1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_ialloc.h"
16 #include "xfs_alloc.h"
17 #include "xfs_error.h"
18 #include "xfs_trans.h"
19 #include "xfs_buf_item.h"
20 #include "xfs_bmap_btree.h"
21 #include "xfs_alloc_btree.h"
22 #include "xfs_log.h"
23 #include "xfs_rmap_btree.h"
24 #include "xfs_refcount_btree.h"
25 #include "xfs_da_format.h"
26 #include "xfs_health.h"
27 #include "xfs_ag.h"
28 #include "xfs_rtbitmap.h"
29
30 /*
31 * Physical superblock buffer manipulations. Shared with libxfs in userspace.
32 */
33
34 /*
35 * Check that all the V4 feature bits that the V5 filesystem format requires are
36 * correctly set.
37 */
38 static bool
xfs_sb_validate_v5_features(struct xfs_sb * sbp)39 xfs_sb_validate_v5_features(
40 struct xfs_sb *sbp)
41 {
42 /* We must not have any unknown V4 feature bits set */
43 if (sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS)
44 return false;
45
46 /*
47 * The CRC bit is considered an invalid V4 flag, so we have to add it
48 * manually to the OKBITS mask.
49 */
50 if (sbp->sb_features2 & ~(XFS_SB_VERSION2_OKBITS |
51 XFS_SB_VERSION2_CRCBIT))
52 return false;
53
54 /* Now check all the required V4 feature flags are set. */
55
56 #define V5_VERS_FLAGS (XFS_SB_VERSION_NLINKBIT | \
57 XFS_SB_VERSION_ALIGNBIT | \
58 XFS_SB_VERSION_LOGV2BIT | \
59 XFS_SB_VERSION_EXTFLGBIT | \
60 XFS_SB_VERSION_DIRV2BIT | \
61 XFS_SB_VERSION_MOREBITSBIT)
62
63 #define V5_FEAT_FLAGS (XFS_SB_VERSION2_LAZYSBCOUNTBIT | \
64 XFS_SB_VERSION2_ATTR2BIT | \
65 XFS_SB_VERSION2_PROJID32BIT | \
66 XFS_SB_VERSION2_CRCBIT)
67
68 if ((sbp->sb_versionnum & V5_VERS_FLAGS) != V5_VERS_FLAGS)
69 return false;
70 if ((sbp->sb_features2 & V5_FEAT_FLAGS) != V5_FEAT_FLAGS)
71 return false;
72 return true;
73 }
74
75 /*
76 * We current support XFS v5 formats with known features and v4 superblocks with
77 * at least V2 directories.
78 */
79 bool
xfs_sb_good_version(struct xfs_sb * sbp)80 xfs_sb_good_version(
81 struct xfs_sb *sbp)
82 {
83 /*
84 * All v5 filesystems are supported, but we must check that all the
85 * required v4 feature flags are enabled correctly as the code checks
86 * those flags and not for v5 support.
87 */
88 if (xfs_sb_is_v5(sbp))
89 return xfs_sb_validate_v5_features(sbp);
90
91 /* versions prior to v4 are not supported */
92 if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_4)
93 return false;
94
95 /* We must not have any unknown v4 feature bits set */
96 if ((sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS) ||
97 ((sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) &&
98 (sbp->sb_features2 & ~XFS_SB_VERSION2_OKBITS)))
99 return false;
100
101 /* V4 filesystems need v2 directories and unwritten extents */
102 if (!(sbp->sb_versionnum & XFS_SB_VERSION_DIRV2BIT))
103 return false;
104 if (!(sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT))
105 return false;
106
107 /* It's a supported v4 filesystem */
108 return true;
109 }
110
111 uint64_t
xfs_sb_version_to_features(struct xfs_sb * sbp)112 xfs_sb_version_to_features(
113 struct xfs_sb *sbp)
114 {
115 uint64_t features = 0;
116
117 /* optional V4 features */
118 if (sbp->sb_rblocks > 0)
119 features |= XFS_FEAT_REALTIME;
120 if (sbp->sb_versionnum & XFS_SB_VERSION_NLINKBIT)
121 features |= XFS_FEAT_NLINK;
122 if (sbp->sb_versionnum & XFS_SB_VERSION_ATTRBIT)
123 features |= XFS_FEAT_ATTR;
124 if (sbp->sb_versionnum & XFS_SB_VERSION_QUOTABIT)
125 features |= XFS_FEAT_QUOTA;
126 if (sbp->sb_versionnum & XFS_SB_VERSION_ALIGNBIT)
127 features |= XFS_FEAT_ALIGN;
128 if (sbp->sb_versionnum & XFS_SB_VERSION_LOGV2BIT)
129 features |= XFS_FEAT_LOGV2;
130 if (sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT)
131 features |= XFS_FEAT_DALIGN;
132 if (sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT)
133 features |= XFS_FEAT_EXTFLG;
134 if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT)
135 features |= XFS_FEAT_SECTOR;
136 if (sbp->sb_versionnum & XFS_SB_VERSION_BORGBIT)
137 features |= XFS_FEAT_ASCIICI;
138 if (sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) {
139 if (sbp->sb_features2 & XFS_SB_VERSION2_LAZYSBCOUNTBIT)
140 features |= XFS_FEAT_LAZYSBCOUNT;
141 if (sbp->sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
142 features |= XFS_FEAT_ATTR2;
143 if (sbp->sb_features2 & XFS_SB_VERSION2_PROJID32BIT)
144 features |= XFS_FEAT_PROJID32;
145 if (sbp->sb_features2 & XFS_SB_VERSION2_FTYPE)
146 features |= XFS_FEAT_FTYPE;
147 }
148
149 if (!xfs_sb_is_v5(sbp))
150 return features;
151
152 /* Always on V5 features */
153 features |= XFS_FEAT_ALIGN | XFS_FEAT_LOGV2 | XFS_FEAT_EXTFLG |
154 XFS_FEAT_LAZYSBCOUNT | XFS_FEAT_ATTR2 | XFS_FEAT_PROJID32 |
155 XFS_FEAT_V3INODES | XFS_FEAT_CRC | XFS_FEAT_PQUOTINO;
156
157 /* Optional V5 features */
158 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_FINOBT)
159 features |= XFS_FEAT_FINOBT;
160 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_RMAPBT)
161 features |= XFS_FEAT_RMAPBT;
162 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_REFLINK)
163 features |= XFS_FEAT_REFLINK;
164 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_INOBTCNT)
165 features |= XFS_FEAT_INOBTCNT;
166 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_FTYPE)
167 features |= XFS_FEAT_FTYPE;
168 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES)
169 features |= XFS_FEAT_SPINODES;
170 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
171 features |= XFS_FEAT_META_UUID;
172 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_BIGTIME)
173 features |= XFS_FEAT_BIGTIME;
174 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR)
175 features |= XFS_FEAT_NEEDSREPAIR;
176 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NREXT64)
177 features |= XFS_FEAT_NREXT64;
178
179 return features;
180 }
181
182 /* Check all the superblock fields we care about when reading one in. */
183 STATIC int
xfs_validate_sb_read(struct xfs_mount * mp,struct xfs_sb * sbp)184 xfs_validate_sb_read(
185 struct xfs_mount *mp,
186 struct xfs_sb *sbp)
187 {
188 if (!xfs_sb_is_v5(sbp))
189 return 0;
190
191 /*
192 * Version 5 superblock feature mask validation. Reject combinations
193 * the kernel cannot support up front before checking anything else.
