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_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_trans_priv.h"
  26 #include "xfs_sb.h"
  27 #include "xfs_ag.h"
  28 #include "xfs_mount.h"
  29 #include "xfs_da_btree.h"
  30 #include "xfs_dir2_format.h"
  31 #include "xfs_dir2.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_ialloc_btree.h"
  35 #include "xfs_dinode.h"
  36 #include "xfs_inode.h"
  37 #include "xfs_btree.h"
  38 #include "xfs_ialloc.h"
  39 #include "xfs_alloc.h"
  40 #include "xfs_rtalloc.h"
  41 #include "xfs_bmap.h"
  42 #include "xfs_error.h"
  43 #include "xfs_quota.h"
  44 #include "xfs_fsops.h"
  45 #include "xfs_trace.h"
  46 #include "xfs_cksum.h"
  47 #include "xfs_buf_item.h"
  48
  49 /*
  50  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
  51  */
  52
  53 static const struct {
  54         short offset;
  55         short type;     /* 0 = integer
  56                          * 1 = binary / string (no translation)
  57                          */
  58 } xfs_sb_info[] = {
  59         { offsetof(xfs_sb_t, sb_magicnum),      0 },
  60         { offsetof(xfs_sb_t, sb_blocksize),     0 },
  61         { offsetof(xfs_sb_t, sb_dblocks),       0 },
  62         { offsetof(xfs_sb_t, sb_rblocks),       0 },
  63         { offsetof(xfs_sb_t, sb_rextents),      0 },
  64         { offsetof(xfs_sb_t, sb_uuid),          1 },
  65         { offsetof(xfs_sb_t, sb_logstart),      0 },
  66         { offsetof(xfs_sb_t, sb_rootino),       0 },
  67         { offsetof(xfs_sb_t, sb_rbmino),        0 },
  68         { offsetof(xfs_sb_t, sb_rsumino),       0 },
  69         { offsetof(xfs_sb_t, sb_rextsize),      0 },
  70         { offsetof(xfs_sb_t, sb_agblocks),      0 },
  71         { offsetof(xfs_sb_t, sb_agcount),       0 },
  72         { offsetof(xfs_sb_t, sb_rbmblocks),     0 },
  73         { offsetof(xfs_sb_t, sb_logblocks),     0 },
  74         { offsetof(xfs_sb_t, sb_versionnum),    0 },
  75         { offsetof(xfs_sb_t, sb_sectsize),      0 },
  76         { offsetof(xfs_sb_t, sb_inodesize),     0 },
  77         { offsetof(xfs_sb_t, sb_inopblock),     0 },
  78         { offsetof(xfs_sb_t, sb_fname[0]),      1 },
  79         { offsetof(xfs_sb_t, sb_blocklog),      0 },
  80         { offsetof(xfs_sb_t, sb_sectlog),       0 },
  81         { offsetof(xfs_sb_t, sb_inodelog),      0 },
  82         { offsetof(xfs_sb_t, sb_inopblog),      0 },
  83         { offsetof(xfs_sb_t, sb_agblklog),      0 },
  84         { offsetof(xfs_sb_t, sb_rextslog),      0 },
  85         { offsetof(xfs_sb_t, sb_inprogress),    0 },
  86         { offsetof(xfs_sb_t, sb_imax_pct),      0 },
  87         { offsetof(xfs_sb_t, sb_icount),        0 },
  88         { offsetof(xfs_sb_t, sb_ifree),         0 },
  89         { offsetof(xfs_sb_t, sb_fdblocks),      0 },
  90         { offsetof(xfs_sb_t, sb_frextents),     0 },
  91         { offsetof(xfs_sb_t, sb_uquotino),      0 },
  92         { offsetof(xfs_sb_t, sb_gquotino),      0 },
  93         { offsetof(xfs_sb_t, sb_qflags),        0 },
  94         { offsetof(xfs_sb_t, sb_flags),         0 },
  95         { offsetof(xfs_sb_t, sb_shared_vn),     0 },
  96         { offsetof(xfs_sb_t, sb_inoalignmt),    0 },
  97         { offsetof(xfs_sb_t, sb_unit),          0 },
  98         { offsetof(xfs_sb_t, sb_width),         0 },
  99         { offsetof(xfs_sb_t, sb_dirblklog),     0 },
 100         { offsetof(xfs_sb_t, sb_logsectlog),    0 },
 101         { offsetof(xfs_sb_t, sb_logsectsize),   0 },
 102         { offsetof(xfs_sb_t, sb_logsunit),      0 },
 103         { offsetof(xfs_sb_t, sb_features2),     0 },
 104         { offsetof(xfs_sb_t, sb_bad_features2), 0 },
 105         { offsetof(xfs_sb_t, sb_features_compat),       0 },
 106         { offsetof(xfs_sb_t, sb_features_ro_compat),    0 },
 107         { offsetof(xfs_sb_t, sb_features_incompat),     0 },
 108         { offsetof(xfs_sb_t, sb_features_log_incompat), 0 },
 109         { offsetof(xfs_sb_t, sb_crc),           0 },
 110         { offsetof(xfs_sb_t, sb_pad),           0 },
 111         { offsetof(xfs_sb_t, sb_pquotino),      0 },
 112         { offsetof(xfs_sb_t, sb_lsn),           0 },
 113         { sizeof(xfs_sb_t),                     0 }
 114 };
 115
 116 /*
 117  * Reference counting access wrappers to the perag structures.
