xref: /linux/fs/xfs/libxfs/xfs_sb.c (revision 37744feebc086908fd89760650f458ab19071750)
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_trace.h"
19 #include "xfs_trans.h"
20 #include "xfs_buf_item.h"
21 #include "xfs_bmap_btree.h"
22 #include "xfs_alloc_btree.h"
23 #include "xfs_log.h"
24 #include "xfs_rmap_btree.h"
25 #include "xfs_refcount_btree.h"
26 #include "xfs_da_format.h"
27 #include "xfs_health.h"
28 
29 /*
30  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
31  */
32 
33 /*
34  * Reference counting access wrappers to the perag structures.
35  * Because we never free per-ag structures, the only thing we
36  * have to protect against changes is the tree structure itself.
37  */
38 struct xfs_perag *
39 xfs_perag_get(
40 	struct xfs_mount	*mp,
41 	xfs_agnumber_t		agno)
42 {
43 	struct xfs_perag	*pag;
44 	int			ref = 0;
45 
46 	rcu_read_lock();
47 	pag = radix_tree_lookup(&mp->m_perag_tree, agno);
48 	if (pag) {
49 		ASSERT(atomic_read(&pag->pag_ref) >= 0);
50 		ref = atomic_inc_return(&pag->pag_ref);
51 	}
52 	rcu_read_unlock();
53 	trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
54 	return pag;
55 }
56 
57 /*
58  * search from @first to find the next perag with the given tag set.
59  */
60 struct xfs_perag *
61 xfs_perag_get_tag(
62 	struct xfs_mount	*mp,
63 	xfs_agnumber_t		first,
64 	int			tag)
65 {
66 	struct xfs_perag	*pag;
67 	int			found;
68 	int			ref;
69 
70 	rcu_read_lock();
71 	found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
72 					(void **)&pag, first, 1, tag);
73 	if (found <= 0) {
74 		rcu_read_unlock();
75 		return NULL;
76 	}
77 	ref = atomic_inc_return(&pag->pag_ref);
78 	rcu_read_unlock();
79 	trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
80 	return pag;
81 }
82 
83 void
84 xfs_perag_put(
85 	struct xfs_perag	*pag)
86 {
87 	int	ref;
88 
89 	ASSERT(atomic_read(&pag->pag_ref) > 0);
90 	ref = atomic_dec_return(&pag->pag_ref);
91 	trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
92 }
93 
94 /* Check all the superblock fields we care about when reading one in. */
95 STATIC int
96 xfs_validate_sb_read(
97 	struct xfs_mount	*mp,
98 	struct xfs_sb		*sbp)
99 {
100 	if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_5)
101 		return 0;
102 
103 	/*
104 	 * Version 5 superblock feature mask validation. Reject combinations
105 	 * the kernel cannot support up front before checking anything else.
106 	 */
107 	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
108 		xfs_warn(mp,
109 "Superblock has unknown compatible features (0x%x) enabled.",
110 			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
111 		xfs_warn(mp,
112 "Using a more recent kernel is recommended.");
113 	}
114 
115 	if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
116 		xfs_alert(mp,
117 "Superblock has unknown read-only compatible features (0x%x) enabled.",
118 			(sbp->sb_features_ro_compat &
119 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
120 		if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
121 			xfs_warn(mp,
122 "Attempted to mount read-only compatible filesystem read-write.");
123 			xfs_warn(mp,
124 "Filesystem can only be safely mounted read only.");
125 
126 			return -EINVAL;
127 		}
128 	}
129 	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
130 		xfs_warn(mp,
131 "Superblock has unknown incompatible features (0x%x) enabled.",
132 			(sbp->sb_features_incompat &
133 					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
134 		xfs_warn(mp,
135 "Filesystem cannot be safely mounted by this kernel.");
136 		return -EINVAL;
137 	}
138 
139 	return 0;
140 }
141 
142 /* Check all the superblock fields we care about when writing one out. */
143 STATIC int
144 xfs_validate_sb_write(
145 	struct xfs_mount	*mp,
146 	struct xfs_buf		*bp,
147 	struct xfs_sb		*sbp)
148 {
149 	/*
150 	 * Carry out additional sb summary counter sanity checks when we write
151 	 * the superblock.  We skip this in the read validator because there
152 	 * could be newer superblocks in the log and if the values are garbage
153 	 * even after replay we'll recalculate them at the end of log mount.
154 	 *
155 	 * mkfs has traditionally written zeroed counters to inprogress and
156 	 * secondary superblocks, so allow this usage to continue because
157 	 * we never read counters from such superblocks.
158 	 */
159 	if (XFS_BUF_ADDR(bp) == XFS_SB_DADDR && !sbp->sb_inprogress &&
160 	    (sbp->sb_fdblocks > sbp->sb_dblocks ||
161 	     !xfs_verify_icount(mp, sbp->sb_icount) ||
162 	     sbp->sb_ifree > sbp->sb_icount)) {
163 		xfs_warn(mp, "SB summary counter sanity check failed");
164 		return -EFSCORRUPTED;
165 	}
166 
167 	if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_5)
168 		return 0;
169 
170 	/*
171 	 * Version 5 superblock feature mask validation. Reject combinations
172 	 * the kernel cannot support since we checked for unsupported bits in
173 	 * the read verifier, which means that memory is corrupt.
