fs/xfs/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_ag.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_inode.h"
  29 #include "xfs_ialloc.h"
  30 #include "xfs_alloc.h"
  31 #include "xfs_error.h"
  32 #include "xfs_trace.h"
  33 #include "xfs_cksum.h"
  34 #include "xfs_trans.h"
  35 #include "xfs_buf_item.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_bmap_btree.h"
  38 #include "xfs_alloc_btree.h"
  39 #include "xfs_ialloc_btree.h"
  40
  41 /*
  42  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
  43  */
  44
  45 static const struct {
  46         short offset;
  47         short type;     /* 0 = integer
  48                          * 1 = binary / string (no translation)
  49                          */
  50 } xfs_sb_info[] = {
  51         { offsetof(xfs_sb_t, sb_magicnum),      0 },
  52         { offsetof(xfs_sb_t, sb_blocksize),     0 },
  53         { offsetof(xfs_sb_t, sb_dblocks),       0 },
  54         { offsetof(xfs_sb_t, sb_rblocks),       0 },
  55         { offsetof(xfs_sb_t, sb_rextents),      0 },
  56         { offsetof(xfs_sb_t, sb_uuid),          1 },
  57         { offsetof(xfs_sb_t, sb_logstart),      0 },
  58         { offsetof(xfs_sb_t, sb_rootino),       0 },
  59         { offsetof(xfs_sb_t, sb_rbmino),        0 },
  60         { offsetof(xfs_sb_t, sb_rsumino),       0 },
  61         { offsetof(xfs_sb_t, sb_rextsize),      0 },
  62         { offsetof(xfs_sb_t, sb_agblocks),      0 },
  63         { offsetof(xfs_sb_t, sb_agcount),       0 },
  64         { offsetof(xfs_sb_t, sb_rbmblocks),     0 },
  65         { offsetof(xfs_sb_t, sb_logblocks),     0 },
  66         { offsetof(xfs_sb_t, sb_versionnum),    0 },
  67         { offsetof(xfs_sb_t, sb_sectsize),      0 },
  68         { offsetof(xfs_sb_t, sb_inodesize),     0 },
  69         { offsetof(xfs_sb_t, sb_inopblock),     0 },
  70         { offsetof(xfs_sb_t, sb_fname[0]),      1 },
  71         { offsetof(xfs_sb_t, sb_blocklog),      0 },
  72         { offsetof(xfs_sb_t, sb_sectlog),       0 },
  73         { offsetof(xfs_sb_t, sb_inodelog),      0 },
  74         { offsetof(xfs_sb_t, sb_inopblog),      0 },
  75         { offsetof(xfs_sb_t, sb_agblklog),      0 },
  76         { offsetof(xfs_sb_t, sb_rextslog),      0 },
  77         { offsetof(xfs_sb_t, sb_inprogress),    0 },
  78         { offsetof(xfs_sb_t, sb_imax_pct),      0 },
  79         { offsetof(xfs_sb_t, sb_icount),        0 },
  80         { offsetof(xfs_sb_t, sb_ifree),         0 },
  81         { offsetof(xfs_sb_t, sb_fdblocks),      0 },
  82         { offsetof(xfs_sb_t, sb_frextents),     0 },
  83         { offsetof(xfs_sb_t, sb_uquotino),      0 },
  84         { offsetof(xfs_sb_t, sb_gquotino),      0 },
  85         { offsetof(xfs_sb_t, sb_qflags),        0 },
  86         { offsetof(xfs_sb_t, sb_flags),         0 },
  87         { offsetof(xfs_sb_t, sb_shared_vn),     0 },
  88         { offsetof(xfs_sb_t, sb_inoalignmt),    0 },
  89         { offsetof(xfs_sb_t, sb_unit),          0 },
  90         { offsetof(xfs_sb_t, sb_width),         0 },
  91         { offsetof(xfs_sb_t, sb_dirblklog),     0 },
  92         { offsetof(xfs_sb_t, sb_logsectlog),    0 },
  93         { offsetof(xfs_sb_t, sb_logsectsize),   0 },
  94         { offsetof(xfs_sb_t, sb_logsunit),      0 },
  95         { offsetof(xfs_sb_t, sb_features2),     0 },
  96         { offsetof(xfs_sb_t, sb_bad_features2), 0 },
  97         { offsetof(xfs_sb_t, sb_features_compat),       0 },
  98         { offsetof(xfs_sb_t, sb_features_ro_compat),    0 },
  99         { offsetof(xfs_sb_t, sb_features_incompat),     0 },
 100         { offsetof(xfs_sb_t, sb_features_log_incompat), 0 },
 101         { offsetof(xfs_sb_t, sb_crc),           0 },
 102         { offsetof(xfs_sb_t, sb_pad),           0 },
 103         { offsetof(xfs_sb_t, sb_pquotino),      0 },
 104         { offsetof(xfs_sb_t, sb_lsn),           0 },
 105         { sizeof(xfs_sb_t),                     0 }
 106 };
 107
 108 /*
 109  * Reference counting access wrappers to the perag structures.
