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 #include "xfs_exchrange.h"
30 #include "xfs_rtgroup.h"
31
32 /*
33 * Physical superblock buffer manipulations. Shared with libxfs in userspace.
34 */
35
36 /*
37 * Check that all the V4 feature bits that the V5 filesystem format requires are
38 * correctly set.
39 */
40 static bool
xfs_sb_validate_v5_features(struct xfs_sb * sbp)41 xfs_sb_validate_v5_features(
42 struct xfs_sb *sbp)
43 {
44 /* We must not have any unknown V4 feature bits set */
45 if (sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS)
46 return false;
47
48 /*
49 * The CRC bit is considered an invalid V4 flag, so we have to add it
50 * manually to the OKBITS mask.
51 */
52 if (sbp->sb_features2 & ~(XFS_SB_VERSION2_OKBITS |
53 XFS_SB_VERSION2_CRCBIT))
54 return false;
55
56 /* Now check all the required V4 feature flags are set. */
57
58 #define V5_VERS_FLAGS (XFS_SB_VERSION_NLINKBIT | \
59 XFS_SB_VERSION_ALIGNBIT | \
60 XFS_SB_VERSION_LOGV2BIT | \
61 XFS_SB_VERSION_EXTFLGBIT | \
62 XFS_SB_VERSION_DIRV2BIT | \
63 XFS_SB_VERSION_MOREBITSBIT)
64
65 #define V5_FEAT_FLAGS (XFS_SB_VERSION2_LAZYSBCOUNTBIT | \
66 XFS_SB_VERSION2_ATTR2BIT | \
67 XFS_SB_VERSION2_PROJID32BIT | \
68 XFS_SB_VERSION2_CRCBIT)
69
70 if ((sbp->sb_versionnum & V5_VERS_FLAGS) != V5_VERS_FLAGS)
71 return false;
72 if ((sbp->sb_features2 & V5_FEAT_FLAGS) != V5_FEAT_FLAGS)
73 return false;
74 return true;
75 }
76
77 /*
78 * We current support XFS v5 formats with known features and v4 superblocks with
79 * at least V2 directories.
80 */
81 bool
xfs_sb_good_version(struct xfs_sb * sbp)82 xfs_sb_good_version(
83 struct xfs_sb *sbp)
84 {
85 /*
86 * All v5 filesystems are supported, but we must check that all the
87 * required v4 feature flags are enabled correctly as the code checks
88 * those flags and not for v5 support.
89 */
90 if (xfs_sb_is_v5(sbp))
91 return xfs_sb_validate_v5_features(sbp);
92
93 /* versions prior to v4 are not supported */
94 if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_4)
95 return false;
96
97 /* We must not have any unknown v4 feature bits set */
98 if ((sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS) ||
99 ((sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) &&
100 (sbp->sb_features2 & ~XFS_SB_VERSION2_OKBITS)))
101 return false;
102
103 /* V4 filesystems need v2 directories and unwritten extents */
104 if (!(sbp->sb_versionnum & XFS_SB_VERSION_DIRV2BIT))
105 return false;
106 if (!(sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT))
107 return false;
108
109 /* It's a supported v4 filesystem */
110 return true;
111 }
112
113 uint64_t
xfs_sb_version_to_features(struct xfs_sb * sbp)114 xfs_sb_version_to_features(
115 struct xfs_sb *sbp)
116 {
117 uint64_t features = 0;
118
119 /* optional V4 features */
120 if (sbp->sb_rblocks > 0)
121 features |= XFS_FEAT_REALTIME;
122 if (sbp->sb_versionnum & XFS_SB_VERSION_NLINKBIT)
123 features |= XFS_FEAT_NLINK;
124 if (sbp->sb_versionnum & XFS_SB_VERSION_ATTRBIT)
125 features |= XFS_FEAT_ATTR;
126 if (sbp->sb_versionnum & XFS_SB_VERSION_QUOTABIT)
127 features |= XFS_FEAT_QUOTA;
128 if (sbp->sb_versionnum & XFS_SB_VERSION_ALIGNBIT)
129 features |= XFS_FEAT_ALIGN;
130 if (sbp->sb_versionnum & XFS_SB_VERSION_LOGV2BIT)
131 features |= XFS_FEAT_LOGV2;
132 if (sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT)
133 features |= XFS_FEAT_DALIGN;
134 if (sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT)
135 features |= XFS_FEAT_EXTFLG;
136 if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT)
137 features |= XFS_FEAT_SECTOR;
138 if (sbp->sb_versionnum & XFS_SB_VERSION_BORGBIT)
139 features |= XFS_FEAT_ASCIICI;
140 if (sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) {
141 if (sbp->sb_features2 & XFS_SB_VERSION2_LAZYSBCOUNTBIT)
142 features |= XFS_FEAT_LAZYSBCOUNT;
143 if (sbp->sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
144 features |= XFS_FEAT_ATTR2;
145 if (sbp->sb_features2 & XFS_SB_VERSION2_PROJID32BIT)
146 features |= XFS_FEAT_PROJID32;
147 if (sbp->sb_features2 & XFS_SB_VERSION2_FTYPE)
148 features |= XFS_FEAT_FTYPE;
149 }
150
151 if (!xfs_sb_is_v5(sbp))
152 return features;
153
154 /* Always on V5 features */
155 features |= XFS_FEAT_ALIGN | XFS_FEAT_LOGV2 | XFS_FEAT_EXTFLG |
156 XFS_FEAT_LAZYSBCOUNT | XFS_FEAT_ATTR2 | XFS_FEAT_PROJID32 |
157 XFS_FEAT_V3INODES | XFS_FEAT_CRC | XFS_FEAT_PQUOTINO;
158
159 /* Optional V5 features */
160 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_FINOBT)
161 features |= XFS_FEAT_FINOBT;
162 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_RMAPBT)
163 features |= XFS_FEAT_RMAPBT;
164 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_REFLINK)
165 features |= XFS_FEAT_REFLINK;
166 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_INOBTCNT)
167 features |= XFS_FEAT_INOBTCNT;
168 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_FTYPE)
169 features |= XFS_FEAT_FTYPE;
170 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES)
171 features |= XFS_FEAT_SPINODES;
172 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
173 features |= XFS_FEAT_META_UUID;
174 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_BIGTIME)
175 features |= XFS_FEAT_BIGTIME;
176 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR)
177 features |= XFS_FEAT_NEEDSREPAIR;
178 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NREXT64)
179 features |= XFS_FEAT_NREXT64;
180 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_EXCHRANGE)
181 features |= XFS_FEAT_EXCHANGE_RANGE;
182 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_PARENT)
183 features |= XFS_FEAT_PARENT;
184 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR)
185 features |= XFS_FEAT_METADIR;
186
187 return features;
188 }
189
190 /* Check all the superblock fields we care about when reading one in. */
191 STATIC int
xfs_validate_sb_read(struct xfs_mount * mp,struct xfs_sb * sbp)192 xfs_validate_sb_read(
193 struct xfs_mount *mp,
194 struct xfs_sb *sbp)
195 {
196 if (!xfs_sb_is_v5(sbp))
197 return 0;
198
199 /*
200 * Version 5 superblock feature mask validation. Reject combinations
201 * the kernel cannot support up front before checking anything else.