174 	 */
175 	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
176 		xfs_warn(mp,
177 "Corruption detected in superblock compatible features (0x%x)!",
178 			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
179 		return -EFSCORRUPTED;
180 	}
181 
182 	if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
183 		xfs_alert(mp,
184 "Corruption detected in superblock read-only compatible features (0x%x)!",
185 			(sbp->sb_features_ro_compat &
186 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
187 		return -EFSCORRUPTED;
188 	}
189 	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
190 		xfs_warn(mp,
191 "Corruption detected in superblock incompatible features (0x%x)!",
192 			(sbp->sb_features_incompat &
193 					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
194 		return -EFSCORRUPTED;
195 	}
196 	if (xfs_sb_has_incompat_log_feature(sbp,
197 			XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
198 		xfs_warn(mp,
199 "Corruption detected in superblock incompatible log features (0x%x)!",
200 			(sbp->sb_features_log_incompat &
201 					XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
202 		return -EFSCORRUPTED;
203 	}
204 
205 	/*
206 	 * We can't read verify the sb LSN because the read verifier is called
207 	 * before the log is allocated and processed. We know the log is set up
208 	 * before write verifier calls, so check it here.
209 	 */
210 	if (!xfs_log_check_lsn(mp, sbp->sb_lsn))
211 		return -EFSCORRUPTED;
212 
213 	return 0;
214 }
215 
216 /* Check the validity of the SB. */
217 STATIC int
218 xfs_validate_sb_common(
219 	struct xfs_mount	*mp,
220 	struct xfs_buf		*bp,
221 	struct xfs_sb		*sbp)
222 {
223 	struct xfs_dsb		*dsb = bp->b_addr;
224 	uint32_t		agcount = 0;
225 	uint32_t		rem;
226 
227 	if (!xfs_verify_magic(bp, dsb->sb_magicnum)) {
228 		xfs_warn(mp, "bad magic number");
229 		return -EWRONGFS;
230 	}
231 
232 	if (!xfs_sb_good_version(sbp)) {
233 		xfs_warn(mp, "bad version");
234 		return -EWRONGFS;
235 	}
236 
237 	if (xfs_sb_version_has_pquotino(sbp)) {
238 		if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
239 			xfs_notice(mp,
240 			   "Version 5 of Super block has XFS_OQUOTA bits.");
241 			return -EFSCORRUPTED;
242 		}
243 	} else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
244 				XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
245 			xfs_notice(mp,
246 "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.");
247 			return -EFSCORRUPTED;
248 	}
249 
250 	/*
251 	 * Full inode chunks must be aligned to inode chunk size when
252 	 * sparse inodes are enabled to support the sparse chunk
253 	 * allocation algorithm and prevent overlapping inode records.
254 	 */
255 	if (xfs_sb_version_hassparseinodes(sbp)) {
256 		uint32_t	align;
257 
258 		align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
259 				>> sbp->sb_blocklog;
260 		if (sbp->sb_inoalignmt != align) {
261 			xfs_warn(mp,
262 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
263 				 sbp->sb_inoalignmt, align);
264 			return -EINVAL;
265 		}
266 	}
267 
268 	if (unlikely(
269 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
270 		xfs_warn(mp,
271 		"filesystem is marked as having an external log; "
272 		"specify logdev on the mount command line.");
273 		return -EINVAL;
274 	}
275 
276 	if (unlikely(
277 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
278 		xfs_warn(mp,
279 		"filesystem is marked as having an internal log; "
280 		"do not specify logdev on the mount command line.");
281 		return -EINVAL;
282 	}
283 
284 	/* Compute agcount for this number of dblocks and agblocks */
285 	if (sbp->sb_agblocks) {
286 		agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem);
287 		if (rem)
288 			agcount++;
289 	}
290 
291 	/*
292 	 * More sanity checking.  Most of these were stolen directly from
293 	 * xfs_repair.
294 	 */
295 	if (unlikely(
296 	    sbp->sb_agcount <= 0					||
297 	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
298 	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
299 	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
300 	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
301 	    sbp->sb_sectsize != (1 << sbp->sb_sectlog)			||
302 	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
303 	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
304 	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
305 	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
306 	    sbp->sb_blocksize != (1 << sbp->sb_blocklog)		||
307 	    sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
308 	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
309 	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
310 	    sbp->sb_inodelog < XFS_DINODE_MIN_LOG			||
311 	    sbp->sb_inodelog > XFS_DINODE_MAX_LOG			||
312 	    sbp->sb_inodesize != (1 << sbp->sb_inodelog)		||
313 	    sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE			||
314 	    sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
315 	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES	||
316 	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES	||
317 	    sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1	||
318 	    agcount == 0 || agcount != sbp->sb_agcount			||
319 	    (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)	||
320 	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
321 	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
