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 #include "xfs_pwork.h"
  39 #include "xfs_ag.h"
  40 #include "xfs_defer.h"
  41 #include "xfs_attr_item.h"
  42 #include "xfs_xattr.h"
  43 #include "xfs_iunlink_item.h"
  44 #include "xfs_dahash_test.h"
  45
  46 #include <linux/magic.h>
  47 #include <linux/fs_context.h>
  48 #include <linux/fs_parser.h>
  49
  50 static const struct super_operations xfs_super_operations;
  51
  52 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  53 #ifdef DEBUG
  54 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  55 #endif
  56
  57 #ifdef CONFIG_HOTPLUG_CPU
  58 static LIST_HEAD(xfs_mount_list);
  59 static DEFINE_SPINLOCK(xfs_mount_list_lock);
  60
  61 static inline void xfs_mount_list_add(struct xfs_mount *mp)
  62 {
  63         spin_lock(&xfs_mount_list_lock);
  64         list_add(&mp->m_mount_list, &xfs_mount_list);
  65         spin_unlock(&xfs_mount_list_lock);
  66 }
  67
  68 static inline void xfs_mount_list_del(struct xfs_mount *mp)
  69 {
  70         spin_lock(&xfs_mount_list_lock);
  71         list_del(&mp->m_mount_list);
  72         spin_unlock(&xfs_mount_list_lock);
  73 }
  74 #else /* !CONFIG_HOTPLUG_CPU */
  75 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
  76 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
  77 #endif
  78
  79 enum xfs_dax_mode {
  80         XFS_DAX_INODE = 0,
  81         XFS_DAX_ALWAYS = 1,
  82         XFS_DAX_NEVER = 2,
  83 };
  84
  85 static void
  86 xfs_mount_set_dax_mode(
  87         struct xfs_mount        *mp,
  88         enum xfs_dax_mode       mode)
  89 {
  90         switch (mode) {
  91         case XFS_DAX_INODE:
  92                 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
  93                 break;
  94         case XFS_DAX_ALWAYS:
  95                 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
  96                 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
  97                 break;
  98         case XFS_DAX_NEVER:
  99                 mp->m_features |= XFS_FEAT_DAX_NEVER;
 100                 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
 101                 break;
 102         }
 103 }
 104
 105 static const struct constant_table dax_param_enums[] = {
 106         {"inode",       XFS_DAX_INODE },
 107         {"always",      XFS_DAX_ALWAYS },
 108         {"never",       XFS_DAX_NEVER },
 109         {}
 110 };
 111
 112 /*
 113  * Table driven mount option parser.
 114  */
 115 enum {
 116         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
 117         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
 118         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
 119         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
 120         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
 121         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
 122         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
 123         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
 124         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
 125 };
 126
 127 static const struct fs_parameter_spec xfs_fs_parameters[] = {
 128         fsparam_u32("logbufs",          Opt_logbufs),
 129         fsparam_string("logbsize",      Opt_logbsize),
 130         fsparam_string("logdev",        Opt_logdev),
 131         fsparam_string("rtdev",         Opt_rtdev),
 132         fsparam_flag("wsync",           Opt_wsync),
 133         fsparam_flag("noalign",         Opt_noalign),
 134         fsparam_flag("swalloc",         Opt_swalloc),
 135         fsparam_u32("sunit",            Opt_sunit),
 136         fsparam_u32("swidth",           Opt_swidth),
 137         fsparam_flag("nouuid",          Opt_nouuid),
 138         fsparam_flag("grpid",           Opt_grpid),
 139         fsparam_flag("nogrpid",         Opt_nogrpid),
 140         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 141         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 142         fsparam_string("allocsize",     Opt_allocsize),
 143         fsparam_flag("norecovery",      Opt_norecovery),
 144         fsparam_flag("inode64",         Opt_inode64),
 145         fsparam_flag("inode32",         Opt_inode32),
 146         fsparam_flag("ikeep",           Opt_ikeep),
 147         fsparam_flag("noikeep",         Opt_noikeep),
 148         fsparam_flag("largeio",         Opt_largeio),
 149         fsparam_flag("nolargeio",       Opt_nolargeio),
 150         fsparam_flag("attr2",           Opt_attr2),
 151         fsparam_flag("noattr2",         Opt_noattr2),
 152         fsparam_flag("filestreams",     Opt_filestreams),
 153         fsparam_flag("quota",           Opt_quota),
 154         fsparam_flag("noquota",         Opt_noquota),
 155         fsparam_flag("usrquota",        Opt_usrquota),
 156         fsparam_flag("grpquota",        Opt_grpquota),
 157         fsparam_flag("prjquota",        Opt_prjquota),
 158         fsparam_flag("uquota",          Opt_uquota),
 159         fsparam_flag("gquota",          Opt_gquota),
 160         fsparam_flag("pquota",          Opt_pquota),
 161         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 162         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 163         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 164         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 165         fsparam_flag("discard",         Opt_discard),
 166         fsparam_flag("nodiscard",       Opt_nodiscard),
 167         fsparam_flag("dax",             Opt_dax),
 168         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 169         {}
 170 };
 171
 172 struct proc_xfs_info {
 173         uint64_t        flag;
 174         char            *str;
 175 };
 176
 177 static int
 178 xfs_fs_show_options(
 179         struct seq_file         *m,
 180         struct dentry           *root)
 181 {
 182         static struct proc_xfs_info xfs_info_set[] = {
 183                 /* the few simple ones we can get from the mount struct */
 184                 { XFS_FEAT_IKEEP,               ",ikeep" },
 185                 { XFS_FEAT_WSYNC,               ",wsync" },
 186                 { XFS_FEAT_NOALIGN,             ",noalign" },
 187                 { XFS_FEAT_SWALLOC,             ",swalloc" },
 188                 { XFS_FEAT_NOUUID,              ",nouuid" },
 189                 { XFS_FEAT_NORECOVERY,          ",norecovery" },
 190                 { XFS_FEAT_ATTR2,               ",attr2" },
 191                 { XFS_FEAT_FILESTREAMS,         ",filestreams" },
 192                 { XFS_FEAT_GRPID,               ",grpid" },
 193                 { XFS_FEAT_DISCARD,             ",discard" },
 194                 { XFS_FEAT_LARGE_IOSIZE,        ",largeio" },
 195                 { XFS_FEAT_DAX_ALWAYS,          ",dax=always" },
 196                 { XFS_FEAT_DAX_NEVER,           ",dax=never" },
 197                 { 0, NULL }
 198         };
 199         struct xfs_mount        *mp = XFS_M(root->d_sb);
 200         struct proc_xfs_info    *xfs_infop;
 201
 202         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 203                 if (mp->m_features & xfs_infop->flag)
 204                         seq_puts(m, xfs_infop->str);
 205         }
 206
 207         seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
 208
 209         if (xfs_has_allocsize(mp))
 210                 seq_printf(m, ",allocsize=%dk",
 211                            (1 << mp->m_allocsize_log) >> 10);
 212
 213         if (mp->m_logbufs > 0)
 214                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 215         if (mp->m_logbsize > 0)
 216                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 217
 218         if (mp->m_logname)
 219                 seq_show_option(m, "logdev", mp->m_logname);
 220         if (mp->m_rtname)
 221                 seq_show_option(m, "rtdev", mp->m_rtname);
 222
 223         if (mp->m_dalign > 0)
 224                 seq_printf(m, ",sunit=%d",
 225                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 226         if (mp->m_swidth > 0)
 227                 seq_printf(m, ",swidth=%d",
 228                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 229
 230         if (mp->m_qflags & XFS_UQUOTA_ENFD)
 231                 seq_puts(m, ",usrquota");
 232         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 233                 seq_puts(m, ",uqnoenforce");
 234
 235         if (mp->m_qflags & XFS_PQUOTA_ENFD)
 236                 seq_puts(m, ",prjquota");
 237         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
 238                 seq_puts(m, ",pqnoenforce");
 239
 240         if (mp->m_qflags & XFS_GQUOTA_ENFD)
 241                 seq_puts(m, ",grpquota");
 242         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
 243                 seq_puts(m, ",gqnoenforce");
 244
 245         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 246                 seq_puts(m, ",noquota");
 247
 248         return 0;
 249 }
 250
 251 static bool
 252 xfs_set_inode_alloc_perag(
 253         struct xfs_perag        *pag,
 254         xfs_ino_t               ino,
 255         xfs_agnumber_t          max_metadata)
 256 {
 257         if (!xfs_is_inode32(pag->pag_mount)) {
 258                 set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
 259                 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 260                 return false;
 261         }
 262
 263         if (ino > XFS_MAXINUMBER_32) {
 264                 clear_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
 265                 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 266                 return false;
 267         }
 268
 269         set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
 270         if (pag->pag_agno < max_metadata)
 271                 set_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 272         else
 273                 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 274         return true;
 275 }
 276
 277 /*
 278  * Set parameters for inode allocation heuristics, taking into account
 279  * filesystem size and inode32/inode64 mount options; i.e. specifically
 280  * whether or not XFS_FEAT_SMALL_INUMS is set.
