xref: /openbmc/linux/fs/xfs/libxfs/xfs_sb.c (revision 9a8f3203)
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_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_alloc.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_cksum.h"
22 #include "xfs_trans.h"
23 #include "xfs_buf_item.h"
24 #include "xfs_bmap_btree.h"
25 #include "xfs_alloc_btree.h"
26 #include "xfs_ialloc_btree.h"
27 #include "xfs_log.h"
28 #include "xfs_rmap_btree.h"
29 #include "xfs_bmap.h"
30 #include "xfs_refcount_btree.h"
31 #include "xfs_da_format.h"
32 #include "xfs_da_btree.h"
33 
34 /*
35  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
36  */
37 
38 /*
39  * Reference counting access wrappers to the perag structures.
40  * Because we never free per-ag structures, the only thing we
41  * have to protect against changes is the tree structure itself.
42  */
43 struct xfs_perag *
44 xfs_perag_get(
45 	struct xfs_mount	*mp,
46 	xfs_agnumber_t		agno)
47 {
48 	struct xfs_perag	*pag;
49 	int			ref = 0;
50 
51 	rcu_read_lock();
52 	pag = radix_tree_lookup(&mp->m_perag_tree, agno);
53 	if (pag) {
54 		ASSERT(atomic_read(&pag->pag_ref) >= 0);
55 		ref = atomic_inc_return(&pag->pag_ref);
56 	}
57 	rcu_read_unlock();
58 	trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
59 	return pag;
60 }
61 
62 /*
63  * search from @first to find the next perag with the given tag set.
64  */
65 struct xfs_perag *
66 xfs_perag_get_tag(
67 	struct xfs_mount	*mp,
68 	xfs_agnumber_t		first,
69 	int			tag)
70 {
71 	struct xfs_perag	*pag;
72 	int			found;
73 	int			ref;
74 
75 	rcu_read_lock();
76 	found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
77 					(void **)&pag, first, 1, tag);
78 	if (found <= 0) {
79 		rcu_read_unlock();
80 		return NULL;
81 	}
82 	ref = atomic_inc_return(&pag->pag_ref);
83 	rcu_read_unlock();
84 	trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
85 	return pag;
86 }
87 
88 void
89 xfs_perag_put(
90 	struct xfs_perag	*pag)
91 {
92 	int	ref;
93 
94 	ASSERT(atomic_read(&pag->pag_ref) > 0);
95 	ref = atomic_dec_return(&pag->pag_ref);
96 	trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
97 }
98 
99 /* Check all the superblock fields we care about when reading one in. */
100 STATIC int
101 xfs_validate_sb_read(
102 	struct xfs_mount	*mp,
103 	struct xfs_sb		*sbp)
104 {
105 	if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_5)
106 		return 0;
107 
108 	/*
109 	 * Version 5 superblock feature mask validation. Reject combinations
110 	 * the kernel cannot support up front before checking anything else.
111 	 */
112 	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
113 		xfs_warn(mp,
114 "Superblock has unknown compatible features (0x%x) enabled.",
115 			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
116 		xfs_warn(mp,
117 "Using a more recent kernel is recommended.");
118 	}
119 
120 	if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
121 		xfs_alert(mp,
122 "Superblock has unknown read-only compatible features (0x%x) enabled.",
123 			(sbp->sb_features_ro_compat &
124 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
125 		if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
126 			xfs_warn(mp,
127 "Attempted to mount read-only compatible filesystem read-write.");
128 			xfs_warn(mp,
129 "Filesystem can only be safely mounted read only.");
130 
131 			return -EINVAL;
132 		}
133 	}
134 	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
135 		xfs_warn(mp,
136 "Superblock has unknown incompatible features (0x%x) enabled.",
137 			(sbp->sb_features_incompat &
138 					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
139 		xfs_warn(mp,
140 "Filesystem cannot be safely mounted by this kernel.");
141 		return -EINVAL;
142 	}
143 
144 	return 0;
145 }
146 
147 /* Check all the superblock fields we care about when writing one out. */
148 STATIC int
149 xfs_validate_sb_write(
150 	struct xfs_mount	*mp,
151 	struct xfs_buf		*bp,
152 	struct xfs_sb		*sbp)
153 {
154 	/*
155 	 * Carry out additional sb summary counter sanity checks when we write
156 	 * the superblock.  We skip this in the read validator because there
157 	 * could be newer superblocks in the log and if the values are garbage
158 	 * even after replay we'll recalculate them at the end of log mount.
159 	 *
160 	 * mkfs has traditionally written zeroed counters to inprogress and
161 	 * secondary superblocks, so allow this usage to continue because
162 	 * we never read counters from such superblocks.
163 	 */
164 	if (XFS_BUF_ADDR(bp) == XFS_SB_DADDR && !sbp->sb_inprogress &&
165 	    (sbp->sb_fdblocks > sbp->sb_dblocks ||
166 	     !xfs_verify_icount(mp, sbp->sb_icount) ||
167 	     sbp->sb_ifree > sbp->sb_icount)) {
168 		xfs_warn(mp, "SB summary counter sanity check failed");
169 		return -EFSCORRUPTED;
170 	}
171 
172 	if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_5)
173 		return 0;
174 
175 	/*
176 	 * Version 5 superblock feature mask validation. Reject combinations
177 	 * the kernel cannot support since we checked for unsupported bits in
178 	 * the read verifier, which means that memory is corrupt.
