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_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 -EWRONGFS;
 189         }
 190
 191
 192         if (!xfs_sb_good_version(sbp)) {
 193                 xfs_warn(mp, "bad version");
 194                 return -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                 if (xfs_sb_has_compat_feature(sbp,
 204                                         XFS_SB_FEAT_COMPAT_UNKNOWN)) {
 205                         xfs_warn(mp,
 206 "Superblock has unknown compatible features (0x%x) enabled.\n"
 207 "Using a more recent kernel is recommended.",
 208                                 (sbp->sb_features_compat &
 209                                                 XFS_SB_FEAT_COMPAT_UNKNOWN));
 210                 }
 211
 212                 if (xfs_sb_has_ro_compat_feature(sbp,
 213                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
 214                         xfs_alert(mp,
 215 "Superblock has unknown read-only compatible features (0x%x) enabled.",
 216                                 (sbp->sb_features_ro_compat &
 217                                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
 218                         if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 219                                 xfs_warn(mp,
 220 "Attempted to mount read-only compatible filesystem read-write.\n"
 221 "Filesystem can only be safely mounted read only.");
 222                                 return -EINVAL;
 223                         }
 224                 }
 225                 if (xfs_sb_has_incompat_feature(sbp,
 226                                         XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
 227                         xfs_warn(mp,
 228 "Superblock has unknown incompatible features (0x%x) enabled.\n"
 229 "Filesystem can not be safely mounted by this kernel.",
 230                                 (sbp->sb_features_incompat &
 231                                                 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
 232                         return -EINVAL;
 233                 }
 234         }
 235
 236         if (xfs_sb_version_has_pquotino(sbp)) {
 237                 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
 238                         xfs_notice(mp,
 239                            "Version 5 of Super block has XFS_OQUOTA bits.");
 240                         return -EFSCORRUPTED;
 241                 }
 242         } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
 243                                 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
 244                         xfs_notice(mp,
 245 "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.");
 246                         return -EFSCORRUPTED;
 247         }
 248
 249         if (unlikely(
 250             sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
 251                 xfs_warn(mp,
 252                 "filesystem is marked as having an external log; "
 253                 "specify logdev on the mount command line.");
 254                 return -EINVAL;
 255         }
 256
 257         if (unlikely(
 258             sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
 259                 xfs_warn(mp,
 260                 "filesystem is marked as having an internal log; "
 261                 "do not specify logdev on the mount command line.");
 262                 return -EINVAL;
 263         }
 264
 265         /*
 266          * More sanity checking.  Most of these were stolen directly from
 267          * xfs_repair.
 268          */
 269         if (unlikely(
 270             sbp->sb_agcount <= 0                                        ||
 271             sbp->sb_sectsize < XFS_MIN_SECTORSIZE                       ||
 272             sbp->sb_sectsize > XFS_MAX_SECTORSIZE                       ||
 273             sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG                    ||
 274             sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG                    ||
 275             sbp->sb_sectsize != (1 << sbp->sb_sectlog)                  ||
 276             sbp->sb_blocksize < XFS_MIN_BLOCKSIZE                       ||
 277             sbp->sb_blocksize > XFS_MAX_BLOCKSIZE                       ||
 278             sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG                    ||
 279             sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG                    ||
 280             sbp->sb_blocksize != (1 << sbp->sb_blocklog)                ||
 281             sbp->sb_dirblklog > XFS_MAX_BLOCKSIZE_LOG                   ||
 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_logsunit > XLOG_MAX_RECORD_BSIZE                    ||
 288             sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
 289             (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
 290             (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
 291             (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
 292             (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)    ||
 293             sbp->sb_dblocks == 0                                        ||
 294             sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)                      ||
 295             sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp)                      ||
 296             sbp->sb_shared_vn != 0)) {
 297                 xfs_notice(mp, "SB sanity check failed");
 298                 return -EFSCORRUPTED;
 299         }
 300
 301         /*
 302          * Until this is fixed only page-sized or smaller data blocks work.
