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_mount.h"
  27 #include "xfs_inode.h"
  28 #include "xfs_ialloc.h"
  29 #include "xfs_alloc.h"
  30 #include "xfs_error.h"
  31 #include "xfs_trace.h"
  32 #include "xfs_cksum.h"
  33 #include "xfs_trans.h"
  34 #include "xfs_buf_item.h"
  35 #include "xfs_bmap_btree.h"
  36 #include "xfs_alloc_btree.h"
  37 #include "xfs_ialloc_btree.h"
  38
  39 /*
  40  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
  41  */
  42
  43 /*
  44  * Reference counting access wrappers to the perag structures.
  45  * Because we never free per-ag structures, the only thing we
  46  * have to protect against changes is the tree structure itself.
  47  */
  48 struct xfs_perag *
  49 xfs_perag_get(
  50         struct xfs_mount        *mp,
  51         xfs_agnumber_t          agno)
  52 {
  53         struct xfs_perag        *pag;
  54         int                     ref = 0;
  55
  56         rcu_read_lock();
  57         pag = radix_tree_lookup(&mp->m_perag_tree, agno);
  58         if (pag) {
  59                 ASSERT(atomic_read(&pag->pag_ref) >= 0);
  60                 ref = atomic_inc_return(&pag->pag_ref);
  61         }
  62         rcu_read_unlock();
  63         trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
  64         return pag;
  65 }
  66
  67 /*
  68  * search from @first to find the next perag with the given tag set.
  69  */
  70 struct xfs_perag *
  71 xfs_perag_get_tag(
  72         struct xfs_mount        *mp,
  73         xfs_agnumber_t          first,
  74         int                     tag)
  75 {
  76         struct xfs_perag        *pag;
  77         int                     found;
  78         int                     ref;
  79
  80         rcu_read_lock();
  81         found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
  82                                         (void **)&pag, first, 1, tag);
  83         if (found <= 0) {
  84                 rcu_read_unlock();
  85                 return NULL;
  86         }
  87         ref = atomic_inc_return(&pag->pag_ref);
  88         rcu_read_unlock();
  89         trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
  90         return pag;
  91 }
  92
  93 void
  94 xfs_perag_put(
  95         struct xfs_perag        *pag)
  96 {
  97         int     ref;
  98
  99         ASSERT(atomic_read(&pag->pag_ref) > 0);
 100         ref = atomic_dec_return(&pag->pag_ref);
 101         trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
 102 }
 103
 104 /*
 105  * Check the validity of the SB found.
 106  */
 107 STATIC int
 108 xfs_mount_validate_sb(
 109         xfs_mount_t     *mp,
 110         xfs_sb_t        *sbp,
 111         bool            check_inprogress,
 112         bool            check_version)
 113 {
 114         if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 115                 xfs_warn(mp, "bad magic number");
 116                 return -EWRONGFS;
 117         }
 118
 119
 120         if (!xfs_sb_good_version(sbp)) {
 121                 xfs_warn(mp, "bad version");
 122                 return -EWRONGFS;
 123         }
 124
 125         /*
 126          * Version 5 superblock feature mask validation. Reject combinations the
 127          * kernel cannot support up front before checking anything else. For
 128          * write validation, we don't need to check feature masks.