202 */
203 if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
204 xfs_warn(mp,
205 "Superblock has unknown compatible features (0x%x) enabled.",
206 (sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
207 xfs_warn(mp,
208 "Using a more recent kernel is recommended.");
209 }
210
211 if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
212 xfs_alert(mp,
213 "Superblock has unknown read-only compatible features (0x%x) enabled.",
214 (sbp->sb_features_ro_compat &
215 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
216 if (!xfs_is_readonly(mp)) {
217 xfs_warn(mp,
218 "Attempted to mount read-only compatible filesystem read-write.");
219 xfs_warn(mp,
220 "Filesystem can only be safely mounted read only.");
221
222 return -EINVAL;
223 }
224 }
225 if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
226 xfs_warn(mp,
227 "Superblock has unknown incompatible features (0x%x) enabled.",
228 (sbp->sb_features_incompat &
229 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
230 xfs_warn(mp,
231 "Filesystem cannot be safely mounted by this kernel.");
232 return -EINVAL;
233 }
234
235 return 0;
236 }
237
238 /* Return the number of extents covered by a single rt bitmap file */
239 static xfs_rtbxlen_t
xfs_extents_per_rbm(struct xfs_sb * sbp)240 xfs_extents_per_rbm(
241 struct xfs_sb *sbp)
242 {
243 if (xfs_sb_is_v5(sbp) &&
244 (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR))
245 return sbp->sb_rgextents;
246 return sbp->sb_rextents;
247 }
248
249 /*
250 * Return the payload size of a single rt bitmap block (without the metadata
251 * header if any).
252 */
253 static inline unsigned int
xfs_rtbmblock_size(struct xfs_sb * sbp)254 xfs_rtbmblock_size(
255 struct xfs_sb *sbp)
256 {
257 if (xfs_sb_is_v5(sbp) &&
258 (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR))
259 return sbp->sb_blocksize - sizeof(struct xfs_rtbuf_blkinfo);
260 return sbp->sb_blocksize;
261 }
262
263 static uint64_t
xfs_expected_rbmblocks(struct xfs_sb * sbp)264 xfs_expected_rbmblocks(
265 struct xfs_sb *sbp)
266 {
267 return howmany_64(xfs_extents_per_rbm(sbp),
268 NBBY * xfs_rtbmblock_size(sbp));
269 }
270
271 /* Validate the realtime geometry */
272 bool
xfs_validate_rt_geometry(struct xfs_sb * sbp)273 xfs_validate_rt_geometry(
274 struct xfs_sb *sbp)
275 {
276 if (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE ||
277 sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)
278 return false;
279
280 if (sbp->sb_rblocks == 0) {
281 if (sbp->sb_rextents != 0 || sbp->sb_rbmblocks != 0 ||
282 sbp->sb_rextslog != 0 || sbp->sb_frextents != 0)
283 return false;
284 return true;
285 }
286
287 if (sbp->sb_rextents == 0 ||
288 sbp->sb_rextents != div_u64(sbp->sb_rblocks, sbp->sb_rextsize) ||
289 sbp->sb_rextslog != xfs_compute_rextslog(sbp->sb_rextents) ||
290 sbp->sb_rbmblocks != xfs_expected_rbmblocks(sbp))
291 return false;
292
293 return true;
294 }
295
296 /* Check all the superblock fields we care about when writing one out. */
297 STATIC int
xfs_validate_sb_write(struct xfs_mount * mp,struct xfs_buf * bp,struct xfs_sb * sbp)298 xfs_validate_sb_write(
299 struct xfs_mount *mp,
300 struct xfs_buf *bp,
301 struct xfs_sb *sbp)
302 {
303 /*
304 * Carry out additional sb summary counter sanity checks when we write
305 * the superblock. We skip this in the read validator because there
306 * could be newer superblocks in the log and if the values are garbage
307 * even after replay we'll recalculate them at the end of log mount.
308 *
309 * mkfs has traditionally written zeroed counters to inprogress and
310 * secondary superblocks, so allow this usage to continue because
311 * we never read counters from such superblocks.
312 */
313 if (xfs_buf_daddr(bp) == XFS_SB_DADDR && !sbp->sb_inprogress &&
314 (sbp->sb_fdblocks > sbp->sb_dblocks ||
315 !xfs_verify_icount(mp, sbp->sb_icount) ||
316 sbp->sb_ifree > sbp->sb_icount)) {
317 xfs_warn(mp, "SB summary counter sanity check failed");
318 return -EFSCORRUPTED;
319 }
320
321 if (!xfs_sb_is_v5(sbp))
322 return 0;
323
324 /*
325 * Version 5 superblock feature mask validation. Reject combinations
326 * the kernel cannot support since we checked for unsupported bits in
327 * the read verifier, which means that memory is corrupt.
328 */
329 if (!xfs_is_readonly(mp) &&
330 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
331 xfs_alert(mp,
332 "Corruption detected in superblock read-only compatible features (0x%x)!",
333 (sbp->sb_features_ro_compat &
334 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
335 return -EFSCORRUPTED;
336 }
337 if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
338 xfs_warn(mp,
339 "Corruption detected in superblock incompatible features (0x%x)!",
340 (sbp->sb_features_incompat &
341 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
342 return -EFSCORRUPTED;
343 }
344 if (xfs_sb_has_incompat_log_feature(sbp,
345 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
346 xfs_warn(mp,
347 "Corruption detected in superblock incompatible log features (0x%x)!",
348 (sbp->sb_features_log_incompat &
349 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
350 return -EFSCORRUPTED;
351 }
352
353 /*
354 * We can't read verify the sb LSN because the read verifier is called
355 * before the log is allocated and processed. We know the log is set up
356 * before write verifier calls, so check it here.
357 */
358 if (!xfs_log_check_lsn(mp, sbp->sb_lsn))
359 return -EFSCORRUPTED;
360
361 return 0;
362 }
363
364 int
xfs_compute_rgblklog(xfs_rtxlen_t rgextents,xfs_rgblock_t rextsize)365 xfs_compute_rgblklog(
366 xfs_rtxlen_t rgextents,
367 xfs_rgblock_t rextsize)
368 {
369 uint64_t rgblocks = (uint64_t)rgextents * rextsize;
370
371 return xfs_highbit64(rgblocks - 1) + 1;
372 }
373
374 static int
xfs_validate_sb_rtgroups(struct xfs_mount * mp,struct xfs_sb * sbp)375 xfs_validate_sb_rtgroups(
376 struct xfs_mount *mp,
377 struct xfs_sb *sbp)
378 {
379 uint64_t groups;
380 int rgblklog;
381
382 if (sbp->sb_rextsize == 0) {
383 xfs_warn(mp,
384 "Realtime extent size must not be zero.");
385 return -EINVAL;
386 }
387
388 if (sbp->sb_rgextents > XFS_MAX_RGBLOCKS / sbp->sb_rextsize) {
389 xfs_warn(mp,
390 "Realtime group size (%u) must be less than %u rt extents.",
391 sbp->sb_rgextents,
392 XFS_MAX_RGBLOCKS / sbp->sb_rextsize);
393 return -EINVAL;
394 }
395
396 if (sbp->sb_rgextents < XFS_MIN_RGEXTENTS) {
397 xfs_warn(mp,
398 "Realtime group size (%u) must be at least %u rt extents.",
399 sbp->sb_rgextents, XFS_MIN_RGEXTENTS);
400 return -EINVAL;
401 }
402
403 if (sbp->sb_rgcount > XFS_MAX_RGNUMBER) {
404 xfs_warn(mp,
405 "Realtime groups (%u) must be less than %u.",
406 sbp->sb_rgcount, XFS_MAX_RGNUMBER);
407 return -EINVAL;
408 }
409
410 groups = howmany_64(sbp->sb_rextents, sbp->sb_rgextents);
411 if (groups != sbp->sb_rgcount) {
412 xfs_warn(mp,
413 "Realtime groups (%u) do not cover the entire rt section; need (%llu) groups.",
414 sbp->sb_rgcount, groups);
415 return -EINVAL;
416 }
417
418 /* Exchange-range is required for fsr to work on realtime files */
419 if (!(sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_EXCHRANGE)) {
420 xfs_warn(mp,
421 "Realtime groups feature requires exchange-range support.");
422 return -EINVAL;
423 }
424
425 rgblklog = xfs_compute_rgblklog(sbp->sb_rgextents, sbp->sb_rextsize);
426 if (sbp->sb_rgblklog != rgblklog) {
427 xfs_warn(mp,
428 "Realtime group log (%d) does not match expected value (%d).",
429 sbp->sb_rgblklog, rgblklog);
430 return -EINVAL;
431 }
432
433 return 0;
434 }
435
436 /* Check the validity of the SB. */
