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, max_metadata;
 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                /*
 505                 * Calculate how much should be reserved for inodes to meet
 506                 * the max inode percentage.
 507                 */
 508                if (mp->m_maxicount) {
 509                        __uint64_t      icount;
 510
 511                        icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 512                        do_div(icount, 100);
 513                        icount += sbp->sb_agblocks - 1;
 514                        do_div(icount, sbp->sb_agblocks);
 515                        max_metadata = icount;
 516                } else {
 517                        max_metadata = agcount;
 518                }
 519
 520                for (index = 0; index < agcount; index++) {
 521                        ino = XFS_AGINO_TO_INO(mp, index, agino);
 522                        if (ino > XFS_MAXINUMBER_32) {
 523                                index++;
 524                                break;
 525                        }
 526
 527                        pag = xfs_perag_get(mp, index);
 528                        pag->pagi_inodeok = 1;
 529                        if (index < max_metadata)
 530                                pag->pagf_metadata = 1;
 531                        xfs_perag_put(pag);
 532                }
 533        } else {
 534                for (index = 0; index < agcount; index++) {
 535                        pag = xfs_perag_get(mp, index);
 536                        pag->pagi_inodeok = 1;
 537                        xfs_perag_put(pag);
 538                }
 539        }
 540
 541        if (maxagi)
 542                *maxagi = index;
 543        return 0;
 544
 545out_unwind:
 546        kmem_free(pag);
 547        for (; index > first_initialised; index--) {
 548                pag = radix_tree_delete(&mp->m_perag_tree, index);
 549                kmem_free(pag);
 550        }
 551        return error;
 552}
 553
 554void
 555xfs_sb_from_disk(
 556        struct xfs_mount        *mp,
 557        xfs_dsb_t       *from)
 558{
 559        struct xfs_sb *to = &mp->m_sb;
 560
 561        to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
 562        to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
 563        to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
 564        to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
 565        to->sb_rextents = be64_to_cpu(from->sb_rextents);
 566        memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
 567        to->sb_logstart = be64_to_cpu(from->sb_logstart);
 568        to->sb_rootino = be64_to_cpu(from->sb_rootino);
 569        to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
 570        to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
 571        to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
 572        to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
 573        to->sb_agcount = be32_to_cpu(from->sb_agcount);
 574        to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
 575        to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
 576        to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
 577        to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
 578        to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
 579        to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
 580        memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
 581        to->sb_blocklog = from->sb_blocklog;
 582        to->sb_sectlog = from->sb_sectlog;
 583        to->sb_inodelog = from->sb_inodelog;
 584        to->sb_inopblog = from->sb_inopblog;
 585        to->sb_agblklog = from->sb_agblklog;
 586        to->sb_rextslog = from->sb_rextslog;
 587        to->sb_inprogress = from->sb_inprogress;
 588        to->sb_imax_pct = from->sb_imax_pct;
 589        to->sb_icount = be64_to_cpu(from->sb_icount);
 590        to->sb_ifree = be64_to_cpu(from->sb_ifree);
 591        to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
 592        to->sb_frextents = be64_to_cpu(from->sb_frextents);
 593        to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
 594        to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
 595        to->sb_qflags = be16_to_cpu(from->sb_qflags);
 596        to->sb_flags = from->sb_flags;
 597        to->sb_shared_vn = from->sb_shared_vn;
 598        to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
 599        to->sb_unit = be32_to_cpu(from->sb_unit);
 600        to->sb_width = be32_to_cpu(from->sb_width);
 601        to->sb_dirblklog = from->sb_dirblklog;
 602        to->sb_logsectlog = from->sb_logsectlog;
 603        to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
 604        to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
 605        to->sb_features2 = be32_to_cpu(from->sb_features2);
 606        to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
 607}
 608
 609/*
 610 * Copy in core superblock to ondisk one.
 611 *
 612 * The fields argument is mask of superblock fields to copy.
 613 */
 614void
 615xfs_sb_to_disk(
 616        xfs_dsb_t       *to,
 617        xfs_sb_t        *from,
 618        __int64_t       fields)
 619{
 620        xfs_caddr_t     to_ptr = (xfs_caddr_t)to;
 621        xfs_caddr_t     from_ptr = (xfs_caddr_t)from;
 622        xfs_sb_field_t  f;
 623        int             first;
 624        int             size;
 625
 626        ASSERT(fields);
 627        if (!fields)
 628                return;
 629
 630        while (fields) {
 631                f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 632                first = xfs_sb_info[f].offset;
 633                size = xfs_sb_info[f + 1].offset - first;
 634
 635                ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
 636
 637                if (size == 1 || xfs_sb_info[f].type == 1) {
 638                        memcpy(to_ptr + first, from_ptr + first, size);
 639                } else {
 640                        switch (size) {
 641                        case 2:
 642                                *(__be16 *)(to_ptr + first) =
 643                                        cpu_to_be16(*(__u16 *)(from_ptr + first));
 644                                break;
 645                        case 4:
 646                                *(__be32 *)(to_ptr + first) =
 647                                        cpu_to_be32(*(__u32 *)(from_ptr + first));
 648                                break;
 649                        case 8:
 650                                *(__be64 *)(to_ptr + first) =
 651                                        cpu_to_be64(*(__u64 *)(from_ptr + first));
 652                                break;
 653                        default:
 654                                ASSERT(0);
 655                        }
 656                }
 657
 658                fields &= ~(1LL << f);
 659        }
 660}
 661
 662/*
 663 * xfs_readsb
 664 *
 665 * Does the initial read of the superblock.
 666 */
 667int
 668xfs_readsb(xfs_mount_t *mp, int flags)
 669{
 670        unsigned int    sector_size;
 671        xfs_buf_t       *bp;
 672        int             error;
 673        int             loud = !(flags & XFS_MFSI_QUIET);
 674
 675        ASSERT(mp->m_sb_bp == NULL);
 676        ASSERT(mp->m_ddev_targp != NULL);
 677
 678        /*
 679         * Allocate a (locked) buffer to hold the superblock.
 680         * This will be kept around at all times to optimize
 681         * access to the superblock.
 682         */
 683        sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
 684
 685reread:
 686        bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
 687                                        BTOBB(sector_size), 0);
 688        if (!bp) {
 689                if (loud)
 690                        xfs_warn(mp, "SB buffer read failed");
 691                return EIO;
 692        }
 693
 694        /*
 695         * Initialize the mount structure from the superblock.
 696         * But first do some basic consistency checking.
 697         */
 698        xfs_sb_from_disk(mp, XFS_BUF_TO_SBP(bp));
 699        error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
 700        if (error) {
 701                if (loud)
 702                        xfs_warn(mp, "SB validate failed");
 703                goto release_buf;
 704        }
 705
 706        /*
 707         * We must be able to do sector-sized and sector-aligned IO.
 708         */
 709        if (sector_size > mp->m_sb.sb_sectsize) {
 710                if (loud)
 711                        xfs_warn(mp, "device supports %u byte sectors (not %u)",
 712                                sector_size, mp->m_sb.sb_sectsize);
 713                error = ENOSYS;
 714                goto release_buf;
 715        }
 716
 717        /*
 718         * If device sector size is smaller than the superblock size,
 719         * re-read the superblock so the buffer is correctly sized.
 720         */
 721        if (sector_size < mp->m_sb.sb_sectsize) {
 722                xfs_buf_relse(bp);
 723                sector_size = mp->m_sb.sb_sectsize;
 724                goto reread;
 725        }
 726
 727        /* Initialize per-cpu counters */
 728        xfs_icsb_reinit_counters(mp);
 729
 730        mp->m_sb_bp = bp;
 731        xfs_buf_unlock(bp);
 732        return 0;
 733
 734release_buf:
 735        xfs_buf_relse(bp);
 736        return error;
 737}
 738
 739
 740/*
 741 * xfs_mount_common
 742 *
 743 * Mount initialization code establishing various mount
 744 * fields from the superblock associated with the given
 745 * mount structure
 746 */
 747STATIC void
 748xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
 749{
 750        mp->m_agfrotor = mp->m_agirotor = 0;
 751        spin_lock_init(&mp->m_agirotor_lock);
 752        mp->m_maxagi = mp->m_sb.sb_agcount;
 753        mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
 754        mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
 755        mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
 756        mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
 757        mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
 758        mp->m_blockmask = sbp->sb_blocksize - 1;
 759        mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
 760        mp->m_blockwmask = mp->m_blockwsize - 1;
 761
 762        mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
 763        mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
 764        mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
 765        mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
 766
 767        mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
 768        mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
 769        mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
 770        mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
 771
 772        mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
 773        mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
 774        mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
 775        mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
 776
 777        mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
 778        mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 779                                        sbp->sb_inopblock);
 780        mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 781}
 782
 783/*
 784 * xfs_initialize_perag_data
 785 *
 786 * Read in each per-ag structure so we can count up the number of
 787 * allocated inodes, free inodes and used filesystem blocks as this
 788 * information is no longer persistent in the superblock. Once we have
 789 * this information, write it into the in-core superblock structure.
