]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - fs/xfs/libxfs/xfs_sb.c
xfs: Use consistent logging message prefixes
[karo-tx-linux.git] / fs / xfs / libxfs / xfs_sb.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_inode.h"
28 #include "xfs_ialloc.h"
29 #include "xfs_alloc.h"
30 #include "xfs_error.h"
31 #include "xfs_trace.h"
32 #include "xfs_cksum.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_alloc_btree.h"
37 #include "xfs_ialloc_btree.h"
38
39 /*
40  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
41  */
42
43 /*
44  * Reference counting access wrappers to the perag structures.
45  * Because we never free per-ag structures, the only thing we
46  * have to protect against changes is the tree structure itself.
47  */
48 struct xfs_perag *
49 xfs_perag_get(
50         struct xfs_mount        *mp,
51         xfs_agnumber_t          agno)
52 {
53         struct xfs_perag        *pag;
54         int                     ref = 0;
55
56         rcu_read_lock();
57         pag = radix_tree_lookup(&mp->m_perag_tree, agno);
58         if (pag) {
59                 ASSERT(atomic_read(&pag->pag_ref) >= 0);
60                 ref = atomic_inc_return(&pag->pag_ref);
61         }
62         rcu_read_unlock();
63         trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
64         return pag;
65 }
66
67 /*
68  * search from @first to find the next perag with the given tag set.
69  */
70 struct xfs_perag *
71 xfs_perag_get_tag(
72         struct xfs_mount        *mp,
73         xfs_agnumber_t          first,
74         int                     tag)
75 {
76         struct xfs_perag        *pag;
77         int                     found;
78         int                     ref;
79
80         rcu_read_lock();
81         found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
82                                         (void **)&pag, first, 1, tag);
83         if (found <= 0) {
84                 rcu_read_unlock();
85                 return NULL;
86         }
87         ref = atomic_inc_return(&pag->pag_ref);
88         rcu_read_unlock();
89         trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
90         return pag;
91 }
92
93 void
94 xfs_perag_put(
95         struct xfs_perag        *pag)
96 {
97         int     ref;
98
99         ASSERT(atomic_read(&pag->pag_ref) > 0);
100         ref = atomic_dec_return(&pag->pag_ref);
101         trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
102 }
103
104 /*
105  * Check the validity of the SB found.
106  */
107 STATIC int
108 xfs_mount_validate_sb(
109         xfs_mount_t     *mp,
110         xfs_sb_t        *sbp,
111         bool            check_inprogress,
112         bool            check_version)
113 {
114         if (sbp->sb_magicnum != XFS_SB_MAGIC) {
115                 xfs_warn(mp, "bad magic number");
116                 return -EWRONGFS;
117         }
118
119
120         if (!xfs_sb_good_version(sbp)) {
121                 xfs_warn(mp, "bad version");
122                 return -EWRONGFS;
123         }
124
125         /*
126          * Version 5 superblock feature mask validation. Reject combinations the
127          * kernel cannot support up front before checking anything else. For
128          * write validation, we don't need to check feature masks.
129          */
130         if (check_version && XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5) {
131                 if (xfs_sb_has_compat_feature(sbp,
132                                         XFS_SB_FEAT_COMPAT_UNKNOWN)) {
133                         xfs_warn(mp,
134 "Superblock has unknown compatible features (0x%x) enabled.",
135                                 (sbp->sb_features_compat &
136                                                 XFS_SB_FEAT_COMPAT_UNKNOWN));
137                         xfs_warn(mp,
138 "Using a more recent kernel is recommended.");
139                 }
140
141                 if (xfs_sb_has_ro_compat_feature(sbp,
142                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
143                         xfs_alert(mp,
144 "Superblock has unknown read-only compatible features (0x%x) enabled.",
145                                 (sbp->sb_features_ro_compat &
146                                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
147                         if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
148                                 xfs_warn(mp,
149 "Attempted to mount read-only compatible filesystem read-write.");
150                                 xfs_warn(mp,
151 "Filesystem can only be safely mounted read only.");
152
153                                 return -EINVAL;
154                         }
155                 }
156                 if (xfs_sb_has_incompat_feature(sbp,
157                                         XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
158                         xfs_warn(mp,
159 "Superblock has unknown incompatible features (0x%x) enabled.",
160                                 (sbp->sb_features_incompat &
161                                                 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
162                         xfs_warn(mp,
163 "Filesystem can not be safely mounted by this kernel.");
164                         return -EINVAL;
165                 }
166         }
167
168         if (xfs_sb_version_has_pquotino(sbp)) {
169                 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
170                         xfs_notice(mp,
171                            "Version 5 of Super block has XFS_OQUOTA bits.");
172                         return -EFSCORRUPTED;
173                 }
174         } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
175                                 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
176                         xfs_notice(mp,
177 "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.");
178                         return -EFSCORRUPTED;
179         }
180
181         /*
182          * Full inode chunks must be aligned to inode chunk size when
183          * sparse inodes are enabled to support the sparse chunk
184          * allocation algorithm and prevent overlapping inode records.
