xref: /openbmc/linux/fs/xfs/libxfs/xfs_sb.c (revision 68f436a8)
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 
29 /*
30  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
31  */
32 
33 /*
34  * Check that all the V4 feature bits that the V5 filesystem format requires are
35  * correctly set.
36  */
37 static bool
38 xfs_sb_validate_v5_features(
39 	struct xfs_sb	*sbp)
40 {
41 	/* We must not have any unknown V4 feature bits set */
42 	if (sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS)
43 		return false;
44 
45 	/*
46 	 * The CRC bit is considered an invalid V4 flag, so we have to add it
47 	 * manually to the OKBITS mask.
48 	 */
49 	if (sbp->sb_features2 & ~(XFS_SB_VERSION2_OKBITS |
50 				  XFS_SB_VERSION2_CRCBIT))
51 		return false;
52 
53 	/* Now check all the required V4 feature flags are set. */
54 
55 #define V5_VERS_FLAGS	(XFS_SB_VERSION_NLINKBIT	| \
56 			XFS_SB_VERSION_ALIGNBIT		| \
57 			XFS_SB_VERSION_LOGV2BIT		| \
58 			XFS_SB_VERSION_EXTFLGBIT	| \
59 			XFS_SB_VERSION_DIRV2BIT		| \
60 			XFS_SB_VERSION_MOREBITSBIT)
61 
62 #define V5_FEAT_FLAGS	(XFS_SB_VERSION2_LAZYSBCOUNTBIT	| \
63 			XFS_SB_VERSION2_ATTR2BIT	| \
64 			XFS_SB_VERSION2_PROJID32BIT	| \
65 			XFS_SB_VERSION2_CRCBIT)
66 
67 	if ((sbp->sb_versionnum & V5_VERS_FLAGS) != V5_VERS_FLAGS)
68 		return false;
69 	if ((sbp->sb_features2 & V5_FEAT_FLAGS) != V5_FEAT_FLAGS)
70 		return false;
71 	return true;
72 }
73 
74 /*
75  * We current support XFS v5 formats with known features and v4 superblocks with
76  * at least V2 directories.
77  */
78 bool
79 xfs_sb_good_version(
80 	struct xfs_sb	*sbp)
81 {
82 	/*
83 	 * All v5 filesystems are supported, but we must check that all the
84 	 * required v4 feature flags are enabled correctly as the code checks
85 	 * those flags and not for v5 support.
86 	 */
87 	if (xfs_sb_is_v5(sbp))
88 		return xfs_sb_validate_v5_features(sbp);
89 
90 	/* versions prior to v4 are not supported */
91 	if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_4)
92 		return false;
93 
94 	/* We must not have any unknown v4 feature bits set */
95 	if ((sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS) ||
96 	    ((sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) &&
97 	     (sbp->sb_features2 & ~XFS_SB_VERSION2_OKBITS)))
98 		return false;
99 
100 	/* V4 filesystems need v2 directories and unwritten extents */
101 	if (!(sbp->sb_versionnum & XFS_SB_VERSION_DIRV2BIT))
102 		return false;
103 	if (!(sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT))
104 		return false;
105 
106 	/* It's a supported v4 filesystem */
107 	return true;
108 }
109 
110 uint64_t
111 xfs_sb_version_to_features(
112 	struct xfs_sb	*sbp)
113 {
114 	uint64_t	features = 0;
115 
116 	/* optional V4 features */
117 	if (sbp->sb_rblocks > 0)
118 		features |= XFS_FEAT_REALTIME;
119 	if (sbp->sb_versionnum & XFS_SB_VERSION_NLINKBIT)
120 		features |= XFS_FEAT_NLINK;
121 	if (sbp->sb_versionnum & XFS_SB_VERSION_ATTRBIT)
122 		features |= XFS_FEAT_ATTR;
123 	if (sbp->sb_versionnum & XFS_SB_VERSION_QUOTABIT)
124 		features |= XFS_FEAT_QUOTA;
125 	if (sbp->sb_versionnum & XFS_SB_VERSION_ALIGNBIT)
126 		features |= XFS_FEAT_ALIGN;
127 	if (sbp->sb_versionnum & XFS_SB_VERSION_LOGV2BIT)
128 		features |= XFS_FEAT_LOGV2;
129 	if (sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT)
130 		features |= XFS_FEAT_DALIGN;
131 	if (sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT)
132 		features |= XFS_FEAT_EXTFLG;
133 	if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT)
134 		features |= XFS_FEAT_SECTOR;
135 	if (sbp->sb_versionnum & XFS_SB_VERSION_BORGBIT)
136 		features |= XFS_FEAT_ASCIICI;
137 	if (sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) {
138 		if (sbp->sb_features2 & XFS_SB_VERSION2_LAZYSBCOUNTBIT)
139 			features |= XFS_FEAT_LAZYSBCOUNT;
140 		if (sbp->sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
141 			features |= XFS_FEAT_ATTR2;
142 		if (sbp->sb_features2 & XFS_SB_VERSION2_PROJID32BIT)
143 			features |= XFS_FEAT_PROJID32;
144 		if (sbp->sb_features2 & XFS_SB_VERSION2_FTYPE)
145 			features |= XFS_FEAT_FTYPE;
146 	}
147 
148 	if (!xfs_sb_is_v5(sbp))
149 		return features;
150 
151 	/* Always on V5 features */
152 	features |= XFS_FEAT_ALIGN | XFS_FEAT_LOGV2 | XFS_FEAT_EXTFLG |
153 		    XFS_FEAT_LAZYSBCOUNT | XFS_FEAT_ATTR2 | XFS_FEAT_PROJID32 |
154 		    XFS_FEAT_V3INODES | XFS_FEAT_CRC | XFS_FEAT_PQUOTINO;
155 
156 	/* Optional V5 features */
157 	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_FINOBT)
158 		features |= XFS_FEAT_FINOBT;
159 	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_RMAPBT)
160 		features |= XFS_FEAT_RMAPBT;
161 	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_REFLINK)
162 		features |= XFS_FEAT_REFLINK;
163 	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_INOBTCNT)
164 		features |= XFS_FEAT_INOBTCNT;
165 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_FTYPE)
166 		features |= XFS_FEAT_FTYPE;
167 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES)
168 		features |= XFS_FEAT_SPINODES;
169 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
170 		features |= XFS_FEAT_META_UUID;
171 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_BIGTIME)
172 		features |= XFS_FEAT_BIGTIME;
173 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR)
174 		features |= XFS_FEAT_NEEDSREPAIR;
175 	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NREXT64)
176 		features |= XFS_FEAT_NREXT64;
177 
178 	return features;
179 }
180 
181 /* Check all the superblock fields we care about when reading one in. */
182 STATIC int
183 xfs_validate_sb_read(
184 	struct xfs_mount	*mp,
185 	struct xfs_sb		*sbp)
186 {
187 	if (!xfs_sb_is_v5(sbp))
188 		return 0;
189 
190 	/*
191 	 * Version 5 superblock feature mask validation. Reject combinations
192 	 * the kernel cannot support up front before checking anything else.
