fs/xfs/xfs_super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6
   7 #include "xfs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_sb.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_btree.h"
  16 #include "xfs_bmap.h"
  17 #include "xfs_alloc.h"
  18 #include "xfs_fsops.h"
  19 #include "xfs_trans.h"
  20 #include "xfs_buf_item.h"
  21 #include "xfs_log.h"
  22 #include "xfs_log_priv.h"
  23 #include "xfs_dir2.h"
  24 #include "xfs_extfree_item.h"
  25 #include "xfs_mru_cache.h"
  26 #include "xfs_inode_item.h"
  27 #include "xfs_icache.h"
  28 #include "xfs_trace.h"
  29 #include "xfs_icreate_item.h"
  30 #include "xfs_filestream.h"
  31 #include "xfs_quota.h"
  32 #include "xfs_sysfs.h"
  33 #include "xfs_ondisk.h"
  34 #include "xfs_rmap_item.h"
  35 #include "xfs_refcount_item.h"
  36 #include "xfs_bmap_item.h"
  37 #include "xfs_reflink.h"
  38
  39 #include <linux/magic.h>
  40 #include <linux/fs_context.h>
  41 #include <linux/fs_parser.h>
  42
  43 static const struct super_operations xfs_super_operations;
  44
  45 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  46 #ifdef DEBUG
  47 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  48 #endif
  49
  50 /*
  51  * Table driven mount option parser.
  52  */
  53 enum {
  54         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
  55         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
  56         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
  57         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
  58         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
  59         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
  60         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
  61         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
  62         Opt_discard, Opt_nodiscard, Opt_dax,
  63 };
  64
  65 static const struct fs_parameter_spec xfs_fs_parameters[] = {
  66         fsparam_u32("logbufs",          Opt_logbufs),
  67         fsparam_string("logbsize",      Opt_logbsize),
  68         fsparam_string("logdev",        Opt_logdev),
  69         fsparam_string("rtdev",         Opt_rtdev),
  70         fsparam_flag("wsync",           Opt_wsync),
  71         fsparam_flag("noalign",         Opt_noalign),
  72         fsparam_flag("swalloc",         Opt_swalloc),
  73         fsparam_u32("sunit",            Opt_sunit),
  74         fsparam_u32("swidth",           Opt_swidth),
  75         fsparam_flag("nouuid",          Opt_nouuid),
  76         fsparam_flag("grpid",           Opt_grpid),
  77         fsparam_flag("nogrpid",         Opt_nogrpid),
  78         fsparam_flag("bsdgroups",       Opt_bsdgroups),
  79         fsparam_flag("sysvgroups",      Opt_sysvgroups),
  80         fsparam_string("allocsize",     Opt_allocsize),
  81         fsparam_flag("norecovery",      Opt_norecovery),
  82         fsparam_flag("inode64",         Opt_inode64),
  83         fsparam_flag("inode32",         Opt_inode32),
  84         fsparam_flag("ikeep",           Opt_ikeep),
  85         fsparam_flag("noikeep",         Opt_noikeep),
  86         fsparam_flag("largeio",         Opt_largeio),
  87         fsparam_flag("nolargeio",       Opt_nolargeio),
  88         fsparam_flag("attr2",           Opt_attr2),
  89         fsparam_flag("noattr2",         Opt_noattr2),
  90         fsparam_flag("filestreams",     Opt_filestreams),
  91         fsparam_flag("quota",           Opt_quota),
  92         fsparam_flag("noquota",         Opt_noquota),
  93         fsparam_flag("usrquota",        Opt_usrquota),
  94         fsparam_flag("grpquota",        Opt_grpquota),
  95         fsparam_flag("prjquota",        Opt_prjquota),
  96         fsparam_flag("uquota",          Opt_uquota),
  97         fsparam_flag("gquota",          Opt_gquota),
  98         fsparam_flag("pquota",          Opt_pquota),
  99         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 100         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 101         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 102         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 103         fsparam_flag("discard",         Opt_discard),
 104         fsparam_flag("nodiscard",       Opt_nodiscard),
 105         fsparam_flag("dax",             Opt_dax),
 106         {}
 107 };
 108
 109 struct proc_xfs_info {
 110         uint64_t        flag;
 111         char            *str;
 112 };
 113
 114 static int
 115 xfs_fs_show_options(
 116         struct seq_file         *m,
 117         struct dentry           *root)
 118 {
 119         static struct proc_xfs_info xfs_info_set[] = {
 120                 /* the few simple ones we can get from the mount struct */
 121                 { XFS_MOUNT_IKEEP,              ",ikeep" },
 122                 { XFS_MOUNT_WSYNC,              ",wsync" },
 123                 { XFS_MOUNT_NOALIGN,            ",noalign" },
 124                 { XFS_MOUNT_SWALLOC,            ",swalloc" },
 125                 { XFS_MOUNT_NOUUID,             ",nouuid" },
 126                 { XFS_MOUNT_NORECOVERY,         ",norecovery" },
 127                 { XFS_MOUNT_ATTR2,              ",attr2" },
 128                 { XFS_MOUNT_FILESTREAMS,        ",filestreams" },
 129                 { XFS_MOUNT_GRPID,              ",grpid" },
 130                 { XFS_MOUNT_DISCARD,            ",discard" },
 131                 { XFS_MOUNT_LARGEIO,            ",largeio" },
 132                 { XFS_MOUNT_DAX,                ",dax" },
 133                 { 0, NULL }
 134         };
 135         struct xfs_mount        *mp = XFS_M(root->d_sb);
 136         struct proc_xfs_info    *xfs_infop;
 137
 138         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 139                 if (mp->m_flags & xfs_infop->flag)
 140                         seq_puts(m, xfs_infop->str);
 141         }
 142
 143         seq_printf(m, ",inode%d",
 144                 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
 145
 146         if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
 147                 seq_printf(m, ",allocsize=%dk",
 148                            (1 << mp->m_allocsize_log) >> 10);
 149
 150         if (mp->m_logbufs > 0)
 151                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 152         if (mp->m_logbsize > 0)
 153                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 154
 155         if (mp->m_logname)
 156                 seq_show_option(m, "logdev", mp->m_logname);
 157         if (mp->m_rtname)
 158                 seq_show_option(m, "rtdev", mp->m_rtname);
 159
 160         if (mp->m_dalign > 0)
 161                 seq_printf(m, ",sunit=%d",
 162                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 163         if (mp->m_swidth > 0)
 164                 seq_printf(m, ",swidth=%d",
 165                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 166
 167         if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
 168                 seq_puts(m, ",usrquota");
 169         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 170                 seq_puts(m, ",uqnoenforce");
 171
 172         if (mp->m_qflags & XFS_PQUOTA_ACCT) {
 173                 if (mp->m_qflags & XFS_PQUOTA_ENFD)
 174                         seq_puts(m, ",prjquota");
 175                 else
 176                         seq_puts(m, ",pqnoenforce");
 177         }
 178         if (mp->m_qflags & XFS_GQUOTA_ACCT) {
 179                 if (mp->m_qflags & XFS_GQUOTA_ENFD)
 180                         seq_puts(m, ",grpquota");
 181                 else
 182                         seq_puts(m, ",gqnoenforce");
 183         }
 184
 185         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 186                 seq_puts(m, ",noquota");
 187
 188         return 0;
 189 }
 190
 191 static uint64_t
 192 xfs_max_file_offset(
 193         unsigned int            blockshift)
 194 {
 195         unsigned int            pagefactor = 1;
 196         unsigned int            bitshift = BITS_PER_LONG - 1;
 197
 198         /* Figure out maximum filesize, on Linux this can depend on
 199          * the filesystem blocksize (on 32 bit platforms).
 200          * __block_write_begin does this in an [unsigned] long long...
 201          *      page->index << (PAGE_SHIFT - bbits)
 202          * So, for page sized blocks (4K on 32 bit platforms),
 203          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 204          *      (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
 205          * but for smaller blocksizes it is less (bbits = log2 bsize).
 206          */
 207
 208 #if BITS_PER_LONG == 32
 209         ASSERT(sizeof(sector_t) == 8);
 210         pagefactor = PAGE_SIZE;
 211         bitshift = BITS_PER_LONG;
 212 #endif
 213
 214         return (((uint64_t)pagefactor) << bitshift) - 1;
 215 }
 216
 217 /*
 218  * Set parameters for inode allocation heuristics, taking into account
 219  * filesystem size and inode32/inode64 mount options; i.e. specifically
 220  * whether or not XFS_MOUNT_SMALL_INUMS is set.
 221  *
 222  * Inode allocation patterns are altered only if inode32 is requested
 223  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
 224  * If altered, XFS_MOUNT_32BITINODES is set as well.
 225  *
 226  * An agcount independent of that in the mount structure is provided
 227  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 228  * to the potentially higher ag count.