 790 */
 791STATIC int
 792xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
 793{
 794        xfs_agnumber_t  index;
 795        xfs_perag_t     *pag;
 796        xfs_sb_t        *sbp = &mp->m_sb;
 797        uint64_t        ifree = 0;
 798        uint64_t        ialloc = 0;
 799        uint64_t        bfree = 0;
 800        uint64_t        bfreelst = 0;
 801        uint64_t        btree = 0;
 802        int             error;
 803
 804        for (index = 0; index < agcount; index++) {
 805                /*
 806                 * read the agf, then the agi. This gets us
 807                 * all the information we need and populates the
 808                 * per-ag structures for us.
 809                 */
 810                error = xfs_alloc_pagf_init(mp, NULL, index, 0);
 811                if (error)
 812                        return error;
 813
 814                error = xfs_ialloc_pagi_init(mp, NULL, index);
 815                if (error)
 816                        return error;
 817                pag = xfs_perag_get(mp, index);
 818                ifree += pag->pagi_freecount;
 819                ialloc += pag->pagi_count;
 820                bfree += pag->pagf_freeblks;
 821                bfreelst += pag->pagf_flcount;
 822                btree += pag->pagf_btreeblks;
 823                xfs_perag_put(pag);
 824        }
 825        /*
 826         * Overwrite incore superblock counters with just-read data
 827         */
 828        spin_lock(&mp->m_sb_lock);
 829        sbp->sb_ifree = ifree;
 830        sbp->sb_icount = ialloc;
 831        sbp->sb_fdblocks = bfree + bfreelst + btree;
 832        spin_unlock(&mp->m_sb_lock);
 833
 834        /* Fixup the per-cpu counters as well. */
 835        xfs_icsb_reinit_counters(mp);
 836
 837        return 0;
 838}
 839
 840/*
 841 * Update alignment values based on mount options and sb values
 842 */
 843STATIC int
 844xfs_update_alignment(xfs_mount_t *mp)
 845{
 846        xfs_sb_t        *sbp = &(mp->m_sb);
 847
 848        if (mp->m_dalign) {
 849                /*
 850                 * If stripe unit and stripe width are not multiples
 851                 * of the fs blocksize turn off alignment.
 852                 */
 853                if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
 854                    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
 855                        if (mp->m_flags & XFS_MOUNT_RETERR) {
 856                                xfs_warn(mp, "alignment check failed: "
 857                                         "(sunit/swidth vs. blocksize)");
 858                                return XFS_ERROR(EINVAL);
 859                        }
 860                        mp->m_dalign = mp->m_swidth = 0;
 861                } else {
 862                        /*
 863                         * Convert the stripe unit and width to FSBs.
 864                         */
 865                        mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
 866                        if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
 867                                if (mp->m_flags & XFS_MOUNT_RETERR) {
 868                                        xfs_warn(mp, "alignment check failed: "
 869                                                 "(sunit/swidth vs. ag size)");
 870                                        return XFS_ERROR(EINVAL);
 871                                }
 872                                xfs_warn(mp,
 873                "stripe alignment turned off: sunit(%d)/swidth(%d) "
 874                "incompatible with agsize(%d)",
 875                                        mp->m_dalign, mp->m_swidth,
 876                                        sbp->sb_agblocks);
 877
 878                                mp->m_dalign = 0;
 879                                mp->m_swidth = 0;
 880                        } else if (mp->m_dalign) {
 881                                mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
 882                        } else {
 883                                if (mp->m_flags & XFS_MOUNT_RETERR) {
 884                                        xfs_warn(mp, "alignment check failed: "
 885                                                "sunit(%d) less than bsize(%d)",
 886                                                mp->m_dalign,
 887                                                mp->m_blockmask +1);
 888                                        return XFS_ERROR(EINVAL);
 889                                }
 890                                mp->m_swidth = 0;
 891                        }
 892                }
 893
 894                /*
 895                 * Update superblock with new values
 896                 * and log changes
 897                 */
 898                if (xfs_sb_version_hasdalign(sbp)) {
 899                        if (sbp->sb_unit != mp->m_dalign) {
 900                                sbp->sb_unit = mp->m_dalign;
 901                                mp->m_update_flags |= XFS_SB_UNIT;
 902                        }
 903                        if (sbp->sb_width != mp->m_swidth) {
 904                                sbp->sb_width = mp->m_swidth;
 905                                mp->m_update_flags |= XFS_SB_WIDTH;
 906                        }
 907                }
 908        } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
 909                    xfs_sb_version_hasdalign(&mp->m_sb)) {
 910                        mp->m_dalign = sbp->sb_unit;
 911                        mp->m_swidth = sbp->sb_width;
 912        }
 913
 914        return 0;
 915}
 916
 917/*
 918 * Set the maximum inode count for this filesystem
 919 */
 920STATIC void
 921xfs_set_maxicount(xfs_mount_t *mp)
 922{
 923        xfs_sb_t        *sbp = &(mp->m_sb);
 924        __uint64_t      icount;
 925
 926        if (sbp->sb_imax_pct) {
 927                /*
 928                 * Make sure the maximum inode count is a multiple
 929                 * of the units we allocate inodes in.
 930                 */
 931                icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 932                do_div(icount, 100);
 933                do_div(icount, mp->m_ialloc_blks);
 934                mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
 935                                   sbp->sb_inopblog;
 936        } else {
 937                mp->m_maxicount = 0;
 938        }
 939}
 940
 941/*
 942 * Set the default minimum read and write sizes unless
 943 * already specified in a mount option.
 944 * We use smaller I/O sizes when the file system
 945 * is being used for NFS service (wsync mount option).
 946 */
 947STATIC void
 948xfs_set_rw_sizes(xfs_mount_t *mp)
 949{
 950        xfs_sb_t        *sbp = &(mp->m_sb);
 951        int             readio_log, writeio_log;
 952
 953        if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
 954                if (mp->m_flags & XFS_MOUNT_WSYNC) {
 955                        readio_log = XFS_WSYNC_READIO_LOG;
 956                        writeio_log = XFS_WSYNC_WRITEIO_LOG;
 957                } else {
 958                        readio_log = XFS_READIO_LOG_LARGE;
 959                        writeio_log = XFS_WRITEIO_LOG_LARGE;
 960                }
 961        } else {
 962                readio_log = mp->m_readio_log;
 963                writeio_log = mp->m_writeio_log;
 964        }
 965
 966        if (sbp->sb_blocklog > readio_log) {
 967                mp->m_readio_log = sbp->sb_blocklog;
 968        } else {
 969                mp->m_readio_log = readio_log;
 970        }
 971        mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
 972        if (sbp->sb_blocklog > writeio_log) {
 973                mp->m_writeio_log = sbp->sb_blocklog;
 974        } else {
 975                mp->m_writeio_log = writeio_log;
 976        }
 977        mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
 978}
 979
 980/*
 981 * precalculate the low space thresholds for dynamic speculative preallocation.
 982 */
 983void
 984xfs_set_low_space_thresholds(
 985        struct xfs_mount        *mp)
 986{
 987        int i;
 988
 989        for (i = 0; i < XFS_LOWSP_MAX; i++) {
 990                __uint64_t space = mp->m_sb.sb_dblocks;
 991
 992                do_div(space, 100);
 993                mp->m_low_space[i] = space * (i + 1);
 994        }
 995}
 996
 997
 998/*
 999 * Set whether we're using inode alignment.
1000 */
1001STATIC void
1002xfs_set_inoalignment(xfs_mount_t *mp)
1003{
1004        if (xfs_sb_version_hasalign(&mp->m_sb) &&
1005            mp->m_sb.sb_inoalignmt >=
1006            XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1007                mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1008        else
1009                mp->m_inoalign_mask = 0;
1010        /*
1011         * If we are using stripe alignment, check whether
1012         * the stripe unit is a multiple of the inode alignment
1013         */
1014        if (mp->m_dalign && mp->m_inoalign_mask &&
1015            !(mp->m_dalign & mp->m_inoalign_mask))
1016                mp->m_sinoalign = mp->m_dalign;
1017        else
1018                mp->m_sinoalign = 0;
1019}
1020
1021/*
1022 * Check that the data (and log if separate) are an ok size.
