
4835 | struct xfs_trans *tp; | |
4836 | int64_t freeblks; | |
4837 | uint resblks; | |
4838 | int error; | |
4839 | ||
4840 | /* | |
4841 | * We're finishing the defer_ops that accumulated as a result of | |
4842 | * recovering unfinished intent items during log recovery. We | |
4843 | * reserve an itruncate transaction because it is the largest | |
4844 | * permanent transaction type. Since we're the only user of the fs | |
4845 | * right now, take 93% (15/16) of the available free blocks. Use | |
4846 | * weird math to avoid a 64-bit division. | |
4847 | */ | |
4848 | freeblks = percpu_counter_sum(&mp->m_fdblocks); | |
4849 | if (freeblks <= 0) | |
4850 | return -ENOSPC; | |
4851 | resblks = min_t(int64_t, UINT_MAX, freeblks); | |
4852 | resblks = (resblks * 15) >> 4; | |
4853 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks, | |
4854 | 0, XFS_TRANS_RESERVE, &tp); | |
4855 | if (error) | |
4856 | return error; | |
4857 | ||
4858 | error = xfs_defer_finish(&tp, dfops); | |
4859 | if (error) | |
4860 | goto out_cancel; | |
4861 | ||
4862 | return xfs_trans_commit(tp); | |
4863 | ||
4864 | out_cancel: | |
4865 | xfs_trans_cancel(tp); | |
4866 | return error; | |
4867 | } | |
4868 | ||
1da177e4 | 4869 | /* |
dc42375d DW |
4870 | * When this is called, all of the log intent items which did not have |
4871 | * corresponding log done items should be in the AIL. What we do now | |
4872 | * is update the data structures associated with each one. | |
1da177e4 | 4873 | * |
dc42375d DW |
4874 | * Since we process the log intent items in normal transactions, they |
4875 | * will be removed at some point after the commit. This prevents us | |
4876 | * from just walking down the list processing each one. We'll use a | |
4877 | * flag in the intent item to skip those that we've already processed | |
4878 | * and use the AIL iteration mechanism's generation count to try to | |
4879 | * speed this up at least a bit. | |
1da177e4 | 4880 | * |
dc42375d DW |
4881 | * When we start, we know that the intents are the only things in the |
4882 | * AIL. As we process them, however, other items are added to the | |
4883 | * AIL. | |
1da177e4 | 4884 | */ |
3c1e2bbe | 4885 | STATIC int |
dc42375d | 4886 | xlog_recover_process_intents( |
f0b2efad | 4887 | struct xlog *log) |
1da177e4 | 4888 | { |
50995582 | 4889 | struct xfs_defer_ops dfops; |
27d8d5fe | 4890 | struct xfs_ail_cursor cur; |
50995582 | 4891 | struct xfs_log_item *lip; |
a9c21c1b | 4892 | struct xfs_ail *ailp; |
50995582 DW |
4893 | xfs_fsblock_t firstfsb; |
4894 | int error = 0; | |
7bf7a193 | 4895 | #if defined(DEBUG) || defined(XFS_WARN) |
dc42375d | 4896 | xfs_lsn_t last_lsn; |
7bf7a193 | 4897 | #endif |
1da177e4 | 4898 | |
a9c21c1b | 4899 | ailp = log->l_ailp; |
57e80956 | 4900 | spin_lock(&ailp->ail_lock); |
a9c21c1b | 4901 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
7bf7a193 | 4902 | #if defined(DEBUG) || defined(XFS_WARN) |
dc42375d | 4903 | last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block); |
7bf7a193 | 4904 | #endif |
50995582 | 4905 | xfs_defer_init(&dfops, &firstfsb); |
1da177e4 LT |
4906 | while (lip != NULL) { |
4907 | /* | |
dc42375d DW |
4908 | * We're done when we see something other than an intent. |
4909 | * There should be no intents left in the AIL now. | |
1da177e4 | 4910 | */ |
dc42375d | 4911 | if (!xlog_item_is_intent(lip)) { |
27d8d5fe | 4912 | #ifdef DEBUG |
a9c21c1b | 4913 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
dc42375d | 4914 | ASSERT(!xlog_item_is_intent(lip)); |
27d8d5fe | 4915 | #endif |
1da177e4 LT |
4916 | break; |
4917 | } | |
4918 | ||
4919 | /* | |
dc42375d DW |
4920 | * We should never see a redo item with a LSN higher than |
4921 | * the last transaction we found in the log at the start | |
4922 | * of recovery. | |
1da177e4 | 4923 | */ |
dc42375d | 4924 | ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0); |
1da177e4 | 4925 | |
50995582 DW |
4926 | /* |
4927 | * NOTE: If your intent processing routine can create more | |
4928 | * deferred ops, you /must/ attach them to the dfops in this | |
4929 | * routine or else those subsequent intents will get | |
4930 | * replayed in the wrong order! | |
4931 | */ | |
dc42375d DW |
4932 | switch (lip->li_type) { |
4933 | case XFS_LI_EFI: | |
4934 | error = xlog_recover_process_efi(log->l_mp, ailp, lip); | |
4935 | break; | |
9e88b5d8 DW |
4936 | case XFS_LI_RUI: |
4937 | error = xlog_recover_process_rui(log->l_mp, ailp, lip); | |
4938 | break; | |
f997ee21 | 4939 | case XFS_LI_CUI: |
50995582 DW |
4940 | error = xlog_recover_process_cui(log->l_mp, ailp, lip, |
4941 | &dfops); | |
f997ee21 | 4942 | break; |
77d61fe4 | 4943 | case XFS_LI_BUI: |
50995582 DW |
4944 | error = xlog_recover_process_bui(log->l_mp, ailp, lip, |
4945 | &dfops); | |
77d61fe4 | 4946 | break; |
dc42375d | 4947 | } |
27d8d5fe DC |
4948 | if (error) |
4949 | goto out; | |
a9c21c1b | 4950 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
1da177e4 | 4951 | } |
27d8d5fe | 4952 | out: |
e4a1e29c | 4953 | xfs_trans_ail_cursor_done(&cur); |
57e80956 | 4954 | spin_unlock(&ailp->ail_lock); |
50995582 DW |
4955 | if (error) |
4956 | xfs_defer_cancel(&dfops); | |
4957 | else | |
4958 | error = xlog_finish_defer_ops(log->l_mp, &dfops); | |
4959 | ||
3c1e2bbe | 4960 | return error; |
1da177e4 LT |
4961 | } |
4962 | ||
f0b2efad | 4963 | /* |
dc42375d DW |
4964 | * A cancel occurs when the mount has failed and we're bailing out. |