437 STATIC int
xfs_validate_sb_common(struct xfs_mount * mp,struct xfs_buf * bp,struct xfs_sb * sbp)438 xfs_validate_sb_common(
439 struct xfs_mount *mp,
440 struct xfs_buf *bp,
441 struct xfs_sb *sbp)
442 {
443 struct xfs_dsb *dsb = bp->b_addr;
444 uint32_t agcount = 0;
445 uint32_t rem;
446 bool has_dalign;
447 int error;
448
449 if (!xfs_verify_magic(bp, dsb->sb_magicnum)) {
450 xfs_warn(mp,
451 "Superblock has bad magic number 0x%x. Not an XFS filesystem?",
452 be32_to_cpu(dsb->sb_magicnum));
453 return -EWRONGFS;
454 }
455
456 if (!xfs_sb_good_version(sbp)) {
457 xfs_warn(mp,
458 "Superblock has unknown features enabled or corrupted feature masks.");
459 return -EWRONGFS;
460 }
461
462 /*
463 * Validate feature flags and state
464 */
465 if (xfs_sb_is_v5(sbp)) {
466 if (sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) {
467 xfs_notice(mp,
468 "Block size (%u bytes) too small for Version 5 superblock (minimum %d bytes)",
469 sbp->sb_blocksize, XFS_MIN_CRC_BLOCKSIZE);
470 return -EFSCORRUPTED;
471 }
472
473 /* V5 has a separate project quota inode */
474 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
475 xfs_notice(mp,
476 "Version 5 of Super block has XFS_OQUOTA bits.");
477 return -EFSCORRUPTED;
478 }
479
480 /*
481 * Full inode chunks must be aligned to inode chunk size when
482 * sparse inodes are enabled to support the sparse chunk
483 * allocation algorithm and prevent overlapping inode records.
484 */
485 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES) {
486 uint32_t align;
487
488 align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
489 >> sbp->sb_blocklog;
490 if (sbp->sb_inoalignmt != align) {
491 xfs_warn(mp,
492 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
493 sbp->sb_inoalignmt, align);
494 return -EINVAL;
495 }
496
497 if (sbp->sb_spino_align &&
498 (sbp->sb_spino_align > sbp->sb_inoalignmt ||
499 (sbp->sb_inoalignmt % sbp->sb_spino_align) != 0)) {
500 xfs_warn(mp,
501 "Sparse inode alignment (%u) is invalid, must be integer factor of (%u).",
502 sbp->sb_spino_align,
503 sbp->sb_inoalignmt);
504 return -EINVAL;
505 }
506 } else if (sbp->sb_spino_align) {
507 xfs_warn(mp,
508 "Sparse inode alignment (%u) should be zero.",
509 sbp->sb_spino_align);
510 return -EINVAL;
511 }
512
513 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR) {
514 if (memchr_inv(sbp->sb_pad, 0, sizeof(sbp->sb_pad))) {
515 xfs_warn(mp,
516 "Metadir superblock padding fields must be zero.");
517 return -EINVAL;
518 }
519
520 error = xfs_validate_sb_rtgroups(mp, sbp);
521 if (error)
522 return error;
523 }
524 } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
525 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
526 xfs_notice(mp,
527 "Superblock earlier than Version 5 has XFS_{P|G}QUOTA_{ENFD|CHKD} bits.");
528 return -EFSCORRUPTED;
529 }
530
531 if (unlikely(
532 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
533 xfs_warn(mp,
534 "filesystem is marked as having an external log; "
535 "specify logdev on the mount command line.");
536 return -EINVAL;
537 }
538
539 if (unlikely(
540 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
541 xfs_warn(mp,
542 "filesystem is marked as having an internal log; "
543 "do not specify logdev on the mount command line.");
544 return -EINVAL;
545 }
546
547 /* Compute agcount for this number of dblocks and agblocks */
548 if (sbp->sb_agblocks) {
549 agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem);
550 if (rem)
551 agcount++;
552 }
553
554 /*
555 * More sanity checking. Most of these were stolen directly from
556 * xfs_repair.
557 */
558 if (unlikely(
559 sbp->sb_agcount <= 0 ||
560 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
561 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
562 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
563 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
564 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
565 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
566 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
567 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
568 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
569 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
570 sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
571 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
572 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
573 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
574 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
575 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
576 sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
577 XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES ||
578 XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES ||
579 sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1 ||
580 agcount == 0 || agcount != sbp->sb_agcount ||
581 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
582 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
583 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
584 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) ||
585 sbp->sb_dblocks == 0 ||
586 sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) ||
587 sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp) ||
588 sbp->sb_shared_vn != 0)) {
589 xfs_notice(mp, "SB sanity check failed");
590 return -EFSCORRUPTED;
591 }
592
593 /*
594 * Logs that are too large are not supported at all. Reject them
595 * outright. Logs that are too small are tolerated on v4 filesystems,
596 * but we can only check that when mounting the log. Hence we skip
597 * those checks here.
598 */
599 if (sbp->sb_logblocks > XFS_MAX_LOG_BLOCKS) {
600 xfs_notice(mp,
601 "Log size 0x%x blocks too large, maximum size is 0x%llx blocks",
602 sbp->sb_logblocks, XFS_MAX_LOG_BLOCKS);
603 return -EFSCORRUPTED;
604 }
605
606 if (XFS_FSB_TO_B(mp, sbp->sb_logblocks) > XFS_MAX_LOG_BYTES) {
607 xfs_warn(mp,
608 "log size 0x%llx bytes too large, maximum size is 0x%llx bytes",
609 XFS_FSB_TO_B(mp, sbp->sb_logblocks),
610 XFS_MAX_LOG_BYTES);
611 return -EFSCORRUPTED;
612 }
613
614 /*
615 * Do not allow filesystems with corrupted log sector or stripe units to
616 * be mounted. We cannot safely size the iclogs or write to the log if
617 * the log stripe unit is not valid.
618 */
619 if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT) {
620 if (sbp->sb_logsectsize != (1U << sbp->sb_logsectlog)) {
621 xfs_notice(mp,
622 "log sector size in bytes/log2 (0x%x/0x%x) must match",
623 sbp->sb_logsectsize, 1U << sbp->sb_logsectlog);
624 return -EFSCORRUPTED;
625 }
626 } else if (sbp->sb_logsectsize || sbp->sb_logsectlog) {
627 xfs_notice(mp,
628 "log sector size in bytes/log2 (0x%x/0x%x) are not zero",
629 sbp->sb_logsectsize, sbp->sb_logsectlog);
630 return -EFSCORRUPTED;
631 }
632
633 if (sbp->sb_logsunit > 1) {
634 if (sbp->sb_logsunit % sbp->sb_blocksize) {
635 xfs_notice(mp,
636 "log stripe unit 0x%x bytes must be a multiple of block size",
637 sbp->sb_logsunit);
638 return -EFSCORRUPTED;
639 }
640 if (sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE) {
641 xfs_notice(mp,
642 "log stripe unit 0x%x bytes over maximum size (0x%x bytes)",
643 sbp->sb_logsunit, XLOG_MAX_RECORD_BSIZE);
644 return -EFSCORRUPTED;
645 }
646 }
647
648 if (!xfs_validate_rt_geometry(sbp)) {
649 xfs_notice(mp,
650 "realtime %sgeometry check failed",
651 sbp->sb_rblocks ? "" : "zeroed ");
652 return -EFSCORRUPTED;
653 }
654
655 /*
656 * Either (sb_unit and !hasdalign) or (!sb_unit and hasdalign)
657 * would imply the image is corrupted.
