linux/fs/xfs/xfs_mount.c
<<
>>
Prefs
   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_types.h"
  21#include "xfs_bit.h"
  22#include "xfs_log.h"
  23#include "xfs_inum.h"
  24#include "xfs_trans.h"
  25#include "xfs_trans_priv.h"
  26#include "xfs_sb.h"
  27#include "xfs_ag.h"
  28#include "xfs_dir2.h"
  29#include "xfs_mount.h"
  30#include "xfs_bmap_btree.h"
  31#include "xfs_alloc_btree.h"
  32#include "xfs_ialloc_btree.h"
  33#include "xfs_dinode.h"
  34#include "xfs_inode.h"
  35#include "xfs_btree.h"
  36#include "xfs_ialloc.h"
  37#include "xfs_alloc.h"
  38#include "xfs_rtalloc.h"
  39#include "xfs_bmap.h"
  40#include "xfs_error.h"
  41#include "xfs_quota.h"
  42#include "xfs_fsops.h"
  43#include "xfs_utils.h"
  44#include "xfs_trace.h"
  45#include "xfs_icache.h"
  46#include "xfs_cksum.h"
  47#include "xfs_buf_item.h"
  48
  49
  50#ifdef HAVE_PERCPU_SB
  51STATIC void     xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
  52                                                int);
  53STATIC void     xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
  54                                                int);
  55STATIC void     xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
  56#else
  57
  58#define xfs_icsb_balance_counter(mp, a, b)              do { } while (0)
  59#define xfs_icsb_balance_counter_locked(mp, a, b)       do { } while (0)
  60#endif
  61
  62static const struct {
  63        short offset;
  64        short type;     /* 0 = integer
  65                         * 1 = binary / string (no translation)
  66                         */
  67} xfs_sb_info[] = {
  68    { offsetof(xfs_sb_t, sb_magicnum),   0 },
  69    { offsetof(xfs_sb_t, sb_blocksize),  0 },
  70    { offsetof(xfs_sb_t, sb_dblocks),    0 },
  71    { offsetof(xfs_sb_t, sb_rblocks),    0 },
  72    { offsetof(xfs_sb_t, sb_rextents),   0 },
  73    { offsetof(xfs_sb_t, sb_uuid),       1 },
  74    { offsetof(xfs_sb_t, sb_logstart),   0 },
  75    { offsetof(xfs_sb_t, sb_rootino),    0 },
  76    { offsetof(xfs_sb_t, sb_rbmino),     0 },
  77    { offsetof(xfs_sb_t, sb_rsumino),    0 },
  78    { offsetof(xfs_sb_t, sb_rextsize),   0 },
  79    { offsetof(xfs_sb_t, sb_agblocks),   0 },
  80    { offsetof(xfs_sb_t, sb_agcount),    0 },
  81    { offsetof(xfs_sb_t, sb_rbmblocks),  0 },
  82    { offsetof(xfs_sb_t, sb_logblocks),  0 },
  83    { offsetof(xfs_sb_t, sb_versionnum), 0 },
  84    { offsetof(xfs_sb_t, sb_sectsize),   0 },
  85    { offsetof(xfs_sb_t, sb_inodesize),  0 },
  86    { offsetof(xfs_sb_t, sb_inopblock),  0 },
  87    { offsetof(xfs_sb_t, sb_fname[0]),   1 },
  88    { offsetof(xfs_sb_t, sb_blocklog),   0 },
  89    { offsetof(xfs_sb_t, sb_sectlog),    0 },
  90    { offsetof(xfs_sb_t, sb_inodelog),   0 },
  91    { offsetof(xfs_sb_t, sb_inopblog),   0 },
  92    { offsetof(xfs_sb_t, sb_agblklog),   0 },
  93    { offsetof(xfs_sb_t, sb_rextslog),   0 },
  94    { offsetof(xfs_sb_t, sb_inprogress), 0 },
  95    { offsetof(xfs_sb_t, sb_imax_pct),   0 },
  96    { offsetof(xfs_sb_t, sb_icount),     0 },
  97    { offsetof(xfs_sb_t, sb_ifree),      0 },
  98    { offsetof(xfs_sb_t, sb_fdblocks),   0 },
  99    { offsetof(xfs_sb_t, sb_frextents),  0 },
 100    { offsetof(xfs_sb_t, sb_uquotino),   0 },
 101    { offsetof(xfs_sb_t, sb_gquotino),   0 },
 102    { offsetof(xfs_sb_t, sb_qflags),     0 },
 103    { offsetof(xfs_sb_t, sb_flags),      0 },
 104    { offsetof(xfs_sb_t, sb_shared_vn),  0 },
 105    { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
 106    { offsetof(xfs_sb_t, sb_unit),       0 },
 107    { offsetof(xfs_sb_t, sb_width),      0 },
 108    { offsetof(xfs_sb_t, sb_dirblklog),  0 },
 109    { offsetof(xfs_sb_t, sb_logsectlog), 0 },
 110    { offsetof(xfs_sb_t, sb_logsectsize),0 },
 111    { offsetof(xfs_sb_t, sb_logsunit),   0 },
 112    { offsetof(xfs_sb_t, sb_features2),  0 },
 113    { offsetof(xfs_sb_t, sb_bad_features2), 0 },
 114    { offsetof(xfs_sb_t, sb_features_compat), 0 },
 115    { offsetof(xfs_sb_t, sb_features_ro_compat), 0 },
 116    { offsetof(xfs_sb_t, sb_features_incompat), 0 },
 117    { offsetof(xfs_sb_t, sb_features_log_incompat), 0 },
 118    { offsetof(xfs_sb_t, sb_crc),        0 },
 119    { offsetof(xfs_sb_t, sb_pad),        0 },
 120    { offsetof(xfs_sb_t, sb_pquotino),   0 },
 121    { offsetof(xfs_sb_t, sb_lsn),        0 },
 122    { sizeof(xfs_sb_t),                  0 }
 123};
 124
 125static DEFINE_MUTEX(xfs_uuid_table_mutex);
 126static int xfs_uuid_table_size;
 127static uuid_t *xfs_uuid_table;
 128
 129/*
 130 * See if the UUID is unique among mounted XFS filesystems.
 131 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
 132 */
 133STATIC int
 134xfs_uuid_mount(
 135        struct xfs_mount        *mp)
 136{
 137        uuid_t                  *uuid = &mp->m_sb.sb_uuid;
 138        int                     hole, i;
 139
 140        if (mp->m_flags & XFS_MOUNT_NOUUID)
 141                return 0;
 142
 143        if (uuid_is_nil(uuid)) {
 144                xfs_warn(mp, "Filesystem has nil UUID - can't mount");
 145                return XFS_ERROR(EINVAL);
 146        }
 147
 148        mutex_lock(&xfs_uuid_table_mutex);
 149        for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
 150                if (uuid_is_nil(&xfs_uuid_table[i])) {
 151                        hole = i;
 152                        continue;
 153                }
 154                if (uuid_equal(uuid, &xfs_uuid_table[i]))
 155                        goto out_duplicate;
 156        }
 157
 158        if (hole < 0) {
 159                xfs_uuid_table = kmem_realloc(xfs_uuid_table,
 160                        (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
 161                        xfs_uuid_table_size  * sizeof(*xfs_uuid_table),
 162                        KM_SLEEP);
 163                hole = xfs_uuid_table_size++;
 164        }
 165        xfs_uuid_table[hole] = *uuid;
 166        mutex_unlock(&xfs_uuid_table_mutex);
 167
 168        return 0;
 169
 170 out_duplicate:
 171        mutex_unlock(&xfs_uuid_table_mutex);
 172        xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
 173        return XFS_ERROR(EINVAL);
 174}
 175
 176STATIC void
 177xfs_uuid_unmount(
 178        struct xfs_mount        *mp)
 179{
 180        uuid_t                  *uuid = &mp->m_sb.sb_uuid;
 181        int                     i;
 182
 183        if (mp->m_flags & XFS_MOUNT_NOUUID)
 184                return;
 185
 186        mutex_lock(&xfs_uuid_table_mutex);
 187        for (i = 0; i < xfs_uuid_table_size; i++) {
 188                if (uuid_is_nil(&xfs_uuid_table[i]))
 189                        continue;
 190                if (!uuid_equal(uuid, &xfs_uuid_table[i]))
 191                        continue;
 192                memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
 193                break;
 194        }
 195        ASSERT(i < xfs_uuid_table_size);
 196        mutex_unlock(&xfs_uuid_table_mutex);
 197}
 198
 199
 200/*
 201 * Reference counting access wrappers to the perag structures.
 202 * Because we never free per-ag structures, the only thing we
 203 * have to protect against changes is the tree structure itself.
 204 */
 205struct xfs_perag *
 206xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
 207{
 208        struct xfs_perag        *pag;
 209        int                     ref = 0;
 210
 211        rcu_read_lock();
 212        pag = radix_tree_lookup(&mp->m_perag_tree, agno);
 213        if (pag) {
 214                ASSERT(atomic_read(&pag->pag_ref) >= 0);
 215                ref = atomic_inc_return(&pag->pag_ref);
 216        }
 217        rcu_read_unlock();
 218        trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
 219        return pag;
 220}
 221
 222/*
 223 * search from @first to find the next perag with the given tag set.
 224 */
 225struct xfs_perag *
 226xfs_perag_get_tag(
 227        struct xfs_mount        *mp,
 228        xfs_agnumber_t          first,
 229        int                     tag)
 230{
 231        struct xfs_perag        *pag;
 232        int                     found;
 233        int                     ref;
 234
 235        rcu_read_lock();
 236        found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
 237                                        (void **)&pag, first, 1, tag);
 238        if (found <= 0) {
 239                rcu_read_unlock();
 240                return NULL;
 241        }
 242        ref = atomic_inc_return(&pag->pag_ref);
 243        rcu_read_unlock();
 244        trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
 245        return pag;
 246}
 247
 248void
 249xfs_perag_put(struct xfs_perag *pag)
 250{
 251        int     ref;
 252
 253        ASSERT(atomic_read(&pag->pag_ref) > 0);
 254        ref = atomic_dec_return(&pag->pag_ref);
 255        trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
 256}
 257
 258STATIC void
 259__xfs_free_perag(
 260        struct rcu_head *head)
 261{
 262        struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
 263
 264        ASSERT(atomic_read(&pag->pag_ref) == 0);
 265        kmem_free(pag);
 266}
 267
 268/*
 269 * Free up the per-ag resources associated with the mount structure.
 270 */
 271STATIC void
 272xfs_free_perag(
 273        xfs_mount_t     *mp)
 274{
 275        xfs_agnumber_t  agno;
 276        struct xfs_perag *pag;
 277
 278        for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
 279                spin_lock(&mp->m_perag_lock);
 280                pag = radix_tree_delete(&mp->m_perag_tree, agno);
 281                spin_unlock(&mp->m_perag_lock);
 282                ASSERT(pag);
 283                ASSERT(atomic_read(&pag->pag_ref) == 0);
 284                call_rcu(&pag->rcu_head, __xfs_free_perag);
 285        }
 286}
 287
 288/*
 289 * Check size of device based on the (data/realtime) block count.
 290 * Note: this check is used by the growfs code as well as mount.
