sys/gnu/fs/xfs/xfs_mount.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_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_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_dir_sf.h"
  35 #include "xfs_dir2_sf.h"
  36 #include "xfs_attr_sf.h"
  37 #include "xfs_dinode.h"
  38 #include "xfs_inode.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_ialloc.h"
  41 #include "xfs_alloc.h"
  42 #include "xfs_rtalloc.h"
  43 #include "xfs_bmap.h"
  44 #include "xfs_error.h"
  45 #include "xfs_rw.h"
  46 #include "xfs_quota.h"
  47 #include "xfs_fsops.h"
  48
  49 STATIC void     xfs_mount_log_sbunit(xfs_mount_t *, __int64_t);
  50 STATIC int      xfs_uuid_mount(xfs_mount_t *);
  51 STATIC void     xfs_uuid_unmount(xfs_mount_t *mp);
  52 STATIC void     xfs_unmountfs_wait(xfs_mount_t *);
  53
  54
  55 #ifdef HAVE_PERCPU_SB
  56 STATIC void     xfs_icsb_destroy_counters(xfs_mount_t *);
  57 STATIC void     xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, int);
  58 STATIC void     xfs_icsb_sync_counters(xfs_mount_t *);
  59 STATIC int      xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
  60                                                 int, int);
  61 STATIC int      xfs_icsb_modify_counters_locked(xfs_mount_t *, xfs_sb_field_t,
  62                                                 int, int);
  63 STATIC int      xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
  64
  65 #else
  66
  67 #define xfs_icsb_destroy_counters(mp)                   do { } while (0)
  68 #define xfs_icsb_balance_counter(mp, a, b)              do { } while (0)
  69 #define xfs_icsb_sync_counters(mp)                      do { } while (0)
  70 #define xfs_icsb_modify_counters(mp, a, b, c)           do { } while (0)
  71 #define xfs_icsb_modify_counters_locked(mp, a, b, c)    do { } while (0)
  72
  73 #endif
  74
  75 static const struct {
  76         short offset;
  77         short type;     /* 0 = integer
  78                          * 1 = binary / string (no translation)
  79                          */
  80 } xfs_sb_info[] = {
  81     { offsetof(xfs_sb_t, sb_magicnum),   0 },
  82     { offsetof(xfs_sb_t, sb_blocksize),  0 },
  83     { offsetof(xfs_sb_t, sb_dblocks),    0 },
  84     { offsetof(xfs_sb_t, sb_rblocks),    0 },
  85     { offsetof(xfs_sb_t, sb_rextents),   0 },
  86     { offsetof(xfs_sb_t, sb_uuid),       1 },
  87     { offsetof(xfs_sb_t, sb_logstart),   0 },
  88     { offsetof(xfs_sb_t, sb_rootino),    0 },
  89     { offsetof(xfs_sb_t, sb_rbmino),     0 },
  90     { offsetof(xfs_sb_t, sb_rsumino),    0 },
  91     { offsetof(xfs_sb_t, sb_rextsize),   0 },
  92     { offsetof(xfs_sb_t, sb_agblocks),   0 },
  93     { offsetof(xfs_sb_t, sb_agcount),    0 },
  94     { offsetof(xfs_sb_t, sb_rbmblocks),  0 },
  95     { offsetof(xfs_sb_t, sb_logblocks),  0 },
  96     { offsetof(xfs_sb_t, sb_versionnum), 0 },
  97     { offsetof(xfs_sb_t, sb_sectsize),   0 },
  98     { offsetof(xfs_sb_t, sb_inodesize),  0 },
  99     { offsetof(xfs_sb_t, sb_inopblock),  0 },
 100     { offsetof(xfs_sb_t, sb_fname[0]),   1 },
 101     { offsetof(xfs_sb_t, sb_blocklog),   0 },
 102     { offsetof(xfs_sb_t, sb_sectlog),    0 },
 103     { offsetof(xfs_sb_t, sb_inodelog),   0 },
 104     { offsetof(xfs_sb_t, sb_inopblog),   0 },
 105     { offsetof(xfs_sb_t, sb_agblklog),   0 },
 106     { offsetof(xfs_sb_t, sb_rextslog),   0 },
 107     { offsetof(xfs_sb_t, sb_inprogress), 0 },
 108     { offsetof(xfs_sb_t, sb_imax_pct),   0 },
 109     { offsetof(xfs_sb_t, sb_icount),     0 },
 110     { offsetof(xfs_sb_t, sb_ifree),      0 },
 111     { offsetof(xfs_sb_t, sb_fdblocks),   0 },
 112     { offsetof(xfs_sb_t, sb_frextents),  0 },
 113     { offsetof(xfs_sb_t, sb_uquotino),   0 },
 114     { offsetof(xfs_sb_t, sb_gquotino),   0 },
 115     { offsetof(xfs_sb_t, sb_qflags),     0 },
 116     { offsetof(xfs_sb_t, sb_flags),      0 },
 117     { offsetof(xfs_sb_t, sb_shared_vn),  0 },
 118     { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
 119     { offsetof(xfs_sb_t, sb_unit),       0 },
 120     { offsetof(xfs_sb_t, sb_width),      0 },
 121     { offsetof(xfs_sb_t, sb_dirblklog),  0 },
 122     { offsetof(xfs_sb_t, sb_logsectlog), 0 },
 123     { offsetof(xfs_sb_t, sb_logsectsize),0 },
 124     { offsetof(xfs_sb_t, sb_logsunit),   0 },
 125     { offsetof(xfs_sb_t, sb_features2),  0 },
 126     { sizeof(xfs_sb_t),                  0 }
 127 };
 128
 129 /*
 130  * Return a pointer to an initialized xfs_mount structure.
 131  */
 132 xfs_mount_t *
 133 xfs_mount_init(void)
 134 {
 135         xfs_mount_t *mp;
 136
 137         mp = kmem_zalloc(sizeof(xfs_mount_t), KM_SLEEP);
 138
 139         if (xfs_icsb_init_counters(mp)) {
 140                 mp->m_flags |= XFS_MOUNT_NO_PERCPU_SB;
 141         }
 142
 143         AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail");
 144         spinlock_init(&mp->m_sb_lock, "xfs_sb");
 145         /* FreeBSD specfic */
 146         sx_init(&mp->m_ilock, "xfs_mnt");
 147         initnsema(&mp->m_growlock, 1, "xfs_grow");
 148         /*
 149          * Initialize the AIL.
 150          */
 151         xfs_trans_ail_init(mp);
 152
 153         atomic_set(&mp->m_active_trans, 0);
 154
 155         return mp;
 156 }
 157
 158 /*
 159  * Free up the resources associated with a mount structure.  Assume that
 160  * the structure was initially zeroed, so we can tell which fields got
 161  * initialized.
 162  */
 163 void
 164 xfs_mount_free(
 165         xfs_mount_t     *mp,
 166         int             remove_bhv)
 167 {
 168         if (mp->m_ihash)
 169                 xfs_ihash_free(mp);
 170         if (mp->m_chash)
 171                 xfs_chash_free(mp);
 172
 173         if (mp->m_perag) {
 174                 int     agno;
 175
 176                 for (agno = 0; agno < mp->m_maxagi; agno++)
 177                         if (mp->m_perag[agno].pagb_list)
 178                                 kmem_free(mp->m_perag[agno].pagb_list,
 179                                                 sizeof(xfs_perag_busy_t) *
 180                                                         XFS_PAGB_NUM_SLOTS);
 181                 kmem_free(mp->m_perag,
 182                           sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
 183         }
 184
 185         AIL_LOCK_DESTROY(&mp->m_ail_lock);
 186         spinlock_destroy(&mp->m_sb_lock);
 187         /* FreeBSD specfic */
 188         sx_destroy(&mp->m_ilock);
 189         freesema(&mp->m_growlock);
 190         if (mp->m_quotainfo)
 191                 XFS_QM_DONE(mp);
 192
 193         if (mp->m_fsname != NULL)
 194                 kmem_free(mp->m_fsname, mp->m_fsname_len);
 195         if (mp->m_rtname != NULL)
 196                 kmem_free(mp->m_rtname, strlen(mp->m_rtname) + 1);
 197         if (mp->m_logname != NULL)
 198                 kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
 199
 200         if (remove_bhv) {
 201                 xfs_vfs_t       *vfsp = XFS_MTOVFS(mp);
 202
 203                 bhv_remove_all_vfsops(vfsp, 0);
 204                 VFS_REMOVEBHV(vfsp, &mp->m_bhv);
 205         }
 206
 207         xfs_icsb_destroy_counters(mp);
 208         kmem_free(mp, sizeof(xfs_mount_t));
 209 }
 210
 211
 212 /*
 213  * Check the validity of the SB found.
 214  */
 215 STATIC int
 216 xfs_mount_validate_sb(
 217         xfs_mount_t     *mp,
 218         xfs_sb_t        *sbp,
 219         int             flags)
 220 {
 221         /*
 222          * If the log device and data device have the
 223          * same device number, the log is internal.
 224          * Consequently, the sb_logstart should be non-zero.  If
 225          * we have a zero sb_logstart in this case, we may be trying to mount
 226          * a volume filesystem in a non-volume manner.