322 	    (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)	||
323 	    sbp->sb_dblocks == 0					||
324 	    sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)			||
325 	    sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp)			||
326 	    sbp->sb_shared_vn != 0)) {
327 		xfs_notice(mp, "SB sanity check failed");
328 		return -EFSCORRUPTED;
329 	}
330 
331 	/* Validate the realtime geometry; stolen from xfs_repair */
332 	if (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE ||
333 	    sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) {
334 		xfs_notice(mp,
335 			"realtime extent sanity check failed");
336 		return -EFSCORRUPTED;
337 	}
338 
339 	if (sbp->sb_rblocks == 0) {
340 		if (sbp->sb_rextents != 0 || sbp->sb_rbmblocks != 0 ||
341 		    sbp->sb_rextslog != 0 || sbp->sb_frextents != 0) {
342 			xfs_notice(mp,
343 				"realtime zeroed geometry check failed");
344 			return -EFSCORRUPTED;
345 		}
346 	} else {
347 		uint64_t	rexts;
348 		uint64_t	rbmblocks;
349 
350 		rexts = div_u64(sbp->sb_rblocks, sbp->sb_rextsize);
351 		rbmblocks = howmany_64(sbp->sb_rextents,
352 				       NBBY * sbp->sb_blocksize);
353 
354 		if (sbp->sb_rextents != rexts ||
355 		    sbp->sb_rextslog != xfs_highbit32(sbp->sb_rextents) ||
356 		    sbp->sb_rbmblocks != rbmblocks) {
357 			xfs_notice(mp,
358 				"realtime geometry sanity check failed");
359 			return -EFSCORRUPTED;
360 		}
361 	}
362 
363 	if (sbp->sb_unit) {
364 		if (!xfs_sb_version_hasdalign(sbp) ||
365 		    sbp->sb_unit > sbp->sb_width ||
366 		    (sbp->sb_width % sbp->sb_unit) != 0) {
367 			xfs_notice(mp, "SB stripe unit sanity check failed");
368 			return -EFSCORRUPTED;
369 		}
370 	} else if (xfs_sb_version_hasdalign(sbp)) {
371 		xfs_notice(mp, "SB stripe alignment sanity check failed");
372 		return -EFSCORRUPTED;
373 	} else if (sbp->sb_width) {
374 		xfs_notice(mp, "SB stripe width sanity check failed");
375 		return -EFSCORRUPTED;
376 	}
377 
378 
379 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
380 	    sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) {
381 		xfs_notice(mp, "v5 SB sanity check failed");
382 		return -EFSCORRUPTED;
383 	}
384 
385 	/*
386 	 * Until this is fixed only page-sized or smaller data blocks work.
387 	 */
388 	if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
389 		xfs_warn(mp,
390 		"File system with blocksize %d bytes. "
391 		"Only pagesize (%ld) or less will currently work.",
392 				sbp->sb_blocksize, PAGE_SIZE);
393 		return -ENOSYS;
394 	}
395 
396 	/*
397 	 * Currently only very few inode sizes are supported.
398 	 */
399 	switch (sbp->sb_inodesize) {
400 	case 256:
401 	case 512:
402 	case 1024:
403 	case 2048:
404 		break;
405 	default:
406 		xfs_warn(mp, "inode size of %d bytes not supported",
407 				sbp->sb_inodesize);
408 		return -ENOSYS;
409 	}
410 
411 	if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
412 	    xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
413 		xfs_warn(mp,
414 		"file system too large to be mounted on this system.");
415 		return -EFBIG;
416 	}
417 
418 	/*
419 	 * Don't touch the filesystem if a user tool thinks it owns the primary
420 	 * superblock.  mkfs doesn't clear the flag from secondary supers, so
421 	 * we don't check them at all.
422 	 */
423 	if (XFS_BUF_ADDR(bp) == XFS_SB_DADDR && sbp->sb_inprogress) {
424 		xfs_warn(mp, "Offline file system operation in progress!");
425 		return -EFSCORRUPTED;
426 	}
427 	return 0;
428 }
429 
430 void
431 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
432 {
433 	/*
434 	 * older mkfs doesn't initialize quota inodes to NULLFSINO. This
435 	 * leads to in-core values having two different values for a quota
436 	 * inode to be invalid: 0 and NULLFSINO. Change it to a single value
437 	 * NULLFSINO.
438 	 *
439 	 * Note that this change affect only the in-core values. These
440 	 * values are not written back to disk unless any quota information
441 	 * is written to the disk. Even in that case, sb_pquotino field is
442 	 * not written to disk unless the superblock supports pquotino.
443 	 */
444 	if (sbp->sb_uquotino == 0)
445 		sbp->sb_uquotino = NULLFSINO;
446 	if (sbp->sb_gquotino == 0)
447 		sbp->sb_gquotino = NULLFSINO;
448 	if (sbp->sb_pquotino == 0)
449 		sbp->sb_pquotino = NULLFSINO;
450 
451 	/*
452 	 * We need to do these manipilations only if we are working
453 	 * with an older version of on-disk superblock.
454 	 */
455 	if (xfs_sb_version_has_pquotino(sbp))
456 		return;
457 
458 	if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
459 		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
460 					XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
461 	if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
462 		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
463 					XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
464 	sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
465 
466 	if (sbp->sb_qflags & XFS_PQUOTA_ACCT &&
467 	    sbp->sb_gquotino != NULLFSINO)  {
468 		/*
469 		 * In older version of superblock, on-disk superblock only
470 		 * has sb_gquotino, and in-core superblock has both sb_gquotino
471 		 * and sb_pquotino. But, only one of them is supported at any
472 		 * point of time. So, if PQUOTA is set in disk superblock,
473 		 * copy over sb_gquotino to sb_pquotino.  The NULLFSINO test
474 		 * above is to make sure we don't do this twice and wipe them
475 		 * both out!