 281  *
 282  * Inode allocation patterns are altered only if inode32 is requested
 283  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
 284  * If altered, XFS_OPSTATE_INODE32 is set as well.
 285  *
 286  * An agcount independent of that in the mount structure is provided
 287  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 288  * to the potentially higher ag count.
 289  *
 290  * Returns the maximum AG index which may contain inodes.
 291  */
 292 xfs_agnumber_t
 293 xfs_set_inode_alloc(
 294         struct xfs_mount *mp,
 295         xfs_agnumber_t  agcount)
 296 {
 297         xfs_agnumber_t  index;
 298         xfs_agnumber_t  maxagi = 0;
 299         xfs_sb_t        *sbp = &mp->m_sb;
 300         xfs_agnumber_t  max_metadata;
 301         xfs_agino_t     agino;
 302         xfs_ino_t       ino;
 303
 304         /*
 305          * Calculate how much should be reserved for inodes to meet
 306          * the max inode percentage.  Used only for inode32.
 307          */
 308         if (M_IGEO(mp)->maxicount) {
 309                 uint64_t        icount;
 310
 311                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 312                 do_div(icount, 100);
 313                 icount += sbp->sb_agblocks - 1;
 314                 do_div(icount, sbp->sb_agblocks);
 315                 max_metadata = icount;
 316         } else {
 317                 max_metadata = agcount;
 318         }
 319
 320         /* Get the last possible inode in the filesystem */
 321         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 322         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 323
 324         /*
 325          * If user asked for no more than 32-bit inodes, and the fs is
 326          * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
 327          * the allocator to accommodate the request.
 328          */
 329         if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
 330                 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 331         else
 332                 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 333
 334         for (index = 0; index < agcount; index++) {
 335                 struct xfs_perag        *pag;
 336
 337                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 338
 339                 pag = xfs_perag_get(mp, index);
 340                 if (xfs_set_inode_alloc_perag(pag, ino, max_metadata))
 341                         maxagi++;
 342                 xfs_perag_put(pag);
 343         }
 344
 345         return xfs_is_inode32(mp) ? maxagi : agcount;
 346 }
 347
 348 static int
 349 xfs_setup_dax_always(
 350         struct xfs_mount        *mp)
 351 {
 352         if (!mp->m_ddev_targp->bt_daxdev &&
 353             (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
 354                 xfs_alert(mp,
 355                         "DAX unsupported by block device. Turning off DAX.");
 356                 goto disable_dax;
 357         }
 358
 359         if (mp->m_super->s_blocksize != PAGE_SIZE) {
 360                 xfs_alert(mp,
 361                         "DAX not supported for blocksize. Turning off DAX.");
 362                 goto disable_dax;
 363         }
 364
 365         if (xfs_has_reflink(mp) &&
 366             bdev_is_partition(mp->m_ddev_targp->bt_bdev)) {
 367                 xfs_alert(mp,
 368                         "DAX and reflink cannot work with multi-partitions!");
 369                 return -EINVAL;
 370         }
 371
 372         xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
 373         return 0;
 374
 375 disable_dax:
 376         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
 377         return 0;
 378 }
 379
 380 static void
 381 xfs_bdev_mark_dead(
 382         struct block_device     *bdev)
 383 {
 384         xfs_force_shutdown(bdev->bd_holder, SHUTDOWN_DEVICE_REMOVED);
 385 }
 386
 387 static const struct blk_holder_ops xfs_holder_ops = {
 388         .mark_dead              = xfs_bdev_mark_dead,
 389 };
 390
 391 STATIC int
 392 xfs_blkdev_get(
 393         xfs_mount_t             *mp,
 394         const char              *name,
 395         struct block_device     **bdevp)
 396 {
 397         int                     error = 0;
 398
 399         *bdevp = blkdev_get_by_path(name, BLK_OPEN_READ | BLK_OPEN_WRITE, mp,
 400                                     &xfs_holder_ops);
 401         if (IS_ERR(*bdevp)) {
 402                 error = PTR_ERR(*bdevp);
 403                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 404         }
 405
 406         return error;
 407 }
 408
 409 STATIC void
 410 xfs_blkdev_put(
 411         struct xfs_mount        *mp,
 412         struct block_device     *bdev)
 413 {
 414         if (bdev)
 415                 blkdev_put(bdev, mp);
 416 }
 417
 418 STATIC void
 419 xfs_close_devices(
 420         struct xfs_mount        *mp)
 421 {
 422         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 423                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 424
 425                 xfs_free_buftarg(mp->m_logdev_targp);
 426                 xfs_blkdev_put(mp, logdev);
 427         }
 428         if (mp->m_rtdev_targp) {
 429                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 430
 431                 xfs_free_buftarg(mp->m_rtdev_targp);
 432                 xfs_blkdev_put(mp, rtdev);
 433         }
 434         xfs_free_buftarg(mp->m_ddev_targp);
 435 }
 436
 437 /*
 438  * The file system configurations are:
 439  *      (1) device (partition) with data and internal log
 440  *      (2) logical volume with data and log subvolumes.
 441  *      (3) logical volume with data, log, and realtime subvolumes.
 442  *
 443  * We only have to handle opening the log and realtime volumes here if
 444  * they are present.  The data subvolume has already been opened by
 445  * get_sb_bdev() and is stored in sb->s_bdev.
 446  */
 447 STATIC int
 448 xfs_open_devices(
 449         struct xfs_mount        *mp)
 450 {
 451         struct block_device     *ddev = mp->m_super->s_bdev;
 452         struct block_device     *logdev = NULL, *rtdev = NULL;
 453         int                     error;
 454
 455         /*
 456          * Open real time and log devices - order is important.