 227          */
 228         if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 229                 xfs_fs_mount_cmn_err(flags, "bad magic number");
 230                 return XFS_ERROR(EWRONGFS);
 231         }
 232
 233         if (!XFS_SB_GOOD_VERSION(sbp)) {
 234                 xfs_fs_mount_cmn_err(flags, "bad version");
 235                 return XFS_ERROR(EWRONGFS);
 236         }
 237
 238         if (unlikely(
 239             sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
 240                 xfs_fs_mount_cmn_err(flags,
 241                         "filesystem is marked as having an external log; "
 242                         "specify logdev on the\nmount command line.");
 243                 return XFS_ERROR(EINVAL);
 244         }
 245
 246         if (unlikely(
 247             sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
 248                 xfs_fs_mount_cmn_err(flags,
 249                         "filesystem is marked as having an internal log; "
 250                         "do not specify logdev on\nthe mount command line.");
 251                 return XFS_ERROR(EINVAL);
 252         }
 253
 254         /*
 255          * More sanity checking. These were stolen directly from
 256          * xfs_repair.
 257          */
 258         if (unlikely(
 259             sbp->sb_agcount <= 0                                        ||
 260             sbp->sb_sectsize < XFS_MIN_SECTORSIZE                       ||
 261             sbp->sb_sectsize > XFS_MAX_SECTORSIZE                       ||
 262             sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG                    ||
 263             sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG                    ||
 264             sbp->sb_blocksize < XFS_MIN_BLOCKSIZE                       ||
 265             sbp->sb_blocksize > XFS_MAX_BLOCKSIZE                       ||
 266             sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG                    ||
 267             sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG                    ||
 268             sbp->sb_inodesize < XFS_DINODE_MIN_SIZE                     ||
 269             sbp->sb_inodesize > XFS_DINODE_MAX_SIZE                     ||
 270             sbp->sb_inodelog < XFS_DINODE_MIN_LOG                       ||
 271             sbp->sb_inodelog > XFS_DINODE_MAX_LOG                       ||
 272             (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
 273             (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
 274             (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
 275             (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
 276                 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
 277                 return XFS_ERROR(EFSCORRUPTED);
 278         }
 279
 280         /*
 281          * Sanity check AG count, size fields against data size field
 282          */
 283         if (unlikely(
 284             sbp->sb_dblocks == 0 ||
 285             sbp->sb_dblocks >
 286              (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
 287             sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
 288                               sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
 289                 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
 290                 return XFS_ERROR(EFSCORRUPTED);
 291         }
 292
 293         ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
 294         ASSERT(sbp->sb_blocklog >= BBSHIFT);
 295
 296 #if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
 297         if (unlikely(
 298             (sbp->sb_dblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX ||
 299             (sbp->sb_rblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX)) {
 300 #else                  /* Limited by UINT_MAX of sectors */
 301         if (unlikely(
 302             (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX ||
 303             (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) {
 304 #endif
 305                 xfs_fs_mount_cmn_err(flags,
 306                         "file system too large to be mounted on this system.");
 307                 return XFS_ERROR(E2BIG);
 308         }
 309
 310         if (unlikely(sbp->sb_inprogress)) {
 311                 xfs_fs_mount_cmn_err(flags, "file system busy");
 312                 return XFS_ERROR(EFSCORRUPTED);
 313         }
 314
 315         /*
 316          * Version 1 directory format has never worked on Linux.
 317          */
 318         if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp))) {
 319                 xfs_fs_mount_cmn_err(flags,
 320                         "file system using version 1 directory format");
 321                 return XFS_ERROR(ENOSYS);
 322         }
 323
 324         /*
 325          * Until this is fixed only page-sized or smaller data blocks work.
 326          */
 327         if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
 328                 xfs_fs_mount_cmn_err(flags,
 329                         "file system with blocksize %d bytes",
 330                         sbp->sb_blocksize);
 331                 xfs_fs_mount_cmn_err(flags,
 332                         "only pagesize (%ld) or less will currently work.",
 333                         PAGE_SIZE);
 334                 return XFS_ERROR(ENOSYS);
 335         }
 336
 337         return 0;
 338 }
 339
 340 xfs_agnumber_t
 341 xfs_initialize_perag(
 342         struct xfs_vfs  *vfs,
 343         xfs_mount_t     *mp,
 344         xfs_agnumber_t  agcount)
 345 {
 346         xfs_agnumber_t  index, max_metadata;
 347         xfs_perag_t     *pag;
 348         xfs_agino_t     agino;
 349         xfs_ino_t       ino;
 350         xfs_sb_t        *sbp = &mp->m_sb;
 351         xfs_ino_t       max_inum = XFS_MAXINUMBER_32;
 352
 353         /* Check to see if the filesystem can overflow 32 bit inodes */
 354         agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
 355         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 356
 357         /* Clear the mount flag if no inode can overflow 32 bits
 358          * on this filesystem, or if specifically requested..
 359          */
 360         if ((vfs->vfs_flag & VFS_32BITINODES) && ino > max_inum) {
 361                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 362         } else {
 363                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
 364         }
 365
 366         /* If we can overflow then setup the ag headers accordingly */
 367         if (mp->m_flags & XFS_MOUNT_32BITINODES) {
 368                 /* Calculate how much should be reserved for inodes to
 369                  * meet the max inode percentage.
 370                  */
 371                 if (mp->m_maxicount) {
 372                         __uint64_t      icount;
 373
 374                         icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 375                         do_div(icount, 100);
 376                         icount += sbp->sb_agblocks - 1;
 377                         do_div(icount, sbp->sb_agblocks);
 378                         max_metadata = icount;
 379                 } else {
 380                         max_metadata = agcount;
 381                 }
 382                 for (index = 0; index < agcount; index++) {
 383                         ino = XFS_AGINO_TO_INO(mp, index, agino);
 384                         if (ino > max_inum) {
 385                                 index++;
 386                                 break;
 387                         }
 388
 389                         /* This ag is preferred for inodes */
 390                         pag = &mp->m_perag[index];
 391                         pag->pagi_inodeok = 1;
 392                         if (index < max_metadata)
 393                                 pag->pagf_metadata = 1;
 394                 }
 395         } else {
 396                 /* Setup default behavior for smaller filesystems */
 397                 for (index = 0; index < agcount; index++) {
 398                         pag = &mp->m_perag[index];
 399                         pag->pagi_inodeok = 1;
 400                 }
 401         }
 402         return index;
 403 }
 404
 405 /*
 406  * xfs_xlatesb
 407  *
 408  *     data       - on disk version of sb
 409  *     sb         - a superblock
 410  *     dir        - conversion direction: <0 - convert sb to buf
 411  *                                        >0 - convert buf to sb
 412  *     fields     - which fields to copy (bitmask)
 413  */
 414 void
 415 xfs_xlatesb(
 416         void            *data,
 417         xfs_sb_t        *sb,
 418         int             dir,
 419         __int64_t       fields)
 420 {
 421         xfs_caddr_t     buf_ptr;
 422         xfs_caddr_t     mem_ptr;
 423         xfs_sb_field_t  f;
 424         int             first;
 425         int             size;
 426
 427         ASSERT(dir);
 428         ASSERT(fields);
 429
 430         if (!fields)
 431                 return;
 432
 433         buf_ptr = (xfs_caddr_t)data;
 434         mem_ptr = (xfs_caddr_t)sb;
 435
 436         while (fields) {
 437                 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
 438                 first = xfs_sb_info[f].offset;
 439                 size = xfs_sb_info[f + 1].offset - first;
 440
 441                 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
 442
 443                 if (size == 1 || xfs_sb_info[f].type == 1) {
 444                         if (dir > 0) {
 445                                 memcpy(mem_ptr + first, buf_ptr + first, size);
 446                         } else {
 447                                 memcpy(buf_ptr + first, mem_ptr + first, size);
 448                         }
 449                 } else {
 450                         switch (size) {
 451                         case 2:
 452                                 INT_XLATE(*(__uint16_t*)(buf_ptr+first),
 453                                           *(__uint16_t*)(mem_ptr+first),
 454                                           dir, ARCH_CONVERT);
 455                                 break;
 456                         case 4:
 457                                 INT_XLATE(*(__uint32_t*)(buf_ptr+first),
 458                                           *(__uint32_t*)(mem_ptr+first),
 459                                           dir, ARCH_CONVERT);
 460                                 break;
 461                         case 8:
 462                                 INT_XLATE(*(__uint64_t*)(buf_ptr+first),
 463                                           *(__uint64_t*)(mem_ptr+first), dir, ARCH_CONVERT);
 464                                 break;
 465                         default:
 466                                 ASSERT(0);
 467                         }
 468                 }
 469
 470                 fields &= ~(1LL << f);
 471         }
 472 }
 473
 474 /*
 475  * xfs_readsb
 476  *
 477  * Does the initial read of the superblock.
 478  */
 479 int
 480 xfs_readsb(xfs_mount_t *mp, int flags)
 481 {
 482         unsigned int    sector_size;
 483         unsigned int    extra_flags;
 484         xfs_buf_t       *bp;
 485         xfs_sb_t        *sbp;
 486         int             error;
 487
 488         ASSERT(mp->m_sb_bp == NULL);
 489         ASSERT(mp->m_ddev_targp != NULL);
 490
 491         /*
 492          * Allocate a (locked) buffer to hold the superblock.
 493          * This will be kept around at all times to optimize
 494          * access to the superblock.
 495          */
 496         sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
 497         extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
 498
 499         bp = xfs_getsb(mp,0);
 500
 501         if (!bp || XFS_BUF_ISERROR(bp)) {
 502                 xfs_fs_mount_cmn_err(flags, "SB read failed");
 503                 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
 504                 goto fail;
 505         }
 506         ASSERT(XFS_BUF_ISBUSY(bp));
 507         ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
 508
 509         /*
 510          * Initialize the mount structure from the superblock.
 511          * But first do some basic consistency checking.