 291 */
 292int
 293xfs_sb_validate_fsb_count(
 294        xfs_sb_t        *sbp,
 295        __uint64_t      nblocks)
 296{
 297        ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
 298        ASSERT(sbp->sb_blocklog >= BBSHIFT);
 299
 300#if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
 301        if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
 302                return EFBIG;
 303#else                  /* Limited by UINT_MAX of sectors */
 304        if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
 305                return EFBIG;
 306#endif
 307        return 0;
 308}
 309
 310/*
 311 * Check the validity of the SB found.
 312 */
 313STATIC int
 314xfs_mount_validate_sb(
 315        xfs_mount_t     *mp,
 316        xfs_sb_t        *sbp,
 317        bool            check_inprogress,
 318        bool            check_version)
 319{
 320
 321        /*
 322         * If the log device and data device have the
 323         * same device number, the log is internal.
 324         * Consequently, the sb_logstart should be non-zero.  If
 325         * we have a zero sb_logstart in this case, we may be trying to mount
 326         * a volume filesystem in a non-volume manner.
 327         */
 328        if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 329                xfs_warn(mp, "bad magic number");
 330                return XFS_ERROR(EWRONGFS);
 331        }
 332
 333
 334        if (!xfs_sb_good_version(sbp)) {
 335                xfs_warn(mp, "bad version");
 336                return XFS_ERROR(EWRONGFS);
 337        }
 338
 339        if ((sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) &&
 340                        (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
 341                                XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))) {
 342                xfs_notice(mp,
 343"Super block has XFS_OQUOTA bits along with XFS_PQUOTA and/or XFS_GQUOTA bits.\n");
 344                return XFS_ERROR(EFSCORRUPTED);
 345        }
 346
 347        /*
 348         * Version 5 superblock feature mask validation. Reject combinations the
 349         * kernel cannot support up front before checking anything else. For
 350         * write validation, we don't need to check feature masks.
 351         */
 352        if (check_version && XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5) {
 353                xfs_alert(mp,
 354"Version 5 superblock detected. This kernel has EXPERIMENTAL support enabled!\n"
 355"Use of these features in this kernel is at your own risk!");
 356
 357                if (xfs_sb_has_compat_feature(sbp,
 358                                        XFS_SB_FEAT_COMPAT_UNKNOWN)) {
 359                        xfs_warn(mp,
 360"Superblock has unknown compatible features (0x%x) enabled.\n"
 361"Using a more recent kernel is recommended.",
 362                                (sbp->sb_features_compat &
 363                                                XFS_SB_FEAT_COMPAT_UNKNOWN));
 364                }
 365
 366                if (xfs_sb_has_ro_compat_feature(sbp,
 367                                        XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
 368                        xfs_alert(mp,
 369"Superblock has unknown read-only compatible features (0x%x) enabled.",
 370                                (sbp->sb_features_ro_compat &
 371                                                XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
 372                        if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 373                                xfs_warn(mp,
 374"Attempted to mount read-only compatible filesystem read-write.\n"
 375"Filesystem can only be safely mounted read only.");
 376                                return XFS_ERROR(EINVAL);
 377                        }
 378                }
 379                if (xfs_sb_has_incompat_feature(sbp,
 380                                        XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
 381                        xfs_warn(mp,
 382"Superblock has unknown incompatible features (0x%x) enabled.\n"
 383"Filesystem can not be safely mounted by this kernel.",
 384                                (sbp->sb_features_incompat &
 385                                                XFS_SB_FEAT_INCOMPAT_UNKNOWN));
 386                        return XFS_ERROR(EINVAL);
 387                }
 388        }
 389
 390        if (unlikely(
 391            sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
 392                xfs_warn(mp,
 393                "filesystem is marked as having an external log; "
 394                "specify logdev on the mount command line.");
 395                return XFS_ERROR(EINVAL);
 396        }
 397
 398        if (unlikely(
 399            sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
 400                xfs_warn(mp,
 401                "filesystem is marked as having an internal log; "
 402                "do not specify logdev on the mount command line.");
 403                return XFS_ERROR(EINVAL);
 404        }
 405
 406        /*
 407         * More sanity checking.  Most of these were stolen directly from
 408         * xfs_repair.
 409         */
 410        if (unlikely(
 411            sbp->sb_agcount <= 0                                        ||
 412            sbp->sb_sectsize < XFS_MIN_SECTORSIZE                       ||
 413            sbp->sb_sectsize > XFS_MAX_SECTORSIZE                       ||
 414            sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG                    ||
 415            sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG                    ||
 416            sbp->sb_sectsize != (1 << sbp->sb_sectlog)                  ||
 417            sbp->sb_blocksize < XFS_MIN_BLOCKSIZE                       ||
 418            sbp->sb_blocksize > XFS_MAX_BLOCKSIZE                       ||
 419            sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG                    ||
 420            sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG                    ||
 421            sbp->sb_blocksize != (1 << sbp->sb_blocklog)                ||
 422            sbp->sb_inodesize < XFS_DINODE_MIN_SIZE                     ||
 423            sbp->sb_inodesize > XFS_DINODE_MAX_SIZE                     ||
 424            sbp->sb_inodelog < XFS_DINODE_MIN_LOG                       ||
 425            sbp->sb_inodelog > XFS_DINODE_MAX_LOG                       ||
 426            sbp->sb_inodesize != (1 << sbp->sb_inodelog)                ||
 427            (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
 428            (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
 429            (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
 430            (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)    ||
 431            sbp->sb_dblocks == 0                                        ||
 432            sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)                      ||
 433            sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
 434                XFS_CORRUPTION_ERROR("SB sanity check failed",
 435                                XFS_ERRLEVEL_LOW, mp, sbp);
 436                return XFS_ERROR(EFSCORRUPTED);
 437        }
 438
 439        /*
 440         * Until this is fixed only page-sized or smaller data blocks work.
 441         */
 442        if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
 443                xfs_warn(mp,
 444                "File system with blocksize %d bytes. "
 445                "Only pagesize (%ld) or less will currently work.",
 446                                sbp->sb_blocksize, PAGE_SIZE);
 447                return XFS_ERROR(ENOSYS);
 448        }
 449
 450        /*
 451         * Currently only very few inode sizes are supported.
 452         */
 453        switch (sbp->sb_inodesize) {
 454        case 256:
 455        case 512:
 456        case 1024:
 457        case 2048:
 458                break;
 459        default:
 460                xfs_warn(mp, "inode size of %d bytes not supported",
 461                                sbp->sb_inodesize);
 462                return XFS_ERROR(ENOSYS);
 463        }
 464
 465        if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
 466            xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
 467                xfs_warn(mp,
 468                "file system too large to be mounted on this system.");
 469                return XFS_ERROR(EFBIG);
 470        }
 471
 472        if (check_inprogress && sbp->sb_inprogress) {
 473                xfs_warn(mp, "Offline file system operation in progress!");
 474                return XFS_ERROR(EFSCORRUPTED);
 475        }
 476
 477        /*
 478         * Version 1 directory format has never worked on Linux.
 479         */
 480        if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
 481                xfs_warn(mp, "file system using version 1 directory format");
 482                return XFS_ERROR(ENOSYS);
 483        }
 484
 485        return 0;
 486}
 487
 488int
 489xfs_initialize_perag(
 490        xfs_mount_t     *mp,
 491        xfs_agnumber_t  agcount,
 492        xfs_agnumber_t  *maxagi)
 493{
 494        xfs_agnumber_t  index;
 495        xfs_agnumber_t  first_initialised = 0;
 496        xfs_perag_t     *pag;
 497        xfs_agino_t     agino;
 498        xfs_ino_t       ino;
 499        xfs_sb_t        *sbp = &mp->m_sb;
 500        int             error = -ENOMEM;
 501
 502        /*
 503         * Walk the current per-ag tree so we don't try to initialise AGs
 504         * that already exist (growfs case). Allocate and insert all the
 505         * AGs we don't find ready for initialisation.
 506         */
 507        for (index = 0; index < agcount; index++) {
 508                pag = xfs_perag_get(mp, index);
 509                if (pag) {
 510                        xfs_perag_put(pag);
 511                        continue;
 512                }
 513                if (!first_initialised)
 514                        first_initialised = index;
 515
 516                pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
 517                if (!pag)
 518                        goto out_unwind;
 519                pag->pag_agno = index;
 520                pag->pag_mount = mp;
 521                spin_lock_init(&pag->pag_ici_lock);
 522                mutex_init(&pag->pag_ici_reclaim_lock);
 523                INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
 524                spin_lock_init(&pag->pag_buf_lock);
 525                pag->pag_buf_tree = RB_ROOT;
 526
 527                if (radix_tree_preload(GFP_NOFS))
 528                        goto out_unwind;
 529
 530                spin_lock(&mp->m_perag_lock);
 531                if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
 532                        BUG();
 533                        spin_unlock(&mp->m_perag_lock);
 534                        radix_tree_preload_end();
 535                        error = -EEXIST;
 536                        goto out_unwind;
 537                }
 538                spin_unlock(&mp->m_perag_lock);
 539                radix_tree_preload_end();
 540        }
 541
 542        /*
 543         * If we mount with the inode64 option, or no inode overflows
 544         * the legacy 32-bit address space clear the inode32 option.
 545         */
 546        agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
 547        ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 548
 549        if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
 550                mp->m_flags |= XFS_MOUNT_32BITINODES;
 551        else
 552                mp->m_flags &= ~XFS_MOUNT_32BITINODES;
 553
 554        if (mp->m_flags & XFS_MOUNT_32BITINODES)
 555                index = xfs_set_inode32(mp);
 556        else
 557                index = xfs_set_inode64(mp);
 558
 559        if (maxagi)
 560                *maxagi = index;
 561        return 0;
 562
 563out_unwind:
 564        kmem_free(pag);
 565        for (; index > first_initialised; index--) {
 566                pag = radix_tree_delete(&mp->m_perag_tree, index);
 567                kmem_free(pag);
 568        }
 569        return error;
 570}
 571
 572static void
 573xfs_sb_quota_from_disk(struct xfs_sb *sbp)
 574{
 575        if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
 576                sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 577                                        XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
 578        if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
 579                sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
 580                                        XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
 581        sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
 582}
 583
 584void
 585xfs_sb_from_disk(
 586        struct xfs_sb   *to,
 587        xfs_dsb_t       *from)
 588{
 589        to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
 590        to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
 591        to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
 592        to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
 593        to->sb_rextents = be64_to_cpu(from->sb_rextents);
 594        memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
 595        to->sb_logstart = be64_to_cpu(from->sb_logstart);
 596        to->sb_rootino = be64_to_cpu(from->sb_rootino);
 597        to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
 598        to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
 599        to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
 600        to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
 601        to->sb_agcount = be32_to_cpu(from->sb_agcount);
 602        to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
 603        to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
 604        to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
 605        to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
 606        to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
 607        to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
 608        memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
 609        to->sb_blocklog = from->sb_blocklog;
 610        to->sb_sectlog = from->sb_sectlog;
 611        to->sb_inodelog = from->sb_inodelog;
 612        to->sb_inopblog = from->sb_inopblog;
 613        to->sb_agblklog = from->sb_agblklog;
 614        to->sb_rextslog = from->sb_rextslog;
 615        to->sb_inprogress = from->sb_inprogress;
 616        to->sb_imax_pct = from->sb_imax_pct;
 617        to->sb_icount = be64_to_cpu(from->sb_icount);
 618        to->sb_ifree = be64_to_cpu(from->sb_ifree);
 619        to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
 620        to->sb_frextents = be64_to_cpu(from->sb_frextents);
 621        to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
 622        to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
 623        to->sb_qflags = be16_to_cpu(from->sb_qflags);
 624        to->sb_flags = from->sb_flags;
 625        to->sb_shared_vn = from->sb_shared_vn;
 626        to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
 627        to->sb_unit = be32_to_cpu(from->sb_unit);
 628        to->sb_width = be32_to_cpu(from->sb_width);
 629        to->sb_dirblklog = from->sb_dirblklog;
 630        to->sb_logsectlog = from->sb_logsectlog;
 631        to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
 632        to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
 633        to->sb_features2 = be32_to_cpu(from->sb_features2);
 634        to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
 635        to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
 636        to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
 637        to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
 638        to->sb_features_log_incompat =
 639                                be32_to_cpu(from->sb_features_log_incompat);
 640        to->sb_pad = 0;
 641        to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
 642        to->sb_lsn = be64_to_cpu(from->sb_lsn);
 643}
 644
 645static inline void
 646xfs_sb_quota_to_disk(
 647        xfs_dsb_t       *to,
 648        xfs_sb_t        *from,
 649        __int64_t       *fields)
 650{
 651        __uint16_t      qflags = from->sb_qflags;
 652
 653        if (*fields & XFS_SB_QFLAGS) {
 654                /*
 655                 * The in-core version of sb_qflags do not have
 656                 * XFS_OQUOTA_* flags, whereas the on-disk version
 657                 * does.  So, convert incore XFS_{PG}QUOTA_* flags
 658                 * to on-disk XFS_OQUOTA_* flags.