179 	 */
180 	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
181 		xfs_warn(mp,
182 "Corruption detected in superblock compatible features (0x%x)!",
183 			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
184 		return -EFSCORRUPTED;
185 	}
186 
187 	if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
188 		xfs_alert(mp,
189 "Corruption detected in superblock read-only compatible features (0x%x)!",
190 			(sbp->sb_features_ro_compat &
191 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
192 		return -EFSCORRUPTED;
193 	}
194 	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
195 		xfs_warn(mp,
196 "Corruption detected in superblock incompatible features (0x%x)!",
197 			(sbp->sb_features_incompat &
198 					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
199 		return -EFSCORRUPTED;
200 	}
201 	if (xfs_sb_has_incompat_log_feature(sbp,
202 			XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
203 		xfs_warn(mp,
204 "Corruption detected in superblock incompatible log features (0x%x)!",
205 			(sbp->sb_features_log_incompat &
206 					XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
207 		return -EFSCORRUPTED;
208 	}
209 
210 	/*
211 	 * We can't read verify the sb LSN because the read verifier is called
212 	 * before the log is allocated and processed. We know the log is set up
213 	 * before write verifier calls, so check it here.
214 	 */
215 	if (!xfs_log_check_lsn(mp, sbp->sb_lsn))
216 		return -EFSCORRUPTED;
217 
218 	return 0;
219 }
220 
221 /* Check the validity of the SB. */
222 STATIC int
223 xfs_validate_sb_common(
224 	struct xfs_mount	*mp,
225 	struct xfs_buf		*bp,
226 	struct xfs_sb		*sbp)
227 {
228 	struct xfs_dsb		*dsb = XFS_BUF_TO_SBP(bp);
229 	uint32_t		agcount = 0;
230 	uint32_t		rem;
231 
232 	if (!xfs_verify_magic(bp, dsb->sb_magicnum)) {
233 		xfs_warn(mp, "bad magic number");
234 		return -EWRONGFS;
235 	}
236 
237 	if (!xfs_sb_good_version(sbp)) {
238 		xfs_warn(mp, "bad version");
239 		return -EWRONGFS;
240 	}
241 
242 	if (xfs_sb_version_has_pquotino(sbp)) {
243 		if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
244 			xfs_notice(mp,
245 			   "Version 5 of Super block has XFS_OQUOTA bits.");
246 			return -EFSCORRUPTED;
247 		}
248 	} else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
249 				XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
250 			xfs_notice(mp,
251 "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.");
252 			return -EFSCORRUPTED;
253 	}
254 
255 	/*
256 	 * Full inode chunks must be aligned to inode chunk size when
257 	 * sparse inodes are enabled to support the sparse chunk
258 	 * allocation algorithm and prevent overlapping inode records.
259 	 */
260 	if (xfs_sb_version_hassparseinodes(sbp)) {
261 		uint32_t	align;
262 
263 		align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
264 				>> sbp->sb_blocklog;
265 		if (sbp->sb_inoalignmt != align) {
266 			xfs_warn(mp,
267 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
268 				 sbp->sb_inoalignmt, align);
269 			return -EINVAL;
270 		}
271 	}
272 
273 	if (unlikely(
274 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
275 		xfs_warn(mp,
276 		"filesystem is marked as having an external log; "
277 		"specify logdev on the mount command line.");
278 		return -EINVAL;
279 	}
280 
281 	if (unlikely(
282 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
283 		xfs_warn(mp,
284 		"filesystem is marked as having an internal log; "
285 		"do not specify logdev on the mount command line.");
286 		return -EINVAL;
287 	}
288 
289 	/* Compute agcount for this number of dblocks and agblocks */
290 	if (sbp->sb_agblocks) {
291 		agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem);
292 		if (rem)
293 			agcount++;
294 	}
295 
296 	/*
297 	 * More sanity checking.  Most of these were stolen directly from
298 	 * xfs_repair.
299 	 */
300 	if (unlikely(
301 	    sbp->sb_agcount <= 0					||
302 	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
303 	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
304 	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
305 	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
306 	    sbp->sb_sectsize != (1 << sbp->sb_sectlog)			||
307 	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
308 	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
309 	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
310 	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
311 	    sbp->sb_blocksize != (1 << sbp->sb_blocklog)		||
312 	    sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
313 	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
314 	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
315 	    sbp->sb_inodelog < XFS_DINODE_MIN_LOG			||
316 	    sbp->sb_inodelog > XFS_DINODE_MAX_LOG			||
317 	    sbp->sb_inodesize != (1 << sbp->sb_inodelog)		||
318 	    sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE			||
319 	    sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
320 	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES	||
321 	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES	||
322 	    sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1	||
323 	    agcount == 0 || agcount != sbp->sb_agcount			||
324 	    (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)	||
325 	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
326 	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
327 	    (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)	||
328 	    sbp->sb_dblocks == 0					||
329 	    sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)			||
330 	    sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp)			||
331 	    sbp->sb_shared_vn != 0)) {
332 		xfs_notice(mp, "SB sanity check failed");
333 		return -EFSCORRUPTED;
334 	}
335 
336 	if (sbp->sb_unit) {
337 		if (!xfs_sb_version_hasdalign(sbp) ||
338 		    sbp->sb_unit > sbp->sb_width ||
339 		    (sbp->sb_width % sbp->sb_unit) != 0) {
340 			xfs_notice(mp, "SB stripe unit sanity check failed");
341 			return -EFSCORRUPTED;
342 		}
343 	} else if (xfs_sb_version_hasdalign(sbp)) {
344 		xfs_notice(mp, "SB stripe alignment sanity check failed");
345 		return -EFSCORRUPTED;
346 	} else if (sbp->sb_width) {
347 		xfs_notice(mp, "SB stripe width sanity check failed");
348 		return -EFSCORRUPTED;
349 	}
350 
351 
352 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
353 	    sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) {
354 		xfs_notice(mp, "v5 SB sanity check failed");
355 		return -EFSCORRUPTED;
356 	}
357 
358 	/*
359 	 * Until this is fixed only page-sized or smaller data blocks work.