 229  *
 230  * Returns the maximum AG index which may contain inodes.
 231  */
 232 xfs_agnumber_t
 233 xfs_set_inode_alloc(
 234         struct xfs_mount *mp,
 235         xfs_agnumber_t  agcount)
 236 {
 237         xfs_agnumber_t  index;
 238         xfs_agnumber_t  maxagi = 0;
 239         xfs_sb_t        *sbp = &mp->m_sb;
 240         xfs_agnumber_t  max_metadata;
 241         xfs_agino_t     agino;
 242         xfs_ino_t       ino;
 243
 244         /*
 245          * Calculate how much should be reserved for inodes to meet
 246          * the max inode percentage.  Used only for inode32.
 247          */
 248         if (M_IGEO(mp)->maxicount) {
 249                 uint64_t        icount;
 250
 251                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 252                 do_div(icount, 100);
 253                 icount += sbp->sb_agblocks - 1;
 254                 do_div(icount, sbp->sb_agblocks);
 255                 max_metadata = icount;
 256         } else {
 257                 max_metadata = agcount;
 258         }
 259
 260         /* Get the last possible inode in the filesystem */
 261         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 262         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 263
 264         /*
 265          * If user asked for no more than 32-bit inodes, and the fs is
 266          * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
 267          * the allocator to accommodate the request.
 268          */
 269         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
 270                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 271         else
 272                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
 273
 274         for (index = 0; index < agcount; index++) {
 275                 struct xfs_perag        *pag;
 276
 277                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 278
 279                 pag = xfs_perag_get(mp, index);
 280
 281                 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
 282                         if (ino > XFS_MAXINUMBER_32) {
 283                                 pag->pagi_inodeok = 0;
 284                                 pag->pagf_metadata = 0;
 285                         } else {
 286                                 pag->pagi_inodeok = 1;
 287                                 maxagi++;
 288                                 if (index < max_metadata)
 289                                         pag->pagf_metadata = 1;
 290                                 else
 291                                         pag->pagf_metadata = 0;
 292                         }
 293                 } else {
 294                         pag->pagi_inodeok = 1;
 295                         pag->pagf_metadata = 0;
 296                 }
 297
 298                 xfs_perag_put(pag);
 299         }
 300
 301         return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
 302 }
 303
 304 STATIC int
 305 xfs_blkdev_get(
 306         xfs_mount_t             *mp,
 307         const char              *name,
 308         struct block_device     **bdevp)
 309 {
 310         int                     error = 0;
 311
 312         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 313                                     mp);
 314         if (IS_ERR(*bdevp)) {
 315                 error = PTR_ERR(*bdevp);
 316                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 317         }
 318
 319         return error;
 320 }
 321
 322 STATIC void
 323 xfs_blkdev_put(
 324         struct block_device     *bdev)
 325 {
 326         if (bdev)
 327                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 328 }
 329
 330 void
 331 xfs_blkdev_issue_flush(
 332         xfs_buftarg_t           *buftarg)
 333 {
 334         blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
 335 }
 336
 337 STATIC void
 338 xfs_close_devices(
 339         struct xfs_mount        *mp)
 340 {
 341         struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
 342
 343         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 344                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 345                 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
 346
 347                 xfs_free_buftarg(mp->m_logdev_targp);
 348                 xfs_blkdev_put(logdev);
 349                 fs_put_dax(dax_logdev);
 350         }
 351         if (mp->m_rtdev_targp) {
 352                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 353                 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
 354
 355                 xfs_free_buftarg(mp->m_rtdev_targp);
 356                 xfs_blkdev_put(rtdev);
 357                 fs_put_dax(dax_rtdev);
 358         }
 359         xfs_free_buftarg(mp->m_ddev_targp);
 360         fs_put_dax(dax_ddev);
 361 }
 362
 363 /*
 364  * The file system configurations are:
 365  *      (1) device (partition) with data and internal log
 366  *      (2) logical volume with data and log subvolumes.
 367  *      (3) logical volume with data, log, and realtime subvolumes.
 368  *
 369  * We only have to handle opening the log and realtime volumes here if
 370  * they are present.  The data subvolume has already been opened by
 371  * get_sb_bdev() and is stored in sb->s_bdev.
 372  */
 373 STATIC int
 374 xfs_open_devices(
 375         struct xfs_mount        *mp)
 376 {
 377         struct block_device     *ddev = mp->m_super->s_bdev;
 378         struct dax_device       *dax_ddev = fs_dax_get_by_bdev(ddev);
 379         struct dax_device       *dax_logdev = NULL, *dax_rtdev = NULL;
 380         struct block_device     *logdev = NULL, *rtdev = NULL;
 381         int                     error;
 382
 383         /*
 384          * Open real time and log devices - order is important.
 385          */
 386         if (mp->m_logname) {
 387                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 388                 if (error)
 389                         goto out;
 390                 dax_logdev = fs_dax_get_by_bdev(logdev);
 391         }
 392
 393         if (mp->m_rtname) {
 394                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 395                 if (error)
 396                         goto out_close_logdev;
 397
 398                 if (rtdev == ddev || rtdev == logdev) {
 399                         xfs_warn(mp,
 400         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 401                         error = -EINVAL;
 402                         goto out_close_rtdev;
 403                 }
 404                 dax_rtdev = fs_dax_get_by_bdev(rtdev);
 405         }
 406
 407         /*
 408          * Setup xfs_mount buffer target pointers
 409          */
 410         error = -ENOMEM;
 411         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
 412         if (!mp->m_ddev_targp)
 413                 goto out_close_rtdev;
 414
 415         if (rtdev) {
 416                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
 417                 if (!mp->m_rtdev_targp)
 418                         goto out_free_ddev_targ;
 419         }
 420
 421         if (logdev && logdev != ddev) {
 422                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
 423                 if (!mp->m_logdev_targp)
 424                         goto out_free_rtdev_targ;
 425         } else {
 426                 mp->m_logdev_targp = mp->m_ddev_targp;
 427         }
 428
 429         return 0;
 430
 431  out_free_rtdev_targ:
 432         if (mp->m_rtdev_targp)
 433                 xfs_free_buftarg(mp->m_rtdev_targp);
 434  out_free_ddev_targ:
 435         xfs_free_buftarg(mp->m_ddev_targp);
 436  out_close_rtdev:
 437         xfs_blkdev_put(rtdev);
 438         fs_put_dax(dax_rtdev);
 439  out_close_logdev:
 440         if (logdev && logdev != ddev) {
 441                 xfs_blkdev_put(logdev);
 442                 fs_put_dax(dax_logdev);
 443         }
 444  out:
 445         fs_put_dax(dax_ddev);
 446         return error;
 447 }
 448
 449 /*
 450  * Setup xfs_mount buffer target pointers based on superblock
 451  */
 452 STATIC int
 453 xfs_setup_devices(
 454         struct xfs_mount        *mp)
 455 {
 456         int                     error;
 457
 458         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 459         if (error)
 460                 return error;
 461
 462         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 463                 unsigned int    log_sector_size = BBSIZE;
 464
 465                 if (xfs_sb_version_hassector(&mp->m_sb))
 466                         log_sector_size = mp->m_sb.sb_logsectsize;
 467                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 468                                             log_sector_size);
 469                 if (error)
 470                         return error;
 471         }
 472         if (mp->m_rtdev_targp) {
 473                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 474                                             mp->m_sb.sb_sectsize);
 475                 if (error)
 476                         return error;
 477         }
 478
 479         return 0;
 480 }
 481
 482 STATIC int
 483 xfs_init_mount_workqueues(
 484         struct xfs_mount        *mp)
 485 {
 486         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 487                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_super->s_id);
 488         if (!mp->m_buf_workqueue)
 489                 goto out;
 490
 491         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 492                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
 493         if (!mp->m_unwritten_workqueue)
 494                 goto out_destroy_buf;
 495
 496         mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
 497                         WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND,
 498                         0, mp->m_super->s_id);
 499         if (!mp->m_cil_workqueue)
 500                 goto out_destroy_unwritten;
 501
 502         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 503                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
 504         if (!mp->m_reclaim_workqueue)
 505                 goto out_destroy_cil;
 506
 507         mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
 508                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
 509         if (!mp->m_eofblocks_workqueue)
 510                 goto out_destroy_reclaim;
 511
 512         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
 513                                                mp->m_super->s_id);
 514         if (!mp->m_sync_workqueue)
 515                 goto out_destroy_eofb;
 516
 517         return 0;
 518
 519 out_destroy_eofb:
 520         destroy_workqueue(mp->m_eofblocks_workqueue);
 521 out_destroy_reclaim:
 522         destroy_workqueue(mp->m_reclaim_workqueue);
 523 out_destroy_cil:
 524         destroy_workqueue(mp->m_cil_workqueue);
 525 out_destroy_unwritten:
 526         destroy_workqueue(mp->m_unwritten_workqueue);
 527 out_destroy_buf:
 528         destroy_workqueue(mp->m_buf_workqueue);
 529 out:
 530         return -ENOMEM;
 531 }
 532
 533 STATIC void
 534 xfs_destroy_mount_workqueues(
 535         struct xfs_mount        *mp)
 536 {
 537         destroy_workqueue(mp->m_sync_workqueue);
 538         destroy_workqueue(mp->m_eofblocks_workqueue);
 539         destroy_workqueue(mp->m_reclaim_workqueue);
 540         destroy_workqueue(mp->m_cil_workqueue);
 541         destroy_workqueue(mp->m_unwritten_workqueue);
 542         destroy_workqueue(mp->m_buf_workqueue);
 543 }
 544
 545 /*
 546  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 547  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 548  * for IO to complete so that we effectively throttle multiple callers to the
 549  * rate at which IO is completing.