 659                 */
 660                qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
 661                                XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
 662
 663                if (from->sb_qflags &
 664                                (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
 665                        qflags |= XFS_OQUOTA_ENFD;
 666                if (from->sb_qflags &
 667                                (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
 668                        qflags |= XFS_OQUOTA_CHKD;
 669                to->sb_qflags = cpu_to_be16(qflags);
 670                *fields &= ~XFS_SB_QFLAGS;
 671        }
 672}
 673
 674/*
 675 * Copy in core superblock to ondisk one.
 676 *
 677 * The fields argument is mask of superblock fields to copy.
 678 */
 679void
 680xfs_sb_to_disk(
 681        xfs_dsb_t       *to,
 682        xfs_sb_t        *from,
 683        __int64_t       fields)
 684{
 685        xfs_caddr_t     to_ptr = (xfs_caddr_t)to;
 686        xfs_caddr_t     from_ptr = (xfs_caddr_t)from;
 687        xfs_sb_field_t  f;
 688        int             first;
 689        int             size;
 690
 691        ASSERT(fields);
 692        if (!fields)
 693                return;
 694
 695        xfs_sb_quota_to_disk(to, from, &fields);
 696        while (fields) {
 697                f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 698                first = xfs_sb_info[f].offset;
 699                size = xfs_sb_info[f + 1].offset - first;
 700
 701                ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
 702
 703                if (size == 1 || xfs_sb_info[f].type == 1) {
 704                        memcpy(to_ptr + first, from_ptr + first, size);
 705                } else {
 706                        switch (size) {
 707                        case 2:
 708                                *(__be16 *)(to_ptr + first) =
 709                                        cpu_to_be16(*(__u16 *)(from_ptr + first));
 710                                break;
 711                        case 4:
 712                                *(__be32 *)(to_ptr + first) =
 713                                        cpu_to_be32(*(__u32 *)(from_ptr + first));
 714                                break;
 715                        case 8:
 716                                *(__be64 *)(to_ptr + first) =
 717                                        cpu_to_be64(*(__u64 *)(from_ptr + first));
 718                                break;
 719                        default:
 720                                ASSERT(0);
 721                        }
 722                }
 723
 724                fields &= ~(1LL << f);
 725        }
 726}
 727
 728static int
 729xfs_sb_verify(
 730        struct xfs_buf  *bp,
 731        bool            check_version)
 732{
 733        struct xfs_mount *mp = bp->b_target->bt_mount;
 734        struct xfs_sb   sb;
 735
 736        xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp));
 737
 738        /*
 739         * Only check the in progress field for the primary superblock as
 740         * mkfs.xfs doesn't clear it from secondary superblocks.
 741         */
 742        return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR,
 743                                     check_version);
 744}
 745
 746/*
 747 * If the superblock has the CRC feature bit set or the CRC field is non-null,
 748 * check that the CRC is valid.  We check the CRC field is non-null because a
 749 * single bit error could clear the feature bit and unused parts of the
 750 * superblock are supposed to be zero. Hence a non-null crc field indicates that
 751 * we've potentially lost a feature bit and we should check it anyway.
 752 */
 753static void
 754xfs_sb_read_verify(
 755        struct xfs_buf  *bp)
 756{
 757        struct xfs_mount *mp = bp->b_target->bt_mount;
 758        struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 759        int             error;
 760
 761        /*
 762         * open code the version check to avoid needing to convert the entire
 763         * superblock from disk order just to check the version number
 764         */
 765        if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
 766            (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
 767                                                XFS_SB_VERSION_5) ||
 768             dsb->sb_crc != 0)) {
 769
 770                if (!xfs_verify_cksum(bp->b_addr, be16_to_cpu(dsb->sb_sectsize),
 771                                      offsetof(struct xfs_sb, sb_crc))) {
 772                        error = EFSCORRUPTED;
 773                        goto out_error;
 774                }
 775        }
 776        error = xfs_sb_verify(bp, true);
 777
 778out_error:
 779        if (error) {
 780                XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 781                xfs_buf_ioerror(bp, error);
 782        }
 783}
 784
 785/*
 786 * We may be probed for a filesystem match, so we may not want to emit
 787 * messages when the superblock buffer is not actually an XFS superblock.
 788 * If we find an XFS superblock, the run a normal, noisy mount because we are
 789 * really going to mount it and want to know about errors.
 790 */
 791static void
 792xfs_sb_quiet_read_verify(
 793        struct xfs_buf  *bp)
 794{
 795        struct xfs_dsb  *dsb = XFS_BUF_TO_SBP(bp);
 796
 797
 798        if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
 799                /* XFS filesystem, verify noisily! */
 800                xfs_sb_read_verify(bp);
 801                return;
 802        }
 803        /* quietly fail */
 804        xfs_buf_ioerror(bp, EWRONGFS);
 805}
 806
 807static void
 808xfs_sb_write_verify(
 809        struct xfs_buf          *bp)
 810{
 811        struct xfs_mount        *mp = bp->b_target->bt_mount;
 812        struct xfs_buf_log_item *bip = bp->b_fspriv;
 813        int                     error;
 814
 815        error = xfs_sb_verify(bp, false);
 816        if (error) {
 817                XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 818                xfs_buf_ioerror(bp, error);
 819                return;
 820        }
 821
 822        if (!xfs_sb_version_hascrc(&mp->m_sb))
 823                return;
 824
 825        if (bip)
 826                XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 827
 828        xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
 829                         offsetof(struct xfs_sb, sb_crc));
 830}
 831
 832const struct xfs_buf_ops xfs_sb_buf_ops = {
 833        .verify_read = xfs_sb_read_verify,
 834        .verify_write = xfs_sb_write_verify,
 835};
 836
 837static const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
 838        .verify_read = xfs_sb_quiet_read_verify,
 839        .verify_write = xfs_sb_write_verify,
 840};
 841
 842/*
 843 * xfs_readsb
 844 *
 845 * Does the initial read of the superblock.
 846 */
 847int
 848xfs_readsb(xfs_mount_t *mp, int flags)
 849{
 850        unsigned int    sector_size;
 851        struct xfs_buf  *bp;
 852        struct xfs_sb   *sbp = &mp->m_sb;
 853        int             error;
 854        int             loud = !(flags & XFS_MFSI_QUIET);
 855
 856        ASSERT(mp->m_sb_bp == NULL);
 857        ASSERT(mp->m_ddev_targp != NULL);
 858
 859        /*
 860         * Allocate a (locked) buffer to hold the superblock.
 861         * This will be kept around at all times to optimize
 862         * access to the superblock.
 863         */
 864        sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
 865
 866reread:
 867        bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
 868                                   BTOBB(sector_size), 0,
 869                                   loud ? &xfs_sb_buf_ops
 870                                        : &xfs_sb_quiet_buf_ops);
 871        if (!bp) {
 872                if (loud)
 873                        xfs_warn(mp, "SB buffer read failed");
 874                return EIO;
 875        }
 876        if (bp->b_error) {
 877                error = bp->b_error;
 878                if (loud)
 879                        xfs_warn(mp, "SB validate failed with error %d.", error);
 880                goto release_buf;
 881        }
 882
 883        /*
 884         * Initialize the mount structure from the superblock.
 885         */
 886        xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
 887
 888        xfs_sb_quota_from_disk(&mp->m_sb);
 889        /*
 890         * We must be able to do sector-sized and sector-aligned IO.
 891         */
 892        if (sector_size > sbp->sb_sectsize) {
 893                if (loud)
 894                        xfs_warn(mp, "device supports %u byte sectors (not %u)",
 895                                sector_size, sbp->sb_sectsize);
 896                error = ENOSYS;
 897                goto release_buf;
 898        }
 899
 900        /*
 901         * If device sector size is smaller than the superblock size,
 902         * re-read the superblock so the buffer is correctly sized.
 903         */
 904        if (sector_size < sbp->sb_sectsize) {
 905                xfs_buf_relse(bp);
 906                sector_size = sbp->sb_sectsize;
 907                goto reread;
 908        }
 909
 910        /* Initialize per-cpu counters */
 911        xfs_icsb_reinit_counters(mp);
 912
 913        /* no need to be quiet anymore, so reset the buf ops */
 914        bp->b_ops = &xfs_sb_buf_ops;
 915
 916        mp->m_sb_bp = bp;
 917        xfs_buf_unlock(bp);
 918        return 0;
 919
 920release_buf:
 921        xfs_buf_relse(bp);
 922        return error;
 923}
 924
 925
 926/*
 927 * xfs_mount_common
 928 *
 929 * Mount initialization code establishing various mount
 930 * fields from the superblock associated with the given
 931 * mount structure
 932 */
 933STATIC void
 934xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
 935{
 936        mp->m_agfrotor = mp->m_agirotor = 0;
 937        spin_lock_init(&mp->m_agirotor_lock);
 938        mp->m_maxagi = mp->m_sb.sb_agcount;
 939        mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
 940        mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
 941        mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
 942        mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
 943        mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
 944        mp->m_blockmask = sbp->sb_blocksize - 1;
 945        mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
 946        mp->m_blockwmask = mp->m_blockwsize - 1;
 947
 948        mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
 949        mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
 950        mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
 951        mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
 952
 953        mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
 954        mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
 955        mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
 956        mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
 957
 958        mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
 959        mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
 960        mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
 961        mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
 962
 963        mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
 964        mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 965                                        sbp->sb_inopblock);
 966        mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 967}
 968
 969/*
 970 * xfs_initialize_perag_data
 971 *
 972 * Read in each per-ag structure so we can count up the number of
 973 * allocated inodes, free inodes and used filesystem blocks as this
 974 * information is no longer persistent in the superblock. Once we have
 975 * this information, write it into the in-core superblock structure.