360 	 */
361 	if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
362 		xfs_warn(mp,
363 		"File system with blocksize %d bytes. "
364 		"Only pagesize (%ld) or less will currently work.",
365 				sbp->sb_blocksize, PAGE_SIZE);
366 		return -ENOSYS;
367 	}
368 
369 	/*
370 	 * Currently only very few inode sizes are supported.
371 	 */
372 	switch (sbp->sb_inodesize) {
373 	case 256:
374 	case 512:
375 	case 1024:
376 	case 2048:
377 		break;
378 	default:
379 		xfs_warn(mp, "inode size of %d bytes not supported",
380 				sbp->sb_inodesize);
381 		return -ENOSYS;
382 	}
383 
384 	if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
385 	    xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
386 		xfs_warn(mp,
387 		"file system too large to be mounted on this system.");
388 		return -EFBIG;
389 	}
390 
391 	/*
392 	 * Don't touch the filesystem if a user tool thinks it owns the primary
393 	 * superblock.  mkfs doesn't clear the flag from secondary supers, so
394 	 * we don't check them at all.
395 	 */
396 	if (XFS_BUF_ADDR(bp) == XFS_SB_DADDR && sbp->sb_inprogress) {
397 		xfs_warn(mp, "Offline file system operation in progress!");
398 		return -EFSCORRUPTED;
399 	}
400 	return 0;
401 }
402 
403 void
404 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
405 {
406 	/*
407 	 * older mkfs doesn't initialize quota inodes to NULLFSINO. This
408 	 * leads to in-core values having two different values for a quota
409 	 * inode to be invalid: 0 and NULLFSINO. Change it to a single value
410 	 * NULLFSINO.
411 	 *
412 	 * Note that this change affect only the in-core values. These
413 	 * values are not written back to disk unless any quota information
414 	 * is written to the disk. Even in that case, sb_pquotino field is
415 	 * not written to disk unless the superblock supports pquotino.
416 	 */
417 	if (sbp->sb_uquotino == 0)
418 		sbp->sb_uquotino = NULLFSINO;
419 	if (sbp->sb_gquotino == 0)
420 		sbp->sb_gquotino = NULLFSINO;
421 	if (sbp->sb_pquotino == 0)
422 		sbp->sb_pquotino = NULLFSINO;
423 
424 	/*
425 	 * We need to do these manipilations only if we are working
426 	 * with an older version of on-disk superblock.
427 	 */
428 	if (xfs_sb_version_has_pquotino(sbp))
429 		return;
430 
431 	if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
432 		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
433 					XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
434 	if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
435 		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
436 					XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
437 	sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
438 
439 	if (sbp->sb_qflags & XFS_PQUOTA_ACCT &&
440 	    sbp->sb_gquotino != NULLFSINO)  {
441 		/*
442 		 * In older version of superblock, on-disk superblock only
443 		 * has sb_gquotino, and in-core superblock has both sb_gquotino
444 		 * and sb_pquotino. But, only one of them is supported at any
445 		 * point of time. So, if PQUOTA is set in disk superblock,
446 		 * copy over sb_gquotino to sb_pquotino.  The NULLFSINO test
447 		 * above is to make sure we don't do this twice and wipe them
448 		 * both out!
