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