 550  */
 551 void
 552 xfs_flush_inodes(
 553         struct xfs_mount        *mp)
 554 {
 555         struct super_block      *sb = mp->m_super;
 556
 557         if (down_read_trylock(&sb->s_umount)) {
 558                 sync_inodes_sb(sb);
 559                 up_read(&sb->s_umount);
 560         }
 561 }
 562
 563 /* Catch misguided souls that try to use this interface on XFS */
 564 STATIC struct inode *
 565 xfs_fs_alloc_inode(
 566         struct super_block      *sb)
 567 {
 568         BUG();
 569         return NULL;
 570 }
 571
 572 #ifdef DEBUG
 573 static void
 574 xfs_check_delalloc(
 575         struct xfs_inode        *ip,
 576         int                     whichfork)
 577 {
 578         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 579         struct xfs_bmbt_irec    got;
 580         struct xfs_iext_cursor  icur;
 581
 582         if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
 583                 return;
 584         do {
 585                 if (isnullstartblock(got.br_startblock)) {
 586                         xfs_warn(ip->i_mount,
 587         "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
 588                                 ip->i_ino,
 589                                 whichfork == XFS_DATA_FORK ? "data" : "cow",
 590                                 got.br_startoff, got.br_blockcount);
 591                 }
 592         } while (xfs_iext_next_extent(ifp, &icur, &got));
 593 }
 594 #else
 595 #define xfs_check_delalloc(ip, whichfork)       do { } while (0)
 596 #endif
 597
 598 /*
 599  * Now that the generic code is guaranteed not to be accessing
 600  * the linux inode, we can inactivate and reclaim the inode.
 601  */
 602 STATIC void
 603 xfs_fs_destroy_inode(
 604         struct inode            *inode)
 605 {
 606         struct xfs_inode        *ip = XFS_I(inode);
 607
 608         trace_xfs_destroy_inode(ip);
 609
 610         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 611         XFS_STATS_INC(ip->i_mount, vn_rele);
 612         XFS_STATS_INC(ip->i_mount, vn_remove);
 613
 614         xfs_inactive(ip);
 615
 616         if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
 617                 xfs_check_delalloc(ip, XFS_DATA_FORK);
 618                 xfs_check_delalloc(ip, XFS_COW_FORK);
 619                 ASSERT(0);
 620         }
 621
 622         XFS_STATS_INC(ip->i_mount, vn_reclaim);
 623
 624         /*
 625          * We should never get here with one of the reclaim flags already set.
 626          */
 627         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
 628         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
 629
 630         /*
 631          * We always use background reclaim here because even if the
 632          * inode is clean, it still may be under IO and hence we have
 633          * to take the flush lock. The background reclaim path handles
 634          * this more efficiently than we can here, so simply let background
 635          * reclaim tear down all inodes.
 636          */
 637         xfs_inode_set_reclaim_tag(ip);
 638 }
 639
 640 static void
 641 xfs_fs_dirty_inode(
 642         struct inode                    *inode,
 643         int                             flag)
 644 {
 645         struct xfs_inode                *ip = XFS_I(inode);
 646         struct xfs_mount                *mp = ip->i_mount;
 647         struct xfs_trans                *tp;
 648
 649         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 650                 return;
 651         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 652                 return;
 653
 654         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 655                 return;
 656         xfs_ilock(ip, XFS_ILOCK_EXCL);
 657         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 658         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 659         xfs_trans_commit(tp);
 660 }
 661
 662 /*
 663  * Slab object creation initialisation for the XFS inode.
 664  * This covers only the idempotent fields in the XFS inode;
 665  * all other fields need to be initialised on allocation
 666  * from the slab. This avoids the need to repeatedly initialise
 667  * fields in the xfs inode that left in the initialise state
 668  * when freeing the inode.
 669  */
 670 STATIC void
 671 xfs_fs_inode_init_once(
 672         void                    *inode)
 673 {
 674         struct xfs_inode        *ip = inode;
 675
 676         memset(ip, 0, sizeof(struct xfs_inode));
 677
 678         /* vfs inode */
 679         inode_init_once(VFS_I(ip));
 680
 681         /* xfs inode */
 682         atomic_set(&ip->i_pincount, 0);
 683         spin_lock_init(&ip->i_flags_lock);
 684
 685         mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 686                      "xfsino", ip->i_ino);
 687         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 688                      "xfsino", ip->i_ino);
 689 }
 690
 691 /*
 692  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 693  * serialised against cache hits here via the inode->i_lock and igrab() in
 694  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 695  * racing with us, and it avoids needing to grab a spinlock here for every inode
 696  * we drop the final reference on.
 697  */
 698 STATIC int
 699 xfs_fs_drop_inode(
 700         struct inode            *inode)
 701 {
 702         struct xfs_inode        *ip = XFS_I(inode);
 703
 704         /*
 705          * If this unlinked inode is in the middle of recovery, don't
 706          * drop the inode just yet; log recovery will take care of
 707          * that.  See the comment for this inode flag.
 708          */
 709         if (ip->i_flags & XFS_IRECOVERY) {
 710                 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
 711                 return 0;
 712         }
 713
 714         return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
 715 }
 716
 717 static void
 718 xfs_mount_free(
 719         struct xfs_mount        *mp)
 720 {
 721         kfree(mp->m_rtname);
 722         kfree(mp->m_logname);
 723         kmem_free(mp);
 724 }
 725
 726 STATIC int
 727 xfs_fs_sync_fs(
 728         struct super_block      *sb,
 729         int                     wait)
 730 {
 731         struct xfs_mount        *mp = XFS_M(sb);
 732
 733         /*
 734          * Doing anything during the async pass would be counterproductive.
 735          */
 736         if (!wait)
 737                 return 0;
 738
 739         xfs_log_force(mp, XFS_LOG_SYNC);
 740         if (laptop_mode) {
 741                 /*
 742                  * The disk must be active because we're syncing.
 743                  * We schedule log work now (now that the disk is
 744                  * active) instead of later (when it might not be).
 745                  */
 746                 flush_delayed_work(&mp->m_log->l_work);
 747         }
 748
 749         return 0;
 750 }
 751
 752 STATIC int
 753 xfs_fs_statfs(
 754         struct dentry           *dentry,
 755         struct kstatfs          *statp)
 756 {
 757         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 758         xfs_sb_t                *sbp = &mp->m_sb;
 759         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 760         uint64_t                fakeinos, id;
 761         uint64_t                icount;
 762         uint64_t                ifree;
 763         uint64_t                fdblocks;
 764         xfs_extlen_t            lsize;
 765         int64_t                 ffree;
 766
 767         statp->f_type = XFS_SUPER_MAGIC;
 768         statp->f_namelen = MAXNAMELEN - 1;
 769
 770         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 771         statp->f_fsid.val[0] = (u32)id;
 772         statp->f_fsid.val[1] = (u32)(id >> 32);
 773
 774         icount = percpu_counter_sum(&mp->m_icount);
 775         ifree = percpu_counter_sum(&mp->m_ifree);
 776         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 777
 778         spin_lock(&mp->m_sb_lock);
 779         statp->f_bsize = sbp->sb_blocksize;
 780         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 781         statp->f_blocks = sbp->sb_dblocks - lsize;
 782         spin_unlock(&mp->m_sb_lock);
 783
 784         statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
 785         statp->f_bavail = statp->f_bfree;
 786
 787         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 788         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 789         if (M_IGEO(mp)->maxicount)
 790                 statp->f_files = min_t(typeof(statp->f_files),
 791                                         statp->f_files,
 792                                         M_IGEO(mp)->maxicount);
 793
 794         /* If sb_icount overshot maxicount, report actual allocation */
 795         statp->f_files = max_t(typeof(statp->f_files),
 796                                         statp->f_files,
 797                                         sbp->sb_icount);
 798
 799         /* make sure statp->f_ffree does not underflow */
 800         ffree = statp->f_files - (icount - ifree);
 801         statp->f_ffree = max_t(int64_t, ffree, 0);
 802
 803
 804         if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
 805             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 806                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 807                 xfs_qm_statvfs(ip, statp);
 808
 809         if (XFS_IS_REALTIME_MOUNT(mp) &&
 810             (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 811                 statp->f_blocks = sbp->sb_rblocks;
 812                 statp->f_bavail = statp->f_bfree =
 813                         sbp->sb_frextents * sbp->sb_rextsize;
 814         }
 815
 816         return 0;
 817 }
 818
 819 STATIC void
 820 xfs_save_resvblks(struct xfs_mount *mp)
 821 {
 822         uint64_t resblks = 0;
 823
 824         mp->m_resblks_save = mp->m_resblks;
 825         xfs_reserve_blocks(mp, &resblks, NULL);
 826 }
 827
 828 STATIC void
 829 xfs_restore_resvblks(struct xfs_mount *mp)
 830 {
 831         uint64_t resblks;
 832
 833         if (mp->m_resblks_save) {
 834                 resblks = mp->m_resblks_save;
 835                 mp->m_resblks_save = 0;
 836         } else
 837                 resblks = xfs_default_resblks(mp);
 838
 839         xfs_reserve_blocks(mp, &resblks, NULL);
 840 }
 841
 842 /*
 843  * Trigger writeback of all the dirty metadata in the file system.