 457          */
 458         if (mp->m_logname) {
 459                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 460                 if (error)
 461                         return error;
 462         }
 463
 464         if (mp->m_rtname) {
 465                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 466                 if (error)
 467                         goto out_close_logdev;
 468
 469                 if (rtdev == ddev || rtdev == logdev) {
 470                         xfs_warn(mp,
 471         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 472                         error = -EINVAL;
 473                         goto out_close_rtdev;
 474                 }
 475         }
 476
 477         /*
 478          * Setup xfs_mount buffer target pointers
 479          */
 480         error = -ENOMEM;
 481         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
 482         if (!mp->m_ddev_targp)
 483                 goto out_close_rtdev;
 484
 485         if (rtdev) {
 486                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
 487                 if (!mp->m_rtdev_targp)
 488                         goto out_free_ddev_targ;
 489         }
 490
 491         if (logdev && logdev != ddev) {
 492                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
 493                 if (!mp->m_logdev_targp)
 494                         goto out_free_rtdev_targ;
 495         } else {
 496                 mp->m_logdev_targp = mp->m_ddev_targp;
 497         }
 498
 499         return 0;
 500
 501  out_free_rtdev_targ:
 502         if (mp->m_rtdev_targp)
 503                 xfs_free_buftarg(mp->m_rtdev_targp);
 504  out_free_ddev_targ:
 505         xfs_free_buftarg(mp->m_ddev_targp);
 506  out_close_rtdev:
 507         xfs_blkdev_put(mp, rtdev);
 508  out_close_logdev:
 509         if (logdev && logdev != ddev)
 510                 xfs_blkdev_put(mp, logdev);
 511         return error;
 512 }
 513
 514 /*
 515  * Setup xfs_mount buffer target pointers based on superblock
 516  */
 517 STATIC int
 518 xfs_setup_devices(
 519         struct xfs_mount        *mp)
 520 {
 521         int                     error;
 522
 523         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 524         if (error)
 525                 return error;
 526
 527         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 528                 unsigned int    log_sector_size = BBSIZE;
 529
 530                 if (xfs_has_sector(mp))
 531                         log_sector_size = mp->m_sb.sb_logsectsize;
 532                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 533                                             log_sector_size);
 534                 if (error)
 535                         return error;
 536         }
 537         if (mp->m_rtdev_targp) {
 538                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 539                                             mp->m_sb.sb_sectsize);
 540                 if (error)
 541                         return error;
 542         }
 543
 544         return 0;
 545 }
 546
 547 STATIC int
 548 xfs_init_mount_workqueues(
 549         struct xfs_mount        *mp)
 550 {
 551         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 552                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 553                         1, mp->m_super->s_id);
 554         if (!mp->m_buf_workqueue)
 555                 goto out;
 556
 557         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 558                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 559                         0, mp->m_super->s_id);
 560         if (!mp->m_unwritten_workqueue)
 561                 goto out_destroy_buf;
 562
 563         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 564                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 565                         0, mp->m_super->s_id);
 566         if (!mp->m_reclaim_workqueue)
 567                 goto out_destroy_unwritten;
 568
 569         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
 570                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
 571                         0, mp->m_super->s_id);
 572         if (!mp->m_blockgc_wq)
 573                 goto out_destroy_reclaim;
 574
 575         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
 576                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 577                         1, mp->m_super->s_id);
 578         if (!mp->m_inodegc_wq)
 579                 goto out_destroy_blockgc;
 580
 581         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 582                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 583         if (!mp->m_sync_workqueue)
 584                 goto out_destroy_inodegc;
 585
 586         return 0;
 587
 588 out_destroy_inodegc:
 589         destroy_workqueue(mp->m_inodegc_wq);
 590 out_destroy_blockgc:
 591         destroy_workqueue(mp->m_blockgc_wq);
 592 out_destroy_reclaim:
 593         destroy_workqueue(mp->m_reclaim_workqueue);
 594 out_destroy_unwritten:
 595         destroy_workqueue(mp->m_unwritten_workqueue);
 596 out_destroy_buf:
 597         destroy_workqueue(mp->m_buf_workqueue);
 598 out:
 599         return -ENOMEM;
 600 }
 601
 602 STATIC void
 603 xfs_destroy_mount_workqueues(
 604         struct xfs_mount        *mp)
 605 {
 606         destroy_workqueue(mp->m_sync_workqueue);
 607         destroy_workqueue(mp->m_blockgc_wq);
 608         destroy_workqueue(mp->m_inodegc_wq);
 609         destroy_workqueue(mp->m_reclaim_workqueue);
 610         destroy_workqueue(mp->m_unwritten_workqueue);
 611         destroy_workqueue(mp->m_buf_workqueue);
 612 }
 613
 614 static void
 615 xfs_flush_inodes_worker(
 616         struct work_struct      *work)
 617 {
 618         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 619                                                    m_flush_inodes_work);
 620         struct super_block      *sb = mp->m_super;
 621
 622         if (down_read_trylock(&sb->s_umount)) {
 623                 sync_inodes_sb(sb);
 624                 up_read(&sb->s_umount);
 625         }
 626 }
 627
 628 /*
 629  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 630  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 631  * for IO to complete so that we effectively throttle multiple callers to the
 632  * rate at which IO is completing.
 633  */
 634 void
 635 xfs_flush_inodes(
 636         struct xfs_mount        *mp)
 637 {
 638         /*
 639          * If flush_work() returns true then that means we waited for a flush
 640          * which was already in progress.  Don't bother running another scan.
 641          */
 642         if (flush_work(&mp->m_flush_inodes_work))
 643                 return;
 644
 645         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 646         flush_work(&mp->m_flush_inodes_work);
 647 }
 648
 649 /* Catch misguided souls that try to use this interface on XFS */
 650 STATIC struct inode *
 651 xfs_fs_alloc_inode(
 652         struct super_block      *sb)
 653 {
 654         BUG();
 655         return NULL;
 656 }
 657
 658 /*
 659  * Now that the generic code is guaranteed not to be accessing
 660  * the linux inode, we can inactivate and reclaim the inode.
 661  */
 662 STATIC void
 663 xfs_fs_destroy_inode(
 664         struct inode            *inode)
 665 {
 666         struct xfs_inode        *ip = XFS_I(inode);
 667
 668         trace_xfs_destroy_inode(ip);
 669
 670         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 671         XFS_STATS_INC(ip->i_mount, vn_rele);
 672         XFS_STATS_INC(ip->i_mount, vn_remove);
 673         xfs_inode_mark_reclaimable(ip);
 674 }
 675
 676 static void
 677 xfs_fs_dirty_inode(
 678         struct inode                    *inode,
 679         int                             flags)
 680 {
 681         struct xfs_inode                *ip = XFS_I(inode);
 682         struct xfs_mount                *mp = ip->i_mount;
 683         struct xfs_trans                *tp;
 684
 685         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 686                 return;
 687
 688         /*
 689          * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC)
 690          * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed
 691          * in flags possibly together with I_DIRTY_SYNC.
 692          */
 693         if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME))
 694                 return;
 695
 696         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 697                 return;
 698         xfs_ilock(ip, XFS_ILOCK_EXCL);
 699         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 700         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 701         xfs_trans_commit(tp);
 702 }
 703
 704 /*
 705  * Slab object creation initialisation for the XFS inode.
 706  * This covers only the idempotent fields in the XFS inode;
 707  * all other fields need to be initialised on allocation
 708  * from the slab. This avoids the need to repeatedly initialise
 709  * fields in the xfs inode that left in the initialise state
 710  * when freeing the inode.
 711  */
 712 STATIC void
 713 xfs_fs_inode_init_once(
 714         void                    *inode)
 715 {
 716         struct xfs_inode        *ip = inode;
 717
 718         memset(ip, 0, sizeof(struct xfs_inode));
 719
 720         /* vfs inode */
 721         inode_init_once(VFS_I(ip));
 722
 723         /* xfs inode */
 724         atomic_set(&ip->i_pincount, 0);
 725         spin_lock_init(&ip->i_flags_lock);
 726
 727         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 728                      "xfsino", ip->i_ino);
 729 }
 730
 731 /*
 732  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 733  * serialised against cache hits here via the inode->i_lock and igrab() in
 734  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 735  * racing with us, and it avoids needing to grab a spinlock here for every inode
 736  * we drop the final reference on.
 737  */
 738 STATIC int
 739 xfs_fs_drop_inode(
 740         struct inode            *inode)
 741 {
 742         struct xfs_inode        *ip = XFS_I(inode);
 743
 744         /*
 745          * If this unlinked inode is in the middle of recovery, don't
 746          * drop the inode just yet; log recovery will take care of
 747          * that.  See the comment for this inode flag.
 748          */
 749         if (ip->i_flags & XFS_IRECOVERY) {
 750                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
 751                 return 0;
 752         }
 753
 754         return generic_drop_inode(inode);
 755 }
 756
 757 static void
 758 xfs_mount_free(
 759         struct xfs_mount        *mp)
 760 {
 761         kfree(mp->m_rtname);
 762         kfree(mp->m_logname);
 763         kmem_free(mp);
 764 }
 765
 766 STATIC int
 767 xfs_fs_sync_fs(
 768         struct super_block      *sb,
 769         int                     wait)
 770 {
 771         struct xfs_mount        *mp = XFS_M(sb);
 772         int                     error;
 773
 774         trace_xfs_fs_sync_fs(mp, __return_address);
 775
 776         /*
 777          * Doing anything during the async pass would be counterproductive.
 778          */
 779         if (!wait)
 780                 return 0;
 781
 782         error = xfs_log_force(mp, XFS_LOG_SYNC);
 783         if (error)
 784                 return error;
 785
 786         if (laptop_mode) {
 787                 /*
 788                  * The disk must be active because we're syncing.
 789                  * We schedule log work now (now that the disk is
 790                  * active) instead of later (when it might not be).
 791                  */
 792                 flush_delayed_work(&mp->m_log->l_work);
 793         }
 794
 795         /*
 796          * If we are called with page faults frozen out, it means we are about
 797          * to freeze the transaction subsystem. Take the opportunity to shut
 798          * down inodegc because once SB_FREEZE_FS is set it's too late to
 799          * prevent inactivation races with freeze. The fs doesn't get called
 800          * again by the freezing process until after SB_FREEZE_FS has been set,
 801          * so it's now or never.  Same logic applies to speculative allocation
 802          * garbage collection.