1023 */
1024STATIC int
1025xfs_check_sizes(xfs_mount_t *mp)
1026{
1027        xfs_buf_t       *bp;
1028        xfs_daddr_t     d;
1029
1030        d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1031        if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1032                xfs_warn(mp, "filesystem size mismatch detected");
1033                return XFS_ERROR(EFBIG);
1034        }
1035        bp = xfs_buf_read_uncached(mp->m_ddev_targp,
1036                                        d - XFS_FSS_TO_BB(mp, 1),
1037                                        XFS_FSS_TO_BB(mp, 1), 0);
1038        if (!bp) {
1039                xfs_warn(mp, "last sector read failed");
1040                return EIO;
1041        }
1042        xfs_buf_relse(bp);
1043
1044        if (mp->m_logdev_targp != mp->m_ddev_targp) {
1045                d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1046                if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1047                        xfs_warn(mp, "log size mismatch detected");
1048                        return XFS_ERROR(EFBIG);
1049                }
1050                bp = xfs_buf_read_uncached(mp->m_logdev_targp,
1051                                        d - XFS_FSB_TO_BB(mp, 1),
1052                                        XFS_FSB_TO_BB(mp, 1), 0);
1053                if (!bp) {
1054                        xfs_warn(mp, "log device read failed");
1055                        return EIO;
1056                }
1057                xfs_buf_relse(bp);
1058        }
1059        return 0;
1060}
1061
1062/*
1063 * Clear the quotaflags in memory and in the superblock.
1064 */
1065int
1066xfs_mount_reset_sbqflags(
1067        struct xfs_mount        *mp)
1068{
1069        int                     error;
1070        struct xfs_trans        *tp;
1071
1072        mp->m_qflags = 0;
1073
1074        /*
1075         * It is OK to look at sb_qflags here in mount path,
1076         * without m_sb_lock.
1077         */
1078        if (mp->m_sb.sb_qflags == 0)
1079                return 0;
1080        spin_lock(&mp->m_sb_lock);
1081        mp->m_sb.sb_qflags = 0;
1082        spin_unlock(&mp->m_sb_lock);
1083
1084        /*
1085         * If the fs is readonly, let the incore superblock run
1086         * with quotas off but don't flush the update out to disk
1087         */
1088        if (mp->m_flags & XFS_MOUNT_RDONLY)
1089                return 0;
1090
1091        tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1092        error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1093                                      XFS_DEFAULT_LOG_COUNT);
1094        if (error) {
1095                xfs_trans_cancel(tp, 0);
1096                xfs_alert(mp, "%s: Superblock update failed!", __func__);
1097                return error;
1098        }
1099
1100        xfs_mod_sb(tp, XFS_SB_QFLAGS);
1101        return xfs_trans_commit(tp, 0);
1102}
1103
1104__uint64_t
1105xfs_default_resblks(xfs_mount_t *mp)
1106{
1107        __uint64_t resblks;
1108
1109        /*
1110         * We default to 5% or 8192 fsbs of space reserved, whichever is
1111         * smaller.  This is intended to cover concurrent allocation
1112         * transactions when we initially hit enospc. These each require a 4
1113         * block reservation. Hence by default we cover roughly 2000 concurrent
1114         * allocation reservations.
1115         */
1116        resblks = mp->m_sb.sb_dblocks;
1117        do_div(resblks, 20);
1118        resblks = min_t(__uint64_t, resblks, 8192);
1119        return resblks;
1120}
1121
1122/*
1123 * This function does the following on an initial mount of a file system:
1124 *      - reads the superblock from disk and init the mount struct
1125 *      - if we're a 32-bit kernel, do a size check on the superblock
1126 *              so we don't mount terabyte filesystems
1127 *      - init mount struct realtime fields
1128 *      - allocate inode hash table for fs
1129 *      - init directory manager
1130 *      - perform recovery and init the log manager
1131 */
1132int
1133xfs_mountfs(
1134        xfs_mount_t     *mp)
1135{
1136        xfs_sb_t        *sbp = &(mp->m_sb);
1137        xfs_inode_t     *rip;
1138        __uint64_t      resblks;
1139        uint            quotamount = 0;
1140        uint            quotaflags = 0;
1141        int             error = 0;
1142
1143        xfs_mount_common(mp, sbp);
1144
1145        /*
1146         * Check for a mismatched features2 values.  Older kernels
1147         * read & wrote into the wrong sb offset for sb_features2
1148         * on some platforms due to xfs_sb_t not being 64bit size aligned
1149         * when sb_features2 was added, which made older superblock
1150         * reading/writing routines swap it as a 64-bit value.
1151         *
1152         * For backwards compatibility, we make both slots equal.
1153         *
1154         * If we detect a mismatched field, we OR the set bits into the
1155         * existing features2 field in case it has already been modified; we
1156         * don't want to lose any features.  We then update the bad location
1157         * with the ORed value so that older kernels will see any features2
1158         * flags, and mark the two fields as needing updates once the
1159         * transaction subsystem is online.
1160         */
1161        if (xfs_sb_has_mismatched_features2(sbp)) {
1162                xfs_warn(mp, "correcting sb_features alignment problem");
1163                sbp->sb_features2 |= sbp->sb_bad_features2;
1164                sbp->sb_bad_features2 = sbp->sb_features2;
1165                mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1166
1167                /*
1168                 * Re-check for ATTR2 in case it was found in bad_features2
1169                 * slot.
1170                 */
1171                if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1172                   !(mp->m_flags & XFS_MOUNT_NOATTR2))
1173                        mp->m_flags |= XFS_MOUNT_ATTR2;
1174        }
1175
1176        if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1177           (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1178                xfs_sb_version_removeattr2(&mp->m_sb);
1179                mp->m_update_flags |= XFS_SB_FEATURES2;
1180
1181                /* update sb_versionnum for the clearing of the morebits */
1182                if (!sbp->sb_features2)
1183                        mp->m_update_flags |= XFS_SB_VERSIONNUM;
1184        }
1185
1186        /*
1187         * Check if sb_agblocks is aligned at stripe boundary
1188         * If sb_agblocks is NOT aligned turn off m_dalign since
1189         * allocator alignment is within an ag, therefore ag has
1190         * to be aligned at stripe boundary.
1191         */
1192        error = xfs_update_alignment(mp);
1193        if (error)
1194                goto out;
1195
1196        xfs_alloc_compute_maxlevels(mp);
1197        xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1198        xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1199        xfs_ialloc_compute_maxlevels(mp);
1200
1201        xfs_set_maxicount(mp);
1202
1203        error = xfs_uuid_mount(mp);
1204        if (error)
1205                goto out;
1206
1207        /*
1208         * Set the minimum read and write sizes
1209         */
1210        xfs_set_rw_sizes(mp);
1211
1212        /* set the low space thresholds for dynamic preallocation */
1213        xfs_set_low_space_thresholds(mp);
1214
1215        /*
1216         * Set the inode cluster size.
1217         * This may still be overridden by the file system
1218         * block size if it is larger than the chosen cluster size.
1219         */
1220        mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1221
1222        /*
1223         * Set inode alignment fields
1224         */
1225        xfs_set_inoalignment(mp);
1226
1227        /*
1228         * Check that the data (and log if separate) are an ok size.
1229         */
1230        error = xfs_check_sizes(mp);
1231        if (error)
1232                goto out_remove_uuid;
1233
1234        /*
1235         * Initialize realtime fields in the mount structure
1236         */
1237        error = xfs_rtmount_init(mp);
1238        if (error) {
1239                xfs_warn(mp, "RT mount failed");
1240                goto out_remove_uuid;
1241        }
1242
1243        /*
1244         *  Copies the low order bits of the timestamp and the randomly
1245         *  set "sequence" number out of a UUID.
1246         */
1247        uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1248
1249        mp->m_dmevmask = 0;     /* not persistent; set after each mount */
1250
1251        xfs_dir_mount(mp);
1252
1253        /*
1254         * Initialize the attribute manager's entries.
1255         */
1256        mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1257
1258        /*
1259         * Initialize the precomputed transaction reservations values.