4965 | * Release all pending log intent items so they don't pin the AIL. | |
f0b2efad BF |
4966 | */ |
4967 | STATIC int | |
dc42375d | 4968 | xlog_recover_cancel_intents( |
f0b2efad BF |
4969 | struct xlog *log) |
4970 | { | |
4971 | struct xfs_log_item *lip; | |
f0b2efad BF |
4972 | int error = 0; |
4973 | struct xfs_ail_cursor cur; | |
4974 | struct xfs_ail *ailp; | |
4975 | ||
4976 | ailp = log->l_ailp; | |
57e80956 | 4977 | spin_lock(&ailp->ail_lock); |
f0b2efad BF |
4978 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
4979 | while (lip != NULL) { | |
4980 | /* | |
dc42375d DW |
4981 | * We're done when we see something other than an intent. |
4982 | * There should be no intents left in the AIL now. | |
f0b2efad | 4983 | */ |
dc42375d | 4984 | if (!xlog_item_is_intent(lip)) { |
f0b2efad BF |
4985 | #ifdef DEBUG |
4986 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) | |
dc42375d | 4987 | ASSERT(!xlog_item_is_intent(lip)); |
f0b2efad BF |
4988 | #endif |
4989 | break; | |
4990 | } | |
4991 | ||
dc42375d DW |
4992 | switch (lip->li_type) { |
4993 | case XFS_LI_EFI: | |
4994 | xlog_recover_cancel_efi(log->l_mp, ailp, lip); | |
4995 | break; | |
9e88b5d8 DW |
4996 | case XFS_LI_RUI: |
4997 | xlog_recover_cancel_rui(log->l_mp, ailp, lip); | |
4998 | break; | |
f997ee21 DW |
4999 | case XFS_LI_CUI: |
5000 | xlog_recover_cancel_cui(log->l_mp, ailp, lip); | |
5001 | break; | |
77d61fe4 DW |
5002 | case XFS_LI_BUI: |
5003 | xlog_recover_cancel_bui(log->l_mp, ailp, lip); | |
5004 | break; | |
dc42375d | 5005 | } |
f0b2efad BF |
5006 | |
5007 | lip = xfs_trans_ail_cursor_next(ailp, &cur); | |
5008 | } | |
5009 | ||
5010 | xfs_trans_ail_cursor_done(&cur); | |
57e80956 | 5011 | spin_unlock(&ailp->ail_lock); |
f0b2efad BF |
5012 | return error; |
5013 | } | |
5014 | ||
1da177e4 LT |
5015 | /* |
5016 | * This routine performs a transaction to null out a bad inode pointer | |
5017 | * in an agi unlinked inode hash bucket. | |
5018 | */ | |
5019 | STATIC void | |
5020 | xlog_recover_clear_agi_bucket( | |
5021 | xfs_mount_t *mp, | |
5022 | xfs_agnumber_t agno, | |
5023 | int bucket) | |
5024 | { | |
5025 | xfs_trans_t *tp; | |
5026 | xfs_agi_t *agi; | |
5027 | xfs_buf_t *agibp; | |
5028 | int offset; | |
5029 | int error; | |
5030 | ||
253f4911 | 5031 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp); |
e5720eec | 5032 | if (error) |
253f4911 | 5033 | goto out_error; |
1da177e4 | 5034 | |
5e1be0fb CH |
5035 | error = xfs_read_agi(mp, tp, agno, &agibp); |
5036 | if (error) | |
e5720eec | 5037 | goto out_abort; |
1da177e4 | 5038 | |
5e1be0fb | 5039 | agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 5040 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
5041 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
5042 | (sizeof(xfs_agino_t) * bucket); | |
5043 | xfs_trans_log_buf(tp, agibp, offset, | |
5044 | (offset + sizeof(xfs_agino_t) - 1)); | |
5045 | ||
70393313 | 5046 | error = xfs_trans_commit(tp); |
e5720eec DC |
5047 | if (error) |
5048 | goto out_error; | |
5049 | return; | |
5050 | ||
5051 | out_abort: | |
4906e215 | 5052 | xfs_trans_cancel(tp); |
e5720eec | 5053 | out_error: |
a0fa2b67 | 5054 | xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno); |
e5720eec | 5055 | return; |
1da177e4 LT |
5056 | } |
5057 | ||
23fac50f CH |
5058 | STATIC xfs_agino_t |
5059 | xlog_recover_process_one_iunlink( | |
5060 | struct xfs_mount *mp, | |
5061 | xfs_agnumber_t agno, | |
5062 | xfs_agino_t agino, | |
5063 | int bucket) | |
5064 | { | |
5065 | struct xfs_buf *ibp; | |
5066 | struct xfs_dinode *dip; | |
5067 | struct xfs_inode *ip; | |
5068 | xfs_ino_t ino; | |
5069 | int error; | |
5070 | ||
5071 | ino = XFS_AGINO_TO_INO(mp, agno, agino); | |
7b6259e7 | 5072 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip); |
23fac50f CH |
5073 | if (error) |
5074 | goto fail; | |
5075 | ||
5076 | /* | |
5077 | * Get the on disk inode to find the next inode in the bucket. | |
5078 | */ | |
475ee413 | 5079 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0); |
23fac50f | 5080 | if (error) |
0e446673 | 5081 | goto fail_iput; |
23fac50f | 5082 | |
17c12bcd | 5083 | xfs_iflags_clear(ip, XFS_IRECOVERY); |
54d7b5c1 | 5084 | ASSERT(VFS_I(ip)->i_nlink == 0); |
c19b3b05 | 5085 | ASSERT(VFS_I(ip)->i_mode != 0); |
23fac50f CH |
5086 | |
5087 | /* setup for the next pass */ | |
5088 | agino = be32_to_cpu(dip->di_next_unlinked); | |
5089 | xfs_buf_relse(ibp); | |
5090 | ||
5091 | /* | |
5092 | * Prevent any DMAPI event from being sent when the reference on | |
5093 | * the inode is dropped. | |
5094 | */ | |
5095 | ip->i_d.di_dmevmask = 0; | |
5096 | ||
0e446673 | 5097 | IRELE(ip); |
23fac50f CH |
5098 | return agino; |
5099 | ||
0e446673 CH |
5100 | fail_iput: |
5101 | IRELE(ip); | |
23fac50f CH |
5102 | fail: |
5103 | /* | |
5104 | * We can't read in the inode this bucket points to, or this inode | |
5105 | * is messed up. Just ditch this bucket of inodes. We will lose | |
5106 | * some inodes and space, but at least we won't hang. | |
5107 | * | |
5108 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to | |
5109 | * clear the inode pointer in the bucket. | |
5110 | */ | |
5111 | xlog_recover_clear_agi_bucket(mp, agno, bucket); | |
5112 | return NULLAGINO; | |
5113 | } | |
5114 | ||
1da177e4 LT |
5115 | /* |
5116 | * xlog_iunlink_recover | |
5117 | * | |
5118 | * This is called during recovery to process any inodes which | |
5119 | * we unlinked but not freed when the system crashed. These | |