 512          */
 513         sbp = XFS_BUF_TO_SBP(bp);
 514         xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1, XFS_SB_ALL_BITS);
 515
 516         error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
 517         if (error) {
 518                 xfs_fs_mount_cmn_err(flags, "SB validate failed");
 519                 goto fail;
 520         }
 521
 522         /*
 523          * We must be able to do sector-sized and sector-aligned IO.
 524          */
 525         if (sector_size > mp->m_sb.sb_sectsize) {
 526                 xfs_fs_mount_cmn_err(flags,
 527                         "device supports only %u byte sectors (not %u)",
 528                         sector_size, mp->m_sb.sb_sectsize);
 529                 error = ENOSYS;
 530                 goto fail;
 531         }
 532
 533         /*
 534          * If device sector size is smaller than the superblock size,
 535          * re-read the superblock so the buffer is correctly sized.
 536          */
 537         if (sector_size < mp->m_sb.sb_sectsize) {
 538                 XFS_BUF_UNMANAGE(bp);
 539                 xfs_buf_relse(bp);
 540                 sector_size = mp->m_sb.sb_sectsize;
 541                 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
 542                                         BTOBB(sector_size), extra_flags);
 543                 if (!bp || XFS_BUF_ISERROR(bp)) {
 544                         xfs_fs_mount_cmn_err(flags, "SB re-read failed");
 545                         error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
 546                         goto fail;
 547                 }
 548                 ASSERT(XFS_BUF_ISBUSY(bp));
 549                 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
 550         }
 551
 552         xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
 553         xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
 554         xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
 555
 556         mp->m_sb_bp = bp;
 557         xfs_buf_relse(bp);
 558         ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
 559         return 0;
 560
 561  fail:
 562         if (bp) {
 563                 XFS_BUF_UNMANAGE(bp);
 564                 xfs_buf_relse(bp);
 565         }
 566         return error;
 567 }
 568
 569
 570 /*
 571  * xfs_mount_common
 572  *
 573  * Mount initialization code establishing various mount
 574  * fields from the superblock associated with the given
 575  * mount structure
 576  */
 577 STATIC void
 578 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
 579 {
 580         int     i;
 581
 582         mp->m_agfrotor = mp->m_agirotor = 0;
 583         spinlock_init(&mp->m_agirotor_lock, "m_agirotor_lock");
 584         mp->m_maxagi = mp->m_sb.sb_agcount;
 585         mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
 586         mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
 587         mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
 588         mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
 589         mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
 590         mp->m_litino = sbp->sb_inodesize -
 591                 ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
 592         mp->m_blockmask = sbp->sb_blocksize - 1;
 593         mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
 594         mp->m_blockwmask = mp->m_blockwsize - 1;
 595 #ifdef RMC
 596         INIT_LIST_HEAD(&mp->m_del_inodes);
 597 #endif
 598         TAILQ_INIT(&mp->m_del_inodes);
 599
 600         /*
 601          * Setup for attributes, in case they get created.
 602          * This value is for inodes getting attributes for the first time,
 603          * the per-inode value is for old attribute values.
 604          */
 605         ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
 606         switch (sbp->sb_inodesize) {
 607         case 256:
 608                 mp->m_attroffset = XFS_LITINO(mp) -
 609                                    XFS_BMDR_SPACE_CALC(MINABTPTRS);
 610                 break;
 611         case 512:
 612         case 1024:
 613         case 2048:
 614                 mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 615                 break;
 616         default:
 617                 ASSERT(0);
 618         }
 619         ASSERT(mp->m_attroffset < XFS_LITINO(mp));
 620
 621         for (i = 0; i < 2; i++) {
 622                 mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
 623                         xfs_alloc, i == 0);
 624                 mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
 625                         xfs_alloc, i == 0);
 626         }
 627         for (i = 0; i < 2; i++) {
 628                 mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
 629                         xfs_bmbt, i == 0);
 630                 mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
 631                         xfs_bmbt, i == 0);
 632         }
 633         for (i = 0; i < 2; i++) {
 634                 mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
 635                         xfs_inobt, i == 0);
 636                 mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
 637                         xfs_inobt, i == 0);
 638         }
 639
 640         mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
 641         mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 642                                         sbp->sb_inopblock);
 643         mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 644 }
 645 /*
 646  * xfs_mountfs
 647  *
 648  * This function does the following on an initial mount of a file system:
 649  *      - reads the superblock from disk and init the mount struct
 650  *      - if we're a 32-bit kernel, do a size check on the superblock
 651  *              so we don't mount terabyte filesystems
 652  *      - init mount struct realtime fields
 653  *      - allocate inode hash table for fs
 654  *      - init directory manager
 655  *      - perform recovery and init the log manager
 656  */
 657 int
 658 xfs_mountfs(
 659         xfs_vfs_t       *vfsp,
 660         xfs_mount_t     *mp,
 661         int             mfsi_flags)
 662 {
 663         xfs_buf_t       *bp;
 664         xfs_sb_t        *sbp = &(mp->m_sb);
 665         xfs_inode_t     *rip;
 666         xfs_vnode_t     *rvp = NULL;
 667         int             readio_log, writeio_log;
 668         xfs_daddr_t     d;
 669         __uint64_t      ret64;
 670         __int64_t       update_flags;
 671         uint            quotamount, quotaflags;
 672         int             agno;
 673         int             uuid_mounted = 0;
 674         int             error = 0;
 675
 676         if (mp->m_sb_bp == NULL) {
 677                 if ((error = xfs_readsb(mp, mfsi_flags))) {
 678                         return error;
 679                 }
 680         }
 681         xfs_mount_common(mp, sbp);
 682
 683         /*
 684          * Check if sb_agblocks is aligned at stripe boundary
 685          * If sb_agblocks is NOT aligned turn off m_dalign since
 686          * allocator alignment is within an ag, therefore ag has
 687          * to be aligned at stripe boundary.
 688          */
 689         update_flags = 0LL;
 690         if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
 691                 /*
 692                  * If stripe unit and stripe width are not multiples
 693                  * of the fs blocksize turn off alignment.
 694                  */
 695                 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
 696                     (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
 697                         if (mp->m_flags & XFS_MOUNT_RETERR) {
 698                                 cmn_err(CE_WARN,
 699                                         "XFS: alignment check 1 failed");
 700                                 error = XFS_ERROR(EINVAL);
 701                                 goto error1;
 702                         }
 703                         mp->m_dalign = mp->m_swidth = 0;
 704                 } else {
 705                         /*
 706                          * Convert the stripe unit and width to FSBs.
 707                          */
 708                         mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
 709                         if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
 710                                 if (mp->m_flags & XFS_MOUNT_RETERR) {
 711                                         error = XFS_ERROR(EINVAL);
 712                                         goto error1;
 713                                 }
 714                                 xfs_fs_cmn_err(CE_WARN, mp,
 715 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
 716                                         mp->m_dalign, mp->m_swidth,
 717                                         sbp->sb_agblocks);
 718
 719                                 mp->m_dalign = 0;
 720                                 mp->m_swidth = 0;
 721                         } else if (mp->m_dalign) {
 722                                 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
 723                         } else {
 724                                 if (mp->m_flags & XFS_MOUNT_RETERR) {
 725                                         xfs_fs_cmn_err(CE_WARN, mp,
 726 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
 727                                                 mp->m_dalign,
 728                                                 mp->m_blockmask +1);
 729                                         error = XFS_ERROR(EINVAL);
 730                                         goto error1;
 731                                 }
 732                                 mp->m_swidth = 0;
 733                         }
 734                 }
 735
 736                 /*
 737                  * Update superblock with new values
 738                  * and log changes
 739                  */
 740                 if (XFS_SB_VERSION_HASDALIGN(sbp)) {
 741                         if (sbp->sb_unit != mp->m_dalign) {
 742                                 sbp->sb_unit = mp->m_dalign;
 743                                 update_flags |= XFS_SB_UNIT;
 744                         }
 745                         if (sbp->sb_width != mp->m_swidth) {
 746                                 sbp->sb_width = mp->m_swidth;
 747                                 update_flags |= XFS_SB_WIDTH;
 748                         }
 749                 }
 750         } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
 751                     XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
 752                         mp->m_dalign = sbp->sb_unit;
 753                         mp->m_swidth = sbp->sb_width;
 754         }
 755
 756         xfs_alloc_compute_maxlevels(mp);
 757         xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
 758         xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
 759         xfs_ialloc_compute_maxlevels(mp);
 760
 761         if (sbp->sb_imax_pct) {
 762                 __uint64_t      icount;
 763
 764                 /* Make sure the maximum inode count is a multiple of the
 765                  * units we allocate inodes in.
 766                  */
 767
 768                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 769                 do_div(icount, 100);
 770                 do_div(icount, mp->m_ialloc_blks);
 771                 mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
 772                                    sbp->sb_inopblog;
 773         } else
 774                 mp->m_maxicount = 0;
 775
 776         mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
 777
 778         /*
 779          * XFS uses the uuid from the superblock as the unique
 780          * identifier for fsid.  We can not use the uuid from the volume
 781          * since a single partition filesystem is identical to a single
 782          * partition volume/filesystem.
 783          */
 784         if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
 785             (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
 786                 if (xfs_uuid_mount(mp)) {
 787                         error = XFS_ERROR(EINVAL);
 788                         goto error1;
 789                 }
 790                 uuid_mounted=1;
 791                 ret64 = uuid_hash64(&sbp->sb_uuid);
 792                 memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
 793         }
 794
 795         /*
 796          * Set the default minimum read and write sizes unless
 797          * already specified in a mount option.