 303          */
 304         if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
 305                 xfs_warn(mp,
 306                 "File system with blocksize %d bytes. "
 307                 "Only pagesize (%ld) or less will currently work.",
 308                                 sbp->sb_blocksize, PAGE_SIZE);
 309                 return -ENOSYS;
 310         }
 311
 312         /*
 313          * Currently only very few inode sizes are supported.
 314          */
 315         switch (sbp->sb_inodesize) {
 316         case 256:
 317         case 512:
 318         case 1024:
 319         case 2048:
 320                 break;
 321         default:
 322                 xfs_warn(mp, "inode size of %d bytes not supported",
 323                                 sbp->sb_inodesize);
 324                 return -ENOSYS;
 325         }
 326
 327         if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
 328             xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
 329                 xfs_warn(mp,
 330                 "file system too large to be mounted on this system.");
 331                 return -EFBIG;
 332         }
 333
 334         if (check_inprogress && sbp->sb_inprogress) {
 335                 xfs_warn(mp, "Offline file system operation in progress!");
 336                 return -EFSCORRUPTED;
 337         }
 338         return 0;
 339 }
 340
 341 void
 342 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
 343 {
 344         /*
 345          * older mkfs doesn't initialize quota inodes to NULLFSINO. This
 346          * leads to in-core values having two different values for a quota
 347          * inode to be invalid: 0 and NULLFSINO. Change it to a single value
 348          * NULLFSINO.
 349          *
 350          * Note that this change affect only the in-core values. These
 351          * values are not written back to disk unless any quota information
 352          * is written to the disk. Even in that case, sb_pquotino field is
 353          * not written to disk unless the superblock supports pquotino.
 354          */
 355         if (sbp->sb_uquotino == 0)
 356                 sbp->sb_uquotino = NULLFSINO;
 357         if (sbp->sb_gquotino == 0)
 358                 sbp->sb_gquotino = NULLFSINO;
 359         if (sbp->sb_pquotino == 0)
 360                 sbp->sb_pquotino = NULLFSINO;
 361
 362         /*
 363          * We need to do these manipilations only if we are working
 364          * with an older version of on-disk superblock.
 365          */
 366         if (xfs_sb_version_has_pquotino(sbp))
 367                 return;
 368
 369         if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
 370                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 371                                         XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
 372         if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
 373                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 374                                         XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
 375         sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
 376
 377         if (sbp->sb_qflags & XFS_PQUOTA_ACCT)  {
 378                 /*
 379                  * In older version of superblock, on-disk superblock only
 380                  * has sb_gquotino, and in-core superblock has both sb_gquotino
 381                  * and sb_pquotino. But, only one of them is supported at any
 382                  * point of time. So, if PQUOTA is set in disk superblock,
 383                  * copy over sb_gquotino to sb_pquotino.