 844  *
 845  * This ensures that the metadata is written to their location on disk rather
 846  * than just existing in transactions in the log. This means after a quiesce
 847  * there is no log replay required to write the inodes to disk - this is the
 848  * primary difference between a sync and a quiesce.
 849  *
 850  * Note: xfs_log_quiesce() stops background log work - the callers must ensure
 851  * it is started again when appropriate.
 852  */
 853 void
 854 xfs_quiesce_attr(
 855         struct xfs_mount        *mp)
 856 {
 857         int     error = 0;
 858
 859         /* wait for all modifications to complete */
 860         while (atomic_read(&mp->m_active_trans) > 0)
 861                 delay(100);
 862
 863         /* force the log to unpin objects from the now complete transactions */
 864         xfs_log_force(mp, XFS_LOG_SYNC);
 865
 866         /* reclaim inodes to do any IO before the freeze completes */
 867         xfs_reclaim_inodes(mp, 0);
 868         xfs_reclaim_inodes(mp, SYNC_WAIT);
 869
 870         /* Push the superblock and write an unmount record */
 871         error = xfs_log_sbcount(mp);
 872         if (error)
 873                 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
 874                                 "Frozen image may not be consistent.");
 875         /*
 876          * Just warn here till VFS can correctly support
 877          * read-only remount without racing.
 878          */
 879         WARN_ON(atomic_read(&mp->m_active_trans) != 0);
 880
 881         xfs_log_quiesce(mp);
 882 }
 883
 884 /*
 885  * Second stage of a freeze. The data is already frozen so we only
 886  * need to take care of the metadata. Once that's done sync the superblock
 887  * to the log to dirty it in case of a crash while frozen. This ensures that we
 888  * will recover the unlinked inode lists on the next mount.
 889  */
 890 STATIC int
 891 xfs_fs_freeze(
 892         struct super_block      *sb)
 893 {
 894         struct xfs_mount        *mp = XFS_M(sb);
 895
 896         xfs_stop_block_reaping(mp);
 897         xfs_save_resvblks(mp);
 898         xfs_quiesce_attr(mp);
 899         return xfs_sync_sb(mp, true);
 900 }
 901
 902 STATIC int
 903 xfs_fs_unfreeze(
 904         struct super_block      *sb)
 905 {
 906         struct xfs_mount        *mp = XFS_M(sb);
 907
 908         xfs_restore_resvblks(mp);
 909         xfs_log_work_queue(mp);
 910         xfs_start_block_reaping(mp);
 911         return 0;
 912 }
 913
 914 /*
 915  * This function fills in xfs_mount_t fields based on mount args.
 916  * Note: the superblock _has_ now been read in.
 917  */
 918 STATIC int
 919 xfs_finish_flags(
 920         struct xfs_mount        *mp)
 921 {
 922         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
 923
 924         /* Fail a mount where the logbuf is smaller than the log stripe */
 925         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
 926                 if (mp->m_logbsize <= 0 &&
 927                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 928                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 929                 } else if (mp->m_logbsize > 0 &&
 930                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 931                         xfs_warn(mp,
 932                 "logbuf size must be greater than or equal to log stripe size");
 933                         return -EINVAL;
 934                 }
 935         } else {
 936                 /* Fail a mount if the logbuf is larger than 32K */
 937                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 938                         xfs_warn(mp,
 939                 "logbuf size for version 1 logs must be 16K or 32K");
 940                         return -EINVAL;
 941                 }
 942         }
 943
 944         /*
 945          * V5 filesystems always use attr2 format for attributes.
 946          */
 947         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 948             (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 949                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 950                              "attr2 is always enabled for V5 filesystems.");
 951                 return -EINVAL;
 952         }
 953
 954         /*
 955          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
 956          * told by noattr2 to turn it off
 957          */
 958         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 959             !(mp->m_flags & XFS_MOUNT_NOATTR2))
 960                 mp->m_flags |= XFS_MOUNT_ATTR2;
 961
 962         /*
 963          * prohibit r/w mounts of read-only filesystems
 964          */
 965         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 966                 xfs_warn(mp,
 967                         "cannot mount a read-only filesystem as read-write");
 968                 return -EROFS;
 969         }
 970
 971         if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
 972             (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
 973             !xfs_sb_version_has_pquotino(&mp->m_sb)) {
 974                 xfs_warn(mp,
 975                   "Super block does not support project and group quota together");
 976                 return -EINVAL;
 977         }
 978
 979         return 0;
 980 }
 981
 982 static int
 983 xfs_init_percpu_counters(
 984         struct xfs_mount        *mp)
 985 {
 986         int             error;
 987
 988         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
 989         if (error)
 990                 return -ENOMEM;
 991
 992         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
 993         if (error)
 994                 goto free_icount;
 995
 996         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
 997         if (error)
 998                 goto free_ifree;
 999
1000         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1001         if (error)
1002                 goto free_fdblocks;
1003
1004         return 0;
1005
1006 free_fdblocks:
1007         percpu_counter_destroy(&mp->m_fdblocks);
1008 free_ifree:
1009         percpu_counter_destroy(&mp->m_ifree);
1010 free_icount:
1011         percpu_counter_destroy(&mp->m_icount);
1012         return -ENOMEM;
1013 }
1014
1015 void
1016 xfs_reinit_percpu_counters(
1017         struct xfs_mount        *mp)
1018 {
1019         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1020         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1021         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1022 }
1023
1024 static void
1025 xfs_destroy_percpu_counters(
1026         struct xfs_mount        *mp)
1027 {
1028         percpu_counter_destroy(&mp->m_icount);
1029         percpu_counter_destroy(&mp->m_ifree);
1030         percpu_counter_destroy(&mp->m_fdblocks);
1031         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1032                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1033         percpu_counter_destroy(&mp->m_delalloc_blks);
1034 }
1035
1036 static void
1037 xfs_fs_put_super(
1038         struct super_block      *sb)
1039 {
1040         struct xfs_mount        *mp = XFS_M(sb);
1041
1042         /* if ->fill_super failed, we have no mount to tear down */
1043         if (!sb->s_fs_info)
1044                 return;
1045
1046         xfs_notice(mp, "Unmounting Filesystem");
1047         xfs_filestream_unmount(mp);
1048         xfs_unmountfs(mp);
1049
1050         xfs_freesb(mp);
1051         free_percpu(mp->m_stats.xs_stats);
1052         xfs_destroy_percpu_counters(mp);
1053         xfs_destroy_mount_workqueues(mp);
1054         xfs_close_devices(mp);
1055
1056         sb->s_fs_info = NULL;
1057         xfs_mount_free(mp);
1058 }
1059
1060 static long
1061 xfs_fs_nr_cached_objects(
1062         struct super_block      *sb,
1063         struct shrink_control   *sc)
1064 {
1065         /* Paranoia: catch incorrect calls during mount setup or teardown */
1066         if (WARN_ON_ONCE(!sb->s_fs_info))
1067                 return 0;
1068         return xfs_reclaim_inodes_count(XFS_M(sb));
1069 }
1070
1071 static long
1072 xfs_fs_free_cached_objects(
1073         struct super_block      *sb,
1074         struct shrink_control   *sc)
1075 {
1076         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1077 }
1078
1079 static const struct super_operations xfs_super_operations = {
1080         .alloc_inode            = xfs_fs_alloc_inode,
1081         .destroy_inode          = xfs_fs_destroy_inode,
1082         .dirty_inode            = xfs_fs_dirty_inode,
1083         .drop_inode             = xfs_fs_drop_inode,
1084         .put_super              = xfs_fs_put_super,
1085         .sync_fs                = xfs_fs_sync_fs,
1086         .freeze_fs              = xfs_fs_freeze,
1087         .unfreeze_fs            = xfs_fs_unfreeze,
1088         .statfs                 = xfs_fs_statfs,
1089         .show_options           = xfs_fs_show_options,
1090         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1091         .free_cached_objects    = xfs_fs_free_cached_objects,
1092 };
1093
1094 static int
1095 suffix_kstrtoint(
1096         const char      *s,
1097         unsigned int    base,
1098         int             *res)
1099 {
1100         int             last, shift_left_factor = 0, _res;
1101         char            *value;
1102         int             ret = 0;
1103
1104         value = kstrdup(s, GFP_KERNEL);
1105         if (!value)
1106                 return -ENOMEM;
1107
1108         last = strlen(value) - 1;
1109         if (value[last] == 'K' || value[last] == 'k') {
1110                 shift_left_factor = 10;
1111                 value[last] = '\0';
1112         }
1113         if (value[last] == 'M' || value[last] == 'm') {
1114                 shift_left_factor = 20;
1115                 value[last] = '\0';
1116         }
1117         if (value[last] == 'G' || value[last] == 'g') {
1118                 shift_left_factor = 30;
1119                 value[last] = '\0';
1120         }
1121
1122         if (kstrtoint(value, base, &_res))
1123                 ret = -EINVAL;
1124         kfree(value);
1125         *res = _res << shift_left_factor;
1126         return ret;
1127 }
1128
1129 /*
1130  * Set mount state from a mount option.