1260         */
1261        xfs_trans_init(mp);
1262
1263        /*
1264         * Allocate and initialize the per-ag data.
1265         */
1266        spin_lock_init(&mp->m_perag_lock);
1267        INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1268        error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1269        if (error) {
1270                xfs_warn(mp, "Failed per-ag init: %d", error);
1271                goto out_remove_uuid;
1272        }
1273
1274        if (!sbp->sb_logblocks) {
1275                xfs_warn(mp, "no log defined");
1276                XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1277                error = XFS_ERROR(EFSCORRUPTED);
1278                goto out_free_perag;
1279        }
1280
1281        /*
1282         * log's mount-time initialization. Perform 1st part recovery if needed
1283         */
1284        error = xfs_log_mount(mp, mp->m_logdev_targp,
1285                              XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1286                              XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1287        if (error) {
1288                xfs_warn(mp, "log mount failed");
1289                goto out_fail_wait;
1290        }
1291
1292        /*
1293         * Now the log is mounted, we know if it was an unclean shutdown or
1294         * not. If it was, with the first phase of recovery has completed, we
1295         * have consistent AG blocks on disk. We have not recovered EFIs yet,
1296         * but they are recovered transactionally in the second recovery phase
1297         * later.
1298         *
1299         * Hence we can safely re-initialise incore superblock counters from
1300         * the per-ag data. These may not be correct if the filesystem was not
1301         * cleanly unmounted, so we need to wait for recovery to finish before
1302         * doing this.
1303         *
1304         * If the filesystem was cleanly unmounted, then we can trust the
1305         * values in the superblock to be correct and we don't need to do
1306         * anything here.
1307         *
1308         * If we are currently making the filesystem, the initialisation will
1309         * fail as the perag data is in an undefined state.
1310         */
1311        if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1312            !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1313             !mp->m_sb.sb_inprogress) {
1314                error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1315                if (error)
1316                        goto out_fail_wait;
1317        }
1318
1319        /*
1320         * Get and sanity-check the root inode.
1321         * Save the pointer to it in the mount structure.
1322         */
1323        error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1324        if (error) {
1325                xfs_warn(mp, "failed to read root inode");
1326                goto out_log_dealloc;
1327        }
1328
1329        ASSERT(rip != NULL);
1330
1331        if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
1332                xfs_warn(mp, "corrupted root inode %llu: not a directory",
1333                        (unsigned long long)rip->i_ino);
1334                xfs_iunlock(rip, XFS_ILOCK_EXCL);
1335                XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1336                                 mp);
1337                error = XFS_ERROR(EFSCORRUPTED);
1338                goto out_rele_rip;
1339        }
1340        mp->m_rootip = rip;     /* save it */
1341
1342        xfs_iunlock(rip, XFS_ILOCK_EXCL);
1343
1344        /*
1345         * Initialize realtime inode pointers in the mount structure
1346         */
1347        error = xfs_rtmount_inodes(mp);
1348        if (error) {
1349                /*
1350                 * Free up the root inode.
1351                 */
1352                xfs_warn(mp, "failed to read RT inodes");
1353                goto out_rele_rip;
1354        }
1355
1356        /*
1357         * If this is a read-only mount defer the superblock updates until
1358         * the next remount into writeable mode.  Otherwise we would never
1359         * perform the update e.g. for the root filesystem.
1360         */
1361        if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1362                error = xfs_mount_log_sb(mp, mp->m_update_flags);
1363                if (error) {
1364                        xfs_warn(mp, "failed to write sb changes");
1365                        goto out_rtunmount;
1366                }
1367        }
1368
1369        /*
1370         * Initialise the XFS quota management subsystem for this mount
1371         */
1372        if (XFS_IS_QUOTA_RUNNING(mp)) {
1373                error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1374                if (error)
1375                        goto out_rtunmount;
1376        } else {
1377                ASSERT(!XFS_IS_QUOTA_ON(mp));
1378
1379                /*
1380                 * If a file system had quotas running earlier, but decided to
1381                 * mount without -o uquota/pquota/gquota options, revoke the
1382                 * quotachecked license.
1383                 */
1384                if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1385                        xfs_notice(mp, "resetting quota flags");
1386                        error = xfs_mount_reset_sbqflags(mp);
1387                        if (error)
1388                                return error;
1389                }
1390        }
1391
1392        /*
1393         * Finish recovering the file system.  This part needed to be
1394         * delayed until after the root and real-time bitmap inodes
1395         * were consistently read in.
1396         */
1397        error = xfs_log_mount_finish(mp);
1398        if (error) {
1399                xfs_warn(mp, "log mount finish failed");
1400                goto out_rtunmount;
1401        }
1402
1403        /*
1404         * Complete the quota initialisation, post-log-replay component.
1405         */
1406        if (quotamount) {
1407                ASSERT(mp->m_qflags == 0);
1408                mp->m_qflags = quotaflags;
1409
1410                xfs_qm_mount_quotas(mp);
1411        }
1412
1413        /*
1414         * Now we are mounted, reserve a small amount of unused space for
1415         * privileged transactions. This is needed so that transaction
1416         * space required for critical operations can dip into this pool
1417         * when at ENOSPC. This is needed for operations like create with
1418         * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1419         * are not allowed to use this reserved space.
1420         *
1421         * This may drive us straight to ENOSPC on mount, but that implies
1422         * we were already there on the last unmount. Warn if this occurs.
1423         */
1424        if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1425                resblks = xfs_default_resblks(mp);
1426                error = xfs_reserve_blocks(mp, &resblks, NULL);
1427                if (error)
1428                        xfs_warn(mp,
1429        "Unable to allocate reserve blocks. Continuing without reserve pool.");
1430        }
1431
1432        return 0;
1433
1434 out_rtunmount:
1435        xfs_rtunmount_inodes(mp);
1436 out_rele_rip:
1437        IRELE(rip);
1438 out_log_dealloc:
1439        xfs_log_unmount(mp);
1440 out_fail_wait:
1441        if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1442                xfs_wait_buftarg(mp->m_logdev_targp);
1443        xfs_wait_buftarg(mp->m_ddev_targp);
1444 out_free_perag:
1445        xfs_free_perag(mp);
1446 out_remove_uuid:
1447        xfs_uuid_unmount(mp);
1448 out:
1449        return error;
1450}
1451
1452/*
1453 * This flushes out the inodes,dquots and the superblock, unmounts the
1454 * log and makes sure that incore structures are freed.
1455 */
1456void
1457xfs_unmountfs(
1458        struct xfs_mount        *mp)
1459{
1460        __uint64_t              resblks;
1461        int                     error;
1462
1463        xfs_qm_unmount_quotas(mp);
1464        xfs_rtunmount_inodes(mp);
1465        IRELE(mp->m_rootip);
1466
1467        /*
1468         * We can potentially deadlock here if we have an inode cluster
1469         * that has been freed has its buffer still pinned in memory because
1470         * the transaction is still sitting in a iclog. The stale inodes
1471         * on that buffer will have their flush locks held until the
1472         * transaction hits the disk and the callbacks run. the inode
1473         * flush takes the flush lock unconditionally and with nothing to
1474         * push out the iclog we will never get that unlocked. hence we
1475         * need to force the log first.
1476         */
1477        xfs_log_force(mp, XFS_LOG_SYNC);
1478
1479        /*
1480         * Flush all pending changes from the AIL.
1481         */
1482        xfs_ail_push_all_sync(mp->m_ail);
1483
1484        /*
1485         * And reclaim all inodes.  At this point there should be no dirty
1486         * inode, and none should be pinned or locked, but use synchronous
1487         * reclaim just to be sure.
1488         */
1489        xfs_reclaim_inodes(mp, SYNC_WAIT);
1490
1491        xfs_qm_unmount(mp);
1492
1493        /*
1494         * Flush out the log synchronously so that we know for sure
1495         * that nothing is pinned.  This is important because bflush()
1496         * will skip pinned buffers.
1497         */
1498        xfs_log_force(mp, XFS_LOG_SYNC);
1499
1500        /*
1501         * Unreserve any blocks we have so that when we unmount we don't account
1502         * the reserved free space as used. This is really only necessary for
1503         * lazy superblock counting because it trusts the incore superblock
1504         * counters to be absolutely correct on clean unmount.
1505         *
1506         * We don't bother correcting this elsewhere for lazy superblock
1507         * counting because on mount of an unclean filesystem we reconstruct the
1508         * correct counter value and this is irrelevant.