185          */
186         if (xfs_sb_version_hassparseinodes(sbp)) {
187                 uint32_t        align;
188
189                 xfs_alert(mp,
190         "EXPERIMENTAL sparse inode feature enabled. Use at your own risk!");
191
192                 align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
193                                 >> sbp->sb_blocklog;
194                 if (sbp->sb_inoalignmt != align) {
195                         xfs_warn(mp,
196 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
197                                  sbp->sb_inoalignmt, align);
198                         return -EINVAL;
199                 }
200         }
201
202         if (unlikely(
203             sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
204                 xfs_warn(mp,
205                 "filesystem is marked as having an external log; "
206                 "specify logdev on the mount command line.");
207                 return -EINVAL;
208         }
209
210         if (unlikely(
211             sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
212                 xfs_warn(mp,
213                 "filesystem is marked as having an internal log; "
214                 "do not specify logdev on the mount command line.");
215                 return -EINVAL;
216         }
217
218         /*
219          * More sanity checking.  Most of these were stolen directly from
220          * xfs_repair.
221          */
222         if (unlikely(
223             sbp->sb_agcount <= 0                                        ||
224             sbp->sb_sectsize < XFS_MIN_SECTORSIZE                       ||
225             sbp->sb_sectsize > XFS_MAX_SECTORSIZE                       ||
226             sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG                    ||
227             sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG                    ||
228             sbp->sb_sectsize != (1 << sbp->sb_sectlog)                  ||
229             sbp->sb_blocksize < XFS_MIN_BLOCKSIZE                       ||
230             sbp->sb_blocksize > XFS_MAX_BLOCKSIZE                       ||
231             sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG                    ||
232             sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG                    ||
233             sbp->sb_blocksize != (1 << sbp->sb_blocklog)                ||
234             sbp->sb_dirblklog > XFS_MAX_BLOCKSIZE_LOG                   ||
235             sbp->sb_inodesize < XFS_DINODE_MIN_SIZE                     ||
236             sbp->sb_inodesize > XFS_DINODE_MAX_SIZE                     ||
237             sbp->sb_inodelog < XFS_DINODE_MIN_LOG                       ||
238             sbp->sb_inodelog > XFS_DINODE_MAX_LOG                       ||
239             sbp->sb_inodesize != (1 << sbp->sb_inodelog)                ||
240             sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE                    ||
241             sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
242             (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
243             (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
244             (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
245             (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)    ||
246             sbp->sb_dblocks == 0                                        ||
247             sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)                      ||
248             sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp)                      ||
249             sbp->sb_shared_vn != 0)) {
250                 xfs_notice(mp, "SB sanity check failed");
251                 return -EFSCORRUPTED;
252         }
253
254         /*
255          * Until this is fixed only page-sized or smaller data blocks work.
256          */
257         if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
258                 xfs_warn(mp,
259                 "File system with blocksize %d bytes. "
260                 "Only pagesize (%ld) or less will currently work.",
261                                 sbp->sb_blocksize, PAGE_SIZE);
262                 return -ENOSYS;
263         }
264
265         /*
266          * Currently only very few inode sizes are supported.
267          */
268         switch (sbp->sb_inodesize) {
269         case 256:
270         case 512:
271         case 1024:
272         case 2048:
273                 break;
274         default:
275                 xfs_warn(mp, "inode size of %d bytes not supported",
276                                 sbp->sb_inodesize);
277                 return -ENOSYS;
278         }
279
280         if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
281             xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
282                 xfs_warn(mp,
283                 "file system too large to be mounted on this system.");
284                 return -EFBIG;
285         }
286
287         if (check_inprogress && sbp->sb_inprogress) {
288                 xfs_warn(mp, "Offline file system operation in progress!");
289                 return -EFSCORRUPTED;
290         }
291         return 0;
292 }
293
294 void
295 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
296 {
297         /*
298          * older mkfs doesn't initialize quota inodes to NULLFSINO. This
299          * leads to in-core values having two different values for a quota
300          * inode to be invalid: 0 and NULLFSINO. Change it to a single value
301          * NULLFSINO.