476 		 */
477 		sbp->sb_pquotino = sbp->sb_gquotino;
478 		sbp->sb_gquotino = NULLFSINO;
479 	}
480 }
481 
482 static void
483 __xfs_sb_from_disk(
484 	struct xfs_sb	*to,
485 	xfs_dsb_t	*from,
486 	bool		convert_xquota)
487 {
488 	to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
489 	to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
490 	to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
491 	to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
492 	to->sb_rextents = be64_to_cpu(from->sb_rextents);
493 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
494 	to->sb_logstart = be64_to_cpu(from->sb_logstart);
495 	to->sb_rootino = be64_to_cpu(from->sb_rootino);
496 	to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
497 	to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
498 	to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
499 	to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
500 	to->sb_agcount = be32_to_cpu(from->sb_agcount);
501 	to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
502 	to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
503 	to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
504 	to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
505 	to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
506 	to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
507 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
508 	to->sb_blocklog = from->sb_blocklog;
509 	to->sb_sectlog = from->sb_sectlog;
510 	to->sb_inodelog = from->sb_inodelog;
511 	to->sb_inopblog = from->sb_inopblog;
512 	to->sb_agblklog = from->sb_agblklog;
513 	to->sb_rextslog = from->sb_rextslog;
514 	to->sb_inprogress = from->sb_inprogress;
515 	to->sb_imax_pct = from->sb_imax_pct;
516 	to->sb_icount = be64_to_cpu(from->sb_icount);
517 	to->sb_ifree = be64_to_cpu(from->sb_ifree);
518 	to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
519 	to->sb_frextents = be64_to_cpu(from->sb_frextents);
520 	to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
521 	to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
522 	to->sb_qflags = be16_to_cpu(from->sb_qflags);
523 	to->sb_flags = from->sb_flags;
524 	to->sb_shared_vn = from->sb_shared_vn;
525 	to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
526 	to->sb_unit = be32_to_cpu(from->sb_unit);
527 	to->sb_width = be32_to_cpu(from->sb_width);
528 	to->sb_dirblklog = from->sb_dirblklog;
529 	to->sb_logsectlog = from->sb_logsectlog;
530 	to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
531 	to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
532 	to->sb_features2 = be32_to_cpu(from->sb_features2);
533 	to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
534 	to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
535 	to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
536 	to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
537 	to->sb_features_log_incompat =
538 				be32_to_cpu(from->sb_features_log_incompat);
539 	/* crc is only used on disk, not in memory; just init to 0 here. */
540 	to->sb_crc = 0;
541 	to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
542 	to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
543 	to->sb_lsn = be64_to_cpu(from->sb_lsn);
544 	/*
545 	 * sb_meta_uuid is only on disk if it differs from sb_uuid and the
546 	 * feature flag is set; if not set we keep it only in memory.
547 	 */
548 	if (xfs_sb_version_hasmetauuid(to))
549 		uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
550 	else
551 		uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
552 	/* Convert on-disk flags to in-memory flags? */
553 	if (convert_xquota)
554 		xfs_sb_quota_from_disk(to);
555 }
556 
557 void
558 xfs_sb_from_disk(
559 	struct xfs_sb	*to,
560 	xfs_dsb_t	*from)
561 {
562 	__xfs_sb_from_disk(to, from, true);
563 }
564 
565 static void
566 xfs_sb_quota_to_disk(
567 	struct xfs_dsb	*to,
568 	struct xfs_sb	*from)
569 {
570 	uint16_t	qflags = from->sb_qflags;
571 
572 	to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
573 	if (xfs_sb_version_has_pquotino(from)) {
574 		to->sb_qflags = cpu_to_be16(from->sb_qflags);
575 		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
576 		to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
577 		return;
578 	}
579 
580 	/*
581 	 * The in-core version of sb_qflags do not have XFS_OQUOTA_*
582 	 * flags, whereas the on-disk version does.  So, convert incore
583 	 * XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
584 	 */
585 	qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
586 			XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
587 
588 	if (from->sb_qflags &
589 			(XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
590 		qflags |= XFS_OQUOTA_ENFD;
591 	if (from->sb_qflags &
592 			(XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
593 		qflags |= XFS_OQUOTA_CHKD;
594 	to->sb_qflags = cpu_to_be16(qflags);
595 
596 	/*
597 	 * GQUOTINO and PQUOTINO cannot be used together in versions
598 	 * of superblock that do not have pquotino. from->sb_flags
599 	 * tells us which quota is active and should be copied to
600 	 * disk. If neither are active, we should NULL the inode.
601 	 *
602 	 * In all cases, the separate pquotino must remain 0 because it
603 	 * it beyond the "end" of the valid non-pquotino superblock.
604 	 */
605 	if (from->sb_qflags & XFS_GQUOTA_ACCT)
606 		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
607 	else if (from->sb_qflags & XFS_PQUOTA_ACCT)
608 		to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
609 	else {
610 		/*
611 		 * We can't rely on just the fields being logged to tell us
612 		 * that it is safe to write NULLFSINO - we should only do that
613 		 * if quotas are not actually enabled. Hence only write
614 		 * NULLFSINO if both in-core quota inodes are NULL.