449 		 */
450 		sbp->sb_pquotino = sbp->sb_gquotino;
451 		sbp->sb_gquotino = NULLFSINO;
452 	}
453 }
454 
455 static void
456 __xfs_sb_from_disk(
457 	struct xfs_sb	*to,
458 	xfs_dsb_t	*from,
459 	bool		convert_xquota)
460 {
461 	to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
462 	to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
463 	to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
464 	to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
465 	to->sb_rextents = be64_to_cpu(from->sb_rextents);
466 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
467 	to->sb_logstart = be64_to_cpu(from->sb_logstart);
468 	to->sb_rootino = be64_to_cpu(from->sb_rootino);
469 	to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
470 	to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
471 	to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
472 	to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
473 	to->sb_agcount = be32_to_cpu(from->sb_agcount);
474 	to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
475 	to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
476 	to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
477 	to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
478 	to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
479 	to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
480 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
481 	to->sb_blocklog = from->sb_blocklog;
482 	to->sb_sectlog = from->sb_sectlog;
483 	to->sb_inodelog = from->sb_inodelog;
484 	to->sb_inopblog = from->sb_inopblog;
485 	to->sb_agblklog = from->sb_agblklog;
486 	to->sb_rextslog = from->sb_rextslog;
487 	to->sb_inprogress = from->sb_inprogress;
488 	to->sb_imax_pct = from->sb_imax_pct;
489 	to->sb_icount = be64_to_cpu(from->sb_icount);
490 	to->sb_ifree = be64_to_cpu(from->sb_ifree);
491 	to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
492 	to->sb_frextents = be64_to_cpu(from->sb_frextents);
493 	to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
494 	to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
495 	to->sb_qflags = be16_to_cpu(from->sb_qflags);
496 	to->sb_flags = from->sb_flags;
497 	to->sb_shared_vn = from->sb_shared_vn;
498 	to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
499 	to->sb_unit = be32_to_cpu(from->sb_unit);
500 	to->sb_width = be32_to_cpu(from->sb_width);
501 	to->sb_dirblklog = from->sb_dirblklog;
502 	to->sb_logsectlog = from->sb_logsectlog;
503 	to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
504 	to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
505 	to->sb_features2 = be32_to_cpu(from->sb_features2);
506 	to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
507 	to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
508 	to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
509 	to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
510 	to->sb_features_log_incompat =
511 				be32_to_cpu(from->sb_features_log_incompat);
512 	/* crc is only used on disk, not in memory; just init to 0 here. */
513 	to->sb_crc = 0;
514 	to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
515 	to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
516 	to->sb_lsn = be64_to_cpu(from->sb_lsn);
517 	/*
518 	 * sb_meta_uuid is only on disk if it differs from sb_uuid and the
519 	 * feature flag is set; if not set we keep it only in memory.
520 	 */
521 	if (xfs_sb_version_hasmetauuid(to))
522 		uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
523 	else
524 		uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
525 	/* Convert on-disk flags to in-memory flags? */
526 	if (convert_xquota)
527 		xfs_sb_quota_from_disk(to);
528 }
529 
530 void
531 xfs_sb_from_disk(
532 	struct xfs_sb	*to,
533 	xfs_dsb_t	*from)
534 {
535 	__xfs_sb_from_disk(to, from, true);
536 }
537 
538 static void
539 xfs_sb_quota_to_disk(
540 	struct xfs_dsb	*to,
541 	struct xfs_sb	*from)
542 {
543 	uint16_t	qflags = from->sb_qflags;
544 
545 	to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
546 	if (xfs_sb_version_has_pquotino(from)) {
547 		to->sb_qflags = cpu_to_be16(from->sb_qflags);
548 		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
549 		to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
550 		return;
551 	}
552 
553 	/*
554 	 * The in-core version of sb_qflags do not have XFS_OQUOTA_*
555 	 * flags, whereas the on-disk version does.  So, convert incore
556 	 * XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
557 	 */
558 	qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
559 			XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
560 
561 	if (from->sb_qflags &
562 			(XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
563 		qflags |= XFS_OQUOTA_ENFD;
564 	if (from->sb_qflags &
565 			(XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
566 		qflags |= XFS_OQUOTA_CHKD;
567 	to->sb_qflags = cpu_to_be16(qflags);
568 
569 	/*
570 	 * GQUOTINO and PQUOTINO cannot be used together in versions
571 	 * of superblock that do not have pquotino. from->sb_flags
572 	 * tells us which quota is active and should be copied to
573 	 * disk. If neither are active, we should NULL the inode.
574 	 *
575 	 * In all cases, the separate pquotino must remain 0 because it
576 	 * it beyond the "end" of the valid non-pquotino superblock.
577 	 */
578 	if (from->sb_qflags & XFS_GQUOTA_ACCT)
579 		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
580 	else if (from->sb_qflags & XFS_PQUOTA_ACCT)
581 		to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
582 	else {
583 		/*
584 		 * We can't rely on just the fields being logged to tell us
585 		 * that it is safe to write NULLFSINO - we should only do that
586 		 * if quotas are not actually enabled. Hence only write
587 		 * NULLFSINO if both in-core quota inodes are NULL.
588 		 */
589 		if (from->sb_gquotino == NULLFSINO &&
590 		    from->sb_pquotino == NULLFSINO)
591 			to->sb_gquotino = cpu_to_be64(NULLFSINO);
592 	}
593 
594 	to->sb_pquotino = 0;
595 }
596 
597 void
598 xfs_sb_to_disk(
599 	struct xfs_dsb	*to,
600 	struct xfs_sb	*from)
601 {
602 	xfs_sb_quota_to_disk(to, from);
603 
604 	to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
605 	to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
606 	to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
607 	to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
608 	to->sb_rextents = cpu_to_be64(from->sb_rextents);
609 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
610 	to->sb_logstart = cpu_to_be64(from->sb_logstart);
611 	to->sb_rootino = cpu_to_be64(from->sb_rootino);
612 	to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
613 	to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
614 	to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
615 	to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
616 	to->sb_agcount = cpu_to_be32(from->sb_agcount);
617 	to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
618 	to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
619 	to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
620 	to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
621 	to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
622 	to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
623 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
624 	to->sb_blocklog = from->sb_blocklog;
625 	to->sb_sectlog = from->sb_sectlog;
626 	to->sb_inodelog = from->sb_inodelog;
627 	to->sb_inopblog = from->sb_inopblog;
628 	to->sb_agblklog = from->sb_agblklog;
629 	to->sb_rextslog = from->sb_rextslog;
630 	to->sb_inprogress = from->sb_inprogress;
631 	to->sb_imax_pct = from->sb_imax_pct;
632 	to->sb_icount = cpu_to_be64(from->sb_icount);
633 	to->sb_ifree = cpu_to_be64(from->sb_ifree);
634 	to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
635 	to->sb_frextents = cpu_to_be64(from->sb_frextents);
636 
637 	to->sb_flags = from->sb_flags;
638 	to->sb_shared_vn = from->sb_shared_vn;
639 	to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
640 	to->sb_unit = cpu_to_be32(from->sb_unit);
641 	to->sb_width = cpu_to_be32(from->sb_width);
642 	to->sb_dirblklog = from->sb_dirblklog;
643 	to->sb_logsectlog = from->sb_logsectlog;
644 	to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
645 	to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
646 
647 	/*
648 	 * We need to ensure that bad_features2 always matches features2.
