xref: /linux/fs/xfs/libxfs/xfs_sb.c (revision 90d32e92011eaae8e70a9169b4e7acf4ca8f9d3a)
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 
31 /*
32  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
33  */
34 
35 /*
36  * Check that all the V4 feature bits that the V5 filesystem format requires are
37  * correctly set.
38  */
39 static bool
40 xfs_sb_validate_v5_features(
41 	struct xfs_sb	*sbp)
42 {
43 	/* We must not have any unknown V4 feature bits set */
44 	if (sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS)
45 		return false;
46 
47 	/*
48 	 * The CRC bit is considered an invalid V4 flag, so we have to add it
49 	 * manually to the OKBITS mask.
50 	 */
51 	if (sbp->sb_features2 & ~(XFS_SB_VERSION2_OKBITS |
52 				  XFS_SB_VERSION2_CRCBIT))
53 		return false;
54 
55 	/* Now check all the required V4 feature flags are set. */
56 
57 #define V5_VERS_FLAGS	(XFS_SB_VERSION_NLINKBIT	| \
58 			XFS_SB_VERSION_ALIGNBIT		| \
59 			XFS_SB_VERSION_LOGV2BIT		| \
60 			XFS_SB_VERSION_EXTFLGBIT	| \
61 			XFS_SB_VERSION_DIRV2BIT		| \
62 			XFS_SB_VERSION_MOREBITSBIT)
63 
64 #define V5_FEAT_FLAGS	(XFS_SB_VERSION2_LAZYSBCOUNTBIT	| \
65 			XFS_SB_VERSION2_ATTR2BIT	| \
66 			XFS_SB_VERSION2_PROJID32BIT	| \
67 			XFS_SB_VERSION2_CRCBIT)
68 
69 	if ((sbp->sb_versionnum & V5_VERS_FLAGS) != V5_VERS_FLAGS)
70 		return false;
71 	if ((sbp->sb_features2 & V5_FEAT_FLAGS) != V5_FEAT_FLAGS)
72 		return false;
73 	return true;
74 }
75 
76 /*
77  * We current support XFS v5 formats with known features and v4 superblocks with
78  * at least V2 directories.
79  */
80 bool
81 xfs_sb_good_version(
82 	struct xfs_sb	*sbp)
83 {
84 	/*
85 	 * All v5 filesystems are supported, but we must check that all the
86 	 * required v4 feature flags are enabled correctly as the code checks
87 	 * those flags and not for v5 support.
88 	 */
89 	if (xfs_sb_is_v5(sbp))
90 		return xfs_sb_validate_v5_features(sbp);
91 
92 	/* versions prior to v4 are not supported */
93 	if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_4)
94 		return false;
95 
96 	/* We must not have any unknown v4 feature bits set */
97 	if ((sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS) ||
98 	    ((sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) &&
99 	     (sbp->sb_features2 & ~XFS_SB_VERSION2_OKBITS)))
100 		return false;
101 
102 	/* V4 filesystems need v2 directories and unwritten extents */
103 	if (!(sbp->sb_versionnum & XFS_SB_VERSION_DIRV2BIT))
104 		return false;
105 	if (!(sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT))
106 		return false;
107 
108 	/* It's a supported v4 filesystem */
109 	return true;
110 }
111 
112 uint64_t
113 xfs_sb_version_to_features(
114 	struct xfs_sb	*sbp)
115 {
116 	uint64_t	features = 0;
117 
118 	/* optional V4 features */
119 	if (sbp->sb_rblocks > 0)
120 		features |= XFS_FEAT_REALTIME;
121 	if (sbp->sb_versionnum & XFS_SB_VERSION_NLINKBIT)
122 		features |= XFS_FEAT_NLINK;
123 	if (sbp->sb_versionnum & XFS_SB_VERSION_ATTRBIT)
124 		features |= XFS_FEAT_ATTR;
125 	if (sbp->sb_versionnum & XFS_SB_VERSION_QUOTABIT)
126 		features |= XFS_FEAT_QUOTA;
127 	if (sbp->sb_versionnum & XFS_SB_VERSION_ALIGNBIT)
128 		features |= XFS_FEAT_ALIGN;
129 	if (sbp->sb_versionnum & XFS_SB_VERSION_LOGV2BIT)
130 		features |= XFS_FEAT_LOGV2;
131 	if (sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT)
132 		features |= XFS_FEAT_DALIGN;
133 	if (sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT)
134 		features |= XFS_FEAT_EXTFLG;
135 	if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT)
136 		features |= XFS_FEAT_SECTOR;
137 	if (sbp->sb_versionnum & XFS_SB_VERSION_BORGBIT)
138 		features |= XFS_FEAT_ASCIICI;
139 	if (sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) {
140 		if (sbp->sb_features2 & XFS_SB_VERSION2_LAZYSBCOUNTBIT)
141 			features |= XFS_FEAT_LAZYSBCOUNT;
142 		if (sbp->sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
143 			features |= XFS_FEAT_ATTR2;
144 		if (sbp->sb_features2 & XFS_SB_VERSION2_PROJID32BIT)
145 			features |= XFS_FEAT_PROJID32;
146 		if (sbp->sb_features2 & XFS_SB_VERSION2_FTYPE)
147 			features |= XFS_FEAT_FTYPE;
148 	}
149 
150 	if (!xfs_sb_is_v5(sbp))
151 		return features;
152 
153 	/* Always on V5 features */
154 	features |= XFS_FEAT_ALIGN | XFS_FEAT_LOGV2 | XFS_FEAT_EXTFLG |
155 		    XFS_FEAT_LAZYSBCOUNT | XFS_FEAT_ATTR2 | XFS_FEAT_PROJID32 |
156 		    XFS_FEAT_V3INODES | XFS_FEAT_CRC | XFS_FEAT_PQUOTINO;
157 
158 	/* Optional V5 features */
159 	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_FINOBT)
160 		features |= XFS_FEAT_FINOBT;
161 	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_RMAPBT)
162 		features |= XFS_FEAT_RMAPBT;
163 	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_REFLINK)
164 		features |= XFS_FEAT_REFLINK;
165 	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_INOBTCNT)
166 		features |= XFS_FEAT_INOBTCNT;
167 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_FTYPE)
168 		features |= XFS_FEAT_FTYPE;
169 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES)
170 		features |= XFS_FEAT_SPINODES;
171 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
172 		features |= XFS_FEAT_META_UUID;
173 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_BIGTIME)
174 		features |= XFS_FEAT_BIGTIME;
175 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR)
176 		features |= XFS_FEAT_NEEDSREPAIR;
177 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NREXT64)
178 		features |= XFS_FEAT_NREXT64;
179 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_EXCHRANGE)
180 		features |= XFS_FEAT_EXCHANGE_RANGE;
181 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_PARENT)
182 		features |= XFS_FEAT_PARENT;
183 
184 	return features;
185 }
186 
187 /* Check all the superblock fields we care about when reading one in. */
188 STATIC int
189 xfs_validate_sb_read(
190 	struct xfs_mount	*mp,
191 	struct xfs_sb		*sbp)
192 {
193 	if (!xfs_sb_is_v5(sbp))
194 		return 0;
195 
196 	/*
197 	 * Version 5 superblock feature mask validation. Reject combinations
198 	 * the kernel cannot support up front before checking anything else.
