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