302          *
303          * Note that this change affect only the in-core values. These
304          * values are not written back to disk unless any quota information
305          * is written to the disk. Even in that case, sb_pquotino field is
306          * not written to disk unless the superblock supports pquotino.
307          */
308         if (sbp->sb_uquotino == 0)
309                 sbp->sb_uquotino = NULLFSINO;
310         if (sbp->sb_gquotino == 0)
311                 sbp->sb_gquotino = NULLFSINO;
312         if (sbp->sb_pquotino == 0)
313                 sbp->sb_pquotino = NULLFSINO;
314
315         /*
316          * We need to do these manipilations only if we are working
317          * with an older version of on-disk superblock.
318          */
319         if (xfs_sb_version_has_pquotino(sbp))
320                 return;
321
322         if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
323                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
324                                         XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
325         if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
326                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
327                                         XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
328         sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
329
330         if (sbp->sb_qflags & XFS_PQUOTA_ACCT)  {
331                 /*
332                  * In older version of superblock, on-disk superblock only
333                  * has sb_gquotino, and in-core superblock has both sb_gquotino
334                  * and sb_pquotino. But, only one of them is supported at any
335                  * point of time. So, if PQUOTA is set in disk superblock,
336                  * copy over sb_gquotino to sb_pquotino.
337                  */
338                 sbp->sb_pquotino = sbp->sb_gquotino;
339                 sbp->sb_gquotino = NULLFSINO;
340         }
341 }
342
343 static void
344 __xfs_sb_from_disk(
345         struct xfs_sb   *to,
346         xfs_dsb_t       *from,
347         bool            convert_xquota)
348 {
349         to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
350         to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
351         to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
352         to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
353         to->sb_rextents = be64_to_cpu(from->sb_rextents);
354         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
355         to->sb_logstart = be64_to_cpu(from->sb_logstart);
356         to->sb_rootino = be64_to_cpu(from->sb_rootino);
357         to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
358         to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
359         to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
360         to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
361         to->sb_agcount = be32_to_cpu(from->sb_agcount);
362         to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
363         to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
364         to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
365         to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
366         to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
367         to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
368         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
369         to->sb_blocklog = from->sb_blocklog;
370         to->sb_sectlog = from->sb_sectlog;
371         to->sb_inodelog = from->sb_inodelog;
372         to->sb_inopblog = from->sb_inopblog;
373         to->sb_agblklog = from->sb_agblklog;
374         to->sb_rextslog = from->sb_rextslog;
375         to->sb_inprogress = from->sb_inprogress;
376         to->sb_imax_pct = from->sb_imax_pct;
377         to->sb_icount = be64_to_cpu(from->sb_icount);
378         to->sb_ifree = be64_to_cpu(from->sb_ifree);
379         to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
380         to->sb_frextents = be64_to_cpu(from->sb_frextents);
381         to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
382         to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
383         to->sb_qflags = be16_to_cpu(from->sb_qflags);
384         to->sb_flags = from->sb_flags;
385         to->sb_shared_vn = from->sb_shared_vn;
386         to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
387         to->sb_unit = be32_to_cpu(from->sb_unit);
388         to->sb_width = be32_to_cpu(from->sb_width);
389         to->sb_dirblklog = from->sb_dirblklog;
390         to->sb_logsectlog = from->sb_logsectlog;
391         to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
392         to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
393         to->sb_features2 = be32_to_cpu(from->sb_features2);
394         to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
395         to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
396         to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
397         to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
398         to->sb_features_log_incompat =
399                                 be32_to_cpu(from->sb_features_log_incompat);
400         /* crc is only used on disk, not in memory; just init to 0 here. */
401         to->sb_crc = 0;
402         to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
403         to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
404         to->sb_lsn = be64_to_cpu(from->sb_lsn);
405         /* Convert on-disk flags to in-memory flags? */
406         if (convert_xquota)
407                 xfs_sb_quota_from_disk(to);
408 }
409
410 void
411 xfs_sb_from_disk(
412         struct xfs_sb   *to,
413         xfs_dsb_t       *from)
414 {
415         __xfs_sb_from_disk(to, from, true);
416 }
417
418 static void
419 xfs_sb_quota_to_disk(
420         struct xfs_dsb  *to,
421         struct xfs_sb   *from)
422 {
423         __uint16_t      qflags = from->sb_qflags;
424
425         to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
426         if (xfs_sb_version_has_pquotino(from)) {
427                 to->sb_qflags = cpu_to_be16(from->sb_qflags);
428                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
429                 to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
430                 return;
431         }
432
433         /*
434          * The in-core version of sb_qflags do not have XFS_OQUOTA_*
435          * flags, whereas the on-disk version does.  So, convert incore
436          * XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
437          */
438         qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
439                         XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
440
441         if (from->sb_qflags &
442                         (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
443                 qflags |= XFS_OQUOTA_ENFD;
444         if (from->sb_qflags &
445                         (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
446                 qflags |= XFS_OQUOTA_CHKD;
447         to->sb_qflags = cpu_to_be16(qflags);
448
449         /*
450          * GQUOTINO and PQUOTINO cannot be used together in versions
451          * of superblock that do not have pquotino. from->sb_flags
452          * tells us which quota is active and should be copied to
453          * disk. If neither are active, we should NULL the inode.