 129          */
 130         if (check_version && XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5) {
 131                 if (xfs_sb_has_compat_feature(sbp,
 132                                         XFS_SB_FEAT_COMPAT_UNKNOWN)) {
 133                         xfs_warn(mp,
 134 "Superblock has unknown compatible features (0x%x) enabled.\n"
 135 "Using a more recent kernel is recommended.",
 136                                 (sbp->sb_features_compat &
 137                                                 XFS_SB_FEAT_COMPAT_UNKNOWN));
 138                 }
 139
 140                 if (xfs_sb_has_ro_compat_feature(sbp,
 141                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
 142                         xfs_alert(mp,
 143 "Superblock has unknown read-only compatible features (0x%x) enabled.",
 144                                 (sbp->sb_features_ro_compat &
 145                                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
 146                         if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 147                                 xfs_warn(mp,
 148 "Attempted to mount read-only compatible filesystem read-write.\n"
 149 "Filesystem can only be safely mounted read only.");
 150                                 return -EINVAL;
 151                         }
 152                 }
 153                 if (xfs_sb_has_incompat_feature(sbp,
 154                                         XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
 155                         xfs_warn(mp,
 156 "Superblock has unknown incompatible features (0x%x) enabled.\n"
 157 "Filesystem can not be safely mounted by this kernel.",
 158                                 (sbp->sb_features_incompat &
 159                                                 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
 160                         return -EINVAL;
 161                 }
 162         }
 163
 164         if (xfs_sb_version_has_pquotino(sbp)) {
 165                 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
 166                         xfs_notice(mp,
 167                            "Version 5 of Super block has XFS_OQUOTA bits.");
 168                         return -EFSCORRUPTED;
 169                 }
 170         } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
 171                                 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
 172                         xfs_notice(mp,
 173 "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.");
 174                         return -EFSCORRUPTED;
 175         }
 176
 177         /*
 178          * Full inode chunks must be aligned to inode chunk size when
 179          * sparse inodes are enabled to support the sparse chunk
 180          * allocation algorithm and prevent overlapping inode records.
 181          */
 182         if (xfs_sb_version_hassparseinodes(sbp)) {
 183                 uint32_t        align;
 184
 185                 align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
 186                                 >> sbp->sb_blocklog;
 187                 if (sbp->sb_inoalignmt != align) {
 188                         xfs_warn(mp,
 189 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
 190                                  sbp->sb_inoalignmt, align);
 191                         return -EINVAL;
 192                 }
 193         }
 194
 195         if (unlikely(
 196             sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
 197                 xfs_warn(mp,
 198                 "filesystem is marked as having an external log; "
 199                 "specify logdev on the mount command line.");
 200                 return -EINVAL;
 201         }
 202
 203         if (unlikely(
 204             sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
 205                 xfs_warn(mp,
 206                 "filesystem is marked as having an internal log; "
 207                 "do not specify logdev on the mount command line.");
 208                 return -EINVAL;
 209         }
 210
 211         /*
 212          * More sanity checking.  Most of these were stolen directly from
 213          * xfs_repair.
 214          */
 215         if (unlikely(
 216             sbp->sb_agcount <= 0                                        ||
 217             sbp->sb_sectsize < XFS_MIN_SECTORSIZE                       ||
 218             sbp->sb_sectsize > XFS_MAX_SECTORSIZE                       ||
 219             sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG                    ||
 220             sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG                    ||
 221             sbp->sb_sectsize != (1 << sbp->sb_sectlog)                  ||
 222             sbp->sb_blocksize < XFS_MIN_BLOCKSIZE                       ||
 223             sbp->sb_blocksize > XFS_MAX_BLOCKSIZE                       ||
 224             sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG                    ||
 225             sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG                    ||
 226             sbp->sb_blocksize != (1 << sbp->sb_blocklog)                ||
 227             sbp->sb_dirblklog > XFS_MAX_BLOCKSIZE_LOG                   ||
 228             sbp->sb_inodesize < XFS_DINODE_MIN_SIZE                     ||
 229             sbp->sb_inodesize > XFS_DINODE_MAX_SIZE                     ||
 230             sbp->sb_inodelog < XFS_DINODE_MIN_LOG                       ||
 231             sbp->sb_inodelog > XFS_DINODE_MAX_LOG                       ||
 232             sbp->sb_inodesize != (1 << sbp->sb_inodelog)                ||
 233             sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE                    ||
 234             sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
 235             (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
 236             (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
 237             (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
 238             (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)    ||
 239             sbp->sb_dblocks == 0                                        ||
 240             sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)                      ||
 241             sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp)                      ||
 242             sbp->sb_shared_vn != 0)) {
 243                 xfs_notice(mp, "SB sanity check failed");
 244                 return -EFSCORRUPTED;
 245         }
 246
 247         /*
 248          * Until this is fixed only page-sized or smaller data blocks work.