1509         *
1510         * For non-lazy counter filesystems, this doesn't matter at all because
1511         * we only every apply deltas to the superblock and hence the incore
1512         * value does not matter....
1513         */
1514        resblks = 0;
1515        error = xfs_reserve_blocks(mp, &resblks, NULL);
1516        if (error)
1517                xfs_warn(mp, "Unable to free reserved block pool. "
1518                                "Freespace may not be correct on next mount.");
1519
1520        error = xfs_log_sbcount(mp);
1521        if (error)
1522                xfs_warn(mp, "Unable to update superblock counters. "
1523                                "Freespace may not be correct on next mount.");
1524
1525        /*
1526         * At this point we might have modified the superblock again and thus
1527         * added an item to the AIL, thus flush it again.
1528         */
1529        xfs_ail_push_all_sync(mp->m_ail);
1530        xfs_wait_buftarg(mp->m_ddev_targp);
1531
1532        /*
1533         * The superblock buffer is uncached and xfsaild_push() will lock and
1534         * set the XBF_ASYNC flag on the buffer. We cannot do xfs_buf_iowait()
1535         * here but a lock on the superblock buffer will block until iodone()
1536         * has completed.
1537         */
1538        xfs_buf_lock(mp->m_sb_bp);
1539        xfs_buf_unlock(mp->m_sb_bp);
1540
1541        xfs_log_unmount_write(mp);
1542        xfs_log_unmount(mp);
1543        xfs_uuid_unmount(mp);
1544
1545#if defined(DEBUG)
1546        xfs_errortag_clearall(mp, 0);
1547#endif
1548        xfs_free_perag(mp);
1549}
1550
1551int
1552xfs_fs_writable(xfs_mount_t *mp)
1553{
1554        return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
1555                (mp->m_flags & XFS_MOUNT_RDONLY));
1556}
1557
1558/*
1559 * xfs_log_sbcount
1560 *
1561 * Sync the superblock counters to disk.
1562 *
1563 * Note this code can be called during the process of freezing, so
1564 * we may need to use the transaction allocator which does not
1565 * block when the transaction subsystem is in its frozen state.
1566 */
1567int
1568xfs_log_sbcount(xfs_mount_t *mp)
1569{
1570        xfs_trans_t     *tp;
1571        int             error;
1572
1573        if (!xfs_fs_writable(mp))
1574                return 0;
1575
1576        xfs_icsb_sync_counters(mp, 0);
1577
1578        /*
1579         * we don't need to do this if we are updating the superblock
1580         * counters on every modification.
1581         */
1582        if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1583                return 0;
1584
1585        tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1586        error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1587                                        XFS_DEFAULT_LOG_COUNT);
1588        if (error) {
1589                xfs_trans_cancel(tp, 0);
1590                return error;
1591        }
1592
1593        xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1594        xfs_trans_set_sync(tp);
1595        error = xfs_trans_commit(tp, 0);
1596        return error;
1597}
1598
1599/*
1600 * xfs_mod_sb() can be used to copy arbitrary changes to the
1601 * in-core superblock into the superblock buffer to be logged.
1602 * It does not provide the higher level of locking that is
1603 * needed to protect the in-core superblock from concurrent
1604 * access.
1605 */
1606void
1607xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1608{
1609        xfs_buf_t       *bp;
1610        int             first;
1611        int             last;
1612        xfs_mount_t     *mp;
1613        xfs_sb_field_t  f;
1614
1615        ASSERT(fields);
1616        if (!fields)
1617                return;
1618        mp = tp->t_mountp;
1619        bp = xfs_trans_getsb(tp, mp, 0);
1620        first = sizeof(xfs_sb_t);
1621        last = 0;
1622
1623        /* translate/copy */
1624
1625        xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1626
1627        /* find modified range */
1628        f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1629        ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1630        last = xfs_sb_info[f + 1].offset - 1;
1631
1632        f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1633        ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1634        first = xfs_sb_info[f].offset;
1635
1636        xfs_trans_log_buf(tp, bp, first, last);
1637}
1638
1639
1640/*
1641 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1642 * a delta to a specified field in the in-core superblock.  Simply
1643 * switch on the field indicated and apply the delta to that field.
1644 * Fields are not allowed to dip below zero, so if the delta would
1645 * do this do not apply it and return EINVAL.
1646 *
1647 * The m_sb_lock must be held when this routine is called.
1648 */
1649STATIC int
1650xfs_mod_incore_sb_unlocked(
1651        xfs_mount_t     *mp,
1652        xfs_sb_field_t  field,
1653        int64_t         delta,
1654        int             rsvd)
1655{
1656        int             scounter;       /* short counter for 32 bit fields */
1657        long long       lcounter;       /* long counter for 64 bit fields */
1658        long long       res_used, rem;
1659
1660        /*
1661         * With the in-core superblock spin lock held, switch
1662         * on the indicated field.  Apply the delta to the
1663         * proper field.  If the fields value would dip below
1664         * 0, then do not apply the delta and return EINVAL.
1665         */
1666        switch (field) {
1667        case XFS_SBS_ICOUNT:
1668                lcounter = (long long)mp->m_sb.sb_icount;
1669                lcounter += delta;
1670                if (lcounter < 0) {
1671                        ASSERT(0);
1672                        return XFS_ERROR(EINVAL);
1673                }
1674                mp->m_sb.sb_icount = lcounter;
1675                return 0;
1676        case XFS_SBS_IFREE:
1677                lcounter = (long long)mp->m_sb.sb_ifree;
1678                lcounter += delta;
1679                if (lcounter < 0) {
1680                        ASSERT(0);
1681                        return XFS_ERROR(EINVAL);
1682                }
1683                mp->m_sb.sb_ifree = lcounter;
1684                return 0;
1685        case XFS_SBS_FDBLOCKS:
1686                lcounter = (long long)
1687                        mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1688                res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1689
1690                if (delta > 0) {                /* Putting blocks back */
1691                        if (res_used > delta) {
1692                                mp->m_resblks_avail += delta;
1693                        } else {
1694                                rem = delta - res_used;
1695                                mp->m_resblks_avail = mp->m_resblks;
1696                                lcounter += rem;
1697                        }
1698                } else {                                /* Taking blocks away */
1699                        lcounter += delta;
1700                        if (lcounter >= 0) {
1701                                mp->m_sb.sb_fdblocks = lcounter +
1702                                                        XFS_ALLOC_SET_ASIDE(mp);
1703                                return 0;
1704                        }
1705
1706                        /*
1707                         * We are out of blocks, use any available reserved
1708                         * blocks if were allowed to.