615 		 */
616 		if (from->sb_gquotino == NULLFSINO &&
617 		    from->sb_pquotino == NULLFSINO)
618 			to->sb_gquotino = cpu_to_be64(NULLFSINO);
619 	}
620 
621 	to->sb_pquotino = 0;
622 }
623 
624 void
625 xfs_sb_to_disk(
626 	struct xfs_dsb	*to,
627 	struct xfs_sb	*from)
628 {
629 	xfs_sb_quota_to_disk(to, from);
630 
631 	to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
632 	to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
633 	to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
634 	to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
635 	to->sb_rextents = cpu_to_be64(from->sb_rextents);
636 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
637 	to->sb_logstart = cpu_to_be64(from->sb_logstart);
638 	to->sb_rootino = cpu_to_be64(from->sb_rootino);
639 	to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
640 	to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
641 	to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
642 	to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
643 	to->sb_agcount = cpu_to_be32(from->sb_agcount);
644 	to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
645 	to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
646 	to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
647 	to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
648 	to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
649 	to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
650 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
651 	to->sb_blocklog = from->sb_blocklog;
652 	to->sb_sectlog = from->sb_sectlog;
653 	to->sb_inodelog = from->sb_inodelog;
654 	to->sb_inopblog = from->sb_inopblog;
655 	to->sb_agblklog = from->sb_agblklog;
656 	to->sb_rextslog = from->sb_rextslog;
657 	to->sb_inprogress = from->sb_inprogress;
658 	to->sb_imax_pct = from->sb_imax_pct;
659 	to->sb_icount = cpu_to_be64(from->sb_icount);
660 	to->sb_ifree = cpu_to_be64(from->sb_ifree);
661 	to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
662 	to->sb_frextents = cpu_to_be64(from->sb_frextents);
663 
664 	to->sb_flags = from->sb_flags;
665 	to->sb_shared_vn = from->sb_shared_vn;
666 	to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
667 	to->sb_unit = cpu_to_be32(from->sb_unit);
668 	to->sb_width = cpu_to_be32(from->sb_width);
669 	to->sb_dirblklog = from->sb_dirblklog;
670 	to->sb_logsectlog = from->sb_logsectlog;
671 	to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
672 	to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
673 
674 	/*
675 	 * We need to ensure that bad_features2 always matches features2.
676 	 * Hence we enforce that here rather than having to remember to do it
677 	 * everywhere else that updates features2.
678 	 */
679 	from->sb_bad_features2 = from->sb_features2;
680 	to->sb_features2 = cpu_to_be32(from->sb_features2);
681 	to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
682 
683 	if (xfs_sb_version_hascrc(from)) {
684 		to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
685 		to->sb_features_ro_compat =
686 				cpu_to_be32(from->sb_features_ro_compat);
687 		to->sb_features_incompat =
688 				cpu_to_be32(from->sb_features_incompat);
689 		to->sb_features_log_incompat =
690 				cpu_to_be32(from->sb_features_log_incompat);
691 		to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
692 		to->sb_lsn = cpu_to_be64(from->sb_lsn);
693 		if (xfs_sb_version_hasmetauuid(from))
694 			uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
695 	}
696 }
697 
698 /*
699  * If the superblock has the CRC feature bit set or the CRC field is non-null,
700  * check that the CRC is valid.  We check the CRC field is non-null because a
701  * single bit error could clear the feature bit and unused parts of the
702  * superblock are supposed to be zero. Hence a non-null crc field indicates that
703  * we've potentially lost a feature bit and we should check it anyway.
704  *
705  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
706  * last field in V4 secondary superblocks.  So for secondary superblocks,
707  * we are more forgiving, and ignore CRC failures if the primary doesn't
708  * indicate that the fs version is V5.
709  */
710 static void
711 xfs_sb_read_verify(
712 	struct xfs_buf		*bp)
713 {
714 	struct xfs_sb		sb;
715 	struct xfs_mount	*mp = bp->b_mount;
716 	struct xfs_dsb		*dsb = bp->b_addr;
717 	int			error;
718 
719 	/*
720 	 * open code the version check to avoid needing to convert the entire
721 	 * superblock from disk order just to check the version number
722 	 */
723 	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
724 	    (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
725 						XFS_SB_VERSION_5) ||
726 	     dsb->sb_crc != 0)) {
727 
728 		if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
729 			/* Only fail bad secondaries on a known V5 filesystem */
730 			if (bp->b_bn == XFS_SB_DADDR ||
731 			    xfs_sb_version_hascrc(&mp->m_sb)) {
732 				error = -EFSBADCRC;
733 				goto out_error;
734 			}
735 		}
736 	}
737 
738 	/*
739 	 * Check all the superblock fields.  Don't byteswap the xquota flags
740 	 * because _verify_common checks the on-disk values.
741 	 */
742 	__xfs_sb_from_disk(&sb, dsb, false);
743 	error = xfs_validate_sb_common(mp, bp, &sb);
744 	if (error)
745 		goto out_error;
746 	error = xfs_validate_sb_read(mp, &sb);
747 
748 out_error:
749 	if (error == -EFSCORRUPTED || error == -EFSBADCRC)
750 		xfs_verifier_error(bp, error, __this_address);
751 	else if (error)
752 		xfs_buf_ioerror(bp, error);
753 }
754 
755 /*
756  * We may be probed for a filesystem match, so we may not want to emit
757  * messages when the superblock buffer is not actually an XFS superblock.
758  * If we find an XFS superblock, then run a normal, noisy mount because we are
759  * really going to mount it and want to know about errors.