 118  * Because we never free per-ag structures, the only thing we
 119  * have to protect against changes is the tree structure itself.
 120  */
 121 struct xfs_perag *
 122 xfs_perag_get(
 123         struct xfs_mount        *mp,
 124         xfs_agnumber_t          agno)
 125 {
 126         struct xfs_perag        *pag;
 127         int                     ref = 0;
 128
 129         rcu_read_lock();
 130         pag = radix_tree_lookup(&mp->m_perag_tree, agno);
 131         if (pag) {
 132                 ASSERT(atomic_read(&pag->pag_ref) >= 0);
 133                 ref = atomic_inc_return(&pag->pag_ref);
 134         }
 135         rcu_read_unlock();
 136         trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
 137         return pag;
 138 }
 139
 140 /*
 141  * search from @first to find the next perag with the given tag set.
 142  */
 143 struct xfs_perag *
 144 xfs_perag_get_tag(
 145         struct xfs_mount        *mp,
 146         xfs_agnumber_t          first,
 147         int                     tag)
 148 {
 149         struct xfs_perag        *pag;
 150         int                     found;
 151         int                     ref;
 152
 153         rcu_read_lock();
 154         found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
 155                                         (void **)&pag, first, 1, tag);
 156         if (found <= 0) {
 157                 rcu_read_unlock();
 158                 return NULL;
 159         }
 160         ref = atomic_inc_return(&pag->pag_ref);
 161         rcu_read_unlock();
 162         trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
 163         return pag;
 164 }
 165
 166 void
 167 xfs_perag_put(
 168         struct xfs_perag        *pag)
 169 {
 170         int     ref;
 171
 172         ASSERT(atomic_read(&pag->pag_ref) > 0);
 173         ref = atomic_dec_return(&pag->pag_ref);
 174         trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
 175 }
 176
 177 /*
 178  * Check the validity of the SB found.
 179  */
 180 STATIC int
 181 xfs_mount_validate_sb(
 182         xfs_mount_t     *mp,
 183         xfs_sb_t        *sbp,
 184         bool            check_inprogress,
 185         bool            check_version)
 186 {
 187
 188         /*
 189          * If the log device and data device have the
 190          * same device number, the log is internal.
 191          * Consequently, the sb_logstart should be non-zero.  If
 192          * we have a zero sb_logstart in this case, we may be trying to mount
 193          * a volume filesystem in a non-volume manner.
 194          */
 195         if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 196                 xfs_warn(mp, "bad magic number");
 197                 return XFS_ERROR(EWRONGFS);
 198         }
 199
 200
 201         if (!xfs_sb_good_version(sbp)) {
 202                 xfs_warn(mp, "bad version");
 203                 return XFS_ERROR(EWRONGFS);
 204         }
 205
 206         /*
 207          * Version 5 superblock feature mask validation. Reject combinations the
 208          * kernel cannot support up front before checking anything else. For
 209          * write validation, we don't need to check feature masks.