194 */
195 if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
196 xfs_warn(mp,
197 "Superblock has unknown compatible features (0x%x) enabled.",
198 (sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
199 xfs_warn(mp,
200 "Using a more recent kernel is recommended.");
201 }
202
203 if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
204 xfs_alert(mp,
205 "Superblock has unknown read-only compatible features (0x%x) enabled.",
206 (sbp->sb_features_ro_compat &
207 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
208 if (!xfs_is_readonly(mp)) {
209 xfs_warn(mp,
210 "Attempted to mount read-only compatible filesystem read-write.");
211 xfs_warn(mp,
212 "Filesystem can only be safely mounted read only.");
213
214 return -EINVAL;
215 }
216 }
217 if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
218 xfs_warn(mp,
219 "Superblock has unknown incompatible features (0x%x) enabled.",
220 (sbp->sb_features_incompat &
221 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
222 xfs_warn(mp,
223 "Filesystem cannot be safely mounted by this kernel.");
224 return -EINVAL;
225 }
226
227 return 0;
228 }
229
230 /* Check all the superblock fields we care about when writing one out. */
231 STATIC int
xfs_validate_sb_write(struct xfs_mount * mp,struct xfs_buf * bp,struct xfs_sb * sbp)232 xfs_validate_sb_write(
233 struct xfs_mount *mp,
234 struct xfs_buf *bp,
235 struct xfs_sb *sbp)
236 {
237 /*
238 * Carry out additional sb summary counter sanity checks when we write
239 * the superblock. We skip this in the read validator because there
240 * could be newer superblocks in the log and if the values are garbage
241 * even after replay we'll recalculate them at the end of log mount.
242 *
243 * mkfs has traditionally written zeroed counters to inprogress and
244 * secondary superblocks, so allow this usage to continue because
245 * we never read counters from such superblocks.
246 */
247 if (xfs_buf_daddr(bp) == XFS_SB_DADDR && !sbp->sb_inprogress &&
248 (sbp->sb_fdblocks > sbp->sb_dblocks ||
249 !xfs_verify_icount(mp, sbp->sb_icount) ||
250 sbp->sb_ifree > sbp->sb_icount)) {
251 xfs_warn(mp, "SB summary counter sanity check failed");
252 return -EFSCORRUPTED;
253 }
254
255 if (!xfs_sb_is_v5(sbp))
256 return 0;
257
258 /*
259 * Version 5 superblock feature mask validation. Reject combinations
260 * the kernel cannot support since we checked for unsupported bits in
261 * the read verifier, which means that memory is corrupt.
262 */
263 if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
264 xfs_warn(mp,
265 "Corruption detected in superblock compatible features (0x%x)!",
266 (sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
267 return -EFSCORRUPTED;
268 }
269
270 if (!xfs_is_readonly(mp) &&
271 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
272 xfs_alert(mp,
273 "Corruption detected in superblock read-only compatible features (0x%x)!",
274 (sbp->sb_features_ro_compat &
275 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
276 return -EFSCORRUPTED;
277 }
278 if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
279 xfs_warn(mp,
280 "Corruption detected in superblock incompatible features (0x%x)!",
281 (sbp->sb_features_incompat &
282 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
283 return -EFSCORRUPTED;
284 }
285 if (xfs_sb_has_incompat_log_feature(sbp,
286 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
287 xfs_warn(mp,
288 "Corruption detected in superblock incompatible log features (0x%x)!",
289 (sbp->sb_features_log_incompat &
290 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
291 return -EFSCORRUPTED;
292 }
293
294 /*
295 * We can't read verify the sb LSN because the read verifier is called
296 * before the log is allocated and processed. We know the log is set up
297 * before write verifier calls, so check it here.
298 */
299 if (!xfs_log_check_lsn(mp, sbp->sb_lsn))
300 return -EFSCORRUPTED;
301
302 return 0;
303 }
304
305 /* Check the validity of the SB. */
306 STATIC int
xfs_validate_sb_common(struct xfs_mount * mp,struct xfs_buf * bp,struct xfs_sb * sbp)307 xfs_validate_sb_common(
308 struct xfs_mount *mp,
309 struct xfs_buf *bp,
310 struct xfs_sb *sbp)
311 {
312 struct xfs_dsb *dsb = bp->b_addr;
313 uint32_t agcount = 0;
314 uint32_t rem;
315 bool has_dalign;
316
317 if (!xfs_verify_magic(bp, dsb->sb_magicnum)) {
318 xfs_warn(mp,
319 "Superblock has bad magic number 0x%x. Not an XFS filesystem?",
320 be32_to_cpu(dsb->sb_magicnum));
321 return -EWRONGFS;
322 }
323
324 if (!xfs_sb_good_version(sbp)) {
325 xfs_warn(mp,
326 "Superblock has unknown features enabled or corrupted feature masks.");
327 return -EWRONGFS;
328 }
329
330 /*
331 * Validate feature flags and state
332 */
333 if (xfs_sb_is_v5(sbp)) {
334 if (sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) {
335 xfs_notice(mp,
336 "Block size (%u bytes) too small for Version 5 superblock (minimum %d bytes)",
337 sbp->sb_blocksize, XFS_MIN_CRC_BLOCKSIZE);
338 return -EFSCORRUPTED;
339 }
340
341 /* V5 has a separate project quota inode */
342 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
343 xfs_notice(mp,
344 "Version 5 of Super block has XFS_OQUOTA bits.");
345 return -EFSCORRUPTED;
346 }
347
348 /*
349 * Full inode chunks must be aligned to inode chunk size when
350 * sparse inodes are enabled to support the sparse chunk
351 * allocation algorithm and prevent overlapping inode records.