658 */
659 has_dalign = sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT;
660 if (!!sbp->sb_unit ^ has_dalign) {
661 xfs_notice(mp, "SB stripe alignment sanity check failed");
662 return -EFSCORRUPTED;
663 }
664
665 if (!xfs_validate_stripe_geometry(mp, XFS_FSB_TO_B(mp, sbp->sb_unit),
666 XFS_FSB_TO_B(mp, sbp->sb_width), 0,
667 xfs_buf_daddr(bp) == XFS_SB_DADDR, false))
668 return -EFSCORRUPTED;
669
670 /*
671 * Currently only very few inode sizes are supported.
672 */
673 switch (sbp->sb_inodesize) {
674 case 256:
675 case 512:
676 case 1024:
677 case 2048:
678 break;
679 default:
680 xfs_warn(mp, "inode size of %d bytes not supported",
681 sbp->sb_inodesize);
682 return -ENOSYS;
683 }
684
685 return 0;
686 }
687
688 void
xfs_sb_quota_from_disk(struct xfs_sb * sbp)689 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
690 {
691 if (xfs_sb_is_v5(sbp) &&
692 (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR)) {
693 sbp->sb_uquotino = NULLFSINO;
694 sbp->sb_gquotino = NULLFSINO;
695 sbp->sb_pquotino = NULLFSINO;
696 return;
697 }
698
699 /*
700 * older mkfs doesn't initialize quota inodes to NULLFSINO. This
701 * leads to in-core values having two different values for a quota
702 * inode to be invalid: 0 and NULLFSINO. Change it to a single value
703 * NULLFSINO.
704 *
705 * Note that this change affect only the in-core values. These
706 * values are not written back to disk unless any quota information
707 * is written to the disk. Even in that case, sb_pquotino field is
708 * not written to disk unless the superblock supports pquotino.
709 */
710 if (sbp->sb_uquotino == 0)
711 sbp->sb_uquotino = NULLFSINO;
712 if (sbp->sb_gquotino == 0)
713 sbp->sb_gquotino = NULLFSINO;
714 if (sbp->sb_pquotino == 0)
715 sbp->sb_pquotino = NULLFSINO;
716
717 /*
718 * We need to do these manipilations only if we are working
719 * with an older version of on-disk superblock.
720 */
721 if (xfs_sb_is_v5(sbp))
722 return;
723
724 if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
725 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
726 XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
727 if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
728 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
729 XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
730 sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
731
732 if (sbp->sb_qflags & XFS_PQUOTA_ACCT &&
733 sbp->sb_gquotino != NULLFSINO) {
734 /*
735 * In older version of superblock, on-disk superblock only
736 * has sb_gquotino, and in-core superblock has both sb_gquotino
737 * and sb_pquotino. But, only one of them is supported at any
738 * point of time. So, if PQUOTA is set in disk superblock,
739 * copy over sb_gquotino to sb_pquotino. The NULLFSINO test
740 * above is to make sure we don't do this twice and wipe them
741 * both out!
742 */
743 sbp->sb_pquotino = sbp->sb_gquotino;
744 sbp->sb_gquotino = NULLFSINO;
745 }
746 }
747
748 static void
__xfs_sb_from_disk(struct xfs_sb * to,struct xfs_dsb * from,bool convert_xquota)749 __xfs_sb_from_disk(
750 struct xfs_sb *to,
751 struct xfs_dsb *from,
752 bool convert_xquota)
753 {
754 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
755 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
756 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
757 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
758 to->sb_rextents = be64_to_cpu(from->sb_rextents);
759 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
760 to->sb_logstart = be64_to_cpu(from->sb_logstart);
761 to->sb_rootino = be64_to_cpu(from->sb_rootino);
762 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
763 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
764 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
765 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
766 to->sb_agcount = be32_to_cpu(from->sb_agcount);
767 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
768 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
769 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
770 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
771 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
772 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
773 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
774 to->sb_blocklog = from->sb_blocklog;
775 to->sb_sectlog = from->sb_sectlog;
776 to->sb_inodelog = from->sb_inodelog;
777 to->sb_inopblog = from->sb_inopblog;
778 to->sb_agblklog = from->sb_agblklog;
779 to->sb_rextslog = from->sb_rextslog;
780 to->sb_inprogress = from->sb_inprogress;
781 to->sb_imax_pct = from->sb_imax_pct;
782 to->sb_icount = be64_to_cpu(from->sb_icount);
783 to->sb_ifree = be64_to_cpu(from->sb_ifree);
784 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
785 to->sb_frextents = be64_to_cpu(from->sb_frextents);
786 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
787 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
788 to->sb_qflags = be16_to_cpu(from->sb_qflags);
789 to->sb_flags = from->sb_flags;
790 to->sb_shared_vn = from->sb_shared_vn;
791 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
792 to->sb_unit = be32_to_cpu(from->sb_unit);
793 to->sb_width = be32_to_cpu(from->sb_width);
794 to->sb_dirblklog = from->sb_dirblklog;
795 to->sb_logsectlog = from->sb_logsectlog;
796 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
797 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
798 to->sb_features2 = be32_to_cpu(from->sb_features2);
799 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
800 to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
801 to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
802 to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
803 to->sb_features_log_incompat =
804 be32_to_cpu(from->sb_features_log_incompat);
805 /* crc is only used on disk, not in memory; just init to 0 here. */
806 to->sb_crc = 0;
807 to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
808 to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
809 to->sb_lsn = be64_to_cpu(from->sb_lsn);
810 /*
811 * sb_meta_uuid is only on disk if it differs from sb_uuid and the
812 * feature flag is set; if not set we keep it only in memory.