5120 | * inodes will be on the lists in the AGI blocks. What we do | |
5121 | * here is scan all the AGIs and fully truncate and free any | |
5122 | * inodes found on the lists. Each inode is removed from the | |
5123 | * lists when it has been fully truncated and is freed. The | |
5124 | * freeing of the inode and its removal from the list must be | |
5125 | * atomic. | |
5126 | */ | |
d96f8f89 | 5127 | STATIC void |
1da177e4 | 5128 | xlog_recover_process_iunlinks( |
9a8d2fdb | 5129 | struct xlog *log) |
1da177e4 LT |
5130 | { |
5131 | xfs_mount_t *mp; | |
5132 | xfs_agnumber_t agno; | |
5133 | xfs_agi_t *agi; | |
5134 | xfs_buf_t *agibp; | |
1da177e4 | 5135 | xfs_agino_t agino; |
1da177e4 LT |
5136 | int bucket; |
5137 | int error; | |
1da177e4 LT |
5138 | |
5139 | mp = log->l_mp; | |
5140 | ||
1da177e4 LT |
5141 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { |
5142 | /* | |
5143 | * Find the agi for this ag. | |
5144 | */ | |
5e1be0fb CH |
5145 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
5146 | if (error) { | |
5147 | /* | |
5148 | * AGI is b0rked. Don't process it. | |
5149 | * | |
5150 | * We should probably mark the filesystem as corrupt | |
5151 | * after we've recovered all the ag's we can.... | |
5152 | */ | |
5153 | continue; | |
1da177e4 | 5154 | } |
d97d32ed JK |
5155 | /* |
5156 | * Unlock the buffer so that it can be acquired in the normal | |
5157 | * course of the transaction to truncate and free each inode. | |
5158 | * Because we are not racing with anyone else here for the AGI | |
5159 | * buffer, we don't even need to hold it locked to read the | |
5160 | * initial unlinked bucket entries out of the buffer. We keep | |
5161 | * buffer reference though, so that it stays pinned in memory | |
5162 | * while we need the buffer. | |
5163 | */ | |
1da177e4 | 5164 | agi = XFS_BUF_TO_AGI(agibp); |
d97d32ed | 5165 | xfs_buf_unlock(agibp); |
1da177e4 LT |
5166 | |
5167 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { | |
16259e7d | 5168 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
1da177e4 | 5169 | while (agino != NULLAGINO) { |
23fac50f CH |
5170 | agino = xlog_recover_process_one_iunlink(mp, |
5171 | agno, agino, bucket); | |
1da177e4 LT |
5172 | } |
5173 | } | |
d97d32ed | 5174 | xfs_buf_rele(agibp); |
1da177e4 | 5175 | } |
1da177e4 LT |
5176 | } |
5177 | ||
0e446be4 | 5178 | STATIC int |
1da177e4 | 5179 | xlog_unpack_data( |
9a8d2fdb | 5180 | struct xlog_rec_header *rhead, |
b2a922cd | 5181 | char *dp, |
9a8d2fdb | 5182 | struct xlog *log) |
1da177e4 LT |
5183 | { |
5184 | int i, j, k; | |
1da177e4 | 5185 | |
b53e675d | 5186 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
1da177e4 | 5187 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
b53e675d | 5188 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
1da177e4 LT |
5189 | dp += BBSIZE; |
5190 | } | |
5191 | ||
62118709 | 5192 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
b28708d6 | 5193 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
b53e675d | 5194 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
1da177e4 LT |
5195 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
5196 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 5197 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
1da177e4 LT |
5198 | dp += BBSIZE; |
5199 | } | |
5200 | } | |
0e446be4 CH |
5201 | |
5202 | return 0; | |
1da177e4 LT |
5203 | } |
5204 | ||
9d94901f | 5205 | /* |
b94fb2d1 | 5206 | * CRC check, unpack and process a log record. |
9d94901f BF |
5207 | */ |
5208 | STATIC int | |
5209 | xlog_recover_process( | |
5210 | struct xlog *log, | |
5211 | struct hlist_head rhash[], | |
5212 | struct xlog_rec_header *rhead, | |
5213 | char *dp, | |
12818d24 BF |
5214 | int pass, |
5215 | struct list_head *buffer_list) | |
9d94901f BF |
5216 | { |
5217 | int error; | |
cae028df | 5218 | __le32 old_crc = rhead->h_crc; |
b94fb2d1 BF |
5219 | __le32 crc; |
5220 | ||
cae028df | 5221 | |
6528250b BF |
5222 | crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len)); |
5223 | ||
b94fb2d1 | 5224 | /* |
6528250b BF |
5225 | * Nothing else to do if this is a CRC verification pass. Just return |
5226 | * if this a record with a non-zero crc. Unfortunately, mkfs always | |
cae028df | 5227 | * sets old_crc to 0 so we must consider this valid even on v5 supers. |
6528250b BF |
5228 | * Otherwise, return EFSBADCRC on failure so the callers up the stack |
5229 | * know precisely what failed. | |
5230 | */ | |
5231 | if (pass == XLOG_RECOVER_CRCPASS) { | |
cae028df | 5232 | if (old_crc && crc != old_crc) |
6528250b BF |
5233 | return -EFSBADCRC; |
5234 | return 0; | |
5235 | } | |
5236 | ||
5237 | /* | |
5238 | * We're in the normal recovery path. Issue a warning if and only if the | |
5239 | * CRC in the header is non-zero. This is an advisory warning and the | |
5240 | * zero CRC check prevents warnings from being emitted when upgrading | |
5241 | * the kernel from one that does not add CRCs by default. | |
b94fb2d1 | 5242 | */ |
cae028df DC |
5243 | if (crc != old_crc) { |
5244 | if (old_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) { | |
b94fb2d1 BF |
5245 | xfs_alert(log->l_mp, |
5246 | "log record CRC mismatch: found 0x%x, expected 0x%x.", | |
cae028df | 5247 | le32_to_cpu(old_crc), |
b94fb2d1 BF |
5248 | le32_to_cpu(crc)); |
5249 | xfs_hex_dump(dp, 32); | |
5250 | } | |
5251 | ||
5252 | /* | |
5253 | * If the filesystem is CRC enabled, this mismatch becomes a | |