 798          * We use smaller I/O sizes when the file system
 799          * is being used for NFS service (wsync mount option).
 800          */
 801         if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
 802                 if (mp->m_flags & XFS_MOUNT_WSYNC) {
 803                         readio_log = XFS_WSYNC_READIO_LOG;
 804                         writeio_log = XFS_WSYNC_WRITEIO_LOG;
 805                 } else {
 806                         readio_log = XFS_READIO_LOG_LARGE;
 807                         writeio_log = XFS_WRITEIO_LOG_LARGE;
 808                 }
 809         } else {
 810                 readio_log = mp->m_readio_log;
 811                 writeio_log = mp->m_writeio_log;
 812         }
 813
 814         /*
 815          * Set the number of readahead buffers to use based on
 816          * physical memory size.
 817          */
 818         if (xfs_physmem <= 4096)                /* <= 16MB */
 819                 mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
 820         else if (xfs_physmem <= 8192)   /* <= 32MB */
 821                 mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
 822         else
 823                 mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
 824         if (sbp->sb_blocklog > readio_log) {
 825                 mp->m_readio_log = sbp->sb_blocklog;
 826         } else {
 827                 mp->m_readio_log = readio_log;
 828         }
 829         mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
 830         if (sbp->sb_blocklog > writeio_log) {
 831                 mp->m_writeio_log = sbp->sb_blocklog;
 832         } else {
 833                 mp->m_writeio_log = writeio_log;
 834         }
 835         mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
 836
 837         /*
 838          * Set the inode cluster size based on the physical memory
 839          * size.  This may still be overridden by the file system
 840          * block size if it is larger than the chosen cluster size.
 841          */
 842         if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
 843                 mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
 844         } else {
 845                 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
 846         }
 847         /*
 848          * Set whether we're using inode alignment.
 849          */
 850         if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
 851             mp->m_sb.sb_inoalignmt >=
 852             XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
 853                 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
 854         else
 855                 mp->m_inoalign_mask = 0;
 856         /*
 857          * If we are using stripe alignment, check whether
 858          * the stripe unit is a multiple of the inode alignment
 859          */
 860         if (mp->m_dalign && mp->m_inoalign_mask &&
 861             !(mp->m_dalign & mp->m_inoalign_mask))
 862                 mp->m_sinoalign = mp->m_dalign;
 863         else
 864                 mp->m_sinoalign = 0;
 865         /*
 866          * Check that the data (and log if separate) are an ok size.
 867          */
 868         d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
 869         if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
 870                 cmn_err(CE_WARN, "XFS: size check 1 failed");
 871                 error = XFS_ERROR(E2BIG);
 872                 goto error1;
 873         }
 874         error = xfs_read_buf(mp, mp->m_ddev_targp,
 875                              d - XFS_FSS_TO_BB(mp, 1),
 876                              XFS_FSS_TO_BB(mp, 1), 0, &bp);
 877         if (!error) {
 878                 xfs_buf_relse(bp);
 879         } else {
 880                 cmn_err(CE_WARN, "XFS: size check 2 failed");
 881                 if (error == ENOSPC) {
 882                         error = XFS_ERROR(E2BIG);
 883                 }
 884                 goto error1;
 885         }
 886
 887         if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
 888             mp->m_logdev_targp != mp->m_ddev_targp) {
 889                 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
 890                 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
 891                         cmn_err(CE_WARN, "XFS: size check 3 failed");
 892                         error = XFS_ERROR(E2BIG);
 893                         goto error1;
 894                 }
 895                 error = xfs_read_buf(mp, mp->m_logdev_targp,
 896                                      d - XFS_FSB_TO_BB(mp, 1),
 897                                      XFS_FSB_TO_BB(mp, 1), 0, &bp);
 898                 if (!error) {
 899                         xfs_buf_relse(bp);
 900                 } else {
 901                         cmn_err(CE_WARN, "XFS: size check 3 failed");
 902                         if (error == ENOSPC) {
 903                                 error = XFS_ERROR(E2BIG);
 904                         }
 905                         goto error1;
 906                 }
 907         }
 908
 909         /*
 910          * Initialize realtime fields in the mount structure
 911          */
 912         if ((error = xfs_rtmount_init(mp))) {
 913                 cmn_err(CE_WARN, "XFS: RT mount failed");
 914                 goto error1;
 915         }
 916
 917         /*
 918          * For client case we are done now
 919          */
 920         if (mfsi_flags & XFS_MFSI_CLIENT) {
 921                 return 0;
 922         }
 923
 924         /*
 925          *  Copies the low order bits of the timestamp and the randomly
 926          *  set "sequence" number out of a UUID.
 927          */
 928         uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
 929
 930         /*
 931          *  The vfs structure needs to have a file system independent
 932          *  way of checking for the invariant file system ID.  Since it
 933          *  can't look at mount structures it has a pointer to the data
 934          *  in the mount structure.
 935          *
 936          *  File systems that don't support user level file handles (i.e.
 937          *  all of them except for XFS) will leave vfs_altfsid as NULL.
 938          */
 939         vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
 940         mp->m_dmevmask = 0;     /* not persistent; set after each mount */
 941
 942         /*
 943          * Select the right directory manager.
 944          */
 945         mp->m_dirops =
 946                 XFS_SB_VERSION_HASDIRV2(&mp->m_sb) ?
 947                         xfsv2_dirops :
 948                         xfsv1_dirops;
 949
 950         /*
 951          * Initialize directory manager's entries.
 952          */
 953         XFS_DIR_MOUNT(mp);
 954
 955         /*
 956          * Initialize the attribute manager's entries.
 957          */
 958         mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
 959
 960         /*
 961          * Initialize the precomputed transaction reservations values.
 962          */
 963         xfs_trans_init(mp);
 964
 965         /*
 966          * Allocate and initialize the inode hash table for this
 967          * file system.
 968          */
 969         xfs_ihash_init(mp);
 970         xfs_chash_init(mp);
 971
 972         /*
 973          * Allocate and initialize the per-ag data.
 974          */
 975         init_rwsem(&mp->m_peraglock);
 976         mp->m_perag =
 977                 kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
 978
 979         mp->m_maxagi = xfs_initialize_perag(vfsp, mp, sbp->sb_agcount);
 980
 981         /*
 982          * log's mount-time initialization. Perform 1st part recovery if needed
 983          */
 984         if (likely(sbp->sb_logblocks > 0)) {    /* check for volume case */
 985                 error = xfs_log_mount(mp, mp->m_logdev_targp,
 986                                       XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
 987                                       XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
 988                 if (error) {
 989                         cmn_err(CE_WARN, "XFS: log mount failed");
 990                         goto error2;
 991                 }
 992         } else {        /* No log has been defined */
 993                 cmn_err(CE_WARN, "XFS: no log defined");
 994                 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
 995                 error = XFS_ERROR(EFSCORRUPTED);
 996                 goto error2;
 997         }
 998
 999         /*
1000          * Get and sanity-check the root inode.
1001          * Save the pointer to it in the mount structure.
1002          */
1003         error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
1004         if (error) {
1005                 cmn_err(CE_WARN, "XFS: failed to read root inode");
1006                 goto error3;
1007         }
1008
1009         ASSERT(rip != NULL);
1010         rvp = XFS_ITOV(rip);
1011
1012         if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1013                 cmn_err(CE_WARN, "XFS: corrupted root inode");
1014                 printf("Root inode %p is not a directory: %llu",
1015                        mp->m_ddev_targp, (unsigned long long)rip->i_ino);
1016                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1017                 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1018                                  mp);
1019                 error = XFS_ERROR(EFSCORRUPTED);
1020                 goto error4;
1021         }
1022         mp->m_rootip = rip;     /* save it */
1023
1024         xfs_iunlock(rip, XFS_ILOCK_EXCL);
1025
1026         /*
1027          * Initialize realtime inode pointers in the mount structure
1028          */
1029         if ((error = xfs_rtmount_inodes(mp))) {
1030                 /*
1031                  * Free up the root inode.
1032                  */
1033                 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1034                 goto error4;
1035         }
1036
1037         /*
1038          * If fs is not mounted readonly, then update the superblock
1039          * unit and width changes.
1040          */
1041         if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
1042                 xfs_mount_log_sbunit(mp, update_flags);
1043
1044         /*
1045          * Initialise the XFS quota management subsystem for this mount
1046          */
1047         if ((error = XFS_QM_INIT(mp, &quotamount, &quotaflags)))
1048                 goto error4;
1049
1050         /*
1051          * Finish recovering the file system.  This part needed to be
1052          * delayed until after the root and real-time bitmap inodes
1053          * were consistently read in.
1054          */
1055         error = xfs_log_mount_finish(mp, mfsi_flags);
1056         if (error) {
1057                 cmn_err(CE_WARN, "XFS: log mount finish failed");
1058                 goto error4;
1059         }
1060
1061         /*
1062          * Complete the quota initialisation, post-log-replay component.
1063          */
1064         if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags)))
1065                 goto error4;
1066
1067         return 0;
1068
1069  error4:
1070         /*
1071          * Free up the root inode.
1072          */
1073         VN_RELE(rvp);
1074  error3:
1075         xfs_log_unmount_dealloc(mp);
1076  error2:
1077         xfs_ihash_free(mp);
1078         xfs_chash_free(mp);
1079         for (agno = 0; agno < sbp->sb_agcount; agno++)
1080                 if (mp->m_perag[agno].pagb_list)
1081                         kmem_free(mp->m_perag[agno].pagb_list,
1082                           sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1083         kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1084         mp->m_perag = NULL;
1085         /* FALLTHROUGH */
1086  error1:
1087         if (uuid_mounted)
1088                 xfs_uuid_unmount(mp);
1089         xfs_freesb(mp);
1090         return error;
1091 }
1092
1093 /*
1094  * xfs_unmountfs
1095  *
1096  * This flushes out the inodes,dquots and the superblock, unmounts the
1097  * log and makes sure that incore structures are freed.