813 */
814 if (xfs_sb_is_v5(to) &&
815 (to->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID))
816 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
817 else
818 uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
819 /* Convert on-disk flags to in-memory flags? */
820 if (convert_xquota)
821 xfs_sb_quota_from_disk(to);
822
823 if (to->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR) {
824 to->sb_metadirino = be64_to_cpu(from->sb_metadirino);
825 to->sb_rgblklog = from->sb_rgblklog;
826 memcpy(to->sb_pad, from->sb_pad, sizeof(to->sb_pad));
827 to->sb_rgcount = be32_to_cpu(from->sb_rgcount);
828 to->sb_rgextents = be32_to_cpu(from->sb_rgextents);
829 to->sb_rbmino = NULLFSINO;
830 to->sb_rsumino = NULLFSINO;
831 } else {
832 to->sb_metadirino = NULLFSINO;
833 to->sb_rgcount = 1;
834 to->sb_rgextents = 0;
835 }
836 }
837
838 void
xfs_sb_from_disk(struct xfs_sb * to,struct xfs_dsb * from)839 xfs_sb_from_disk(
840 struct xfs_sb *to,
841 struct xfs_dsb *from)
842 {
843 __xfs_sb_from_disk(to, from, true);
844 }
845
846 static void
xfs_sb_quota_to_disk(struct xfs_dsb * to,struct xfs_sb * from)847 xfs_sb_quota_to_disk(
848 struct xfs_dsb *to,
849 struct xfs_sb *from)
850 {
851 uint16_t qflags = from->sb_qflags;
852
853 if (xfs_sb_is_v5(from) &&
854 (from->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR)) {
855 to->sb_qflags = cpu_to_be16(from->sb_qflags);
856 to->sb_uquotino = cpu_to_be64(0);
857 to->sb_gquotino = cpu_to_be64(0);
858 to->sb_pquotino = cpu_to_be64(0);
859 return;
860 }
861
862 to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
863
864 /*
865 * The in-memory superblock quota state matches the v5 on-disk format so
866 * just write them out and return
867 */
868 if (xfs_sb_is_v5(from)) {
869 to->sb_qflags = cpu_to_be16(from->sb_qflags);
870 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
871 to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
872 return;
873 }
874
875 /*
876 * For older superblocks (v4), the in-core version of sb_qflags do not
877 * have XFS_OQUOTA_* flags, whereas the on-disk version does. So,
878 * convert incore XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
879 */
880 qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
881 XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
882
883 if (from->sb_qflags &
884 (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
885 qflags |= XFS_OQUOTA_ENFD;
886 if (from->sb_qflags &
887 (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
888 qflags |= XFS_OQUOTA_CHKD;
889 to->sb_qflags = cpu_to_be16(qflags);
890
891 /*
892 * GQUOTINO and PQUOTINO cannot be used together in versions
893 * of superblock that do not have pquotino. from->sb_flags
894 * tells us which quota is active and should be copied to
895 * disk. If neither are active, we should NULL the inode.
896 *
897 * In all cases, the separate pquotino must remain 0 because it
898 * is beyond the "end" of the valid non-pquotino superblock.
899 */
900 if (from->sb_qflags & XFS_GQUOTA_ACCT)
901 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
902 else if (from->sb_qflags & XFS_PQUOTA_ACCT)
903 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
904 else {
905 /*
906 * We can't rely on just the fields being logged to tell us
907 * that it is safe to write NULLFSINO - we should only do that
908 * if quotas are not actually enabled. Hence only write
909 * NULLFSINO if both in-core quota inodes are NULL.
910 */
911 if (from->sb_gquotino == NULLFSINO &&
912 from->sb_pquotino == NULLFSINO)
913 to->sb_gquotino = cpu_to_be64(NULLFSINO);
914 }
915
916 to->sb_pquotino = 0;
917 }
918
919 void
xfs_sb_to_disk(struct xfs_dsb * to,struct xfs_sb * from)920 xfs_sb_to_disk(
921 struct xfs_dsb *to,
922 struct xfs_sb *from)
923 {
924 xfs_sb_quota_to_disk(to, from);
925
926 to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
927 to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
928 to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
929 to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
930 to->sb_rextents = cpu_to_be64(from->sb_rextents);
931 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
932 to->sb_logstart = cpu_to_be64(from->sb_logstart);
933 to->sb_rootino = cpu_to_be64(from->sb_rootino);
934 to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
935 to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
936 to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
937 to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
938 to->sb_agcount = cpu_to_be32(from->sb_agcount);
939 to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
940 to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
941 to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
942 to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
943 to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
944 to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
945 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
946 to->sb_blocklog = from->sb_blocklog;
947 to->sb_sectlog = from->sb_sectlog;
948 to->sb_inodelog = from->sb_inodelog;
949 to->sb_inopblog = from->sb_inopblog;
950 to->sb_agblklog = from->sb_agblklog;
951 to->sb_rextslog = from->sb_rextslog;
952 to->sb_inprogress = from->sb_inprogress;
953 to->sb_imax_pct = from->sb_imax_pct;
954 to->sb_icount = cpu_to_be64(from->sb_icount);
955 to->sb_ifree = cpu_to_be64(from->sb_ifree);
956 to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
957 to->sb_frextents = cpu_to_be64(from->sb_frextents);
958
959 to->sb_flags = from->sb_flags;
960 to->sb_shared_vn = from->sb_shared_vn;
961 to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
962 to->sb_unit = cpu_to_be32(from->sb_unit);
963 to->sb_width = cpu_to_be32(from->sb_width);
964 to->sb_dirblklog = from->sb_dirblklog;
965 to->sb_logsectlog = from->sb_logsectlog;
966 to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
967 to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
968
969 /*
970 * We need to ensure that bad_features2 always matches features2.
971 * Hence we enforce that here rather than having to remember to do it
972 * everywhere else that updates features2.
973 */
974 from->sb_bad_features2 = from->sb_features2;
975 to->sb_features2 = cpu_to_be32(from->sb_features2);
976 to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
977
978 if (!xfs_sb_is_v5(from))
979 return;
980
981 to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
982 to->sb_features_ro_compat =
983 cpu_to_be32(from->sb_features_ro_compat);
984 to->sb_features_incompat =
985 cpu_to_be32(from->sb_features_incompat);
986 to->sb_features_log_incompat =
987 cpu_to_be32(from->sb_features_log_incompat);
988 to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
989 to->sb_lsn = cpu_to_be64(from->sb_lsn);
990 if (from->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
991 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
992
993 if (from->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR) {
994 to->sb_metadirino = cpu_to_be64(from->sb_metadirino);
995 to->sb_rgblklog = from->sb_rgblklog;
996 memset(to->sb_pad, 0, sizeof(to->sb_pad));
997 to->sb_rgcount = cpu_to_be32(from->sb_rgcount);
998 to->sb_rgextents = cpu_to_be32(from->sb_rgextents);
999 to->sb_rbmino = cpu_to_be64(0);
1000 to->sb_rsumino = cpu_to_be64(0);
1001 }
1002 }
1003
1004 /*
1005 * If the superblock has the CRC feature bit set or the CRC field is non-null,
1006 * check that the CRC is valid. We check the CRC field is non-null because a
1007 * single bit error could clear the feature bit and unused parts of the
1008 * superblock are supposed to be zero. Hence a non-null crc field indicates that
1009 * we've potentially lost a feature bit and we should check it anyway.
1010 *
1011 * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
1012 * last field in V4 secondary superblocks. So for secondary superblocks,
1013 * we are more forgiving, and ignore CRC failures if the primary doesn't
1014 * indicate that the fs version is V5.
1015 */
1016 static void
xfs_sb_read_verify(struct xfs_buf * bp)1017 xfs_sb_read_verify(
1018 struct xfs_buf *bp)
1019 {
1020 struct xfs_sb sb;
1021 struct xfs_mount *mp = bp->b_mount;
1022 struct xfs_dsb *dsb = bp->b_addr;
1023 int error;
1024
1025 /*
1026 * open code the version check to avoid needing to convert the entire
1027 * superblock from disk order just to check the version number
1028 */
1029 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
1030 (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
1031 XFS_SB_VERSION_5) ||
1032 dsb->sb_crc != 0)) {
1033
1034 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
1035 /* Only fail bad secondaries on a known V5 filesystem */
1036 if (xfs_buf_daddr(bp) == XFS_SB_DADDR ||
1037 xfs_has_crc(mp)) {
1038 error = -EFSBADCRC;
1039 goto out_error;
1040 }
1041 }
1042 }
1043
1044 /*
1045 * Check all the superblock fields. Don't byteswap the xquota flags
1046 * because _verify_common checks the on-disk values.
1047 */
1048 __xfs_sb_from_disk(&sb, dsb, false);
1049 error = xfs_validate_sb_common(mp, bp, &sb);
1050 if (error)
1051 goto out_error;
1052 error = xfs_validate_sb_read(mp, &sb);
1053
1054 out_error:
1055 if (error == -EFSCORRUPTED || error == -EFSBADCRC)
1056 xfs_verifier_error(bp, error, __this_address);
1057 else if (error)
1058 xfs_buf_ioerror(bp, error);
1059 }
1060
1061 /*
1062 * We may be probed for a filesystem match, so we may not want to emit
1063 * messages when the superblock buffer is not actually an XFS superblock.