352 */
353 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES) {
354 uint32_t align;
355
356 align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
357 >> sbp->sb_blocklog;
358 if (sbp->sb_inoalignmt != align) {
359 xfs_warn(mp,
360 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
361 sbp->sb_inoalignmt, align);
362 return -EINVAL;
363 }
364 }
365 } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
366 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
367 xfs_notice(mp,
368 "Superblock earlier than Version 5 has XFS_{P|G}QUOTA_{ENFD|CHKD} bits.");
369 return -EFSCORRUPTED;
370 }
371
372 if (unlikely(
373 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
374 xfs_warn(mp,
375 "filesystem is marked as having an external log; "
376 "specify logdev on the mount command line.");
377 return -EINVAL;
378 }
379
380 if (unlikely(
381 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
382 xfs_warn(mp,
383 "filesystem is marked as having an internal log; "
384 "do not specify logdev on the mount command line.");
385 return -EINVAL;
386 }
387
388 /* Compute agcount for this number of dblocks and agblocks */
389 if (sbp->sb_agblocks) {
390 agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem);
391 if (rem)
392 agcount++;
393 }
394
395 /*
396 * More sanity checking. Most of these were stolen directly from
397 * xfs_repair.
398 */
399 if (unlikely(
400 sbp->sb_agcount <= 0 ||
401 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
402 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
403 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
404 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
405 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
406 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
407 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
408 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
409 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
410 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
411 sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
412 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
413 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
414 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
415 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
416 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
417 sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
418 XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES ||
419 XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES ||
420 sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1 ||
421 agcount == 0 || agcount != sbp->sb_agcount ||
422 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
423 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
424 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
425 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) ||
426 sbp->sb_dblocks == 0 ||
427 sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) ||
428 sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp) ||
429 sbp->sb_shared_vn != 0)) {
430 xfs_notice(mp, "SB sanity check failed");
431 return -EFSCORRUPTED;
432 }
433
434 /*
435 * Logs that are too large are not supported at all. Reject them
436 * outright. Logs that are too small are tolerated on v4 filesystems,
437 * but we can only check that when mounting the log. Hence we skip
438 * those checks here.
439 */
440 if (sbp->sb_logblocks > XFS_MAX_LOG_BLOCKS) {
441 xfs_notice(mp,
442 "Log size 0x%x blocks too large, maximum size is 0x%llx blocks",
443 sbp->sb_logblocks, XFS_MAX_LOG_BLOCKS);
444 return -EFSCORRUPTED;
445 }
446
447 if (XFS_FSB_TO_B(mp, sbp->sb_logblocks) > XFS_MAX_LOG_BYTES) {
448 xfs_warn(mp,
449 "log size 0x%llx bytes too large, maximum size is 0x%llx bytes",
450 XFS_FSB_TO_B(mp, sbp->sb_logblocks),
451 XFS_MAX_LOG_BYTES);
452 return -EFSCORRUPTED;
453 }
454
455 /*
456 * Do not allow filesystems with corrupted log sector or stripe units to
457 * be mounted. We cannot safely size the iclogs or write to the log if
458 * the log stripe unit is not valid.
459 */
460 if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT) {
461 if (sbp->sb_logsectsize != (1U << sbp->sb_logsectlog)) {
462 xfs_notice(mp,
463 "log sector size in bytes/log2 (0x%x/0x%x) must match",
464 sbp->sb_logsectsize, 1U << sbp->sb_logsectlog);
465 return -EFSCORRUPTED;
466 }
467 } else if (sbp->sb_logsectsize || sbp->sb_logsectlog) {
468 xfs_notice(mp,
469 "log sector size in bytes/log2 (0x%x/0x%x) are not zero",
470 sbp->sb_logsectsize, sbp->sb_logsectlog);
471 return -EFSCORRUPTED;
472 }
473
474 if (sbp->sb_logsunit > 1) {
475 if (sbp->sb_logsunit % sbp->sb_blocksize) {
476 xfs_notice(mp,
477 "log stripe unit 0x%x bytes must be a multiple of block size",
478 sbp->sb_logsunit);
479 return -EFSCORRUPTED;
480 }
481 if (sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE) {
482 xfs_notice(mp,
483 "log stripe unit 0x%x bytes over maximum size (0x%x bytes)",
484 sbp->sb_logsunit, XLOG_MAX_RECORD_BSIZE);
485 return -EFSCORRUPTED;
486 }
487 }
488
489 /* Validate the realtime geometry; stolen from xfs_repair */
490 if (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE ||
491 sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) {
492 xfs_notice(mp,
493 "realtime extent sanity check failed");
494 return -EFSCORRUPTED;
495 }
496
497 if (sbp->sb_rblocks == 0) {
498 if (sbp->sb_rextents != 0 || sbp->sb_rbmblocks != 0 ||
499 sbp->sb_rextslog != 0 || sbp->sb_frextents != 0) {
500 xfs_notice(mp,
501 "realtime zeroed geometry check failed");
502 return -EFSCORRUPTED;
503 }
504 } else {
505 uint64_t rexts;
506 uint64_t rbmblocks;
507
508 rexts = div_u64(sbp->sb_rblocks, sbp->sb_rextsize);
509 rbmblocks = howmany_64(sbp->sb_rextents,
510 NBBY * sbp->sb_blocksize);
511
512 if (!xfs_validate_rtextents(rexts) ||
513 sbp->sb_rextents != rexts ||
514 sbp->sb_rextslog != xfs_compute_rextslog(rexts) ||
515 sbp->sb_rbmblocks != rbmblocks) {
516 xfs_notice(mp,
517 "realtime geometry sanity check failed");
518 return -EFSCORRUPTED;
519 }
520 }
521
522 /*
523 * Either (sb_unit and !hasdalign) or (!sb_unit and hasdalign)
524 * would imply the image is corrupted.