5254 | * fatal log corruption failure. | |
5255 | */ | |
5256 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) | |
5257 | return -EFSCORRUPTED; | |
5258 | } | |
9d94901f BF |
5259 | |
5260 | error = xlog_unpack_data(rhead, dp, log); | |
5261 | if (error) | |
5262 | return error; | |
5263 | ||
12818d24 BF |
5264 | return xlog_recover_process_data(log, rhash, rhead, dp, pass, |
5265 | buffer_list); | |
9d94901f BF |
5266 | } |
5267 | ||
1da177e4 LT |
5268 | STATIC int |
5269 | xlog_valid_rec_header( | |
9a8d2fdb MT |
5270 | struct xlog *log, |
5271 | struct xlog_rec_header *rhead, | |
1da177e4 LT |
5272 | xfs_daddr_t blkno) |
5273 | { | |
5274 | int hlen; | |
5275 | ||
69ef921b | 5276 | if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) { |
1da177e4 LT |
5277 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
5278 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 5279 | return -EFSCORRUPTED; |
1da177e4 LT |
5280 | } |
5281 | if (unlikely( | |
5282 | (!rhead->h_version || | |
b53e675d | 5283 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
a0fa2b67 | 5284 | xfs_warn(log->l_mp, "%s: unrecognised log version (%d).", |
34a622b2 | 5285 | __func__, be32_to_cpu(rhead->h_version)); |
2451337d | 5286 | return -EIO; |
1da177e4 LT |
5287 | } |
5288 | ||
5289 | /* LR body must have data or it wouldn't have been written */ | |
b53e675d | 5290 | hlen = be32_to_cpu(rhead->h_len); |
1da177e4 LT |
5291 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
5292 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", | |
5293 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 5294 | return -EFSCORRUPTED; |
1da177e4 LT |
5295 | } |
5296 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { | |
5297 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", | |
5298 | XFS_ERRLEVEL_LOW, log->l_mp); | |
2451337d | 5299 | return -EFSCORRUPTED; |
1da177e4 LT |
5300 | } |
5301 | return 0; | |
5302 | } | |
5303 | ||
5304 | /* | |
5305 | * Read the log from tail to head and process the log records found. | |
5306 | * Handle the two cases where the tail and head are in the same cycle | |
5307 | * and where the active portion of the log wraps around the end of | |
5308 | * the physical log separately. The pass parameter is passed through | |
5309 | * to the routines called to process the data and is not looked at | |
5310 | * here. | |
5311 | */ | |
5312 | STATIC int | |
5313 | xlog_do_recovery_pass( | |
9a8d2fdb | 5314 | struct xlog *log, |
1da177e4 LT |
5315 | xfs_daddr_t head_blk, |
5316 | xfs_daddr_t tail_blk, | |
d7f37692 BF |
5317 | int pass, |
5318 | xfs_daddr_t *first_bad) /* out: first bad log rec */ | |
1da177e4 LT |
5319 | { |
5320 | xlog_rec_header_t *rhead; | |
284f1c2c | 5321 | xfs_daddr_t blk_no, rblk_no; |
d7f37692 | 5322 | xfs_daddr_t rhead_blk; |
b2a922cd | 5323 | char *offset; |
1da177e4 | 5324 | xfs_buf_t *hbp, *dbp; |
a70f9fe5 | 5325 | int error = 0, h_size, h_len; |
12818d24 | 5326 | int error2 = 0; |
1da177e4 LT |
5327 | int bblks, split_bblks; |
5328 | int hblks, split_hblks, wrapped_hblks; | |
39775431 | 5329 | int i; |
f0a76953 | 5330 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
12818d24 | 5331 | LIST_HEAD (buffer_list); |
1da177e4 LT |
5332 | |
5333 | ASSERT(head_blk != tail_blk); | |
a4c9b34d | 5334 | blk_no = rhead_blk = tail_blk; |
1da177e4 | 5335 | |
39775431 BF |
5336 | for (i = 0; i < XLOG_RHASH_SIZE; i++) |
5337 | INIT_HLIST_HEAD(&rhash[i]); | |
5338 | ||
1da177e4 LT |
5339 | /* |
5340 | * Read the header of the tail block and get the iclog buffer size from | |
5341 | * h_size. Use this to tell how many sectors make up the log header. | |
5342 | */ | |
62118709 | 5343 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
1da177e4 LT |
5344 | /* |
5345 | * When using variable length iclogs, read first sector of | |
5346 | * iclog header and extract the header size from it. Get a | |
5347 | * new hbp that is the correct size. | |
5348 | */ | |
5349 | hbp = xlog_get_bp(log, 1); | |
5350 | if (!hbp) | |
2451337d | 5351 | return -ENOMEM; |
076e6acb CH |
5352 | |
5353 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); | |
5354 | if (error) | |
1da177e4 | 5355 | goto bread_err1; |
076e6acb | 5356 | |
1da177e4 LT |
5357 | rhead = (xlog_rec_header_t *)offset; |
5358 | error = xlog_valid_rec_header(log, rhead, tail_blk); | |
5359 | if (error) | |
5360 | goto bread_err1; | |
a70f9fe5 BF |
5361 | |
5362 | /* | |
5363 | * xfsprogs has a bug where record length is based on lsunit but | |
5364 | * h_size (iclog size) is hardcoded to 32k. Now that we | |
5365 | * unconditionally CRC verify the unmount record, this means the | |
5366 | * log buffer can be too small for the record and cause an | |
5367 | * overrun. | |
5368 | * | |
5369 | * Detect this condition here. Use lsunit for the buffer size as | |
5370 | * long as this looks like the mkfs case. Otherwise, return an | |
5371 | * error to avoid a buffer overrun. | |
5372 | */ | |
b53e675d | 5373 | h_size = be32_to_cpu(rhead->h_size); |
a70f9fe5 BF |
5374 | h_len = be32_to_cpu(rhead->h_len); |
5375 | if (h_len > h_size) { | |
5376 | if (h_len <= log->l_mp->m_logbsize && | |
5377 | be32_to_cpu(rhead->h_num_logops) == 1) { | |
5378 | xfs_warn(log->l_mp, | |
5379 | "invalid iclog size (%d bytes), using lsunit (%d bytes)", | |
5380 | h_size, log->l_mp->m_logbsize); | |
5381 | h_size = log->l_mp->m_logbsize; | |
5382 | } else | |
5383 | return -EFSCORRUPTED; | |
5384 | } | |
5385 | ||
b53e675d | 5386 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && |
1da177e4 LT |
5387 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
5388 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
5389 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
5390 | hblks++; | |
5391 | xlog_put_bp(hbp); | |
5392 | hbp = xlog_get_bp(log, hblks); | |
5393 | } else { | |
5394 | hblks = 1; | |
5395 | } | |
5396 | } else { | |
69ce58f0 | 5397 | ASSERT(log->l_sectBBsize == 1); |
1da177e4 LT |
5398 | hblks = 1; |
5399 | hbp = xlog_get_bp(log, 1); | |
5400 | h_size = XLOG_BIG_RECORD_BSIZE; | |
5401 | } | |
5402 | ||
5403 | if (!hbp) | |
2451337d | 5404 | return -ENOMEM; |
1da177e4 LT |
5405 | dbp = xlog_get_bp(log, BTOBB(h_size)); |
5406 | if (!dbp) { | |
5407 | xlog_put_bp(hbp); | |
2451337d | 5408 | return -ENOMEM; |
1da177e4 LT |
5409 | } |
5410 | ||
5411 | memset(rhash, 0, sizeof(rhash)); | |
970fd3f0 | 5412 | if (tail_blk > head_blk) { |
1da177e4 LT |
5413 | /* |
5414 | * Perform recovery around the end of the physical log. | |
5415 | * When the head is not on the same cycle number as the tail, | |
970fd3f0 | 5416 | * we can't do a sequential recovery. |
1da177e4 | 5417 | */ |
1da177e4 LT |
5418 | while (blk_no < log->l_logBBsize) { |
5419 | /* | |
5420 | * Check for header wrapping around physical end-of-log | |
5421 | */ | |
62926044 | 5422 | offset = hbp->b_addr; |
1da177e4 LT |
5423 | split_hblks = 0; |
5424 | wrapped_hblks = 0; | |
5425 | if (blk_no + hblks <= log->l_logBBsize) { | |
5426 | /* Read header in one read */ | |
076e6acb CH |
5427 | error = xlog_bread(log, blk_no, hblks, hbp, |
5428 | &offset); | |
1da177e4 LT |
5429 | if (error) |
5430 | goto bread_err2; | |
1da177e4 LT |
5431 | } else { |
5432 | /* This LR is split across physical log end */ | |
5433 | if (blk_no != log->l_logBBsize) { | |
5434 | /* some data before physical log end */ | |
5435 | ASSERT(blk_no <= INT_MAX); | |
5436 | split_hblks = log->l_logBBsize - (int)blk_no; | |
5437 | ASSERT(split_hblks > 0); | |
076e6acb CH |
5438 | error = xlog_bread(log, blk_no, |
5439 | split_hblks, hbp, | |
5440 | &offset); | |
5441 | if (error) | |
1da177e4 | 5442 | goto bread_err2; |
1da177e4 | 5443 | } |
076e6acb | 5444 | |
1da177e4 LT |
5445 | /* |
5446 | * Note: this black magic still works with | |
5447 | * large sector sizes (non-512) only because: | |
5448 | * - we increased the buffer size originally | |
5449 | * by 1 sector giving us enough extra space | |
5450 | * for the second read; | |
5451 | * - the log start is guaranteed to be sector | |
5452 | * aligned; | |
5453 | * - we read the log end (LR header start) | |
5454 | * _first_, then the log start (LR header end) | |
5455 | * - order is important. | |
5456 | */ | |
234f56ac | 5457 | wrapped_hblks = hblks - split_hblks; |
44396476 DC |
5458 | error = xlog_bread_offset(log, 0, |
5459 | wrapped_hblks, hbp, | |
5460 | offset + BBTOB(split_hblks)); | |
1da177e4 LT |
5461 | if (error) |
5462 | goto bread_err2; | |
1da177e4 LT |
5463 | } |
5464 | rhead = (xlog_rec_header_t *)offset; | |
5465 | error = xlog_valid_rec_header(log, rhead, | |
5466 | split_hblks ? blk_no : 0); | |
5467 | if (error) | |
5468 | goto bread_err2; | |
5469 | ||
b53e675d | 5470 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
1da177e4 LT |
5471 | blk_no += hblks; |
5472 | ||
284f1c2c BF |
5473 | /* |
5474 | * Read the log record data in multiple reads if it | |
5475 | * wraps around the end of the log. Note that if the | |
5476 | * header already wrapped, blk_no could point past the | |
5477 | * end of the log. The record data is contiguous in | |
5478 | * that case. | |
5479 | */ | |
5480 | if (blk_no + bblks <= log->l_logBBsize || | |
5481 | blk_no >= log->l_logBBsize) { | |
0703a8e1 | 5482 | rblk_no = xlog_wrap_logbno(log, blk_no); |
284f1c2c | 5483 | error = xlog_bread(log, rblk_no, bblks, dbp, |
076e6acb | 5484 | &offset); |
1da177e4 LT |
5485 | if (error) |
5486 | goto bread_err2; | |
1da177e4 LT |
5487 | } else { |
5488 | /* This log record is split across the | |
5489 | * physical end of log */ | |
62926044 | 5490 | offset = dbp->b_addr; |
1da177e4 LT |
5491 | split_bblks = 0; |
5492 | if (blk_no != log->l_logBBsize) { | |
5493 | /* some data is before the physical | |
5494 | * end of log */ | |
5495 | ASSERT(!wrapped_hblks); | |
5496 | ASSERT(blk_no <= INT_MAX); | |
5497 | split_bblks = | |
5498 | log->l_logBBsize - (int)blk_no; | |
5499 | ASSERT(split_bblks > 0); | |
076e6acb CH |
5500 | error = xlog_bread(log, blk_no, |
5501 | split_bblks, dbp, | |
5502 | &offset); | |
5503 | if (error) | |
1da177e4 | 5504 | goto bread_err2; |
1da177e4 | 5505 | } |
076e6acb | 5506 | |
1da177e4 LT |
5507 | /* |
5508 | * Note: this black magic still works with | |
5509 | * large sector sizes (non-512) only because: | |
5510 | * - we increased the buffer size originally | |
5511 | * by 1 sector giving us enough extra space | |
5512 | * for the second read; | |
5513 | * - the log start is guaranteed to be sector | |
5514 | * aligned; | |
5515 | * - we read the log end (LR header start) | |
5516 | * _first_, then the log start (LR header end) | |
5517 | * - order is important. | |
5518 | */ | |
44396476 | 5519 | error = xlog_bread_offset(log, 0, |
009507b0 | 5520 | bblks - split_bblks, dbp, |
44396476 | 5521 | offset + BBTOB(split_bblks)); |
076e6acb CH |
5522 | if (error) |
5523 | goto bread_err2; | |
1da177e4 | 5524 | } |
0e446be4 | 5525 | |
9d94901f | 5526 | error = xlog_recover_process(log, rhash, rhead, offset, |