1098  */
1099 int
1100 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1101 {
1102         struct xfs_vfs  *vfsp = XFS_MTOVFS(mp);
1103 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1104         int64_t         fsid;
1105 #endif
1106
1107         xfs_iflush_all(mp);
1108
1109         XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
1110
1111         /*
1112          * Flush out the log synchronously so that we know for sure
1113          * that nothing is pinned.  This is important because bflush()
1114          * will skip pinned buffers.
1115          */
1116         xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1117
1118         xfs_binval(mp->m_ddev_targp);
1119         if (mp->m_rtdev_targp) {
1120                 xfs_binval(mp->m_rtdev_targp);
1121         }
1122
1123         xfs_unmountfs_writesb(mp);
1124
1125         xfs_unmountfs_wait(mp);                 /* wait for async bufs */
1126
1127         xfs_log_unmount(mp);                    /* Done! No more fs ops. */
1128
1129         xfs_freesb(mp);
1130
1131         /*
1132          * All inodes from this mount point should be freed.
1133          */
1134         //ASSERT(mp->m_inodes == NULL);
1135         if (mp->m_inodes != NULL ) {
1136                 printf("WRONG: mp->m_ireclaims: %d\n", mp->m_ireclaims);
1137                 printf("WRONG: mp->m_inodes: %p\n", mp->m_inodes);
1138         }
1139
1140         xfs_unmountfs_close(mp, cr);
1141         if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1142                 xfs_uuid_unmount(mp);
1143
1144 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1145         /*
1146          * clear all error tags on this filesystem
1147          */
1148         memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
1149         xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
1150 #endif
1151         XFS_IODONE(vfsp);
1152         xfs_mount_free(mp, 1);
1153         return 0;
1154 }
1155
1156 void
1157 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1158 {
1159         if (mp->m_logdev_targp != mp->m_ddev_targp)
1160                 xfs_free_buftarg(mp->m_logdev_targp, 1);
1161         if (mp->m_rtdev_targp)
1162                 xfs_free_buftarg(mp->m_rtdev_targp, 1);
1163         xfs_free_buftarg(mp->m_ddev_targp, 0);
1164 }
1165
1166 STATIC void
1167 xfs_unmountfs_wait(xfs_mount_t *mp)
1168 {
1169         if (mp->m_logdev_targp != mp->m_ddev_targp)
1170                 xfs_wait_buftarg(mp->m_logdev_targp);
1171         if (mp->m_rtdev_targp)
1172                 xfs_wait_buftarg(mp->m_rtdev_targp);
1173         xfs_wait_buftarg(mp->m_ddev_targp);
1174 }
1175
1176 int
1177 xfs_unmountfs_writesb(xfs_mount_t *mp)
1178 {
1179         xfs_buf_t       *sbp;
1180         xfs_sb_t        *sb;
1181         int             error = 0;
1182
1183         /*
1184          * skip superblock write if fs is read-only, or
1185          * if we are doing a forced umount.
1186          */
1187         sbp = xfs_getsb(mp, 0);
1188         if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
1189                 XFS_FORCED_SHUTDOWN(mp))) {
1190
1191                 xfs_icsb_sync_counters(mp);
1192
1193                 /*
1194                  * mark shared-readonly if desired
1195                  */
1196                 sb = XFS_BUF_TO_SBP(sbp);
1197                 if (mp->m_mk_sharedro) {
1198                         if (!(sb->sb_flags & XFS_SBF_READONLY))
1199                                 sb->sb_flags |= XFS_SBF_READONLY;
1200                         if (!XFS_SB_VERSION_HASSHARED(sb))
1201                                 XFS_SB_VERSION_ADDSHARED(sb);
1202                         xfs_fs_cmn_err(CE_NOTE, mp,
1203                                 "Unmounting, marking shared read-only");
1204                 }
1205                 XFS_BUF_UNDONE(sbp);
1206                 XFS_BUF_UNREAD(sbp);
1207                 XFS_BUF_UNDELAYWRITE(sbp);
1208                 XFS_BUF_WRITE(sbp);
1209                 XFS_BUF_UNASYNC(sbp);
1210                 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1211                 xfsbdstrat(mp, sbp);
1212                 /* Nevermind errors we might get here. */
1213                 error = xfs_iowait(sbp);
1214                 if (error)
1215                         xfs_ioerror_alert("xfs_unmountfs_writesb",
1216                                           mp, sbp, XFS_BUF_ADDR(sbp));
1217                 if (error && mp->m_mk_sharedro)
1218                         xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting.  Filesystem may not be marked shared readonly");
1219         }
1220         xfs_buf_relse(sbp);
1221         return error;
1222 }
1223
1224 /*
1225  * xfs_mod_sb() can be used to copy arbitrary changes to the
1226  * in-core superblock into the superblock buffer to be logged.
1227  * It does not provide the higher level of locking that is
1228  * needed to protect the in-core superblock from concurrent
1229  * access.
1230  */
1231 void
1232 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1233 {
1234         xfs_buf_t       *bp;
1235         int             first;
1236         int             last;
1237         xfs_mount_t     *mp;
1238         xfs_sb_t        *sbp;
1239         xfs_sb_field_t  f;
1240
1241         ASSERT(fields);
1242         if (!fields)
1243                 return;
1244         mp = tp->t_mountp;
1245         bp = xfs_trans_getsb(tp, mp, 0);
1246         sbp = XFS_BUF_TO_SBP(bp);
1247         first = sizeof(xfs_sb_t);
1248         last = 0;
1249
1250         /* translate/copy */
1251
1252         xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, fields);
1253
1254         /* find modified range */
1255
1256         f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1257         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1258         first = xfs_sb_info[f].offset;
1259
1260         f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1261         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1262         last = xfs_sb_info[f + 1].offset - 1;
1263
1264         xfs_trans_log_buf(tp, bp, first, last);
1265 }
1266 /*
1267  * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1268  * a delta to a specified field in the in-core superblock.  Simply
1269  * switch on the field indicated and apply the delta to that field.
1270  * Fields are not allowed to dip below zero, so if the delta would
1271  * do this do not apply it and return EINVAL.
1272  *
1273  * The SB_LOCK must be held when this routine is called.
1274  */
1275 int
1276 xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field,
1277                         int delta, int rsvd)
1278 {
1279         int             scounter;       /* short counter for 32 bit fields */
1280         long long       lcounter;       /* long counter for 64 bit fields */
1281         long long       res_used, rem;
1282
1283         /*
1284          * With the in-core superblock spin lock held, switch
1285          * on the indicated field.  Apply the delta to the
1286          * proper field.  If the fields value would dip below
1287          * 0, then do not apply the delta and return EINVAL.
1288          */
1289         switch (field) {
1290         case XFS_SBS_ICOUNT:
1291                 lcounter = (long long)mp->m_sb.sb_icount;
1292                 lcounter += delta;
1293                 if (lcounter < 0) {
1294                         ASSERT(0);
1295                         return XFS_ERROR(EINVAL);
1296                 }
1297                 mp->m_sb.sb_icount = lcounter;
1298                 return 0;
1299         case XFS_SBS_IFREE:
1300                 lcounter = (long long)mp->m_sb.sb_ifree;
1301                 lcounter += delta;
1302                 if (lcounter < 0) {
1303                         ASSERT(0);
1304                         return XFS_ERROR(EINVAL);
1305                 }
1306                 mp->m_sb.sb_ifree = lcounter;
1307                 return 0;
1308         case XFS_SBS_FDBLOCKS:
1309
1310                 lcounter = (long long)mp->m_sb.sb_fdblocks;
1311                 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1312
1313                 if (delta > 0) {                /* Putting blocks back */
1314                         if (res_used > delta) {
1315                                 mp->m_resblks_avail += delta;
1316                         } else {
1317                                 rem = delta - res_used;
1318                                 mp->m_resblks_avail = mp->m_resblks;
1319                                 lcounter += rem;
1320                         }
1321                 } else {                                /* Taking blocks away */
1322
1323                         lcounter += delta;
1324
1325                 /*
1326                  * If were out of blocks, use any available reserved blocks if
1327                  * were allowed to.