199 	 */
200 	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
201 		xfs_warn(mp,
202 "Superblock has unknown compatible features (0x%x) enabled.",
203 			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
204 		xfs_warn(mp,
205 "Using a more recent kernel is recommended.");
206 	}
207 
208 	if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
209 		xfs_alert(mp,
210 "Superblock has unknown read-only compatible features (0x%x) enabled.",
211 			(sbp->sb_features_ro_compat &
212 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
213 		if (!xfs_is_readonly(mp)) {
214 			xfs_warn(mp,
215 "Attempted to mount read-only compatible filesystem read-write.");
216 			xfs_warn(mp,
217 "Filesystem can only be safely mounted read only.");
218 
219 			return -EINVAL;
220 		}
221 	}
222 	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
223 		xfs_warn(mp,
224 "Superblock has unknown incompatible features (0x%x) enabled.",
225 			(sbp->sb_features_incompat &
226 					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
227 		xfs_warn(mp,
228 "Filesystem cannot be safely mounted by this kernel.");
229 		return -EINVAL;
230 	}
231 
232 	return 0;
233 }
234 
235 /* Check all the superblock fields we care about when writing one out. */
236 STATIC int
237 xfs_validate_sb_write(
238 	struct xfs_mount	*mp,
239 	struct xfs_buf		*bp,
240 	struct xfs_sb		*sbp)
241 {
242 	/*
243 	 * Carry out additional sb summary counter sanity checks when we write
244 	 * the superblock.  We skip this in the read validator because there
245 	 * could be newer superblocks in the log and if the values are garbage
246 	 * even after replay we'll recalculate them at the end of log mount.
247 	 *
248 	 * mkfs has traditionally written zeroed counters to inprogress and
249 	 * secondary superblocks, so allow this usage to continue because
250 	 * we never read counters from such superblocks.
251 	 */
252 	if (xfs_buf_daddr(bp) == XFS_SB_DADDR && !sbp->sb_inprogress &&
253 	    (sbp->sb_fdblocks > sbp->sb_dblocks ||
254 	     !xfs_verify_icount(mp, sbp->sb_icount) ||
255 	     sbp->sb_ifree > sbp->sb_icount)) {
256 		xfs_warn(mp, "SB summary counter sanity check failed");
257 		return -EFSCORRUPTED;
258 	}
259 
260 	if (!xfs_sb_is_v5(sbp))
261 		return 0;
262 
263 	/*
264 	 * Version 5 superblock feature mask validation. Reject combinations
265 	 * the kernel cannot support since we checked for unsupported bits in
266 	 * the read verifier, which means that memory is corrupt.
267 	 */
268 	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
269 		xfs_warn(mp,
270 "Corruption detected in superblock compatible features (0x%x)!",
271 			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
272 		return -EFSCORRUPTED;
273 	}
274 
275 	if (!xfs_is_readonly(mp) &&
276 	    xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
277 		xfs_alert(mp,
278 "Corruption detected in superblock read-only compatible features (0x%x)!",
279 			(sbp->sb_features_ro_compat &
280 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
281 		return -EFSCORRUPTED;
282 	}
283 	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
284 		xfs_warn(mp,
285 "Corruption detected in superblock incompatible features (0x%x)!",
286 			(sbp->sb_features_incompat &
287 					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
288 		return -EFSCORRUPTED;
289 	}
290 	if (xfs_sb_has_incompat_log_feature(sbp,
291 			XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
292 		xfs_warn(mp,
293 "Corruption detected in superblock incompatible log features (0x%x)!",
294 			(sbp->sb_features_log_incompat &
295 					XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
296 		return -EFSCORRUPTED;
297 	}
298 
299 	/*
300 	 * We can't read verify the sb LSN because the read verifier is called
301 	 * before the log is allocated and processed. We know the log is set up
302 	 * before write verifier calls, so check it here.
303 	 */
304 	if (!xfs_log_check_lsn(mp, sbp->sb_lsn))
305 		return -EFSCORRUPTED;
306 
307 	return 0;
308 }
309 
310 /* Check the validity of the SB. */
311 STATIC int
312 xfs_validate_sb_common(
313 	struct xfs_mount	*mp,
314 	struct xfs_buf		*bp,
315 	struct xfs_sb		*sbp)
316 {
317 	struct xfs_dsb		*dsb = bp->b_addr;
318 	uint32_t		agcount = 0;
319 	uint32_t		rem;
320 	bool			has_dalign;
321 
322 	if (!xfs_verify_magic(bp, dsb->sb_magicnum)) {
323 		xfs_warn(mp,
324 "Superblock has bad magic number 0x%x. Not an XFS filesystem?",
325 			be32_to_cpu(dsb->sb_magicnum));
326 		return -EWRONGFS;
327 	}
328 
329 	if (!xfs_sb_good_version(sbp)) {
330 		xfs_warn(mp,
331 "Superblock has unknown features enabled or corrupted feature masks.");
332 		return -EWRONGFS;
333 	}
334 
335 	/*
336 	 * Validate feature flags and state
337 	 */
338 	if (xfs_sb_is_v5(sbp)) {
339 		if (sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) {
340 			xfs_notice(mp,
341 "Block size (%u bytes) too small for Version 5 superblock (minimum %d bytes)",
342 				sbp->sb_blocksize, XFS_MIN_CRC_BLOCKSIZE);
343 			return -EFSCORRUPTED;
344 		}
345 
346 		/* V5 has a separate project quota inode */
347 		if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
348 			xfs_notice(mp,
349 			   "Version 5 of Super block has XFS_OQUOTA bits.");
350 			return -EFSCORRUPTED;
351 		}
352 
353 		/*
354 		 * Full inode chunks must be aligned to inode chunk size when
355 		 * sparse inodes are enabled to support the sparse chunk
356 		 * allocation algorithm and prevent overlapping inode records.