193 	 */
194 	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
195 		xfs_warn(mp,
196 "Superblock has unknown compatible features (0x%x) enabled.",
197 			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
198 		xfs_warn(mp,
199 "Using a more recent kernel is recommended.");
200 	}
201 
202 	if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
203 		xfs_alert(mp,
204 "Superblock has unknown read-only compatible features (0x%x) enabled.",
205 			(sbp->sb_features_ro_compat &
206 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
207 		if (!xfs_is_readonly(mp)) {
208 			xfs_warn(mp,
209 "Attempted to mount read-only compatible filesystem read-write.");
210 			xfs_warn(mp,
211 "Filesystem can only be safely mounted read only.");
212 
213 			return -EINVAL;
214 		}
215 	}
216 	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
217 		xfs_warn(mp,
218 "Superblock has unknown incompatible features (0x%x) enabled.",
219 			(sbp->sb_features_incompat &
220 					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
221 		xfs_warn(mp,
222 "Filesystem cannot be safely mounted by this kernel.");
223 		return -EINVAL;
224 	}
225 
226 	return 0;
227 }
228 
229 /* Check all the superblock fields we care about when writing one out. */
230 STATIC int
231 xfs_validate_sb_write(
232 	struct xfs_mount	*mp,
233 	struct xfs_buf		*bp,
234 	struct xfs_sb		*sbp)
235 {
236 	/*
237 	 * Carry out additional sb summary counter sanity checks when we write
238 	 * the superblock.  We skip this in the read validator because there
239 	 * could be newer superblocks in the log and if the values are garbage
240 	 * even after replay we'll recalculate them at the end of log mount.
241 	 *
242 	 * mkfs has traditionally written zeroed counters to inprogress and
243 	 * secondary superblocks, so allow this usage to continue because
244 	 * we never read counters from such superblocks.
245 	 */
246 	if (xfs_buf_daddr(bp) == XFS_SB_DADDR && !sbp->sb_inprogress &&
247 	    (sbp->sb_fdblocks > sbp->sb_dblocks ||
248 	     !xfs_verify_icount(mp, sbp->sb_icount) ||
249 	     sbp->sb_ifree > sbp->sb_icount)) {
250 		xfs_warn(mp, "SB summary counter sanity check failed");
251 		return -EFSCORRUPTED;
252 	}
253 
254 	if (!xfs_sb_is_v5(sbp))
255 		return 0;
256 
257 	/*
258 	 * Version 5 superblock feature mask validation. Reject combinations
259 	 * the kernel cannot support since we checked for unsupported bits in
260 	 * the read verifier, which means that memory is corrupt.
261 	 */
262 	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
263 		xfs_warn(mp,
264 "Corruption detected in superblock compatible features (0x%x)!",
265 			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
266 		return -EFSCORRUPTED;
267 	}
268 
269 	if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
270 		xfs_alert(mp,
271 "Corruption detected in superblock read-only compatible features (0x%x)!",
272 			(sbp->sb_features_ro_compat &
273 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
274 		return -EFSCORRUPTED;
275 	}
276 	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
277 		xfs_warn(mp,
278 "Corruption detected in superblock incompatible features (0x%x)!",
279 			(sbp->sb_features_incompat &
280 					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
281 		return -EFSCORRUPTED;
282 	}
283 	if (xfs_sb_has_incompat_log_feature(sbp,
284 			XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
285 		xfs_warn(mp,
286 "Corruption detected in superblock incompatible log features (0x%x)!",
287 			(sbp->sb_features_log_incompat &
288 					XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
289 		return -EFSCORRUPTED;
290 	}
291 
292 	/*
293 	 * We can't read verify the sb LSN because the read verifier is called
294 	 * before the log is allocated and processed. We know the log is set up
295 	 * before write verifier calls, so check it here.
296 	 */
297 	if (!xfs_log_check_lsn(mp, sbp->sb_lsn))
298 		return -EFSCORRUPTED;
299 
300 	return 0;
301 }
302 
303 /* Check the validity of the SB. */
304 STATIC int
305 xfs_validate_sb_common(
306 	struct xfs_mount	*mp,
307 	struct xfs_buf		*bp,
308 	struct xfs_sb		*sbp)
309 {
310 	struct xfs_dsb		*dsb = bp->b_addr;
311 	uint32_t		agcount = 0;
312 	uint32_t		rem;
313 	bool			has_dalign;
314 
315 	if (!xfs_verify_magic(bp, dsb->sb_magicnum)) {
316 		xfs_warn(mp,
317 "Superblock has bad magic number 0x%x. Not an XFS filesystem?",
318 			be32_to_cpu(dsb->sb_magicnum));
319 		return -EWRONGFS;
320 	}
321 
322 	if (!xfs_sb_good_version(sbp)) {
323 		xfs_warn(mp,
324 "Superblock has unknown features enabled or corrupted feature masks.");
325 		return -EWRONGFS;
326 	}
327 
328 	/*
329 	 * Validate feature flags and state
330 	 */
331 	if (xfs_sb_is_v5(sbp)) {
332 		if (sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) {
333 			xfs_notice(mp,
334 "Block size (%u bytes) too small for Version 5 superblock (minimum %d bytes)",
335 				sbp->sb_blocksize, XFS_MIN_CRC_BLOCKSIZE);
336 			return -EFSCORRUPTED;
337 		}
338 
339 		/* V5 has a separate project quota inode */
340 		if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
341 			xfs_notice(mp,
342 			   "Version 5 of Super block has XFS_OQUOTA bits.");
343 			return -EFSCORRUPTED;
344 		}
345 
346 		/*
347 		 * Full inode chunks must be aligned to inode chunk size when
348 		 * sparse inodes are enabled to support the sparse chunk
349 		 * allocation algorithm and prevent overlapping inode records.