 210          */
 211         if (check_version && XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5) {
 212                 xfs_alert(mp,
 213 "Version 5 superblock detected. This kernel has EXPERIMENTAL support enabled!\n"
 214 "Use of these features in this kernel is at your own risk!");
 215
 216                 if (xfs_sb_has_compat_feature(sbp,
 217                                         XFS_SB_FEAT_COMPAT_UNKNOWN)) {
 218                         xfs_warn(mp,
 219 "Superblock has unknown compatible features (0x%x) enabled.\n"
 220 "Using a more recent kernel is recommended.",
 221                                 (sbp->sb_features_compat &
 222                                                 XFS_SB_FEAT_COMPAT_UNKNOWN));
 223                 }
 224
 225                 if (xfs_sb_has_ro_compat_feature(sbp,
 226                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
 227                         xfs_alert(mp,
 228 "Superblock has unknown read-only compatible features (0x%x) enabled.",
 229                                 (sbp->sb_features_ro_compat &
 230                                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
 231                         if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 232                                 xfs_warn(mp,
 233 "Attempted to mount read-only compatible filesystem read-write.\n"
 234 "Filesystem can only be safely mounted read only.");
 235                                 return XFS_ERROR(EINVAL);
 236                         }
 237                 }
 238                 if (xfs_sb_has_incompat_feature(sbp,
 239                                         XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
 240                         xfs_warn(mp,
 241 "Superblock has unknown incompatible features (0x%x) enabled.\n"
 242 "Filesystem can not be safely mounted by this kernel.",
 243                                 (sbp->sb_features_incompat &
 244                                                 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
 245                         return XFS_ERROR(EINVAL);
 246                 }
 247         }
 248
 249         if (xfs_sb_version_has_pquotino(sbp)) {
 250                 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
 251                         xfs_notice(mp,
 252                            "Version 5 of Super block has XFS_OQUOTA bits.\n");
 253                         return XFS_ERROR(EFSCORRUPTED);
 254                 }
 255         } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
 256                                 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
 257                         xfs_notice(mp,
 258 "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.\n");
 259                         return XFS_ERROR(EFSCORRUPTED);
 260         }
 261
 262         if (unlikely(
 263             sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
 264                 xfs_warn(mp,
 265                 "filesystem is marked as having an external log; "
 266                 "specify logdev on the mount command line.");
 267                 return XFS_ERROR(EINVAL);
 268         }
 269
 270         if (unlikely(
 271             sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
 272                 xfs_warn(mp,
 273                 "filesystem is marked as having an internal log; "
 274                 "do not specify logdev on the mount command line.");
 275                 return XFS_ERROR(EINVAL);
 276         }
 277
 278         /*
 279          * More sanity checking.  Most of these were stolen directly from
 280          * xfs_repair.
 281          */
 282         if (unlikely(
 283             sbp->sb_agcount <= 0                                        ||
 284             sbp->sb_sectsize < XFS_MIN_SECTORSIZE                       ||
 285             sbp->sb_sectsize > XFS_MAX_SECTORSIZE                       ||
 286             sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG                    ||
 287             sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG                    ||
 288             sbp->sb_sectsize != (1 << sbp->sb_sectlog)                  ||
 289             sbp->sb_blocksize < XFS_MIN_BLOCKSIZE                       ||
 290             sbp->sb_blocksize > XFS_MAX_BLOCKSIZE                       ||
 291             sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG                    ||
 292             sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG                    ||
 293             sbp->sb_blocksize != (1 << sbp->sb_blocklog)                ||
 294             sbp->sb_inodesize < XFS_DINODE_MIN_SIZE                     ||
 295             sbp->sb_inodesize > XFS_DINODE_MAX_SIZE                     ||
 296             sbp->sb_inodelog < XFS_DINODE_MIN_LOG                       ||
 297             sbp->sb_inodelog > XFS_DINODE_MAX_LOG                       ||
 298             sbp->sb_inodesize != (1 << sbp->sb_inodelog)                ||
 299             (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
 300             (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
 301             (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
 302             (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)    ||
 303             sbp->sb_dblocks == 0                                        ||
 304             sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)                      ||
 305             sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
 306                 XFS_CORRUPTION_ERROR("SB sanity check failed",
 307                                 XFS_ERRLEVEL_LOW, mp, sbp);
 308                 return XFS_ERROR(EFSCORRUPTED);
 309         }
 310
 311         /*
 312          * Until this is fixed only page-sized or smaller data blocks work.
 313          */
 314         if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
 315                 xfs_warn(mp,
 316                 "File system with blocksize %d bytes. "
 317                 "Only pagesize (%ld) or less will currently work.",
 318                                 sbp->sb_blocksize, PAGE_SIZE);
 319                 return XFS_ERROR(ENOSYS);
 320         }
 321
 322         /*
 323          * Currently only very few inode sizes are supported.