649 	 * Hence we enforce that here rather than having to remember to do it
650 	 * everywhere else that updates features2.
651 	 */
652 	from->sb_bad_features2 = from->sb_features2;
653 	to->sb_features2 = cpu_to_be32(from->sb_features2);
654 	to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
655 
656 	if (xfs_sb_version_hascrc(from)) {
657 		to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
658 		to->sb_features_ro_compat =
659 				cpu_to_be32(from->sb_features_ro_compat);
660 		to->sb_features_incompat =
661 				cpu_to_be32(from->sb_features_incompat);
662 		to->sb_features_log_incompat =
663 				cpu_to_be32(from->sb_features_log_incompat);
664 		to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
665 		to->sb_lsn = cpu_to_be64(from->sb_lsn);
666 		if (xfs_sb_version_hasmetauuid(from))
667 			uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
668 	}
669 }
670 
671 /*
672  * If the superblock has the CRC feature bit set or the CRC field is non-null,
673  * check that the CRC is valid.  We check the CRC field is non-null because a
674  * single bit error could clear the feature bit and unused parts of the
675  * superblock are supposed to be zero. Hence a non-null crc field indicates that
676  * we've potentially lost a feature bit and we should check it anyway.
677  *
678  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
679  * last field in V4 secondary superblocks.  So for secondary superblocks,
680  * we are more forgiving, and ignore CRC failures if the primary doesn't
681  * indicate that the fs version is V5.
682  */
683 static void
684 xfs_sb_read_verify(
685 	struct xfs_buf		*bp)
686 {
687 	struct xfs_sb		sb;
688 	struct xfs_mount	*mp = bp->b_target->bt_mount;
689 	struct xfs_dsb		*dsb = XFS_BUF_TO_SBP(bp);
690 	int			error;
691 
692 	/*
693 	 * open code the version check to avoid needing to convert the entire
694 	 * superblock from disk order just to check the version number
695 	 */
696 	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
697 	    (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
698 						XFS_SB_VERSION_5) ||
699 	     dsb->sb_crc != 0)) {
700 
701 		if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
702 			/* Only fail bad secondaries on a known V5 filesystem */
703 			if (bp->b_bn == XFS_SB_DADDR ||
704 			    xfs_sb_version_hascrc(&mp->m_sb)) {
705 				error = -EFSBADCRC;
706 				goto out_error;
707 			}
708 		}
709 	}
710 
711 	/*
712 	 * Check all the superblock fields.  Don't byteswap the xquota flags
713 	 * because _verify_common checks the on-disk values.
714 	 */
715 	__xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp), false);
716 	error = xfs_validate_sb_common(mp, bp, &sb);
717 	if (error)
718 		goto out_error;
719 	error = xfs_validate_sb_read(mp, &sb);
720 
721 out_error:
722 	if (error == -EFSCORRUPTED || error == -EFSBADCRC)
723 		xfs_verifier_error(bp, error, __this_address);
724 	else if (error)
725 		xfs_buf_ioerror(bp, error);
726 }
727 
728 /*
729  * We may be probed for a filesystem match, so we may not want to emit
730  * messages when the superblock buffer is not actually an XFS superblock.
731  * If we find an XFS superblock, then run a normal, noisy mount because we are
732  * really going to mount it and want to know about errors.
733  */
734 static void
735 xfs_sb_quiet_read_verify(
736 	struct xfs_buf	*bp)
737 {
738 	struct xfs_dsb	*dsb = XFS_BUF_TO_SBP(bp);
739 
740 	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
741 		/* XFS filesystem, verify noisily! */
742 		xfs_sb_read_verify(bp);
743 		return;
744 	}
745 	/* quietly fail */
746 	xfs_buf_ioerror(bp, -EWRONGFS);
747 }
748 
749 static void
750 xfs_sb_write_verify(
751 	struct xfs_buf		*bp)
752 {
753 	struct xfs_sb		sb;
754 	struct xfs_mount	*mp = bp->b_target->bt_mount;
755 	struct xfs_buf_log_item	*bip = bp->b_log_item;
756 	int			error;
757 
758 	/*
759 	 * Check all the superblock fields.  Don't byteswap the xquota flags
760 	 * because _verify_common checks the on-disk values.