 249          */
 250         if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
 251                 xfs_warn(mp,
 252                 "File system with blocksize %d bytes. "
 253                 "Only pagesize (%ld) or less will currently work.",
 254                                 sbp->sb_blocksize, PAGE_SIZE);
 255                 return -ENOSYS;
 256         }
 257
 258         /*
 259          * Currently only very few inode sizes are supported.
 260          */
 261         switch (sbp->sb_inodesize) {
 262         case 256:
 263         case 512:
 264         case 1024:
 265         case 2048:
 266                 break;
 267         default:
 268                 xfs_warn(mp, "inode size of %d bytes not supported",
 269                                 sbp->sb_inodesize);
 270                 return -ENOSYS;
 271         }
 272
 273         if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
 274             xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
 275                 xfs_warn(mp,
 276                 "file system too large to be mounted on this system.");
 277                 return -EFBIG;
 278         }
 279
 280         if (check_inprogress && sbp->sb_inprogress) {
 281                 xfs_warn(mp, "Offline file system operation in progress!");
 282                 return -EFSCORRUPTED;
 283         }
 284         return 0;
 285 }
 286
 287 void
 288 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
 289 {
 290         /*
 291          * older mkfs doesn't initialize quota inodes to NULLFSINO. This
 292          * leads to in-core values having two different values for a quota
 293          * inode to be invalid: 0 and NULLFSINO. Change it to a single value
 294          * NULLFSINO.
 295          *
 296          * Note that this change affect only the in-core values. These
 297          * values are not written back to disk unless any quota information
 298          * is written to the disk. Even in that case, sb_pquotino field is
 299          * not written to disk unless the superblock supports pquotino.
 300          */
 301         if (sbp->sb_uquotino == 0)
 302                 sbp->sb_uquotino = NULLFSINO;
 303         if (sbp->sb_gquotino == 0)
 304                 sbp->sb_gquotino = NULLFSINO;
 305         if (sbp->sb_pquotino == 0)
 306                 sbp->sb_pquotino = NULLFSINO;
 307
 308         /*
 309          * We need to do these manipilations only if we are working
 310          * with an older version of on-disk superblock.
 311          */
 312         if (xfs_sb_version_has_pquotino(sbp))
 313                 return;
 314
 315         if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
 316                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 317                                         XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
 318         if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
 319                 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 320                                         XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
 321         sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
 322
 323         if (sbp->sb_qflags & XFS_PQUOTA_ACCT)  {
 324                 /*
 325                  * In older version of superblock, on-disk superblock only
 326                  * has sb_gquotino, and in-core superblock has both sb_gquotino
 327                  * and sb_pquotino. But, only one of them is supported at any
 328                  * point of time. So, if PQUOTA is set in disk superblock,
 329                  * copy over sb_gquotino to sb_pquotino.
 330                  */
 331                 sbp->sb_pquotino = sbp->sb_gquotino;
 332                 sbp->sb_gquotino = NULLFSINO;
 333         }
 334 }
 335
 336 static void
 337 __xfs_sb_from_disk(
 338         struct xfs_sb   *to,
 339         xfs_dsb_t       *from,
 340         bool            convert_xquota)
 341 {
 342         to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
 343         to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
 344         to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
 345         to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
 346         to->sb_rextents = be64_to_cpu(from->sb_rextents);
 347         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
 348         to->sb_logstart = be64_to_cpu(from->sb_logstart);
 349         to->sb_rootino = be64_to_cpu(from->sb_rootino);
 350         to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
 351         to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
 352         to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
 353         to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
 354         to->sb_agcount = be32_to_cpu(from->sb_agcount);
 355         to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
 356         to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
 357         to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
 358         to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
 359         to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
 360         to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
 361         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