 324          */
 325         switch (sbp->sb_inodesize) {
 326         case 256:
 327         case 512:
 328         case 1024:
 329         case 2048:
 330                 break;
 331         default:
 332                 xfs_warn(mp, "inode size of %d bytes not supported",
 333                                 sbp->sb_inodesize);
 334                 return XFS_ERROR(ENOSYS);
 335         }
 336
 337         if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
 338             xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
 339                 xfs_warn(mp,
 340                 "file system too large to be mounted on this system.");
 341                 return XFS_ERROR(EFBIG);
 342         }
 343
 344         if (check_inprogress && sbp->sb_inprogress) {
 345                 xfs_warn(mp, "Offline file system operation in progress!");
 346                 return XFS_ERROR(EFSCORRUPTED);
 347         }
 348
 349         /*
 350          * Version 1 directory format has never worked on Linux.
 351          */
 352         if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
 353                 xfs_warn(mp, "file system using version 1 directory format");
 354                 return XFS_ERROR(ENOSYS);
 355         }
 356
 357         return 0;
 358 }
 359
 360 void
 361 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
 362 {
 363         /*
 364          * older mkfs doesn't initialize quota inodes to NULLFSINO. This
 365          * leads to in-core values having two different values for a quota
 366          * inode to be invalid: 0 and NULLFSINO. Change it to a single value
 367          * NULLFSINO.
 368          *
 369          * Note that this change affect only the in-core values. These
 370          * values are not written back to disk unless any quota information
 371          * is written to the disk. Even in that case, sb_pquotino field is
 372          * not written to disk unless the superblock supports pquotino.
 373          */
 374         if (sbp->sb_uquotino == 0)
 375                 sbp->sb_uquotino = NULLFSINO;
 376         if (sbp->sb_gquotino == 0)
 377                 sbp->sb_gquotino = NULLFSINO;
 378         if (sbp->sb_pquotino == 0)
 379                 sbp->sb_pquotino = NULLFSINO;
 380
 381         /*
 382          * We need to do these manipilations only if we are working
 383          * with an older version of on-disk superblock.
 384          */
 385         if (xfs_sb_version_has_pquotino(sbp))
 386                 return;
 387
 388         if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
 389                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 390                                         XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
 391         if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
 392                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 393                                         XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
 394         sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
 395
 396         if (sbp->sb_qflags & XFS_PQUOTA_ACCT)  {
 397                 /*
 398                  * In older version of superblock, on-disk superblock only
 399                  * has sb_gquotino, and in-core superblock has both sb_gquotino
 400                  * and sb_pquotino. But, only one of them is supported at any
 401                  * point of time. So, if PQUOTA is set in disk superblock,
 402                  * copy over sb_gquotino to sb_pquotino.
 403                  */
 404                 sbp->sb_pquotino = sbp->sb_gquotino;
 405                 sbp->sb_gquotino = NULLFSINO;
 406         }
 407 }
 408
 409 void
 410 xfs_sb_from_disk(
 411         struct xfs_sb   *to,
 412         xfs_dsb_t       *from)
 413 {
 414         to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
 415         to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
 416         to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
 417         to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
 418         to->sb_rextents = be64_to_cpu(from->sb_rextents);
 419         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
 420         to->sb_logstart = be64_to_cpu(from->sb_logstart);
 421         to->sb_rootino = be64_to_cpu(from->sb_rootino);
 422         to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
 423         to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
 424         to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
 425         to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
 426         to->sb_agcount = be32_to_cpu(from->sb_agcount);
 427         to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
 428         to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
 429         to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
 430         to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
 431         to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
 432         to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
 433         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
 434         to->sb_blocklog = from->sb_blocklog;
 435         to->sb_sectlog = from->sb_sectlog;
 436         to->sb_inodelog = from->sb_inodelog;
 437         to->sb_inopblog = from->sb_inopblog;
 438         to->sb_agblklog = from->sb_agblklog;
 439         to->sb_rextslog = from->sb_rextslog;
 440         to->sb_inprogress = from->sb_inprogress;
 441         to->sb_imax_pct = from->sb_imax_pct;
 442         to->sb_icount = be64_to_cpu(from->sb_icount);
 443         to->sb_ifree = be64_to_cpu(from->sb_ifree);
 444         to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
 445         to->sb_frextents = be64_to_cpu(from->sb_frextents);
 446         to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
 447         to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
 448         to->sb_qflags = be16_to_cpu(from->sb_qflags);