12818d24 | 5527 | pass, &buffer_list); |
0e446be4 | 5528 | if (error) |
1da177e4 | 5529 | goto bread_err2; |
d7f37692 | 5530 | |
1da177e4 | 5531 | blk_no += bblks; |
d7f37692 | 5532 | rhead_blk = blk_no; |
1da177e4 LT |
5533 | } |
5534 | ||
5535 | ASSERT(blk_no >= log->l_logBBsize); | |
5536 | blk_no -= log->l_logBBsize; | |
d7f37692 | 5537 | rhead_blk = blk_no; |
970fd3f0 | 5538 | } |
1da177e4 | 5539 | |
970fd3f0 ES |
5540 | /* read first part of physical log */ |
5541 | while (blk_no < head_blk) { | |
5542 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); | |
5543 | if (error) | |
5544 | goto bread_err2; | |
076e6acb | 5545 | |
970fd3f0 ES |
5546 | rhead = (xlog_rec_header_t *)offset; |
5547 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
5548 | if (error) | |
5549 | goto bread_err2; | |
076e6acb | 5550 | |
970fd3f0 ES |
5551 | /* blocks in data section */ |
5552 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); | |
5553 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, | |
5554 | &offset); | |
5555 | if (error) | |
5556 | goto bread_err2; | |
076e6acb | 5557 | |
12818d24 BF |
5558 | error = xlog_recover_process(log, rhash, rhead, offset, pass, |
5559 | &buffer_list); | |
970fd3f0 ES |
5560 | if (error) |
5561 | goto bread_err2; | |
d7f37692 | 5562 | |
970fd3f0 | 5563 | blk_no += bblks + hblks; |
d7f37692 | 5564 | rhead_blk = blk_no; |
1da177e4 LT |
5565 | } |
5566 | ||
5567 | bread_err2: | |
5568 | xlog_put_bp(dbp); | |
5569 | bread_err1: | |
5570 | xlog_put_bp(hbp); | |
d7f37692 | 5571 | |
12818d24 BF |
5572 | /* |
5573 | * Submit buffers that have been added from the last record processed, | |
5574 | * regardless of error status. | |
5575 | */ | |
5576 | if (!list_empty(&buffer_list)) | |
5577 | error2 = xfs_buf_delwri_submit(&buffer_list); | |
5578 | ||
d7f37692 BF |
5579 | if (error && first_bad) |
5580 | *first_bad = rhead_blk; | |
5581 | ||
39775431 BF |
5582 | /* |
5583 | * Transactions are freed at commit time but transactions without commit | |
5584 | * records on disk are never committed. Free any that may be left in the | |
5585 | * hash table. | |
5586 | */ | |
5587 | for (i = 0; i < XLOG_RHASH_SIZE; i++) { | |
5588 | struct hlist_node *tmp; | |
5589 | struct xlog_recover *trans; | |
5590 | ||
5591 | hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list) | |
5592 | xlog_recover_free_trans(trans); | |
5593 | } | |
5594 | ||
12818d24 | 5595 | return error ? error : error2; |
1da177e4 LT |
5596 | } |
5597 | ||
5598 | /* | |
5599 | * Do the recovery of the log. We actually do this in two phases. | |
5600 | * The two passes are necessary in order to implement the function | |
5601 | * of cancelling a record written into the log. The first pass | |
5602 | * determines those things which have been cancelled, and the | |
5603 | * second pass replays log items normally except for those which | |
5604 | * have been cancelled. The handling of the replay and cancellations | |
5605 | * takes place in the log item type specific routines. | |
5606 | * | |
5607 | * The table of items which have cancel records in the log is allocated | |
5608 | * and freed at this level, since only here do we know when all of | |
5609 | * the log recovery has been completed. | |
5610 | */ | |
5611 | STATIC int | |
5612 | xlog_do_log_recovery( | |
9a8d2fdb | 5613 | struct xlog *log, |
1da177e4 LT |
5614 | xfs_daddr_t head_blk, |
5615 | xfs_daddr_t tail_blk) | |
5616 | { | |
d5689eaa | 5617 | int error, i; |
1da177e4 LT |
5618 | |
5619 | ASSERT(head_blk != tail_blk); | |
5620 | ||
5621 | /* | |
5622 | * First do a pass to find all of the cancelled buf log items. | |
5623 | * Store them in the buf_cancel_table for use in the second pass. | |
5624 | */ | |
d5689eaa CH |
5625 | log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE * |
5626 | sizeof(struct list_head), | |
1da177e4 | 5627 | KM_SLEEP); |
d5689eaa CH |
5628 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
5629 | INIT_LIST_HEAD(&log->l_buf_cancel_table[i]); | |
5630 | ||
1da177e4 | 5631 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
d7f37692 | 5632 | XLOG_RECOVER_PASS1, NULL); |
1da177e4 | 5633 | if (error != 0) { |
f0e2d93c | 5634 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
5635 | log->l_buf_cancel_table = NULL; |
5636 | return error; | |
5637 | } | |
5638 | /* | |
5639 | * Then do a second pass to actually recover the items in the log. | |
5640 | * When it is complete free the table of buf cancel items. | |
5641 | */ | |
5642 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
d7f37692 | 5643 | XLOG_RECOVER_PASS2, NULL); |
1da177e4 | 5644 | #ifdef DEBUG |
6d192a9b | 5645 | if (!error) { |
1da177e4 LT |
5646 | int i; |
5647 | ||
5648 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
d5689eaa | 5649 | ASSERT(list_empty(&log->l_buf_cancel_table[i])); |
1da177e4 LT |
5650 | } |
5651 | #endif /* DEBUG */ | |
5652 | ||
f0e2d93c | 5653 | kmem_free(log->l_buf_cancel_table); |
1da177e4 LT |
5654 | log->l_buf_cancel_table = NULL; |
5655 | ||
5656 | return error; | |
5657 | } | |
5658 | ||
5659 | /* | |
5660 | * Do the actual recovery | |
5661 | */ | |
5662 | STATIC int | |
5663 | xlog_do_recover( | |
9a8d2fdb | 5664 | struct xlog *log, |
1da177e4 LT |
5665 | xfs_daddr_t head_blk, |
5666 | xfs_daddr_t tail_blk) | |
5667 | { | |
a798011c | 5668 | struct xfs_mount *mp = log->l_mp; |
1da177e4 LT |
5669 | int error; |
5670 | xfs_buf_t *bp; | |
5671 | xfs_sb_t *sbp; | |
5672 | ||
e67d3d42 BF |