 362         to->sb_blocklog = from->sb_blocklog;
 363         to->sb_sectlog = from->sb_sectlog;
 364         to->sb_inodelog = from->sb_inodelog;
 365         to->sb_inopblog = from->sb_inopblog;
 366         to->sb_agblklog = from->sb_agblklog;
 367         to->sb_rextslog = from->sb_rextslog;
 368         to->sb_inprogress = from->sb_inprogress;
 369         to->sb_imax_pct = from->sb_imax_pct;
 370         to->sb_icount = be64_to_cpu(from->sb_icount);
 371         to->sb_ifree = be64_to_cpu(from->sb_ifree);
 372         to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
 373         to->sb_frextents = be64_to_cpu(from->sb_frextents);
 374         to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
 375         to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
 376         to->sb_qflags = be16_to_cpu(from->sb_qflags);
 377         to->sb_flags = from->sb_flags;
 378         to->sb_shared_vn = from->sb_shared_vn;
 379         to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
 380         to->sb_unit = be32_to_cpu(from->sb_unit);
 381         to->sb_width = be32_to_cpu(from->sb_width);
 382         to->sb_dirblklog = from->sb_dirblklog;
 383         to->sb_logsectlog = from->sb_logsectlog;
 384         to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
 385         to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
 386         to->sb_features2 = be32_to_cpu(from->sb_features2);
 387         to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
 388         to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
 389         to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
 390         to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
 391         to->sb_features_log_incompat =
 392                                 be32_to_cpu(from->sb_features_log_incompat);
 393         /* crc is only used on disk, not in memory; just init to 0 here. */
 394         to->sb_crc = 0;
 395         to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
 396         to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
 397         to->sb_lsn = be64_to_cpu(from->sb_lsn);
 398         /*
 399          * sb_meta_uuid is only on disk if it differs from sb_uuid and the
 400          * feature flag is set; if not set we keep it only in memory.
 401          */
 402         if (xfs_sb_version_hasmetauuid(to))
 403                 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
 404         else
 405                 uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
 406         /* Convert on-disk flags to in-memory flags? */
 407         if (convert_xquota)
 408                 xfs_sb_quota_from_disk(to);
 409 }
 410
 411 void
 412 xfs_sb_from_disk(
 413         struct xfs_sb   *to,
 414         xfs_dsb_t       *from)
 415 {
 416         __xfs_sb_from_disk(to, from, true);
 417 }
 418
 419 static void
 420 xfs_sb_quota_to_disk(
 421         struct xfs_dsb  *to,
 422         struct xfs_sb   *from)
 423 {
 424         __uint16_t      qflags = from->sb_qflags;
 425
 426         to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
 427         if (xfs_sb_version_has_pquotino(from)) {
 428                 to->sb_qflags = cpu_to_be16(from->sb_qflags);
 429                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
 430                 to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
 431                 return;
 432         }
 433
 434         /*
 435          * The in-core version of sb_qflags do not have XFS_OQUOTA_*
 436          * flags, whereas the on-disk version does.  So, convert incore
 437          * XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
 438          */
 439         qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
 440                         XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
 441
 442         if (from->sb_qflags &
 443                         (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
 444                 qflags |= XFS_OQUOTA_ENFD;
 445         if (from->sb_qflags &
 446                         (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
 447                 qflags |= XFS_OQUOTA_CHKD;
 448         to->sb_qflags = cpu_to_be16(qflags);
 449
 450         /*
 451          * GQUOTINO and PQUOTINO cannot be used together in versions
 452          * of superblock that do not have pquotino. from->sb_flags
 453          * tells us which quota is active and should be copied to
 454          * disk. If neither are active, we should NULL the inode.
 455          *
 456          * In all cases, the separate pquotino must remain 0 because it
 457          * it beyond the "end" of the valid non-pquotino superblock.
 458          */
 459         if (from->sb_qflags & XFS_GQUOTA_ACCT)
 460                 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
 461         else if (from->sb_qflags & XFS_PQUOTA_ACCT)
 462                 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
 463         else {
 464                 /*
 465                  * We can't rely on just the fields being logged to tell us
 466                  * that it is safe to write NULLFSINO - we should only do that
 467                  * if quotas are not actually enabled. Hence only write
 468                  * NULLFSINO if both in-core quota inodes are NULL.