 803          *
 804          * We don't care if this is a normal syncfs call that does this or
 805          * freeze that does this - we can run this multiple times without issue
 806          * and we won't race with a restart because a restart can only occur
 807          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
 808          */
 809         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
 810                 xfs_inodegc_stop(mp);
 811                 xfs_blockgc_stop(mp);
 812         }
 813
 814         return 0;
 815 }
 816
 817 STATIC int
 818 xfs_fs_statfs(
 819         struct dentry           *dentry,
 820         struct kstatfs          *statp)
 821 {
 822         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 823         xfs_sb_t                *sbp = &mp->m_sb;
 824         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 825         uint64_t                fakeinos, id;
 826         uint64_t                icount;
 827         uint64_t                ifree;
 828         uint64_t                fdblocks;
 829         xfs_extlen_t            lsize;
 830         int64_t                 ffree;
 831
 832         /*
 833          * Expedite background inodegc but don't wait. We do not want to block
 834          * here waiting hours for a billion extent file to be truncated.
 835          */
 836         xfs_inodegc_push(mp);
 837
 838         statp->f_type = XFS_SUPER_MAGIC;
 839         statp->f_namelen = MAXNAMELEN - 1;
 840
 841         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 842         statp->f_fsid = u64_to_fsid(id);
 843
 844         icount = percpu_counter_sum(&mp->m_icount);
 845         ifree = percpu_counter_sum(&mp->m_ifree);
 846         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 847
 848         spin_lock(&mp->m_sb_lock);
 849         statp->f_bsize = sbp->sb_blocksize;
 850         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 851         statp->f_blocks = sbp->sb_dblocks - lsize;
 852         spin_unlock(&mp->m_sb_lock);
 853
 854         /* make sure statp->f_bfree does not underflow */
 855         statp->f_bfree = max_t(int64_t, 0,
 856                                 fdblocks - xfs_fdblocks_unavailable(mp));
 857         statp->f_bavail = statp->f_bfree;
 858
 859         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 860         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 861         if (M_IGEO(mp)->maxicount)
 862                 statp->f_files = min_t(typeof(statp->f_files),
 863                                         statp->f_files,
 864                                         M_IGEO(mp)->maxicount);
 865
 866         /* If sb_icount overshot maxicount, report actual allocation */
 867         statp->f_files = max_t(typeof(statp->f_files),
 868                                         statp->f_files,
 869                                         sbp->sb_icount);
 870
 871         /* make sure statp->f_ffree does not underflow */
 872         ffree = statp->f_files - (icount - ifree);
 873         statp->f_ffree = max_t(int64_t, ffree, 0);
 874
 875
 876         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 877             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 878                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 879                 xfs_qm_statvfs(ip, statp);
 880
 881         if (XFS_IS_REALTIME_MOUNT(mp) &&
 882             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 883                 s64     freertx;
 884
 885                 statp->f_blocks = sbp->sb_rblocks;
 886                 freertx = percpu_counter_sum_positive(&mp->m_frextents);
 887                 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
 888         }
 889
 890         return 0;
 891 }
 892
 893 STATIC void
 894 xfs_save_resvblks(struct xfs_mount *mp)
 895 {
 896         uint64_t resblks = 0;
 897
 898         mp->m_resblks_save = mp->m_resblks;
 899         xfs_reserve_blocks(mp, &resblks, NULL);
 900 }
 901
 902 STATIC void
 903 xfs_restore_resvblks(struct xfs_mount *mp)
 904 {
 905         uint64_t resblks;
 906
 907         if (mp->m_resblks_save) {
 908                 resblks = mp->m_resblks_save;
 909                 mp->m_resblks_save = 0;
 910         } else
 911                 resblks = xfs_default_resblks(mp);
 912
 913         xfs_reserve_blocks(mp, &resblks, NULL);
 914 }
 915
 916 /*
 917  * Second stage of a freeze. The data is already frozen so we only
 918  * need to take care of the metadata. Once that's done sync the superblock
 919  * to the log to dirty it in case of a crash while frozen. This ensures that we
 920  * will recover the unlinked inode lists on the next mount.
 921  */
 922 STATIC int
 923 xfs_fs_freeze(
 924         struct super_block      *sb)
 925 {
 926         struct xfs_mount        *mp = XFS_M(sb);
 927         unsigned int            flags;
 928         int                     ret;
 929
 930         /*
 931          * The filesystem is now frozen far enough that memory reclaim
 932          * cannot safely operate on the filesystem. Hence we need to
 933          * set a GFP_NOFS context here to avoid recursion deadlocks.
 934          */
 935         flags = memalloc_nofs_save();
 936         xfs_save_resvblks(mp);
 937         ret = xfs_log_quiesce(mp);
 938         memalloc_nofs_restore(flags);
 939
 940         /*
 941          * For read-write filesystems, we need to restart the inodegc on error
 942          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
 943          * going to be run to restart it now.  We are at SB_FREEZE_FS level
 944          * here, so we can restart safely without racing with a stop in
 945          * xfs_fs_sync_fs().
 946          */
 947         if (ret && !xfs_is_readonly(mp)) {
 948                 xfs_blockgc_start(mp);
 949                 xfs_inodegc_start(mp);
 950         }
 951
 952         return ret;
 953 }
 954
 955 STATIC int
 956 xfs_fs_unfreeze(
 957         struct super_block      *sb)
 958 {
 959         struct xfs_mount        *mp = XFS_M(sb);
 960
 961         xfs_restore_resvblks(mp);
 962         xfs_log_work_queue(mp);
 963
 964         /*
 965          * Don't reactivate the inodegc worker on a readonly filesystem because
 966          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
 967          * worker because there are no speculative preallocations on a readonly
 968          * filesystem.
 969          */
 970         if (!xfs_is_readonly(mp)) {
 971                 xfs_blockgc_start(mp);
 972                 xfs_inodegc_start(mp);
 973         }
 974
 975         return 0;
 976 }
 977
 978 /*
 979  * This function fills in xfs_mount_t fields based on mount args.
 980  * Note: the superblock _has_ now been read in.
 981  */
 982 STATIC int
 983 xfs_finish_flags(
 984         struct xfs_mount        *mp)
 985 {
 986         /* Fail a mount where the logbuf is smaller than the log stripe */
 987         if (xfs_has_logv2(mp)) {
 988                 if (mp->m_logbsize <= 0 &&
 989                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 990                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 991                 } else if (mp->m_logbsize > 0 &&
 992                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 993                         xfs_warn(mp,
 994                 "logbuf size must be greater than or equal to log stripe size");
 995                         return -EINVAL;
 996                 }
 997         } else {
 998                 /* Fail a mount if the logbuf is larger than 32K */
 999                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1000                         xfs_warn(mp,
1001                 "logbuf size for version 1 logs must be 16K or 32K");
1002                         return -EINVAL;
1003                 }
1004         }
1005
1006         /*
1007          * V5 filesystems always use attr2 format for attributes.