357 		 */
358 		if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES) {
359 			uint32_t	align;
360 
361 			align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
362 					>> sbp->sb_blocklog;
363 			if (sbp->sb_inoalignmt != align) {
364 				xfs_warn(mp,
365 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
366 					 sbp->sb_inoalignmt, align);
367 				return -EINVAL;
368 			}
369 		}
370 	} else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
371 				XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
372 			xfs_notice(mp,
373 "Superblock earlier than Version 5 has XFS_{P|G}QUOTA_{ENFD|CHKD} bits.");
374 			return -EFSCORRUPTED;
375 	}
376 
377 	if (unlikely(
378 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
379 		xfs_warn(mp,
380 		"filesystem is marked as having an external log; "
381 		"specify logdev on the mount command line.");
382 		return -EINVAL;
383 	}
384 
385 	if (unlikely(
386 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
387 		xfs_warn(mp,
388 		"filesystem is marked as having an internal log; "
389 		"do not specify logdev on the mount command line.");
390 		return -EINVAL;
391 	}
392 
393 	/* Compute agcount for this number of dblocks and agblocks */
394 	if (sbp->sb_agblocks) {
395 		agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem);
396 		if (rem)
397 			agcount++;
398 	}
399 
400 	/*
401 	 * More sanity checking.  Most of these were stolen directly from
402 	 * xfs_repair.
403 	 */
404 	if (unlikely(
405 	    sbp->sb_agcount <= 0					||
406 	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
407 	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
408 	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
409 	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
410 	    sbp->sb_sectsize != (1 << sbp->sb_sectlog)			||
411 	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
412 	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
413 	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
414 	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
415 	    sbp->sb_blocksize != (1 << sbp->sb_blocklog)		||
416 	    sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
417 	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
418 	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
419 	    sbp->sb_inodelog < XFS_DINODE_MIN_LOG			||
420 	    sbp->sb_inodelog > XFS_DINODE_MAX_LOG			||
421 	    sbp->sb_inodesize != (1 << sbp->sb_inodelog)		||
422 	    sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
423 	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES	||
424 	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES	||
425 	    sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1	||
426 	    agcount == 0 || agcount != sbp->sb_agcount			||
427 	    (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)	||
428 	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
429 	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
430 	    (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)	||
431 	    sbp->sb_dblocks == 0					||
432 	    sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)			||
433 	    sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp)			||
434 	    sbp->sb_shared_vn != 0)) {
435 		xfs_notice(mp, "SB sanity check failed");
436 		return -EFSCORRUPTED;
437 	}
438 
439 	/*
440 	 * Logs that are too large are not supported at all. Reject them
441 	 * outright. Logs that are too small are tolerated on v4 filesystems,
442 	 * but we can only check that when mounting the log. Hence we skip
443 	 * those checks here.
444 	 */
445 	if (sbp->sb_logblocks > XFS_MAX_LOG_BLOCKS) {
446 		xfs_notice(mp,
447 		"Log size 0x%x blocks too large, maximum size is 0x%llx blocks",
448 			 sbp->sb_logblocks, XFS_MAX_LOG_BLOCKS);
449 		return -EFSCORRUPTED;
450 	}
451 
452 	if (XFS_FSB_TO_B(mp, sbp->sb_logblocks) > XFS_MAX_LOG_BYTES) {
453 		xfs_warn(mp,
454 		"log size 0x%llx bytes too large, maximum size is 0x%llx bytes",
455 			 XFS_FSB_TO_B(mp, sbp->sb_logblocks),
456 			 XFS_MAX_LOG_BYTES);
457 		return -EFSCORRUPTED;
458 	}
459 
460 	/*
461 	 * Do not allow filesystems with corrupted log sector or stripe units to
462 	 * be mounted. We cannot safely size the iclogs or write to the log if
463 	 * the log stripe unit is not valid.
464 	 */
465 	if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT) {
466 		if (sbp->sb_logsectsize != (1U << sbp->sb_logsectlog)) {
467 			xfs_notice(mp,
468 			"log sector size in bytes/log2 (0x%x/0x%x) must match",
469 				sbp->sb_logsectsize, 1U << sbp->sb_logsectlog);
470 			return -EFSCORRUPTED;
471 		}
472 	} else if (sbp->sb_logsectsize || sbp->sb_logsectlog) {
473 		xfs_notice(mp,
474 		"log sector size in bytes/log2 (0x%x/0x%x) are not zero",
475 			sbp->sb_logsectsize, sbp->sb_logsectlog);
476 		return -EFSCORRUPTED;
477 	}
478 
479 	if (sbp->sb_logsunit > 1) {
480 		if (sbp->sb_logsunit % sbp->sb_blocksize) {
481 			xfs_notice(mp,
482 		"log stripe unit 0x%x bytes must be a multiple of block size",
483 				sbp->sb_logsunit);
484 			return -EFSCORRUPTED;
485 		}
486 		if (sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE) {
487 			xfs_notice(mp,
488 		"log stripe unit 0x%x bytes over maximum size (0x%x bytes)",
489 				sbp->sb_logsunit, XLOG_MAX_RECORD_BSIZE);
490 			return -EFSCORRUPTED;
491 		}
492 	}
493 
494 	/* Validate the realtime geometry; stolen from xfs_repair */
495 	if (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE ||
496 	    sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) {
497 		xfs_notice(mp,
498 			"realtime extent sanity check failed");
499 		return -EFSCORRUPTED;
500 	}
501 
502 	if (sbp->sb_rblocks == 0) {
503 		if (sbp->sb_rextents != 0 || sbp->sb_rbmblocks != 0 ||
504 		    sbp->sb_rextslog != 0 || sbp->sb_frextents != 0) {
505 			xfs_notice(mp,
506 				"realtime zeroed geometry check failed");
507 			return -EFSCORRUPTED;
508 		}
509 	} else {
510 		uint64_t	rexts;
511 		uint64_t	rbmblocks;
512 
513 		rexts = div_u64(sbp->sb_rblocks, sbp->sb_rextsize);
514 		rbmblocks = howmany_64(sbp->sb_rextents,
515 				       NBBY * sbp->sb_blocksize);
516 
517 		if (!xfs_validate_rtextents(rexts) ||
518 		    sbp->sb_rextents != rexts ||
519 		    sbp->sb_rextslog != xfs_compute_rextslog(rexts) ||
520 		    sbp->sb_rbmblocks != rbmblocks) {
521 			xfs_notice(mp,
522 				"realtime geometry sanity check failed");
523 			return -EFSCORRUPTED;
524 		}
525 	}
526 
527 	/*
528 	 * Either (sb_unit and !hasdalign) or (!sb_unit and hasdalign)
529 	 * would imply the image is corrupted.