1064 * If we find an XFS superblock, then run a normal, noisy mount because we are
1065 * really going to mount it and want to know about errors.
1066 */
1067 static void
xfs_sb_quiet_read_verify(struct xfs_buf * bp)1068 xfs_sb_quiet_read_verify(
1069 struct xfs_buf *bp)
1070 {
1071 struct xfs_dsb *dsb = bp->b_addr;
1072
1073 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
1074 /* XFS filesystem, verify noisily! */
1075 xfs_sb_read_verify(bp);
1076 return;
1077 }
1078 /* quietly fail */
1079 xfs_buf_ioerror(bp, -EWRONGFS);
1080 }
1081
1082 static void
xfs_sb_write_verify(struct xfs_buf * bp)1083 xfs_sb_write_verify(
1084 struct xfs_buf *bp)
1085 {
1086 struct xfs_sb sb;
1087 struct xfs_mount *mp = bp->b_mount;
1088 struct xfs_buf_log_item *bip = bp->b_log_item;
1089 struct xfs_dsb *dsb = bp->b_addr;
1090 int error;
1091
1092 /*
1093 * Check all the superblock fields. Don't byteswap the xquota flags
1094 * because _verify_common checks the on-disk values.
1095 */
1096 __xfs_sb_from_disk(&sb, dsb, false);
1097 error = xfs_validate_sb_common(mp, bp, &sb);
1098 if (error)
1099 goto out_error;
1100 error = xfs_validate_sb_write(mp, bp, &sb);
1101 if (error)
1102 goto out_error;
1103
1104 if (!xfs_sb_is_v5(&sb))
1105 return;
1106
1107 if (bip)
1108 dsb->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
1109
1110 xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
1111 return;
1112
1113 out_error:
1114 xfs_verifier_error(bp, error, __this_address);
1115 }
1116
1117 const struct xfs_buf_ops xfs_sb_buf_ops = {
1118 .name = "xfs_sb",
1119 .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
1120 .verify_read = xfs_sb_read_verify,
1121 .verify_write = xfs_sb_write_verify,
1122 };
1123
1124 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
1125 .name = "xfs_sb_quiet",
1126 .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
1127 .verify_read = xfs_sb_quiet_read_verify,
1128 .verify_write = xfs_sb_write_verify,
1129 };
1130
1131 /* Compute cached rt geometry from the incore sb. */
1132 void
xfs_sb_mount_rextsize(struct xfs_mount * mp,struct xfs_sb * sbp)1133 xfs_sb_mount_rextsize(
1134 struct xfs_mount *mp,
1135 struct xfs_sb *sbp)
1136 {
1137 struct xfs_groups *rgs = &mp->m_groups[XG_TYPE_RTG];
1138
1139 mp->m_rtxblklog = log2_if_power2(sbp->sb_rextsize);
1140 mp->m_rtxblkmask = mask64_if_power2(sbp->sb_rextsize);
1141
1142 if (xfs_sb_is_v5(sbp) &&
1143 (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR)) {
1144 rgs->blocks = sbp->sb_rgextents * sbp->sb_rextsize;
1145 rgs->blklog = mp->m_sb.sb_rgblklog;
1146 rgs->blkmask = xfs_mask32lo(mp->m_sb.sb_rgblklog);
1147 } else {
1148 rgs->blocks = 0;
1149 rgs->blklog = 0;
1150 rgs->blkmask = (uint64_t)-1;
1151 }
1152 }
1153
1154 /* Update incore sb rt extent size, then recompute the cached rt geometry. */
1155 void
xfs_mount_sb_set_rextsize(struct xfs_mount * mp,struct xfs_sb * sbp,xfs_agblock_t rextsize)1156 xfs_mount_sb_set_rextsize(
1157 struct xfs_mount *mp,
1158 struct xfs_sb *sbp,
1159 xfs_agblock_t rextsize)
1160 {
1161 sbp->sb_rextsize = rextsize;
1162 if (xfs_sb_is_v5(sbp) &&
1163 (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_METADIR))
1164 sbp->sb_rgblklog = xfs_compute_rgblklog(sbp->sb_rgextents,
1165 rextsize);
1166
1167 xfs_sb_mount_rextsize(mp, sbp);
1168 }
1169
1170 /*
1171 * xfs_mount_common
1172 *
1173 * Mount initialization code establishing various mount
1174 * fields from the superblock associated with the given
1175 * mount structure.
1176 *
1177 * Inode geometry are calculated in xfs_ialloc_setup_geometry.
1178 */
1179 void
xfs_sb_mount_common(struct xfs_mount * mp,struct xfs_sb * sbp)1180 xfs_sb_mount_common(
1181 struct xfs_mount *mp,
1182 struct xfs_sb *sbp)
1183 {
1184 struct xfs_groups *ags = &mp->m_groups[XG_TYPE_AG];
1185
1186 mp->m_agfrotor = 0;
1187 atomic_set(&mp->m_agirotor, 0);
1188 mp->m_maxagi = mp->m_sb.sb_agcount;
1189 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
1190 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
1191 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
1192 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
1193 mp->m_blockmask = sbp->sb_blocksize - 1;
1194 mp->m_blockwsize = xfs_rtbmblock_size(sbp) >> XFS_WORDLOG;
1195 mp->m_rtx_per_rbmblock = mp->m_blockwsize << XFS_NBWORDLOG;
1196
1197 ags->blocks = mp->m_sb.sb_agblocks;
1198 ags->blklog = mp->m_sb.sb_agblklog;
1199 ags->blkmask = xfs_mask32lo(mp->m_sb.sb_agblklog);
1200
1201 xfs_sb_mount_rextsize(mp, sbp);
1202
1203 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, true);
1204 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, false);
1205 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
1206 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
1207
1208 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, true);
1209 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, false);
1210 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
1211 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
1212
1213 mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(mp, sbp->sb_blocksize, true);
1214 mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(mp, sbp->sb_blocksize, false);
1215 mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
1216 mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
1217
1218 mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(mp, sbp->sb_blocksize, true);
1219 mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(mp, sbp->sb_blocksize, false);
1220 mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
1221 mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
1222
1223 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
1224 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
1225 mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp);
1226 }
1227
1228 /*
1229 * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
1230 * into the superblock buffer to be logged. It does not provide the higher
1231 * level of locking that is needed to protect the in-core superblock from
1232 * concurrent access.
1233 */
1234 void
xfs_log_sb(struct xfs_trans * tp)1235 xfs_log_sb(
1236 struct xfs_trans *tp)
1237 {
1238 struct xfs_mount *mp = tp->t_mountp;
1239 struct xfs_buf *bp = xfs_trans_getsb(tp);
1240
1241 /*
1242 * Lazy sb counters don't update the in-core superblock so do that now.
1243 * If this is at unmount, the counters will be exactly correct, but at
1244 * any other time they will only be ballpark correct because of
1245 * reservations that have been taken out percpu counters. If we have an
1246 * unclean shutdown, this will be corrected by log recovery rebuilding
1247 * the counters from the AGF block counts.
