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