 976 */
 977STATIC int
 978xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
 979{
 980        xfs_agnumber_t  index;
 981        xfs_perag_t     *pag;
 982        xfs_sb_t        *sbp = &mp->m_sb;
 983        uint64_t        ifree = 0;
 984        uint64_t        ialloc = 0;
 985        uint64_t        bfree = 0;
 986        uint64_t        bfreelst = 0;
 987        uint64_t        btree = 0;
 988        int             error;
 989
 990        for (index = 0; index < agcount; index++) {
 991                /*
 992                 * read the agf, then the agi. This gets us
 993                 * all the information we need and populates the
 994                 * per-ag structures for us.
 995                 */
 996                error = xfs_alloc_pagf_init(mp, NULL, index, 0);
 997                if (error)
 998                        return error;
 999
1000                error = xfs_ialloc_pagi_init(mp, NULL, index);
1001                if (error)
1002                        return error;
1003                pag = xfs_perag_get(mp, index);
1004                ifree += pag->pagi_freecount;
1005                ialloc += pag->pagi_count;
1006                bfree += pag->pagf_freeblks;
1007                bfreelst += pag->pagf_flcount;
1008                btree += pag->pagf_btreeblks;
1009                xfs_perag_put(pag);
1010        }
1011        /*
1012         * Overwrite incore superblock counters with just-read data
1013         */
1014        spin_lock(&mp->m_sb_lock);
1015        sbp->sb_ifree = ifree;
1016        sbp->sb_icount = ialloc;
1017        sbp->sb_fdblocks = bfree + bfreelst + btree;
1018        spin_unlock(&mp->m_sb_lock);
1019
1020        /* Fixup the per-cpu counters as well. */
1021        xfs_icsb_reinit_counters(mp);
1022
1023        return 0;
1024}
1025
1026/*
1027 * Update alignment values based on mount options and sb values
1028 */
1029STATIC int
1030xfs_update_alignment(xfs_mount_t *mp)
1031{
1032        xfs_sb_t        *sbp = &(mp->m_sb);
1033
1034        if (mp->m_dalign) {
1035                /*
1036                 * If stripe unit and stripe width are not multiples
1037                 * of the fs blocksize turn off alignment.
1038                 */
1039                if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
1040                    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
1041                        xfs_warn(mp,
1042                "alignment check failed: sunit/swidth vs. blocksize(%d)",
1043                                sbp->sb_blocksize);
1044                        return XFS_ERROR(EINVAL);
1045                } else {
1046                        /*
1047                         * Convert the stripe unit and width to FSBs.
1048                         */
1049                        mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
1050                        if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
1051                                xfs_warn(mp,
1052                        "alignment check failed: sunit/swidth vs. agsize(%d)",
1053                                         sbp->sb_agblocks);
1054                                return XFS_ERROR(EINVAL);
1055                        } else if (mp->m_dalign) {
1056                                mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
1057                        } else {
1058                                xfs_warn(mp,
1059                        "alignment check failed: sunit(%d) less than bsize(%d)",
1060                                         mp->m_dalign, sbp->sb_blocksize);
1061                                return XFS_ERROR(EINVAL);
1062                        }
1063                }
1064
1065                /*
1066                 * Update superblock with new values
1067                 * and log changes
1068                 */
1069                if (xfs_sb_version_hasdalign(sbp)) {
1070                        if (sbp->sb_unit != mp->m_dalign) {
1071                                sbp->sb_unit = mp->m_dalign;
1072                                mp->m_update_flags |= XFS_SB_UNIT;
1073                        }
1074                        if (sbp->sb_width != mp->m_swidth) {
1075                                sbp->sb_width = mp->m_swidth;
1076                                mp->m_update_flags |= XFS_SB_WIDTH;
1077                        }
1078                } else {
1079                        xfs_warn(mp,
1080        "cannot change alignment: superblock does not support data alignment");
1081                        return XFS_ERROR(EINVAL);
1082                }
1083        } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
1084                    xfs_sb_version_hasdalign(&mp->m_sb)) {
1085                        mp->m_dalign = sbp->sb_unit;
1086                        mp->m_swidth = sbp->sb_width;
1087        }
1088
1089        return 0;
1090}
1091
1092/*
1093 * Set the maximum inode count for this filesystem
1094 */
1095STATIC void
1096xfs_set_maxicount(xfs_mount_t *mp)
1097{
1098        xfs_sb_t        *sbp = &(mp->m_sb);
1099        __uint64_t      icount;
1100
1101        if (sbp->sb_imax_pct) {
1102                /*
1103                 * Make sure the maximum inode count is a multiple
1104                 * of the units we allocate inodes in.
1105                 */
1106                icount = sbp->sb_dblocks * sbp->sb_imax_pct;
1107                do_div(icount, 100);
1108                do_div(icount, mp->m_ialloc_blks);
1109                mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
1110                                   sbp->sb_inopblog;
1111        } else {
1112                mp->m_maxicount = 0;
1113        }
1114}
1115
1116/*
1117 * Set the default minimum read and write sizes unless
1118 * already specified in a mount option.
1119 * We use smaller I/O sizes when the file system
1120 * is being used for NFS service (wsync mount option).
1121 */
1122STATIC void
1123xfs_set_rw_sizes(xfs_mount_t *mp)
1124{
1125        xfs_sb_t        *sbp = &(mp->m_sb);
1126        int             readio_log, writeio_log;
1127
1128        if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
1129                if (mp->m_flags & XFS_MOUNT_WSYNC) {
1130                        readio_log = XFS_WSYNC_READIO_LOG;
1131                        writeio_log = XFS_WSYNC_WRITEIO_LOG;
1132                } else {
1133                        readio_log = XFS_READIO_LOG_LARGE;
1134                        writeio_log = XFS_WRITEIO_LOG_LARGE;
1135                }
1136        } else {
1137                readio_log = mp->m_readio_log;
1138                writeio_log = mp->m_writeio_log;
1139        }
1140
1141        if (sbp->sb_blocklog > readio_log) {
1142                mp->m_readio_log = sbp->sb_blocklog;
1143        } else {
1144                mp->m_readio_log = readio_log;
1145        }
1146        mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
1147        if (sbp->sb_blocklog > writeio_log) {
1148                mp->m_writeio_log = sbp->sb_blocklog;
1149        } else {
1150                mp->m_writeio_log = writeio_log;
1151        }
1152        mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
1153}
1154
1155/*
1156 * precalculate the low space thresholds for dynamic speculative preallocation.
1157 */
1158void
1159xfs_set_low_space_thresholds(
1160        struct xfs_mount        *mp)
1161{
1162        int i;
1163
1164        for (i = 0; i < XFS_LOWSP_MAX; i++) {
1165                __uint64_t space = mp->m_sb.sb_dblocks;
1166
1167                do_div(space, 100);
1168                mp->m_low_space[i] = space * (i + 1);
1169        }
1170}
1171
1172
1173/*
1174 * Set whether we're using inode alignment.
1175 */
1176STATIC void
1177xfs_set_inoalignment(xfs_mount_t *mp)
1178{
1179        if (xfs_sb_version_hasalign(&mp->m_sb) &&
1180            mp->m_sb.sb_inoalignmt >=
1181            XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1182                mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1183        else
1184                mp->m_inoalign_mask = 0;
1185        /*
1186         * If we are using stripe alignment, check whether
1187         * the stripe unit is a multiple of the inode alignment
1188         */
1189        if (mp->m_dalign && mp->m_inoalign_mask &&
1190            !(mp->m_dalign & mp->m_inoalign_mask))
1191                mp->m_sinoalign = mp->m_dalign;
1192        else
1193                mp->m_sinoalign = 0;
1194}
1195
1196/*
1197 * Check that the data (and log if separate) are an ok size.
1198 */
1199STATIC int
1200xfs_check_sizes(xfs_mount_t *mp)
1201{
1202        xfs_buf_t       *bp;
1203        xfs_daddr_t     d;
1204
1205        d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1206        if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1207                xfs_warn(mp, "filesystem size mismatch detected");
1208                return XFS_ERROR(EFBIG);
1209        }
1210        bp = xfs_buf_read_uncached(mp->m_ddev_targp,
1211                                        d - XFS_FSS_TO_BB(mp, 1),
1212                                        XFS_FSS_TO_BB(mp, 1), 0, NULL);
1213        if (!bp) {
1214                xfs_warn(mp, "last sector read failed");
1215                return EIO;
1216        }
1217        xfs_buf_relse(bp);
1218
1219        if (mp->m_logdev_targp != mp->m_ddev_targp) {
1220                d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1221                if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1222                        xfs_warn(mp, "log size mismatch detected");
1223                        return XFS_ERROR(EFBIG);
1224                }
1225                bp = xfs_buf_read_uncached(mp->m_logdev_targp,
1226                                        d - XFS_FSB_TO_BB(mp, 1),
1227                                        XFS_FSB_TO_BB(mp, 1), 0, NULL);
1228                if (!bp) {
1229                        xfs_warn(mp, "log device read failed");
1230                        return EIO;
1231                }
1232                xfs_buf_relse(bp);
1233        }
1234        return 0;
1235}
1236
1237/*
1238 * Clear the quotaflags in memory and in the superblock.
1239 */
1240int
1241xfs_mount_reset_sbqflags(
1242        struct xfs_mount        *mp)
1243{
1244        int                     error;
1245        struct xfs_trans        *tp;
1246
1247        mp->m_qflags = 0;
1248
1249        /*
1250         * It is OK to look at sb_qflags here in mount path,
1251         * without m_sb_lock.
1252         */
1253        if (mp->m_sb.sb_qflags == 0)
1254                return 0;
1255        spin_lock(&mp->m_sb_lock);
1256        mp->m_sb.sb_qflags = 0;
1257        spin_unlock(&mp->m_sb_lock);
1258
1259        /*
1260         * If the fs is readonly, let the incore superblock run
1261         * with quotas off but don't flush the update out to disk
1262         */
1263        if (mp->m_flags & XFS_MOUNT_RDONLY)
1264                return 0;
1265
1266        tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1267        error = xfs_trans_reserve(tp, 0, XFS_QM_SBCHANGE_LOG_RES(mp),
1268                                  0, 0, XFS_DEFAULT_LOG_COUNT);
1269        if (error) {
1270                xfs_trans_cancel(tp, 0);
1271                xfs_alert(mp, "%s: Superblock update failed!", __func__);
1272                return error;
1273        }
1274
1275        xfs_mod_sb(tp, XFS_SB_QFLAGS);
1276        return xfs_trans_commit(tp, 0);
1277}
1278
1279__uint64_t
1280xfs_default_resblks(xfs_mount_t *mp)
1281{
1282        __uint64_t resblks;
1283
1284        /*
1285         * We default to 5% or 8192 fsbs of space reserved, whichever is
1286         * smaller.  This is intended to cover concurrent allocation
1287         * transactions when we initially hit enospc. These each require a 4
1288         * block reservation. Hence by default we cover roughly 2000 concurrent
1289         * allocation reservations.