 469                  */
 470                 if (from->sb_gquotino == NULLFSINO &&
 471                     from->sb_pquotino == NULLFSINO)
 472                         to->sb_gquotino = cpu_to_be64(NULLFSINO);
 473         }
 474
 475         to->sb_pquotino = 0;
 476 }
 477
 478 void
 479 xfs_sb_to_disk(
 480         struct xfs_dsb  *to,
 481         struct xfs_sb   *from)
 482 {
 483         xfs_sb_quota_to_disk(to, from);
 484
 485         to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
 486         to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
 487         to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
 488         to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
 489         to->sb_rextents = cpu_to_be64(from->sb_rextents);
 490         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
 491         to->sb_logstart = cpu_to_be64(from->sb_logstart);
 492         to->sb_rootino = cpu_to_be64(from->sb_rootino);
 493         to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
 494         to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
 495         to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
 496         to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
 497         to->sb_agcount = cpu_to_be32(from->sb_agcount);
 498         to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
 499         to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
 500         to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
 501         to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
 502         to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
 503         to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
 504         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
 505         to->sb_blocklog = from->sb_blocklog;
 506         to->sb_sectlog = from->sb_sectlog;
 507         to->sb_inodelog = from->sb_inodelog;
 508         to->sb_inopblog = from->sb_inopblog;
 509         to->sb_agblklog = from->sb_agblklog;
 510         to->sb_rextslog = from->sb_rextslog;
 511         to->sb_inprogress = from->sb_inprogress;
 512         to->sb_imax_pct = from->sb_imax_pct;
 513         to->sb_icount = cpu_to_be64(from->sb_icount);
 514         to->sb_ifree = cpu_to_be64(from->sb_ifree);
 515         to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
 516         to->sb_frextents = cpu_to_be64(from->sb_frextents);
 517
 518         to->sb_flags = from->sb_flags;
 519         to->sb_shared_vn = from->sb_shared_vn;
 520         to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
 521         to->sb_unit = cpu_to_be32(from->sb_unit);
 522         to->sb_width = cpu_to_be32(from->sb_width);
 523         to->sb_dirblklog = from->sb_dirblklog;
 524         to->sb_logsectlog = from->sb_logsectlog;
 525         to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
 526         to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
 527
 528         /*
 529          * We need to ensure that bad_features2 always matches features2.
 530          * Hence we enforce that here rather than having to remember to do it
 531          * everywhere else that updates features2.
 532          */
 533         from->sb_bad_features2 = from->sb_features2;
 534         to->sb_features2 = cpu_to_be32(from->sb_features2);
 535         to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
 536
 537         if (xfs_sb_version_hascrc(from)) {
 538                 to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
 539                 to->sb_features_ro_compat =
 540                                 cpu_to_be32(from->sb_features_ro_compat);
 541                 to->sb_features_incompat =
 542                                 cpu_to_be32(from->sb_features_incompat);
 543                 to->sb_features_log_incompat =
 544                                 cpu_to_be32(from->sb_features_log_incompat);
 545                 to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
 546                 to->sb_lsn = cpu_to_be64(from->sb_lsn);
 547                 if (xfs_sb_version_hasmetauuid(from))
 548                         uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
 549         }
 550 }
 551
 552 static int
 553 xfs_sb_verify(
 554         struct xfs_buf  *bp,
 555         bool            check_version)
 556 {
 557         struct xfs_mount *mp = bp->b_target->bt_mount;
 558         struct xfs_sb   sb;
 559
 560         /*
 561          * Use call variant which doesn't convert quota flags from disk
 562          * format, because xfs_mount_validate_sb checks the on-disk flags.
 563          */
 564         __xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp), false);
 565
 566         /*
 567          * Only check the in progress field for the primary superblock as
 568          * mkfs.xfs doesn't clear it from secondary superblocks.
 569          */
 570         return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR,
 571                                      check_version);
 572 }
 573
 574 /*
 575  * If the superblock has the CRC feature bit set or the CRC field is non-null,
 576  * check that the CRC is valid.  We check the CRC field is non-null because a
 577  * single bit error could clear the feature bit and unused parts of the
 578  * superblock are supposed to be zero. Hence a non-null crc field indicates that
 579  * we've potentially lost a feature bit and we should check it anyway.