350 		 */
351 		if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES) {
352 			uint32_t	align;
353 
354 			align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
355 					>> sbp->sb_blocklog;
356 			if (sbp->sb_inoalignmt != align) {
357 				xfs_warn(mp,
358 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
359 					 sbp->sb_inoalignmt, align);
360 				return -EINVAL;
361 			}
362 		}
363 	} else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
364 				XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
365 			xfs_notice(mp,
366 "Superblock earlier than Version 5 has XFS_{P|G}QUOTA_{ENFD|CHKD} bits.");
367 			return -EFSCORRUPTED;
368 	}
369 
370 	if (unlikely(
371 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
372 		xfs_warn(mp,
373 		"filesystem is marked as having an external log; "
374 		"specify logdev on the mount command line.");
375 		return -EINVAL;
376 	}
377 
378 	if (unlikely(
379 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
380 		xfs_warn(mp,
381 		"filesystem is marked as having an internal log; "
382 		"do not specify logdev on the mount command line.");
383 		return -EINVAL;
384 	}
385 
386 	/* Compute agcount for this number of dblocks and agblocks */
387 	if (sbp->sb_agblocks) {
388 		agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem);
389 		if (rem)
390 			agcount++;
391 	}
392 
393 	/*
394 	 * More sanity checking.  Most of these were stolen directly from
395 	 * xfs_repair.
396 	 */
397 	if (unlikely(
398 	    sbp->sb_agcount <= 0					||
399 	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
400 	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
401 	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
402 	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
403 	    sbp->sb_sectsize != (1 << sbp->sb_sectlog)			||
404 	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
405 	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
406 	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
407 	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
408 	    sbp->sb_blocksize != (1 << sbp->sb_blocklog)		||
409 	    sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
410 	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
411 	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
412 	    sbp->sb_inodelog < XFS_DINODE_MIN_LOG			||
413 	    sbp->sb_inodelog > XFS_DINODE_MAX_LOG			||
414 	    sbp->sb_inodesize != (1 << sbp->sb_inodelog)		||
415 	    sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
416 	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES	||
417 	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES	||
418 	    sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1	||
419 	    agcount == 0 || agcount != sbp->sb_agcount			||
420 	    (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)	||
421 	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
422 	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
423 	    (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)	||
424 	    sbp->sb_dblocks == 0					||
425 	    sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)			||
426 	    sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp)			||
427 	    sbp->sb_shared_vn != 0)) {
428 		xfs_notice(mp, "SB sanity check failed");
429 		return -EFSCORRUPTED;
430 	}
431 
432 	/*
433 	 * Logs that are too large are not supported at all. Reject them
434 	 * outright. Logs that are too small are tolerated on v4 filesystems,
435 	 * but we can only check that when mounting the log. Hence we skip
436 	 * those checks here.
437 	 */
438 	if (sbp->sb_logblocks > XFS_MAX_LOG_BLOCKS) {
439 		xfs_notice(mp,
440 		"Log size 0x%x blocks too large, maximum size is 0x%llx blocks",
441 			 sbp->sb_logblocks, XFS_MAX_LOG_BLOCKS);
442 		return -EFSCORRUPTED;
443 	}
444 
445 	if (XFS_FSB_TO_B(mp, sbp->sb_logblocks) > XFS_MAX_LOG_BYTES) {
446 		xfs_warn(mp,
447 		"log size 0x%llx bytes too large, maximum size is 0x%llx bytes",
448 			 XFS_FSB_TO_B(mp, sbp->sb_logblocks),
449 			 XFS_MAX_LOG_BYTES);
450 		return -EFSCORRUPTED;
451 	}
452 
453 	/*
454 	 * Do not allow filesystems with corrupted log sector or stripe units to
455 	 * be mounted. We cannot safely size the iclogs or write to the log if
456 	 * the log stripe unit is not valid.
457 	 */
458 	if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT) {
459 		if (sbp->sb_logsectsize != (1U << sbp->sb_logsectlog)) {
460 			xfs_notice(mp,
461 			"log sector size in bytes/log2 (0x%x/0x%x) must match",
462 				sbp->sb_logsectsize, 1U << sbp->sb_logsectlog);
463 			return -EFSCORRUPTED;
464 		}
465 	} else if (sbp->sb_logsectsize || sbp->sb_logsectlog) {
466 		xfs_notice(mp,
467 		"log sector size in bytes/log2 (0x%x/0x%x) are not zero",
468 			sbp->sb_logsectsize, sbp->sb_logsectlog);
469 		return -EFSCORRUPTED;
470 	}
471 
472 	if (sbp->sb_logsunit > 1) {
473 		if (sbp->sb_logsunit % sbp->sb_blocksize) {
474 			xfs_notice(mp,
475 		"log stripe unit 0x%x bytes must be a multiple of block size",
476 				sbp->sb_logsunit);
477 			return -EFSCORRUPTED;
478 		}
479 		if (sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE) {
480 			xfs_notice(mp,
481 		"log stripe unit 0x%x bytes over maximum size (0x%x bytes)",
482 				sbp->sb_logsunit, XLOG_MAX_RECORD_BSIZE);
483 			return -EFSCORRUPTED;
484 		}
485 	}
486 
487 	/* Validate the realtime geometry; stolen from xfs_repair */
488 	if (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE ||
489 	    sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) {
490 		xfs_notice(mp,
491 			"realtime extent sanity check failed");
492 		return -EFSCORRUPTED;
493 	}
494 
495 	if (sbp->sb_rblocks == 0) {
496 		if (sbp->sb_rextents != 0 || sbp->sb_rbmblocks != 0 ||
497 		    sbp->sb_rextslog != 0 || sbp->sb_frextents != 0) {
498 			xfs_notice(mp,
499 				"realtime zeroed geometry check failed");
500 			return -EFSCORRUPTED;
501 		}
502 	} else {
503 		uint64_t	rexts;
504 		uint64_t	rbmblocks;
505 
506 		rexts = div_u64(sbp->sb_rblocks, sbp->sb_rextsize);
507 		rbmblocks = howmany_64(sbp->sb_rextents,
508 				       NBBY * sbp->sb_blocksize);
509 
510 		if (sbp->sb_rextents != rexts ||
511 		    sbp->sb_rextslog != xfs_highbit32(sbp->sb_rextents) ||
512 		    sbp->sb_rbmblocks != rbmblocks) {
513 			xfs_notice(mp,
514 				"realtime geometry sanity check failed");
515 			return -EFSCORRUPTED;
516 		}
517 	}
518 
519 	/*
520 	 * Either (sb_unit and !hasdalign) or (!sb_unit and hasdalign)
521 	 * would imply the image is corrupted.