761 	 */
762 	__xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp), false);
763 	error = xfs_validate_sb_common(mp, bp, &sb);
764 	if (error)
765 		goto out_error;
766 	error = xfs_validate_sb_write(mp, bp, &sb);
767 	if (error)
768 		goto out_error;
769 
770 	if (!xfs_sb_version_hascrc(&mp->m_sb))
771 		return;
772 
773 	if (bip)
774 		XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
775 
776 	xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
777 	return;
778 
779 out_error:
780 	xfs_verifier_error(bp, error, __this_address);
781 }
782 
783 const struct xfs_buf_ops xfs_sb_buf_ops = {
784 	.name = "xfs_sb",
785 	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
786 	.verify_read = xfs_sb_read_verify,
787 	.verify_write = xfs_sb_write_verify,
788 };
789 
790 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
791 	.name = "xfs_sb_quiet",
792 	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
793 	.verify_read = xfs_sb_quiet_read_verify,
794 	.verify_write = xfs_sb_write_verify,
795 };
796 
797 /*
798  * xfs_mount_common
799  *
800  * Mount initialization code establishing various mount
801  * fields from the superblock associated with the given
802  * mount structure
803  */
804 void
805 xfs_sb_mount_common(
806 	struct xfs_mount *mp,
807 	struct xfs_sb	*sbp)
808 {
809 	mp->m_agfrotor = mp->m_agirotor = 0;
810 	mp->m_maxagi = mp->m_sb.sb_agcount;
811 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
812 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
813 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
814 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
815 	mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
816 	mp->m_blockmask = sbp->sb_blocksize - 1;
817 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
818 	mp->m_blockwmask = mp->m_blockwsize - 1;
819 
820 	mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
821 	mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
822 	mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
823 	mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
824 
825 	mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
826 	mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
827 	mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
828 	mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
829 
830 	mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
831 	mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
832 	mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
833 	mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
834 
835 	mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 1);
836 	mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 0);
837 	mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
838 	mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
839 
840 	mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, true);
841 	mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, false);
842 	mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
843 	mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
844 
845 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
846 	mp->m_ialloc_inos = max_t(uint16_t, XFS_INODES_PER_CHUNK,
847 					sbp->sb_inopblock);
848 	mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
849 
850 	if (sbp->sb_spino_align)
851 		mp->m_ialloc_min_blks = sbp->sb_spino_align;
852 	else
853 		mp->m_ialloc_min_blks = mp->m_ialloc_blks;
854 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
855 	mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp);
856 }
857 
858 /*
859  * xfs_initialize_perag_data
860  *
861  * Read in each per-ag structure so we can count up the number of
862  * allocated inodes, free inodes and used filesystem blocks as this
863  * information is no longer persistent in the superblock. Once we have
864  * this information, write it into the in-core superblock structure.
865  */
866 int
867 xfs_initialize_perag_data(
868 	struct xfs_mount *mp,
869 	xfs_agnumber_t	agcount)
870 {
871 	xfs_agnumber_t	index;
872 	xfs_perag_t	*pag;
873 	xfs_sb_t	*sbp = &mp->m_sb;
874 	uint64_t	ifree = 0;
875 	uint64_t	ialloc = 0;
876 	uint64_t	bfree = 0;
877 	uint64_t	bfreelst = 0;
878 	uint64_t	btree = 0;
879 	uint64_t	fdblocks;
880 	int		error = 0;
881 
882 	for (index = 0; index < agcount; index++) {
883 		/*
884 		 * read the agf, then the agi. This gets us
885 		 * all the information we need and populates the
886 		 * per-ag structures for us.
887 		 */
888 		error = xfs_alloc_pagf_init(mp, NULL, index, 0);
889 		if (error)
890 			return error;
891 
892 		error = xfs_ialloc_pagi_init(mp, NULL, index);
893 		if (error)
894 			return error;
895 		pag = xfs_perag_get(mp, index);
896 		ifree += pag->pagi_freecount;
897 		ialloc += pag->pagi_count;
898 		bfree += pag->pagf_freeblks;
899 		bfreelst += pag->pagf_flcount;
900 		btree += pag->pagf_btreeblks;
901 		xfs_perag_put(pag);
902 	}
903 	fdblocks = bfree + bfreelst + btree;
904 
905 	/*
906 	 * If the new summary counts are obviously incorrect, fail the
907 	 * mount operation because that implies the AGFs are also corrupt.
908 	 * Clear BAD_SUMMARY so that we don't unmount with a dirty log, which
909 	 * will prevent xfs_repair from fixing anything.
910 	 */
911 	if (fdblocks > sbp->sb_dblocks || ifree > ialloc) {
912 		xfs_alert(mp, "AGF corruption. Please run xfs_repair.");
913 		error = -EFSCORRUPTED;
914 		goto out;
915 	}
916 
917 	/* Overwrite incore superblock counters with just-read data */
918 	spin_lock(&mp->m_sb_lock);
919 	sbp->sb_ifree = ifree;
920 	sbp->sb_icount = ialloc;
921 	sbp->sb_fdblocks = fdblocks;
922 	spin_unlock(&mp->m_sb_lock);
923 
924 	xfs_reinit_percpu_counters(mp);
925 out:
926 	mp->m_flags &= ~XFS_MOUNT_BAD_SUMMARY;
927 	return error;
928 }
929 
930 /*
931  * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
932  * into the superblock buffer to be logged.  It does not provide the higher
933  * level of locking that is needed to protect the in-core superblock from
934  * concurrent access.