 580  *
 581  * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
 582  * last field in V4 secondary superblocks.  So for secondary superblocks,
 583  * we are more forgiving, and ignore CRC failures if the primary doesn't
 584  * indicate that the fs version is V5.
 585  */
 586 static void
 587 xfs_sb_read_verify(
 588         struct xfs_buf  *bp)
 589 {
 590         struct xfs_mount *mp = bp->b_target->bt_mount;
 591         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 592         int             error;
 593
 594         /*
 595          * open code the version check to avoid needing to convert the entire
 596          * superblock from disk order just to check the version number
 597          */
 598         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
 599             (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
 600                                                 XFS_SB_VERSION_5) ||
 601              dsb->sb_crc != 0)) {
 602
 603                 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
 604                         /* Only fail bad secondaries on a known V5 filesystem */
 605                         if (bp->b_bn == XFS_SB_DADDR ||
 606                             xfs_sb_version_hascrc(&mp->m_sb)) {
 607                                 error = -EFSBADCRC;
 608                                 goto out_error;
 609                         }
 610                 }
 611         }
 612         error = xfs_sb_verify(bp, true);
 613
 614 out_error:
 615         if (error) {
 616                 xfs_buf_ioerror(bp, error);
 617                 if (error == -EFSCORRUPTED || error == -EFSBADCRC)
 618                         xfs_verifier_error(bp);
 619         }
 620 }
 621
 622 /*
 623  * We may be probed for a filesystem match, so we may not want to emit
 624  * messages when the superblock buffer is not actually an XFS superblock.
 625  * If we find an XFS superblock, then run a normal, noisy mount because we are
 626  * really going to mount it and want to know about errors.
 627  */
 628 static void
 629 xfs_sb_quiet_read_verify(
 630         struct xfs_buf  *bp)
 631 {
 632         struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 633
 634         if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
 635                 /* XFS filesystem, verify noisily! */
 636                 xfs_sb_read_verify(bp);
 637                 return;
 638         }
 639         /* quietly fail */
 640         xfs_buf_ioerror(bp, -EWRONGFS);
 641 }
 642
 643 static void
 644 xfs_sb_write_verify(
 645         struct xfs_buf          *bp)
 646 {
 647         struct xfs_mount        *mp = bp->b_target->bt_mount;
 648         struct xfs_buf_log_item *bip = bp->b_fspriv;
 649         int                     error;
 650
 651         error = xfs_sb_verify(bp, false);
 652         if (error) {
 653                 xfs_buf_ioerror(bp, error);
 654                 xfs_verifier_error(bp);
 655                 return;
 656         }
 657
 658         if (!xfs_sb_version_hascrc(&mp->m_sb))
 659                 return;
 660
 661         if (bip)
 662                 XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 663
 664         xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
 665 }
 666
 667 const struct xfs_buf_ops xfs_sb_buf_ops = {
 668         .verify_read = xfs_sb_read_verify,
 669         .verify_write = xfs_sb_write_verify,
 670 };
 671
 672 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
 673         .verify_read = xfs_sb_quiet_read_verify,
 674         .verify_write = xfs_sb_write_verify,
 675 };
 676
 677 /*
 678  * xfs_mount_common
 679  *
 680  * Mount initialization code establishing various mount
 681  * fields from the superblock associated with the given
 682  * mount structure
 683  */
 684 void
 685 xfs_sb_mount_common(
 686         struct xfs_mount *mp,
 687         struct xfs_sb   *sbp)
 688 {
 689         mp->m_agfrotor = mp->m_agirotor = 0;
 690         spin_lock_init(&mp->m_agirotor_lock);
 691         mp->m_maxagi = mp->m_sb.sb_agcount;
 692         mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
 693         mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
 694         mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
 695         mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
 696         mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
 697         mp->m_blockmask = sbp->sb_blocksize - 1;
 698         mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
 699         mp->m_blockwmask = mp->m_blockwsize - 1;
 700
 701         mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
 702         mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
 703         mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
 704         mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
 705
 706         mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
 707         mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
 708         mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
 709         mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
 710
 711         mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
 712         mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
 713         mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
 714         mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
 715
 716         mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
 717         mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 718                                         sbp->sb_inopblock);
 719         mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 720
 721         if (sbp->sb_spino_align)
 722                 mp->m_ialloc_min_blks = sbp->sb_spino_align;
 723         else
 724                 mp->m_ialloc_min_blks = mp->m_ialloc_blks;
 725 }
 726
 727 /*
 728  * xfs_initialize_perag_data
 729  *
 730  * Read in each per-ag structure so we can count up the number of
 731  * allocated inodes, free inodes and used filesystem blocks as this
 732  * information is no longer persistent in the superblock. Once we have
 733  * this information, write it into the in-core superblock structure.
