fs/xfs/libxfs/xfs_sb.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_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 -EWRONGFS;
 190         }
 191
 192
 193         if (!xfs_sb_good_version(sbp)) {
 194                 xfs_warn(mp, "bad version");
 195                 return -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 -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 -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 -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 -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 -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 -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             sbp->sb_shared_vn != 0)) {
 296                 xfs_notice(mp, "SB sanity check failed");
 297                 return -EFSCORRUPTED;
 298         }
 299
 300         /*
 301          * Until this is fixed only page-sized or smaller data blocks work.
 302          */
 303         if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
 304                 xfs_warn(mp,
 305                 "File system with blocksize %d bytes. "
 306                 "Only pagesize (%ld) or less will currently work.",
 307                                 sbp->sb_blocksize, PAGE_SIZE);
 308                 return -ENOSYS;
 309         }
 310
 311         /*
 312          * Currently only very few inode sizes are supported.
 313          */
 314         switch (sbp->sb_inodesize) {
 315         case 256:
 316         case 512:
 317         case 1024:
 318         case 2048:
 319                 break;
 320         default:
 321                 xfs_warn(mp, "inode size of %d bytes not supported",
 322                                 sbp->sb_inodesize);
 323                 return -ENOSYS;
 324         }
 325
 326         if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
 327             xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
 328                 xfs_warn(mp,
 329                 "file system too large to be mounted on this system.");
 330                 return -EFBIG;
 331         }
 332
 333         if (check_inprogress && sbp->sb_inprogress) {
 334                 xfs_warn(mp, "Offline file system operation in progress!");
 335                 return -EFSCORRUPTED;
 336         }
 337         return 0;
 338 }
 339
 340 void
 341 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
 342 {
 343         /*
 344          * older mkfs doesn't initialize quota inodes to NULLFSINO. This
 345          * leads to in-core values having two different values for a quota
 346          * inode to be invalid: 0 and NULLFSINO. Change it to a single value
 347          * NULLFSINO.
 348          *
 349          * Note that this change affect only the in-core values. These
 350          * values are not written back to disk unless any quota information
 351          * is written to the disk. Even in that case, sb_pquotino field is
 352          * not written to disk unless the superblock supports pquotino.
 353          */
 354         if (sbp->sb_uquotino == 0)
 355                 sbp->sb_uquotino = NULLFSINO;
 356         if (sbp->sb_gquotino == 0)
 357                 sbp->sb_gquotino = NULLFSINO;
 358         if (sbp->sb_pquotino == 0)
 359                 sbp->sb_pquotino = NULLFSINO;
 360
 361         /*
 362          * We need to do these manipilations only if we are working
 363          * with an older version of on-disk superblock.
 364          */
 365         if (xfs_sb_version_has_pquotino(sbp))
 366                 return;
 367
 368         if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
 369                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 370                                         XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
 371         if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
 372                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 373                                         XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
 374         sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
 375
 376         if (sbp->sb_qflags & XFS_PQUOTA_ACCT)  {
 377                 /*
 378                  * In older version of superblock, on-disk superblock only
 379                  * has sb_gquotino, and in-core superblock has both sb_gquotino
 380                  * and sb_pquotino. But, only one of them is supported at any
 381                  * point of time. So, if PQUOTA is set in disk superblock,
 382                  * copy over sb_gquotino to sb_pquotino.