935  */
936 void
937 xfs_log_sb(
938 	struct xfs_trans	*tp)
939 {
940 	struct xfs_mount	*mp = tp->t_mountp;
941 	struct xfs_buf		*bp = xfs_trans_getsb(tp, mp, 0);
942 
943 	mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
944 	mp->m_sb.sb_ifree = percpu_counter_sum(&mp->m_ifree);
945 	mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
946 
947 	xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
948 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
949 	xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb));
950 }
951 
952 /*
953  * xfs_sync_sb
954  *
955  * Sync the superblock to disk.
956  *
957  * Note that the caller is responsible for checking the frozen state of the
958  * filesystem. This procedure uses the non-blocking transaction allocator and
959  * thus will allow modifications to a frozen fs. This is required because this
960  * code can be called during the process of freezing where use of the high-level
961  * allocator would deadlock.
962  */
963 int
964 xfs_sync_sb(
965 	struct xfs_mount	*mp,
966 	bool			wait)
967 {
968 	struct xfs_trans	*tp;
969 	int			error;
970 
971 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0,
972 			XFS_TRANS_NO_WRITECOUNT, &tp);
973 	if (error)
974 		return error;
975 
976 	xfs_log_sb(tp);
977 	if (wait)
978 		xfs_trans_set_sync(tp);
979 	return xfs_trans_commit(tp);
980 }
981 
982 /*
983  * Update all the secondary superblocks to match the new state of the primary.
984  * Because we are completely overwriting all the existing fields in the
985  * secondary superblock buffers, there is no need to read them in from disk.
986  * Just get a new buffer, stamp it and write it.
987  *
988  * The sb buffers need to be cached here so that we serialise against other
989  * operations that access the secondary superblocks, but we don't want to keep
990  * them in memory once it is written so we mark it as a one-shot buffer.
991  */
992 int
993 xfs_update_secondary_sbs(
994 	struct xfs_mount	*mp)
995 {
996 	xfs_agnumber_t		agno;
997 	int			saved_error = 0;
998 	int			error = 0;
999 	LIST_HEAD		(buffer_list);
1000 
1001 	/* update secondary superblocks. */
1002 	for (agno = 1; agno < mp->m_sb.sb_agcount; agno++) {
1003 		struct xfs_buf		*bp;
1004 
1005 		bp = xfs_buf_get(mp->m_ddev_targp,
1006 				 XFS_AG_DADDR(mp, agno, XFS_SB_DADDR),
1007 				 XFS_FSS_TO_BB(mp, 1), 0);
1008 		/*
1009 		 * If we get an error reading or writing alternate superblocks,
1010 		 * continue.  xfs_repair chooses the "best" superblock based
1011 		 * on most matches; if we break early, we'll leave more
1012 		 * superblocks un-updated than updated, and xfs_repair may
1013 		 * pick them over the properly-updated primary.
1014 		 */
1015 		if (!bp) {
1016 			xfs_warn(mp,
1017 		"error allocating secondary superblock for ag %d",
1018 				agno);
1019 			if (!saved_error)
1020 				saved_error = -ENOMEM;
1021 			continue;
1022 		}
1023 
1024 		bp->b_ops = &xfs_sb_buf_ops;
1025 		xfs_buf_oneshot(bp);
1026 		xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
1027 		xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
1028 		xfs_buf_delwri_queue(bp, &buffer_list);
1029 		xfs_buf_relse(bp);
1030 
1031 		/* don't hold too many buffers at once */
1032 		if (agno % 16)
1033 			continue;
1034 
1035 		error = xfs_buf_delwri_submit(&buffer_list);
1036 		if (error) {
1037 			xfs_warn(mp,
1038 		"write error %d updating a secondary superblock near ag %d",
1039 				error, agno);
1040 			if (!saved_error)
1041 				saved_error = error;
1042 			continue;
1043 		}
1044 	}
1045 	error = xfs_buf_delwri_submit(&buffer_list);
1046 	if (error) {
1047 		xfs_warn(mp,
1048 		"write error %d updating a secondary superblock near ag %d",
1049 			error, agno);
1050 	}
1051 
1052 	return saved_error ? saved_error : error;
1053 }
1054 
1055 /*
1056  * Same behavior as xfs_sync_sb, except that it is always synchronous and it
1057  * also writes the superblock buffer to disk sector 0 immediately.