1709                         */
1710                        if (!rsvd)
1711                                return XFS_ERROR(ENOSPC);
1712
1713                        lcounter = (long long)mp->m_resblks_avail + delta;
1714                        if (lcounter >= 0) {
1715                                mp->m_resblks_avail = lcounter;
1716                                return 0;
1717                        }
1718                        printk_once(KERN_WARNING
1719                                "Filesystem \"%s\": reserve blocks depleted! "
1720                                "Consider increasing reserve pool size.",
1721                                mp->m_fsname);
1722                        return XFS_ERROR(ENOSPC);
1723                }
1724
1725                mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1726                return 0;
1727        case XFS_SBS_FREXTENTS:
1728                lcounter = (long long)mp->m_sb.sb_frextents;
1729                lcounter += delta;
1730                if (lcounter < 0) {
1731                        return XFS_ERROR(ENOSPC);
1732                }
1733                mp->m_sb.sb_frextents = lcounter;
1734                return 0;
1735        case XFS_SBS_DBLOCKS:
1736                lcounter = (long long)mp->m_sb.sb_dblocks;
1737                lcounter += delta;
1738                if (lcounter < 0) {
1739                        ASSERT(0);
1740                        return XFS_ERROR(EINVAL);
1741                }
1742                mp->m_sb.sb_dblocks = lcounter;
1743                return 0;
1744        case XFS_SBS_AGCOUNT:
1745                scounter = mp->m_sb.sb_agcount;
1746                scounter += delta;
1747                if (scounter < 0) {
1748                        ASSERT(0);
1749                        return XFS_ERROR(EINVAL);
1750                }
1751                mp->m_sb.sb_agcount = scounter;
1752                return 0;
1753        case XFS_SBS_IMAX_PCT:
1754                scounter = mp->m_sb.sb_imax_pct;
1755                scounter += delta;
1756                if (scounter < 0) {
1757                        ASSERT(0);
1758                        return XFS_ERROR(EINVAL);
1759                }
1760                mp->m_sb.sb_imax_pct = scounter;
1761                return 0;
1762        case XFS_SBS_REXTSIZE:
1763                scounter = mp->m_sb.sb_rextsize;
1764                scounter += delta;
1765                if (scounter < 0) {
1766                        ASSERT(0);
1767                        return XFS_ERROR(EINVAL);
1768                }
1769                mp->m_sb.sb_rextsize = scounter;
1770                return 0;
1771        case XFS_SBS_RBMBLOCKS:
1772                scounter = mp->m_sb.sb_rbmblocks;
1773                scounter += delta;
1774                if (scounter < 0) {
1775                        ASSERT(0);
1776                        return XFS_ERROR(EINVAL);
1777                }
1778                mp->m_sb.sb_rbmblocks = scounter;
1779                return 0;
1780        case XFS_SBS_RBLOCKS:
1781                lcounter = (long long)mp->m_sb.sb_rblocks;
1782                lcounter += delta;
1783                if (lcounter < 0) {
1784                        ASSERT(0);
1785                        return XFS_ERROR(EINVAL);
1786                }
1787                mp->m_sb.sb_rblocks = lcounter;
1788                return 0;
1789        case XFS_SBS_REXTENTS:
1790                lcounter = (long long)mp->m_sb.sb_rextents;
1791                lcounter += delta;
1792                if (lcounter < 0) {
1793                        ASSERT(0);
1794                        return XFS_ERROR(EINVAL);
1795                }
1796                mp->m_sb.sb_rextents = lcounter;
1797                return 0;
1798        case XFS_SBS_REXTSLOG:
1799                scounter = mp->m_sb.sb_rextslog;
1800                scounter += delta;
1801                if (scounter < 0) {
1802                        ASSERT(0);
1803                        return XFS_ERROR(EINVAL);
1804                }
1805                mp->m_sb.sb_rextslog = scounter;
1806                return 0;
1807        default:
1808                ASSERT(0);
1809                return XFS_ERROR(EINVAL);
1810        }
1811}
1812
1813/*
1814 * xfs_mod_incore_sb() is used to change a field in the in-core
1815 * superblock structure by the specified delta.  This modification
1816 * is protected by the m_sb_lock.  Just use the xfs_mod_incore_sb_unlocked()
1817 * routine to do the work.
1818 */
1819int
1820xfs_mod_incore_sb(
1821        struct xfs_mount        *mp,
1822        xfs_sb_field_t          field,
1823        int64_t                 delta,
1824        int                     rsvd)
1825{
1826        int                     status;
1827
1828#ifdef HAVE_PERCPU_SB
1829        ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1830#endif
1831        spin_lock(&mp->m_sb_lock);
1832        status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1833        spin_unlock(&mp->m_sb_lock);
1834
1835        return status;
1836}
1837
1838/*
1839 * Change more than one field in the in-core superblock structure at a time.
1840 *
1841 * The fields and changes to those fields are specified in the array of
1842 * xfs_mod_sb structures passed in.  Either all of the specified deltas
1843 * will be applied or none of them will.  If any modified field dips below 0,
1844 * then all modifications will be backed out and EINVAL will be returned.
1845 *
1846 * Note that this function may not be used for the superblock values that
1847 * are tracked with the in-memory per-cpu counters - a direct call to
1848 * xfs_icsb_modify_counters is required for these.
1849 */
1850int
1851xfs_mod_incore_sb_batch(
1852        struct xfs_mount        *mp,
1853        xfs_mod_sb_t            *msb,
1854        uint                    nmsb,
1855        int                     rsvd)
1856{
1857        xfs_mod_sb_t            *msbp;
1858        int                     error = 0;
1859
1860        /*
1861         * Loop through the array of mod structures and apply each individually.
1862         * If any fail, then back out all those which have already been applied.
1863         * Do all of this within the scope of the m_sb_lock so that all of the
1864         * changes will be atomic.
1865         */
1866        spin_lock(&mp->m_sb_lock);
1867        for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1868                ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1869                       msbp->msb_field > XFS_SBS_FDBLOCKS);
1870
1871                error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1872                                                   msbp->msb_delta, rsvd);
1873                if (error)
1874                        goto unwind;
1875        }
1876        spin_unlock(&mp->m_sb_lock);
1877        return 0;
1878
1879unwind:
1880        while (--msbp >= msb) {
1881                error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1882                                                   -msbp->msb_delta, rsvd);
1883                ASSERT(error == 0);
1884        }
1885        spin_unlock(&mp->m_sb_lock);
1886        return error;
1887}
1888
1889/*
1890 * xfs_getsb() is called to obtain the buffer for the superblock.
1891 * The buffer is returned locked and read in from disk.
1892 * The buffer should be released with a call to xfs_brelse().
1893 *
1894 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1895 * the superblock buffer if it can be locked without sleeping.
1896 * If it can't then we'll return NULL.
1897 */
1898struct xfs_buf *
1899xfs_getsb(
1900        struct xfs_mount        *mp,
1901        int                     flags)
1902{
1903        struct xfs_buf          *bp = mp->m_sb_bp;
1904
1905        if (!xfs_buf_trylock(bp)) {
1906                if (flags & XBF_TRYLOCK)
1907                        return NULL;
1908                xfs_buf_lock(bp);
1909        }
1910
1911        xfs_buf_hold(bp);
1912        ASSERT(XFS_BUF_ISDONE(bp));
1913        return bp;
1914}
1915
1916/*
1917 * Used to free the superblock along various error paths.
1918 */
1919void
1920xfs_freesb(
1921        struct xfs_mount        *mp)
1922{
1923        struct xfs_buf          *bp = mp->m_sb_bp;
1924
1925        xfs_buf_lock(bp);
1926        mp->m_sb_bp = NULL;
1927        xfs_buf_relse(bp);
1928}
1929
1930/*
1931 * Used to log changes to the superblock unit and width fields which could
1932 * be altered by the mount options, as well as any potential sb_features2
1933 * fixup. Only the first superblock is updated.
1934 */
1935int
1936xfs_mount_log_sb(
1937        xfs_mount_t     *mp,
1938        __int64_t       fields)
1939{
1940        xfs_trans_t     *tp;
1941        int             error;
1942
1943        ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1944                         XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1945                         XFS_SB_VERSIONNUM));
1946
1947        tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1948        error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1949                                XFS_DEFAULT_LOG_COUNT);
1950        if (error) {
1951                xfs_trans_cancel(tp, 0);
1952                return error;
1953        }
1954        xfs_mod_sb(tp, fields);
1955        error = xfs_trans_commit(tp, 0);
1956        return error;
1957}
1958
1959/*
1960 * If the underlying (data/log/rt) device is readonly, there are some
1961 * operations that cannot proceed.
1962 */
1963int
1964xfs_dev_is_read_only(
1965        struct xfs_mount        *mp,
1966        char                    *message)
1967{
1968        if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1969            xfs_readonly_buftarg(mp->m_logdev_targp) ||
1970            (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1971                xfs_notice(mp, "%s required on read-only device.", message);
1972                xfs_notice(mp, "write access unavailable, cannot proceed.");
1973                return EROFS;
1974        }
1975        return 0;
1976}
1977
1978#ifdef HAVE_PERCPU_SB
1979/*
1980 * Per-cpu incore superblock counters
1981 *
1982 * Simple concept, difficult implementation
1983 *
1984 * Basically, replace the incore superblock counters with a distributed per cpu
1985 * counter for contended fields (e.g.  free block count).
1986 *
1987 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1988 * hence needs to be accurately read when we are running low on space. Hence
1989 * there is a method to enable and disable the per-cpu counters based on how
1990 * much "stuff" is available in them.
1991 *
1992 * Basically, a counter is enabled if there is enough free resource to justify
1993 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1994 * ENOSPC), then we disable the counters to synchronise all callers and
1995 * re-distribute the available resources.
1996 *
1997 * If, once we redistributed the available resources, we still get a failure,
1998 * we disable the per-cpu counter and go through the slow path.
1999 *
2000 * The slow path is the current xfs_mod_incore_sb() function.  This means that
2001 * when we disable a per-cpu counter, we need to drain its resources back to
2002 * the global superblock. We do this after disabling the counter to prevent
2003 * more threads from queueing up on the counter.
2004 *
2005 * Essentially, this means that we still need a lock in the fast path to enable
2006 * synchronisation between the global counters and the per-cpu counters. This
2007 * is not a problem because the lock will be local to a CPU almost all the time
2008 * and have little contention except when we get to ENOSPC conditions.