1328                  */
1329
1330                         if (lcounter < 0) {
1331                                 if (rsvd) {
1332                                         lcounter = (long long)mp->m_resblks_avail + delta;
1333                                         if (lcounter < 0) {
1334                                                 return XFS_ERROR(ENOSPC);
1335                                         }
1336                                         mp->m_resblks_avail = lcounter;
1337                                         return 0;
1338                                 } else {        /* not reserved */
1339                                         return XFS_ERROR(ENOSPC);
1340                                 }
1341                         }
1342                 }
1343
1344                 mp->m_sb.sb_fdblocks = lcounter;
1345                 return 0;
1346         case XFS_SBS_FREXTENTS:
1347                 lcounter = (long long)mp->m_sb.sb_frextents;
1348                 lcounter += delta;
1349                 if (lcounter < 0) {
1350                         return XFS_ERROR(ENOSPC);
1351                 }
1352                 mp->m_sb.sb_frextents = lcounter;
1353                 return 0;
1354         case XFS_SBS_DBLOCKS:
1355                 lcounter = (long long)mp->m_sb.sb_dblocks;
1356                 lcounter += delta;
1357                 if (lcounter < 0) {
1358                         ASSERT(0);
1359                         return XFS_ERROR(EINVAL);
1360                 }
1361                 mp->m_sb.sb_dblocks = lcounter;
1362                 return 0;
1363         case XFS_SBS_AGCOUNT:
1364                 scounter = mp->m_sb.sb_agcount;
1365                 scounter += delta;
1366                 if (scounter < 0) {
1367                         ASSERT(0);
1368                         return XFS_ERROR(EINVAL);
1369                 }
1370                 mp->m_sb.sb_agcount = scounter;
1371                 return 0;
1372         case XFS_SBS_IMAX_PCT:
1373                 scounter = mp->m_sb.sb_imax_pct;
1374                 scounter += delta;
1375                 if (scounter < 0) {
1376                         ASSERT(0);
1377                         return XFS_ERROR(EINVAL);
1378                 }
1379                 mp->m_sb.sb_imax_pct = scounter;
1380                 return 0;
1381         case XFS_SBS_REXTSIZE:
1382                 scounter = mp->m_sb.sb_rextsize;
1383                 scounter += delta;
1384                 if (scounter < 0) {
1385                         ASSERT(0);
1386                         return XFS_ERROR(EINVAL);
1387                 }
1388                 mp->m_sb.sb_rextsize = scounter;
1389                 return 0;
1390         case XFS_SBS_RBMBLOCKS:
1391                 scounter = mp->m_sb.sb_rbmblocks;
1392                 scounter += delta;
1393                 if (scounter < 0) {
1394                         ASSERT(0);
1395                         return XFS_ERROR(EINVAL);
1396                 }
1397                 mp->m_sb.sb_rbmblocks = scounter;
1398                 return 0;
1399         case XFS_SBS_RBLOCKS:
1400                 lcounter = (long long)mp->m_sb.sb_rblocks;
1401                 lcounter += delta;
1402                 if (lcounter < 0) {
1403                         ASSERT(0);
1404                         return XFS_ERROR(EINVAL);
1405                 }
1406                 mp->m_sb.sb_rblocks = lcounter;
1407                 return 0;
1408         case XFS_SBS_REXTENTS:
1409                 lcounter = (long long)mp->m_sb.sb_rextents;
1410                 lcounter += delta;
1411                 if (lcounter < 0) {
1412                         ASSERT(0);
1413                         return XFS_ERROR(EINVAL);
1414                 }
1415                 mp->m_sb.sb_rextents = lcounter;
1416                 return 0;
1417         case XFS_SBS_REXTSLOG:
1418                 scounter = mp->m_sb.sb_rextslog;
1419                 scounter += delta;
1420                 if (scounter < 0) {
1421                         ASSERT(0);
1422                         return XFS_ERROR(EINVAL);
1423                 }
1424                 mp->m_sb.sb_rextslog = scounter;
1425                 return 0;
1426         default:
1427                 ASSERT(0);
1428                 return XFS_ERROR(EINVAL);
1429         }
1430 }
1431
1432 /*
1433  * xfs_mod_incore_sb() is used to change a field in the in-core
1434  * superblock structure by the specified delta.  This modification
1435  * is protected by the SB_LOCK.  Just use the xfs_mod_incore_sb_unlocked()
1436  * routine to do the work.
1437  */
1438 int
1439 xfs_mod_incore_sb(xfs_mount_t *mp, xfs_sb_field_t field, int delta, int rsvd)
1440 {
1441         unsigned long   s;
1442         int     status;
1443
1444         /* check for per-cpu counters */
1445         switch (field) {
1446 #ifdef HAVE_PERCPU_SB
1447         case XFS_SBS_ICOUNT:
1448         case XFS_SBS_IFREE:
1449         case XFS_SBS_FDBLOCKS:
1450                 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1451                         status = xfs_icsb_modify_counters(mp, field,
1452                                                         delta, rsvd);
1453                         break;
1454                 }
1455                 /* FALLTHROUGH */
1456 #endif
1457         default:
1458                 s = XFS_SB_LOCK(mp);
1459                 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1460                 XFS_SB_UNLOCK(mp, s);
1461                 break;
1462         }
1463
1464         return status;
1465 }
1466
1467 /*
1468  * xfs_mod_incore_sb_batch() is used to change more than one field
1469  * in the in-core superblock structure at a time.  This modification
1470  * is protected by a lock internal to this module.  The fields and
1471  * changes to those fields are specified in the array of xfs_mod_sb
1472  * structures passed in.
1473  *
1474  * Either all of the specified deltas will be applied or none of
1475  * them will.  If any modified field dips below 0, then all modifications
1476  * will be backed out and EINVAL will be returned.
1477  */
1478 int
1479 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1480 {
1481         unsigned long   s;
1482         int             status=0;
1483         xfs_mod_sb_t    *msbp;
1484
1485         /*
1486          * Loop through the array of mod structures and apply each
1487          * individually.  If any fail, then back out all those
1488          * which have already been applied.  Do all of this within
1489          * the scope of the SB_LOCK so that all of the changes will
1490          * be atomic.
1491          */
1492         s = XFS_SB_LOCK(mp);
1493         msbp = &msb[0];
1494         for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1495                 /*
1496                  * Apply the delta at index n.  If it fails, break
1497                  * from the loop so we'll fall into the undo loop
1498                  * below.
1499                  */
1500                 switch (msbp->msb_field) {
1501 #ifdef HAVE_PERCPU_SB
1502                 case XFS_SBS_ICOUNT:
1503                 case XFS_SBS_IFREE:
1504                 case XFS_SBS_FDBLOCKS:
1505                         if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1506                                 status = xfs_icsb_modify_counters_locked(mp,
1507                                                         msbp->msb_field,
1508                                                         msbp->msb_delta, rsvd);
1509                                 break;
1510                         }
1511                         /* FALLTHROUGH */
1512 #endif
1513                 default:
1514                         status = xfs_mod_incore_sb_unlocked(mp,
1515                                                 msbp->msb_field,
1516                                                 msbp->msb_delta, rsvd);
1517                         break;
1518                 }
1519
1520                 if (status != 0) {
1521                         break;
1522                 }
1523         }
1524
1525         /*
1526          * If we didn't complete the loop above, then back out
1527          * any changes made to the superblock.  If you add code
1528          * between the loop above and here, make sure that you
1529          * preserve the value of status. Loop back until
1530          * we step below the beginning of the array.  Make sure
1531          * we don't touch anything back there.
1532          */
1533         if (status != 0) {
1534                 msbp--;
1535                 while (msbp >= msb) {
1536                         switch (msbp->msb_field) {
1537 #ifdef HAVE_PERCPU_SB
1538                         case XFS_SBS_ICOUNT:
1539                         case XFS_SBS_IFREE:
1540                         case XFS_SBS_FDBLOCKS:
1541                                 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1542                                         status =
1543                                             xfs_icsb_modify_counters_locked(mp,
1544                                                         msbp->msb_field,
1545                                                         -(msbp->msb_delta),
1546                                                         rsvd);
1547                                         break;
1548                                 }
1549                                 /* FALLTHROUGH */
1550 #endif
1551                         default:
1552                                 status = xfs_mod_incore_sb_unlocked(mp,
1553                                                         msbp->msb_field,
1554                                                         -(msbp->msb_delta),
1555                                                         rsvd);
1556                                 break;
1557                         }
1558                         ASSERT(status == 0);
1559                         msbp--;
1560                 }
1561         }
1562         XFS_SB_UNLOCK(mp, s);
1563         return status;
1564 }
1565
1566 /*
1567  * xfs_getsb() is called to obtain the buffer for the superblock.
1568  * The buffer is returned locked and read in from disk.
1569  * The buffer should be released with a call to xfs_brelse().
1570  *
1571  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1572  * the superblock buffer if it can be locked without sleeping.
1573  * If it can't then we'll return NULL.
1574  */
1575 xfs_buf_t *
1576 xfs_getsb(
1577         xfs_mount_t     *mp,
1578         int             flags)
1579 {
1580         xfs_buf_t       *bp;
1581         int             extra_flags = 0;
1582         unsigned int    sector_size;
1583
1584
1585         bp = mp->m_sb_bp;
1586         sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
1587 #ifdef NOT
1588         /* MANAGED buf's appear broken in FreeBSD
1589          * but it's unclear if we need a persistant superblock?
1590          * since we now translate the ondisk superblock to
1591          * a separate translated structure and then translate that
1592          * structure back when we want to write the superblock
1593          */
1594         extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
1595         extra_flags = XFS_BUF_MANAGE;
1596 #endif
1597
1598         mp->m_sb_bp = bp
1599           = xfs_buf_read_flags(mp->m_ddev_targp,
1600                                XFS_SB_DADDR,
1601                                BTOBB(sector_size),
1602                                extra_flags);
1603
1604         XFS_BUF_HOLD(bp);
1605         ASSERT(XFS_BUF_ISDONE(bp));
1606         if (!XFS_BUF_ISDONE(bp)){
1607                 printf("xfs_getsb: %p bp flags 0x%x\n",bp,bp->b_flags);
1608         }
1609         return bp;
1610 }
1611
1612 /*
1613  * Used to free the superblock along various error paths.
1614  */
1615 void
1616 xfs_freesb(
1617         xfs_mount_t     *mp)
1618 {
1619         xfs_buf_t       *bp;
1620
1621         /*
1622          * Use xfs_getsb() so that the buffer will be locked
1623          * when we call xfs_buf_relse().