1008          */
1009         if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
1010                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1011                              "attr2 is always enabled for V5 filesystems.");
1012                 return -EINVAL;
1013         }
1014
1015         /*
1016          * prohibit r/w mounts of read-only filesystems
1017          */
1018         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
1019                 xfs_warn(mp,
1020                         "cannot mount a read-only filesystem as read-write");
1021                 return -EROFS;
1022         }
1023
1024         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
1025             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
1026             !xfs_has_pquotino(mp)) {
1027                 xfs_warn(mp,
1028                   "Super block does not support project and group quota together");
1029                 return -EINVAL;
1030         }
1031
1032         return 0;
1033 }
1034
1035 static int
1036 xfs_init_percpu_counters(
1037         struct xfs_mount        *mp)
1038 {
1039         int             error;
1040
1041         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1042         if (error)
1043                 return -ENOMEM;
1044
1045         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1046         if (error)
1047                 goto free_icount;
1048
1049         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1050         if (error)
1051                 goto free_ifree;
1052
1053         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1054         if (error)
1055                 goto free_fdblocks;
1056
1057         error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1058         if (error)
1059                 goto free_delalloc;
1060
1061         return 0;
1062
1063 free_delalloc:
1064         percpu_counter_destroy(&mp->m_delalloc_blks);
1065 free_fdblocks:
1066         percpu_counter_destroy(&mp->m_fdblocks);
1067 free_ifree:
1068         percpu_counter_destroy(&mp->m_ifree);
1069 free_icount:
1070         percpu_counter_destroy(&mp->m_icount);
1071         return -ENOMEM;
1072 }
1073
1074 void
1075 xfs_reinit_percpu_counters(
1076         struct xfs_mount        *mp)
1077 {
1078         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1079         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1080         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1081         percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1082 }
1083
1084 static void
1085 xfs_destroy_percpu_counters(
1086         struct xfs_mount        *mp)
1087 {
1088         percpu_counter_destroy(&mp->m_icount);
1089         percpu_counter_destroy(&mp->m_ifree);
1090         percpu_counter_destroy(&mp->m_fdblocks);
1091         ASSERT(xfs_is_shutdown(mp) ||
1092                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1093         percpu_counter_destroy(&mp->m_delalloc_blks);
1094         percpu_counter_destroy(&mp->m_frextents);
1095 }
1096
1097 static int
1098 xfs_inodegc_init_percpu(
1099         struct xfs_mount        *mp)
1100 {
1101         struct xfs_inodegc      *gc;
1102         int                     cpu;
1103
1104         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1105         if (!mp->m_inodegc)
1106                 return -ENOMEM;
1107
1108         for_each_possible_cpu(cpu) {
1109                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1110 #if defined(DEBUG) || defined(XFS_WARN)
1111                 gc->cpu = cpu;
1112 #endif
1113                 init_llist_head(&gc->list);
1114                 gc->items = 0;
1115                 gc->error = 0;
1116                 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1117         }
1118         return 0;
1119 }
1120
1121 static void
1122 xfs_inodegc_free_percpu(
1123         struct xfs_mount        *mp)
1124 {
1125         if (!mp->m_inodegc)
1126                 return;
1127         free_percpu(mp->m_inodegc);
1128 }
1129
1130 static void
1131 xfs_fs_put_super(
1132         struct super_block      *sb)
1133 {
1134         struct xfs_mount        *mp = XFS_M(sb);
1135
1136         /* if ->fill_super failed, we have no mount to tear down */
1137         if (!sb->s_fs_info)
1138                 return;
1139
1140         xfs_notice(mp, "Unmounting Filesystem %pU", &mp->m_sb.sb_uuid);
1141         xfs_filestream_unmount(mp);
1142         xfs_unmountfs(mp);
1143
1144         xfs_freesb(mp);
1145         free_percpu(mp->m_stats.xs_stats);
1146         xfs_mount_list_del(mp);
1147         xfs_inodegc_free_percpu(mp);
1148         xfs_destroy_percpu_counters(mp);
1149         xfs_destroy_mount_workqueues(mp);
1150         xfs_close_devices(mp);
1151
1152         sb->s_fs_info = NULL;
1153         xfs_mount_free(mp);
1154 }
1155
1156 static long
1157 xfs_fs_nr_cached_objects(
1158         struct super_block      *sb,
1159         struct shrink_control   *sc)
1160 {
1161         /* Paranoia: catch incorrect calls during mount setup or teardown */
1162         if (WARN_ON_ONCE(!sb->s_fs_info))
1163                 return 0;
1164         return xfs_reclaim_inodes_count(XFS_M(sb));
1165 }
1166
1167 static long
1168 xfs_fs_free_cached_objects(
1169         struct super_block      *sb,
1170         struct shrink_control   *sc)
1171 {
1172         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1173 }
1174
1175 static void
1176 xfs_fs_shutdown(
1177         struct super_block      *sb)
1178 {
1179         xfs_force_shutdown(XFS_M(sb), SHUTDOWN_DEVICE_REMOVED);
1180 }
1181
1182 static const struct super_operations xfs_super_operations = {
1183         .alloc_inode            = xfs_fs_alloc_inode,
1184         .destroy_inode          = xfs_fs_destroy_inode,
1185         .dirty_inode            = xfs_fs_dirty_inode,
1186         .drop_inode             = xfs_fs_drop_inode,
1187         .put_super              = xfs_fs_put_super,
1188         .sync_fs                = xfs_fs_sync_fs,
1189         .freeze_fs              = xfs_fs_freeze,
1190         .unfreeze_fs            = xfs_fs_unfreeze,
1191         .statfs                 = xfs_fs_statfs,
1192         .show_options           = xfs_fs_show_options,
1193         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1194         .free_cached_objects    = xfs_fs_free_cached_objects,
1195         .shutdown               = xfs_fs_shutdown,
1196 };
1197
1198 static int
1199 suffix_kstrtoint(
1200         const char      *s,
1201         unsigned int    base,
1202         int             *res)
1203 {
1204         int             last, shift_left_factor = 0, _res;
1205         char            *value;
1206         int             ret = 0;
1207
1208         value = kstrdup(s, GFP_KERNEL);
1209         if (!value)
1210                 return -ENOMEM;
1211
1212         last = strlen(value) - 1;
1213         if (value[last] == 'K' || value[last] == 'k') {
1214                 shift_left_factor = 10;
1215                 value[last] = '\0';
1216         }
1217         if (value[last] == 'M' || value[last] == 'm') {
1218                 shift_left_factor = 20;
1219                 value[last] = '\0';
1220         }
1221         if (value[last] == 'G' || value[last] == 'g') {
1222                 shift_left_factor = 30;
1223                 value[last] = '\0';
1224         }
1225
1226         if (kstrtoint(value, base, &_res))
1227                 ret = -EINVAL;
1228         kfree(value);
1229         *res = _res << shift_left_factor;
1230         return ret;
1231 }
1232
1233 static inline void
1234 xfs_fs_warn_deprecated(
1235         struct fs_context       *fc,
1236         struct fs_parameter     *param,
1237         uint64_t                flag,
1238         bool                    value)
1239 {
1240         /* Don't print the warning if reconfiguring and current mount point
1241          * already had the flag set
1242          */
1243         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1244             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1245                 return;
1246         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1247 }
1248
1249 /*
1250  * Set mount state from a mount option.
1251  *
1252  * NOTE: mp->m_super is NULL here!