 110  * Because we never free per-ag structures, the only thing we
 111  * have to protect against changes is the tree structure itself.
 112  */
 113 struct xfs_perag *
 114 xfs_perag_get(
 115         struct xfs_mount        *mp,
 116         xfs_agnumber_t          agno)
 117 {
 118         struct xfs_perag        *pag;
 119         int                     ref = 0;
 120
 121         rcu_read_lock();
 122         pag = radix_tree_lookup(&mp->m_perag_tree, agno);
 123         if (pag) {
 124                 ASSERT(atomic_read(&pag->pag_ref) >= 0);
 125                 ref = atomic_inc_return(&pag->pag_ref);
 126         }
 127         rcu_read_unlock();
 128         trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
 129         return pag;
 130 }
 131
 132 /*
 133  * search from @first to find the next perag with the given tag set.
 134  */
 135 struct xfs_perag *
 136 xfs_perag_get_tag(
 137         struct xfs_mount        *mp,
 138         xfs_agnumber_t          first,
 139         int                     tag)
 140 {
 141         struct xfs_perag        *pag;
 142         int                     found;
 143         int                     ref;
 144
 145         rcu_read_lock();
 146         found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
 147                                         (void **)&pag, first, 1, tag);
 148         if (found <= 0) {
 149                 rcu_read_unlock();
 150                 return NULL;
 151         }
 152         ref = atomic_inc_return(&pag->pag_ref);
 153         rcu_read_unlock();
 154         trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
 155         return pag;
 156 }
 157
 158 void
 159 xfs_perag_put(
 160         struct xfs_perag        *pag)
 161 {
 162         int     ref;
 163
 164         ASSERT(atomic_read(&pag->pag_ref) > 0);
 165         ref = atomic_dec_return(&pag->pag_ref);
 166         trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
 167 }
 168
 169 /*
 170  * Check the validity of the SB found.
 171  */
 172 STATIC int
 173 xfs_mount_validate_sb(
 174         xfs_mount_t     *mp,
 175         xfs_sb_t        *sbp,
 176         bool            check_inprogress,
 177         bool            check_version)
 178 {
 179
 180         /*
 181          * If the log device and data device have the
 182          * same device number, the log is internal.
 183          * Consequently, the sb_logstart should be non-zero.  If
 184          * we have a zero sb_logstart in this case, we may be trying to mount
 185          * a volume filesystem in a non-volume manner.
 186          */
 187         if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 188                 xfs_warn(mp, "bad magic number");
 189                 return XFS_ERROR(EWRONGFS);
 190         }
 191
 192
 193         if (!xfs_sb_good_version(sbp)) {
 194                 xfs_warn(mp, "bad version");
 195                 return XFS_ERROR(EWRONGFS);
 196         }
 197
 198         /*
 199          * Version 5 superblock feature mask validation. Reject combinations the
 200          * kernel cannot support up front before checking anything else. For
 201          * write validation, we don't need to check feature masks.