1624          */
1625         bp = xfs_getsb(mp, 0);
1626         XFS_BUF_UNMANAGE(bp);
1627         xfs_buf_relse(bp);
1628         mp->m_sb_bp = NULL;
1629 }
1630
1631 /*
1632  * See if the UUID is unique among mounted XFS filesystems.
1633  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1634  */
1635 STATIC int
1636 xfs_uuid_mount(
1637         xfs_mount_t     *mp)
1638 {
1639         if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1640                 cmn_err(CE_WARN,
1641                         "XFS: Filesystem %s has nil UUID - can't mount",
1642                         mp->m_fsname);
1643                 return -1;
1644         }
1645         if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1646                 cmn_err(CE_WARN,
1647                         "XFS: Filesystem %s has duplicate UUID - can't mount",
1648                         mp->m_fsname);
1649                 return -1;
1650         }
1651         return 0;
1652 }
1653
1654 /*
1655  * Remove filesystem from the UUID table.
1656  */
1657 STATIC void
1658 xfs_uuid_unmount(
1659         xfs_mount_t     *mp)
1660 {
1661         uuid_table_remove(&mp->m_sb.sb_uuid);
1662 }
1663
1664 /*
1665  * Used to log changes to the superblock unit and width fields which could
1666  * be altered by the mount options. Only the first superblock is updated.
1667  */
1668 STATIC void
1669 xfs_mount_log_sbunit(
1670         xfs_mount_t     *mp,
1671         __int64_t       fields)
1672 {
1673         xfs_trans_t     *tp;
1674
1675         ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
1676
1677         tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1678         if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1679                                 XFS_DEFAULT_LOG_COUNT)) {
1680                 xfs_trans_cancel(tp, 0);
1681                 return;
1682         }
1683         xfs_mod_sb(tp, fields);
1684         xfs_trans_commit(tp, 0, NULL);
1685 }
1686
1687
1688 #ifdef HAVE_PERCPU_SB
1689 /*
1690  * Per-cpu incore superblock counters
1691  *
1692  * Simple concept, difficult implementation
1693  *
1694  * Basically, replace the incore superblock counters with a distributed per cpu
1695  * counter for contended fields (e.g.  free block count).
1696  *
1697  * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1698  * hence needs to be accurately read when we are running low on space. Hence
1699  * there is a method to enable and disable the per-cpu counters based on how
1700  * much "stuff" is available in them.
1701  *
1702  * Basically, a counter is enabled if there is enough free resource to justify
1703  * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1704  * ENOSPC), then we disable the counters to synchronise all callers and
1705  * re-distribute the available resources.
1706  *
1707  * If, once we redistributed the available resources, we still get a failure,
1708  * we disable the per-cpu counter and go through the slow path.
1709  *
1710  * The slow path is the current xfs_mod_incore_sb() function.  This means that
1711  * when we disable a per-cpu counter, we need to drain it's resources back to
1712  * the global superblock. We do this after disabling the counter to prevent
1713  * more threads from queueing up on the counter.
1714  *
1715  * Essentially, this means that we still need a lock in the fast path to enable
1716  * synchronisation between the global counters and the per-cpu counters. This
1717  * is not a problem because the lock will be local to a CPU almost all the time
1718  * and have little contention except when we get to ENOSPC conditions.
1719  *
1720  * Basically, this lock becomes a barrier that enables us to lock out the fast
1721  * path while we do things like enabling and disabling counters and
1722  * synchronising the counters.
1723  *
1724  * Locking rules:
1725  *
1726  *      1. XFS_SB_LOCK() before picking up per-cpu locks
1727  *      2. per-cpu locks always picked up via for_each_online_cpu() order
1728  *      3. accurate counter sync requires XFS_SB_LOCK + per cpu locks
1729  *      4. modifying per-cpu counters requires holding per-cpu lock
1730  *      5. modifying global counters requires holding XFS_SB_LOCK
1731  *      6. enabling or disabling a counter requires holding the XFS_SB_LOCK
1732  *         and _none_ of the per-cpu locks.
1733  *
1734  * Disabled counters are only ever re-enabled by a balance operation
1735  * that results in more free resources per CPU than a given threshold.
1736  * To ensure counters don't remain disabled, they are rebalanced when
1737  * the global resource goes above a higher threshold (i.e. some hysteresis
1738  * is present to prevent thrashing).
1739  */
1740
1741 /*
1742  * hot-plug CPU notifier support.
1743  *
1744  * We cannot use the hotcpu_register() function because it does
1745  * not allow notifier instances. We need a notifier per filesystem
1746  * as we need to be able to identify the filesystem to balance
1747  * the counters out. This is achieved by having a notifier block
1748  * embedded in the xfs_mount_t and doing pointer magic to get the
1749  * mount pointer from the notifier block address.
1750  */
1751 STATIC int
1752 xfs_icsb_cpu_notify(
1753         struct notifier_block *nfb,
1754         unsigned long action,
1755         void *hcpu)
1756 {
1757         xfs_icsb_cnts_t *cntp;
1758         xfs_mount_t     *mp;
1759         int             s;
1760
1761         mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1762         cntp = (xfs_icsb_cnts_t *)
1763                         per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1764         switch (action) {
1765         case CPU_UP_PREPARE:
1766                 /* Easy Case - initialize the area and locks, and
1767                  * then rebalance when online does everything else for us. */
1768                 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1769                 break;
1770         case CPU_ONLINE:
1771                 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1772                 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1773                 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1774                 break;
1775         case CPU_DEAD:
1776                 /* Disable all the counters, then fold the dead cpu's
1777                  * count into the total on the global superblock and
1778                  * re-enable the counters. */
1779                 s = XFS_SB_LOCK(mp);
1780                 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1781                 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1782                 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1783
1784                 mp->m_sb.sb_icount += cntp->icsb_icount;
1785                 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1786                 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1787
1788                 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1789
1790                 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, XFS_ICSB_SB_LOCKED);
1791                 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, XFS_ICSB_SB_LOCKED);
1792                 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, XFS_ICSB_SB_LOCKED);
1793                 XFS_SB_UNLOCK(mp, s);
1794                 break;
1795         }
1796
1797         return NOTIFY_OK;
1798 }
1799
1800 int
1801 xfs_icsb_init_counters(
1802         xfs_mount_t     *mp)
1803 {
1804         xfs_icsb_cnts_t *cntp;
1805         int             i;
1806
1807         mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1808         if (mp->m_sb_cnts == NULL)
1809                 return -ENOMEM;
1810
1811         mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1812         mp->m_icsb_notifier.priority = 0;
1813         register_cpu_notifier(&mp->m_icsb_notifier);
1814
1815         for_each_online_cpu(i) {
1816                 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1817                 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1818         }
1819         /*
1820          * start with all counters disabled so that the
1821          * initial balance kicks us off correctly
1822          */
1823         mp->m_icsb_counters = -1;
1824         return 0;
1825 }
1826
1827 STATIC void
1828 xfs_icsb_destroy_counters(
1829         xfs_mount_t     *mp)
1830 {
1831         if (mp->m_sb_cnts) {
1832                 unregister_cpu_notifier(&mp->m_icsb_notifier);
1833                 free_percpu(mp->m_sb_cnts);
1834         }
1835 }
1836
1837 STATIC inline void
1838 xfs_icsb_lock_cntr(
1839         xfs_icsb_cnts_t *icsbp)
1840 {
1841         while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1842                 ndelay(1000);
1843         }
1844 }
1845
1846 STATIC inline void
1847 xfs_icsb_unlock_cntr(
1848         xfs_icsb_cnts_t *icsbp)
1849 {
1850         clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1851 }
1852
1853
1854 STATIC inline void
1855 xfs_icsb_lock_all_counters(
1856         xfs_mount_t     *mp)
1857 {
1858         xfs_icsb_cnts_t *cntp;
1859         int             i;
1860
1861         for_each_online_cpu(i) {
1862                 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1863                 xfs_icsb_lock_cntr(cntp);
1864         }
1865 }
1866
1867 STATIC inline void
1868 xfs_icsb_unlock_all_counters(
1869         xfs_mount_t     *mp)
1870 {
1871         xfs_icsb_cnts_t *cntp;
1872         int             i;
1873
1874         for_each_online_cpu(i) {
1875                 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1876                 xfs_icsb_unlock_cntr(cntp);
1877         }
1878 }
1879
1880 STATIC void
1881 xfs_icsb_count(
1882         xfs_mount_t     *mp,
1883         xfs_icsb_cnts_t *cnt,
1884         int             flags)
1885 {
1886         xfs_icsb_cnts_t *cntp;
1887         int             i;
1888
1889         memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1890
1891         if (!(flags & XFS_ICSB_LAZY_COUNT))
1892                 xfs_icsb_lock_all_counters(mp);
1893
1894         for_each_online_cpu(i) {
1895                 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1896                 cnt->icsb_icount += cntp->icsb_icount;
1897                 cnt->icsb_ifree += cntp->icsb_ifree;
1898                 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1899         }
1900
1901         if (!(flags & XFS_ICSB_LAZY_COUNT))
1902                 xfs_icsb_unlock_all_counters(mp);