 734  */
 735 int
 736 xfs_initialize_perag_data(
 737         struct xfs_mount *mp,
 738         xfs_agnumber_t  agcount)
 739 {
 740         xfs_agnumber_t  index;
 741         xfs_perag_t     *pag;
 742         xfs_sb_t        *sbp = &mp->m_sb;
 743         uint64_t        ifree = 0;
 744         uint64_t        ialloc = 0;
 745         uint64_t        bfree = 0;
 746         uint64_t        bfreelst = 0;
 747         uint64_t        btree = 0;
 748         int             error;
 749
 750         for (index = 0; index < agcount; index++) {
 751                 /*
 752                  * read the agf, then the agi. This gets us
 753                  * all the information we need and populates the
 754                  * per-ag structures for us.
 755                  */
 756                 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
 757                 if (error)
 758                         return error;
 759
 760                 error = xfs_ialloc_pagi_init(mp, NULL, index);
 761                 if (error)
 762                         return error;
 763                 pag = xfs_perag_get(mp, index);
 764                 ifree += pag->pagi_freecount;
 765                 ialloc += pag->pagi_count;
 766                 bfree += pag->pagf_freeblks;
 767                 bfreelst += pag->pagf_flcount;
 768                 btree += pag->pagf_btreeblks;
 769                 xfs_perag_put(pag);
 770         }
 771
 772         /* Overwrite incore superblock counters with just-read data */
 773         spin_lock(&mp->m_sb_lock);
 774         sbp->sb_ifree = ifree;
 775         sbp->sb_icount = ialloc;
 776         sbp->sb_fdblocks = bfree + bfreelst + btree;
 777         spin_unlock(&mp->m_sb_lock);
 778
 779         xfs_reinit_percpu_counters(mp);
 780
 781         return 0;
 782 }
 783
 784 /*
 785  * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
 786  * into the superblock buffer to be logged.  It does not provide the higher
 787  * level of locking that is needed to protect the in-core superblock from
 788  * concurrent access.
 789  */
 790 void
 791 xfs_log_sb(
 792         struct xfs_trans        *tp)
 793 {
 794         struct xfs_mount        *mp = tp->t_mountp;
 795         struct xfs_buf          *bp = xfs_trans_getsb(tp, mp, 0);
 796
 797         mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
 798         mp->m_sb.sb_ifree = percpu_counter_sum(&mp->m_ifree);
 799         mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 800
 801         xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
 802         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
 803         xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb));
 804 }
 805
 806 /*
 807  * xfs_sync_sb
 808  *
 809  * Sync the superblock to disk.
 810  *
 811  * Note that the caller is responsible for checking the frozen state of the
 812  * filesystem. This procedure uses the non-blocking transaction allocator and
 813  * thus will allow modifications to a frozen fs. This is required because this
 814  * code can be called during the process of freezing where use of the high-level
 815  * allocator would deadlock.
 816  */
 817 int
 818 xfs_sync_sb(
 819         struct xfs_mount        *mp,
 820         bool                    wait)
 821 {
 822         struct xfs_trans        *tp;
 823         int                     error;
 824
 825         tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_CHANGE, KM_SLEEP);
 826         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 827         if (error) {
 828                 xfs_trans_cancel(tp);
 829                 return error;
 830         }
 831
 832         xfs_log_sb(tp);
 833         if (wait)
 834                 xfs_trans_set_sync(tp);
 835         return xfs_trans_commit(tp);
 836 }