1131  *
1132  * NOTE: mp->m_super is NULL here!
1133  */
1134 static int
1135 xfs_fc_parse_param(
1136         struct fs_context       *fc,
1137         struct fs_parameter     *param)
1138 {
1139         struct xfs_mount        *mp = fc->s_fs_info;
1140         struct fs_parse_result  result;
1141         int                     size = 0;
1142         int                     opt;
1143
1144         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1145         if (opt < 0)
1146                 return opt;
1147
1148         switch (opt) {
1149         case Opt_logbufs:
1150                 mp->m_logbufs = result.uint_32;
1151                 return 0;
1152         case Opt_logbsize:
1153                 if (suffix_kstrtoint(param->string, 10, &mp->m_logbsize))
1154                         return -EINVAL;
1155                 return 0;
1156         case Opt_logdev:
1157                 kfree(mp->m_logname);
1158                 mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1159                 if (!mp->m_logname)
1160                         return -ENOMEM;
1161                 return 0;
1162         case Opt_rtdev:
1163                 kfree(mp->m_rtname);
1164                 mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1165                 if (!mp->m_rtname)
1166                         return -ENOMEM;
1167                 return 0;
1168         case Opt_allocsize:
1169                 if (suffix_kstrtoint(param->string, 10, &size))
1170                         return -EINVAL;
1171                 mp->m_allocsize_log = ffs(size) - 1;
1172                 mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1173                 return 0;
1174         case Opt_grpid:
1175         case Opt_bsdgroups:
1176                 mp->m_flags |= XFS_MOUNT_GRPID;
1177                 return 0;
1178         case Opt_nogrpid:
1179         case Opt_sysvgroups:
1180                 mp->m_flags &= ~XFS_MOUNT_GRPID;
1181                 return 0;
1182         case Opt_wsync:
1183                 mp->m_flags |= XFS_MOUNT_WSYNC;
1184                 return 0;
1185         case Opt_norecovery:
1186                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
1187                 return 0;
1188         case Opt_noalign:
1189                 mp->m_flags |= XFS_MOUNT_NOALIGN;
1190                 return 0;
1191         case Opt_swalloc:
1192                 mp->m_flags |= XFS_MOUNT_SWALLOC;
1193                 return 0;
1194         case Opt_sunit:
1195                 mp->m_dalign = result.uint_32;
1196                 return 0;
1197         case Opt_swidth:
1198                 mp->m_swidth = result.uint_32;
1199                 return 0;
1200         case Opt_inode32:
1201                 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1202                 return 0;
1203         case Opt_inode64:
1204                 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1205                 return 0;
1206         case Opt_nouuid:
1207                 mp->m_flags |= XFS_MOUNT_NOUUID;
1208                 return 0;
1209         case Opt_ikeep:
1210                 mp->m_flags |= XFS_MOUNT_IKEEP;
1211                 return 0;
1212         case Opt_noikeep:
1213                 mp->m_flags &= ~XFS_MOUNT_IKEEP;
1214                 return 0;
1215         case Opt_largeio:
1216                 mp->m_flags |= XFS_MOUNT_LARGEIO;
1217                 return 0;
1218         case Opt_nolargeio:
1219                 mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1220                 return 0;
1221         case Opt_attr2:
1222                 mp->m_flags |= XFS_MOUNT_ATTR2;
1223                 return 0;
1224         case Opt_noattr2:
1225                 mp->m_flags &= ~XFS_MOUNT_ATTR2;
1226                 mp->m_flags |= XFS_MOUNT_NOATTR2;
1227                 return 0;
1228         case Opt_filestreams:
1229                 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1230                 return 0;
1231         case Opt_noquota:
1232                 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1233                 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1234                 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1235                 return 0;
1236         case Opt_quota:
1237         case Opt_uquota:
1238         case Opt_usrquota:
1239                 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1240                                  XFS_UQUOTA_ENFD);
1241                 return 0;
1242         case Opt_qnoenforce:
1243         case Opt_uqnoenforce:
1244                 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1245                 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1246                 return 0;
1247         case Opt_pquota:
1248         case Opt_prjquota:
1249                 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1250                                  XFS_PQUOTA_ENFD);
1251                 return 0;
1252         case Opt_pqnoenforce:
1253                 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1254                 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1255                 return 0;
1256         case Opt_gquota:
1257         case Opt_grpquota:
1258                 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1259                                  XFS_GQUOTA_ENFD);
1260                 return 0;
1261         case Opt_gqnoenforce:
1262                 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1263                 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1264                 return 0;
1265         case Opt_discard:
1266                 mp->m_flags |= XFS_MOUNT_DISCARD;
1267                 return 0;
1268         case Opt_nodiscard:
1269                 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1270                 return 0;
1271 #ifdef CONFIG_FS_DAX
1272         case Opt_dax:
1273                 mp->m_flags |= XFS_MOUNT_DAX;
1274                 return 0;
1275 #endif
1276         default:
1277                 xfs_warn(mp, "unknown mount option [%s].", param->key);
1278                 return -EINVAL;
1279         }
1280
1281         return 0;
1282 }
1283
1284 static int
1285 xfs_fc_validate_params(
1286         struct xfs_mount        *mp)
1287 {
1288         /*
1289          * no recovery flag requires a read-only mount
1290          */
1291         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1292             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1293                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1294                 return -EINVAL;
1295         }
1296
1297         if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1298             (mp->m_dalign || mp->m_swidth)) {
1299                 xfs_warn(mp,
1300         "sunit and swidth options incompatible with the noalign option");
1301                 return -EINVAL;
1302         }
1303
1304         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1305                 xfs_warn(mp, "quota support not available in this kernel.");
1306                 return -EINVAL;
1307         }
1308
1309         if ((mp->m_dalign && !mp->m_swidth) ||
1310             (!mp->m_dalign && mp->m_swidth)) {
1311                 xfs_warn(mp, "sunit and swidth must be specified together");
1312                 return -EINVAL;
1313         }
1314
1315         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1316                 xfs_warn(mp,
1317         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1318                         mp->m_swidth, mp->m_dalign);
1319                 return -EINVAL;
1320         }
1321
1322         if (mp->m_logbufs != -1 &&
1323             mp->m_logbufs != 0 &&
1324             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1325              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1326                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1327                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1328                 return -EINVAL;
1329         }
1330
1331         if (mp->m_logbsize != -1 &&
1332             mp->m_logbsize !=  0 &&
1333             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1334              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1335              !is_power_of_2(mp->m_logbsize))) {
1336                 xfs_warn(mp,
1337                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1338                         mp->m_logbsize);
1339                 return -EINVAL;
1340         }
1341
1342         if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1343             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1344              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1345                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1346                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1347                 return -EINVAL;
1348         }
1349
1350         return 0;
1351 }
1352
1353 static int
1354 xfs_fc_fill_super(
1355         struct super_block      *sb,
1356         struct fs_context       *fc)
1357 {
1358         struct xfs_mount        *mp = sb->s_fs_info;
1359         struct inode            *root;
1360         int                     flags = 0, error;
1361
1362         mp->m_super = sb;
1363
1364         error = xfs_fc_validate_params(mp);
1365         if (error)
1366                 goto out_free_names;
1367
1368         sb_min_blocksize(sb, BBSIZE);
1369         sb->s_xattr = xfs_xattr_handlers;
1370         sb->s_export_op = &xfs_export_operations;
1371 #ifdef CONFIG_XFS_QUOTA
1372         sb->s_qcop = &xfs_quotactl_operations;
1373         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1374 #endif
1375         sb->s_op = &xfs_super_operations;
1376
1377         /*
1378          * Delay mount work if the debug hook is set. This is debug
1379          * instrumention to coordinate simulation of xfs mount failures with
1380          * VFS superblock operations
1381          */
1382         if (xfs_globals.mount_delay) {
1383                 xfs_notice(mp, "Delaying mount for %d seconds.",
1384                         xfs_globals.mount_delay);
1385                 msleep(xfs_globals.mount_delay * 1000);
1386         }
1387
1388         if (fc->sb_flags & SB_SILENT)
1389                 flags |= XFS_MFSI_QUIET;
1390
1391         error = xfs_open_devices(mp);
1392         if (error)
1393                 goto out_free_names;
1394
1395         error = xfs_init_mount_workqueues(mp);
1396         if (error)
1397                 goto out_close_devices;
1398
1399         error = xfs_init_percpu_counters(mp);
1400         if (error)
1401                 goto out_destroy_workqueues;
1402
1403         /* Allocate stats memory before we do operations that might use it */
1404         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1405         if (!mp->m_stats.xs_stats) {
1406                 error = -ENOMEM;
1407                 goto out_destroy_counters;
1408         }
1409
1410         error = xfs_readsb(mp, flags);
1411         if (error)
1412                 goto out_free_stats;
1413
1414         error = xfs_finish_flags(mp);
1415         if (error)
1416                 goto out_free_sb;
1417
1418         error = xfs_setup_devices(mp);
1419         if (error)
1420                 goto out_free_sb;
1421
1422         error = xfs_filestream_mount(mp);
1423         if (error)
1424                 goto out_free_sb;
1425
1426         /*
1427          * we must configure the block size in the superblock before we run the
1428          * full mount process as the mount process can lookup and cache inodes.