530 	 */
531 	has_dalign = sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT;
532 	if (!!sbp->sb_unit ^ has_dalign) {
533 		xfs_notice(mp, "SB stripe alignment sanity check failed");
534 		return -EFSCORRUPTED;
535 	}
536 
537 	if (!xfs_validate_stripe_geometry(mp, XFS_FSB_TO_B(mp, sbp->sb_unit),
538 			XFS_FSB_TO_B(mp, sbp->sb_width), 0,
539 			xfs_buf_daddr(bp) == XFS_SB_DADDR, false))
540 		return -EFSCORRUPTED;
541 
542 	/*
543 	 * Currently only very few inode sizes are supported.
544 	 */
545 	switch (sbp->sb_inodesize) {
546 	case 256:
547 	case 512:
548 	case 1024:
549 	case 2048:
550 		break;
551 	default:
552 		xfs_warn(mp, "inode size of %d bytes not supported",
553 				sbp->sb_inodesize);
554 		return -ENOSYS;
555 	}
556 
557 	return 0;
558 }
559 
560 void
561 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
562 {
563 	/*
564 	 * older mkfs doesn't initialize quota inodes to NULLFSINO. This
565 	 * leads to in-core values having two different values for a quota
566 	 * inode to be invalid: 0 and NULLFSINO. Change it to a single value
567 	 * NULLFSINO.
568 	 *
569 	 * Note that this change affect only the in-core values. These
570 	 * values are not written back to disk unless any quota information
571 	 * is written to the disk. Even in that case, sb_pquotino field is
572 	 * not written to disk unless the superblock supports pquotino.
573 	 */
574 	if (sbp->sb_uquotino == 0)
575 		sbp->sb_uquotino = NULLFSINO;
576 	if (sbp->sb_gquotino == 0)
577 		sbp->sb_gquotino = NULLFSINO;
578 	if (sbp->sb_pquotino == 0)
579 		sbp->sb_pquotino = NULLFSINO;
580 
581 	/*
582 	 * We need to do these manipilations only if we are working
583 	 * with an older version of on-disk superblock.
584 	 */
585 	if (xfs_sb_is_v5(sbp))
586 		return;
587 
588 	if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
589 		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
590 					XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
591 	if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
592 		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
593 					XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
594 	sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
595 
596 	if (sbp->sb_qflags & XFS_PQUOTA_ACCT &&
597 	    sbp->sb_gquotino != NULLFSINO)  {
598 		/*
599 		 * In older version of superblock, on-disk superblock only
600 		 * has sb_gquotino, and in-core superblock has both sb_gquotino
601 		 * and sb_pquotino. But, only one of them is supported at any
602 		 * point of time. So, if PQUOTA is set in disk superblock,
603 		 * copy over sb_gquotino to sb_pquotino.  The NULLFSINO test
604 		 * above is to make sure we don't do this twice and wipe them
605 		 * both out!
606 		 */
607 		sbp->sb_pquotino = sbp->sb_gquotino;
608 		sbp->sb_gquotino = NULLFSINO;
609 	}
610 }
611 
612 static void
613 __xfs_sb_from_disk(
614 	struct xfs_sb	*to,
615 	struct xfs_dsb	*from,
616 	bool		convert_xquota)
617 {
618 	to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
619 	to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
620 	to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
621 	to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
622 	to->sb_rextents = be64_to_cpu(from->sb_rextents);
623 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
624 	to->sb_logstart = be64_to_cpu(from->sb_logstart);
625 	to->sb_rootino = be64_to_cpu(from->sb_rootino);
626 	to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
627 	to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
628 	to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
629 	to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
630 	to->sb_agcount = be32_to_cpu(from->sb_agcount);
631 	to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
632 	to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
633 	to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
634 	to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
635 	to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
636 	to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
637 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
638 	to->sb_blocklog = from->sb_blocklog;
639 	to->sb_sectlog = from->sb_sectlog;
640 	to->sb_inodelog = from->sb_inodelog;
641 	to->sb_inopblog = from->sb_inopblog;
642 	to->sb_agblklog = from->sb_agblklog;
643 	to->sb_rextslog = from->sb_rextslog;
644 	to->sb_inprogress = from->sb_inprogress;
645 	to->sb_imax_pct = from->sb_imax_pct;
646 	to->sb_icount = be64_to_cpu(from->sb_icount);
647 	to->sb_ifree = be64_to_cpu(from->sb_ifree);
648 	to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
649 	to->sb_frextents = be64_to_cpu(from->sb_frextents);
650 	to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
651 	to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
652 	to->sb_qflags = be16_to_cpu(from->sb_qflags);
653 	to->sb_flags = from->sb_flags;
654 	to->sb_shared_vn = from->sb_shared_vn;
655 	to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
656 	to->sb_unit = be32_to_cpu(from->sb_unit);
657 	to->sb_width = be32_to_cpu(from->sb_width);
658 	to->sb_dirblklog = from->sb_dirblklog;
659 	to->sb_logsectlog = from->sb_logsectlog;
660 	to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
661 	to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
662 	to->sb_features2 = be32_to_cpu(from->sb_features2);
663 	to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
664 	to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
665 	to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
666 	to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
667 	to->sb_features_log_incompat =
668 				be32_to_cpu(from->sb_features_log_incompat);
669 	/* crc is only used on disk, not in memory; just init to 0 here. */
670 	to->sb_crc = 0;
671 	to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
672 	to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
673 	to->sb_lsn = be64_to_cpu(from->sb_lsn);
674 	/*
675 	 * sb_meta_uuid is only on disk if it differs from sb_uuid and the
676 	 * feature flag is set; if not set we keep it only in memory.