522 	 */
523 	has_dalign = sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT;
524 	if (!!sbp->sb_unit ^ has_dalign) {
525 		xfs_notice(mp, "SB stripe alignment sanity check failed");
526 		return -EFSCORRUPTED;
527 	}
528 
529 	if (!xfs_validate_stripe_geometry(mp, XFS_FSB_TO_B(mp, sbp->sb_unit),
530 			XFS_FSB_TO_B(mp, sbp->sb_width), 0, false))
531 		return -EFSCORRUPTED;
532 
533 	/*
534 	 * Currently only very few inode sizes are supported.
535 	 */
536 	switch (sbp->sb_inodesize) {
537 	case 256:
538 	case 512:
539 	case 1024:
540 	case 2048:
541 		break;
542 	default:
543 		xfs_warn(mp, "inode size of %d bytes not supported",
544 				sbp->sb_inodesize);
545 		return -ENOSYS;
546 	}
547 
548 	return 0;
549 }
550 
551 void
552 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
553 {
554 	/*
555 	 * older mkfs doesn't initialize quota inodes to NULLFSINO. This
556 	 * leads to in-core values having two different values for a quota
557 	 * inode to be invalid: 0 and NULLFSINO. Change it to a single value
558 	 * NULLFSINO.
559 	 *
560 	 * Note that this change affect only the in-core values. These
561 	 * values are not written back to disk unless any quota information
562 	 * is written to the disk. Even in that case, sb_pquotino field is
563 	 * not written to disk unless the superblock supports pquotino.
564 	 */
565 	if (sbp->sb_uquotino == 0)
566 		sbp->sb_uquotino = NULLFSINO;
567 	if (sbp->sb_gquotino == 0)
568 		sbp->sb_gquotino = NULLFSINO;
569 	if (sbp->sb_pquotino == 0)
570 		sbp->sb_pquotino = NULLFSINO;
571 
572 	/*
573 	 * We need to do these manipilations only if we are working
574 	 * with an older version of on-disk superblock.
575 	 */
576 	if (xfs_sb_is_v5(sbp))
577 		return;
578 
579 	if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
580 		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
581 					XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
582 	if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
583 		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
584 					XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
585 	sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
586 
587 	if (sbp->sb_qflags & XFS_PQUOTA_ACCT &&
588 	    sbp->sb_gquotino != NULLFSINO)  {
589 		/*
590 		 * In older version of superblock, on-disk superblock only
591 		 * has sb_gquotino, and in-core superblock has both sb_gquotino
592 		 * and sb_pquotino. But, only one of them is supported at any
593 		 * point of time. So, if PQUOTA is set in disk superblock,
594 		 * copy over sb_gquotino to sb_pquotino.  The NULLFSINO test
595 		 * above is to make sure we don't do this twice and wipe them
596 		 * both out!
597 		 */
598 		sbp->sb_pquotino = sbp->sb_gquotino;
599 		sbp->sb_gquotino = NULLFSINO;
600 	}
601 }
602 
603 static void
604 __xfs_sb_from_disk(
605 	struct xfs_sb	*to,
606 	struct xfs_dsb	*from,
607 	bool		convert_xquota)
608 {
609 	to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
610 	to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
611 	to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
612 	to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
613 	to->sb_rextents = be64_to_cpu(from->sb_rextents);
614 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
615 	to->sb_logstart = be64_to_cpu(from->sb_logstart);
616 	to->sb_rootino = be64_to_cpu(from->sb_rootino);
617 	to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
618 	to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
619 	to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
620 	to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
621 	to->sb_agcount = be32_to_cpu(from->sb_agcount);
622 	to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
623 	to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
624 	to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
625 	to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
626 	to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
627 	to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
628 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
629 	to->sb_blocklog = from->sb_blocklog;
630 	to->sb_sectlog = from->sb_sectlog;
631 	to->sb_inodelog = from->sb_inodelog;
632 	to->sb_inopblog = from->sb_inopblog;
633 	to->sb_agblklog = from->sb_agblklog;
634 	to->sb_rextslog = from->sb_rextslog;
635 	to->sb_inprogress = from->sb_inprogress;
636 	to->sb_imax_pct = from->sb_imax_pct;
637 	to->sb_icount = be64_to_cpu(from->sb_icount);
638 	to->sb_ifree = be64_to_cpu(from->sb_ifree);
639 	to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
640 	to->sb_frextents = be64_to_cpu(from->sb_frextents);
641 	to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
642 	to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
643 	to->sb_qflags = be16_to_cpu(from->sb_qflags);
644 	to->sb_flags = from->sb_flags;
645 	to->sb_shared_vn = from->sb_shared_vn;
646 	to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
647 	to->sb_unit = be32_to_cpu(from->sb_unit);
648 	to->sb_width = be32_to_cpu(from->sb_width);
649 	to->sb_dirblklog = from->sb_dirblklog;
650 	to->sb_logsectlog = from->sb_logsectlog;
651 	to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
652 	to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
653 	to->sb_features2 = be32_to_cpu(from->sb_features2);
654 	to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
655 	to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
656 	to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
657 	to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
658 	to->sb_features_log_incompat =
659 				be32_to_cpu(from->sb_features_log_incompat);
660 	/* crc is only used on disk, not in memory; just init to 0 here. */
661 	to->sb_crc = 0;
662 	to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
663 	to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
664 	to->sb_lsn = be64_to_cpu(from->sb_lsn);
665 	/*
666 	 * sb_meta_uuid is only on disk if it differs from sb_uuid and the
667 	 * feature flag is set; if not set we keep it only in memory.