1253  */
1254 static int
1255 xfs_fs_parse_param(
1256         struct fs_context       *fc,
1257         struct fs_parameter     *param)
1258 {
1259         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1260         struct fs_parse_result  result;
1261         int                     size = 0;
1262         int                     opt;
1263
1264         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1265         if (opt < 0)
1266                 return opt;
1267
1268         switch (opt) {
1269         case Opt_logbufs:
1270                 parsing_mp->m_logbufs = result.uint_32;
1271                 return 0;
1272         case Opt_logbsize:
1273                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1274                         return -EINVAL;
1275                 return 0;
1276         case Opt_logdev:
1277                 kfree(parsing_mp->m_logname);
1278                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1279                 if (!parsing_mp->m_logname)
1280                         return -ENOMEM;
1281                 return 0;
1282         case Opt_rtdev:
1283                 kfree(parsing_mp->m_rtname);
1284                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1285                 if (!parsing_mp->m_rtname)
1286                         return -ENOMEM;
1287                 return 0;
1288         case Opt_allocsize:
1289                 if (suffix_kstrtoint(param->string, 10, &size))
1290                         return -EINVAL;
1291                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1292                 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1293                 return 0;
1294         case Opt_grpid:
1295         case Opt_bsdgroups:
1296                 parsing_mp->m_features |= XFS_FEAT_GRPID;
1297                 return 0;
1298         case Opt_nogrpid:
1299         case Opt_sysvgroups:
1300                 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1301                 return 0;
1302         case Opt_wsync:
1303                 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1304                 return 0;
1305         case Opt_norecovery:
1306                 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1307                 return 0;
1308         case Opt_noalign:
1309                 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1310                 return 0;
1311         case Opt_swalloc:
1312                 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1313                 return 0;
1314         case Opt_sunit:
1315                 parsing_mp->m_dalign = result.uint_32;
1316                 return 0;
1317         case Opt_swidth:
1318                 parsing_mp->m_swidth = result.uint_32;
1319                 return 0;
1320         case Opt_inode32:
1321                 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1322                 return 0;
1323         case Opt_inode64:
1324                 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1325                 return 0;
1326         case Opt_nouuid:
1327                 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1328                 return 0;
1329         case Opt_largeio:
1330                 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1331                 return 0;
1332         case Opt_nolargeio:
1333                 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1334                 return 0;
1335         case Opt_filestreams:
1336                 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1337                 return 0;
1338         case Opt_noquota:
1339                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1340                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1341                 return 0;
1342         case Opt_quota:
1343         case Opt_uquota:
1344         case Opt_usrquota:
1345                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1346                 return 0;
1347         case Opt_qnoenforce:
1348         case Opt_uqnoenforce:
1349                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1350                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1351                 return 0;
1352         case Opt_pquota:
1353         case Opt_prjquota:
1354                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1355                 return 0;
1356         case Opt_pqnoenforce:
1357                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1358                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1359                 return 0;
1360         case Opt_gquota:
1361         case Opt_grpquota:
1362                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1363                 return 0;
1364         case Opt_gqnoenforce:
1365                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1366                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1367                 return 0;
1368         case Opt_discard:
1369                 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1370                 return 0;
1371         case Opt_nodiscard:
1372                 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1373                 return 0;
1374 #ifdef CONFIG_FS_DAX
1375         case Opt_dax:
1376                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1377                 return 0;
1378         case Opt_dax_enum:
1379                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1380                 return 0;
1381 #endif
1382         /* Following mount options will be removed in September 2025 */
1383         case Opt_ikeep:
1384                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1385                 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1386                 return 0;
1387         case Opt_noikeep:
1388                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1389                 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1390                 return 0;
1391         case Opt_attr2:
1392                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1393                 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1394                 return 0;
1395         case Opt_noattr2:
1396                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1397                 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1398                 return 0;
1399         default:
1400                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1401                 return -EINVAL;
1402         }
1403
1404         return 0;
1405 }
1406
1407 static int
1408 xfs_fs_validate_params(
1409         struct xfs_mount        *mp)
1410 {
1411         /* No recovery flag requires a read-only mount */
1412         if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1413                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1414                 return -EINVAL;
1415         }
1416
1417         /*
1418          * We have not read the superblock at this point, so only the attr2
1419          * mount option can set the attr2 feature by this stage.
1420          */
1421         if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1422                 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1423                 return -EINVAL;
1424         }
1425
1426
1427         if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1428                 xfs_warn(mp,
1429         "sunit and swidth options incompatible with the noalign option");
1430                 return -EINVAL;
1431         }
1432
1433         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1434                 xfs_warn(mp, "quota support not available in this kernel.");
1435                 return -EINVAL;
1436         }
1437
1438         if ((mp->m_dalign && !mp->m_swidth) ||
1439             (!mp->m_dalign && mp->m_swidth)) {
1440                 xfs_warn(mp, "sunit and swidth must be specified together");
1441                 return -EINVAL;
1442         }
1443
1444         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1445                 xfs_warn(mp,
1446         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1447                         mp->m_swidth, mp->m_dalign);
1448                 return -EINVAL;
1449         }
1450
1451         if (mp->m_logbufs != -1 &&
1452             mp->m_logbufs != 0 &&
1453             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1454              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1455                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1456                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1457                 return -EINVAL;
1458         }
1459
1460         if (mp->m_logbsize != -1 &&
1461             mp->m_logbsize !=  0 &&
1462             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1463              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1464              !is_power_of_2(mp->m_logbsize))) {
1465                 xfs_warn(mp,
1466                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1467                         mp->m_logbsize);
1468                 return -EINVAL;
1469         }
1470
1471         if (xfs_has_allocsize(mp) &&
1472             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1473              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1474                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1475                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1476                 return -EINVAL;
1477         }
1478
1479         return 0;
1480 }
1481
1482 static int
1483 xfs_fs_fill_super(
1484         struct super_block      *sb,
1485         struct fs_context       *fc)
1486 {
1487         struct xfs_mount        *mp = sb->s_fs_info;
1488         struct inode            *root;
1489         int                     flags = 0, error;
1490
1491         mp->m_super = sb;
1492
1493         error = xfs_fs_validate_params(mp);
1494         if (error)
1495                 goto out_free_names;
1496
1497         sb_min_blocksize(sb, BBSIZE);
1498         sb->s_xattr = xfs_xattr_handlers;
1499         sb->s_export_op = &xfs_export_operations;
1500 #ifdef CONFIG_XFS_QUOTA
1501         sb->s_qcop = &xfs_quotactl_operations;
1502         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1503 #endif
1504         sb->s_op = &xfs_super_operations;
1505
1506         /*
1507          * Delay mount work if the debug hook is set. This is debug
1508          * instrumention to coordinate simulation of xfs mount failures with
1509          * VFS superblock operations
1510          */
1511         if (xfs_globals.mount_delay) {
1512                 xfs_notice(mp, "Delaying mount for %d seconds.",
1513                         xfs_globals.mount_delay);
1514                 msleep(xfs_globals.mount_delay * 1000);
1515         }
1516
1517         if (fc->sb_flags & SB_SILENT)
1518                 flags |= XFS_MFSI_QUIET;
1519
1520         error = xfs_open_devices(mp);
1521         if (error)
1522                 goto out_free_names;
1523
1524         error = xfs_init_mount_workqueues(mp);
1525         if (error)
1526                 goto out_close_devices;
1527
1528         error = xfs_init_percpu_counters(mp);
1529         if (error)
1530                 goto out_destroy_workqueues;
1531
1532         error = xfs_inodegc_init_percpu(mp);
1533         if (error)
1534                 goto out_destroy_counters;
1535
1536         /*
1537          * All percpu data structures requiring cleanup when a cpu goes offline
1538          * must be allocated before adding this @mp to the cpu-dead handler's
1539          * mount list.
1540          */
1541         xfs_mount_list_add(mp);
1542
1543         /* Allocate stats memory before we do operations that might use it */
1544         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1545         if (!mp->m_stats.xs_stats) {
1546                 error = -ENOMEM;
1547                 goto out_destroy_inodegc;
1548         }
1549
1550         error = xfs_readsb(mp, flags);
1551         if (error)
1552                 goto out_free_stats;
1553
1554         error = xfs_finish_flags(mp);
1555         if (error)
1556                 goto out_free_sb;
1557
1558         error = xfs_setup_devices(mp);
1559         if (error)
1560                 goto out_free_sb;
1561
1562         /* V4 support is undergoing deprecation. */
1563         if (!xfs_has_crc(mp)) {
1564 #ifdef CONFIG_XFS_SUPPORT_V4
1565                 xfs_warn_once(mp,
1566         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1567 #else
1568                 xfs_warn(mp,
1569         "Deprecated V4 format (crc=0) not supported by kernel.");
1570                 error = -EINVAL;
1571                 goto out_free_sb;
1572 #endif
1573         }
1574
1575         /* ASCII case insensitivity is undergoing deprecation. */
1576         if (xfs_has_asciici(mp)) {
1577 #ifdef CONFIG_XFS_SUPPORT_ASCII_CI
1578                 xfs_warn_once(mp,
1579         "Deprecated ASCII case-insensitivity feature (ascii-ci=1) will not be supported after September 2030.");
1580 #else
1581                 xfs_warn(mp,
1582         "Deprecated ASCII case-insensitivity feature (ascii-ci=1) not supported by kernel.");
1583                 error = -EINVAL;
1584                 goto out_free_sb;
1585 #endif
1586         }
1587
1588         /* Filesystem claims it needs repair, so refuse the mount. */
1589         if (xfs_has_needsrepair(mp)) {
1590                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1591                 error = -EFSCORRUPTED;
1592                 goto out_free_sb;
1593         }
1594
1595         /*
1596          * Don't touch the filesystem if a user tool thinks it owns the primary
1597          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1598          * we don't check them at all.
1599          */
1600         if (mp->m_sb.sb_inprogress) {
1601                 xfs_warn(mp, "Offline file system operation in progress!");
1602                 error = -EFSCORRUPTED;
1603                 goto out_free_sb;
1604         }
1605
1606         /*
1607          * Until this is fixed only page-sized or smaller data blocks work.