1290         */
1291        resblks = mp->m_sb.sb_dblocks;
1292        do_div(resblks, 20);
1293        resblks = min_t(__uint64_t, resblks, 8192);
1294        return resblks;
1295}
1296
1297/*
1298 * This function does the following on an initial mount of a file system:
1299 *      - reads the superblock from disk and init the mount struct
1300 *      - if we're a 32-bit kernel, do a size check on the superblock
1301 *              so we don't mount terabyte filesystems
1302 *      - init mount struct realtime fields
1303 *      - allocate inode hash table for fs
1304 *      - init directory manager
1305 *      - perform recovery and init the log manager
1306 */
1307int
1308xfs_mountfs(
1309        xfs_mount_t     *mp)
1310{
1311        xfs_sb_t        *sbp = &(mp->m_sb);
1312        xfs_inode_t     *rip;
1313        __uint64_t      resblks;
1314        uint            quotamount = 0;
1315        uint            quotaflags = 0;
1316        int             error = 0;
1317
1318        xfs_mount_common(mp, sbp);
1319
1320        /*
1321         * Check for a mismatched features2 values.  Older kernels
1322         * read & wrote into the wrong sb offset for sb_features2
1323         * on some platforms due to xfs_sb_t not being 64bit size aligned
1324         * when sb_features2 was added, which made older superblock
1325         * reading/writing routines swap it as a 64-bit value.
1326         *
1327         * For backwards compatibility, we make both slots equal.
1328         *
1329         * If we detect a mismatched field, we OR the set bits into the
1330         * existing features2 field in case it has already been modified; we
1331         * don't want to lose any features.  We then update the bad location
1332         * with the ORed value so that older kernels will see any features2
1333         * flags, and mark the two fields as needing updates once the
1334         * transaction subsystem is online.
1335         */
1336        if (xfs_sb_has_mismatched_features2(sbp)) {
1337                xfs_warn(mp, "correcting sb_features alignment problem");
1338                sbp->sb_features2 |= sbp->sb_bad_features2;
1339                sbp->sb_bad_features2 = sbp->sb_features2;
1340                mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1341
1342                /*
1343                 * Re-check for ATTR2 in case it was found in bad_features2
1344                 * slot.
1345                 */
1346                if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1347                   !(mp->m_flags & XFS_MOUNT_NOATTR2))
1348                        mp->m_flags |= XFS_MOUNT_ATTR2;
1349        }
1350
1351        if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1352           (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1353                xfs_sb_version_removeattr2(&mp->m_sb);
1354                mp->m_update_flags |= XFS_SB_FEATURES2;
1355
1356                /* update sb_versionnum for the clearing of the morebits */
1357                if (!sbp->sb_features2)
1358                        mp->m_update_flags |= XFS_SB_VERSIONNUM;
1359        }
1360
1361        /*
1362         * Check if sb_agblocks is aligned at stripe boundary
1363         * If sb_agblocks is NOT aligned turn off m_dalign since
1364         * allocator alignment is within an ag, therefore ag has
1365         * to be aligned at stripe boundary.
1366         */
1367        error = xfs_update_alignment(mp);
1368        if (error)
1369                goto out;
1370
1371        xfs_alloc_compute_maxlevels(mp);
1372        xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1373        xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1374        xfs_ialloc_compute_maxlevels(mp);
1375
1376        xfs_set_maxicount(mp);
1377
1378        error = xfs_uuid_mount(mp);
1379        if (error)
1380                goto out;
1381
1382        /*
1383         * Set the minimum read and write sizes
1384         */
1385        xfs_set_rw_sizes(mp);
1386
1387        /* set the low space thresholds for dynamic preallocation */
1388        xfs_set_low_space_thresholds(mp);
1389
1390        /*
1391         * Set the inode cluster size.
1392         * This may still be overridden by the file system
1393         * block size if it is larger than the chosen cluster size.
1394         */
1395        mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1396
1397        /*
1398         * Set inode alignment fields
1399         */
1400        xfs_set_inoalignment(mp);
1401
1402        /*
1403         * Check that the data (and log if separate) are an ok size.
1404         */
1405        error = xfs_check_sizes(mp);
1406        if (error)
1407                goto out_remove_uuid;
1408
1409        /*
1410         * Initialize realtime fields in the mount structure
1411         */
1412        error = xfs_rtmount_init(mp);
1413        if (error) {
1414                xfs_warn(mp, "RT mount failed");
1415                goto out_remove_uuid;
1416        }
1417
1418        /*
1419         *  Copies the low order bits of the timestamp and the randomly
1420         *  set "sequence" number out of a UUID.
1421         */
1422        uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1423
1424        mp->m_dmevmask = 0;     /* not persistent; set after each mount */
1425
1426        xfs_dir_mount(mp);
1427
1428        /*
1429         * Initialize the attribute manager's entries.
1430         */
1431        mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1432
1433        /*
1434         * Initialize the precomputed transaction reservations values.
1435         */
1436        xfs_trans_init(mp);
1437
1438        /*
1439         * Allocate and initialize the per-ag data.
1440         */
1441        spin_lock_init(&mp->m_perag_lock);
1442        INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1443        error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1444        if (error) {
1445                xfs_warn(mp, "Failed per-ag init: %d", error);
1446                goto out_remove_uuid;
1447        }
1448
1449        if (!sbp->sb_logblocks) {
1450                xfs_warn(mp, "no log defined");
1451                XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1452                error = XFS_ERROR(EFSCORRUPTED);
1453                goto out_free_perag;
1454        }
1455
1456        /*
1457         * log's mount-time initialization. Perform 1st part recovery if needed
1458         */
1459        error = xfs_log_mount(mp, mp->m_logdev_targp,
1460                              XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1461                              XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1462        if (error) {
1463                xfs_warn(mp, "log mount failed");
1464                goto out_fail_wait;
1465        }
1466
1467        /*
1468         * Now the log is mounted, we know if it was an unclean shutdown or
1469         * not. If it was, with the first phase of recovery has completed, we
1470         * have consistent AG blocks on disk. We have not recovered EFIs yet,
1471         * but they are recovered transactionally in the second recovery phase
1472         * later.
1473         *
1474         * Hence we can safely re-initialise incore superblock counters from
1475         * the per-ag data. These may not be correct if the filesystem was not
1476         * cleanly unmounted, so we need to wait for recovery to finish before
1477         * doing this.
1478         *
1479         * If the filesystem was cleanly unmounted, then we can trust the
1480         * values in the superblock to be correct and we don't need to do
1481         * anything here.
1482         *
1483         * If we are currently making the filesystem, the initialisation will
1484         * fail as the perag data is in an undefined state.
1485         */
1486        if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1487            !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1488             !mp->m_sb.sb_inprogress) {
1489                error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1490                if (error)
1491                        goto out_fail_wait;
1492        }
1493
1494        /*
1495         * Get and sanity-check the root inode.
1496         * Save the pointer to it in the mount structure.
1497         */
1498        error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1499        if (error) {
1500                xfs_warn(mp, "failed to read root inode");
1501                goto out_log_dealloc;
1502        }
1503
1504        ASSERT(rip != NULL);
1505
1506        if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
1507                xfs_warn(mp, "corrupted root inode %llu: not a directory",
1508                        (unsigned long long)rip->i_ino);
1509                xfs_iunlock(rip, XFS_ILOCK_EXCL);
1510                XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1511                                 mp);
1512                error = XFS_ERROR(EFSCORRUPTED);
1513                goto out_rele_rip;
1514        }
1515        mp->m_rootip = rip;     /* save it */
1516
1517        xfs_iunlock(rip, XFS_ILOCK_EXCL);
1518
1519        /*
1520         * Initialize realtime inode pointers in the mount structure
1521         */
1522        error = xfs_rtmount_inodes(mp);
1523        if (error) {
1524                /*
1525                 * Free up the root inode.
1526                 */
1527                xfs_warn(mp, "failed to read RT inodes");
1528                goto out_rele_rip;
1529        }
1530
1531        /*
1532         * If this is a read-only mount defer the superblock updates until
1533         * the next remount into writeable mode.  Otherwise we would never
1534         * perform the update e.g. for the root filesystem.
1535         */
1536        if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1537                error = xfs_mount_log_sb(mp, mp->m_update_flags);
1538                if (error) {
1539                        xfs_warn(mp, "failed to write sb changes");
1540                        goto out_rtunmount;
1541                }
1542        }
1543
1544        /*
1545         * Initialise the XFS quota management subsystem for this mount
1546         */
1547        if (XFS_IS_QUOTA_RUNNING(mp)) {
1548                error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1549                if (error)
1550                        goto out_rtunmount;
1551        } else {
1552                ASSERT(!XFS_IS_QUOTA_ON(mp));
1553
1554                /*
1555                 * If a file system had quotas running earlier, but decided to
1556                 * mount without -o uquota/pquota/gquota options, revoke the
1557                 * quotachecked license.
1558                 */
1559                if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1560                        xfs_notice(mp, "resetting quota flags");
1561                        error = xfs_mount_reset_sbqflags(mp);
1562                        if (error)
1563                                return error;
1564                }
1565        }
1566
1567        /*
1568         * Finish recovering the file system.  This part needed to be
1569         * delayed until after the root and real-time bitmap inodes
1570         * were consistently read in.
1571         */
1572        error = xfs_log_mount_finish(mp);
1573        if (error) {
1574                xfs_warn(mp, "log mount finish failed");
1575                goto out_rtunmount;
1576        }
1577
1578        /*
1579         * Complete the quota initialisation, post-log-replay component.
1580         */
1581        if (quotamount) {
1582                ASSERT(mp->m_qflags == 0);
1583                mp->m_qflags = quotaflags;
1584
1585                xfs_qm_mount_quotas(mp);
1586        }
1587
1588        /*
1589         * Now we are mounted, reserve a small amount of unused space for
1590         * privileged transactions. This is needed so that transaction
1591         * space required for critical operations can dip into this pool
1592         * when at ENOSPC. This is needed for operations like create with
1593         * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1594         * are not allowed to use this reserved space.
1595         *
1596         * This may drive us straight to ENOSPC on mount, but that implies
1597         * we were already there on the last unmount. Warn if this occurs.
1598         */
1599        if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1600                resblks = xfs_default_resblks(mp);
1601                error = xfs_reserve_blocks(mp, &resblks, NULL);
1602                if (error)
1603                        xfs_warn(mp,
1604        "Unable to allocate reserve blocks. Continuing without reserve pool.");
1605        }
1606
1607        return 0;
1608
1609 out_rtunmount:
1610        xfs_rtunmount_inodes(mp);
1611 out_rele_rip:
1612        IRELE(rip);
1613 out_log_dealloc:
1614        xfs_log_unmount(mp);
1615 out_fail_wait:
1616        if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1617                xfs_wait_buftarg(mp->m_logdev_targp);
1618        xfs_wait_buftarg(mp->m_ddev_targp);
1619 out_free_perag:
1620        xfs_free_perag(mp);
1621 out_remove_uuid:
1622        xfs_uuid_unmount(mp);
1623 out:
1624        return error;
1625}
1626
1627/*
1628 * This flushes out the inodes,dquots and the superblock, unmounts the
1629 * log and makes sure that incore structures are freed.
1630 */
1631void
1632xfs_unmountfs(
1633        struct xfs_mount        *mp)
1634{
1635        __uint64_t              resblks;
1636        int                     error;
1637
1638        cancel_delayed_work_sync(&mp->m_eofblocks_work);
1639
1640        xfs_qm_unmount_quotas(mp);
1641        xfs_rtunmount_inodes(mp);
1642        IRELE(mp->m_rootip);
1643
1644        /*
1645         * We can potentially deadlock here if we have an inode cluster
1646         * that has been freed has its buffer still pinned in memory because
1647         * the transaction is still sitting in a iclog. The stale inodes
1648         * on that buffer will have their flush locks held until the
1649         * transaction hits the disk and the callbacks run. the inode
1650         * flush takes the flush lock unconditionally and with nothing to
1651         * push out the iclog we will never get that unlocked. hence we
1652         * need to force the log first.