454          *
455          * In all cases, the separate pquotino must remain 0 because it
456          * it beyond the "end" of the valid non-pquotino superblock.
457          */
458         if (from->sb_qflags & XFS_GQUOTA_ACCT)
459                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
460         else if (from->sb_qflags & XFS_PQUOTA_ACCT)
461                 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
462         else {
463                 /*
464                  * We can't rely on just the fields being logged to tell us
465                  * that it is safe to write NULLFSINO - we should only do that
466                  * if quotas are not actually enabled. Hence only write
467                  * NULLFSINO if both in-core quota inodes are NULL.
468                  */
469                 if (from->sb_gquotino == NULLFSINO &&
470                     from->sb_pquotino == NULLFSINO)
471                         to->sb_gquotino = cpu_to_be64(NULLFSINO);
472         }
473
474         to->sb_pquotino = 0;
475 }
476
477 void
478 xfs_sb_to_disk(
479         struct xfs_dsb  *to,
480         struct xfs_sb   *from)
481 {
482         xfs_sb_quota_to_disk(to, from);
483
484         to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
485         to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
486         to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
487         to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
488         to->sb_rextents = cpu_to_be64(from->sb_rextents);
489         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
490         to->sb_logstart = cpu_to_be64(from->sb_logstart);
491         to->sb_rootino = cpu_to_be64(from->sb_rootino);
492         to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
493         to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
494         to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
495         to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
496         to->sb_agcount = cpu_to_be32(from->sb_agcount);
497         to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
498         to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
499         to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
500         to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
501         to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
502         to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
503         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
504         to->sb_blocklog = from->sb_blocklog;
505         to->sb_sectlog = from->sb_sectlog;
506         to->sb_inodelog = from->sb_inodelog;
507         to->sb_inopblog = from->sb_inopblog;
508         to->sb_agblklog = from->sb_agblklog;
509         to->sb_rextslog = from->sb_rextslog;
510         to->sb_inprogress = from->sb_inprogress;
511         to->sb_imax_pct = from->sb_imax_pct;
512         to->sb_icount = cpu_to_be64(from->sb_icount);
513         to->sb_ifree = cpu_to_be64(from->sb_ifree);
514         to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
515         to->sb_frextents = cpu_to_be64(from->sb_frextents);
516
517         to->sb_flags = from->sb_flags;
518         to->sb_shared_vn = from->sb_shared_vn;
519         to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
520         to->sb_unit = cpu_to_be32(from->sb_unit);
521         to->sb_width = cpu_to_be32(from->sb_width);
522         to->sb_dirblklog = from->sb_dirblklog;
523         to->sb_logsectlog = from->sb_logsectlog;
524         to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
525         to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
526
527         /*
528          * We need to ensure that bad_features2 always matches features2.
529          * Hence we enforce that here rather than having to remember to do it
530          * everywhere else that updates features2.