2009 *
2010 * Basically, this lock becomes a barrier that enables us to lock out the fast
2011 * path while we do things like enabling and disabling counters and
2012 * synchronising the counters.
2013 *
2014 * Locking rules:
2015 *
2016 *      1. m_sb_lock before picking up per-cpu locks
2017 *      2. per-cpu locks always picked up via for_each_online_cpu() order
2018 *      3. accurate counter sync requires m_sb_lock + per cpu locks
2019 *      4. modifying per-cpu counters requires holding per-cpu lock
2020 *      5. modifying global counters requires holding m_sb_lock
2021 *      6. enabling or disabling a counter requires holding the m_sb_lock 
2022 *         and _none_ of the per-cpu locks.
2023 *
2024 * Disabled counters are only ever re-enabled by a balance operation
2025 * that results in more free resources per CPU than a given threshold.
2026 * To ensure counters don't remain disabled, they are rebalanced when
2027 * the global resource goes above a higher threshold (i.e. some hysteresis
2028 * is present to prevent thrashing).
2029 */
2030
2031#ifdef CONFIG_HOTPLUG_CPU
2032/*
2033 * hot-plug CPU notifier support.
2034 *
2035 * We need a notifier per filesystem as we need to be able to identify
2036 * the filesystem to balance the counters out. This is achieved by
2037 * having a notifier block embedded in the xfs_mount_t and doing pointer
2038 * magic to get the mount pointer from the notifier block address.
2039 */
2040STATIC int
2041xfs_icsb_cpu_notify(
2042        struct notifier_block *nfb,
2043        unsigned long action,
2044        void *hcpu)
2045{
2046        xfs_icsb_cnts_t *cntp;
2047        xfs_mount_t     *mp;
2048
2049        mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2050        cntp = (xfs_icsb_cnts_t *)
2051                        per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2052        switch (action) {
2053        case CPU_UP_PREPARE:
2054        case CPU_UP_PREPARE_FROZEN:
2055                /* Easy Case - initialize the area and locks, and
2056                 * then rebalance when online does everything else for us. */
2057                memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2058                break;
2059        case CPU_ONLINE:
2060        case CPU_ONLINE_FROZEN:
2061                xfs_icsb_lock(mp);
2062                xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2063                xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2064                xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2065                xfs_icsb_unlock(mp);
2066                break;
2067        case CPU_DEAD:
2068        case CPU_DEAD_FROZEN:
2069                /* Disable all the counters, then fold the dead cpu's
2070                 * count into the total on the global superblock and
2071                 * re-enable the counters. */
2072                xfs_icsb_lock(mp);
2073                spin_lock(&mp->m_sb_lock);
2074                xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2075                xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2076                xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2077
2078                mp->m_sb.sb_icount += cntp->icsb_icount;
2079                mp->m_sb.sb_ifree += cntp->icsb_ifree;
2080                mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2081
2082                memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2083
2084                xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2085                xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2086                xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2087                spin_unlock(&mp->m_sb_lock);
2088                xfs_icsb_unlock(mp);
2089                break;
2090        }
2091
2092        return NOTIFY_OK;
2093}
2094#endif /* CONFIG_HOTPLUG_CPU */
2095
2096int
2097xfs_icsb_init_counters(
2098        xfs_mount_t     *mp)
2099{
2100        xfs_icsb_cnts_t *cntp;
2101        int             i;
2102
2103        mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2104        if (mp->m_sb_cnts == NULL)
2105                return -ENOMEM;
2106
2107#ifdef CONFIG_HOTPLUG_CPU
2108        mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2109        mp->m_icsb_notifier.priority = 0;
2110        register_hotcpu_notifier(&mp->m_icsb_notifier);
2111#endif /* CONFIG_HOTPLUG_CPU */
2112
2113        for_each_online_cpu(i) {
2114                cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2115                memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2116        }
2117
2118        mutex_init(&mp->m_icsb_mutex);
2119
2120        /*
2121         * start with all counters disabled so that the
2122         * initial balance kicks us off correctly
2123         */
2124        mp->m_icsb_counters = -1;
2125        return 0;
2126}
2127
2128void
2129xfs_icsb_reinit_counters(
2130        xfs_mount_t     *mp)
2131{
2132        xfs_icsb_lock(mp);
2133        /*
2134         * start with all counters disabled so that the
2135         * initial balance kicks us off correctly
2136         */
2137        mp->m_icsb_counters = -1;
2138        xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2139        xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2140        xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2141        xfs_icsb_unlock(mp);
2142}
2143
2144void
2145xfs_icsb_destroy_counters(
2146        xfs_mount_t     *mp)
2147{
2148        if (mp->m_sb_cnts) {
2149                unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2150                free_percpu(mp->m_sb_cnts);
2151        }
2152        mutex_destroy(&mp->m_icsb_mutex);
2153}
2154
2155STATIC void
2156xfs_icsb_lock_cntr(
2157        xfs_icsb_cnts_t *icsbp)
2158{
2159        while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2160                ndelay(1000);
2161        }
2162}
2163
2164STATIC void
2165xfs_icsb_unlock_cntr(
2166        xfs_icsb_cnts_t *icsbp)
2167{
2168        clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2169}
2170
2171
2172STATIC void
2173xfs_icsb_lock_all_counters(
2174        xfs_mount_t     *mp)
2175{
2176        xfs_icsb_cnts_t *cntp;
2177        int             i;
2178
2179        for_each_online_cpu(i) {
2180                cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2181                xfs_icsb_lock_cntr(cntp);
2182        }
2183}
2184
2185STATIC void
2186xfs_icsb_unlock_all_counters(
2187        xfs_mount_t     *mp)
2188{
2189        xfs_icsb_cnts_t *cntp;
2190        int             i;
2191
2192        for_each_online_cpu(i) {
2193                cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2194                xfs_icsb_unlock_cntr(cntp);
2195        }
2196}
2197
2198STATIC void
2199xfs_icsb_count(
2200        xfs_mount_t     *mp,
2201        xfs_icsb_cnts_t *cnt,
2202        int             flags)
2203{
2204        xfs_icsb_cnts_t *cntp;
2205        int             i;
2206
2207        memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2208
2209        if (!(flags & XFS_ICSB_LAZY_COUNT))
2210                xfs_icsb_lock_all_counters(mp);
2211
2212        for_each_online_cpu(i) {
2213                cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2214                cnt->icsb_icount += cntp->icsb_icount;
2215                cnt->icsb_ifree += cntp->icsb_ifree;
2216                cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2217        }
2218
2219        if (!(flags & XFS_ICSB_LAZY_COUNT))
2220                xfs_icsb_unlock_all_counters(mp);
2221}
2222
2223STATIC int
2224xfs_icsb_counter_disabled(
2225        xfs_mount_t     *mp,
2226        xfs_sb_field_t  field)
2227{
2228        ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2229        return test_bit(field, &mp->m_icsb_counters);
2230}
2231
2232STATIC void
2233xfs_icsb_disable_counter(
2234        xfs_mount_t     *mp,
2235        xfs_sb_field_t  field)
2236{
2237        xfs_icsb_cnts_t cnt;
2238
2239        ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2240
2241        /*
2242         * If we are already disabled, then there is nothing to do
2243         * here. We check before locking all the counters to avoid
2244         * the expensive lock operation when being called in the
2245         * slow path and the counter is already disabled. This is
2246         * safe because the only time we set or clear this state is under
2247         * the m_icsb_mutex.