 449         to->sb_flags = from->sb_flags;
 450         to->sb_shared_vn = from->sb_shared_vn;
 451         to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
 452         to->sb_unit = be32_to_cpu(from->sb_unit);
 453         to->sb_width = be32_to_cpu(from->sb_width);
 454         to->sb_dirblklog = from->sb_dirblklog;
 455         to->sb_logsectlog = from->sb_logsectlog;
 456         to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
 457         to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
 458         to->sb_features2 = be32_to_cpu(from->sb_features2);
 459         to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
 460         to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
 461         to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
 462         to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
 463         to->sb_features_log_incompat =
 464                                 be32_to_cpu(from->sb_features_log_incompat);
 465         to->sb_pad = 0;
 466         to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
 467         to->sb_lsn = be64_to_cpu(from->sb_lsn);
 468 }
 469
 470 static inline void
 471 xfs_sb_quota_to_disk(
 472         xfs_dsb_t       *to,
 473         xfs_sb_t        *from,
 474         __int64_t       *fields)
 475 {
 476         __uint16_t      qflags = from->sb_qflags;
 477
 478         /*
 479          * We need to do these manipilations only if we are working
 480          * with an older version of on-disk superblock.
 481          */
 482         if (xfs_sb_version_has_pquotino(from))
 483                 return;
 484
 485         if (*fields & XFS_SB_QFLAGS) {
 486                 /*
 487                  * The in-core version of sb_qflags do not have
 488                  * XFS_OQUOTA_* flags, whereas the on-disk version
 489                  * does.  So, convert incore XFS_{PG}QUOTA_* flags
 490                  * to on-disk XFS_OQUOTA_* flags.
 491                  */
 492                 qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
 493                                 XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
 494
 495                 if (from->sb_qflags &
 496                                 (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
 497                         qflags |= XFS_OQUOTA_ENFD;
 498                 if (from->sb_qflags &
 499                                 (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
 500                         qflags |= XFS_OQUOTA_CHKD;
 501                 to->sb_qflags = cpu_to_be16(qflags);
 502                 *fields &= ~XFS_SB_QFLAGS;
 503         }
 504
 505         /*
 506          * GQUOTINO and PQUOTINO cannot be used together in versions
 507          * of superblock that do not have pquotino. from->sb_flags
 508          * tells us which quota is active and should be copied to
 509          * disk.
 510          */
 511         if ((*fields & XFS_SB_GQUOTINO) &&
 512                                 (from->sb_qflags & XFS_GQUOTA_ACCT))
 513                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
 514         else if ((*fields & XFS_SB_PQUOTINO) &&
 515                                 (from->sb_qflags & XFS_PQUOTA_ACCT))
 516                 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
 517
 518         *fields &= ~(XFS_SB_PQUOTINO | XFS_SB_GQUOTINO);
 519 }
 520
 521 /*
 522  * Copy in core superblock to ondisk one.
 523  *
 524  * The fields argument is mask of superblock fields to copy.
 525  */
 526 void
 527 xfs_sb_to_disk(
 528         xfs_dsb_t       *to,
 529         xfs_sb_t        *from,
 530         __int64_t       fields)
 531 {
 532         xfs_caddr_t     to_ptr = (xfs_caddr_t)to;
 533         xfs_caddr_t     from_ptr = (xfs_caddr_t)from;
 534         xfs_sb_field_t  f;
 535         int             first;
 536         int             size;
 537
 538         ASSERT(fields);
 539         if (!fields)
 540                 return;
 541
 542         xfs_sb_quota_to_disk(to, from, &fields);
 543         while (fields) {
 544                 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 545                 first = xfs_sb_info[f].offset;
 546                 size = xfs_sb_info[f + 1].offset - first;
 547
 548                 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
 549
 550                 if (size == 1 || xfs_sb_info[f].type == 1) {
 551                         memcpy(to_ptr + first, from_ptr + first, size);
 552                 } else {
 553                         switch (size) {
 554                         case 2:
 555                                 *(__be16 *)(to_ptr + first) =
 556                                       cpu_to_be16(*(__u16 *)(from_ptr + first));
 557                                 break;
 558                         case 4:
 559                                 *(__be32 *)(to_ptr + first) =
 560                                       cpu_to_be32(*(__u32 *)(from_ptr + first));
 561                                 break;
 562                         case 8:
 563                                 *(__be64 *)(to_ptr + first) =
 564                                       cpu_to_be64(*(__u64 *)(from_ptr + first));
 565                                 break;
 566                         default:
 567                                 ASSERT(0);
 568                         }
 569                 }
 570
 571                 fields &= ~(1LL << f);
 572         }
 573 }
 574
 575 static int
 576 xfs_sb_verify(
 577         struct xfs_buf  *bp,
 578         bool            check_version)
 579 {
 580         struct xfs_mount *mp = bp->b_target->bt_mount;
 581         struct xfs_sb   sb;
 582
 583         xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp));
 584
 585         /*
 586          * Only check the in progress field for the primary superblock as
 587          * mkfs.xfs doesn't clear it from secondary superblocks.