525 */
526 has_dalign = sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT;
527 if (!!sbp->sb_unit ^ has_dalign) {
528 xfs_notice(mp, "SB stripe alignment sanity check failed");
529 return -EFSCORRUPTED;
530 }
531
532 if (!xfs_validate_stripe_geometry(mp, XFS_FSB_TO_B(mp, sbp->sb_unit),
533 XFS_FSB_TO_B(mp, sbp->sb_width), 0,
534 xfs_buf_daddr(bp) == XFS_SB_DADDR, false))
535 return -EFSCORRUPTED;
536
537 /*
538 * Currently only very few inode sizes are supported.
539 */
540 switch (sbp->sb_inodesize) {
541 case 256:
542 case 512:
543 case 1024:
544 case 2048:
545 break;
546 default:
547 xfs_warn(mp, "inode size of %d bytes not supported",
548 sbp->sb_inodesize);
549 return -ENOSYS;
550 }
551
552 return 0;
553 }
554
555 void
xfs_sb_quota_from_disk(struct xfs_sb * sbp)556 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
557 {
558 /*
559 * older mkfs doesn't initialize quota inodes to NULLFSINO. This
560 * leads to in-core values having two different values for a quota
561 * inode to be invalid: 0 and NULLFSINO. Change it to a single value
562 * NULLFSINO.
563 *
564 * Note that this change affect only the in-core values. These
565 * values are not written back to disk unless any quota information
566 * is written to the disk. Even in that case, sb_pquotino field is
567 * not written to disk unless the superblock supports pquotino.
568 */
569 if (sbp->sb_uquotino == 0)
570 sbp->sb_uquotino = NULLFSINO;
571 if (sbp->sb_gquotino == 0)
572 sbp->sb_gquotino = NULLFSINO;
573 if (sbp->sb_pquotino == 0)
574 sbp->sb_pquotino = NULLFSINO;
575
576 /*
577 * We need to do these manipilations only if we are working
578 * with an older version of on-disk superblock.
579 */
580 if (xfs_sb_is_v5(sbp))
581 return;
582
583 if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
584 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
585 XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
586 if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
587 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
588 XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
589 sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
590
591 if (sbp->sb_qflags & XFS_PQUOTA_ACCT &&
592 sbp->sb_gquotino != NULLFSINO) {
593 /*
594 * In older version of superblock, on-disk superblock only
595 * has sb_gquotino, and in-core superblock has both sb_gquotino
596 * and sb_pquotino. But, only one of them is supported at any
597 * point of time. So, if PQUOTA is set in disk superblock,
598 * copy over sb_gquotino to sb_pquotino. The NULLFSINO test
599 * above is to make sure we don't do this twice and wipe them
600 * both out!
601 */
602 sbp->sb_pquotino = sbp->sb_gquotino;
603 sbp->sb_gquotino = NULLFSINO;
604 }
605 }
606
607 static void
__xfs_sb_from_disk(struct xfs_sb * to,struct xfs_dsb * from,bool convert_xquota)608 __xfs_sb_from_disk(
609 struct xfs_sb *to,
610 struct xfs_dsb *from,
611 bool convert_xquota)
612 {
613 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
614 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
615 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
616 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
617 to->sb_rextents = be64_to_cpu(from->sb_rextents);
618 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
619 to->sb_logstart = be64_to_cpu(from->sb_logstart);
620 to->sb_rootino = be64_to_cpu(from->sb_rootino);
621 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
622 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
623 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
624 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
625 to->sb_agcount = be32_to_cpu(from->sb_agcount);
626 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
627 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
628 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
629 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
630 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
631 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
632 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
633 to->sb_blocklog = from->sb_blocklog;
634 to->sb_sectlog = from->sb_sectlog;
635 to->sb_inodelog = from->sb_inodelog;
636 to->sb_inopblog = from->sb_inopblog;
637 to->sb_agblklog = from->sb_agblklog;
638 to->sb_rextslog = from->sb_rextslog;
639 to->sb_inprogress = from->sb_inprogress;
640 to->sb_imax_pct = from->sb_imax_pct;
641 to->sb_icount = be64_to_cpu(from->sb_icount);
642 to->sb_ifree = be64_to_cpu(from->sb_ifree);
643 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
644 to->sb_frextents = be64_to_cpu(from->sb_frextents);
645 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
646 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
647 to->sb_qflags = be16_to_cpu(from->sb_qflags);
648 to->sb_flags = from->sb_flags;
649 to->sb_shared_vn = from->sb_shared_vn;
650 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
651 to->sb_unit = be32_to_cpu(from->sb_unit);
652 to->sb_width = be32_to_cpu(from->sb_width);
653 to->sb_dirblklog = from->sb_dirblklog;
654 to->sb_logsectlog = from->sb_logsectlog;
655 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
656 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
657 to->sb_features2 = be32_to_cpu(from->sb_features2);
658 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
659 to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
660 to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
661 to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
662 to->sb_features_log_incompat =
663 be32_to_cpu(from->sb_features_log_incompat);
664 /* crc is only used on disk, not in memory; just init to 0 here. */
665 to->sb_crc = 0;
666 to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
667 to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
668 to->sb_lsn = be64_to_cpu(from->sb_lsn);
669 /*
670 * sb_meta_uuid is only on disk if it differs from sb_uuid and the
671 * feature flag is set; if not set we keep it only in memory.