668 	 */
669 	if (xfs_sb_is_v5(to) &&
670 	    (to->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID))
671 		uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
672 	else
673 		uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
674 	/* Convert on-disk flags to in-memory flags? */
675 	if (convert_xquota)
676 		xfs_sb_quota_from_disk(to);
677 }
678 
679 void
680 xfs_sb_from_disk(
681 	struct xfs_sb	*to,
682 	struct xfs_dsb	*from)
683 {
684 	__xfs_sb_from_disk(to, from, true);
685 }
686 
687 static void
688 xfs_sb_quota_to_disk(
689 	struct xfs_dsb	*to,
690 	struct xfs_sb	*from)
691 {
692 	uint16_t	qflags = from->sb_qflags;
693 
694 	to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
695 
696 	/*
697 	 * The in-memory superblock quota state matches the v5 on-disk format so
698 	 * just write them out and return
699 	 */
700 	if (xfs_sb_is_v5(from)) {
701 		to->sb_qflags = cpu_to_be16(from->sb_qflags);
702 		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
703 		to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
704 		return;
705 	}
706 
707 	/*
708 	 * For older superblocks (v4), the in-core version of sb_qflags do not
709 	 * have XFS_OQUOTA_* flags, whereas the on-disk version does.  So,
710 	 * convert incore XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
711 	 */
712 	qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
713 			XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
714 
715 	if (from->sb_qflags &
716 			(XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
717 		qflags |= XFS_OQUOTA_ENFD;
718 	if (from->sb_qflags &
719 			(XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
720 		qflags |= XFS_OQUOTA_CHKD;
721 	to->sb_qflags = cpu_to_be16(qflags);
722 
723 	/*
724 	 * GQUOTINO and PQUOTINO cannot be used together in versions
725 	 * of superblock that do not have pquotino. from->sb_flags
726 	 * tells us which quota is active and should be copied to
727 	 * disk. If neither are active, we should NULL the inode.
728 	 *
729 	 * In all cases, the separate pquotino must remain 0 because it
730 	 * is beyond the "end" of the valid non-pquotino superblock.
731 	 */
732 	if (from->sb_qflags & XFS_GQUOTA_ACCT)
733 		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
734 	else if (from->sb_qflags & XFS_PQUOTA_ACCT)
735 		to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
736 	else {
737 		/*
738 		 * We can't rely on just the fields being logged to tell us
739 		 * that it is safe to write NULLFSINO - we should only do that
740 		 * if quotas are not actually enabled. Hence only write
741 		 * NULLFSINO if both in-core quota inodes are NULL.
742 		 */
743 		if (from->sb_gquotino == NULLFSINO &&
744 		    from->sb_pquotino == NULLFSINO)
745 			to->sb_gquotino = cpu_to_be64(NULLFSINO);
746 	}
747 
748 	to->sb_pquotino = 0;
749 }
750 
751 void
752 xfs_sb_to_disk(
753 	struct xfs_dsb	*to,
754 	struct xfs_sb	*from)
755 {
756 	xfs_sb_quota_to_disk(to, from);
757 
758 	to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
759 	to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
760 	to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
761 	to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
762 	to->sb_rextents = cpu_to_be64(from->sb_rextents);
763 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
764 	to->sb_logstart = cpu_to_be64(from->sb_logstart);
765 	to->sb_rootino = cpu_to_be64(from->sb_rootino);
766 	to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
767 	to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
768 	to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
769 	to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
770 	to->sb_agcount = cpu_to_be32(from->sb_agcount);
771 	to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
772 	to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
773 	to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
774 	to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
775 	to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
776 	to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
777 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
778 	to->sb_blocklog = from->sb_blocklog;
779 	to->sb_sectlog = from->sb_sectlog;
780 	to->sb_inodelog = from->sb_inodelog;
781 	to->sb_inopblog = from->sb_inopblog;
782 	to->sb_agblklog = from->sb_agblklog;
783 	to->sb_rextslog = from->sb_rextslog;
784 	to->sb_inprogress = from->sb_inprogress;
785 	to->sb_imax_pct = from->sb_imax_pct;
786 	to->sb_icount = cpu_to_be64(from->sb_icount);
787 	to->sb_ifree = cpu_to_be64(from->sb_ifree);
788 	to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
789 	to->sb_frextents = cpu_to_be64(from->sb_frextents);
790 
791 	to->sb_flags = from->sb_flags;
792 	to->sb_shared_vn = from->sb_shared_vn;
793 	to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
794 	to->sb_unit = cpu_to_be32(from->sb_unit);
795 	to->sb_width = cpu_to_be32(from->sb_width);
796 	to->sb_dirblklog = from->sb_dirblklog;
797 	to->sb_logsectlog = from->sb_logsectlog;
798 	to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
799 	to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
800 
801 	/*
802 	 * We need to ensure that bad_features2 always matches features2.
803 	 * Hence we enforce that here rather than having to remember to do it
804 	 * everywhere else that updates features2.
805 	 */
806 	from->sb_bad_features2 = from->sb_features2;
807 	to->sb_features2 = cpu_to_be32(from->sb_features2);
808 	to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
809 
810 	if (!xfs_sb_is_v5(from))
811 		return;
812 
813 	to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
814 	to->sb_features_ro_compat =
815 			cpu_to_be32(from->sb_features_ro_compat);
816 	to->sb_features_incompat =
817 			cpu_to_be32(from->sb_features_incompat);
818 	to->sb_features_log_incompat =
819 			cpu_to_be32(from->sb_features_log_incompat);
820 	to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
821 	to->sb_lsn = cpu_to_be64(from->sb_lsn);
822 	if (from->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
823 		uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
824 }
825 
826 /*
827  * If the superblock has the CRC feature bit set or the CRC field is non-null,
828  * check that the CRC is valid.  We check the CRC field is non-null because a
829  * single bit error could clear the feature bit and unused parts of the
830  * superblock are supposed to be zero. Hence a non-null crc field indicates that
831  * we've potentially lost a feature bit and we should check it anyway.
832  *
833  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
834  * last field in V4 secondary superblocks.  So for secondary superblocks,
835  * we are more forgiving, and ignore CRC failures if the primary doesn't
836  * indicate that the fs version is V5.