 588          */
 589         return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR,
 590                                      check_version);
 591 }
 592
 593 /*
 594  * If the superblock has the CRC feature bit set or the CRC field is non-null,
 595  * check that the CRC is valid.  We check the CRC field is non-null because a
 596  * single bit error could clear the feature bit and unused parts of the
 597  * superblock are supposed to be zero. Hence a non-null crc field indicates that
 598  * we've potentially lost a feature bit and we should check it anyway.
 599  */
 600 static void
 601 xfs_sb_read_verify(
 602         struct xfs_buf  *bp)
 603 {
 604         struct xfs_mount *mp = bp->b_target->bt_mount;
 605         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 606         int             error;
 607
 608         /*
 609          * open code the version check to avoid needing to convert the entire
 610          * superblock from disk order just to check the version number
 611          */
 612         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
 613             (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
 614                                                 XFS_SB_VERSION_5) ||
 615              dsb->sb_crc != 0)) {
 616
 617                 if (!xfs_verify_cksum(bp->b_addr, be16_to_cpu(dsb->sb_sectsize),
 618                                       offsetof(struct xfs_sb, sb_crc))) {
 619                         error = EFSCORRUPTED;
 620                         goto out_error;
 621                 }
 622         }
 623         error = xfs_sb_verify(bp, true);
 624
 625 out_error:
 626         if (error) {
 627                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW,
 628                                      mp, bp->b_addr);
 629                 xfs_buf_ioerror(bp, error);
 630         }
 631 }
 632
 633 /*
 634  * We may be probed for a filesystem match, so we may not want to emit
 635  * messages when the superblock buffer is not actually an XFS superblock.
 636  * If we find an XFS superblock, then run a normal, noisy mount because we are
 637  * really going to mount it and want to know about errors.
 638  */
 639 static void
 640 xfs_sb_quiet_read_verify(
 641         struct xfs_buf  *bp)
 642 {
 643         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 644
 645
 646         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
 647                 /* XFS filesystem, verify noisily! */
 648                 xfs_sb_read_verify(bp);
 649                 return;
 650         }
 651         /* quietly fail */
 652         xfs_buf_ioerror(bp, EWRONGFS);
 653 }
 654
 655 static void
 656 xfs_sb_write_verify(
 657         struct xfs_buf          *bp)
 658 {
 659         struct xfs_mount        *mp = bp->b_target->bt_mount;
 660         struct xfs_buf_log_item *bip = bp->b_fspriv;
 661         int                     error;
 662
 663         error = xfs_sb_verify(bp, false);
 664         if (error) {
 665                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW,
 666                                      mp, bp->b_addr);
 667                 xfs_buf_ioerror(bp, error);
 668                 return;
 669         }
 670
 671         if (!xfs_sb_version_hascrc(&mp->m_sb))
 672                 return;
 673
 674         if (bip)
 675                 XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 676
 677         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
 678                          offsetof(struct xfs_sb, sb_crc));
 679 }
 680
 681 const struct xfs_buf_ops xfs_sb_buf_ops = {
 682         .verify_read = xfs_sb_read_verify,
 683         .verify_write = xfs_sb_write_verify,
 684 };
 685
 686 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
 687         .verify_read = xfs_sb_quiet_read_verify,
 688         .verify_write = xfs_sb_write_verify,
 689 };
 690
 691 /*
 692  * xfs_mount_common
 693  *
 694  * Mount initialization code establishing various mount
 695  * fields from the superblock associated with the given
 696  * mount structure
 697  */
 698 void
 699 xfs_sb_mount_common(
 700         struct xfs_mount *mp,
 701         struct xfs_sb   *sbp)
 702 {
 703         mp->m_agfrotor = mp->m_agirotor = 0;
 704         spin_lock_init(&mp->m_agirotor_lock);
 705         mp->m_maxagi = mp->m_sb.sb_agcount;
 706         mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
 707         mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
 708         mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
 709         mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
 710         mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
 711         mp->m_blockmask = sbp->sb_blocksize - 1;
 712         mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