672 */
673 if (xfs_sb_is_v5(to) &&
674 (to->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID))
675 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
676 else
677 uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
678 /* Convert on-disk flags to in-memory flags? */
679 if (convert_xquota)
680 xfs_sb_quota_from_disk(to);
681 }
682
683 void
xfs_sb_from_disk(struct xfs_sb * to,struct xfs_dsb * from)684 xfs_sb_from_disk(
685 struct xfs_sb *to,
686 struct xfs_dsb *from)
687 {
688 __xfs_sb_from_disk(to, from, true);
689 }
690
691 static void
xfs_sb_quota_to_disk(struct xfs_dsb * to,struct xfs_sb * from)692 xfs_sb_quota_to_disk(
693 struct xfs_dsb *to,
694 struct xfs_sb *from)
695 {
696 uint16_t qflags = from->sb_qflags;
697
698 to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
699
700 /*
701 * The in-memory superblock quota state matches the v5 on-disk format so
702 * just write them out and return
703 */
704 if (xfs_sb_is_v5(from)) {
705 to->sb_qflags = cpu_to_be16(from->sb_qflags);
706 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
707 to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
708 return;
709 }
710
711 /*
712 * For older superblocks (v4), the in-core version of sb_qflags do not
713 * have XFS_OQUOTA_* flags, whereas the on-disk version does. So,
714 * convert incore XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
715 */
716 qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
717 XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
718
719 if (from->sb_qflags &
720 (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
721 qflags |= XFS_OQUOTA_ENFD;
722 if (from->sb_qflags &
723 (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
724 qflags |= XFS_OQUOTA_CHKD;
725 to->sb_qflags = cpu_to_be16(qflags);
726
727 /*
728 * GQUOTINO and PQUOTINO cannot be used together in versions
729 * of superblock that do not have pquotino. from->sb_flags
730 * tells us which quota is active and should be copied to
731 * disk. If neither are active, we should NULL the inode.
732 *
733 * In all cases, the separate pquotino must remain 0 because it
734 * is beyond the "end" of the valid non-pquotino superblock.
735 */
736 if (from->sb_qflags & XFS_GQUOTA_ACCT)
737 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
738 else if (from->sb_qflags & XFS_PQUOTA_ACCT)
739 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
740 else {
741 /*
742 * We can't rely on just the fields being logged to tell us
743 * that it is safe to write NULLFSINO - we should only do that
744 * if quotas are not actually enabled. Hence only write
745 * NULLFSINO if both in-core quota inodes are NULL.
746 */
747 if (from->sb_gquotino == NULLFSINO &&
748 from->sb_pquotino == NULLFSINO)
749 to->sb_gquotino = cpu_to_be64(NULLFSINO);
750 }
751
752 to->sb_pquotino = 0;
753 }
754
755 void
xfs_sb_to_disk(struct xfs_dsb * to,struct xfs_sb * from)756 xfs_sb_to_disk(
757 struct xfs_dsb *to,
758 struct xfs_sb *from)
759 {
760 xfs_sb_quota_to_disk(to, from);
761
762 to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
763 to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
764 to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
765 to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
766 to->sb_rextents = cpu_to_be64(from->sb_rextents);
767 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
768 to->sb_logstart = cpu_to_be64(from->sb_logstart);
769 to->sb_rootino = cpu_to_be64(from->sb_rootino);
770 to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
771 to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
772 to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
773 to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
774 to->sb_agcount = cpu_to_be32(from->sb_agcount);
775 to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
776 to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
777 to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
778 to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
779 to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
780 to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
781 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
782 to->sb_blocklog = from->sb_blocklog;
783 to->sb_sectlog = from->sb_sectlog;
784 to->sb_inodelog = from->sb_inodelog;
785 to->sb_inopblog = from->sb_inopblog;
786 to->sb_agblklog = from->sb_agblklog;
787 to->sb_rextslog = from->sb_rextslog;
788 to->sb_inprogress = from->sb_inprogress;
789 to->sb_imax_pct = from->sb_imax_pct;
790 to->sb_icount = cpu_to_be64(from->sb_icount);
791 to->sb_ifree = cpu_to_be64(from->sb_ifree);
792 to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
793 to->sb_frextents = cpu_to_be64(from->sb_frextents);
794
795 to->sb_flags = from->sb_flags;
796 to->sb_shared_vn = from->sb_shared_vn;
797 to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
798 to->sb_unit = cpu_to_be32(from->sb_unit);
799 to->sb_width = cpu_to_be32(from->sb_width);
800 to->sb_dirblklog = from->sb_dirblklog;
801 to->sb_logsectlog = from->sb_logsectlog;
802 to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
803 to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
804
805 /*
806 * We need to ensure that bad_features2 always matches features2.
807 * Hence we enforce that here rather than having to remember to do it
808 * everywhere else that updates features2.
809 */
810 from->sb_bad_features2 = from->sb_features2;
811 to->sb_features2 = cpu_to_be32(from->sb_features2);
812 to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
813
814 if (!xfs_sb_is_v5(from))
815 return;
816
817 to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
818 to->sb_features_ro_compat =
819 cpu_to_be32(from->sb_features_ro_compat);
820 to->sb_features_incompat =
821 cpu_to_be32(from->sb_features_incompat);
822 to->sb_features_log_incompat =
823 cpu_to_be32(from->sb_features_log_incompat);
824 to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
825 to->sb_lsn = cpu_to_be64(from->sb_lsn);
826 if (from->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
827 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
828 }
829
830 /*
831 * If the superblock has the CRC feature bit set or the CRC field is non-null,
832 * check that the CRC is valid. We check the CRC field is non-null because a
833 * single bit error could clear the feature bit and unused parts of the
834 * superblock are supposed to be zero. Hence a non-null crc field indicates that
835 * we've potentially lost a feature bit and we should check it anyway.
836 *
837 * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
838 * last field in V4 secondary superblocks. So for secondary superblocks,
839 * we are more forgiving, and ignore CRC failures if the primary doesn't
840 * indicate that the fs version is V5.
841 */
842 static void
xfs_sb_read_verify(struct xfs_buf * bp)843 xfs_sb_read_verify(
844 struct xfs_buf *bp)
845 {
846 struct xfs_sb sb;
847 struct xfs_mount *mp = bp->b_mount;
848 struct xfs_dsb *dsb = bp->b_addr;
849 int error;
850
851 /*
852 * open code the version check to avoid needing to convert the entire
853 * superblock from disk order just to check the version number
854 */
855 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
856 (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
857 XFS_SB_VERSION_5) ||
858 dsb->sb_crc != 0)) {
859
860 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
861 /* Only fail bad secondaries on a known V5 filesystem */
862 if (xfs_buf_daddr(bp) == XFS_SB_DADDR ||
863 xfs_has_crc(mp)) {
864 error = -EFSBADCRC;
865 goto out_error;
866 }
867 }
868 }
869
870 /*
871 * Check all the superblock fields. Don't byteswap the xquota flags
872 * because _verify_common checks the on-disk values.