2248         */
2249        if (xfs_icsb_counter_disabled(mp, field))
2250                return;
2251
2252        xfs_icsb_lock_all_counters(mp);
2253        if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2254                /* drain back to superblock */
2255
2256                xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2257                switch(field) {
2258                case XFS_SBS_ICOUNT:
2259                        mp->m_sb.sb_icount = cnt.icsb_icount;
2260                        break;
2261                case XFS_SBS_IFREE:
2262                        mp->m_sb.sb_ifree = cnt.icsb_ifree;
2263                        break;
2264                case XFS_SBS_FDBLOCKS:
2265                        mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2266                        break;
2267                default:
2268                        BUG();
2269                }
2270        }
2271
2272        xfs_icsb_unlock_all_counters(mp);
2273}
2274
2275STATIC void
2276xfs_icsb_enable_counter(
2277        xfs_mount_t     *mp,
2278        xfs_sb_field_t  field,
2279        uint64_t        count,
2280        uint64_t        resid)
2281{
2282        xfs_icsb_cnts_t *cntp;
2283        int             i;
2284
2285        ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2286
2287        xfs_icsb_lock_all_counters(mp);
2288        for_each_online_cpu(i) {
2289                cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2290                switch (field) {
2291                case XFS_SBS_ICOUNT:
2292                        cntp->icsb_icount = count + resid;
2293                        break;
2294                case XFS_SBS_IFREE:
2295                        cntp->icsb_ifree = count + resid;
2296                        break;
2297                case XFS_SBS_FDBLOCKS:
2298                        cntp->icsb_fdblocks = count + resid;
2299                        break;
2300                default:
2301                        BUG();
2302                        break;
2303                }
2304                resid = 0;
2305        }
2306        clear_bit(field, &mp->m_icsb_counters);
2307        xfs_icsb_unlock_all_counters(mp);
2308}
2309
2310void
2311xfs_icsb_sync_counters_locked(
2312        xfs_mount_t     *mp,
2313        int             flags)
2314{
2315        xfs_icsb_cnts_t cnt;
2316
2317        xfs_icsb_count(mp, &cnt, flags);
2318
2319        if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2320                mp->m_sb.sb_icount = cnt.icsb_icount;
2321        if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2322                mp->m_sb.sb_ifree = cnt.icsb_ifree;
2323        if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2324                mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2325}
2326
2327/*
2328 * Accurate update of per-cpu counters to incore superblock
2329 */
2330void
2331xfs_icsb_sync_counters(
2332        xfs_mount_t     *mp,
2333        int             flags)
2334{
2335        spin_lock(&mp->m_sb_lock);
2336        xfs_icsb_sync_counters_locked(mp, flags);
2337        spin_unlock(&mp->m_sb_lock);
2338}
2339
2340/*
2341 * Balance and enable/disable counters as necessary.
2342 *
2343 * Thresholds for re-enabling counters are somewhat magic.  inode counts are
2344 * chosen to be the same number as single on disk allocation chunk per CPU, and
2345 * free blocks is something far enough zero that we aren't going thrash when we
2346 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2347 * prevent looping endlessly when xfs_alloc_space asks for more than will
2348 * be distributed to a single CPU but each CPU has enough blocks to be
2349 * reenabled.
2350 *
2351 * Note that we can be called when counters are already disabled.
2352 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2353 * prevent locking every per-cpu counter needlessly.
2354 */
2355
2356#define XFS_ICSB_INO_CNTR_REENABLE      (uint64_t)64
2357#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2358                (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2359STATIC void
2360xfs_icsb_balance_counter_locked(
2361        xfs_mount_t     *mp,
2362        xfs_sb_field_t  field,
2363        int             min_per_cpu)
2364{
2365        uint64_t        count, resid;
2366        int             weight = num_online_cpus();
2367        uint64_t        min = (uint64_t)min_per_cpu;
2368
2369        /* disable counter and sync counter */
2370        xfs_icsb_disable_counter(mp, field);
2371
2372        /* update counters  - first CPU gets residual*/
2373        switch (field) {
2374        case XFS_SBS_ICOUNT:
2375                count = mp->m_sb.sb_icount;
2376                resid = do_div(count, weight);
2377                if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2378                        return;
2379                break;
2380        case XFS_SBS_IFREE:
2381                count = mp->m_sb.sb_ifree;
2382                resid = do_div(count, weight);
2383                if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2384                        return;
2385                break;
2386        case XFS_SBS_FDBLOCKS:
2387                count = mp->m_sb.sb_fdblocks;
2388                resid = do_div(count, weight);
2389                if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2390                        return;
2391                break;
2392        default:
2393                BUG();
2394                count = resid = 0;      /* quiet, gcc */
2395                break;
2396        }
2397
2398        xfs_icsb_enable_counter(mp, field, count, resid);
2399}
2400
2401STATIC void
2402xfs_icsb_balance_counter(
2403        xfs_mount_t     *mp,
2404        xfs_sb_field_t  fields,
2405        int             min_per_cpu)
2406{
2407        spin_lock(&mp->m_sb_lock);
2408        xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2409        spin_unlock(&mp->m_sb_lock);
2410}
2411
2412int
2413xfs_icsb_modify_counters(
2414        xfs_mount_t     *mp,
2415        xfs_sb_field_t  field,
2416        int64_t         delta,
2417        int             rsvd)
2418{
2419        xfs_icsb_cnts_t *icsbp;
2420        long long       lcounter;       /* long counter for 64 bit fields */
2421        int             ret = 0;
2422
2423        might_sleep();
2424again:
2425        preempt_disable();
2426        icsbp = this_cpu_ptr(mp->m_sb_cnts);
2427
2428        /*
2429         * if the counter is disabled, go to slow path
2430         */
2431        if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2432                goto slow_path;
2433        xfs_icsb_lock_cntr(icsbp);
2434        if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2435                xfs_icsb_unlock_cntr(icsbp);
2436                goto slow_path;
2437        }
2438
2439        switch (field) {
2440        case XFS_SBS_ICOUNT:
2441                lcounter = icsbp->icsb_icount;
2442                lcounter += delta;
2443                if (unlikely(lcounter < 0))
2444                        goto balance_counter;
2445                icsbp->icsb_icount = lcounter;
2446                break;
2447
2448        case XFS_SBS_IFREE:
2449                lcounter = icsbp->icsb_ifree;
2450                lcounter += delta;
2451                if (unlikely(lcounter < 0))
2452                        goto balance_counter;
2453                icsbp->icsb_ifree = lcounter;
2454                break;
2455
2456        case XFS_SBS_FDBLOCKS:
2457                BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2458
2459                lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2460                lcounter += delta;
2461                if (unlikely(lcounter < 0))
2462                        goto balance_counter;
2463                icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2464                break;
2465        default:
2466                BUG();
2467                break;
2468        }
2469        xfs_icsb_unlock_cntr(icsbp);
2470        preempt_enable();
2471        return 0;
2472
2473slow_path:
2474        preempt_enable();
2475
2476        /*
2477         * serialise with a mutex so we don't burn lots of cpu on
2478         * the superblock lock. We still need to hold the superblock
2479         * lock, however, when we modify the global structures.
2480         */
2481        xfs_icsb_lock(mp);
2482
2483        /*
2484         * Now running atomically.
2485         *
2486         * If the counter is enabled, someone has beaten us to rebalancing.
2487         * Drop the lock and try again in the fast path....
2488         */
2489        if (!(xfs_icsb_counter_disabled(mp, field))) {
2490                xfs_icsb_unlock(mp);
2491                goto again;
2492        }
2493
2494        /*
2495         * The counter is currently disabled. Because we are
2496         * running atomically here, we know a rebalance cannot
2497         * be in progress. Hence we can go straight to operating
2498         * on the global superblock. We do not call xfs_mod_incore_sb()
2499         * here even though we need to get the m_sb_lock. Doing so
2500         * will cause us to re-enter this function and deadlock.
2501         * Hence we get the m_sb_lock ourselves and then call
2502         * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2503         * directly on the global counters.
2504         */
2505        spin_lock(&mp->m_sb_lock);
2506        ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2507        spin_unlock(&mp->m_sb_lock);
2508
2509        /*
2510         * Now that we've modified the global superblock, we
2511         * may be able to re-enable the distributed counters
2512         * (e.g. lots of space just got freed). After that
2513         * we are done.
2514         */
2515        if (ret != ENOSPC)
2516                xfs_icsb_balance_counter(mp, field, 0);
2517        xfs_icsb_unlock(mp);
2518        return ret;
2519
2520balance_counter:
2521        xfs_icsb_unlock_cntr(icsbp);
2522        preempt_enable();
2523
2524        /*
2525         * We may have multiple threads here if multiple per-cpu
2526         * counters run dry at the same time. This will mean we can
2527         * do more balances than strictly necessary but it is not
2528         * the common slowpath case.
2529         */
2530        xfs_icsb_lock(mp);
2531
2532        /*
2533         * running atomically.
2534         *
2535         * This will leave the counter in the correct state for future
2536         * accesses. After the rebalance, we simply try again and our retry
2537         * will either succeed through the fast path or slow path without
2538         * another balance operation being required.
2539         */
2540        xfs_icsb_balance_counter(mp, field, delta);
2541        xfs_icsb_unlock(mp);
2542        goto again;
2543}
2544
2545#endif
2546
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.