677 	 */
678 	if (xfs_sb_is_v5(to) &&
679 	    (to->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID))
680 		uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
681 	else
682 		uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
683 	/* Convert on-disk flags to in-memory flags? */
684 	if (convert_xquota)
685 		xfs_sb_quota_from_disk(to);
686 }
687 
688 void
689 xfs_sb_from_disk(
690 	struct xfs_sb	*to,
691 	struct xfs_dsb	*from)
692 {
693 	__xfs_sb_from_disk(to, from, true);
694 }
695 
696 static void
697 xfs_sb_quota_to_disk(
698 	struct xfs_dsb	*to,
699 	struct xfs_sb	*from)
700 {
701 	uint16_t	qflags = from->sb_qflags;
702 
703 	to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
704 
705 	/*
706 	 * The in-memory superblock quota state matches the v5 on-disk format so
707 	 * just write them out and return
708 	 */
709 	if (xfs_sb_is_v5(from)) {
710 		to->sb_qflags = cpu_to_be16(from->sb_qflags);
711 		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
712 		to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
713 		return;
714 	}
715 
716 	/*
717 	 * For older superblocks (v4), the in-core version of sb_qflags do not
718 	 * have XFS_OQUOTA_* flags, whereas the on-disk version does.  So,
719 	 * convert incore XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
720 	 */
721 	qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
722 			XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
723 
724 	if (from->sb_qflags &
725 			(XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
726 		qflags |= XFS_OQUOTA_ENFD;
727 	if (from->sb_qflags &
728 			(XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
729 		qflags |= XFS_OQUOTA_CHKD;
730 	to->sb_qflags = cpu_to_be16(qflags);
731 
732 	/*
733 	 * GQUOTINO and PQUOTINO cannot be used together in versions
734 	 * of superblock that do not have pquotino. from->sb_flags
735 	 * tells us which quota is active and should be copied to
736 	 * disk. If neither are active, we should NULL the inode.
737 	 *
738 	 * In all cases, the separate pquotino must remain 0 because it
739 	 * is beyond the "end" of the valid non-pquotino superblock.
740 	 */
741 	if (from->sb_qflags & XFS_GQUOTA_ACCT)
742 		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
743 	else if (from->sb_qflags & XFS_PQUOTA_ACCT)
744 		to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
745 	else {
746 		/*
747 		 * We can't rely on just the fields being logged to tell us
748 		 * that it is safe to write NULLFSINO - we should only do that
749 		 * if quotas are not actually enabled. Hence only write
750 		 * NULLFSINO if both in-core quota inodes are NULL.
751 		 */
752 		if (from->sb_gquotino == NULLFSINO &&
753 		    from->sb_pquotino == NULLFSINO)
754 			to->sb_gquotino = cpu_to_be64(NULLFSINO);
755 	}
756 
757 	to->sb_pquotino = 0;
758 }
759 
760 void
761 xfs_sb_to_disk(
762 	struct xfs_dsb	*to,
763 	struct xfs_sb	*from)
764 {
765 	xfs_sb_quota_to_disk(to, from);
766 
767 	to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
768 	to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
769 	to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
770 	to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
771 	to->sb_rextents = cpu_to_be64(from->sb_rextents);
772 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
773 	to->sb_logstart = cpu_to_be64(from->sb_logstart);
774 	to->sb_rootino = cpu_to_be64(from->sb_rootino);
775 	to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
776 	to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
777 	to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
778 	to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
779 	to->sb_agcount = cpu_to_be32(from->sb_agcount);
780 	to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
781 	to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
782 	to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
783 	to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
784 	to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
785 	to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
786 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
787 	to->sb_blocklog = from->sb_blocklog;
788 	to->sb_sectlog = from->sb_sectlog;
789 	to->sb_inodelog = from->sb_inodelog;
790 	to->sb_inopblog = from->sb_inopblog;
791 	to->sb_agblklog = from->sb_agblklog;
792 	to->sb_rextslog = from->sb_rextslog;
793 	to->sb_inprogress = from->sb_inprogress;
794 	to->sb_imax_pct = from->sb_imax_pct;
795 	to->sb_icount = cpu_to_be64(from->sb_icount);
796 	to->sb_ifree = cpu_to_be64(from->sb_ifree);
797 	to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
798 	to->sb_frextents = cpu_to_be64(from->sb_frextents);
799 
800 	to->sb_flags = from->sb_flags;
801 	to->sb_shared_vn = from->sb_shared_vn;
802 	to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
803 	to->sb_unit = cpu_to_be32(from->sb_unit);
804 	to->sb_width = cpu_to_be32(from->sb_width);
805 	to->sb_dirblklog = from->sb_dirblklog;
806 	to->sb_logsectlog = from->sb_logsectlog;
807 	to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
808 	to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
809 
810 	/*
811 	 * We need to ensure that bad_features2 always matches features2.
812 	 * Hence we enforce that here rather than having to remember to do it
813 	 * everywhere else that updates features2.
814 	 */
815 	from->sb_bad_features2 = from->sb_features2;
816 	to->sb_features2 = cpu_to_be32(from->sb_features2);
817 	to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
818 
819 	if (!xfs_sb_is_v5(from))
820 		return;
821 
822 	to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
823 	to->sb_features_ro_compat =
824 			cpu_to_be32(from->sb_features_ro_compat);
825 	to->sb_features_incompat =
826 			cpu_to_be32(from->sb_features_incompat);
827 	to->sb_features_log_incompat =
828 			cpu_to_be32(from->sb_features_log_incompat);
829 	to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
830 	to->sb_lsn = cpu_to_be64(from->sb_lsn);
831 	if (from->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
832 		uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
833 }
834 
835 /*
836  * If the superblock has the CRC feature bit set or the CRC field is non-null,
837  * check that the CRC is valid.  We check the CRC field is non-null because a
838  * single bit error could clear the feature bit and unused parts of the
839  * superblock are supposed to be zero. Hence a non-null crc field indicates that
840  * we've potentially lost a feature bit and we should check it anyway.
841  *
842  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
843  * last field in V4 secondary superblocks.  So for secondary superblocks,
844  * we are more forgiving, and ignore CRC failures if the primary doesn't
845  * indicate that the fs version is V5.
846  */
847 static void
848 xfs_sb_read_verify(
849 	struct xfs_buf		*bp)
850 {
851 	struct xfs_sb		sb;
852 	struct xfs_mount	*mp = bp->b_mount;
853 	struct xfs_dsb		*dsb = bp->b_addr;
854 	int			error;
855 
856 	/*
857 	 * open code the version check to avoid needing to convert the entire
858 	 * superblock from disk order just to check the version number
859 	 */
860 	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
861 	    (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
862 						XFS_SB_VERSION_5) ||
863 	     dsb->sb_crc != 0)) {
864 
865 		if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
866 			/* Only fail bad secondaries on a known V5 filesystem */
867 			if (xfs_buf_daddr(bp) == XFS_SB_DADDR ||
868 			    xfs_has_crc(mp)) {
869 				error = -EFSBADCRC;
870 				goto out_error;
871 			}
872 		}
873 	}
874 
875 	/*
876 	 * Check all the superblock fields.  Don't byteswap the xquota flags
877 	 * because _verify_common checks the on-disk values.