873 */
874 __xfs_sb_from_disk(&sb, dsb, false);
875 error = xfs_validate_sb_common(mp, bp, &sb);
876 if (error)
877 goto out_error;
878 error = xfs_validate_sb_read(mp, &sb);
879
880 out_error:
881 if (error == -EFSCORRUPTED || error == -EFSBADCRC)
882 xfs_verifier_error(bp, error, __this_address);
883 else if (error)
884 xfs_buf_ioerror(bp, error);
885 }
886
887 /*
888 * We may be probed for a filesystem match, so we may not want to emit
889 * messages when the superblock buffer is not actually an XFS superblock.
890 * If we find an XFS superblock, then run a normal, noisy mount because we are
891 * really going to mount it and want to know about errors.
892 */
893 static void
xfs_sb_quiet_read_verify(struct xfs_buf * bp)894 xfs_sb_quiet_read_verify(
895 struct xfs_buf *bp)
896 {
897 struct xfs_dsb *dsb = bp->b_addr;
898
899 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
900 /* XFS filesystem, verify noisily! */
901 xfs_sb_read_verify(bp);
902 return;
903 }
904 /* quietly fail */
905 xfs_buf_ioerror(bp, -EWRONGFS);
906 }
907
908 static void
xfs_sb_write_verify(struct xfs_buf * bp)909 xfs_sb_write_verify(
910 struct xfs_buf *bp)
911 {
912 struct xfs_sb sb;
913 struct xfs_mount *mp = bp->b_mount;
914 struct xfs_buf_log_item *bip = bp->b_log_item;
915 struct xfs_dsb *dsb = bp->b_addr;
916 int error;
917
918 /*
919 * Check all the superblock fields. Don't byteswap the xquota flags
920 * because _verify_common checks the on-disk values.
921 */
922 __xfs_sb_from_disk(&sb, dsb, false);
923 error = xfs_validate_sb_common(mp, bp, &sb);
924 if (error)
925 goto out_error;
926 error = xfs_validate_sb_write(mp, bp, &sb);
927 if (error)
928 goto out_error;
929
930 if (!xfs_sb_is_v5(&sb))
931 return;
932
933 if (bip)
934 dsb->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
935
936 xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
937 return;
938
939 out_error:
940 xfs_verifier_error(bp, error, __this_address);
941 }
942
943 const struct xfs_buf_ops xfs_sb_buf_ops = {
944 .name = "xfs_sb",
945 .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
946 .verify_read = xfs_sb_read_verify,
947 .verify_write = xfs_sb_write_verify,
948 };
949
950 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
951 .name = "xfs_sb_quiet",
952 .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
953 .verify_read = xfs_sb_quiet_read_verify,
954 .verify_write = xfs_sb_write_verify,
955 };
956
957 /*
958 * xfs_mount_common
959 *
960 * Mount initialization code establishing various mount
961 * fields from the superblock associated with the given
962 * mount structure.
963 *
964 * Inode geometry are calculated in xfs_ialloc_setup_geometry.
965 */
966 void
xfs_sb_mount_common(struct xfs_mount * mp,struct xfs_sb * sbp)967 xfs_sb_mount_common(
968 struct xfs_mount *mp,
969 struct xfs_sb *sbp)
970 {
971 mp->m_agfrotor = 0;
972 atomic_set(&mp->m_agirotor, 0);
973 mp->m_maxagi = mp->m_sb.sb_agcount;
974 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
975 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
976 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
977 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
978 mp->m_blockmask = sbp->sb_blocksize - 1;
979 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
980 mp->m_blockwmask = mp->m_blockwsize - 1;
981
982 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
983 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
984 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
985 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
986
987 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
988 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
989 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
990 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
991
992 mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 1);
993 mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 0);
994 mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
995 mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
996
997 mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, true);
998 mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, false);
999 mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
1000 mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
1001
1002 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
1003 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
1004 mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp);
1005 }
1006
1007 /*
1008 * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
1009 * into the superblock buffer to be logged. It does not provide the higher
1010 * level of locking that is needed to protect the in-core superblock from
1011 * concurrent access.
1012 */
1013 void
xfs_log_sb(struct xfs_trans * tp)1014 xfs_log_sb(
1015 struct xfs_trans *tp)
1016 {
1017 struct xfs_mount *mp = tp->t_mountp;
1018 struct xfs_buf *bp = xfs_trans_getsb(tp);
1019
1020 /*
1021 * Lazy sb counters don't update the in-core superblock so do that now.
1022 * If this is at unmount, the counters will be exactly correct, but at
1023 * any other time they will only be ballpark correct because of
1024 * reservations that have been taken out percpu counters. If we have an
1025 * unclean shutdown, this will be corrected by log recovery rebuilding
1026 * the counters from the AGF block counts.
1027 *
1028 * Do not update sb_frextents here because it is not part of the lazy
1029 * sb counters, despite having a percpu counter. It is always kept
1030 * consistent with the ondisk rtbitmap by xfs_trans_apply_sb_deltas()
1031 * and hence we don't need have to update it here.
1032 */
1033 if (xfs_has_lazysbcount(mp)) {
1034 mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
1035 mp->m_sb.sb_ifree = min_t(uint64_t,
1036 percpu_counter_sum(&mp->m_ifree),
1037 mp->m_sb.sb_icount);
1038 mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1039 }
1040
1041 xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1042 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
1043 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
1044 }
1045
1046 /*
1047 * xfs_sync_sb
1048 *
1049 * Sync the superblock to disk.
1050 *
1051 * Note that the caller is responsible for checking the frozen state of the
1052 * filesystem. This procedure uses the non-blocking transaction allocator and
1053 * thus will allow modifications to a frozen fs. This is required because this
1054 * code can be called during the process of freezing where use of the high-level
1055 * allocator would deadlock.
1056 */
1057 int
xfs_sync_sb(struct xfs_mount * mp,bool wait)1058 xfs_sync_sb(
1059 struct xfs_mount *mp,
1060 bool wait)
1061 {
1062 struct xfs_trans *tp;
1063 int error;
1064
1065 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0,
1066 XFS_TRANS_NO_WRITECOUNT, &tp);
1067 if (error)
1068 return error;
1069
1070 xfs_log_sb(tp);
1071 if (wait)
1072 xfs_trans_set_sync(tp);
1073 return xfs_trans_commit(tp);
1074 }
1075
1076 /*
1077 * Update all the secondary superblocks to match the new state of the primary.