 384                  */
 385                 sbp->sb_pquotino = sbp->sb_gquotino;
 386                 sbp->sb_gquotino = NULLFSINO;
 387         }
 388 }
 389
 390 static void
 391 __xfs_sb_from_disk(
 392         struct xfs_sb   *to,
 393         xfs_dsb_t       *from,
 394         bool            convert_xquota)
 395 {
 396         to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
 397         to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
 398         to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
 399         to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
 400         to->sb_rextents = be64_to_cpu(from->sb_rextents);
 401         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
 402         to->sb_logstart = be64_to_cpu(from->sb_logstart);
 403         to->sb_rootino = be64_to_cpu(from->sb_rootino);
 404         to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
 405         to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
 406         to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
 407         to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
 408         to->sb_agcount = be32_to_cpu(from->sb_agcount);
 409         to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
 410         to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
 411         to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
 412         to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
 413         to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
 414         to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
 415         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
 416         to->sb_blocklog = from->sb_blocklog;
 417         to->sb_sectlog = from->sb_sectlog;
 418         to->sb_inodelog = from->sb_inodelog;
 419         to->sb_inopblog = from->sb_inopblog;
 420         to->sb_agblklog = from->sb_agblklog;
 421         to->sb_rextslog = from->sb_rextslog;
 422         to->sb_inprogress = from->sb_inprogress;
 423         to->sb_imax_pct = from->sb_imax_pct;
 424         to->sb_icount = be64_to_cpu(from->sb_icount);
 425         to->sb_ifree = be64_to_cpu(from->sb_ifree);
 426         to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
 427         to->sb_frextents = be64_to_cpu(from->sb_frextents);
 428         to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
 429         to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
 430         to->sb_qflags = be16_to_cpu(from->sb_qflags);
 431         to->sb_flags = from->sb_flags;
 432         to->sb_shared_vn = from->sb_shared_vn;
 433         to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
 434         to->sb_unit = be32_to_cpu(from->sb_unit);
 435         to->sb_width = be32_to_cpu(from->sb_width);
 436         to->sb_dirblklog = from->sb_dirblklog;
 437         to->sb_logsectlog = from->sb_logsectlog;
 438         to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
 439         to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
 440         to->sb_features2 = be32_to_cpu(from->sb_features2);
 441         to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
 442         to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
 443         to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
 444         to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
 445         to->sb_features_log_incompat =
 446                                 be32_to_cpu(from->sb_features_log_incompat);
 447         /* crc is only used on disk, not in memory; just init to 0 here. */
 448         to->sb_crc = 0;
 449         to->sb_pad = 0;
 450         to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
 451         to->sb_lsn = be64_to_cpu(from->sb_lsn);
 452         /* Convert on-disk flags to in-memory flags? */
 453         if (convert_xquota)
 454                 xfs_sb_quota_from_disk(to);
 455 }
 456
 457 void
 458 xfs_sb_from_disk(
 459         struct xfs_sb   *to,
 460         xfs_dsb_t       *from)
 461 {
 462         __xfs_sb_from_disk(to, from, true);
 463 }
 464
 465 static inline void
 466 xfs_sb_quota_to_disk(
 467         xfs_dsb_t       *to,
 468         xfs_sb_t        *from,
 469         __int64_t       *fields)
 470 {
 471         __uint16_t      qflags = from->sb_qflags;
 472
 473         /*
 474          * We need to do these manipilations only if we are working
 475          * with an older version of on-disk superblock.
 476          */
 477         if (xfs_sb_version_has_pquotino(from))
 478                 return;
 479
 480         if (*fields & XFS_SB_QFLAGS) {
 481                 /*
 482                  * The in-core version of sb_qflags do not have
 483                  * XFS_OQUOTA_* flags, whereas the on-disk version
 484                  * does.  So, convert incore XFS_{PG}QUOTA_* flags
 485                  * to on-disk XFS_OQUOTA_* flags.
 486                  */
 487                 qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
 488                                 XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
 489
 490                 if (from->sb_qflags &
 491                                 (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
 492                         qflags |= XFS_OQUOTA_ENFD;
 493                 if (from->sb_qflags &
 494                                 (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
 495                         qflags |= XFS_OQUOTA_CHKD;
 496                 to->sb_qflags = cpu_to_be16(qflags);
 497                 *fields &= ~XFS_SB_QFLAGS;
 498         }
 499
 500         /*
 501          * GQUOTINO and PQUOTINO cannot be used together in versions of
 502          * superblock that do not have pquotino. from->sb_flags tells us which
 503          * quota is active and should be copied to disk. If neither are active,
 504          * make sure we write NULLFSINO to the sb_gquotino field as a quota
 505          * inode value of "0" is invalid when the XFS_SB_VERSION_QUOTA feature
 506          * bit is set.