1248 */
1249 if (xfs_has_lazysbcount(mp)) {
1250 mp->m_sb.sb_icount = percpu_counter_sum_positive(&mp->m_icount);
1251 mp->m_sb.sb_ifree = min_t(uint64_t,
1252 percpu_counter_sum_positive(&mp->m_ifree),
1253 mp->m_sb.sb_icount);
1254 mp->m_sb.sb_fdblocks =
1255 percpu_counter_sum_positive(&mp->m_fdblocks);
1256 }
1257
1258 /*
1259 * sb_frextents was added to the lazy sb counters when the rt groups
1260 * feature was introduced. This counter can go negative due to the way
1261 * we handle nearly-lockless reservations, so we must use the _positive
1262 * variant here to avoid writing out nonsense frextents.
1263 */
1264 if (xfs_has_rtgroups(mp))
1265 mp->m_sb.sb_frextents =
1266 percpu_counter_sum_positive(&mp->m_frextents);
1267
1268 xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1269 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
1270 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
1271 }
1272
1273 /*
1274 * xfs_sync_sb
1275 *
1276 * Sync the superblock to disk.
1277 *
1278 * Note that the caller is responsible for checking the frozen state of the
1279 * filesystem. This procedure uses the non-blocking transaction allocator and
1280 * thus will allow modifications to a frozen fs. This is required because this
1281 * code can be called during the process of freezing where use of the high-level
1282 * allocator would deadlock.
1283 */
1284 int
xfs_sync_sb(struct xfs_mount * mp,bool wait)1285 xfs_sync_sb(
1286 struct xfs_mount *mp,
1287 bool wait)
1288 {
1289 struct xfs_trans *tp;
1290 int error;
1291
1292 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0,
1293 XFS_TRANS_NO_WRITECOUNT, &tp);
1294 if (error)
1295 return error;
1296
1297 xfs_log_sb(tp);
1298 if (wait)
1299 xfs_trans_set_sync(tp);
1300 return xfs_trans_commit(tp);
1301 }
1302
1303 /*
1304 * Update all the secondary superblocks to match the new state of the primary.
1305 * Because we are completely overwriting all the existing fields in the
1306 * secondary superblock buffers, there is no need to read them in from disk.
1307 * Just get a new buffer, stamp it and write it.
1308 *
1309 * The sb buffers need to be cached here so that we serialise against other
1310 * operations that access the secondary superblocks, but we don't want to keep
1311 * them in memory once it is written so we mark it as a one-shot buffer.
1312 */
1313 int
xfs_update_secondary_sbs(struct xfs_mount * mp)1314 xfs_update_secondary_sbs(
1315 struct xfs_mount *mp)
1316 {
1317 struct xfs_perag *pag = NULL;
1318 int saved_error = 0;
1319 int error = 0;
1320 LIST_HEAD (buffer_list);
1321
1322 /* update secondary superblocks. */
1323 while ((pag = xfs_perag_next_from(mp, pag, 1))) {
1324 struct xfs_buf *bp;
1325
1326 error = xfs_buf_get(mp->m_ddev_targp,
1327 XFS_AG_DADDR(mp, pag_agno(pag), XFS_SB_DADDR),
1328 XFS_FSS_TO_BB(mp, 1), &bp);
1329 /*
1330 * If we get an error reading or writing alternate superblocks,
1331 * continue. xfs_repair chooses the "best" superblock based
1332 * on most matches; if we break early, we'll leave more
1333 * superblocks un-updated than updated, and xfs_repair may
1334 * pick them over the properly-updated primary.
1335 */
1336 if (error) {
1337 xfs_warn(mp,
1338 "error allocating secondary superblock for ag %d",
1339 pag_agno(pag));
1340 if (!saved_error)
1341 saved_error = error;
1342 continue;
1343 }
1344
1345 bp->b_ops = &xfs_sb_buf_ops;
1346 xfs_buf_oneshot(bp);
1347 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
1348 xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1349 xfs_buf_delwri_queue(bp, &buffer_list);
1350 xfs_buf_relse(bp);
1351
1352 /* don't hold too many buffers at once */
1353 if (pag_agno(pag) % 16)
1354 continue;
1355
1356 error = xfs_buf_delwri_submit(&buffer_list);
1357 if (error) {
1358 xfs_warn(mp,
1359 "write error %d updating a secondary superblock near ag %d",
1360 error, pag_agno(pag));
1361 if (!saved_error)
1362 saved_error = error;
1363 continue;
1364 }
1365 }
1366 error = xfs_buf_delwri_submit(&buffer_list);
1367 if (error)
1368 xfs_warn(mp, "error %d writing secondary superblocks", error);
1369 return saved_error ? saved_error : error;
1370 }
1371
1372 /*
1373 * Same behavior as xfs_sync_sb, except that it is always synchronous and it
1374 * also writes the superblock buffer to disk sector 0 immediately.
1375 */
1376 int
xfs_sync_sb_buf(struct xfs_mount * mp,bool update_rtsb)1377 xfs_sync_sb_buf(
1378 struct xfs_mount *mp,
1379 bool update_rtsb)
1380 {
1381 struct xfs_trans *tp;
1382 struct xfs_buf *bp;
1383 struct xfs_buf *rtsb_bp = NULL;
1384 int error;
1385
1386 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp);
1387 if (error)
1388 return error;
1389
1390 bp = xfs_trans_getsb(tp);
1391 xfs_log_sb(tp);
1392 xfs_trans_bhold(tp, bp);
1393 if (update_rtsb) {
1394 rtsb_bp = xfs_log_rtsb(tp, bp);
1395 if (rtsb_bp)
1396 xfs_trans_bhold(tp, rtsb_bp);
1397 }
1398 xfs_trans_set_sync(tp);
1399 error = xfs_trans_commit(tp);
1400 if (error)
1401 goto out;
1402 /*
1403 * write out the sb buffer to get the changes to disk
1404 */
1405 error = xfs_bwrite(bp);
1406 if (!error && rtsb_bp)
1407 error = xfs_bwrite(rtsb_bp);
1408 out:
1409 if (rtsb_bp)
1410 xfs_buf_relse(rtsb_bp);
1411 xfs_buf_relse(bp);
1412 return error;
1413 }
1414
1415 void
xfs_fs_geometry(struct xfs_mount * mp,struct xfs_fsop_geom * geo,int struct_version)1416 xfs_fs_geometry(
1417 struct xfs_mount *mp,
1418 struct xfs_fsop_geom *geo,
1419 int struct_version)
1420 {
1421 struct xfs_sb *sbp = &mp->m_sb;
1422
1423 memset(geo, 0, sizeof(struct xfs_fsop_geom));
1424
1425 geo->blocksize = sbp->sb_blocksize;
1426 geo->rtextsize = sbp->sb_rextsize;
1427 geo->agblocks = sbp->sb_agblocks;
1428 geo->agcount = sbp->sb_agcount;
1429 geo->logblocks = sbp->sb_logblocks;
1430 geo->sectsize = sbp->sb_sectsize;
1431 geo->inodesize = sbp->sb_inodesize;
1432 geo->imaxpct = sbp->sb_imax_pct;
1433 geo->datablocks = sbp->sb_dblocks;
1434 geo->rtblocks = sbp->sb_rblocks;
1435 geo->rtextents = sbp->sb_rextents;
1436 geo->logstart = sbp->sb_logstart;
1437 BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid));
1438 memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid));
1439
1440 if (struct_version < 2)
1441 return;
1442
1443 geo->sunit = sbp->sb_unit;
1444 geo->swidth = sbp->sb_width;
1445
1446 if (struct_version < 3)
1447 return;
1448
1449 geo->version = XFS_FSOP_GEOM_VERSION;
1450 geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
1451 XFS_FSOP_GEOM_FLAGS_DIRV2 |
1452 XFS_FSOP_GEOM_FLAGS_EXTFLG;
1453 if (xfs_has_attr(mp))
1454 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR;
1455 if (xfs_has_quota(mp))
1456 geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA;