1078 * Because we are completely overwriting all the existing fields in the
1079 * secondary superblock buffers, there is no need to read them in from disk.
1080 * Just get a new buffer, stamp it and write it.
1081 *
1082 * The sb buffers need to be cached here so that we serialise against other
1083 * operations that access the secondary superblocks, but we don't want to keep
1084 * them in memory once it is written so we mark it as a one-shot buffer.
1085 */
1086 int
xfs_update_secondary_sbs(struct xfs_mount * mp)1087 xfs_update_secondary_sbs(
1088 struct xfs_mount *mp)
1089 {
1090 struct xfs_perag *pag;
1091 xfs_agnumber_t agno = 1;
1092 int saved_error = 0;
1093 int error = 0;
1094 LIST_HEAD (buffer_list);
1095
1096 /* update secondary superblocks. */
1097 for_each_perag_from(mp, agno, pag) {
1098 struct xfs_buf *bp;
1099
1100 error = xfs_buf_get(mp->m_ddev_targp,
1101 XFS_AG_DADDR(mp, pag->pag_agno, XFS_SB_DADDR),
1102 XFS_FSS_TO_BB(mp, 1), &bp);
1103 /*
1104 * If we get an error reading or writing alternate superblocks,
1105 * continue. xfs_repair chooses the "best" superblock based
1106 * on most matches; if we break early, we'll leave more
1107 * superblocks un-updated than updated, and xfs_repair may
1108 * pick them over the properly-updated primary.
1109 */
1110 if (error) {
1111 xfs_warn(mp,
1112 "error allocating secondary superblock for ag %d",
1113 pag->pag_agno);
1114 if (!saved_error)
1115 saved_error = error;
1116 continue;
1117 }
1118
1119 bp->b_ops = &xfs_sb_buf_ops;
1120 xfs_buf_oneshot(bp);
1121 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
1122 xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1123 xfs_buf_delwri_queue(bp, &buffer_list);
1124 xfs_buf_relse(bp);
1125
1126 /* don't hold too many buffers at once */
1127 if (agno % 16)
1128 continue;
1129
1130 error = xfs_buf_delwri_submit(&buffer_list);
1131 if (error) {
1132 xfs_warn(mp,
1133 "write error %d updating a secondary superblock near ag %d",
1134 error, pag->pag_agno);
1135 if (!saved_error)
1136 saved_error = error;
1137 continue;
1138 }
1139 }
1140 error = xfs_buf_delwri_submit(&buffer_list);
1141 if (error) {
1142 xfs_warn(mp,
1143 "write error %d updating a secondary superblock near ag %d",
1144 error, agno);
1145 }
1146
1147 return saved_error ? saved_error : error;
1148 }
1149
1150 /*
1151 * Same behavior as xfs_sync_sb, except that it is always synchronous and it
1152 * also writes the superblock buffer to disk sector 0 immediately.
1153 */
1154 int
xfs_sync_sb_buf(struct xfs_mount * mp)1155 xfs_sync_sb_buf(
1156 struct xfs_mount *mp)
1157 {
1158 struct xfs_trans *tp;
1159 struct xfs_buf *bp;
1160 int error;
1161
1162 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp);
1163 if (error)
1164 return error;
1165
1166 bp = xfs_trans_getsb(tp);
1167 xfs_log_sb(tp);
1168 xfs_trans_bhold(tp, bp);
1169 xfs_trans_set_sync(tp);
1170 error = xfs_trans_commit(tp);
1171 if (error)
1172 goto out;
1173 /*
1174 * write out the sb buffer to get the changes to disk
1175 */
1176 error = xfs_bwrite(bp);
1177 out:
1178 xfs_buf_relse(bp);
1179 return error;
1180 }
1181
1182 void
xfs_fs_geometry(struct xfs_mount * mp,struct xfs_fsop_geom * geo,int struct_version)1183 xfs_fs_geometry(
1184 struct xfs_mount *mp,
1185 struct xfs_fsop_geom *geo,
1186 int struct_version)
1187 {
1188 struct xfs_sb *sbp = &mp->m_sb;
1189
1190 memset(geo, 0, sizeof(struct xfs_fsop_geom));
1191
1192 geo->blocksize = sbp->sb_blocksize;
1193 geo->rtextsize = sbp->sb_rextsize;
1194 geo->agblocks = sbp->sb_agblocks;
1195 geo->agcount = sbp->sb_agcount;
1196 geo->logblocks = sbp->sb_logblocks;
1197 geo->sectsize = sbp->sb_sectsize;
1198 geo->inodesize = sbp->sb_inodesize;
1199 geo->imaxpct = sbp->sb_imax_pct;
1200 geo->datablocks = sbp->sb_dblocks;
1201 geo->rtblocks = sbp->sb_rblocks;
1202 geo->rtextents = sbp->sb_rextents;
1203 geo->logstart = sbp->sb_logstart;
1204 BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid));
1205 memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid));
1206
1207 if (struct_version < 2)
1208 return;
1209
1210 geo->sunit = sbp->sb_unit;
1211 geo->swidth = sbp->sb_width;
1212
1213 if (struct_version < 3)
1214 return;
1215
1216 geo->version = XFS_FSOP_GEOM_VERSION;
1217 geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
1218 XFS_FSOP_GEOM_FLAGS_DIRV2 |
1219 XFS_FSOP_GEOM_FLAGS_EXTFLG;
1220 if (xfs_has_attr(mp))
1221 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR;
1222 if (xfs_has_quota(mp))
1223 geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA;
1224 if (xfs_has_align(mp))
1225 geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN;
1226 if (xfs_has_dalign(mp))
1227 geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN;
1228 if (xfs_has_asciici(mp))
1229 geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI;
1230 if (xfs_has_lazysbcount(mp))
1231 geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB;
1232 if (xfs_has_attr2(mp))
1233 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2;
1234 if (xfs_has_projid32(mp))
1235 geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32;
1236 if (xfs_has_crc(mp))
1237 geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB;
1238 if (xfs_has_ftype(mp))
1239 geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE;
1240 if (xfs_has_finobt(mp))
1241 geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT;