1903 }
1904
1905 STATIC int
1906 xfs_icsb_counter_disabled(
1907         xfs_mount_t     *mp,
1908         xfs_sb_field_t  field)
1909 {
1910         ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1911         return test_bit(field, &mp->m_icsb_counters);
1912 }
1913
1914 STATIC int
1915 xfs_icsb_disable_counter(
1916         xfs_mount_t     *mp,
1917         xfs_sb_field_t  field)
1918 {
1919         xfs_icsb_cnts_t cnt;
1920
1921         ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1922
1923         xfs_icsb_lock_all_counters(mp);
1924         if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1925                 /* drain back to superblock */
1926
1927                 xfs_icsb_count(mp, &cnt, XFS_ICSB_SB_LOCKED|XFS_ICSB_LAZY_COUNT);
1928                 switch(field) {
1929                 case XFS_SBS_ICOUNT:
1930                         mp->m_sb.sb_icount = cnt.icsb_icount;
1931                         break;
1932                 case XFS_SBS_IFREE:
1933                         mp->m_sb.sb_ifree = cnt.icsb_ifree;
1934                         break;
1935                 case XFS_SBS_FDBLOCKS:
1936                         mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1937                         break;
1938                 default:
1939                         BUG();
1940                 }
1941         }
1942
1943         xfs_icsb_unlock_all_counters(mp);
1944
1945         return 0;
1946 }
1947
1948 STATIC void
1949 xfs_icsb_enable_counter(
1950         xfs_mount_t     *mp,
1951         xfs_sb_field_t  field,
1952         uint64_t        count,
1953         uint64_t        resid)
1954 {
1955         xfs_icsb_cnts_t *cntp;
1956         int             i;
1957
1958         ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1959
1960         xfs_icsb_lock_all_counters(mp);
1961         for_each_online_cpu(i) {
1962                 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1963                 switch (field) {
1964                 case XFS_SBS_ICOUNT:
1965                         cntp->icsb_icount = count + resid;
1966                         break;
1967                 case XFS_SBS_IFREE:
1968                         cntp->icsb_ifree = count + resid;
1969                         break;
1970                 case XFS_SBS_FDBLOCKS:
1971                         cntp->icsb_fdblocks = count + resid;
1972                         break;
1973                 default:
1974                         BUG();
1975                         break;
1976                 }
1977                 resid = 0;
1978         }
1979         clear_bit(field, &mp->m_icsb_counters);
1980         xfs_icsb_unlock_all_counters(mp);
1981 }
1982
1983 STATIC void
1984 xfs_icsb_sync_counters_int(
1985         xfs_mount_t     *mp,
1986         int             flags)
1987 {
1988         xfs_icsb_cnts_t cnt;
1989         int             s;
1990
1991         /* Pass 1: lock all counters */
1992         if ((flags & XFS_ICSB_SB_LOCKED) == 0)
1993                 s = XFS_SB_LOCK(mp);
1994
1995         xfs_icsb_count(mp, &cnt, flags);
1996
1997         /* Step 3: update mp->m_sb fields */
1998         if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
1999                 mp->m_sb.sb_icount = cnt.icsb_icount;
2000         if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2001                 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2002         if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2003                 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2004
2005         if ((flags & XFS_ICSB_SB_LOCKED) == 0)
2006                 XFS_SB_UNLOCK(mp, s);
2007 }
2008
2009 /*
2010  * Accurate update of per-cpu counters to incore superblock
2011  */
2012 STATIC void
2013 xfs_icsb_sync_counters(
2014         xfs_mount_t     *mp)
2015 {
2016         xfs_icsb_sync_counters_int(mp, 0);
2017 }
2018
2019 /*
2020  * lazy addition used for things like df, background sb syncs, etc
2021  */
2022 void
2023 xfs_icsb_sync_counters_lazy(
2024         xfs_mount_t     *mp)
2025 {
2026         xfs_icsb_sync_counters_int(mp, XFS_ICSB_LAZY_COUNT);
2027 }
2028
2029 /*
2030  * Balance and enable/disable counters as necessary.
2031  *
2032  * Thresholds for re-enabling counters are somewhat magic.
2033  * inode counts are chosen to be the same number as single
2034  * on disk allocation chunk per CPU, and free blocks is
2035  * something far enough zero that we aren't going thrash
2036  * when we get near ENOSPC.
2037  */
2038 #define XFS_ICSB_INO_CNTR_REENABLE      64
2039 #define XFS_ICSB_FDBLK_CNTR_REENABLE    512
2040 STATIC void
2041 xfs_icsb_balance_counter(
2042         xfs_mount_t     *mp,
2043         xfs_sb_field_t  field,
2044         int             flags)
2045 {
2046         uint64_t        count, resid = 0;
2047         int             weight = num_online_cpus();
2048         int             s;
2049
2050         if (!(flags & XFS_ICSB_SB_LOCKED))
2051                 s = XFS_SB_LOCK(mp);
2052
2053         /* disable counter and sync counter */
2054         xfs_icsb_disable_counter(mp, field);
2055
2056         /* update counters  - first CPU gets residual*/
2057         switch (field) {
2058         case XFS_SBS_ICOUNT:
2059                 count = mp->m_sb.sb_icount;
2060                 resid = do_div(count, weight);
2061                 if (count < XFS_ICSB_INO_CNTR_REENABLE)
2062                         goto out;
2063                 break;
2064         case XFS_SBS_IFREE:
2065                 count = mp->m_sb.sb_ifree;
2066                 resid = do_div(count, weight);
2067                 if (count < XFS_ICSB_INO_CNTR_REENABLE)
2068                         goto out;
2069                 break;
2070         case XFS_SBS_FDBLOCKS:
2071                 count = mp->m_sb.sb_fdblocks;
2072                 resid = do_div(count, weight);
2073                 if (count < XFS_ICSB_FDBLK_CNTR_REENABLE)
2074                         goto out;
2075                 break;
2076         default:
2077                 BUG();
2078                 break;
2079         }
2080
2081         xfs_icsb_enable_counter(mp, field, count, resid);
2082 out:
2083         if (!(flags & XFS_ICSB_SB_LOCKED))
2084                 XFS_SB_UNLOCK(mp, s);
2085 }
2086
2087 STATIC int
2088 xfs_icsb_modify_counters_int(
2089         xfs_mount_t     *mp,
2090         xfs_sb_field_t  field,
2091         int             delta,
2092         int             rsvd,
2093         int             flags)
2094 {
2095         xfs_icsb_cnts_t *icsbp;
2096         long long       lcounter;       /* long counter for 64 bit fields */
2097         int             cpu, s, locked = 0;
2098         int             ret = 0, balance_done = 0;
2099
2100 again:
2101         cpu = get_cpu();
2102         icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu),
2103         xfs_icsb_lock_cntr(icsbp);
2104         if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2105                 goto slow_path;
2106
2107         switch (field) {
2108         case XFS_SBS_ICOUNT:
2109                 lcounter = icsbp->icsb_icount;
2110                 lcounter += delta;
2111                 if (unlikely(lcounter < 0))
2112                         goto slow_path;
2113                 icsbp->icsb_icount = lcounter;
2114                 break;
2115
2116         case XFS_SBS_IFREE:
2117                 lcounter = icsbp->icsb_ifree;
2118                 lcounter += delta;
2119                 if (unlikely(lcounter < 0))
2120                         goto slow_path;
2121                 icsbp->icsb_ifree = lcounter;
2122                 break;
2123
2124         case XFS_SBS_FDBLOCKS:
2125                 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2126
2127                 lcounter = icsbp->icsb_fdblocks;
2128                 lcounter += delta;
2129                 if (unlikely(lcounter < 0))
2130                         goto slow_path;
2131                 icsbp->icsb_fdblocks = lcounter;
2132                 break;
2133         default:
2134                 BUG();
2135                 break;
2136         }
2137         xfs_icsb_unlock_cntr(icsbp);
2138         put_cpu();
2139         if (locked)
2140                 XFS_SB_UNLOCK(mp, s);
2141         return 0;
2142
2143         /*
2144          * The slow path needs to be run with the SBLOCK
2145          * held so that we prevent other threads from
2146          * attempting to run this path at the same time.
2147          * this provides exclusion for the balancing code,
2148          * and exclusive fallback if the balance does not
2149          * provide enough resources to continue in an unlocked
2150          * manner.
2151          */
2152 slow_path:
2153         xfs_icsb_unlock_cntr(icsbp);
2154         put_cpu();
2155
2156         /* need to hold superblock incase we need
2157          * to disable a counter */
2158         if (!(flags & XFS_ICSB_SB_LOCKED)) {
2159                 s = XFS_SB_LOCK(mp);
2160                 locked = 1;
2161                 flags |= XFS_ICSB_SB_LOCKED;
2162         }
2163         if (!balance_done) {
2164                 xfs_icsb_balance_counter(mp, field, flags);
2165                 balance_done = 1;
2166                 goto again;
2167         } else {
2168                 /*
2169                  * we might not have enough on this local
2170                  * cpu to allocate for a bulk request.
2171                  * We need to drain this field from all CPUs
2172                  * and disable the counter fastpath
2173                  */
2174                 xfs_icsb_disable_counter(mp, field);
2175         }
2176
2177         ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2178
2179         if (locked)
2180                 XFS_SB_UNLOCK(mp, s);
2181         return ret;
2182 }
2183
2184 STATIC int
2185 xfs_icsb_modify_counters(
2186         xfs_mount_t     *mp,
2187         xfs_sb_field_t  field,
2188         int             delta,
2189         int             rsvd)
2190 {
2191         return xfs_icsb_modify_counters_int(mp, field, delta, rsvd, 0);
2192 }
2193
2194 /*
2195  * Called when superblock is already locked
2196  */
2197 STATIC int
2198 xfs_icsb_modify_counters_locked(
2199         xfs_mount_t     *mp,
2200         xfs_sb_field_t  field,
2201         int             delta,
2202         int             rsvd)
2203 {
2204         return xfs_icsb_modify_counters_int(mp, field, delta,
2205                                                 rsvd, XFS_ICSB_SB_LOCKED);
2206 }
2207 #endif