 202          */
 203         if (check_version && XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5) {
 204                 if (xfs_sb_has_compat_feature(sbp,
 205                                         XFS_SB_FEAT_COMPAT_UNKNOWN)) {
 206                         xfs_warn(mp,
 207 "Superblock has unknown compatible features (0x%x) enabled.\n"
 208 "Using a more recent kernel is recommended.",
 209                                 (sbp->sb_features_compat &
 210                                                 XFS_SB_FEAT_COMPAT_UNKNOWN));
 211                 }
 212
 213                 if (xfs_sb_has_ro_compat_feature(sbp,
 214                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
 215                         xfs_alert(mp,
 216 "Superblock has unknown read-only compatible features (0x%x) enabled.",
 217                                 (sbp->sb_features_ro_compat &
 218                                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
 219                         if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 220                                 xfs_warn(mp,
 221 "Attempted to mount read-only compatible filesystem read-write.\n"
 222 "Filesystem can only be safely mounted read only.");
 223                                 return XFS_ERROR(EINVAL);
 224                         }
 225                 }
 226                 if (xfs_sb_has_incompat_feature(sbp,
 227                                         XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
 228                         xfs_warn(mp,
 229 "Superblock has unknown incompatible features (0x%x) enabled.\n"
 230 "Filesystem can not be safely mounted by this kernel.",
 231                                 (sbp->sb_features_incompat &
 232                                                 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
 233                         return XFS_ERROR(EINVAL);
 234                 }
 235         }
 236
 237         if (xfs_sb_version_has_pquotino(sbp)) {
 238                 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
 239                         xfs_notice(mp,
 240                            "Version 5 of Super block has XFS_OQUOTA bits.");
 241                         return XFS_ERROR(EFSCORRUPTED);
 242                 }
 243         } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
 244                                 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
 245                         xfs_notice(mp,
 246 "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.");
 247                         return XFS_ERROR(EFSCORRUPTED);
 248         }
 249
 250         if (unlikely(
 251             sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
 252                 xfs_warn(mp,
 253                 "filesystem is marked as having an external log; "
 254                 "specify logdev on the mount command line.");
 255                 return XFS_ERROR(EINVAL);
 256         }
 257
 258         if (unlikely(
 259             sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
 260                 xfs_warn(mp,
 261                 "filesystem is marked as having an internal log; "
 262                 "do not specify logdev on the mount command line.");
 263                 return XFS_ERROR(EINVAL);
 264         }
 265
 266         /*
 267          * More sanity checking.  Most of these were stolen directly from
 268          * xfs_repair.
 269          */
 270         if (unlikely(
 271             sbp->sb_agcount <= 0                                        ||
 272             sbp->sb_sectsize < XFS_MIN_SECTORSIZE                       ||
 273             sbp->sb_sectsize > XFS_MAX_SECTORSIZE                       ||
 274             sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG                    ||
 275             sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG                    ||
 276             sbp->sb_sectsize != (1 << sbp->sb_sectlog)                  ||
 277             sbp->sb_blocksize < XFS_MIN_BLOCKSIZE                       ||
 278             sbp->sb_blocksize > XFS_MAX_BLOCKSIZE                       ||
 279             sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG                    ||
 280             sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG                    ||
 281             sbp->sb_blocksize != (1 << sbp->sb_blocklog)                ||
 282             sbp->sb_inodesize < XFS_DINODE_MIN_SIZE                     ||
 283             sbp->sb_inodesize > XFS_DINODE_MAX_SIZE                     ||
 284             sbp->sb_inodelog < XFS_DINODE_MIN_LOG                       ||
 285             sbp->sb_inodelog > XFS_DINODE_MAX_LOG                       ||
 286             sbp->sb_inodesize != (1 << sbp->sb_inodelog)                ||
 287             sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
 288             (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
 289             (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
 290             (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
 291             (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)    ||
 292             sbp->sb_dblocks == 0                                        ||
 293             sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)                      ||
 294             sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
 295                 xfs_notice(mp, "SB sanity check failed");
 296                 return XFS_ERROR(EFSCORRUPTED);
 297         }
 298
 299         /*
 300          * Until this is fixed only page-sized or smaller data blocks work.
 301          */
 302         if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
 303                 xfs_warn(mp,
 304                 "File system with blocksize %d bytes. "
 305                 "Only pagesize (%ld) or less will currently work.",
 306                                 sbp->sb_blocksize, PAGE_SIZE);
 307                 return XFS_ERROR(ENOSYS);
 308         }
 309
 310         /*
 311          * Currently only very few inode sizes are supported.