1653         */
1654        xfs_log_force(mp, XFS_LOG_SYNC);
1655
1656        /*
1657         * Flush all pending changes from the AIL.
1658         */
1659        xfs_ail_push_all_sync(mp->m_ail);
1660
1661        /*
1662         * And reclaim all inodes.  At this point there should be no dirty
1663         * inodes and none should be pinned or locked, but use synchronous
1664         * reclaim just to be sure. We can stop background inode reclaim
1665         * here as well if it is still running.
1666         */
1667        cancel_delayed_work_sync(&mp->m_reclaim_work);
1668        xfs_reclaim_inodes(mp, SYNC_WAIT);
1669
1670        xfs_qm_unmount(mp);
1671
1672        /*
1673         * Unreserve any blocks we have so that when we unmount we don't account
1674         * the reserved free space as used. This is really only necessary for
1675         * lazy superblock counting because it trusts the incore superblock
1676         * counters to be absolutely correct on clean unmount.
1677         *
1678         * We don't bother correcting this elsewhere for lazy superblock
1679         * counting because on mount of an unclean filesystem we reconstruct the
1680         * correct counter value and this is irrelevant.
1681         *
1682         * For non-lazy counter filesystems, this doesn't matter at all because
1683         * we only every apply deltas to the superblock and hence the incore
1684         * value does not matter....
1685         */
1686        resblks = 0;
1687        error = xfs_reserve_blocks(mp, &resblks, NULL);
1688        if (error)
1689                xfs_warn(mp, "Unable to free reserved block pool. "
1690                                "Freespace may not be correct on next mount.");
1691
1692        error = xfs_log_sbcount(mp);
1693        if (error)
1694                xfs_warn(mp, "Unable to update superblock counters. "
1695                                "Freespace may not be correct on next mount.");
1696
1697        xfs_log_unmount(mp);
1698        xfs_uuid_unmount(mp);
1699
1700#if defined(DEBUG)
1701        xfs_errortag_clearall(mp, 0);
1702#endif
1703        xfs_free_perag(mp);
1704}
1705
1706int
1707xfs_fs_writable(xfs_mount_t *mp)
1708{
1709        return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
1710                (mp->m_flags & XFS_MOUNT_RDONLY));
1711}
1712
1713/*
1714 * xfs_log_sbcount
1715 *
1716 * Sync the superblock counters to disk.
1717 *
1718 * Note this code can be called during the process of freezing, so
1719 * we may need to use the transaction allocator which does not
1720 * block when the transaction subsystem is in its frozen state.
1721 */
1722int
1723xfs_log_sbcount(xfs_mount_t *mp)
1724{
1725        xfs_trans_t     *tp;
1726        int             error;
1727
1728        if (!xfs_fs_writable(mp))
1729                return 0;
1730
1731        xfs_icsb_sync_counters(mp, 0);
1732
1733        /*
1734         * we don't need to do this if we are updating the superblock
1735         * counters on every modification.
1736         */
1737        if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1738                return 0;
1739
1740        tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1741        error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0,
1742                                  XFS_DEFAULT_LOG_COUNT);
1743        if (error) {
1744                xfs_trans_cancel(tp, 0);
1745                return error;
1746        }
1747
1748        xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1749        xfs_trans_set_sync(tp);
1750        error = xfs_trans_commit(tp, 0);
1751        return error;
1752}
1753
1754/*
1755 * xfs_mod_sb() can be used to copy arbitrary changes to the
1756 * in-core superblock into the superblock buffer to be logged.
1757 * It does not provide the higher level of locking that is
1758 * needed to protect the in-core superblock from concurrent
1759 * access.
1760 */
1761void
1762xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1763{
1764        xfs_buf_t       *bp;
1765        int             first;
1766        int             last;
1767        xfs_mount_t     *mp;
1768        xfs_sb_field_t  f;
1769
1770        ASSERT(fields);
1771        if (!fields)
1772                return;
1773        mp = tp->t_mountp;
1774        bp = xfs_trans_getsb(tp, mp, 0);
1775        first = sizeof(xfs_sb_t);
1776        last = 0;
1777
1778        /* translate/copy */
1779
1780        xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1781
1782        /* find modified range */
1783        f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1784        ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1785        last = xfs_sb_info[f + 1].offset - 1;
1786
1787        f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1788        ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1789        first = xfs_sb_info[f].offset;
1790
1791        xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
1792        xfs_trans_log_buf(tp, bp, first, last);
1793}
1794
1795
1796/*
1797 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1798 * a delta to a specified field in the in-core superblock.  Simply
1799 * switch on the field indicated and apply the delta to that field.
1800 * Fields are not allowed to dip below zero, so if the delta would
1801 * do this do not apply it and return EINVAL.
1802 *
1803 * The m_sb_lock must be held when this routine is called.
1804 */
1805STATIC int
1806xfs_mod_incore_sb_unlocked(
1807        xfs_mount_t     *mp,
1808        xfs_sb_field_t  field,
1809        int64_t         delta,
1810        int             rsvd)
1811{
1812        int             scounter;       /* short counter for 32 bit fields */
1813        long long       lcounter;       /* long counter for 64 bit fields */
1814        long long       res_used, rem;
1815
1816        /*
1817         * With the in-core superblock spin lock held, switch
1818         * on the indicated field.  Apply the delta to the
1819         * proper field.  If the fields value would dip below
1820         * 0, then do not apply the delta and return EINVAL.
1821         */
1822        switch (field) {
1823        case XFS_SBS_ICOUNT:
1824                lcounter = (long long)mp->m_sb.sb_icount;
1825                lcounter += delta;
1826                if (lcounter < 0) {
1827                        ASSERT(0);
1828                        return XFS_ERROR(EINVAL);
1829                }
1830                mp->m_sb.sb_icount = lcounter;
1831                return 0;
1832        case XFS_SBS_IFREE:
1833                lcounter = (long long)mp->m_sb.sb_ifree;
1834                lcounter += delta;
1835                if (lcounter < 0) {
1836                        ASSERT(0);
1837                        return XFS_ERROR(EINVAL);
1838                }
1839                mp->m_sb.sb_ifree = lcounter;
1840                return 0;
1841        case XFS_SBS_FDBLOCKS:
1842                lcounter = (long long)
1843                        mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1844                res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1845
1846                if (delta > 0) {                /* Putting blocks back */
1847                        if (res_used > delta) {
1848                                mp->m_resblks_avail += delta;
1849                        } else {
1850                                rem = delta - res_used;
1851                                mp->m_resblks_avail = mp->m_resblks;
1852                                lcounter += rem;
1853                        }
1854                } else {                                /* Taking blocks away */
1855                        lcounter += delta;
1856                        if (lcounter >= 0) {
1857                                mp->m_sb.sb_fdblocks = lcounter +
1858                                                        XFS_ALLOC_SET_ASIDE(mp);
1859                                return 0;
1860                        }
1861
1862                        /*
1863                         * We are out of blocks, use any available reserved
1864                         * blocks if were allowed to.
1865                         */
1866                        if (!rsvd)
1867                                return XFS_ERROR(ENOSPC);
1868
1869                        lcounter = (long long)mp->m_resblks_avail + delta;
1870                        if (lcounter >= 0) {
1871                                mp->m_resblks_avail = lcounter;
1872                                return 0;
1873                        }
1874                        printk_once(KERN_WARNING
1875                                "Filesystem \"%s\": reserve blocks depleted! "
1876                                "Consider increasing reserve pool size.",
1877                                mp->m_fsname);
1878                        return XFS_ERROR(ENOSPC);
1879                }
1880
1881                mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1882                return 0;
1883        case XFS_SBS_FREXTENTS:
1884                lcounter = (long long)mp->m_sb.sb_frextents;
1885                lcounter += delta;
1886                if (lcounter < 0) {
1887                        return XFS_ERROR(ENOSPC);
1888                }
1889                mp->m_sb.sb_frextents = lcounter;
1890                return 0;
1891        case XFS_SBS_DBLOCKS:
1892                lcounter = (long long)mp->m_sb.sb_dblocks;
1893                lcounter += delta;
1894                if (lcounter < 0) {
1895                        ASSERT(0);
1896                        return XFS_ERROR(EINVAL);
1897                }
1898                mp->m_sb.sb_dblocks = lcounter;
1899                return 0;
1900        case XFS_SBS_AGCOUNT:
1901                scounter = mp->m_sb.sb_agcount;
1902                scounter += delta;
1903                if (scounter < 0) {
1904                        ASSERT(0);
1905                        return XFS_ERROR(EINVAL);
1906                }
1907                mp->m_sb.sb_agcount = scounter;
1908                return 0;
1909        case XFS_SBS_IMAX_PCT:
1910                scounter = mp->m_sb.sb_imax_pct;
1911                scounter += delta;
1912                if (scounter < 0) {
1913                        ASSERT(0);
1914                        return XFS_ERROR(EINVAL);
1915                }
1916                mp->m_sb.sb_imax_pct = scounter;
1917                return 0;
1918        case XFS_SBS_REXTSIZE:
1919                scounter = mp->m_sb.sb_rextsize;
1920                scounter += delta;
1921                if (scounter < 0) {
1922                        ASSERT(0);
1923                        return XFS_ERROR(EINVAL);
1924                }
1925                mp->m_sb.sb_rextsize = scounter;
1926                return 0;
1927        case XFS_SBS_RBMBLOCKS:
1928                scounter = mp->m_sb.sb_rbmblocks;
1929                scounter += delta;
1930                if (scounter < 0) {
1931                        ASSERT(0);
1932                        return XFS_ERROR(EINVAL);
1933                }
1934                mp->m_sb.sb_rbmblocks = scounter;
1935                return 0;
1936        case XFS_SBS_RBLOCKS:
1937                lcounter = (long long)mp->m_sb.sb_rblocks;
1938                lcounter += delta;
1939                if (lcounter < 0) {
1940                        ASSERT(0);
1941                        return XFS_ERROR(EINVAL);
1942                }
1943                mp->m_sb.sb_rblocks = lcounter;
1944                return 0;
1945        case XFS_SBS_REXTENTS:
1946                lcounter = (long long)mp->m_sb.sb_rextents;
1947                lcounter += delta;
1948                if (lcounter < 0) {
1949                        ASSERT(0);
1950                        return XFS_ERROR(EINVAL);
1951                }
1952                mp->m_sb.sb_rextents = lcounter;
1953                return 0;
1954        case XFS_SBS_REXTSLOG:
1955                scounter = mp->m_sb.sb_rextslog;
1956                scounter += delta;
1957                if (scounter < 0) {
1958                        ASSERT(0);
1959                        return XFS_ERROR(EINVAL);
1960                }
1961                mp->m_sb.sb_rextslog = scounter;
1962                return 0;
1963        default:
1964                ASSERT(0);
1965                return XFS_ERROR(EINVAL);
1966        }
1967}
1968
1969/*
1970 * xfs_mod_incore_sb() is used to change a field in the in-core
1971 * superblock structure by the specified delta.  This modification
1972 * is protected by the m_sb_lock.  Just use the xfs_mod_incore_sb_unlocked()
1973 * routine to do the work.