878 	 */
879 	__xfs_sb_from_disk(&sb, dsb, false);
880 	error = xfs_validate_sb_common(mp, bp, &sb);
881 	if (error)
882 		goto out_error;
883 	error = xfs_validate_sb_read(mp, &sb);
884 
885 out_error:
886 	if (error == -EFSCORRUPTED || error == -EFSBADCRC)
887 		xfs_verifier_error(bp, error, __this_address);
888 	else if (error)
889 		xfs_buf_ioerror(bp, error);
890 }
891 
892 /*
893  * We may be probed for a filesystem match, so we may not want to emit
894  * messages when the superblock buffer is not actually an XFS superblock.
895  * If we find an XFS superblock, then run a normal, noisy mount because we are
896  * really going to mount it and want to know about errors.
897  */
898 static void
899 xfs_sb_quiet_read_verify(
900 	struct xfs_buf	*bp)
901 {
902 	struct xfs_dsb	*dsb = bp->b_addr;
903 
904 	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
905 		/* XFS filesystem, verify noisily! */
906 		xfs_sb_read_verify(bp);
907 		return;
908 	}
909 	/* quietly fail */
910 	xfs_buf_ioerror(bp, -EWRONGFS);
911 }
912 
913 static void
914 xfs_sb_write_verify(
915 	struct xfs_buf		*bp)
916 {
917 	struct xfs_sb		sb;
918 	struct xfs_mount	*mp = bp->b_mount;
919 	struct xfs_buf_log_item	*bip = bp->b_log_item;
920 	struct xfs_dsb		*dsb = bp->b_addr;
921 	int			error;
922 
923 	/*
924 	 * Check all the superblock fields.  Don't byteswap the xquota flags
925 	 * because _verify_common checks the on-disk values.
926 	 */
927 	__xfs_sb_from_disk(&sb, dsb, false);
928 	error = xfs_validate_sb_common(mp, bp, &sb);
929 	if (error)
930 		goto out_error;
931 	error = xfs_validate_sb_write(mp, bp, &sb);
932 	if (error)
933 		goto out_error;
934 
935 	if (!xfs_sb_is_v5(&sb))
936 		return;
937 
938 	if (bip)
939 		dsb->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
940 
941 	xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
942 	return;
943 
944 out_error:
945 	xfs_verifier_error(bp, error, __this_address);
946 }
947 
948 const struct xfs_buf_ops xfs_sb_buf_ops = {
949 	.name = "xfs_sb",
950 	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
951 	.verify_read = xfs_sb_read_verify,
952 	.verify_write = xfs_sb_write_verify,
953 };
954 
955 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
956 	.name = "xfs_sb_quiet",
957 	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
958 	.verify_read = xfs_sb_quiet_read_verify,
959 	.verify_write = xfs_sb_write_verify,
960 };
961 
962 /*
963  * xfs_mount_common
964  *
965  * Mount initialization code establishing various mount
966  * fields from the superblock associated with the given
967  * mount structure.
968  *
969  * Inode geometry are calculated in xfs_ialloc_setup_geometry.
970  */
971 void
972 xfs_sb_mount_common(
973 	struct xfs_mount	*mp,
974 	struct xfs_sb		*sbp)
975 {
976 	mp->m_agfrotor = 0;
977 	atomic_set(&mp->m_agirotor, 0);
978 	mp->m_maxagi = mp->m_sb.sb_agcount;
979 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
980 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
981 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
982 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
983 	mp->m_blockmask = sbp->sb_blocksize - 1;
984 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
985 	mp->m_blockwmask = mp->m_blockwsize - 1;
986 	mp->m_rtxblklog = log2_if_power2(sbp->sb_rextsize);
987 	mp->m_rtxblkmask = mask64_if_power2(sbp->sb_rextsize);
988 
989 	mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
990 	mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
991 	mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
992 	mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
993 
994 	mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
995 	mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
996 	mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
997 	mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
998 
999 	mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 1);
1000 	mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 0);
1001 	mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
1002 	mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
1003 
1004 	mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, true);
1005 	mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, false);
1006 	mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
1007 	mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
1008 
1009 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
1010 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
1011 	mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp);
1012 }
1013 
1014 /*
1015  * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
1016  * into the superblock buffer to be logged.  It does not provide the higher
1017  * level of locking that is needed to protect the in-core superblock from
1018  * concurrent access.
1019  */
1020 void
1021 xfs_log_sb(
1022 	struct xfs_trans	*tp)
1023 {
1024 	struct xfs_mount	*mp = tp->t_mountp;
1025 	struct xfs_buf		*bp = xfs_trans_getsb(tp);
1026 
1027 	/*
1028 	 * Lazy sb counters don't update the in-core superblock so do that now.
1029 	 * If this is at unmount, the counters will be exactly correct, but at
1030 	 * any other time they will only be ballpark correct because of
1031 	 * reservations that have been taken out percpu counters. If we have an
1032 	 * unclean shutdown, this will be corrected by log recovery rebuilding
1033 	 * the counters from the AGF block counts.
1034 	 *
1035 	 * Do not update sb_frextents here because it is not part of the lazy
1036 	 * sb counters, despite having a percpu counter. It is always kept
1037 	 * consistent with the ondisk rtbitmap by xfs_trans_apply_sb_deltas()
1038 	 * and hence we don't need have to update it here.
1039 	 */
1040 	if (xfs_has_lazysbcount(mp)) {
1041 		mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
1042 		mp->m_sb.sb_ifree = min_t(uint64_t,
1043 				percpu_counter_sum(&mp->m_ifree),
1044 				mp->m_sb.sb_icount);
1045 		mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1046 	}
1047 
1048 	xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1049 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
1050 	xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
1051 }
1052 
1053 /*
1054  * xfs_sync_sb
1055  *
1056  * Sync the superblock to disk.
1057  *
1058  * Note that the caller is responsible for checking the frozen state of the
1059  * filesystem. This procedure uses the non-blocking transaction allocator and
1060  * thus will allow modifications to a frozen fs. This is required because this
1061  * code can be called during the process of freezing where use of the high-level
1062  * allocator would deadlock.
1063  */
1064 int
1065 xfs_sync_sb(
1066 	struct xfs_mount	*mp,
1067 	bool			wait)
1068 {
1069 	struct xfs_trans	*tp;
1070 	int			error;
1071 
1072 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0,
1073 			XFS_TRANS_NO_WRITECOUNT, &tp);
1074 	if (error)
1075 		return error;
1076 
1077 	xfs_log_sb(tp);
1078 	if (wait)
1079 		xfs_trans_set_sync(tp);
1080 	return xfs_trans_commit(tp);
1081 }
1082 
1083 /*
1084  * Update all the secondary superblocks to match the new state of the primary.
