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