 507          *
 508          * Note that we don't need to handle the sb_uquotino or sb_pquotino here
 509          * as they do not require any translation. Hence the main sb field loop
 510          * will write them appropriately from the in-core superblock.
 511          */
 512         if ((*fields & XFS_SB_GQUOTINO) &&
 513                                 (from->sb_qflags & XFS_GQUOTA_ACCT))
 514                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
 515         else if ((*fields & XFS_SB_PQUOTINO) &&
 516                                 (from->sb_qflags & XFS_PQUOTA_ACCT))
 517                 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
 518         else {
 519                 /*
 520                  * We can't rely on just the fields being logged to tell us
 521                  * that it is safe to write NULLFSINO - we should only do that
 522                  * if quotas are not actually enabled. Hence only write
 523                  * NULLFSINO if both in-core quota inodes are NULL.
 524                  */
 525                 if (from->sb_gquotino == NULLFSINO &&
 526                     from->sb_pquotino == NULLFSINO)
 527                         to->sb_gquotino = cpu_to_be64(NULLFSINO);
 528         }
 529
 530         *fields &= ~(XFS_SB_PQUOTINO | XFS_SB_GQUOTINO);
 531 }
 532
 533 /*
 534  * Copy in core superblock to ondisk one.
 535  *
 536  * The fields argument is mask of superblock fields to copy.
 537  */
 538 void
 539 xfs_sb_to_disk(
 540         xfs_dsb_t       *to,
 541         xfs_sb_t        *from,
 542         __int64_t       fields)
 543 {
 544         xfs_caddr_t     to_ptr = (xfs_caddr_t)to;
 545         xfs_caddr_t     from_ptr = (xfs_caddr_t)from;
 546         xfs_sb_field_t  f;
 547         int             first;
 548         int             size;
 549
 550         ASSERT(fields);
 551         if (!fields)
 552                 return;
 553
 554         /* We should never write the crc here, it's updated in the IO path */
 555         fields &= ~XFS_SB_CRC;
 556
 557         xfs_sb_quota_to_disk(to, from, &fields);
 558         while (fields) {
 559                 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 560                 first = xfs_sb_info[f].offset;
 561                 size = xfs_sb_info[f + 1].offset - first;
 562
 563                 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
 564
 565                 if (size == 1 || xfs_sb_info[f].type == 1) {
 566                         memcpy(to_ptr + first, from_ptr + first, size);
 567                 } else {
 568                         switch (size) {
 569                         case 2:
 570                                 *(__be16 *)(to_ptr + first) =
 571                                       cpu_to_be16(*(__u16 *)(from_ptr + first));
 572                                 break;
 573                         case 4:
 574                                 *(__be32 *)(to_ptr + first) =
 575                                       cpu_to_be32(*(__u32 *)(from_ptr + first));
 576                                 break;
 577                         case 8:
 578                                 *(__be64 *)(to_ptr + first) =
 579                                       cpu_to_be64(*(__u64 *)(from_ptr + first));
 580                                 break;
 581                         default:
 582                                 ASSERT(0);
 583                         }
 584                 }
 585
 586                 fields &= ~(1LL << f);
 587         }
 588 }
 589
 590 static int
 591 xfs_sb_verify(
 592         struct xfs_buf  *bp,
 593         bool            check_version)
 594 {
 595         struct xfs_mount *mp = bp->b_target->bt_mount;
 596         struct xfs_sb   sb;
 597
 598         /*
 599          * Use call variant which doesn't convert quota flags from disk
 600          * format, because xfs_mount_validate_sb checks the on-disk flags.
 601          */
 602         __xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp), false);
 603
 604         /*
 605          * Only check the in progress field for the primary superblock as
 606          * mkfs.xfs doesn't clear it from secondary superblocks.