837  */
838 static void
839 xfs_sb_read_verify(
840 	struct xfs_buf		*bp)
841 {
842 	struct xfs_sb		sb;
843 	struct xfs_mount	*mp = bp->b_mount;
844 	struct xfs_dsb		*dsb = bp->b_addr;
845 	int			error;
846 
847 	/*
848 	 * open code the version check to avoid needing to convert the entire
849 	 * superblock from disk order just to check the version number
850 	 */
851 	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
852 	    (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
853 						XFS_SB_VERSION_5) ||
854 	     dsb->sb_crc != 0)) {
855 
856 		if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
857 			/* Only fail bad secondaries on a known V5 filesystem */
858 			if (xfs_buf_daddr(bp) == XFS_SB_DADDR ||
859 			    xfs_has_crc(mp)) {
860 				error = -EFSBADCRC;
861 				goto out_error;
862 			}
863 		}
864 	}
865 
866 	/*
867 	 * Check all the superblock fields.  Don't byteswap the xquota flags
868 	 * because _verify_common checks the on-disk values.
869 	 */
870 	__xfs_sb_from_disk(&sb, dsb, false);
871 	error = xfs_validate_sb_common(mp, bp, &sb);
872 	if (error)
873 		goto out_error;
874 	error = xfs_validate_sb_read(mp, &sb);
875 
876 out_error:
877 	if (error == -EFSCORRUPTED || error == -EFSBADCRC)
878 		xfs_verifier_error(bp, error, __this_address);
879 	else if (error)
880 		xfs_buf_ioerror(bp, error);
881 }
882 
883 /*
884  * We may be probed for a filesystem match, so we may not want to emit
885  * messages when the superblock buffer is not actually an XFS superblock.
886  * If we find an XFS superblock, then run a normal, noisy mount because we are
887  * really going to mount it and want to know about errors.
888  */
889 static void
890 xfs_sb_quiet_read_verify(
891 	struct xfs_buf	*bp)
892 {
893 	struct xfs_dsb	*dsb = bp->b_addr;
894 
895 	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
896 		/* XFS filesystem, verify noisily! */
897 		xfs_sb_read_verify(bp);
898 		return;
899 	}
900 	/* quietly fail */
901 	xfs_buf_ioerror(bp, -EWRONGFS);
902 }
903 
904 static void
905 xfs_sb_write_verify(
906 	struct xfs_buf		*bp)
907 {
908 	struct xfs_sb		sb;
909 	struct xfs_mount	*mp = bp->b_mount;
910 	struct xfs_buf_log_item	*bip = bp->b_log_item;
911 	struct xfs_dsb		*dsb = bp->b_addr;
912 	int			error;
913 
914 	/*
915 	 * Check all the superblock fields.  Don't byteswap the xquota flags
916 	 * because _verify_common checks the on-disk values.
917 	 */
918 	__xfs_sb_from_disk(&sb, dsb, false);
919 	error = xfs_validate_sb_common(mp, bp, &sb);
920 	if (error)
921 		goto out_error;
922 	error = xfs_validate_sb_write(mp, bp, &sb);
923 	if (error)
924 		goto out_error;
925 
926 	if (!xfs_sb_is_v5(&sb))
927 		return;
928 
929 	if (bip)
930 		dsb->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
931 
932 	xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
933 	return;
934 
935 out_error:
936 	xfs_verifier_error(bp, error, __this_address);
937 }
938 
939 const struct xfs_buf_ops xfs_sb_buf_ops = {
940 	.name = "xfs_sb",
941 	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
942 	.verify_read = xfs_sb_read_verify,
943 	.verify_write = xfs_sb_write_verify,
944 };
945 
946 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
947 	.name = "xfs_sb_quiet",
948 	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
949 	.verify_read = xfs_sb_quiet_read_verify,
950 	.verify_write = xfs_sb_write_verify,
951 };
952 
953 /*
954  * xfs_mount_common
955  *
956  * Mount initialization code establishing various mount
957  * fields from the superblock associated with the given
958  * mount structure.
959  *
960  * Inode geometry are calculated in xfs_ialloc_setup_geometry.
961  */
962 void
963 xfs_sb_mount_common(
964 	struct xfs_mount	*mp,
965 	struct xfs_sb		*sbp)
966 {
967 	mp->m_agfrotor = 0;
968 	atomic_set(&mp->m_agirotor, 0);
969 	mp->m_maxagi = mp->m_sb.sb_agcount;
970 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
971 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
972 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
973 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
974 	mp->m_blockmask = sbp->sb_blocksize - 1;
975 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
976 	mp->m_blockwmask = mp->m_blockwsize - 1;
977 
978 	mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
979 	mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
980 	mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
981 	mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
982 
983 	mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
984 	mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
985 	mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
986 	mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
987 
988 	mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 1);
989 	mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 0);
990 	mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
991 	mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
992 
993 	mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, true);
994 	mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, false);
995 	mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
996 	mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
997 
998 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
999 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
1000 	mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp);
1001 }
1002 
1003 /*
1004  * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
1005  * into the superblock buffer to be logged.  It does not provide the higher
1006  * level of locking that is needed to protect the in-core superblock from
1007  * concurrent access.
1008  */
1009 void
1010 xfs_log_sb(
1011 	struct xfs_trans	*tp)
1012 {
1013 	struct xfs_mount	*mp = tp->t_mountp;
1014 	struct xfs_buf		*bp = xfs_trans_getsb(tp);
1015 
1016 	/*
1017 	 * Lazy sb counters don't update the in-core superblock so do that now.
1018 	 * If this is at unmount, the counters will be exactly correct, but at
1019 	 * any other time they will only be ballpark correct because of
1020 	 * reservations that have been taken out percpu counters. If we have an
1021 	 * unclean shutdown, this will be corrected by log recovery rebuilding
1022 	 * the counters from the AGF block counts.
1023 	 *
1024 	 * Do not update sb_frextents here because it is not part of the lazy
1025 	 * sb counters, despite having a percpu counter. It is always kept
1026 	 * consistent with the ondisk rtbitmap by xfs_trans_apply_sb_deltas()
1027 	 * and hence we don't need have to update it here.