531          */
532         from->sb_bad_features2 = from->sb_features2;
533         to->sb_features2 = cpu_to_be32(from->sb_features2);
534         to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
535
536         if (xfs_sb_version_hascrc(from)) {
537                 to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
538                 to->sb_features_ro_compat =
539                                 cpu_to_be32(from->sb_features_ro_compat);
540                 to->sb_features_incompat =
541                                 cpu_to_be32(from->sb_features_incompat);
542                 to->sb_features_log_incompat =
543                                 cpu_to_be32(from->sb_features_log_incompat);
544                 to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
545                 to->sb_lsn = cpu_to_be64(from->sb_lsn);
546         }
547 }
548
549 static int
550 xfs_sb_verify(
551         struct xfs_buf  *bp,
552         bool            check_version)
553 {
554         struct xfs_mount *mp = bp->b_target->bt_mount;
555         struct xfs_sb   sb;
556
557         /*
558          * Use call variant which doesn't convert quota flags from disk 
559          * format, because xfs_mount_validate_sb checks the on-disk flags.
560          */
561         __xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp), false);
562
563         /*
564          * Only check the in progress field for the primary superblock as
565          * mkfs.xfs doesn't clear it from secondary superblocks.
566          */
567         return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR,
568                                      check_version);
569 }
570
571 /*
572  * If the superblock has the CRC feature bit set or the CRC field is non-null,
573  * check that the CRC is valid.  We check the CRC field is non-null because a
574  * single bit error could clear the feature bit and unused parts of the
575  * superblock are supposed to be zero. Hence a non-null crc field indicates that
576  * we've potentially lost a feature bit and we should check it anyway.
577  *
578  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
579  * last field in V4 secondary superblocks.  So for secondary superblocks,
580  * we are more forgiving, and ignore CRC failures if the primary doesn't
581  * indicate that the fs version is V5.
582  */
583 static void
584 xfs_sb_read_verify(
585         struct xfs_buf  *bp)
586 {
587         struct xfs_mount *mp = bp->b_target->bt_mount;
588         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
589         int             error;
590
591         /*
592          * open code the version check to avoid needing to convert the entire
593          * superblock from disk order just to check the version number
594          */
595         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
596             (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
597                                                 XFS_SB_VERSION_5) ||
598              dsb->sb_crc != 0)) {
599
600                 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
601                         /* Only fail bad secondaries on a known V5 filesystem */
602                         if (bp->b_bn == XFS_SB_DADDR ||
603                             xfs_sb_version_hascrc(&mp->m_sb)) {
604                                 error = -EFSBADCRC;
605                                 goto out_error;
606                         }
607                 }
608         }
609         error = xfs_sb_verify(bp, true);
610
611 out_error:
612         if (error) {
613                 xfs_buf_ioerror(bp, error);
614                 if (error == -EFSCORRUPTED || error == -EFSBADCRC)
615                         xfs_verifier_error(bp);
616         }
617 }
618
619 /*
620  * We may be probed for a filesystem match, so we may not want to emit
621  * messages when the superblock buffer is not actually an XFS superblock.
622  * If we find an XFS superblock, then run a normal, noisy mount because we are
623  * really going to mount it and want to know about errors.
624  */
625 static void
626 xfs_sb_quiet_read_verify(
627         struct xfs_buf  *bp)
628 {
629         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
630
631         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
632                 /* XFS filesystem, verify noisily! */
633                 xfs_sb_read_verify(bp);
634                 return;
635         }
636         /* quietly fail */
637         xfs_buf_ioerror(bp, -EWRONGFS);
638 }
639
640 static void
641 xfs_sb_write_verify(
642         struct xfs_buf          *bp)
643 {
644         struct xfs_mount        *mp = bp->b_target->bt_mount;
645         struct xfs_buf_log_item *bip = bp->b_fspriv;
646         int                     error;
647
648         error = xfs_sb_verify(bp, false);
649         if (error) {
650                 xfs_buf_ioerror(bp, error);
651                 xfs_verifier_error(bp);
652                 return;
653         }
654
655         if (!xfs_sb_version_hascrc(&mp->m_sb))
656                 return;
657
658         if (bip)
659                 XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
660
661         xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
662 }
663
664 const struct xfs_buf_ops xfs_sb_buf_ops = {
665         .verify_read = xfs_sb_read_verify,
666         .verify_write = xfs_sb_write_verify,
667 };
668
669 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
670         .verify_read = xfs_sb_quiet_read_verify,
671         .verify_write = xfs_sb_write_verify,
672 };
673
674 /*
675  * xfs_mount_common
676  *
677  * Mount initialization code establishing various mount
678  * fields from the superblock associated with the given
679  * mount structure
680  */
681 void
682 xfs_sb_mount_common(
683         struct xfs_mount *mp,
684         struct xfs_sb   *sbp)
685 {
686         mp->m_agfrotor = mp->m_agirotor = 0;
687         spin_lock_init(&mp->m_agirotor_lock);
688         mp->m_maxagi = mp->m_sb.sb_agcount;
689         mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
690         mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
691         mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
692         mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
693         mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
694         mp->m_blockmask = sbp->sb_blocksize - 1;
695         mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
696         mp->m_blockwmask = mp->m_blockwsize - 1;
697
698         mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
699         mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
700         mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
701         mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
702
703         mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
704         mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
705         mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
706         mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
707
708         mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
709         mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
710         mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
711         mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
712
713         mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
714         mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
715                                         sbp->sb_inopblock);
716         mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
717
718         if (sbp->sb_spino_align)
719                 mp->m_ialloc_min_blks = sbp->sb_spino_align;
720         else
721                 mp->m_ialloc_min_blks = mp->m_ialloc_blks;
722 }
723
724 /*
725  * xfs_initialize_perag_data
726  *
727  * Read in each per-ag structure so we can count up the number of
728  * allocated inodes, free inodes and used filesystem blocks as this
729  * information is no longer persistent in the superblock. Once we have
730  * this information, write it into the in-core superblock structure.