1457 if (xfs_has_align(mp))
1458 geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN;
1459 if (xfs_has_dalign(mp))
1460 geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN;
1461 if (xfs_has_asciici(mp))
1462 geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI;
1463 if (xfs_has_lazysbcount(mp))
1464 geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB;
1465 if (xfs_has_attr2(mp))
1466 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2;
1467 if (xfs_has_projid32(mp))
1468 geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32;
1469 if (xfs_has_crc(mp))
1470 geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB;
1471 if (xfs_has_ftype(mp))
1472 geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE;
1473 if (xfs_has_finobt(mp))
1474 geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT;
1475 if (xfs_has_sparseinodes(mp))
1476 geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES;
1477 if (xfs_has_rmapbt(mp))
1478 geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT;
1479 if (xfs_has_reflink(mp))
1480 geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK;
1481 if (xfs_has_bigtime(mp))
1482 geo->flags |= XFS_FSOP_GEOM_FLAGS_BIGTIME;
1483 if (xfs_has_inobtcounts(mp))
1484 geo->flags |= XFS_FSOP_GEOM_FLAGS_INOBTCNT;
1485 if (xfs_has_parent(mp))
1486 geo->flags |= XFS_FSOP_GEOM_FLAGS_PARENT;
1487 if (xfs_has_sector(mp)) {
1488 geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR;
1489 geo->logsectsize = sbp->sb_logsectsize;
1490 } else {
1491 geo->logsectsize = BBSIZE;
1492 }
1493 if (xfs_has_large_extent_counts(mp))
1494 geo->flags |= XFS_FSOP_GEOM_FLAGS_NREXT64;
1495 if (xfs_has_exchange_range(mp))
1496 geo->flags |= XFS_FSOP_GEOM_FLAGS_EXCHANGE_RANGE;
1497 if (xfs_has_metadir(mp))
1498 geo->flags |= XFS_FSOP_GEOM_FLAGS_METADIR;
1499 geo->rtsectsize = sbp->sb_blocksize;
1500 geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp);
1501
1502 if (struct_version < 4)
1503 return;
1504
1505 if (xfs_has_logv2(mp))
1506 geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2;
1507
1508 geo->logsunit = sbp->sb_logsunit;
1509
1510 if (struct_version < 5)
1511 return;
1512
1513 geo->version = XFS_FSOP_GEOM_VERSION_V5;
1514
1515 if (xfs_has_rtgroups(mp)) {
1516 geo->rgcount = sbp->sb_rgcount;
1517 geo->rgextents = sbp->sb_rgextents;
1518 }
1519 }
1520
1521 /* Read a secondary superblock. */
1522 int
xfs_sb_read_secondary(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,struct xfs_buf ** bpp)1523 xfs_sb_read_secondary(
1524 struct xfs_mount *mp,
1525 struct xfs_trans *tp,
1526 xfs_agnumber_t agno,
1527 struct xfs_buf **bpp)
1528 {
1529 struct xfs_buf *bp;
1530 int error;
1531
1532 ASSERT(agno != 0 && agno != NULLAGNUMBER);
1533 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1534 XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1535 XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops);
1536 if (xfs_metadata_is_sick(error))
1537 xfs_agno_mark_sick(mp, agno, XFS_SICK_AG_SB);
1538 if (error)
1539 return error;
1540 xfs_buf_set_ref(bp, XFS_SSB_REF);
1541 *bpp = bp;
1542 return 0;
1543 }
1544
1545 /* Get an uninitialised secondary superblock buffer. */
1546 int
xfs_sb_get_secondary(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,struct xfs_buf ** bpp)1547 xfs_sb_get_secondary(
1548 struct xfs_mount *mp,
1549 struct xfs_trans *tp,
1550 xfs_agnumber_t agno,
1551 struct xfs_buf **bpp)
1552 {
1553 struct xfs_buf *bp;
1554 int error;
1555
1556 ASSERT(agno != 0 && agno != NULLAGNUMBER);
1557 error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1558 XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1559 XFS_FSS_TO_BB(mp, 1), 0, &bp);
1560 if (error)
1561 return error;
1562 bp->b_ops = &xfs_sb_buf_ops;
1563 xfs_buf_oneshot(bp);
1564 *bpp = bp;
1565 return 0;
1566 }
1567
1568 /*
1569 * sunit, swidth, sectorsize(optional with 0) should be all in bytes, so users
1570 * won't be confused by values in error messages. This function returns false
1571 * if the stripe geometry is invalid and the caller is unable to repair the
1572 * stripe configuration later in the mount process.
1573 */
1574 bool
xfs_validate_stripe_geometry(struct xfs_mount * mp,__s64 sunit,__s64 swidth,int sectorsize,bool may_repair,bool silent)1575 xfs_validate_stripe_geometry(
1576 struct xfs_mount *mp,
1577 __s64 sunit,
1578 __s64 swidth,
1579 int sectorsize,
1580 bool may_repair,
1581 bool silent)
1582 {
1583 if (swidth > INT_MAX) {
1584 if (!silent)
1585 xfs_notice(mp,
1586 "stripe width (%lld) is too large", swidth);
1587 goto check_override;
1588 }
1589
1590 if (sunit > swidth) {
1591 if (!silent)
1592 xfs_notice(mp,
1593 "stripe unit (%lld) is larger than the stripe width (%lld)", sunit, swidth);
1594 goto check_override;
1595 }
1596
1597 if (sectorsize && (int)sunit % sectorsize) {
1598 if (!silent)
1599 xfs_notice(mp,
1600 "stripe unit (%lld) must be a multiple of the sector size (%d)",
1601 sunit, sectorsize);
1602 goto check_override;
1603 }
1604
1605 if (sunit && !swidth) {
1606 if (!silent)
1607 xfs_notice(mp,
1608 "invalid stripe unit (%lld) and stripe width of 0", sunit);
1609 goto check_override;
1610 }
1611
1612 if (!sunit && swidth) {
1613 if (!silent)
1614 xfs_notice(mp,
1615 "invalid stripe width (%lld) and stripe unit of 0", swidth);
1616 goto check_override;
1617 }
1618
1619 if (sunit && (int)swidth % (int)sunit) {
1620 if (!silent)
1621 xfs_notice(mp,
1622 "stripe width (%lld) must be a multiple of the stripe unit (%lld)",
1623 swidth, sunit);
1624 goto check_override;
1625 }
1626 return true;
1627
1628 check_override:
1629 if (!may_repair)
1630 return false;
1631 /*
1632 * During mount, mp->m_dalign will not be set unless the sunit mount
1633 * option was set. If it was set, ignore the bad stripe alignment values
1634 * and allow the validation and overwrite later in the mount process to
1635 * attempt to overwrite the bad stripe alignment values with the values
1636 * supplied by mount options.
1637 */
1638 if (!mp->m_dalign)
1639 return false;
1640 if (!silent)
1641 xfs_notice(mp,
1642 "Will try to correct with specified mount options sunit (%d) and swidth (%d)",
1643 BBTOB(mp->m_dalign), BBTOB(mp->m_swidth));
1644 return true;
1645 }
1646
1647 /*
1648 * Compute the maximum level number of the realtime summary file, as defined by
1649 * mkfs. The historic use of highbit32 on a 64-bit quantity prohibited correct
1650 * use of rt volumes with more than 2^32 extents.
1651 */
1652 uint8_t
xfs_compute_rextslog(xfs_rtbxlen_t rtextents)1653 xfs_compute_rextslog(
1654 xfs_rtbxlen_t rtextents)
1655 {
1656 if (!rtextents)
1657 return 0;
1658 return xfs_highbit64(rtextents);
1659 }
1660