1429          */
1430         sb->s_magic = XFS_SUPER_MAGIC;
1431         sb->s_blocksize = mp->m_sb.sb_blocksize;
1432         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1433         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1434         sb->s_max_links = XFS_MAXLINK;
1435         sb->s_time_gran = 1;
1436         sb->s_time_min = S32_MIN;
1437         sb->s_time_max = S32_MAX;
1438         sb->s_iflags |= SB_I_CGROUPWB;
1439
1440         set_posix_acl_flag(sb);
1441
1442         /* version 5 superblocks support inode version counters. */
1443         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1444                 sb->s_flags |= SB_I_VERSION;
1445
1446         if (mp->m_flags & XFS_MOUNT_DAX) {
1447                 bool rtdev_is_dax = false, datadev_is_dax;
1448
1449                 xfs_warn(mp,
1450                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1451
1452                 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1453                         sb->s_blocksize);
1454                 if (mp->m_rtdev_targp)
1455                         rtdev_is_dax = bdev_dax_supported(
1456                                 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1457                 if (!rtdev_is_dax && !datadev_is_dax) {
1458                         xfs_alert(mp,
1459                         "DAX unsupported by block device. Turning off DAX.");
1460                         mp->m_flags &= ~XFS_MOUNT_DAX;
1461                 }
1462                 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1463                         xfs_alert(mp,
1464                 "DAX and reflink cannot be used together!");
1465                         error = -EINVAL;
1466                         goto out_filestream_unmount;
1467                 }
1468         }
1469
1470         if (mp->m_flags & XFS_MOUNT_DISCARD) {
1471                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1472
1473                 if (!blk_queue_discard(q)) {
1474                         xfs_warn(mp, "mounting with \"discard\" option, but "
1475                                         "the device does not support discard");
1476                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
1477                 }
1478         }
1479
1480         if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1481                 if (mp->m_sb.sb_rblocks) {
1482                         xfs_alert(mp,
1483         "reflink not compatible with realtime device!");
1484                         error = -EINVAL;
1485                         goto out_filestream_unmount;
1486                 }
1487
1488                 if (xfs_globals.always_cow) {
1489                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1490                         mp->m_always_cow = true;
1491                 }
1492         }
1493
1494         if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1495                 xfs_alert(mp,
1496         "reverse mapping btree not compatible with realtime device!");
1497                 error = -EINVAL;
1498                 goto out_filestream_unmount;
1499         }
1500
1501         error = xfs_mountfs(mp);
1502         if (error)
1503                 goto out_filestream_unmount;
1504
1505         root = igrab(VFS_I(mp->m_rootip));
1506         if (!root) {
1507                 error = -ENOENT;
1508                 goto out_unmount;
1509         }
1510         sb->s_root = d_make_root(root);
1511         if (!sb->s_root) {
1512                 error = -ENOMEM;
1513                 goto out_unmount;
1514         }
1515
1516         return 0;
1517
1518  out_filestream_unmount:
1519         xfs_filestream_unmount(mp);
1520  out_free_sb:
1521         xfs_freesb(mp);
1522  out_free_stats:
1523         free_percpu(mp->m_stats.xs_stats);
1524  out_destroy_counters:
1525         xfs_destroy_percpu_counters(mp);
1526  out_destroy_workqueues:
1527         xfs_destroy_mount_workqueues(mp);
1528  out_close_devices:
1529         xfs_close_devices(mp);
1530  out_free_names:
1531         sb->s_fs_info = NULL;
1532         xfs_mount_free(mp);
1533         return error;
1534
1535  out_unmount:
1536         xfs_filestream_unmount(mp);
1537         xfs_unmountfs(mp);
1538         goto out_free_sb;
1539 }
1540
1541 static int
1542 xfs_fc_get_tree(
1543         struct fs_context       *fc)
1544 {
1545         return get_tree_bdev(fc, xfs_fc_fill_super);
1546 }
1547
1548 static int
1549 xfs_remount_rw(
1550         struct xfs_mount        *mp)
1551 {
1552         struct xfs_sb           *sbp = &mp->m_sb;
1553         int error;
1554
1555         if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1556                 xfs_warn(mp,
1557                         "ro->rw transition prohibited on norecovery mount");
1558                 return -EINVAL;
1559         }
1560
1561         if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1562             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1563                 xfs_warn(mp,
1564         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1565                         (sbp->sb_features_ro_compat &
1566                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1567                 return -EINVAL;
1568         }
1569
1570         mp->m_flags &= ~XFS_MOUNT_RDONLY;
1571
1572         /*
1573          * If this is the first remount to writeable state we might have some
1574          * superblock changes to update.
1575          */
1576         if (mp->m_update_sb) {
1577                 error = xfs_sync_sb(mp, false);
1578                 if (error) {
1579                         xfs_warn(mp, "failed to write sb changes");
1580                         return error;
1581                 }
1582                 mp->m_update_sb = false;
1583         }
1584
1585         /*
1586          * Fill out the reserve pool if it is empty. Use the stashed value if
1587          * it is non-zero, otherwise go with the default.
1588          */
1589         xfs_restore_resvblks(mp);
1590         xfs_log_work_queue(mp);
1591
1592         /* Recover any CoW blocks that never got remapped. */
1593         error = xfs_reflink_recover_cow(mp);
1594         if (error) {
1595                 xfs_err(mp,
1596                         "Error %d recovering leftover CoW allocations.", error);
1597                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1598                 return error;
1599         }
1600         xfs_start_block_reaping(mp);
1601
1602         /* Create the per-AG metadata reservation pool .*/
1603         error = xfs_fs_reserve_ag_blocks(mp);
1604         if (error && error != -ENOSPC)
1605                 return error;
1606
1607         return 0;
1608 }
1609
1610 static int
1611 xfs_remount_ro(
1612         struct xfs_mount        *mp)
1613 {
1614         int error;
1615
1616         /*
1617          * Cancel background eofb scanning so it cannot race with the final
1618          * log force+buftarg wait and deadlock the remount.
1619          */
1620         xfs_stop_block_reaping(mp);
1621
1622         /* Get rid of any leftover CoW reservations... */
1623         error = xfs_icache_free_cowblocks(mp, NULL);
1624         if (error) {
1625                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1626                 return error;
1627         }
1628
1629         /* Free the per-AG metadata reservation pool. */
1630         error = xfs_fs_unreserve_ag_blocks(mp);
1631         if (error) {
1632                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1633                 return error;
1634         }
1635
1636         /*
1637          * Before we sync the metadata, we need to free up the reserve block
1638          * pool so that the used block count in the superblock on disk is
1639          * correct at the end of the remount. Stash the current* reserve pool
1640          * size so that if we get remounted rw, we can return it to the same
1641          * size.