760  */
761 static void
762 xfs_sb_quiet_read_verify(
763 	struct xfs_buf	*bp)
764 {
765 	struct xfs_dsb	*dsb = bp->b_addr;
766 
767 	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
768 		/* XFS filesystem, verify noisily! */
769 		xfs_sb_read_verify(bp);
770 		return;
771 	}
772 	/* quietly fail */
773 	xfs_buf_ioerror(bp, -EWRONGFS);
774 }
775 
776 static void
777 xfs_sb_write_verify(
778 	struct xfs_buf		*bp)
779 {
780 	struct xfs_sb		sb;
781 	struct xfs_mount	*mp = bp->b_mount;
782 	struct xfs_buf_log_item	*bip = bp->b_log_item;
783 	struct xfs_dsb		*dsb = bp->b_addr;
784 	int			error;
785 
786 	/*
787 	 * Check all the superblock fields.  Don't byteswap the xquota flags
788 	 * because _verify_common checks the on-disk values.
789 	 */
790 	__xfs_sb_from_disk(&sb, dsb, false);
791 	error = xfs_validate_sb_common(mp, bp, &sb);
792 	if (error)
793 		goto out_error;
794 	error = xfs_validate_sb_write(mp, bp, &sb);
795 	if (error)
796 		goto out_error;
797 
798 	if (!xfs_sb_version_hascrc(&mp->m_sb))
799 		return;
800 
801 	if (bip)
802 		dsb->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
803 
804 	xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
805 	return;
806 
807 out_error:
808 	xfs_verifier_error(bp, error, __this_address);
809 }
810 
811 const struct xfs_buf_ops xfs_sb_buf_ops = {
812 	.name = "xfs_sb",
813 	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
814 	.verify_read = xfs_sb_read_verify,
815 	.verify_write = xfs_sb_write_verify,
816 };
817 
818 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
819 	.name = "xfs_sb_quiet",
820 	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
821 	.verify_read = xfs_sb_quiet_read_verify,
822 	.verify_write = xfs_sb_write_verify,
823 };
824 
825 /*
826  * xfs_mount_common
827  *
828  * Mount initialization code establishing various mount
829  * fields from the superblock associated with the given
830  * mount structure.
831  *
832  * Inode geometry are calculated in xfs_ialloc_setup_geometry.
833  */
834 void
835 xfs_sb_mount_common(
836 	struct xfs_mount	*mp,
837 	struct xfs_sb		*sbp)
838 {
839 	mp->m_agfrotor = mp->m_agirotor = 0;
840 	mp->m_maxagi = mp->m_sb.sb_agcount;
841 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
842 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
843 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
844 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
845 	mp->m_blockmask = sbp->sb_blocksize - 1;
846 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
847 	mp->m_blockwmask = mp->m_blockwsize - 1;
848 
849 	mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
850 	mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
851 	mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
852 	mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
853 
854 	mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
855 	mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
856 	mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
857 	mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
858 
859 	mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 1);
860 	mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 0);
861 	mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
862 	mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
863 
864 	mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, true);
865 	mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, false);
866 	mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
867 	mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
868 
869 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
870 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
871 	mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp);
872 }
873 
874 /*
875  * xfs_initialize_perag_data
876  *
877  * Read in each per-ag structure so we can count up the number of
878  * allocated inodes, free inodes and used filesystem blocks as this
879  * information is no longer persistent in the superblock. Once we have
880  * this information, write it into the in-core superblock structure.
881  */
882 int
883 xfs_initialize_perag_data(
884 	struct xfs_mount *mp,
885 	xfs_agnumber_t	agcount)
886 {
887 	xfs_agnumber_t	index;
888 	xfs_perag_t	*pag;
889 	xfs_sb_t	*sbp = &mp->m_sb;
890 	uint64_t	ifree = 0;
891 	uint64_t	ialloc = 0;
892 	uint64_t	bfree = 0;
893 	uint64_t	bfreelst = 0;
894 	uint64_t	btree = 0;
895 	uint64_t	fdblocks;
896 	int		error = 0;
897 
898 	for (index = 0; index < agcount; index++) {
899 		/*
900 		 * read the agf, then the agi. This gets us
901 		 * all the information we need and populates the
902 		 * per-ag structures for us.
903 		 */
904 		error = xfs_alloc_pagf_init(mp, NULL, index, 0);
905 		if (error)
906 			return error;
907 
908 		error = xfs_ialloc_pagi_init(mp, NULL, index);
909 		if (error)
910 			return error;
911 		pag = xfs_perag_get(mp, index);
912 		ifree += pag->pagi_freecount;
913 		ialloc += pag->pagi_count;
914 		bfree += pag->pagf_freeblks;
915 		bfreelst += pag->pagf_flcount;
916 		btree += pag->pagf_btreeblks;
917 		xfs_perag_put(pag);
918 	}
919 	fdblocks = bfree + bfreelst + btree;
920 
921 	/*
922 	 * If the new summary counts are obviously incorrect, fail the
923 	 * mount operation because that implies the AGFs are also corrupt.
924 	 * Clear FS_COUNTERS so that we don't unmount with a dirty log, which
925 	 * will prevent xfs_repair from fixing anything.
926 	 */
927 	if (fdblocks > sbp->sb_dblocks || ifree > ialloc) {
928 		xfs_alert(mp, "AGF corruption. Please run xfs_repair.");
929 		error = -EFSCORRUPTED;
930 		goto out;
931 	}
932 
933 	/* Overwrite incore superblock counters with just-read data */
934 	spin_lock(&mp->m_sb_lock);
935 	sbp->sb_ifree = ifree;
936 	sbp->sb_icount = ialloc;
937 	sbp->sb_fdblocks = fdblocks;
938 	spin_unlock(&mp->m_sb_lock);
939 
940 	xfs_reinit_percpu_counters(mp);
941 out:
942 	xfs_fs_mark_healthy(mp, XFS_SICK_FS_COUNTERS);
943 	return error;
944 }
945 
946 /*
947  * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
948  * into the superblock buffer to be logged.  It does not provide the higher
949  * level of locking that is needed to protect the in-core superblock from
950  * concurrent access.