1974 */
1975int
1976xfs_mod_incore_sb(
1977        struct xfs_mount        *mp,
1978        xfs_sb_field_t          field,
1979        int64_t                 delta,
1980        int                     rsvd)
1981{
1982        int                     status;
1983
1984#ifdef HAVE_PERCPU_SB
1985        ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1986#endif
1987        spin_lock(&mp->m_sb_lock);
1988        status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1989        spin_unlock(&mp->m_sb_lock);
1990
1991        return status;
1992}
1993
1994/*
1995 * Change more than one field in the in-core superblock structure at a time.
1996 *
1997 * The fields and changes to those fields are specified in the array of
1998 * xfs_mod_sb structures passed in.  Either all of the specified deltas
1999 * will be applied or none of them will.  If any modified field dips below 0,
2000 * then all modifications will be backed out and EINVAL will be returned.
2001 *
2002 * Note that this function may not be used for the superblock values that
2003 * are tracked with the in-memory per-cpu counters - a direct call to
2004 * xfs_icsb_modify_counters is required for these.
2005 */
2006int
2007xfs_mod_incore_sb_batch(
2008        struct xfs_mount        *mp,
2009        xfs_mod_sb_t            *msb,
2010        uint                    nmsb,
2011        int                     rsvd)
2012{
2013        xfs_mod_sb_t            *msbp;
2014        int                     error = 0;
2015
2016        /*
2017         * Loop through the array of mod structures and apply each individually.
2018         * If any fail, then back out all those which have already been applied.
2019         * Do all of this within the scope of the m_sb_lock so that all of the
2020         * changes will be atomic.
2021         */
2022        spin_lock(&mp->m_sb_lock);
2023        for (msbp = msb; msbp < (msb + nmsb); msbp++) {
2024                ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
2025                       msbp->msb_field > XFS_SBS_FDBLOCKS);
2026
2027                error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
2028                                                   msbp->msb_delta, rsvd);
2029                if (error)
2030                        goto unwind;
2031        }
2032        spin_unlock(&mp->m_sb_lock);
2033        return 0;
2034
2035unwind:
2036        while (--msbp >= msb) {
2037                error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
2038                                                   -msbp->msb_delta, rsvd);
2039                ASSERT(error == 0);
2040        }
2041        spin_unlock(&mp->m_sb_lock);
2042        return error;
2043}
2044
2045/*
2046 * xfs_getsb() is called to obtain the buffer for the superblock.
2047 * The buffer is returned locked and read in from disk.
2048 * The buffer should be released with a call to xfs_brelse().
2049 *
2050 * If the flags parameter is BUF_TRYLOCK, then we'll only return
2051 * the superblock buffer if it can be locked without sleeping.
2052 * If it can't then we'll return NULL.
2053 */
2054struct xfs_buf *
2055xfs_getsb(
2056        struct xfs_mount        *mp,
2057        int                     flags)
2058{
2059        struct xfs_buf          *bp = mp->m_sb_bp;
2060
2061        if (!xfs_buf_trylock(bp)) {
2062                if (flags & XBF_TRYLOCK)
2063                        return NULL;
2064                xfs_buf_lock(bp);
2065        }
2066
2067        xfs_buf_hold(bp);
2068        ASSERT(XFS_BUF_ISDONE(bp));
2069        return bp;
2070}
2071
2072/*
2073 * Used to free the superblock along various error paths.
2074 */
2075void
2076xfs_freesb(
2077        struct xfs_mount        *mp)
2078{
2079        struct xfs_buf          *bp = mp->m_sb_bp;
2080
2081        xfs_buf_lock(bp);
2082        mp->m_sb_bp = NULL;
2083        xfs_buf_relse(bp);
2084}
2085
2086/*
2087 * Used to log changes to the superblock unit and width fields which could
2088 * be altered by the mount options, as well as any potential sb_features2
2089 * fixup. Only the first superblock is updated.
2090 */
2091int
2092xfs_mount_log_sb(
2093        xfs_mount_t     *mp,
2094        __int64_t       fields)
2095{
2096        xfs_trans_t     *tp;
2097        int             error;
2098
2099        ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
2100                         XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
2101                         XFS_SB_VERSIONNUM));
2102
2103        tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
2104        error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0,
2105                                  XFS_DEFAULT_LOG_COUNT);
2106        if (error) {
2107                xfs_trans_cancel(tp, 0);
2108                return error;
2109        }
2110        xfs_mod_sb(tp, fields);
2111        error = xfs_trans_commit(tp, 0);
2112        return error;
2113}
2114
2115/*
2116 * If the underlying (data/log/rt) device is readonly, there are some
2117 * operations that cannot proceed.
2118 */
2119int
2120xfs_dev_is_read_only(
2121        struct xfs_mount        *mp,
2122        char                    *message)
2123{
2124        if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
2125            xfs_readonly_buftarg(mp->m_logdev_targp) ||
2126            (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
2127                xfs_notice(mp, "%s required on read-only device.", message);
2128                xfs_notice(mp, "write access unavailable, cannot proceed.");
2129                return EROFS;
2130        }
2131        return 0;
2132}
2133
2134#ifdef HAVE_PERCPU_SB
2135/*
2136 * Per-cpu incore superblock counters
2137 *
2138 * Simple concept, difficult implementation
2139 *
2140 * Basically, replace the incore superblock counters with a distributed per cpu
2141 * counter for contended fields (e.g.  free block count).
2142 *
2143 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2144 * hence needs to be accurately read when we are running low on space. Hence
2145 * there is a method to enable and disable the per-cpu counters based on how
2146 * much "stuff" is available in them.
2147 *
2148 * Basically, a counter is enabled if there is enough free resource to justify
2149 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2150 * ENOSPC), then we disable the counters to synchronise all callers and
2151 * re-distribute the available resources.
2152 *
2153 * If, once we redistributed the available resources, we still get a failure,
2154 * we disable the per-cpu counter and go through the slow path.
2155 *
2156 * The slow path is the current xfs_mod_incore_sb() function.  This means that
2157 * when we disable a per-cpu counter, we need to drain its resources back to
2158 * the global superblock. We do this after disabling the counter to prevent
2159 * more threads from queueing up on the counter.
2160 *
2161 * Essentially, this means that we still need a lock in the fast path to enable
2162 * synchronisation between the global counters and the per-cpu counters. This
2163 * is not a problem because the lock will be local to a CPU almost all the time
2164 * and have little contention except when we get to ENOSPC conditions.
2165 *
2166 * Basically, this lock becomes a barrier that enables us to lock out the fast
2167 * path while we do things like enabling and disabling counters and
2168 * synchronising the counters.
2169 *
2170 * Locking rules:
2171 *
2172 *      1. m_sb_lock before picking up per-cpu locks
2173 *      2. per-cpu locks always picked up via for_each_online_cpu() order
2174 *      3. accurate counter sync requires m_sb_lock + per cpu locks
2175 *      4. modifying per-cpu counters requires holding per-cpu lock
2176 *      5. modifying global counters requires holding m_sb_lock
2177 *      6. enabling or disabling a counter requires holding the m_sb_lock 
2178 *         and _none_ of the per-cpu locks.
2179 *
2180 * Disabled counters are only ever re-enabled by a balance operation
2181 * that results in more free resources per CPU than a given threshold.
2182 * To ensure counters don't remain disabled, they are rebalanced when
2183 * the global resource goes above a higher threshold (i.e. some hysteresis
2184 * is present to prevent thrashing).
2185 */
2186
2187#ifdef CONFIG_HOTPLUG_CPU
2188/*
2189 * hot-plug CPU notifier support.
2190 *
2191 * We need a notifier per filesystem as we need to be able to identify
2192 * the filesystem to balance the counters out. This is achieved by
2193 * having a notifier block embedded in the xfs_mount_t and doing pointer
2194 * magic to get the mount pointer from the notifier block address.
2195 */
2196STATIC int
2197xfs_icsb_cpu_notify(
2198        struct notifier_block *nfb,
2199        unsigned long action,
2200        void *hcpu)
2201{
2202        xfs_icsb_cnts_t *cntp;
2203        xfs_mount_t     *mp;
2204
2205        mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2206        cntp = (xfs_icsb_cnts_t *)
2207                        per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2208        switch (action) {
2209        case CPU_UP_PREPARE:
2210        case CPU_UP_PREPARE_FROZEN:
2211                /* Easy Case - initialize the area and locks, and
2212                 * then rebalance when online does everything else for us. */
2213                memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2214                break;
2215        case CPU_ONLINE:
2216        case CPU_ONLINE_FROZEN:
2217                xfs_icsb_lock(mp);
2218                xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2219                xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2220                xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2221                xfs_icsb_unlock(mp);
2222                break;
2223        case CPU_DEAD:
2224        case CPU_DEAD_FROZEN:
2225                /* Disable all the counters, then fold the dead cpu's
2226                 * count into the total on the global superblock and
2227                 * re-enable the counters. */
2228                xfs_icsb_lock(mp);
2229                spin_lock(&mp->m_sb_lock);
2230                xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2231                xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2232                xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2233
2234                mp->m_sb.sb_icount += cntp->icsb_icount;
2235                mp->m_sb.sb_ifree += cntp->icsb_ifree;
2236                mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2237
2238                memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2239
2240                xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2241                xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2242                xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2243                spin_unlock(&mp->m_sb_lock);
2244                xfs_icsb_unlock(mp);
2245                break;
2246        }
2247
2248        return NOTIFY_OK;
2249}
2250#endif /* CONFIG_HOTPLUG_CPU */
2251
2252int
2253xfs_icsb_init_counters(
2254        xfs_mount_t     *mp)
2255{
2256        xfs_icsb_cnts_t *cntp;
2257        int             i;
2258
2259        mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2260        if (mp->m_sb_cnts == NULL)
2261                return -ENOMEM;
2262
2263#ifdef CONFIG_HOTPLUG_CPU
2264        mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2265        mp->m_icsb_notifier.priority = 0;
2266        register_hotcpu_notifier(&mp->m_icsb_notifier);
2267#endif /* CONFIG_HOTPLUG_CPU */
2268
2269        for_each_online_cpu(i) {
2270                cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2271                memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2272        }
2273
2274        mutex_init(&mp->m_icsb_mutex);
2275
2276        /*
2277         * start with all counters disabled so that the
2278         * initial balance kicks us off correctly
2279         */
2280        mp->m_icsb_counters = -1;
2281        return 0;
2282}
2283
2284void
2285xfs_icsb_reinit_counters(
2286        xfs_mount_t     *mp)
2287{
2288        xfs_icsb_lock(mp);
2289        /*
2290         * start with all counters disabled so that the
2291         * initial balance kicks us off correctly
2292         */
2293        mp->m_icsb_counters = -1;
2294        xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2295        xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2296        xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2297        xfs_icsb_unlock(mp);
2298}
2299
2300void
2301xfs_icsb_destroy_counters(
2302        xfs_mount_t     *mp)
2303{
2304        if (mp->m_sb_cnts) {
2305                unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2306                free_percpu(mp->m_sb_cnts);
2307        }
2308        mutex_destroy(&mp->m_icsb_mutex);
2309}
2310
2311STATIC void
2312xfs_icsb_lock_cntr(
2313        xfs_icsb_cnts_t *icsbp)
2314{
2315        while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2316                ndelay(1000);
2317        }
2318}
2319
2320STATIC void
2321xfs_icsb_unlock_cntr(
2322        xfs_icsb_cnts_t *icsbp)
2323{
2324        clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2325}
2326
2327
2328STATIC void
2329xfs_icsb_lock_all_counters(
2330        xfs_mount_t     *mp)
2331{
2332        xfs_icsb_cnts_t *cntp;
2333        int             i;
2334
2335        for_each_online_cpu(i) {
2336                cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2337                xfs_icsb_lock_cntr(cntp);
2338        }
2339}
2340
2341STATIC void
2342xfs_icsb_unlock_all_counters(
2343        xfs_mount_t     *mp)
2344{
2345        xfs_icsb_cnts_t *cntp;
2346        int             i;
2347
2348        for_each_online_cpu(i) {
2349                cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2350                xfs_icsb_unlock_cntr(cntp);
2351        }
2352}
2353
2354STATIC void
2355xfs_icsb_count(
2356        xfs_mount_t     *mp,
2357        xfs_icsb_cnts_t *cnt,
2358        int             flags)
2359{
2360        xfs_icsb_cnts_t *cntp;
2361        int             i;
2362
2363        memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2364
2365        if (!(flags & XFS_ICSB_LAZY_COUNT))
2366                xfs_icsb_lock_all_counters(mp);
2367
2368        for_each_online_cpu(i) {
2369                cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2370                cnt->icsb_icount += cntp->icsb_icount;
2371                cnt->icsb_ifree += cntp->icsb_ifree;
2372                cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2373        }
2374
2375        if (!(flags & XFS_ICSB_LAZY_COUNT))
2376                xfs_icsb_unlock_all_counters(mp);
2377}
2378
2379STATIC int
2380xfs_icsb_counter_disabled(
2381        xfs_mount_t     *mp,
2382        xfs_sb_field_t  field)
2383{
2384        ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2385        return test_bit(field, &mp->m_icsb_counters);
2386}
2387
2388STATIC void
2389xfs_icsb_disable_counter(
2390        xfs_mount_t     *mp,
2391        xfs_sb_field_t  field)
2392{
2393        xfs_icsb_cnts_t cnt;
2394
2395        ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2396
2397        /*
2398         * If we are already disabled, then there is nothing to do
2399         * here. We check before locking all the counters to avoid
2400         * the expensive lock operation when being called in the
2401         * slow path and the counter is already disabled. This is
2402         * safe because the only time we set or clear this state is under
2403         * the m_icsb_mutex.