 713         mp->m_blockwmask = mp->m_blockwsize - 1;
 714
 715         mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
 716         mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
 717         mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
 718         mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
 719
 720         mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
 721         mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
 722         mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
 723         mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
 724
 725         mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
 726         mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
 727         mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
 728         mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
 729
 730         mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
 731         mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 732                                         sbp->sb_inopblock);
 733         mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 734 }
 735
 736 /*
 737  * xfs_initialize_perag_data
 738  *
 739  * Read in each per-ag structure so we can count up the number of
 740  * allocated inodes, free inodes and used filesystem blocks as this
 741  * information is no longer persistent in the superblock. Once we have
 742  * this information, write it into the in-core superblock structure.
 743  */
 744 int
 745 xfs_initialize_perag_data(
 746         struct xfs_mount *mp,
 747         xfs_agnumber_t  agcount)
 748 {
 749         xfs_agnumber_t  index;
 750         xfs_perag_t     *pag;
 751         xfs_sb_t        *sbp = &mp->m_sb;
 752         uint64_t        ifree = 0;
 753         uint64_t        ialloc = 0;
 754         uint64_t        bfree = 0;
 755         uint64_t        bfreelst = 0;
 756         uint64_t        btree = 0;
 757         int             error;
 758
 759         for (index = 0; index < agcount; index++) {
 760                 /*
 761                  * read the agf, then the agi. This gets us
 762                  * all the information we need and populates the
 763                  * per-ag structures for us.
 764                  */
 765                 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
 766                 if (error)
 767                         return error;
 768
 769                 error = xfs_ialloc_pagi_init(mp, NULL, index);
 770                 if (error)
 771                         return error;
 772                 pag = xfs_perag_get(mp, index);
 773                 ifree += pag->pagi_freecount;
 774                 ialloc += pag->pagi_count;
 775                 bfree += pag->pagf_freeblks;
 776                 bfreelst += pag->pagf_flcount;
 777                 btree += pag->pagf_btreeblks;
 778                 xfs_perag_put(pag);
 779         }
 780         /*
 781          * Overwrite incore superblock counters with just-read data
 782          */
 783         spin_lock(&mp->m_sb_lock);
 784         sbp->sb_ifree = ifree;
 785         sbp->sb_icount = ialloc;
 786         sbp->sb_fdblocks = bfree + bfreelst + btree;
 787         spin_unlock(&mp->m_sb_lock);
 788
 789         /* Fixup the per-cpu counters as well. */
 790         xfs_icsb_reinit_counters(mp);
 791
 792         return 0;
 793 }
 794
 795 /*
 796  * xfs_mod_sb() can be used to copy arbitrary changes to the
 797  * in-core superblock into the superblock buffer to be logged.
 798  * It does not provide the higher level of locking that is
 799  * needed to protect the in-core superblock from concurrent
 800  * access.
 801  */
 802 void
 803 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
 804 {
 805         xfs_buf_t       *bp;
 806         int             first;
 807         int             last;
 808         xfs_mount_t     *mp;
 809         xfs_sb_field_t  f;
 810
 811         ASSERT(fields);
 812         if (!fields)
 813                 return;
 814         mp = tp->t_mountp;
 815         bp = xfs_trans_getsb(tp, mp, 0);
 816         first = sizeof(xfs_sb_t);
 817         last = 0;
 818
 819         /* translate/copy */
 820
 821         xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
 822
 823         /* find modified range */
 824         f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
 825         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
 826         last = xfs_sb_info[f + 1].offset - 1;
 827
 828         f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 829         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
 830         first = xfs_sb_info[f].offset;
 831
 832         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
 833         xfs_trans_log_buf(tp, bp, first, last);
 834 }