 383                  */
 384                 sbp->sb_pquotino = sbp->sb_gquotino;
 385                 sbp->sb_gquotino = NULLFSINO;
 386         }
 387 }
 388
 389 static void
 390 __xfs_sb_from_disk(
 391         struct xfs_sb   *to,
 392         xfs_dsb_t       *from,
 393         bool            convert_xquota)
 394 {
 395         to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
 396         to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
 397         to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
 398         to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
 399         to->sb_rextents = be64_to_cpu(from->sb_rextents);
 400         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
 401         to->sb_logstart = be64_to_cpu(from->sb_logstart);
 402         to->sb_rootino = be64_to_cpu(from->sb_rootino);
 403         to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
 404         to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
 405         to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
 406         to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
 407         to->sb_agcount = be32_to_cpu(from->sb_agcount);
 408         to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
 409         to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
 410         to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
 411         to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
 412         to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
 413         to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
 414         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
 415         to->sb_blocklog = from->sb_blocklog;
 416         to->sb_sectlog = from->sb_sectlog;
 417         to->sb_inodelog = from->sb_inodelog;
 418         to->sb_inopblog = from->sb_inopblog;
 419         to->sb_agblklog = from->sb_agblklog;
 420         to->sb_rextslog = from->sb_rextslog;
 421         to->sb_inprogress = from->sb_inprogress;
 422         to->sb_imax_pct = from->sb_imax_pct;
 423         to->sb_icount = be64_to_cpu(from->sb_icount);
 424         to->sb_ifree = be64_to_cpu(from->sb_ifree);
 425         to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
 426         to->sb_frextents = be64_to_cpu(from->sb_frextents);
 427         to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
 428         to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
 429         to->sb_qflags = be16_to_cpu(from->sb_qflags);
 430         to->sb_flags = from->sb_flags;
 431         to->sb_shared_vn = from->sb_shared_vn;
 432         to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
 433         to->sb_unit = be32_to_cpu(from->sb_unit);
 434         to->sb_width = be32_to_cpu(from->sb_width);
 435         to->sb_dirblklog = from->sb_dirblklog;
 436         to->sb_logsectlog = from->sb_logsectlog;
 437         to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
 438         to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
 439         to->sb_features2 = be32_to_cpu(from->sb_features2);
 440         to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
 441         to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
 442         to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
 443         to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
 444         to->sb_features_log_incompat =
 445                                 be32_to_cpu(from->sb_features_log_incompat);
 446         to->sb_pad = 0;
 447         to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
 448         to->sb_lsn = be64_to_cpu(from->sb_lsn);
 449         /* Convert on-disk flags to in-memory flags? */
 450         if (convert_xquota)
 451                 xfs_sb_quota_from_disk(to);
 452 }
 453
 454 void
 455 xfs_sb_from_disk(
 456         struct xfs_sb   *to,
 457         xfs_dsb_t       *from)
 458 {
 459         __xfs_sb_from_disk(to, from, true);
 460 }
 461
 462 static inline void
 463 xfs_sb_quota_to_disk(
 464         xfs_dsb_t       *to,
 465         xfs_sb_t        *from,
 466         __int64_t       *fields)
 467 {
 468         __uint16_t      qflags = from->sb_qflags;
 469
 470         /*
 471          * We need to do these manipilations only if we are working
 472          * with an older version of on-disk superblock.
 473          */
 474         if (xfs_sb_version_has_pquotino(from))
 475                 return;
 476
 477         if (*fields & XFS_SB_QFLAGS) {
 478                 /*
 479                  * The in-core version of sb_qflags do not have
 480                  * XFS_OQUOTA_* flags, whereas the on-disk version
 481                  * does.  So, convert incore XFS_{PG}QUOTA_* flags
 482                  * to on-disk XFS_OQUOTA_* flags.
 483                  */
 484                 qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
 485                                 XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
 486
 487                 if (from->sb_qflags &
 488                                 (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
 489                         qflags |= XFS_OQUOTA_ENFD;
 490                 if (from->sb_qflags &
 491                                 (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
 492                         qflags |= XFS_OQUOTA_CHKD;
 493                 to->sb_qflags = cpu_to_be16(qflags);
 494                 *fields &= ~XFS_SB_QFLAGS;
 495         }
 496
 497         /*
 498          * GQUOTINO and PQUOTINO cannot be used together in versions of
 499          * superblock that do not have pquotino. from->sb_flags tells us which
 500          * quota is active and should be copied to disk. If neither are active,
 501          * make sure we write NULLFSINO to the sb_gquotino field as a quota
 502          * inode value of "0" is invalid when the XFS_SB_VERSION_QUOTA feature
 503          * bit is set.