1028 	 */
1029 	if (xfs_has_lazysbcount(mp)) {
1030 		mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
1031 		mp->m_sb.sb_ifree = min_t(uint64_t,
1032 				percpu_counter_sum(&mp->m_ifree),
1033 				mp->m_sb.sb_icount);
1034 		mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1035 	}
1036 
1037 	xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1038 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
1039 	xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
1040 }
1041 
1042 /*
1043  * xfs_sync_sb
1044  *
1045  * Sync the superblock to disk.
1046  *
1047  * Note that the caller is responsible for checking the frozen state of the
1048  * filesystem. This procedure uses the non-blocking transaction allocator and
1049  * thus will allow modifications to a frozen fs. This is required because this
1050  * code can be called during the process of freezing where use of the high-level
1051  * allocator would deadlock.
1052  */
1053 int
1054 xfs_sync_sb(
1055 	struct xfs_mount	*mp,
1056 	bool			wait)
1057 {
1058 	struct xfs_trans	*tp;
1059 	int			error;
1060 
1061 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0,
1062 			XFS_TRANS_NO_WRITECOUNT, &tp);
1063 	if (error)
1064 		return error;
1065 
1066 	xfs_log_sb(tp);
1067 	if (wait)
1068 		xfs_trans_set_sync(tp);
1069 	return xfs_trans_commit(tp);
1070 }
1071 
1072 /*
1073  * Update all the secondary superblocks to match the new state of the primary.
1074  * Because we are completely overwriting all the existing fields in the
1075  * secondary superblock buffers, there is no need to read them in from disk.
1076  * Just get a new buffer, stamp it and write it.
1077  *
1078  * The sb buffers need to be cached here so that we serialise against other
1079  * operations that access the secondary superblocks, but we don't want to keep
1080  * them in memory once it is written so we mark it as a one-shot buffer.
1081  */
1082 int
1083 xfs_update_secondary_sbs(
1084 	struct xfs_mount	*mp)
1085 {
1086 	struct xfs_perag	*pag;
1087 	xfs_agnumber_t		agno = 1;
1088 	int			saved_error = 0;
1089 	int			error = 0;
1090 	LIST_HEAD		(buffer_list);
1091 
1092 	/* update secondary superblocks. */
1093 	for_each_perag_from(mp, agno, pag) {
1094 		struct xfs_buf		*bp;
1095 
1096 		error = xfs_buf_get(mp->m_ddev_targp,
1097 				 XFS_AG_DADDR(mp, pag->pag_agno, XFS_SB_DADDR),
1098 				 XFS_FSS_TO_BB(mp, 1), &bp);
1099 		/*
1100 		 * If we get an error reading or writing alternate superblocks,
1101 		 * continue.  xfs_repair chooses the "best" superblock based
1102 		 * on most matches; if we break early, we'll leave more
1103 		 * superblocks un-updated than updated, and xfs_repair may
1104 		 * pick them over the properly-updated primary.
1105 		 */
1106 		if (error) {
1107 			xfs_warn(mp,
1108 		"error allocating secondary superblock for ag %d",
1109 				pag->pag_agno);
1110 			if (!saved_error)
1111 				saved_error = error;
1112 			continue;
1113 		}
1114 
1115 		bp->b_ops = &xfs_sb_buf_ops;
1116 		xfs_buf_oneshot(bp);
1117 		xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
1118 		xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1119 		xfs_buf_delwri_queue(bp, &buffer_list);
1120 		xfs_buf_relse(bp);
1121 
1122 		/* don't hold too many buffers at once */
1123 		if (agno % 16)
1124 			continue;
1125 
1126 		error = xfs_buf_delwri_submit(&buffer_list);
1127 		if (error) {
1128 			xfs_warn(mp,
1129 		"write error %d updating a secondary superblock near ag %d",
1130 				error, pag->pag_agno);
1131 			if (!saved_error)
1132 				saved_error = error;
1133 			continue;
1134 		}
1135 	}
1136 	error = xfs_buf_delwri_submit(&buffer_list);
1137 	if (error) {
1138 		xfs_warn(mp,
1139 		"write error %d updating a secondary superblock near ag %d",
1140 			error, agno);
1141 	}
1142 
1143 	return saved_error ? saved_error : error;
1144 }
1145 
1146 /*
1147  * Same behavior as xfs_sync_sb, except that it is always synchronous and it
1148  * also writes the superblock buffer to disk sector 0 immediately.
1149  */
1150 int
1151 xfs_sync_sb_buf(
1152 	struct xfs_mount	*mp)
1153 {
1154 	struct xfs_trans	*tp;
1155 	struct xfs_buf		*bp;
1156 	int			error;
1157 
1158 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp);
1159 	if (error)
1160 		return error;
1161 
1162 	bp = xfs_trans_getsb(tp);
1163 	xfs_log_sb(tp);
1164 	xfs_trans_bhold(tp, bp);
1165 	xfs_trans_set_sync(tp);
1166 	error = xfs_trans_commit(tp);
1167 	if (error)
1168 		goto out;
1169 	/*
1170 	 * write out the sb buffer to get the changes to disk
1171 	 */
1172 	error = xfs_bwrite(bp);
1173 out:
1174 	xfs_buf_relse(bp);
1175 	return error;
1176 }
1177 
1178 void
1179 xfs_fs_geometry(
1180 	struct xfs_mount	*mp,
1181 	struct xfs_fsop_geom	*geo,
1182 	int			struct_version)
1183 {
1184 	struct xfs_sb		*sbp = &mp->m_sb;
1185 
1186 	memset(geo, 0, sizeof(struct xfs_fsop_geom));
1187 
1188 	geo->blocksize = sbp->sb_blocksize;
1189 	geo->rtextsize = sbp->sb_rextsize;
1190 	geo->agblocks = sbp->sb_agblocks;
1191 	geo->agcount = sbp->sb_agcount;
1192 	geo->logblocks = sbp->sb_logblocks;
1193 	geo->sectsize = sbp->sb_sectsize;
1194 	geo->inodesize = sbp->sb_inodesize;
1195 	geo->imaxpct = sbp->sb_imax_pct;
1196 	geo->datablocks = sbp->sb_dblocks;
1197 	geo->rtblocks = sbp->sb_rblocks;
1198 	geo->rtextents = sbp->sb_rextents;
1199 	geo->logstart = sbp->sb_logstart;
1200 	BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid));