1608          */
1609         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1610                 xfs_warn(mp,
1611                 "File system with blocksize %d bytes. "
1612                 "Only pagesize (%ld) or less will currently work.",
1613                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1614                 error = -ENOSYS;
1615                 goto out_free_sb;
1616         }
1617
1618         /* Ensure this filesystem fits in the page cache limits */
1619         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1620             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1621                 xfs_warn(mp,
1622                 "file system too large to be mounted on this system.");
1623                 error = -EFBIG;
1624                 goto out_free_sb;
1625         }
1626
1627         /*
1628          * XFS block mappings use 54 bits to store the logical block offset.
1629          * This should suffice to handle the maximum file size that the VFS
1630          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1631          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1632          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1633          * to check this assertion.
1634          *
1635          * Avoid integer overflow by comparing the maximum bmbt offset to the
1636          * maximum pagecache offset in units of fs blocks.
1637          */
1638         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1639                 xfs_warn(mp,
1640 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1641                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1642                          XFS_MAX_FILEOFF);
1643                 error = -EINVAL;
1644                 goto out_free_sb;
1645         }
1646
1647         error = xfs_filestream_mount(mp);
1648         if (error)
1649                 goto out_free_sb;
1650
1651         /*
1652          * we must configure the block size in the superblock before we run the
1653          * full mount process as the mount process can lookup and cache inodes.
1654          */
1655         sb->s_magic = XFS_SUPER_MAGIC;
1656         sb->s_blocksize = mp->m_sb.sb_blocksize;
1657         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1658         sb->s_maxbytes = MAX_LFS_FILESIZE;
1659         sb->s_max_links = XFS_MAXLINK;
1660         sb->s_time_gran = 1;
1661         if (xfs_has_bigtime(mp)) {
1662                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1663                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1664         } else {
1665                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1666                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1667         }
1668         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1669         sb->s_iflags |= SB_I_CGROUPWB;
1670
1671         set_posix_acl_flag(sb);
1672
1673         /* version 5 superblocks support inode version counters. */
1674         if (xfs_has_crc(mp))
1675                 sb->s_flags |= SB_I_VERSION;
1676
1677         if (xfs_has_dax_always(mp)) {
1678                 error = xfs_setup_dax_always(mp);
1679                 if (error)
1680                         goto out_filestream_unmount;
1681         }
1682
1683         if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1684                 xfs_warn(mp,
1685         "mounting with \"discard\" option, but the device does not support discard");
1686                 mp->m_features &= ~XFS_FEAT_DISCARD;
1687         }
1688
1689         if (xfs_has_reflink(mp)) {
1690                 if (mp->m_sb.sb_rblocks) {
1691                         xfs_alert(mp,
1692         "reflink not compatible with realtime device!");
1693                         error = -EINVAL;
1694                         goto out_filestream_unmount;
1695                 }
1696
1697                 if (xfs_globals.always_cow) {
1698                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1699                         mp->m_always_cow = true;
1700                 }
1701         }
1702
1703         if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1704                 xfs_alert(mp,
1705         "reverse mapping btree not compatible with realtime device!");
1706                 error = -EINVAL;
1707                 goto out_filestream_unmount;
1708         }
1709
1710         error = xfs_mountfs(mp);
1711         if (error)
1712                 goto out_filestream_unmount;
1713
1714         root = igrab(VFS_I(mp->m_rootip));
1715         if (!root) {
1716                 error = -ENOENT;
1717                 goto out_unmount;
1718         }
1719         sb->s_root = d_make_root(root);
1720         if (!sb->s_root) {
1721                 error = -ENOMEM;
1722                 goto out_unmount;
1723         }
1724
1725         return 0;
1726
1727  out_filestream_unmount:
1728         xfs_filestream_unmount(mp);
1729  out_free_sb:
1730         xfs_freesb(mp);
1731  out_free_stats:
1732         free_percpu(mp->m_stats.xs_stats);
1733  out_destroy_inodegc:
1734         xfs_mount_list_del(mp);
1735         xfs_inodegc_free_percpu(mp);
1736  out_destroy_counters:
1737         xfs_destroy_percpu_counters(mp);
1738  out_destroy_workqueues:
1739         xfs_destroy_mount_workqueues(mp);
1740  out_close_devices:
1741         xfs_close_devices(mp);
1742  out_free_names:
1743         sb->s_fs_info = NULL;
1744         xfs_mount_free(mp);
1745         return error;
1746
1747  out_unmount:
1748         xfs_filestream_unmount(mp);
1749         xfs_unmountfs(mp);
1750         goto out_free_sb;
1751 }
1752
1753 static int
1754 xfs_fs_get_tree(
1755         struct fs_context       *fc)
1756 {
1757         return get_tree_bdev(fc, xfs_fs_fill_super);
1758 }
1759
1760 static int
1761 xfs_remount_rw(
1762         struct xfs_mount        *mp)
1763 {
1764         struct xfs_sb           *sbp = &mp->m_sb;
1765         int error;
1766
1767         if (xfs_has_norecovery(mp)) {
1768                 xfs_warn(mp,
1769                         "ro->rw transition prohibited on norecovery mount");
1770                 return -EINVAL;
1771         }
1772
1773         if (xfs_sb_is_v5(sbp) &&
1774             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1775                 xfs_warn(mp,
1776         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1777                         (sbp->sb_features_ro_compat &
1778                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1779                 return -EINVAL;
1780         }
1781
1782         clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1783
1784         /*
1785          * If this is the first remount to writeable state we might have some
1786          * superblock changes to update.
1787          */
1788         if (mp->m_update_sb) {
1789                 error = xfs_sync_sb(mp, false);
1790                 if (error) {
1791                         xfs_warn(mp, "failed to write sb changes");
1792                         return error;
1793                 }
1794                 mp->m_update_sb = false;
1795         }
1796
1797         /*
1798          * Fill out the reserve pool if it is empty. Use the stashed value if
1799          * it is non-zero, otherwise go with the default.
1800          */
1801         xfs_restore_resvblks(mp);
1802         xfs_log_work_queue(mp);
1803         xfs_blockgc_start(mp);
1804
1805         /* Create the per-AG metadata reservation pool .*/
1806         error = xfs_fs_reserve_ag_blocks(mp);
1807         if (error && error != -ENOSPC)
1808                 return error;
1809
1810         /* Re-enable the background inode inactivation worker. */
1811         xfs_inodegc_start(mp);
1812
1813         return 0;
1814 }
1815
1816 static int
1817 xfs_remount_ro(
1818         struct xfs_mount        *mp)
1819 {
1820         struct xfs_icwalk       icw = {
1821                 .icw_flags      = XFS_ICWALK_FLAG_SYNC,
1822         };
1823         int                     error;
1824
1825         /* Flush all the dirty data to disk. */
1826         error = sync_filesystem(mp->m_super);
1827         if (error)
1828                 return error;
1829
1830         /*
1831          * Cancel background eofb scanning so it cannot race with the final
1832          * log force+buftarg wait and deadlock the remount.
1833          */
1834         xfs_blockgc_stop(mp);
1835
1836         /*
1837          * Clear out all remaining COW staging extents and speculative post-EOF
1838          * preallocations so that we don't leave inodes requiring inactivation
1839          * cleanups during reclaim on a read-only mount.  We must process every
1840          * cached inode, so this requires a synchronous cache scan.
1841          */
1842         error = xfs_blockgc_free_space(mp, &icw);
1843         if (error) {
1844                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1845                 return error;
1846         }
1847
1848         /*
1849          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1850          * flushed all pending inodegc work when it sync'd the filesystem.
1851          * The VFS holds s_umount, so we know that inodes cannot enter
1852          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1853          * we send inodes straight to reclaim, so no inodes will be queued.
1854          */
1855         xfs_inodegc_stop(mp);
1856
1857         /* Free the per-AG metadata reservation pool. */
1858         error = xfs_fs_unreserve_ag_blocks(mp);
1859         if (error) {
1860                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1861                 return error;
1862         }
1863
1864         /*
1865          * Before we sync the metadata, we need to free up the reserve block
1866          * pool so that the used block count in the superblock on disk is
1867          * correct at the end of the remount. Stash the current* reserve pool
1868          * size so that if we get remounted rw, we can return it to the same
1869          * size.
1870          */
1871         xfs_save_resvblks(mp);
1872
1873         xfs_log_clean(mp);
1874         set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1875
1876         return 0;
1877 }
1878
1879 /*
1880  * Logically we would return an error here to prevent users from believing
1881  * they might have changed mount options using remount which can't be changed.