 504          *
 505          * Note that we don't need to handle the sb_uquotino or sb_pquotino here
 506          * as they do not require any translation. Hence the main sb field loop
 507          * will write them appropriately from the in-core superblock.
 508          */
 509         if ((*fields & XFS_SB_GQUOTINO) &&
 510                                 (from->sb_qflags & XFS_GQUOTA_ACCT))
 511                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
 512         else if ((*fields & XFS_SB_PQUOTINO) &&
 513                                 (from->sb_qflags & XFS_PQUOTA_ACCT))
 514                 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
 515         else {
 516                 /*
 517                  * We can't rely on just the fields being logged to tell us
 518                  * that it is safe to write NULLFSINO - we should only do that
 519                  * if quotas are not actually enabled. Hence only write
 520                  * NULLFSINO if both in-core quota inodes are NULL.
 521                  */
 522                 if (from->sb_gquotino == NULLFSINO &&
 523                     from->sb_pquotino == NULLFSINO)
 524                         to->sb_gquotino = cpu_to_be64(NULLFSINO);
 525         }
 526
 527         *fields &= ~(XFS_SB_PQUOTINO | XFS_SB_GQUOTINO);
 528 }
 529
 530 /*
 531  * Copy in core superblock to ondisk one.
 532  *
 533  * The fields argument is mask of superblock fields to copy.
 534  */
 535 void
 536 xfs_sb_to_disk(
 537         xfs_dsb_t       *to,
 538         xfs_sb_t        *from,
 539         __int64_t       fields)
 540 {
 541         xfs_caddr_t     to_ptr = (xfs_caddr_t)to;
 542         xfs_caddr_t     from_ptr = (xfs_caddr_t)from;
 543         xfs_sb_field_t  f;
 544         int             first;
 545         int             size;
 546
 547         ASSERT(fields);
 548         if (!fields)
 549                 return;
 550
 551         xfs_sb_quota_to_disk(to, from, &fields);
 552         while (fields) {
 553                 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 554                 first = xfs_sb_info[f].offset;
 555                 size = xfs_sb_info[f + 1].offset - first;
 556
 557                 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
 558
 559                 if (size == 1 || xfs_sb_info[f].type == 1) {
 560                         memcpy(to_ptr + first, from_ptr + first, size);
 561                 } else {
 562                         switch (size) {
 563                         case 2:
 564                                 *(__be16 *)(to_ptr + first) =
 565                                       cpu_to_be16(*(__u16 *)(from_ptr + first));
 566                                 break;
 567                         case 4:
 568                                 *(__be32 *)(to_ptr + first) =
 569                                       cpu_to_be32(*(__u32 *)(from_ptr + first));
 570                                 break;
 571                         case 8:
 572                                 *(__be64 *)(to_ptr + first) =
 573                                       cpu_to_be64(*(__u64 *)(from_ptr + first));
 574                                 break;
 575                         default:
 576                                 ASSERT(0);
 577                         }
 578                 }
 579
 580                 fields &= ~(1LL << f);
 581         }
 582 }
 583
 584 static int
 585 xfs_sb_verify(
 586         struct xfs_buf  *bp,
 587         bool            check_version)
 588 {
 589         struct xfs_mount *mp = bp->b_target->bt_mount;
 590         struct xfs_sb   sb;
 591
 592         /*
 593          * Use call variant which doesn't convert quota flags from disk
 594          * format, because xfs_mount_validate_sb checks the on-disk flags.
 595          */
 596         __xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp), false);
 597
 598         /*
 599          * Only check the in progress field for the primary superblock as
 600          * mkfs.xfs doesn't clear it from secondary superblocks.