 312          */
 313         switch (sbp->sb_inodesize) {
 314         case 256:
 315         case 512:
 316         case 1024:
 317         case 2048:
 318                 break;
 319         default:
 320                 xfs_warn(mp, "inode size of %d bytes not supported",
 321                                 sbp->sb_inodesize);
 322                 return XFS_ERROR(ENOSYS);
 323         }
 324
 325         if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
 326             xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
 327                 xfs_warn(mp,
 328                 "file system too large to be mounted on this system.");
 329                 return XFS_ERROR(EFBIG);
 330         }
 331
 332         if (check_inprogress && sbp->sb_inprogress) {
 333                 xfs_warn(mp, "Offline file system operation in progress!");
 334                 return XFS_ERROR(EFSCORRUPTED);
 335         }
 336
 337         /*
 338          * Version 1 directory format has never worked on Linux.
 339          */
 340         if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
 341                 xfs_warn(mp, "file system using version 1 directory format");
 342                 return XFS_ERROR(ENOSYS);
 343         }
 344
 345         return 0;
 346 }
 347
 348 void
 349 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
 350 {
 351         /*
 352          * older mkfs doesn't initialize quota inodes to NULLFSINO. This
 353          * leads to in-core values having two different values for a quota
 354          * inode to be invalid: 0 and NULLFSINO. Change it to a single value
 355          * NULLFSINO.
 356          *
 357          * Note that this change affect only the in-core values. These
 358          * values are not written back to disk unless any quota information
 359          * is written to the disk. Even in that case, sb_pquotino field is
 360          * not written to disk unless the superblock supports pquotino.
 361          */
 362         if (sbp->sb_uquotino == 0)
 363                 sbp->sb_uquotino = NULLFSINO;
 364         if (sbp->sb_gquotino == 0)
 365                 sbp->sb_gquotino = NULLFSINO;
 366         if (sbp->sb_pquotino == 0)
 367                 sbp->sb_pquotino = NULLFSINO;
 368
 369         /*
 370          * We need to do these manipilations only if we are working
 371          * with an older version of on-disk superblock.
 372          */
 373         if (xfs_sb_version_has_pquotino(sbp))
 374                 return;
 375
 376         if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
 377                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 378                                         XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
 379         if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
 380                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 381                                         XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
 382         sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
 383
 384         if (sbp->sb_qflags & XFS_PQUOTA_ACCT)  {
 385                 /*
 386                  * In older version of superblock, on-disk superblock only
 387                  * has sb_gquotino, and in-core superblock has both sb_gquotino
 388                  * and sb_pquotino. But, only one of them is supported at any
 389                  * point of time. So, if PQUOTA is set in disk superblock,
 390                  * copy over sb_gquotino to sb_pquotino.
 391                  */
 392                 sbp->sb_pquotino = sbp->sb_gquotino;
 393                 sbp->sb_gquotino = NULLFSINO;
 394         }
 395 }
 396
 397 void
 398 xfs_sb_from_disk(
 399         struct xfs_sb   *to,
 400         xfs_dsb_t       *from)
 401 {
 402         to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
 403         to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
 404         to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
 405         to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
 406         to->sb_rextents = be64_to_cpu(from->sb_rextents);
 407         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
 408         to->sb_logstart = be64_to_cpu(from->sb_logstart);
 409         to->sb_rootino = be64_to_cpu(from->sb_rootino);
 410         to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
 411         to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
 412         to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
 413         to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
 414         to->sb_agcount = be32_to_cpu(from->sb_agcount);
 415         to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
 416         to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
 417         to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
 418         to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
 419         to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
 420         to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
 421         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
 422         to->sb_blocklog = from->sb_blocklog;
 423         to->sb_sectlog = from->sb_sectlog;
 424         to->sb_inodelog = from->sb_inodelog;
 425         to->sb_inopblog = from->sb_inopblog;
 426         to->sb_agblklog = from->sb_agblklog;
 427         to->sb_rextslog = from->sb_rextslog;
 428         to->sb_inprogress = from->sb_inprogress;
 429         to->sb_imax_pct = from->sb_imax_pct;
 430         to->sb_icount = be64_to_cpu(from->sb_icount);
 431         to->sb_ifree = be64_to_cpu(from->sb_ifree);
 432         to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
 433         to->sb_frextents = be64_to_cpu(from->sb_frextents);
 434         to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
 435         to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
 436         to->sb_qflags = be16_to_cpu(from->sb_qflags);
 437         to->sb_flags = from->sb_flags;
 438         to->sb_shared_vn = from->sb_shared_vn;
 439         to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
 440         to->sb_unit = be32_to_cpu(from->sb_unit);
 441         to->sb_width = be32_to_cpu(from->sb_width);
 442         to->sb_dirblklog = from->sb_dirblklog;
 443         to->sb_logsectlog = from->sb_logsectlog;
 444         to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
 445         to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
 446         to->sb_features2 = be32_to_cpu(from->sb_features2);
 447         to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
 448         to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
 449         to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
 450         to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
 451         to->sb_features_log_incompat =
 452                                 be32_to_cpu(from->sb_features_log_incompat);
 453         to->sb_pad = 0;
 454         to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
 455         to->sb_lsn = be64_to_cpu(from->sb_lsn);
 456 }
 457
 458 static inline void
 459 xfs_sb_quota_to_disk(
 460         xfs_dsb_t       *to,
 461         xfs_sb_t        *from,
 462         __int64_t       *fields)
 463 {
 464         __uint16_t      qflags = from->sb_qflags;
 465
 466         /*
 467          * We need to do these manipilations only if we are working
 468          * with an older version of on-disk superblock.