1642          */
1643         xfs_save_resvblks(mp);
1644
1645         xfs_quiesce_attr(mp);
1646         mp->m_flags |= XFS_MOUNT_RDONLY;
1647
1648         return 0;
1649 }
1650
1651 /*
1652  * Logically we would return an error here to prevent users from believing
1653  * they might have changed mount options using remount which can't be changed.
1654  *
1655  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1656  * arguments in some cases so we can't blindly reject options, but have to
1657  * check for each specified option if it actually differs from the currently
1658  * set option and only reject it if that's the case.
1659  *
1660  * Until that is implemented we return success for every remount request, and
1661  * silently ignore all options that we can't actually change.
1662  */
1663 static int
1664 xfs_fc_reconfigure(
1665         struct fs_context *fc)
1666 {
1667         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1668         struct xfs_mount        *new_mp = fc->s_fs_info;
1669         xfs_sb_t                *sbp = &mp->m_sb;
1670         int                     flags = fc->sb_flags;
1671         int                     error;
1672
1673         error = xfs_fc_validate_params(new_mp);
1674         if (error)
1675                 return error;
1676
1677         sync_filesystem(mp->m_super);
1678
1679         /* inode32 -> inode64 */
1680         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1681             !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1682                 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1683                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1684         }
1685
1686         /* inode64 -> inode32 */
1687         if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1688             (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1689                 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1690                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1691         }
1692
1693         /* ro -> rw */
1694         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1695                 error = xfs_remount_rw(mp);
1696                 if (error)
1697                         return error;
1698         }
1699
1700         /* rw -> ro */
1701         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1702                 error = xfs_remount_ro(mp);
1703                 if (error)
1704                         return error;
1705         }
1706
1707         return 0;
1708 }
1709
1710 static void xfs_fc_free(
1711         struct fs_context       *fc)
1712 {
1713         struct xfs_mount        *mp = fc->s_fs_info;
1714
1715         /*
1716          * mp is stored in the fs_context when it is initialized.
1717          * mp is transferred to the superblock on a successful mount,
1718          * but if an error occurs before the transfer we have to free
1719          * it here.
1720          */
1721         if (mp)
1722                 xfs_mount_free(mp);
1723 }
1724
1725 static const struct fs_context_operations xfs_context_ops = {
1726         .parse_param = xfs_fc_parse_param,
1727         .get_tree    = xfs_fc_get_tree,
1728         .reconfigure = xfs_fc_reconfigure,
1729         .free        = xfs_fc_free,
1730 };
1731
1732 static int xfs_init_fs_context(
1733         struct fs_context       *fc)
1734 {
1735         struct xfs_mount        *mp;
1736
1737         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1738         if (!mp)
1739                 return -ENOMEM;
1740
1741         spin_lock_init(&mp->m_sb_lock);
1742         spin_lock_init(&mp->m_agirotor_lock);
1743         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1744         spin_lock_init(&mp->m_perag_lock);
1745         mutex_init(&mp->m_growlock);
1746         atomic_set(&mp->m_active_trans, 0);
1747         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1748         INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1749         INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1750         mp->m_kobj.kobject.kset = xfs_kset;
1751         /*
1752          * We don't create the finobt per-ag space reservation until after log
1753          * recovery, so we must set this to true so that an ifree transaction
1754          * started during log recovery will not depend on space reservations
1755          * for finobt expansion.
1756          */
1757         mp->m_finobt_nores = true;
1758
1759         /*
1760          * These can be overridden by the mount option parsing.
1761          */
1762         mp->m_logbufs = -1;
1763         mp->m_logbsize = -1;
1764         mp->m_allocsize_log = 16; /* 64k */
1765
1766         /*
1767          * Copy binary VFS mount flags we are interested in.
1768          */
1769         if (fc->sb_flags & SB_RDONLY)
1770                 mp->m_flags |= XFS_MOUNT_RDONLY;
1771         if (fc->sb_flags & SB_DIRSYNC)
1772                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1773         if (fc->sb_flags & SB_SYNCHRONOUS)
1774                 mp->m_flags |= XFS_MOUNT_WSYNC;
1775
1776         fc->s_fs_info = mp;
1777         fc->ops = &xfs_context_ops;
1778
1779         return 0;
1780 }
1781
1782 static struct file_system_type xfs_fs_type = {
1783         .owner                  = THIS_MODULE,
1784         .name                   = "xfs",
1785         .init_fs_context        = xfs_init_fs_context,
1786         .parameters             = xfs_fs_parameters,
1787         .kill_sb                = kill_block_super,
1788         .fs_flags               = FS_REQUIRES_DEV,
1789 };
1790 MODULE_ALIAS_FS("xfs");
1791
1792 STATIC int __init
1793 xfs_init_zones(void)
1794 {
1795         xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1796                                                 sizeof(struct xlog_ticket),
1797                                                 0, 0, NULL);
1798         if (!xfs_log_ticket_zone)
1799                 goto out;
1800
1801         xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1802                                         sizeof(struct xfs_extent_free_item),
1803                                         0, 0, NULL);
1804         if (!xfs_bmap_free_item_zone)
1805                 goto out_destroy_log_ticket_zone;
1806
1807         xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1808                                                sizeof(struct xfs_btree_cur),
1809                                                0, 0, NULL);
1810         if (!xfs_btree_cur_zone)
1811                 goto out_destroy_bmap_free_item_zone;
1812
1813         xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1814                                               sizeof(struct xfs_da_state),
1815                                               0, 0, NULL);
1816         if (!xfs_da_state_zone)
1817                 goto out_destroy_btree_cur_zone;
1818
1819         xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1820                                            sizeof(struct xfs_ifork),
1821                                            0, 0, NULL);
1822         if (!xfs_ifork_zone)
1823                 goto out_destroy_da_state_zone;
1824
1825         xfs_trans_zone = kmem_cache_create("xf_trans",
1826                                            sizeof(struct xfs_trans),
1827                                            0, 0, NULL);
1828         if (!xfs_trans_zone)
1829                 goto out_destroy_ifork_zone;
1830
1831
1832         /*
1833          * The size of the zone allocated buf log item is the maximum
1834          * size possible under XFS.  This wastes a little bit of memory,
1835          * but it is much faster.