1085  * Because we are completely overwriting all the existing fields in the
1086  * secondary superblock buffers, there is no need to read them in from disk.
1087  * Just get a new buffer, stamp it and write it.
1088  *
1089  * The sb buffers need to be cached here so that we serialise against other
1090  * operations that access the secondary superblocks, but we don't want to keep
1091  * them in memory once it is written so we mark it as a one-shot buffer.
1092  */
1093 int
1094 xfs_update_secondary_sbs(
1095 	struct xfs_mount	*mp)
1096 {
1097 	struct xfs_perag	*pag;
1098 	xfs_agnumber_t		agno = 1;
1099 	int			saved_error = 0;
1100 	int			error = 0;
1101 	LIST_HEAD		(buffer_list);
1102 
1103 	/* update secondary superblocks. */
1104 	for_each_perag_from(mp, agno, pag) {
1105 		struct xfs_buf		*bp;
1106 
1107 		error = xfs_buf_get(mp->m_ddev_targp,
1108 				 XFS_AG_DADDR(mp, pag->pag_agno, XFS_SB_DADDR),
1109 				 XFS_FSS_TO_BB(mp, 1), &bp);
1110 		/*
1111 		 * If we get an error reading or writing alternate superblocks,
1112 		 * continue.  xfs_repair chooses the "best" superblock based
1113 		 * on most matches; if we break early, we'll leave more
1114 		 * superblocks un-updated than updated, and xfs_repair may
1115 		 * pick them over the properly-updated primary.
1116 		 */
1117 		if (error) {
1118 			xfs_warn(mp,
1119 		"error allocating secondary superblock for ag %d",
1120 				pag->pag_agno);
1121 			if (!saved_error)
1122 				saved_error = error;
1123 			continue;
1124 		}
1125 
1126 		bp->b_ops = &xfs_sb_buf_ops;
1127 		xfs_buf_oneshot(bp);
1128 		xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
1129 		xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1130 		xfs_buf_delwri_queue(bp, &buffer_list);
1131 		xfs_buf_relse(bp);
1132 
1133 		/* don't hold too many buffers at once */
1134 		if (agno % 16)
1135 			continue;
1136 
1137 		error = xfs_buf_delwri_submit(&buffer_list);
1138 		if (error) {
1139 			xfs_warn(mp,
1140 		"write error %d updating a secondary superblock near ag %d",
1141 				error, pag->pag_agno);
1142 			if (!saved_error)
1143 				saved_error = error;
1144 			continue;
1145 		}
1146 	}
1147 	error = xfs_buf_delwri_submit(&buffer_list);
1148 	if (error) {
1149 		xfs_warn(mp,
1150 		"write error %d updating a secondary superblock near ag %d",
1151 			error, agno);
1152 	}
1153 
1154 	return saved_error ? saved_error : error;
1155 }
1156 
1157 /*
1158  * Same behavior as xfs_sync_sb, except that it is always synchronous and it
1159  * also writes the superblock buffer to disk sector 0 immediately.
1160  */
1161 int
1162 xfs_sync_sb_buf(
1163 	struct xfs_mount	*mp)
1164 {
1165 	struct xfs_trans	*tp;
1166 	struct xfs_buf		*bp;
1167 	int			error;
1168 
1169 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp);
1170 	if (error)
1171 		return error;
1172 
1173 	bp = xfs_trans_getsb(tp);
1174 	xfs_log_sb(tp);
1175 	xfs_trans_bhold(tp, bp);
1176 	xfs_trans_set_sync(tp);
1177 	error = xfs_trans_commit(tp);
1178 	if (error)
1179 		goto out;
1180 	/*
1181 	 * write out the sb buffer to get the changes to disk
1182 	 */
1183 	error = xfs_bwrite(bp);
1184 out:
1185 	xfs_buf_relse(bp);
1186 	return error;
1187 }
1188 
1189 void
1190 xfs_fs_geometry(
1191 	struct xfs_mount	*mp,
1192 	struct xfs_fsop_geom	*geo,
1193 	int			struct_version)
1194 {
1195 	struct xfs_sb		*sbp = &mp->m_sb;
1196 
1197 	memset(geo, 0, sizeof(struct xfs_fsop_geom));
1198 
1199 	geo->blocksize = sbp->sb_blocksize;
1200 	geo->rtextsize = sbp->sb_rextsize;
1201 	geo->agblocks = sbp->sb_agblocks;
1202 	geo->agcount = sbp->sb_agcount;
1203 	geo->logblocks = sbp->sb_logblocks;
1204 	geo->sectsize = sbp->sb_sectsize;
1205 	geo->inodesize = sbp->sb_inodesize;
1206 	geo->imaxpct = sbp->sb_imax_pct;
1207 	geo->datablocks = sbp->sb_dblocks;
1208 	geo->rtblocks = sbp->sb_rblocks;
1209 	geo->rtextents = sbp->sb_rextents;
1210 	geo->logstart = sbp->sb_logstart;
1211 	BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid));
1212 	memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid));
1213 
1214 	if (struct_version < 2)
1215 		return;
1216 
1217 	geo->sunit = sbp->sb_unit;
1218 	geo->swidth = sbp->sb_width;
1219 
1220 	if (struct_version < 3)
1221 		return;
1222 
1223 	geo->version = XFS_FSOP_GEOM_VERSION;
1224 	geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
1225 		     XFS_FSOP_GEOM_FLAGS_DIRV2 |
1226 		     XFS_FSOP_GEOM_FLAGS_EXTFLG;
1227 	if (xfs_has_attr(mp))
1228 		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR;
1229 	if (xfs_has_quota(mp))
1230 		geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA;
1231 	if (xfs_has_align(mp))
1232 		geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN;
1233 	if (xfs_has_dalign(mp))
1234 		geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN;
1235 	if (xfs_has_asciici(mp))
1236 		geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI;
1237 	if (xfs_has_lazysbcount(mp))
1238 		geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB;
1239 	if (xfs_has_attr2(mp))
1240 		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2;
1241 	if (xfs_has_projid32(mp))
1242 		geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32;
1243 	if (xfs_has_crc(mp))
1244 		geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB;
1245 	if (xfs_has_ftype(mp))
1246 		geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE;
1247 	if (xfs_has_finobt(mp))
1248 		geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT;