 607          */
 608         return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR,
 609                                      check_version);
 610 }
 611
 612 /*
 613  * If the superblock has the CRC feature bit set or the CRC field is non-null,
 614  * check that the CRC is valid.  We check the CRC field is non-null because a
 615  * single bit error could clear the feature bit and unused parts of the
 616  * superblock are supposed to be zero. Hence a non-null crc field indicates that
 617  * we've potentially lost a feature bit and we should check it anyway.
 618  *
 619  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
 620  * last field in V4 secondary superblocks.  So for secondary superblocks,
 621  * we are more forgiving, and ignore CRC failures if the primary doesn't
 622  * indicate that the fs version is V5.
 623  */
 624 static void
 625 xfs_sb_read_verify(
 626         struct xfs_buf  *bp)
 627 {
 628         struct xfs_mount *mp = bp->b_target->bt_mount;
 629         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 630         int             error;
 631
 632         /*
 633          * open code the version check to avoid needing to convert the entire
 634          * superblock from disk order just to check the version number
 635          */
 636         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
 637             (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
 638                                                 XFS_SB_VERSION_5) ||
 639              dsb->sb_crc != 0)) {
 640
 641                 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
 642                         /* Only fail bad secondaries on a known V5 filesystem */
 643                         if (bp->b_bn == XFS_SB_DADDR ||
 644                             xfs_sb_version_hascrc(&mp->m_sb)) {
 645                                 error = -EFSBADCRC;
 646                                 goto out_error;
 647                         }
 648                 }
 649         }
 650         error = xfs_sb_verify(bp, true);
 651
 652 out_error:
 653         if (error) {
 654                 xfs_buf_ioerror(bp, error);
 655                 if (error == -EFSCORRUPTED || error == -EFSBADCRC)
 656                         xfs_verifier_error(bp);
 657         }
 658 }
 659
 660 /*
 661  * We may be probed for a filesystem match, so we may not want to emit
 662  * messages when the superblock buffer is not actually an XFS superblock.
 663  * If we find an XFS superblock, then run a normal, noisy mount because we are
 664  * really going to mount it and want to know about errors.
 665  */
 666 static void
 667 xfs_sb_quiet_read_verify(
 668         struct xfs_buf  *bp)
 669 {
 670         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 671
 672         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
 673                 /* XFS filesystem, verify noisily! */
 674                 xfs_sb_read_verify(bp);
 675                 return;
 676         }
 677         /* quietly fail */
 678         xfs_buf_ioerror(bp, -EWRONGFS);
 679 }
 680
 681 static void
 682 xfs_sb_write_verify(
 683         struct xfs_buf          *bp)
 684 {
 685         struct xfs_mount        *mp = bp->b_target->bt_mount;
 686         struct xfs_buf_log_item *bip = bp->b_fspriv;
 687         int                     error;
 688
 689         error = xfs_sb_verify(bp, false);
 690         if (error) {
 691                 xfs_buf_ioerror(bp, error);
 692                 xfs_verifier_error(bp);
 693                 return;
 694         }
 695
 696         if (!xfs_sb_version_hascrc(&mp->m_sb))
 697                 return;
 698
 699         if (bip)
 700                 XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 701
 702         xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
 703 }
 704
 705 const struct xfs_buf_ops xfs_sb_buf_ops = {
 706         .verify_read = xfs_sb_read_verify,
 707         .verify_write = xfs_sb_write_verify,
 708 };
 709
 710 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
 711         .verify_read = xfs_sb_quiet_read_verify,
 712         .verify_write = xfs_sb_write_verify,
 713 };
 714
 715 /*
 716  * xfs_mount_common
 717  *
 718  * Mount initialization code establishing various mount
 719  * fields from the superblock associated with the given
 720  * mount structure
 721  */
 722 void
 723 xfs_sb_mount_common(
 724         struct xfs_mount *mp,
 725         struct xfs_sb   *sbp)
 726 {
 727         mp->m_agfrotor = mp->m_agirotor = 0;
 728         spin_lock_init(&mp->m_agirotor_lock);
 729         mp->m_maxagi = mp->m_sb.sb_agcount;
 730         mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
 731         mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
 732         mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
 733         mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
 734         mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
 735         mp->m_blockmask = sbp->sb_blocksize - 1;
 736         mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
 737         mp->m_blockwmask = mp->m_blockwsize - 1;
 738
 739         mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
 740         mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
 741         mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
 742         mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
 743
 744         mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
 745         mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
 746         mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
 747         mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
 748
 749         mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
 750         mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
 751         mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
 752         mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
 753
 754         mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
 755         mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 756                                         sbp->sb_inopblock);
 757         mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 758 }
 759
 760 /*
 761  * xfs_initialize_perag_data
 762  *
 763  * Read in each per-ag structure so we can count up the number of
 764  * allocated inodes, free inodes and used filesystem blocks as this
 765  * information is no longer persistent in the superblock. Once we have
 766  * this information, write it into the in-core superblock structure.