2404         */
2405        if (xfs_icsb_counter_disabled(mp, field))
2406                return;
2407
2408        xfs_icsb_lock_all_counters(mp);
2409        if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2410                /* drain back to superblock */
2411
2412                xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2413                switch(field) {
2414                case XFS_SBS_ICOUNT:
2415                        mp->m_sb.sb_icount = cnt.icsb_icount;
2416                        break;
2417                case XFS_SBS_IFREE:
2418                        mp->m_sb.sb_ifree = cnt.icsb_ifree;
2419                        break;
2420                case XFS_SBS_FDBLOCKS:
2421                        mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2422                        break;
2423                default:
2424                        BUG();
2425                }
2426        }
2427
2428        xfs_icsb_unlock_all_counters(mp);
2429}
2430
2431STATIC void
2432xfs_icsb_enable_counter(
2433        xfs_mount_t     *mp,
2434        xfs_sb_field_t  field,
2435        uint64_t        count,
2436        uint64_t        resid)
2437{
2438        xfs_icsb_cnts_t *cntp;
2439        int             i;
2440
2441        ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2442
2443        xfs_icsb_lock_all_counters(mp);
2444        for_each_online_cpu(i) {
2445                cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2446                switch (field) {
2447                case XFS_SBS_ICOUNT:
2448                        cntp->icsb_icount = count + resid;
2449                        break;
2450                case XFS_SBS_IFREE:
2451                        cntp->icsb_ifree = count + resid;
2452                        break;
2453                case XFS_SBS_FDBLOCKS:
2454                        cntp->icsb_fdblocks = count + resid;
2455                        break;
2456                default:
2457                        BUG();
2458                        break;
2459                }
2460                resid = 0;
2461        }
2462        clear_bit(field, &mp->m_icsb_counters);
2463        xfs_icsb_unlock_all_counters(mp);
2464}
2465
2466void
2467xfs_icsb_sync_counters_locked(
2468        xfs_mount_t     *mp,
2469        int             flags)
2470{
2471        xfs_icsb_cnts_t cnt;
2472
2473        xfs_icsb_count(mp, &cnt, flags);
2474
2475        if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2476                mp->m_sb.sb_icount = cnt.icsb_icount;
2477        if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2478                mp->m_sb.sb_ifree = cnt.icsb_ifree;
2479        if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2480                mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2481}
2482
2483/*
2484 * Accurate update of per-cpu counters to incore superblock
2485 */
2486void
2487xfs_icsb_sync_counters(
2488        xfs_mount_t     *mp,
2489        int             flags)
2490{
2491        spin_lock(&mp->m_sb_lock);
2492        xfs_icsb_sync_counters_locked(mp, flags);
2493        spin_unlock(&mp->m_sb_lock);
2494}
2495
2496/*
2497 * Balance and enable/disable counters as necessary.
2498 *
2499 * Thresholds for re-enabling counters are somewhat magic.  inode counts are
2500 * chosen to be the same number as single on disk allocation chunk per CPU, and
2501 * free blocks is something far enough zero that we aren't going thrash when we
2502 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2503 * prevent looping endlessly when xfs_alloc_space asks for more than will
2504 * be distributed to a single CPU but each CPU has enough blocks to be
2505 * reenabled.
2506 *
2507 * Note that we can be called when counters are already disabled.
2508 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2509 * prevent locking every per-cpu counter needlessly.
2510 */
2511
2512#define XFS_ICSB_INO_CNTR_REENABLE      (uint64_t)64
2513#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2514                (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2515STATIC void
2516xfs_icsb_balance_counter_locked(
2517        xfs_mount_t     *mp,
2518        xfs_sb_field_t  field,
2519        int             min_per_cpu)
2520{
2521        uint64_t        count, resid;
2522        int             weight = num_online_cpus();
2523        uint64_t        min = (uint64_t)min_per_cpu;
2524
2525        /* disable counter and sync counter */
2526        xfs_icsb_disable_counter(mp, field);
2527
2528        /* update counters  - first CPU gets residual*/
2529        switch (field) {
2530        case XFS_SBS_ICOUNT:
2531                count = mp->m_sb.sb_icount;
2532                resid = do_div(count, weight);
2533                if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2534                        return;
2535                break;
2536        case XFS_SBS_IFREE:
2537                count = mp->m_sb.sb_ifree;
2538                resid = do_div(count, weight);
2539                if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2540                        return;
2541                break;
2542        case XFS_SBS_FDBLOCKS:
2543                count = mp->m_sb.sb_fdblocks;
2544                resid = do_div(count, weight);
2545                if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2546                        return;
2547                break;
2548        default:
2549                BUG();
2550                count = resid = 0;      /* quiet, gcc */
2551                break;
2552        }
2553
2554        xfs_icsb_enable_counter(mp, field, count, resid);
2555}
2556
2557STATIC void
2558xfs_icsb_balance_counter(
2559        xfs_mount_t     *mp,
2560        xfs_sb_field_t  fields,
2561        int             min_per_cpu)
2562{
2563        spin_lock(&mp->m_sb_lock);
2564        xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2565        spin_unlock(&mp->m_sb_lock);
2566}
2567
2568int
2569xfs_icsb_modify_counters(
2570        xfs_mount_t     *mp,
2571        xfs_sb_field_t  field,
2572        int64_t         delta,
2573        int             rsvd)
2574{
2575        xfs_icsb_cnts_t *icsbp;
2576        long long       lcounter;       /* long counter for 64 bit fields */
2577        int             ret = 0;
2578
2579        might_sleep();
2580again:
2581        preempt_disable();
2582        icsbp = this_cpu_ptr(mp->m_sb_cnts);
2583
2584        /*
2585         * if the counter is disabled, go to slow path
2586         */
2587        if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2588                goto slow_path;
2589        xfs_icsb_lock_cntr(icsbp);
2590        if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2591                xfs_icsb_unlock_cntr(icsbp);
2592                goto slow_path;
2593        }
2594
2595        switch (field) {
2596        case XFS_SBS_ICOUNT:
2597                lcounter = icsbp->icsb_icount;
2598                lcounter += delta;
2599                if (unlikely(lcounter < 0))
2600                        goto balance_counter;
2601                icsbp->icsb_icount = lcounter;
2602                break;
2603
2604        case XFS_SBS_IFREE:
2605                lcounter = icsbp->icsb_ifree;
2606                lcounter += delta;
2607                if (unlikely(lcounter < 0))
2608                        goto balance_counter;
2609                icsbp->icsb_ifree = lcounter;
2610                break;
2611
2612        case XFS_SBS_FDBLOCKS:
2613                BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2614
2615                lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2616                lcounter += delta;
2617                if (unlikely(lcounter < 0))
2618                        goto balance_counter;
2619                icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2620                break;
2621        default:
2622                BUG();
2623                break;
2624        }
2625        xfs_icsb_unlock_cntr(icsbp);
2626        preempt_enable();
2627        return 0;
2628
2629slow_path:
2630        preempt_enable();
2631
2632        /*
2633         * serialise with a mutex so we don't burn lots of cpu on
2634         * the superblock lock. We still need to hold the superblock
2635         * lock, however, when we modify the global structures.
2636         */
2637        xfs_icsb_lock(mp);
2638
2639        /*
2640         * Now running atomically.
2641         *
2642         * If the counter is enabled, someone has beaten us to rebalancing.
2643         * Drop the lock and try again in the fast path....
2644         */
2645        if (!(xfs_icsb_counter_disabled(mp, field))) {
2646                xfs_icsb_unlock(mp);
2647                goto again;
2648        }
2649
2650        /*
2651         * The counter is currently disabled. Because we are
2652         * running atomically here, we know a rebalance cannot
2653         * be in progress. Hence we can go straight to operating
2654         * on the global superblock. We do not call xfs_mod_incore_sb()
2655         * here even though we need to get the m_sb_lock. Doing so
2656         * will cause us to re-enter this function and deadlock.
2657         * Hence we get the m_sb_lock ourselves and then call
2658         * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2659         * directly on the global counters.
2660         */
2661        spin_lock(&mp->m_sb_lock);
2662        ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2663        spin_unlock(&mp->m_sb_lock);
2664
2665        /*
2666         * Now that we've modified the global superblock, we
2667         * may be able to re-enable the distributed counters
2668         * (e.g. lots of space just got freed). After that
2669         * we are done.
2670         */
2671        if (ret != ENOSPC)
2672                xfs_icsb_balance_counter(mp, field, 0);
2673        xfs_icsb_unlock(mp);
2674        return ret;
2675
2676balance_counter:
2677        xfs_icsb_unlock_cntr(icsbp);
2678        preempt_enable();
2679
2680        /*
2681         * We may have multiple threads here if multiple per-cpu
2682         * counters run dry at the same time. This will mean we can
2683         * do more balances than strictly necessary but it is not
2684         * the common slowpath case.
2685         */
2686        xfs_icsb_lock(mp);
2687
2688        /*
2689         * running atomically.
2690         *
2691         * This will leave the counter in the correct state for future
2692         * accesses. After the rebalance, we simply try again and our retry
2693         * will either succeed through the fast path or slow path without
2694         * another balance operation being required.
2695         */
2696        xfs_icsb_balance_counter(mp, field, delta);
2697        xfs_icsb_unlock(mp);
2698        goto again;
2699}
2700
2701#endif
2702
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.