 469          */
 470         if (xfs_sb_version_has_pquotino(from))
 471                 return;
 472
 473         if (*fields & XFS_SB_QFLAGS) {
 474                 /*
 475                  * The in-core version of sb_qflags do not have
 476                  * XFS_OQUOTA_* flags, whereas the on-disk version
 477                  * does.  So, convert incore XFS_{PG}QUOTA_* flags
 478                  * to on-disk XFS_OQUOTA_* flags.
 479                  */
 480                 qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
 481                                 XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
 482
 483                 if (from->sb_qflags &
 484                                 (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
 485                         qflags |= XFS_OQUOTA_ENFD;
 486                 if (from->sb_qflags &
 487                                 (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
 488                         qflags |= XFS_OQUOTA_CHKD;
 489                 to->sb_qflags = cpu_to_be16(qflags);
 490                 *fields &= ~XFS_SB_QFLAGS;
 491         }
 492
 493         /*
 494          * GQUOTINO and PQUOTINO cannot be used together in versions
 495          * of superblock that do not have pquotino. from->sb_flags
 496          * tells us which quota is active and should be copied to
 497          * disk.
 498          */
 499         if ((*fields & XFS_SB_GQUOTINO) &&
 500                                 (from->sb_qflags & XFS_GQUOTA_ACCT))
 501                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
 502         else if ((*fields & XFS_SB_PQUOTINO) &&
 503                                 (from->sb_qflags & XFS_PQUOTA_ACCT))
 504                 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
 505
 506         *fields &= ~(XFS_SB_PQUOTINO | XFS_SB_GQUOTINO);
 507 }
 508
 509 /*
 510  * Copy in core superblock to ondisk one.
 511  *
 512  * The fields argument is mask of superblock fields to copy.
 513  */
 514 void
 515 xfs_sb_to_disk(
 516         xfs_dsb_t       *to,
 517         xfs_sb_t        *from,
 518         __int64_t       fields)
 519 {
 520         xfs_caddr_t     to_ptr = (xfs_caddr_t)to;
 521         xfs_caddr_t     from_ptr = (xfs_caddr_t)from;
 522         xfs_sb_field_t  f;
 523         int             first;
 524         int             size;
 525
 526         ASSERT(fields);
 527         if (!fields)
 528                 return;
 529
 530         xfs_sb_quota_to_disk(to, from, &fields);
 531         while (fields) {
 532                 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 533                 first = xfs_sb_info[f].offset;
 534                 size = xfs_sb_info[f + 1].offset - first;
 535
 536                 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
 537
 538                 if (size == 1 || xfs_sb_info[f].type == 1) {
 539                         memcpy(to_ptr + first, from_ptr + first, size);
 540                 } else {
 541                         switch (size) {
 542                         case 2:
 543                                 *(__be16 *)(to_ptr + first) =
 544                                       cpu_to_be16(*(__u16 *)(from_ptr + first));
 545                                 break;
 546                         case 4:
 547                                 *(__be32 *)(to_ptr + first) =
 548                                       cpu_to_be32(*(__u32 *)(from_ptr + first));
 549                                 break;
 550                         case 8:
 551                                 *(__be64 *)(to_ptr + first) =
 552                                       cpu_to_be64(*(__u64 *)(from_ptr + first));
 553                                 break;
 554                         default:
 555                                 ASSERT(0);
 556                         }
 557                 }
 558
 559                 fields &= ~(1LL << f);
 560         }
 561 }
 562
 563 static int
 564 xfs_sb_verify(
 565         struct xfs_buf  *bp,
 566         bool            check_version)
 567 {
 568         struct xfs_mount *mp = bp->b_target->bt_mount;
 569         struct xfs_sb   sb;
 570
 571         xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp));
 572
 573         /*
 574          * Only check the in progress field for the primary superblock as
 575          * mkfs.xfs doesn't clear it from secondary superblocks.