1058  */
1059 int
1060 xfs_sync_sb_buf(
1061 	struct xfs_mount	*mp)
1062 {
1063 	struct xfs_trans	*tp;
1064 	struct xfs_buf		*bp;
1065 	int			error;
1066 
1067 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp);
1068 	if (error)
1069 		return error;
1070 
1071 	bp = xfs_trans_getsb(tp, mp, 0);
1072 	xfs_log_sb(tp);
1073 	xfs_trans_bhold(tp, bp);
1074 	xfs_trans_set_sync(tp);
1075 	error = xfs_trans_commit(tp);
1076 	if (error)
1077 		goto out;
1078 	/*
1079 	 * write out the sb buffer to get the changes to disk
1080 	 */
1081 	error = xfs_bwrite(bp);
1082 out:
1083 	xfs_buf_relse(bp);
1084 	return error;
1085 }
1086 
1087 int
1088 xfs_fs_geometry(
1089 	struct xfs_sb		*sbp,
1090 	struct xfs_fsop_geom	*geo,
1091 	int			struct_version)
1092 {
1093 	memset(geo, 0, sizeof(struct xfs_fsop_geom));
1094 
1095 	geo->blocksize = sbp->sb_blocksize;
1096 	geo->rtextsize = sbp->sb_rextsize;
1097 	geo->agblocks = sbp->sb_agblocks;
1098 	geo->agcount = sbp->sb_agcount;
1099 	geo->logblocks = sbp->sb_logblocks;
1100 	geo->sectsize = sbp->sb_sectsize;
1101 	geo->inodesize = sbp->sb_inodesize;
1102 	geo->imaxpct = sbp->sb_imax_pct;
1103 	geo->datablocks = sbp->sb_dblocks;
1104 	geo->rtblocks = sbp->sb_rblocks;
1105 	geo->rtextents = sbp->sb_rextents;
1106 	geo->logstart = sbp->sb_logstart;
1107 	BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid));
1108 	memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid));
1109 
1110 	if (struct_version < 2)
1111 		return 0;
1112 
1113 	geo->sunit = sbp->sb_unit;
1114 	geo->swidth = sbp->sb_width;
1115 
1116 	if (struct_version < 3)
1117 		return 0;
1118 
1119 	geo->version = XFS_FSOP_GEOM_VERSION;
1120 	geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
1121 		     XFS_FSOP_GEOM_FLAGS_DIRV2 |
1122 		     XFS_FSOP_GEOM_FLAGS_EXTFLG;
1123 	if (xfs_sb_version_hasattr(sbp))
1124 		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR;
1125 	if (xfs_sb_version_hasquota(sbp))
1126 		geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA;
1127 	if (xfs_sb_version_hasalign(sbp))
1128 		geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN;
1129 	if (xfs_sb_version_hasdalign(sbp))
1130 		geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN;
1131 	if (xfs_sb_version_hassector(sbp))
1132 		geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR;
1133 	if (xfs_sb_version_hasasciici(sbp))
1134 		geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI;
1135 	if (xfs_sb_version_haslazysbcount(sbp))
1136 		geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB;
1137 	if (xfs_sb_version_hasattr2(sbp))
1138 		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2;
1139 	if (xfs_sb_version_hasprojid32bit(sbp))
1140 		geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32;
1141 	if (xfs_sb_version_hascrc(sbp))
1142 		geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB;
1143 	if (xfs_sb_version_hasftype(sbp))
1144 		geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE;
1145 	if (xfs_sb_version_hasfinobt(sbp))
1146 		geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT;
1147 	if (xfs_sb_version_hassparseinodes(sbp))
1148 		geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES;
1149 	if (xfs_sb_version_hasrmapbt(sbp))
1150 		geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT;
1151 	if (xfs_sb_version_hasreflink(sbp))
1152 		geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK;
1153 	if (xfs_sb_version_hassector(sbp))
1154 		geo->logsectsize = sbp->sb_logsectsize;
1155 	else
1156 		geo->logsectsize = BBSIZE;
1157 	geo->rtsectsize = sbp->sb_blocksize;
1158 	geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp);
1159 
1160 	if (struct_version < 4)
1161 		return 0;
1162 
1163 	if (xfs_sb_version_haslogv2(sbp))
1164 		geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2;
1165 
1166 	geo->logsunit = sbp->sb_logsunit;
1167 
1168 	return 0;
1169 }
1170 
1171 /* Read a secondary superblock. */
1172 int
1173 xfs_sb_read_secondary(
1174 	struct xfs_mount	*mp,
1175 	struct xfs_trans	*tp,
1176 	xfs_agnumber_t		agno,
1177 	struct xfs_buf		**bpp)
1178 {
1179 	struct xfs_buf		*bp;
1180 	int			error;
1181 
1182 	ASSERT(agno != 0 && agno != NULLAGNUMBER);
1183 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1184 			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1185 			XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops);
1186 	if (error)
1187 		return error;
1188 	xfs_buf_set_ref(bp, XFS_SSB_REF);
1189 	*bpp = bp;
1190 	return 0;
1191 }
1192 
1193 /* Get an uninitialised secondary superblock buffer. */
1194 int
1195 xfs_sb_get_secondary(
1196 	struct xfs_mount	*mp,
1197 	struct xfs_trans	*tp,
1198 	xfs_agnumber_t		agno,
1199 	struct xfs_buf		**bpp)
1200 {
1201 	struct xfs_buf		*bp;
1202 
1203 	ASSERT(agno != 0 && agno != NULLAGNUMBER);
1204 	bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1205 			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1206 			XFS_FSS_TO_BB(mp, 1), 0);
1207 	if (!bp)
1208 		return -ENOMEM;
1209 	bp->b_ops = &xfs_sb_buf_ops;
1210 	xfs_buf_oneshot(bp);
1211 	*bpp = bp;
1212 	return 0;
1213 }
1214