951  */
952 void
953 xfs_log_sb(
954 	struct xfs_trans	*tp)
955 {
956 	struct xfs_mount	*mp = tp->t_mountp;
957 	struct xfs_buf		*bp = xfs_trans_getsb(tp, mp);
958 
959 	mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
960 	mp->m_sb.sb_ifree = percpu_counter_sum(&mp->m_ifree);
961 	mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
962 
963 	xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
964 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
965 	xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
966 }
967 
968 /*
969  * xfs_sync_sb
970  *
971  * Sync the superblock to disk.
972  *
973  * Note that the caller is responsible for checking the frozen state of the
974  * filesystem. This procedure uses the non-blocking transaction allocator and
975  * thus will allow modifications to a frozen fs. This is required because this
976  * code can be called during the process of freezing where use of the high-level
977  * allocator would deadlock.
978  */
979 int
980 xfs_sync_sb(
981 	struct xfs_mount	*mp,
982 	bool			wait)
983 {
984 	struct xfs_trans	*tp;
985 	int			error;
986 
987 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0,
988 			XFS_TRANS_NO_WRITECOUNT, &tp);
989 	if (error)
990 		return error;
991 
992 	xfs_log_sb(tp);
993 	if (wait)
994 		xfs_trans_set_sync(tp);
995 	return xfs_trans_commit(tp);
996 }
997 
998 /*
999  * Update all the secondary superblocks to match the new state of the primary.
1000  * Because we are completely overwriting all the existing fields in the
1001  * secondary superblock buffers, there is no need to read them in from disk.
1002  * Just get a new buffer, stamp it and write it.
1003  *
1004  * The sb buffers need to be cached here so that we serialise against other
1005  * operations that access the secondary superblocks, but we don't want to keep
1006  * them in memory once it is written so we mark it as a one-shot buffer.
1007  */
1008 int
1009 xfs_update_secondary_sbs(
1010 	struct xfs_mount	*mp)
1011 {
1012 	xfs_agnumber_t		agno;
1013 	int			saved_error = 0;
1014 	int			error = 0;
1015 	LIST_HEAD		(buffer_list);
1016 
1017 	/* update secondary superblocks. */
1018 	for (agno = 1; agno < mp->m_sb.sb_agcount; agno++) {
1019 		struct xfs_buf		*bp;
1020 
1021 		error = xfs_buf_get(mp->m_ddev_targp,
1022 				 XFS_AG_DADDR(mp, agno, XFS_SB_DADDR),
1023 				 XFS_FSS_TO_BB(mp, 1), &bp);
1024 		/*
1025 		 * If we get an error reading or writing alternate superblocks,
1026 		 * continue.  xfs_repair chooses the "best" superblock based
1027 		 * on most matches; if we break early, we'll leave more
1028 		 * superblocks un-updated than updated, and xfs_repair may
1029 		 * pick them over the properly-updated primary.
1030 		 */
1031 		if (error) {
1032 			xfs_warn(mp,
1033 		"error allocating secondary superblock for ag %d",
1034 				agno);
1035 			if (!saved_error)
1036 				saved_error = error;
1037 			continue;
1038 		}
1039 
1040 		bp->b_ops = &xfs_sb_buf_ops;
1041 		xfs_buf_oneshot(bp);
1042 		xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
1043 		xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
1044 		xfs_buf_delwri_queue(bp, &buffer_list);
1045 		xfs_buf_relse(bp);
1046 
1047 		/* don't hold too many buffers at once */
1048 		if (agno % 16)
1049 			continue;
1050 
1051 		error = xfs_buf_delwri_submit(&buffer_list);
1052 		if (error) {
1053 			xfs_warn(mp,
1054 		"write error %d updating a secondary superblock near ag %d",
1055 				error, agno);
1056 			if (!saved_error)
1057 				saved_error = error;
1058 			continue;
1059 		}
1060 	}
1061 	error = xfs_buf_delwri_submit(&buffer_list);
1062 	if (error) {
1063 		xfs_warn(mp,
1064 		"write error %d updating a secondary superblock near ag %d",
1065 			error, agno);
1066 	}
1067 
1068 	return saved_error ? saved_error : error;
1069 }
1070 
1071 /*
1072  * Same behavior as xfs_sync_sb, except that it is always synchronous and it
1073  * also writes the superblock buffer to disk sector 0 immediately.