 576          */
 577         return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR,
 578                                      check_version);
 579 }
 580
 581 /*
 582  * If the superblock has the CRC feature bit set or the CRC field is non-null,
 583  * check that the CRC is valid.  We check the CRC field is non-null because a
 584  * single bit error could clear the feature bit and unused parts of the
 585  * superblock are supposed to be zero. Hence a non-null crc field indicates that
 586  * we've potentially lost a feature bit and we should check it anyway.
 587  *
 588  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
 589  * last field in V4 secondary superblocks.  So for secondary superblocks,
 590  * we are more forgiving, and ignore CRC failures if the primary doesn't
 591  * indicate that the fs version is V5.
 592  */
 593 static void
 594 xfs_sb_read_verify(
 595         struct xfs_buf  *bp)
 596 {
 597         struct xfs_mount *mp = bp->b_target->bt_mount;
 598         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 599         int             error;
 600
 601         /*
 602          * open code the version check to avoid needing to convert the entire
 603          * superblock from disk order just to check the version number
 604          */
 605         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
 606             (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
 607                                                 XFS_SB_VERSION_5) ||
 608              dsb->sb_crc != 0)) {
 609
 610                 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
 611                         /* Only fail bad secondaries on a known V5 filesystem */
 612                         if (bp->b_bn == XFS_SB_DADDR ||
 613                             xfs_sb_version_hascrc(&mp->m_sb)) {
 614                                 error = EFSBADCRC;
 615                                 goto out_error;
 616                         }
 617                 }
 618         }
 619         error = xfs_sb_verify(bp, true);
 620
 621 out_error:
 622         if (error) {
 623                 xfs_buf_ioerror(bp, error);
 624                 if (error == EFSCORRUPTED || error == EFSBADCRC)
 625                         xfs_verifier_error(bp);
 626         }
 627 }
 628
 629 /*
 630  * We may be probed for a filesystem match, so we may not want to emit
 631  * messages when the superblock buffer is not actually an XFS superblock.
 632  * If we find an XFS superblock, then run a normal, noisy mount because we are
 633  * really going to mount it and want to know about errors.
 634  */
 635 static void
 636 xfs_sb_quiet_read_verify(
 637         struct xfs_buf  *bp)
 638 {
 639         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 640
 641         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
 642                 /* XFS filesystem, verify noisily! */
 643                 xfs_sb_read_verify(bp);
 644                 return;
 645         }
 646         /* quietly fail */
 647         xfs_buf_ioerror(bp, EWRONGFS);
 648 }
 649
 650 static void
 651 xfs_sb_write_verify(
 652         struct xfs_buf          *bp)
 653 {
 654         struct xfs_mount        *mp = bp->b_target->bt_mount;
 655         struct xfs_buf_log_item *bip = bp->b_fspriv;
 656         int                     error;
 657
 658         error = xfs_sb_verify(bp, false);
 659         if (error) {
 660                 xfs_buf_ioerror(bp, error);
 661                 xfs_verifier_error(bp);
 662                 return;
 663         }
 664
 665         if (!xfs_sb_version_hascrc(&mp->m_sb))
 666                 return;
 667
 668         if (bip)
 669                 XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 670
 671         xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
 672 }
 673
 674 const struct xfs_buf_ops xfs_sb_buf_ops = {
 675         .verify_read = xfs_sb_read_verify,
 676         .verify_write = xfs_sb_write_verify,
 677 };
 678
 679 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
 680         .verify_read = xfs_sb_quiet_read_verify,
 681         .verify_write = xfs_sb_write_verify,
 682 };
 683
 684 /*
 685  * xfs_mount_common
 686  *
 687  * Mount initialization code establishing various mount
 688  * fields from the superblock associated with the given
 689  * mount structure
 690  */
 691 void
 692 xfs_sb_mount_common(
 693         struct xfs_mount *mp,
 694         struct xfs_sb   *sbp)
 695 {
 696         mp->m_agfrotor = mp->m_agirotor = 0;
 697         spin_lock_init(&mp->m_agirotor_lock);
 698         mp->m_maxagi = mp->m_sb.sb_agcount;
 699         mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
 700         mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
 701         mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
 702         mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
 703         mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