 601          */
 602         return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR,
 603                                      check_version);
 604 }
 605
 606 /*
 607  * If the superblock has the CRC feature bit set or the CRC field is non-null,
 608  * check that the CRC is valid.  We check the CRC field is non-null because a
 609  * single bit error could clear the feature bit and unused parts of the
 610  * superblock are supposed to be zero. Hence a non-null crc field indicates that
 611  * we've potentially lost a feature bit and we should check it anyway.
 612  *
 613  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
 614  * last field in V4 secondary superblocks.  So for secondary superblocks,
 615  * we are more forgiving, and ignore CRC failures if the primary doesn't
 616  * indicate that the fs version is V5.
 617  */
 618 static void
 619 xfs_sb_read_verify(
 620         struct xfs_buf  *bp)
 621 {
 622         struct xfs_mount *mp = bp->b_target->bt_mount;
 623         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 624         int             error;
 625
 626         /*
 627          * open code the version check to avoid needing to convert the entire
 628          * superblock from disk order just to check the version number
 629          */
 630         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
 631             (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
 632                                                 XFS_SB_VERSION_5) ||
 633              dsb->sb_crc != 0)) {
 634
 635                 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
 636                         /* Only fail bad secondaries on a known V5 filesystem */
 637                         if (bp->b_bn == XFS_SB_DADDR ||
 638                             xfs_sb_version_hascrc(&mp->m_sb)) {
 639                                 error = -EFSBADCRC;
 640                                 goto out_error;
 641                         }
 642                 }
 643         }
 644         error = xfs_sb_verify(bp, true);
 645
 646 out_error:
 647         if (error) {
 648                 xfs_buf_ioerror(bp, error);
 649                 if (error == -EFSCORRUPTED || error == -EFSBADCRC)
 650                         xfs_verifier_error(bp);
 651         }
 652 }
 653
 654 /*
 655  * We may be probed for a filesystem match, so we may not want to emit
 656  * messages when the superblock buffer is not actually an XFS superblock.
 657  * If we find an XFS superblock, then run a normal, noisy mount because we are
 658  * really going to mount it and want to know about errors.
 659  */
 660 static void
 661 xfs_sb_quiet_read_verify(
 662         struct xfs_buf  *bp)
 663 {
 664         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 665
 666         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
 667                 /* XFS filesystem, verify noisily! */
 668                 xfs_sb_read_verify(bp);
 669                 return;
 670         }
 671         /* quietly fail */
 672         xfs_buf_ioerror(bp, -EWRONGFS);
 673 }
 674
 675 static void
 676 xfs_sb_write_verify(
 677         struct xfs_buf          *bp)
 678 {
 679         struct xfs_mount        *mp = bp->b_target->bt_mount;
 680         struct xfs_buf_log_item *bip = bp->b_fspriv;
 681         int                     error;
 682
 683         error = xfs_sb_verify(bp, false);
 684         if (error) {
 685                 xfs_buf_ioerror(bp, error);
 686                 xfs_verifier_error(bp);
 687                 return;
 688         }
 689
 690         if (!xfs_sb_version_hascrc(&mp->m_sb))
 691                 return;
 692
 693         if (bip)
 694                 XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 695
 696         xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
 697 }
 698
 699 const struct xfs_buf_ops xfs_sb_buf_ops = {
 700         .verify_read = xfs_sb_read_verify,
 701         .verify_write = xfs_sb_write_verify,
 702 };
 703
 704 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
 705         .verify_read = xfs_sb_quiet_read_verify,
 706         .verify_write = xfs_sb_write_verify,
 707 };
 708
 709 /*
 710  * xfs_mount_common
 711  *
 712  * Mount initialization code establishing various mount
 713  * fields from the superblock associated with the given
 714  * mount structure
 715  */
 716 void
 717 xfs_sb_mount_common(
 718         struct xfs_mount *mp,
 719         struct xfs_sb   *sbp)
 720 {
 721         mp->m_agfrotor = mp->m_agirotor = 0;
 722         spin_lock_init(&mp->m_agirotor_lock);
 723         mp->m_maxagi = mp->m_sb.sb_agcount;
 724         mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
 725         mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
 726         mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
 727         mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
 728         mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
 729         mp->m_blockmask = sbp->sb_blocksize - 1;
 730         mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
 731         mp->m_blockwmask = mp->m_blockwsize - 1;
 732
 733         mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
 734         mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
 735         mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
 736         mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
 737
 738         mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
 739         mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
 740         mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
 741         mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
 742
 743         mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
 744         mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
 745         mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
 746         mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
 747
 748         mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
 749         mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 750                                         sbp->sb_inopblock);
 751         mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 752 }
 753
 754 /*
 755  * xfs_initialize_perag_data
 756  *
 757  * Read in each per-ag structure so we can count up the number of
 758  * allocated inodes, free inodes and used filesystem blocks as this
 759  * information is no longer persistent in the superblock. Once we have
 760  * this information, write it into the in-core superblock structure.