1242 if (xfs_has_sparseinodes(mp))
1243 geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES;
1244 if (xfs_has_rmapbt(mp))
1245 geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT;
1246 if (xfs_has_reflink(mp))
1247 geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK;
1248 if (xfs_has_bigtime(mp))
1249 geo->flags |= XFS_FSOP_GEOM_FLAGS_BIGTIME;
1250 if (xfs_has_inobtcounts(mp))
1251 geo->flags |= XFS_FSOP_GEOM_FLAGS_INOBTCNT;
1252 if (xfs_has_sector(mp)) {
1253 geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR;
1254 geo->logsectsize = sbp->sb_logsectsize;
1255 } else {
1256 geo->logsectsize = BBSIZE;
1257 }
1258 if (xfs_has_large_extent_counts(mp))
1259 geo->flags |= XFS_FSOP_GEOM_FLAGS_NREXT64;
1260 geo->rtsectsize = sbp->sb_blocksize;
1261 geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp);
1262
1263 if (struct_version < 4)
1264 return;
1265
1266 if (xfs_has_logv2(mp))
1267 geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2;
1268
1269 geo->logsunit = sbp->sb_logsunit;
1270
1271 if (struct_version < 5)
1272 return;
1273
1274 geo->version = XFS_FSOP_GEOM_VERSION_V5;
1275 }
1276
1277 /* Read a secondary superblock. */
1278 int
xfs_sb_read_secondary(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,struct xfs_buf ** bpp)1279 xfs_sb_read_secondary(
1280 struct xfs_mount *mp,
1281 struct xfs_trans *tp,
1282 xfs_agnumber_t agno,
1283 struct xfs_buf **bpp)
1284 {
1285 struct xfs_buf *bp;
1286 int error;
1287
1288 ASSERT(agno != 0 && agno != NULLAGNUMBER);
1289 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1290 XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1291 XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops);
1292 if (error)
1293 return error;
1294 xfs_buf_set_ref(bp, XFS_SSB_REF);
1295 *bpp = bp;
1296 return 0;
1297 }
1298
1299 /* Get an uninitialised secondary superblock buffer. */
1300 int
xfs_sb_get_secondary(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,struct xfs_buf ** bpp)1301 xfs_sb_get_secondary(
1302 struct xfs_mount *mp,
1303 struct xfs_trans *tp,
1304 xfs_agnumber_t agno,
1305 struct xfs_buf **bpp)
1306 {
1307 struct xfs_buf *bp;
1308 int error;
1309
1310 ASSERT(agno != 0 && agno != NULLAGNUMBER);
1311 error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1312 XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1313 XFS_FSS_TO_BB(mp, 1), 0, &bp);
1314 if (error)
1315 return error;
1316 bp->b_ops = &xfs_sb_buf_ops;
1317 xfs_buf_oneshot(bp);
1318 *bpp = bp;
1319 return 0;
1320 }
1321
1322 /*
1323 * sunit, swidth, sectorsize(optional with 0) should be all in bytes, so users
1324 * won't be confused by values in error messages. This function returns false
1325 * if the stripe geometry is invalid and the caller is unable to repair the
1326 * stripe configuration later in the mount process.
1327 */
1328 bool
xfs_validate_stripe_geometry(struct xfs_mount * mp,__s64 sunit,__s64 swidth,int sectorsize,bool may_repair,bool silent)1329 xfs_validate_stripe_geometry(
1330 struct xfs_mount *mp,
1331 __s64 sunit,
1332 __s64 swidth,
1333 int sectorsize,
1334 bool may_repair,
1335 bool silent)
1336 {
1337 if (swidth > INT_MAX) {
1338 if (!silent)
1339 xfs_notice(mp,
1340 "stripe width (%lld) is too large", swidth);
1341 goto check_override;
1342 }
1343
1344 if (sunit > swidth) {
1345 if (!silent)
1346 xfs_notice(mp,
1347 "stripe unit (%lld) is larger than the stripe width (%lld)", sunit, swidth);
1348 goto check_override;
1349 }
1350
1351 if (sectorsize && (int)sunit % sectorsize) {
1352 if (!silent)
1353 xfs_notice(mp,
1354 "stripe unit (%lld) must be a multiple of the sector size (%d)",
1355 sunit, sectorsize);
1356 goto check_override;
1357 }
1358
1359 if (sunit && !swidth) {
1360 if (!silent)
1361 xfs_notice(mp,
1362 "invalid stripe unit (%lld) and stripe width of 0", sunit);
1363 goto check_override;
1364 }
1365
1366 if (!sunit && swidth) {
1367 if (!silent)
1368 xfs_notice(mp,
1369 "invalid stripe width (%lld) and stripe unit of 0", swidth);
1370 goto check_override;
1371 }
1372
1373 if (sunit && (int)swidth % (int)sunit) {
1374 if (!silent)
1375 xfs_notice(mp,
1376 "stripe width (%lld) must be a multiple of the stripe unit (%lld)",
1377 swidth, sunit);
1378 goto check_override;
1379 }
1380 return true;
1381
1382 check_override:
1383 if (!may_repair)
1384 return false;
1385 /*
1386 * During mount, mp->m_dalign will not be set unless the sunit mount
1387 * option was set. If it was set, ignore the bad stripe alignment values
1388 * and allow the validation and overwrite later in the mount process to
1389 * attempt to overwrite the bad stripe alignment values with the values
1390 * supplied by mount options.
1391 */
1392 if (!mp->m_dalign)
1393 return false;
1394 if (!silent)
1395 xfs_notice(mp,
1396 "Will try to correct with specified mount options sunit (%d) and swidth (%d)",
1397 BBTOB(mp->m_dalign), BBTOB(mp->m_swidth));
1398 return true;
1399 }
1400
1401 /*
1402 * Compute the maximum level number of the realtime summary file, as defined by
1403 * mkfs. The historic use of highbit32 on a 64-bit quantity prohibited correct
1404 * use of rt volumes with more than 2^32 extents.
1405 */
1406 uint8_t
xfs_compute_rextslog(xfs_rtbxlen_t rtextents)1407 xfs_compute_rextslog(
1408 xfs_rtbxlen_t rtextents)
1409 {
1410 if (!rtextents)
1411 return 0;
1412 return xfs_highbit64(rtextents);
1413 }
1414