 704         mp->m_blockmask = sbp->sb_blocksize - 1;
 705         mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
 706         mp->m_blockwmask = mp->m_blockwsize - 1;
 707
 708         mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
 709         mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
 710         mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
 711         mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
 712
 713         mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
 714         mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
 715         mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
 716         mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
 717
 718         mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
 719         mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
 720         mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
 721         mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
 722
 723         mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
 724         mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 725                                         sbp->sb_inopblock);
 726         mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 727 }
 728
 729 /*
 730  * xfs_initialize_perag_data
 731  *
 732  * Read in each per-ag structure so we can count up the number of
 733  * allocated inodes, free inodes and used filesystem blocks as this
 734  * information is no longer persistent in the superblock. Once we have
 735  * this information, write it into the in-core superblock structure.
 736  */
 737 int
 738 xfs_initialize_perag_data(
 739         struct xfs_mount *mp,
 740         xfs_agnumber_t  agcount)
 741 {
 742         xfs_agnumber_t  index;
 743         xfs_perag_t     *pag;
 744         xfs_sb_t        *sbp = &mp->m_sb;
 745         uint64_t        ifree = 0;
 746         uint64_t        ialloc = 0;
 747         uint64_t        bfree = 0;
 748         uint64_t        bfreelst = 0;
 749         uint64_t        btree = 0;
 750         int             error;
 751
 752         for (index = 0; index < agcount; index++) {
 753                 /*
 754                  * read the agf, then the agi. This gets us
 755                  * all the information we need and populates the
 756                  * per-ag structures for us.
 757                  */
 758                 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
 759                 if (error)
 760                         return error;
 761
 762                 error = xfs_ialloc_pagi_init(mp, NULL, index);
 763                 if (error)
 764                         return error;
 765                 pag = xfs_perag_get(mp, index);
 766                 ifree += pag->pagi_freecount;
 767                 ialloc += pag->pagi_count;
 768                 bfree += pag->pagf_freeblks;
 769                 bfreelst += pag->pagf_flcount;
 770                 btree += pag->pagf_btreeblks;
 771                 xfs_perag_put(pag);
 772         }
 773         /*
 774          * Overwrite incore superblock counters with just-read data
 775          */
 776         spin_lock(&mp->m_sb_lock);
 777         sbp->sb_ifree = ifree;
 778         sbp->sb_icount = ialloc;
 779         sbp->sb_fdblocks = bfree + bfreelst + btree;
 780         spin_unlock(&mp->m_sb_lock);
 781
 782         /* Fixup the per-cpu counters as well. */
 783         xfs_icsb_reinit_counters(mp);
 784
 785         return 0;
 786 }
 787
 788 /*
 789  * xfs_mod_sb() can be used to copy arbitrary changes to the
 790  * in-core superblock into the superblock buffer to be logged.
 791  * It does not provide the higher level of locking that is
 792  * needed to protect the in-core superblock from concurrent
 793  * access.
 794  */
 795 void
 796 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
 797 {
 798         xfs_buf_t       *bp;
 799         int             first;
 800         int             last;
 801         xfs_mount_t     *mp;
 802         xfs_sb_field_t  f;
 803
 804         ASSERT(fields);
 805         if (!fields)
 806                 return;
 807         mp = tp->t_mountp;
 808         bp = xfs_trans_getsb(tp, mp, 0);
 809         first = sizeof(xfs_sb_t);
 810         last = 0;
 811
 812         /* translate/copy */
 813
 814         xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
 815
 816         /* find modified range */
 817         f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
 818         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
 819         last = xfs_sb_info[f + 1].offset - 1;
 820
 821         f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 822         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
 823         first = xfs_sb_info[f].offset;
 824
 825         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
 826         xfs_trans_log_buf(tp, bp, first, last);
 827 }