5673 | trace_xfs_log_recover(log, head_blk, tail_blk); |
5674 | ||
1da177e4 LT |
5675 | /* |
5676 | * First replay the images in the log. | |
5677 | */ | |
5678 | error = xlog_do_log_recovery(log, head_blk, tail_blk); | |
43ff2122 | 5679 | if (error) |
1da177e4 | 5680 | return error; |
1da177e4 LT |
5681 | |
5682 | /* | |
5683 | * If IO errors happened during recovery, bail out. | |
5684 | */ | |
a798011c | 5685 | if (XFS_FORCED_SHUTDOWN(mp)) { |
2451337d | 5686 | return -EIO; |
1da177e4 LT |
5687 | } |
5688 | ||
5689 | /* | |
5690 | * We now update the tail_lsn since much of the recovery has completed | |
5691 | * and there may be space available to use. If there were no extent | |
5692 | * or iunlinks, we can free up the entire log and set the tail_lsn to | |
5693 | * be the last_sync_lsn. This was set in xlog_find_tail to be the | |
5694 | * lsn of the last known good LR on disk. If there are extent frees | |
5695 | * or iunlinks they will have some entries in the AIL; so we look at | |
5696 | * the AIL to determine how to set the tail_lsn. | |
5697 | */ | |
a798011c | 5698 | xlog_assign_tail_lsn(mp); |
1da177e4 LT |
5699 | |
5700 | /* | |
5701 | * Now that we've finished replaying all buffer and inode | |
98021821 | 5702 | * updates, re-read in the superblock and reverify it. |
1da177e4 | 5703 | */ |
a798011c | 5704 | bp = xfs_getsb(mp, 0); |
1157b32c | 5705 | bp->b_flags &= ~(XBF_DONE | XBF_ASYNC); |
b68c0821 | 5706 | ASSERT(!(bp->b_flags & XBF_WRITE)); |
0cac682f | 5707 | bp->b_flags |= XBF_READ; |
1813dd64 | 5708 | bp->b_ops = &xfs_sb_buf_ops; |
83a0adc3 | 5709 | |
595bff75 | 5710 | error = xfs_buf_submit_wait(bp); |
d64e31a2 | 5711 | if (error) { |
a798011c | 5712 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
595bff75 DC |
5713 | xfs_buf_ioerror_alert(bp, __func__); |
5714 | ASSERT(0); | |
5715 | } | |
1da177e4 LT |
5716 | xfs_buf_relse(bp); |
5717 | return error; | |
5718 | } | |
5719 | ||
5720 | /* Convert superblock from on-disk format */ | |
a798011c | 5721 | sbp = &mp->m_sb; |
98021821 | 5722 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
1da177e4 LT |
5723 | xfs_buf_relse(bp); |
5724 | ||
a798011c DC |
5725 | /* re-initialise in-core superblock and geometry structures */ |
5726 | xfs_reinit_percpu_counters(mp); | |
5727 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); | |
5728 | if (error) { | |
5729 | xfs_warn(mp, "Failed post-recovery per-ag init: %d", error); | |
5730 | return error; | |
5731 | } | |
52548852 | 5732 | mp->m_alloc_set_aside = xfs_alloc_set_aside(mp); |
5478eead | 5733 | |
1da177e4 LT |
5734 | xlog_recover_check_summary(log); |
5735 | ||
5736 | /* Normal transactions can now occur */ | |
5737 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
5738 | return 0; | |
5739 | } | |
5740 | ||
5741 | /* | |
5742 | * Perform recovery and re-initialize some log variables in xlog_find_tail. | |
5743 | * | |
5744 | * Return error or zero. | |
5745 | */ | |
5746 | int | |
5747 | xlog_recover( | |
9a8d2fdb | 5748 | struct xlog *log) |
1da177e4 LT |
5749 | { |
5750 | xfs_daddr_t head_blk, tail_blk; | |
5751 | int error; | |
5752 | ||
5753 | /* find the tail of the log */ | |
a45086e2 BF |
5754 | error = xlog_find_tail(log, &head_blk, &tail_blk); |
5755 | if (error) | |
1da177e4 LT |
5756 | return error; |
5757 | ||
a45086e2 BF |
5758 | /* |
5759 | * The superblock was read before the log was available and thus the LSN | |
5760 | * could not be verified. Check the superblock LSN against the current | |
5761 | * LSN now that it's known. | |
5762 | */ | |
5763 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb) && | |
5764 | !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn)) | |
5765 | return -EINVAL; | |
5766 | ||
1da177e4 LT |
5767 | if (tail_blk != head_blk) { |
5768 | /* There used to be a comment here: | |
5769 | * | |
5770 | * disallow recovery on read-only mounts. note -- mount | |
5771 | * checks for ENOSPC and turns it into an intelligent | |
5772 | * error message. | |
5773 | * ...but this is no longer true. Now, unless you specify | |
5774 | * NORECOVERY (in which case this function would never be | |
5775 | * called), we just go ahead and recover. We do this all | |
5776 | * under the vfs layer, so we can get away with it unless | |
5777 | * the device itself is read-only, in which case we fail. | |
5778 | */ | |
3a02ee18 | 5779 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
1da177e4 LT |
5780 | return error; |
5781 | } | |
5782 | ||
e721f504 DC |
5783 | /* |
5784 | * Version 5 superblock log feature mask validation. We know the | |
5785 | * log is dirty so check if there are any unknown log features | |
5786 | * in what we need to recover. If there are unknown features | |
5787 | * (e.g. unsupported transactions, then simply reject the | |
5788 | * attempt at recovery before touching anything. | |
5789 | */ | |
5790 | if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 && | |
5791 | xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb, | |
5792 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) { | |
5793 | xfs_warn(log->l_mp, | |
f41febd2 | 5794 | "Superblock has unknown incompatible log features (0x%x) enabled.", |
e721f504 DC |
5795 | (log->l_mp->m_sb.sb_features_log_incompat & |
5796 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); | |
f41febd2 JP |
5797 | xfs_warn(log->l_mp, |
5798 | "The log can not be fully and/or safely recovered by this kernel."); | |
5799 | xfs_warn(log->l_mp, | |