1074  */
1075 int
1076 xfs_sync_sb_buf(
1077 	struct xfs_mount	*mp)
1078 {
1079 	struct xfs_trans	*tp;
1080 	struct xfs_buf		*bp;
1081 	int			error;
1082 
1083 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp);
1084 	if (error)
1085 		return error;
1086 
1087 	bp = xfs_trans_getsb(tp, mp);
1088 	xfs_log_sb(tp);
1089 	xfs_trans_bhold(tp, bp);
1090 	xfs_trans_set_sync(tp);
1091 	error = xfs_trans_commit(tp);
1092 	if (error)
1093 		goto out;
1094 	/*
1095 	 * write out the sb buffer to get the changes to disk
1096 	 */
1097 	error = xfs_bwrite(bp);
1098 out:
1099 	xfs_buf_relse(bp);
1100 	return error;
1101 }
1102 
1103 void
1104 xfs_fs_geometry(
1105 	struct xfs_sb		*sbp,
1106 	struct xfs_fsop_geom	*geo,
1107 	int			struct_version)
1108 {
1109 	memset(geo, 0, sizeof(struct xfs_fsop_geom));
1110 
1111 	geo->blocksize = sbp->sb_blocksize;
1112 	geo->rtextsize = sbp->sb_rextsize;
1113 	geo->agblocks = sbp->sb_agblocks;
1114 	geo->agcount = sbp->sb_agcount;
1115 	geo->logblocks = sbp->sb_logblocks;
1116 	geo->sectsize = sbp->sb_sectsize;
1117 	geo->inodesize = sbp->sb_inodesize;
1118 	geo->imaxpct = sbp->sb_imax_pct;
1119 	geo->datablocks = sbp->sb_dblocks;
1120 	geo->rtblocks = sbp->sb_rblocks;
1121 	geo->rtextents = sbp->sb_rextents;
1122 	geo->logstart = sbp->sb_logstart;
1123 	BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid));
1124 	memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid));
1125 
1126 	if (struct_version < 2)
1127 		return;
1128 
1129 	geo->sunit = sbp->sb_unit;
1130 	geo->swidth = sbp->sb_width;
1131 
1132 	if (struct_version < 3)
1133 		return;
1134 
1135 	geo->version = XFS_FSOP_GEOM_VERSION;
1136 	geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
1137 		     XFS_FSOP_GEOM_FLAGS_DIRV2 |
1138 		     XFS_FSOP_GEOM_FLAGS_EXTFLG;
1139 	if (xfs_sb_version_hasattr(sbp))
1140 		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR;
1141 	if (xfs_sb_version_hasquota(sbp))
1142 		geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA;
1143 	if (xfs_sb_version_hasalign(sbp))
1144 		geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN;
1145 	if (xfs_sb_version_hasdalign(sbp))
1146 		geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN;
1147 	if (xfs_sb_version_hassector(sbp))
1148 		geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR;
1149 	if (xfs_sb_version_hasasciici(sbp))
1150 		geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI;
1151 	if (xfs_sb_version_haslazysbcount(sbp))
1152 		geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB;
1153 	if (xfs_sb_version_hasattr2(sbp))
1154 		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2;
1155 	if (xfs_sb_version_hasprojid32bit(sbp))
1156 		geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32;
1157 	if (xfs_sb_version_hascrc(sbp))
1158 		geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB;
1159 	if (xfs_sb_version_hasftype(sbp))
1160 		geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE;
1161 	if (xfs_sb_version_hasfinobt(sbp))
1162 		geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT;
1163 	if (xfs_sb_version_hassparseinodes(sbp))
1164 		geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES;
1165 	if (xfs_sb_version_hasrmapbt(sbp))
1166 		geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT;
1167 	if (xfs_sb_version_hasreflink(sbp))
1168 		geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK;
1169 	if (xfs_sb_version_hassector(sbp))
1170 		geo->logsectsize = sbp->sb_logsectsize;
1171 	else
1172 		geo->logsectsize = BBSIZE;
1173 	geo->rtsectsize = sbp->sb_blocksize;
1174 	geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp);
1175 
1176 	if (struct_version < 4)
1177 		return;
1178 
1179 	if (xfs_sb_version_haslogv2(sbp))
1180 		geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2;
1181 
1182 	geo->logsunit = sbp->sb_logsunit;
1183 
1184 	if (struct_version < 5)
1185 		return;
1186 
1187 	geo->version = XFS_FSOP_GEOM_VERSION_V5;
1188 }
1189 
1190 /* Read a secondary superblock. */
1191 int
1192 xfs_sb_read_secondary(
1193 	struct xfs_mount	*mp,
1194 	struct xfs_trans	*tp,
1195 	xfs_agnumber_t		agno,
1196 	struct xfs_buf		**bpp)
1197 {
1198 	struct xfs_buf		*bp;
1199 	int			error;
1200 
1201 	ASSERT(agno != 0 && agno != NULLAGNUMBER);
1202 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1203 			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1204 			XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops);
1205 	if (error)
1206 		return error;
1207 	xfs_buf_set_ref(bp, XFS_SSB_REF);
1208 	*bpp = bp;
1209 	return 0;
1210 }
1211 
1212 /* Get an uninitialised secondary superblock buffer. */
1213 int
1214 xfs_sb_get_secondary(
1215 	struct xfs_mount	*mp,
1216 	struct xfs_trans	*tp,
1217 	xfs_agnumber_t		agno,
1218 	struct xfs_buf		**bpp)
1219 {
1220 	struct xfs_buf		*bp;
1221 	int			error;
1222 
1223 	ASSERT(agno != 0 && agno != NULLAGNUMBER);
1224 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1225 			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1226 			XFS_FSS_TO_BB(mp, 1), 0, &bp);
1227 	if (error)
1228 		return error;
1229 	bp->b_ops = &xfs_sb_buf_ops;
1230 	xfs_buf_oneshot(bp);
1231 	*bpp = bp;
1232 	return 0;
1233 }
1234