 767  */
 768 int
 769 xfs_initialize_perag_data(
 770         struct xfs_mount *mp,
 771         xfs_agnumber_t  agcount)
 772 {
 773         xfs_agnumber_t  index;
 774         xfs_perag_t     *pag;
 775         xfs_sb_t        *sbp = &mp->m_sb;
 776         uint64_t        ifree = 0;
 777         uint64_t        ialloc = 0;
 778         uint64_t        bfree = 0;
 779         uint64_t        bfreelst = 0;
 780         uint64_t        btree = 0;
 781         int             error;
 782
 783         for (index = 0; index < agcount; index++) {
 784                 /*
 785                  * read the agf, then the agi. This gets us
 786                  * all the information we need and populates the
 787                  * per-ag structures for us.
 788                  */
 789                 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
 790                 if (error)
 791                         return error;
 792
 793                 error = xfs_ialloc_pagi_init(mp, NULL, index);
 794                 if (error)
 795                         return error;
 796                 pag = xfs_perag_get(mp, index);
 797                 ifree += pag->pagi_freecount;
 798                 ialloc += pag->pagi_count;
 799                 bfree += pag->pagf_freeblks;
 800                 bfreelst += pag->pagf_flcount;
 801                 btree += pag->pagf_btreeblks;
 802                 xfs_perag_put(pag);
 803         }
 804         /*
 805          * Overwrite incore superblock counters with just-read data
 806          */
 807         spin_lock(&mp->m_sb_lock);
 808         sbp->sb_ifree = ifree;
 809         sbp->sb_icount = ialloc;
 810         sbp->sb_fdblocks = bfree + bfreelst + btree;
 811         spin_unlock(&mp->m_sb_lock);
 812
 813         /* Fixup the per-cpu counters as well. */
 814         xfs_icsb_reinit_counters(mp);
 815
 816         return 0;
 817 }
 818
 819 /*
 820  * xfs_mod_sb() can be used to copy arbitrary changes to the
 821  * in-core superblock into the superblock buffer to be logged.
 822  * It does not provide the higher level of locking that is
 823  * needed to protect the in-core superblock from concurrent
 824  * access.
 825  */
 826 void
 827 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
 828 {
 829         xfs_buf_t       *bp;
 830         int             first;
 831         int             last;
 832         xfs_mount_t     *mp;
 833         xfs_sb_field_t  f;
 834
 835         ASSERT(fields);
 836         if (!fields)
 837                 return;
 838         mp = tp->t_mountp;
 839         bp = xfs_trans_getsb(tp, mp, 0);
 840         first = sizeof(xfs_sb_t);
 841         last = 0;
 842
 843         /* translate/copy */
 844
 845         xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
 846
 847         /* find modified range */
 848         f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
 849         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
 850         last = xfs_sb_info[f + 1].offset - 1;
 851
 852         f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 853         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
 854         first = xfs_sb_info[f].offset;
 855
 856         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
 857         xfs_trans_log_buf(tp, bp, first, last);
 858 }