5800 | "Please recover the log on a kernel that supports the unknown features."); | |
2451337d | 5801 | return -EINVAL; |
e721f504 DC |
5802 | } |
5803 | ||
2e227178 BF |
5804 | /* |
5805 | * Delay log recovery if the debug hook is set. This is debug | |
5806 | * instrumention to coordinate simulation of I/O failures with | |
5807 | * log recovery. | |
5808 | */ | |
5809 | if (xfs_globals.log_recovery_delay) { | |
5810 | xfs_notice(log->l_mp, | |
5811 | "Delaying log recovery for %d seconds.", | |
5812 | xfs_globals.log_recovery_delay); | |
5813 | msleep(xfs_globals.log_recovery_delay * 1000); | |
5814 | } | |
5815 | ||
a0fa2b67 DC |
5816 | xfs_notice(log->l_mp, "Starting recovery (logdev: %s)", |
5817 | log->l_mp->m_logname ? log->l_mp->m_logname | |
5818 | : "internal"); | |
1da177e4 LT |
5819 | |
5820 | error = xlog_do_recover(log, head_blk, tail_blk); | |
5821 | log->l_flags |= XLOG_RECOVERY_NEEDED; | |
5822 | } | |
5823 | return error; | |
5824 | } | |
5825 | ||
5826 | /* | |
5827 | * In the first part of recovery we replay inodes and buffers and build | |
5828 | * up the list of extent free items which need to be processed. Here | |
5829 | * we process the extent free items and clean up the on disk unlinked | |
5830 | * inode lists. This is separated from the first part of recovery so | |
5831 | * that the root and real-time bitmap inodes can be read in from disk in | |
5832 | * between the two stages. This is necessary so that we can free space | |
5833 | * in the real-time portion of the file system. | |
5834 | */ | |
5835 | int | |
5836 | xlog_recover_finish( | |
9a8d2fdb | 5837 | struct xlog *log) |
1da177e4 LT |
5838 | { |
5839 | /* | |
5840 | * Now we're ready to do the transactions needed for the | |
5841 | * rest of recovery. Start with completing all the extent | |
5842 | * free intent records and then process the unlinked inode | |
5843 | * lists. At this point, we essentially run in normal mode | |
5844 | * except that we're still performing recovery actions | |
5845 | * rather than accepting new requests. | |
5846 | */ | |
5847 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { | |
3c1e2bbe | 5848 | int error; |
dc42375d | 5849 | error = xlog_recover_process_intents(log); |
3c1e2bbe | 5850 | if (error) { |
dc42375d | 5851 | xfs_alert(log->l_mp, "Failed to recover intents"); |
3c1e2bbe DC |
5852 | return error; |
5853 | } | |
9e88b5d8 | 5854 | |
1da177e4 | 5855 | /* |
dc42375d | 5856 | * Sync the log to get all the intents out of the AIL. |
1da177e4 LT |
5857 | * This isn't absolutely necessary, but it helps in |
5858 | * case the unlink transactions would have problems | |
dc42375d | 5859 | * pushing the intents out of the way. |
1da177e4 | 5860 | */ |
a14a348b | 5861 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
1da177e4 | 5862 | |
4249023a | 5863 | xlog_recover_process_iunlinks(log); |
1da177e4 LT |
5864 | |
5865 | xlog_recover_check_summary(log); | |
5866 | ||
a0fa2b67 DC |
5867 | xfs_notice(log->l_mp, "Ending recovery (logdev: %s)", |
5868 | log->l_mp->m_logname ? log->l_mp->m_logname | |
5869 | : "internal"); | |
1da177e4 LT |
5870 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
5871 | } else { | |
a0fa2b67 | 5872 | xfs_info(log->l_mp, "Ending clean mount"); |
1da177e4 LT |
5873 | } |
5874 | return 0; | |
5875 | } | |
5876 | ||
f0b2efad BF |
5877 | int |
5878 | xlog_recover_cancel( | |
5879 | struct xlog *log) | |
5880 | { | |
5881 | int error = 0; | |
5882 | ||
5883 | if (log->l_flags & XLOG_RECOVERY_NEEDED) | |
dc42375d | 5884 | error = xlog_recover_cancel_intents(log); |
f0b2efad BF |
5885 | |
5886 | return error; | |
5887 | } | |
1da177e4 LT |
5888 | |
5889 | #if defined(DEBUG) | |
5890 | /* | |
5891 | * Read all of the agf and agi counters and check that they | |
5892 | * are consistent with the superblock counters. | |
5893 | */ | |
e89fbb5e | 5894 | STATIC void |
1da177e4 | 5895 | xlog_recover_check_summary( |
9a8d2fdb | 5896 | struct xlog *log) |
1da177e4 LT |
5897 | { |
5898 | xfs_mount_t *mp; | |
5899 | xfs_agf_t *agfp; | |
1da177e4 LT |
5900 | xfs_buf_t *agfbp; |
5901 | xfs_buf_t *agibp; | |
1da177e4 | 5902 | xfs_agnumber_t agno; |
c8ce540d DW |
5903 | uint64_t freeblks; |
5904 | uint64_t itotal; | |
5905 | uint64_t ifree; | |
5e1be0fb | 5906 | int error; |
1da177e4 LT |
5907 | |
5908 | mp = log->l_mp; | |
5909 | ||
5910 | freeblks = 0LL; | |
5911 | itotal = 0LL; | |
5912 | ifree = 0LL; | |
5913 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
4805621a FCH |
5914 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
5915 | if (error) { | |
a0fa2b67 DC |
5916 | xfs_alert(mp, "%s agf read failed agno %d error %d", |
5917 | __func__, agno, error); | |
4805621a FCH |
5918 | } else { |
5919 | agfp = XFS_BUF_TO_AGF(agfbp); | |
5920 | freeblks += be32_to_cpu(agfp->agf_freeblks) + | |
5921 | be32_to_cpu(agfp->agf_flcount); | |
5922 | xfs_buf_relse(agfbp); | |
1da177e4 | 5923 | } |
1da177e4 | 5924 | |
5e1be0fb | 5925 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
a0fa2b67 DC |
5926 | if (error) { |
5927 | xfs_alert(mp, "%s agi read failed agno %d error %d", | |
5928 | __func__, agno, error); | |
5929 | } else { | |
5e1be0fb | 5930 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); |
16259e7d | 5931 | |
5e1be0fb CH |
5932 | itotal += be32_to_cpu(agi->agi_count); |
5933 | ifree += be32_to_cpu(agi->agi_freecount); | |
5934 | xfs_buf_relse(agibp); | |
5935 | } | |
1da177e4 | 5936 | } |
1da177e4 LT |
5937 | } |
5938 | #endif /* DEBUG */ |