1201 	memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid));
1202 
1203 	if (struct_version < 2)
1204 		return;
1205 
1206 	geo->sunit = sbp->sb_unit;
1207 	geo->swidth = sbp->sb_width;
1208 
1209 	if (struct_version < 3)
1210 		return;
1211 
1212 	geo->version = XFS_FSOP_GEOM_VERSION;
1213 	geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
1214 		     XFS_FSOP_GEOM_FLAGS_DIRV2 |
1215 		     XFS_FSOP_GEOM_FLAGS_EXTFLG;
1216 	if (xfs_has_attr(mp))
1217 		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR;
1218 	if (xfs_has_quota(mp))
1219 		geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA;
1220 	if (xfs_has_align(mp))
1221 		geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN;
1222 	if (xfs_has_dalign(mp))
1223 		geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN;
1224 	if (xfs_has_asciici(mp))
1225 		geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI;
1226 	if (xfs_has_lazysbcount(mp))
1227 		geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB;
1228 	if (xfs_has_attr2(mp))
1229 		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2;
1230 	if (xfs_has_projid32(mp))
1231 		geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32;
1232 	if (xfs_has_crc(mp))
1233 		geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB;
1234 	if (xfs_has_ftype(mp))
1235 		geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE;
1236 	if (xfs_has_finobt(mp))
1237 		geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT;
1238 	if (xfs_has_sparseinodes(mp))
1239 		geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES;
1240 	if (xfs_has_rmapbt(mp))
1241 		geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT;
1242 	if (xfs_has_reflink(mp))
1243 		geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK;
1244 	if (xfs_has_bigtime(mp))
1245 		geo->flags |= XFS_FSOP_GEOM_FLAGS_BIGTIME;
1246 	if (xfs_has_inobtcounts(mp))
1247 		geo->flags |= XFS_FSOP_GEOM_FLAGS_INOBTCNT;
1248 	if (xfs_has_sector(mp)) {
1249 		geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR;
1250 		geo->logsectsize = sbp->sb_logsectsize;
1251 	} else {
1252 		geo->logsectsize = BBSIZE;
1253 	}
1254 	if (xfs_has_large_extent_counts(mp))
1255 		geo->flags |= XFS_FSOP_GEOM_FLAGS_NREXT64;
1256 	geo->rtsectsize = sbp->sb_blocksize;
1257 	geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp);
1258 
1259 	if (struct_version < 4)
1260 		return;
1261 
1262 	if (xfs_has_logv2(mp))
1263 		geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2;
1264 
1265 	geo->logsunit = sbp->sb_logsunit;
1266 
1267 	if (struct_version < 5)
1268 		return;
1269 
1270 	geo->version = XFS_FSOP_GEOM_VERSION_V5;
1271 }
1272 
1273 /* Read a secondary superblock. */
1274 int
1275 xfs_sb_read_secondary(
1276 	struct xfs_mount	*mp,
1277 	struct xfs_trans	*tp,
1278 	xfs_agnumber_t		agno,
1279 	struct xfs_buf		**bpp)
1280 {
1281 	struct xfs_buf		*bp;
1282 	int			error;
1283 
1284 	ASSERT(agno != 0 && agno != NULLAGNUMBER);
1285 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1286 			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1287 			XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops);
1288 	if (error)
1289 		return error;
1290 	xfs_buf_set_ref(bp, XFS_SSB_REF);
1291 	*bpp = bp;
1292 	return 0;
1293 }
1294 
1295 /* Get an uninitialised secondary superblock buffer. */
1296 int
1297 xfs_sb_get_secondary(
1298 	struct xfs_mount	*mp,
1299 	struct xfs_trans	*tp,
1300 	xfs_agnumber_t		agno,
1301 	struct xfs_buf		**bpp)
1302 {
1303 	struct xfs_buf		*bp;
1304 	int			error;
1305 
1306 	ASSERT(agno != 0 && agno != NULLAGNUMBER);
1307 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1308 			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1309 			XFS_FSS_TO_BB(mp, 1), 0, &bp);
1310 	if (error)
1311 		return error;
1312 	bp->b_ops = &xfs_sb_buf_ops;
1313 	xfs_buf_oneshot(bp);
1314 	*bpp = bp;
1315 	return 0;
1316 }
1317 
1318 /*
1319  * sunit, swidth, sectorsize(optional with 0) should be all in bytes,
1320  * so users won't be confused by values in error messages.
1321  */
1322 bool
1323 xfs_validate_stripe_geometry(
1324 	struct xfs_mount	*mp,
1325 	__s64			sunit,
1326 	__s64			swidth,
1327 	int			sectorsize,
1328 	bool			silent)
1329 {
1330 	if (swidth > INT_MAX) {
1331 		if (!silent)
1332 			xfs_notice(mp,
1333 "stripe width (%lld) is too large", swidth);
1334 		return false;
1335 	}
1336 
1337 	if (sunit > swidth) {
1338 		if (!silent)
1339 			xfs_notice(mp,
1340 "stripe unit (%lld) is larger than the stripe width (%lld)", sunit, swidth);
1341 		return false;
1342 	}
1343 
1344 	if (sectorsize && (int)sunit % sectorsize) {
1345 		if (!silent)
1346 			xfs_notice(mp,
1347 "stripe unit (%lld) must be a multiple of the sector size (%d)",
1348 				   sunit, sectorsize);
1349 		return false;
1350 	}
1351 
1352 	if (sunit && !swidth) {
1353 		if (!silent)
1354 			xfs_notice(mp,
1355 "invalid stripe unit (%lld) and stripe width of 0", sunit);
1356 		return false;
1357 	}
1358 
1359 	if (!sunit && swidth) {
1360 		if (!silent)
1361 			xfs_notice(mp,
1362 "invalid stripe width (%lld) and stripe unit of 0", swidth);
1363 		return false;
1364 	}
1365 
1366 	if (sunit && (int)swidth % (int)sunit) {
1367 		if (!silent)
1368 			xfs_notice(mp,
1369 "stripe width (%lld) must be a multiple of the stripe unit (%lld)",
1370 				   swidth, sunit);
1371 		return false;
1372 	}
1373 	return true;
1374 }
1375