731  */
732 int
733 xfs_initialize_perag_data(
734         struct xfs_mount *mp,
735         xfs_agnumber_t  agcount)
736 {
737         xfs_agnumber_t  index;
738         xfs_perag_t     *pag;
739         xfs_sb_t        *sbp = &mp->m_sb;
740         uint64_t        ifree = 0;
741         uint64_t        ialloc = 0;
742         uint64_t        bfree = 0;
743         uint64_t        bfreelst = 0;
744         uint64_t        btree = 0;
745         int             error;
746
747         for (index = 0; index < agcount; index++) {
748                 /*
749                  * read the agf, then the agi. This gets us
750                  * all the information we need and populates the
751                  * per-ag structures for us.
752                  */
753                 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
754                 if (error)
755                         return error;
756
757                 error = xfs_ialloc_pagi_init(mp, NULL, index);
758                 if (error)
759                         return error;
760                 pag = xfs_perag_get(mp, index);
761                 ifree += pag->pagi_freecount;
762                 ialloc += pag->pagi_count;
763                 bfree += pag->pagf_freeblks;
764                 bfreelst += pag->pagf_flcount;
765                 btree += pag->pagf_btreeblks;
766                 xfs_perag_put(pag);
767         }
768
769         /* Overwrite incore superblock counters with just-read data */
770         spin_lock(&mp->m_sb_lock);
771         sbp->sb_ifree = ifree;
772         sbp->sb_icount = ialloc;
773         sbp->sb_fdblocks = bfree + bfreelst + btree;
774         spin_unlock(&mp->m_sb_lock);
775
776         xfs_reinit_percpu_counters(mp);
777
778         return 0;
779 }
780
781 /*
782  * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
783  * into the superblock buffer to be logged.  It does not provide the higher
784  * level of locking that is needed to protect the in-core superblock from
785  * concurrent access.
786  */
787 void
788 xfs_log_sb(
789         struct xfs_trans        *tp)
790 {
791         struct xfs_mount        *mp = tp->t_mountp;
792         struct xfs_buf          *bp = xfs_trans_getsb(tp, mp, 0);
793
794         mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
795         mp->m_sb.sb_ifree = percpu_counter_sum(&mp->m_ifree);
796         mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
797
798         xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
799         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
800         xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb));
801 }
802
803 /*
804  * xfs_sync_sb
805  *
806  * Sync the superblock to disk.
807  *
808  * Note that the caller is responsible for checking the frozen state of the
809  * filesystem. This procedure uses the non-blocking transaction allocator and
810  * thus will allow modifications to a frozen fs. This is required because this
811  * code can be called during the process of freezing where use of the high-level
812  * allocator would deadlock.
813  */
814 int
815 xfs_sync_sb(
816         struct xfs_mount        *mp,
817         bool                    wait)
818 {
819         struct xfs_trans        *tp;
820         int                     error;
821
822         tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_CHANGE, KM_SLEEP);
823         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
824         if (error) {
825                 xfs_trans_cancel(tp);
826                 return error;
827         }
828
829         xfs_log_sb(tp);
830         if (wait)
831                 xfs_trans_set_sync(tp);
832         return xfs_trans_commit(tp);
833 }