1249 	if (xfs_has_sparseinodes(mp))
1250 		geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES;
1251 	if (xfs_has_rmapbt(mp))
1252 		geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT;
1253 	if (xfs_has_reflink(mp))
1254 		geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK;
1255 	if (xfs_has_bigtime(mp))
1256 		geo->flags |= XFS_FSOP_GEOM_FLAGS_BIGTIME;
1257 	if (xfs_has_inobtcounts(mp))
1258 		geo->flags |= XFS_FSOP_GEOM_FLAGS_INOBTCNT;
1259 	if (xfs_has_parent(mp))
1260 		geo->flags |= XFS_FSOP_GEOM_FLAGS_PARENT;
1261 	if (xfs_has_sector(mp)) {
1262 		geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR;
1263 		geo->logsectsize = sbp->sb_logsectsize;
1264 	} else {
1265 		geo->logsectsize = BBSIZE;
1266 	}
1267 	if (xfs_has_large_extent_counts(mp))
1268 		geo->flags |= XFS_FSOP_GEOM_FLAGS_NREXT64;
1269 	if (xfs_has_exchange_range(mp))
1270 		geo->flags |= XFS_FSOP_GEOM_FLAGS_EXCHANGE_RANGE;
1271 	geo->rtsectsize = sbp->sb_blocksize;
1272 	geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp);
1273 
1274 	if (struct_version < 4)
1275 		return;
1276 
1277 	if (xfs_has_logv2(mp))
1278 		geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2;
1279 
1280 	geo->logsunit = sbp->sb_logsunit;
1281 
1282 	if (struct_version < 5)
1283 		return;
1284 
1285 	geo->version = XFS_FSOP_GEOM_VERSION_V5;
1286 }
1287 
1288 /* Read a secondary superblock. */
1289 int
1290 xfs_sb_read_secondary(
1291 	struct xfs_mount	*mp,
1292 	struct xfs_trans	*tp,
1293 	xfs_agnumber_t		agno,
1294 	struct xfs_buf		**bpp)
1295 {
1296 	struct xfs_buf		*bp;
1297 	int			error;
1298 
1299 	ASSERT(agno != 0 && agno != NULLAGNUMBER);
1300 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1301 			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1302 			XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops);
1303 	if (xfs_metadata_is_sick(error))
1304 		xfs_agno_mark_sick(mp, agno, XFS_SICK_AG_SB);
1305 	if (error)
1306 		return error;
1307 	xfs_buf_set_ref(bp, XFS_SSB_REF);
1308 	*bpp = bp;
1309 	return 0;
1310 }
1311 
1312 /* Get an uninitialised secondary superblock buffer. */
1313 int
1314 xfs_sb_get_secondary(
1315 	struct xfs_mount	*mp,
1316 	struct xfs_trans	*tp,
1317 	xfs_agnumber_t		agno,
1318 	struct xfs_buf		**bpp)
1319 {
1320 	struct xfs_buf		*bp;
1321 	int			error;
1322 
1323 	ASSERT(agno != 0 && agno != NULLAGNUMBER);
1324 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1325 			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1326 			XFS_FSS_TO_BB(mp, 1), 0, &bp);
1327 	if (error)
1328 		return error;
1329 	bp->b_ops = &xfs_sb_buf_ops;
1330 	xfs_buf_oneshot(bp);
1331 	*bpp = bp;
1332 	return 0;
1333 }
1334 
1335 /*
1336  * sunit, swidth, sectorsize(optional with 0) should be all in bytes, so users
1337  * won't be confused by values in error messages.  This function returns false
1338  * if the stripe geometry is invalid and the caller is unable to repair the
1339  * stripe configuration later in the mount process.
1340  */
1341 bool
1342 xfs_validate_stripe_geometry(
1343 	struct xfs_mount	*mp,
1344 	__s64			sunit,
1345 	__s64			swidth,
1346 	int			sectorsize,
1347 	bool			may_repair,
1348 	bool			silent)
1349 {
1350 	if (swidth > INT_MAX) {
1351 		if (!silent)
1352 			xfs_notice(mp,
1353 "stripe width (%lld) is too large", swidth);
1354 		goto check_override;
1355 	}
1356 
1357 	if (sunit > swidth) {
1358 		if (!silent)
1359 			xfs_notice(mp,
1360 "stripe unit (%lld) is larger than the stripe width (%lld)", sunit, swidth);
1361 		goto check_override;
1362 	}
1363 
1364 	if (sectorsize && (int)sunit % sectorsize) {
1365 		if (!silent)
1366 			xfs_notice(mp,
1367 "stripe unit (%lld) must be a multiple of the sector size (%d)",
1368 				   sunit, sectorsize);
1369 		goto check_override;
1370 	}
1371 
1372 	if (sunit && !swidth) {
1373 		if (!silent)
1374 			xfs_notice(mp,
1375 "invalid stripe unit (%lld) and stripe width of 0", sunit);
1376 		goto check_override;
1377 	}
1378 
1379 	if (!sunit && swidth) {
1380 		if (!silent)
1381 			xfs_notice(mp,
1382 "invalid stripe width (%lld) and stripe unit of 0", swidth);
1383 		goto check_override;
1384 	}
1385 
1386 	if (sunit && (int)swidth % (int)sunit) {
1387 		if (!silent)
1388 			xfs_notice(mp,
1389 "stripe width (%lld) must be a multiple of the stripe unit (%lld)",
1390 				   swidth, sunit);
1391 		goto check_override;
1392 	}
1393 	return true;
1394 
1395 check_override:
1396 	if (!may_repair)
1397 		return false;
1398 	/*
1399 	 * During mount, mp->m_dalign will not be set unless the sunit mount
1400 	 * option was set. If it was set, ignore the bad stripe alignment values
1401 	 * and allow the validation and overwrite later in the mount process to
1402 	 * attempt to overwrite the bad stripe alignment values with the values
1403 	 * supplied by mount options.
1404 	 */
1405 	if (!mp->m_dalign)
1406 		return false;
1407 	if (!silent)
1408 		xfs_notice(mp,
1409 "Will try to correct with specified mount options sunit (%d) and swidth (%d)",
1410 			BBTOB(mp->m_dalign), BBTOB(mp->m_swidth));
1411 	return true;
1412 }
1413 
1414 /*
1415  * Compute the maximum level number of the realtime summary file, as defined by
1416  * mkfs.  The historic use of highbit32 on a 64-bit quantity prohibited correct
1417  * use of rt volumes with more than 2^32 extents.
1418  */
1419 uint8_t
1420 xfs_compute_rextslog(
1421 	xfs_rtbxlen_t		rtextents)
1422 {
1423 	if (!rtextents)
1424 		return 0;
1425 	return xfs_highbit64(rtextents);
1426 }
1427