 761  */
 762 int
 763 xfs_initialize_perag_data(
 764         struct xfs_mount *mp,
 765         xfs_agnumber_t  agcount)
 766 {
 767         xfs_agnumber_t  index;
 768         xfs_perag_t     *pag;
 769         xfs_sb_t        *sbp = &mp->m_sb;
 770         uint64_t        ifree = 0;
 771         uint64_t        ialloc = 0;
 772         uint64_t        bfree = 0;
 773         uint64_t        bfreelst = 0;
 774         uint64_t        btree = 0;
 775         int             error;
 776
 777         for (index = 0; index < agcount; index++) {
 778                 /*
 779                  * read the agf, then the agi. This gets us
 780                  * all the information we need and populates the
 781                  * per-ag structures for us.
 782                  */
 783                 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
 784                 if (error)
 785                         return error;
 786
 787                 error = xfs_ialloc_pagi_init(mp, NULL, index);
 788                 if (error)
 789                         return error;
 790                 pag = xfs_perag_get(mp, index);
 791                 ifree += pag->pagi_freecount;
 792                 ialloc += pag->pagi_count;
 793                 bfree += pag->pagf_freeblks;
 794                 bfreelst += pag->pagf_flcount;
 795                 btree += pag->pagf_btreeblks;
 796                 xfs_perag_put(pag);
 797         }
 798         /*
 799          * Overwrite incore superblock counters with just-read data
 800          */
 801         spin_lock(&mp->m_sb_lock);
 802         sbp->sb_ifree = ifree;
 803         sbp->sb_icount = ialloc;
 804         sbp->sb_fdblocks = bfree + bfreelst + btree;
 805         spin_unlock(&mp->m_sb_lock);
 806
 807         /* Fixup the per-cpu counters as well. */
 808         xfs_icsb_reinit_counters(mp);
 809
 810         return 0;
 811 }
 812
 813 /*
 814  * xfs_mod_sb() can be used to copy arbitrary changes to the
 815  * in-core superblock into the superblock buffer to be logged.
 816  * It does not provide the higher level of locking that is
 817  * needed to protect the in-core superblock from concurrent
 818  * access.
 819  */
 820 void
 821 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
 822 {
 823         xfs_buf_t       *bp;
 824         int             first;
 825         int             last;
 826         xfs_mount_t     *mp;
 827         xfs_sb_field_t  f;
 828
 829         ASSERT(fields);
 830         if (!fields)
 831                 return;
 832         mp = tp->t_mountp;
 833         bp = xfs_trans_getsb(tp, mp, 0);
 834         first = sizeof(xfs_sb_t);
 835         last = 0;
 836
 837         /* translate/copy */
 838
 839         xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
 840
 841         /* find modified range */
 842         f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
 843         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
 844         last = xfs_sb_info[f + 1].offset - 1;
 845
 846         f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 847         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
 848         first = xfs_sb_info[f].offset;
 849
 850         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
 851         xfs_trans_log_buf(tp, bp, first, last);
 852 }