1836          */
1837         xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1838                                               sizeof(struct xfs_buf_log_item),
1839                                               0, 0, NULL);
1840         if (!xfs_buf_item_zone)
1841                 goto out_destroy_trans_zone;
1842
1843         xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1844                                         (sizeof(struct xfs_efd_log_item) +
1845                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1846                                         sizeof(struct xfs_extent)),
1847                                         0, 0, NULL);
1848         if (!xfs_efd_zone)
1849                 goto out_destroy_buf_item_zone;
1850
1851         xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1852                                          (sizeof(struct xfs_efi_log_item) +
1853                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1854                                          sizeof(struct xfs_extent)),
1855                                          0, 0, NULL);
1856         if (!xfs_efi_zone)
1857                 goto out_destroy_efd_zone;
1858
1859         xfs_inode_zone = kmem_cache_create("xfs_inode",
1860                                            sizeof(struct xfs_inode), 0,
1861                                            (SLAB_HWCACHE_ALIGN |
1862                                             SLAB_RECLAIM_ACCOUNT |
1863                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1864                                            xfs_fs_inode_init_once);
1865         if (!xfs_inode_zone)
1866                 goto out_destroy_efi_zone;
1867
1868         xfs_ili_zone = kmem_cache_create("xfs_ili",
1869                                          sizeof(struct xfs_inode_log_item), 0,
1870                                          SLAB_MEM_SPREAD, NULL);
1871         if (!xfs_ili_zone)
1872                 goto out_destroy_inode_zone;
1873
1874         xfs_icreate_zone = kmem_cache_create("xfs_icr",
1875                                              sizeof(struct xfs_icreate_item),
1876                                              0, 0, NULL);
1877         if (!xfs_icreate_zone)
1878                 goto out_destroy_ili_zone;
1879
1880         xfs_rud_zone = kmem_cache_create("xfs_rud_item",
1881                                          sizeof(struct xfs_rud_log_item),
1882                                          0, 0, NULL);
1883         if (!xfs_rud_zone)
1884                 goto out_destroy_icreate_zone;
1885
1886         xfs_rui_zone = kmem_cache_create("xfs_rui_item",
1887                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1888                         0, 0, NULL);
1889         if (!xfs_rui_zone)
1890                 goto out_destroy_rud_zone;
1891
1892         xfs_cud_zone = kmem_cache_create("xfs_cud_item",
1893                                          sizeof(struct xfs_cud_log_item),
1894                                          0, 0, NULL);
1895         if (!xfs_cud_zone)
1896                 goto out_destroy_rui_zone;
1897
1898         xfs_cui_zone = kmem_cache_create("xfs_cui_item",
1899                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1900                         0, 0, NULL);
1901         if (!xfs_cui_zone)
1902                 goto out_destroy_cud_zone;
1903
1904         xfs_bud_zone = kmem_cache_create("xfs_bud_item",
1905                                          sizeof(struct xfs_bud_log_item),
1906                                          0, 0, NULL);
1907         if (!xfs_bud_zone)
1908                 goto out_destroy_cui_zone;
1909
1910         xfs_bui_zone = kmem_cache_create("xfs_bui_item",
1911                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1912                         0, 0, NULL);
1913         if (!xfs_bui_zone)
1914                 goto out_destroy_bud_zone;
1915
1916         return 0;
1917
1918  out_destroy_bud_zone:
1919         kmem_cache_destroy(xfs_bud_zone);
1920  out_destroy_cui_zone:
1921         kmem_cache_destroy(xfs_cui_zone);
1922  out_destroy_cud_zone:
1923         kmem_cache_destroy(xfs_cud_zone);
1924  out_destroy_rui_zone:
1925         kmem_cache_destroy(xfs_rui_zone);
1926  out_destroy_rud_zone:
1927         kmem_cache_destroy(xfs_rud_zone);
1928  out_destroy_icreate_zone:
1929         kmem_cache_destroy(xfs_icreate_zone);
1930  out_destroy_ili_zone:
1931         kmem_cache_destroy(xfs_ili_zone);
1932  out_destroy_inode_zone:
1933         kmem_cache_destroy(xfs_inode_zone);
1934  out_destroy_efi_zone:
1935         kmem_cache_destroy(xfs_efi_zone);
1936  out_destroy_efd_zone:
1937         kmem_cache_destroy(xfs_efd_zone);
1938  out_destroy_buf_item_zone:
1939         kmem_cache_destroy(xfs_buf_item_zone);
1940  out_destroy_trans_zone:
1941         kmem_cache_destroy(xfs_trans_zone);
1942  out_destroy_ifork_zone:
1943         kmem_cache_destroy(xfs_ifork_zone);
1944  out_destroy_da_state_zone:
1945         kmem_cache_destroy(xfs_da_state_zone);
1946  out_destroy_btree_cur_zone:
1947         kmem_cache_destroy(xfs_btree_cur_zone);
1948  out_destroy_bmap_free_item_zone:
1949         kmem_cache_destroy(xfs_bmap_free_item_zone);
1950  out_destroy_log_ticket_zone:
1951         kmem_cache_destroy(xfs_log_ticket_zone);
1952  out:
1953         return -ENOMEM;
1954 }
1955
1956 STATIC void
1957 xfs_destroy_zones(void)
1958 {
1959         /*
1960          * Make sure all delayed rcu free are flushed before we
1961          * destroy caches.
1962          */
1963         rcu_barrier();
1964         kmem_cache_destroy(xfs_bui_zone);
1965         kmem_cache_destroy(xfs_bud_zone);
1966         kmem_cache_destroy(xfs_cui_zone);
1967         kmem_cache_destroy(xfs_cud_zone);
1968         kmem_cache_destroy(xfs_rui_zone);
1969         kmem_cache_destroy(xfs_rud_zone);
1970         kmem_cache_destroy(xfs_icreate_zone);
1971         kmem_cache_destroy(xfs_ili_zone);
1972         kmem_cache_destroy(xfs_inode_zone);
1973         kmem_cache_destroy(xfs_efi_zone);
1974         kmem_cache_destroy(xfs_efd_zone);
1975         kmem_cache_destroy(xfs_buf_item_zone);
1976         kmem_cache_destroy(xfs_trans_zone);
1977         kmem_cache_destroy(xfs_ifork_zone);
1978         kmem_cache_destroy(xfs_da_state_zone);
1979         kmem_cache_destroy(xfs_btree_cur_zone);
1980         kmem_cache_destroy(xfs_bmap_free_item_zone);
1981         kmem_cache_destroy(xfs_log_ticket_zone);
1982 }
1983
1984 STATIC int __init
1985 xfs_init_workqueues(void)
1986 {
1987         /*
1988          * The allocation workqueue can be used in memory reclaim situations
1989          * (writepage path), and parallelism is only limited by the number of
1990          * AGs in all the filesystems mounted. Hence use the default large
1991          * max_active value for this workqueue.
1992          */
1993         xfs_alloc_wq = alloc_workqueue("xfsalloc",
1994                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
1995         if (!xfs_alloc_wq)
1996                 return -ENOMEM;
1997
1998         xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
1999         if (!xfs_discard_wq)
2000                 goto out_free_alloc_wq;
2001
2002         return 0;
2003 out_free_alloc_wq:
2004         destroy_workqueue(xfs_alloc_wq);
2005         return -ENOMEM;
2006 }
2007
2008 STATIC void
2009 xfs_destroy_workqueues(void)
2010 {
2011         destroy_workqueue(xfs_discard_wq);
2012         destroy_workqueue(xfs_alloc_wq);
2013 }
2014
2015 STATIC int __init
2016 init_xfs_fs(void)
2017 {
2018         int                     error;
2019
2020         xfs_check_ondisk_structs();
2021
2022         printk(KERN_INFO XFS_VERSION_STRING " with "
2023                          XFS_BUILD_OPTIONS " enabled\n");
2024
2025         xfs_dir_startup();
2026
2027         error = xfs_init_zones();
2028         if (error)
2029                 goto out;
2030
2031         error = xfs_init_workqueues();
2032         if (error)
2033                 goto out_destroy_zones;
2034
2035         error = xfs_mru_cache_init();
2036         if (error)
2037                 goto out_destroy_wq;
2038
2039         error = xfs_buf_init();
2040         if (error)
2041                 goto out_mru_cache_uninit;
2042
2043         error = xfs_init_procfs();
2044         if (error)
2045                 goto out_buf_terminate;
2046
2047         error = xfs_sysctl_register();
2048         if (error)
2049                 goto out_cleanup_procfs;
2050
2051         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2052         if (!xfs_kset) {
2053                 error = -ENOMEM;
2054                 goto out_sysctl_unregister;
2055         }
2056
2057         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2058
2059         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2060         if (!xfsstats.xs_stats) {
2061                 error = -ENOMEM;
2062                 goto out_kset_unregister;
2063         }
2064
2065         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2066                                "stats");
2067         if (error)
2068                 goto out_free_stats;
2069
2070 #ifdef DEBUG
2071         xfs_dbg_kobj.kobject.kset = xfs_kset;
2072         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2073         if (error)
2074                 goto out_remove_stats_kobj;
2075 #endif
2076
2077         error = xfs_qm_init();
2078         if (error)
2079                 goto out_remove_dbg_kobj;
2080
2081         error = register_filesystem(&xfs_fs_type);
2082         if (error)
2083                 goto out_qm_exit;
2084         return 0;
2085
2086  out_qm_exit:
2087         xfs_qm_exit();
2088  out_remove_dbg_kobj:
2089 #ifdef DEBUG
2090         xfs_sysfs_del(&xfs_dbg_kobj);
2091  out_remove_stats_kobj:
2092 #endif
2093         xfs_sysfs_del(&xfsstats.xs_kobj);
2094  out_free_stats:
2095         free_percpu(xfsstats.xs_stats);
2096  out_kset_unregister:
2097         kset_unregister(xfs_kset);
2098  out_sysctl_unregister:
2099         xfs_sysctl_unregister();
2100  out_cleanup_procfs:
2101         xfs_cleanup_procfs();
2102  out_buf_terminate:
2103         xfs_buf_terminate();
2104  out_mru_cache_uninit:
2105         xfs_mru_cache_uninit();
2106  out_destroy_wq:
2107         xfs_destroy_workqueues();
2108  out_destroy_zones:
2109         xfs_destroy_zones();
2110  out:
2111         return error;
2112 }
2113
2114 STATIC void __exit
2115 exit_xfs_fs(void)
2116 {
2117         xfs_qm_exit();
2118         unregister_filesystem(&xfs_fs_type);
2119 #ifdef DEBUG
2120         xfs_sysfs_del(&xfs_dbg_kobj);
2121 #endif
2122         xfs_sysfs_del(&xfsstats.xs_kobj);
2123         free_percpu(xfsstats.xs_stats);
2124         kset_unregister(xfs_kset);
2125         xfs_sysctl_unregister();
2126         xfs_cleanup_procfs();
2127         xfs_buf_terminate();
2128         xfs_mru_cache_uninit();
2129         xfs_destroy_workqueues();
2130         xfs_destroy_zones();
2131         xfs_uuid_table_free();
2132 }
2133
2134 module_init(init_xfs_fs);
2135 module_exit(exit_xfs_fs);
2136
2137 MODULE_AUTHOR("Silicon Graphics, Inc.");
2138 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2139 MODULE_LICENSE("GPL");