1882  *
1883  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1884  * arguments in some cases so we can't blindly reject options, but have to
1885  * check for each specified option if it actually differs from the currently
1886  * set option and only reject it if that's the case.
1887  *
1888  * Until that is implemented we return success for every remount request, and
1889  * silently ignore all options that we can't actually change.
1890  */
1891 static int
1892 xfs_fs_reconfigure(
1893         struct fs_context *fc)
1894 {
1895         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1896         struct xfs_mount        *new_mp = fc->s_fs_info;
1897         int                     flags = fc->sb_flags;
1898         int                     error;
1899
1900         /* version 5 superblocks always support version counters. */
1901         if (xfs_has_crc(mp))
1902                 fc->sb_flags |= SB_I_VERSION;
1903
1904         error = xfs_fs_validate_params(new_mp);
1905         if (error)
1906                 return error;
1907
1908         /* inode32 -> inode64 */
1909         if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1910                 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1911                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1912         }
1913
1914         /* inode64 -> inode32 */
1915         if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1916                 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1917                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1918         }
1919
1920         /* ro -> rw */
1921         if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1922                 error = xfs_remount_rw(mp);
1923                 if (error)
1924                         return error;
1925         }
1926
1927         /* rw -> ro */
1928         if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1929                 error = xfs_remount_ro(mp);
1930                 if (error)
1931                         return error;
1932         }
1933
1934         return 0;
1935 }
1936
1937 static void xfs_fs_free(
1938         struct fs_context       *fc)
1939 {
1940         struct xfs_mount        *mp = fc->s_fs_info;
1941
1942         /*
1943          * mp is stored in the fs_context when it is initialized.
1944          * mp is transferred to the superblock on a successful mount,
1945          * but if an error occurs before the transfer we have to free
1946          * it here.
1947          */
1948         if (mp)
1949                 xfs_mount_free(mp);
1950 }
1951
1952 static const struct fs_context_operations xfs_context_ops = {
1953         .parse_param = xfs_fs_parse_param,
1954         .get_tree    = xfs_fs_get_tree,
1955         .reconfigure = xfs_fs_reconfigure,
1956         .free        = xfs_fs_free,
1957 };
1958
1959 static int xfs_init_fs_context(
1960         struct fs_context       *fc)
1961 {
1962         struct xfs_mount        *mp;
1963
1964         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1965         if (!mp)
1966                 return -ENOMEM;
1967
1968         spin_lock_init(&mp->m_sb_lock);
1969         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1970         spin_lock_init(&mp->m_perag_lock);
1971         mutex_init(&mp->m_growlock);
1972         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1973         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1974         mp->m_kobj.kobject.kset = xfs_kset;
1975         /*
1976          * We don't create the finobt per-ag space reservation until after log
1977          * recovery, so we must set this to true so that an ifree transaction
1978          * started during log recovery will not depend on space reservations
1979          * for finobt expansion.
1980          */
1981         mp->m_finobt_nores = true;
1982
1983         /*
1984          * These can be overridden by the mount option parsing.
1985          */
1986         mp->m_logbufs = -1;
1987         mp->m_logbsize = -1;
1988         mp->m_allocsize_log = 16; /* 64k */
1989
1990         /*
1991          * Copy binary VFS mount flags we are interested in.
1992          */
1993         if (fc->sb_flags & SB_RDONLY)
1994                 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1995         if (fc->sb_flags & SB_DIRSYNC)
1996                 mp->m_features |= XFS_FEAT_DIRSYNC;
1997         if (fc->sb_flags & SB_SYNCHRONOUS)
1998                 mp->m_features |= XFS_FEAT_WSYNC;
1999
2000         fc->s_fs_info = mp;
2001         fc->ops = &xfs_context_ops;
2002
2003         return 0;
2004 }
2005
2006 static struct file_system_type xfs_fs_type = {
2007         .owner                  = THIS_MODULE,
2008         .name                   = "xfs",
2009         .init_fs_context        = xfs_init_fs_context,
2010         .parameters             = xfs_fs_parameters,
2011         .kill_sb                = kill_block_super,
2012         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
2013 };
2014 MODULE_ALIAS_FS("xfs");
2015
2016 STATIC int __init
2017 xfs_init_caches(void)
2018 {
2019         int             error;
2020
2021         xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
2022                                          SLAB_HWCACHE_ALIGN |
2023                                          SLAB_RECLAIM_ACCOUNT |
2024                                          SLAB_MEM_SPREAD,
2025                                          NULL);
2026         if (!xfs_buf_cache)
2027                 goto out;
2028
2029         xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
2030                                                 sizeof(struct xlog_ticket),
2031                                                 0, 0, NULL);
2032         if (!xfs_log_ticket_cache)
2033                 goto out_destroy_buf_cache;
2034
2035         error = xfs_btree_init_cur_caches();
2036         if (error)
2037                 goto out_destroy_log_ticket_cache;
2038
2039         error = xfs_defer_init_item_caches();
2040         if (error)
2041                 goto out_destroy_btree_cur_cache;
2042
2043         xfs_da_state_cache = kmem_cache_create("xfs_da_state",
2044                                               sizeof(struct xfs_da_state),
2045                                               0, 0, NULL);
2046         if (!xfs_da_state_cache)
2047                 goto out_destroy_defer_item_cache;
2048
2049         xfs_ifork_cache = kmem_cache_create("xfs_ifork",
2050                                            sizeof(struct xfs_ifork),
2051                                            0, 0, NULL);
2052         if (!xfs_ifork_cache)
2053                 goto out_destroy_da_state_cache;
2054
2055         xfs_trans_cache = kmem_cache_create("xfs_trans",
2056                                            sizeof(struct xfs_trans),
2057                                            0, 0, NULL);
2058         if (!xfs_trans_cache)
2059                 goto out_destroy_ifork_cache;
2060
2061
2062         /*
2063          * The size of the cache-allocated buf log item is the maximum
2064          * size possible under XFS.  This wastes a little bit of memory,
2065          * but it is much faster.
2066          */
2067         xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2068                                               sizeof(struct xfs_buf_log_item),
2069                                               0, 0, NULL);
2070         if (!xfs_buf_item_cache)
2071                 goto out_destroy_trans_cache;
2072
2073         xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2074                         xfs_efd_log_item_sizeof(XFS_EFD_MAX_FAST_EXTENTS),
2075                         0, 0, NULL);
2076         if (!xfs_efd_cache)
2077                 goto out_destroy_buf_item_cache;
2078
2079         xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2080                         xfs_efi_log_item_sizeof(XFS_EFI_MAX_FAST_EXTENTS),
2081                         0, 0, NULL);
2082         if (!xfs_efi_cache)
2083                 goto out_destroy_efd_cache;
2084
2085         xfs_inode_cache = kmem_cache_create("xfs_inode",
2086                                            sizeof(struct xfs_inode), 0,
2087                                            (SLAB_HWCACHE_ALIGN |
2088                                             SLAB_RECLAIM_ACCOUNT |
2089                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2090                                            xfs_fs_inode_init_once);
2091         if (!xfs_inode_cache)
2092                 goto out_destroy_efi_cache;
2093
2094         xfs_ili_cache = kmem_cache_create("xfs_ili",
2095                                          sizeof(struct xfs_inode_log_item), 0,
2096                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2097                                          NULL);
2098         if (!xfs_ili_cache)
2099                 goto out_destroy_inode_cache;
2100
2101         xfs_icreate_cache = kmem_cache_create("xfs_icr",
2102                                              sizeof(struct xfs_icreate_item),
2103                                              0, 0, NULL);
2104         if (!xfs_icreate_cache)
2105                 goto out_destroy_ili_cache;
2106
2107         xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2108                                          sizeof(struct xfs_rud_log_item),
2109                                          0, 0, NULL);
2110         if (!xfs_rud_cache)
2111                 goto out_destroy_icreate_cache;
2112
2113         xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2114                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2115                         0, 0, NULL);
2116         if (!xfs_rui_cache)
2117                 goto out_destroy_rud_cache;
2118
2119         xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2120                                          sizeof(struct xfs_cud_log_item),
2121                                          0, 0, NULL);
2122         if (!xfs_cud_cache)
2123                 goto out_destroy_rui_cache;
2124
2125         xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2126                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2127                         0, 0, NULL);
2128         if (!xfs_cui_cache)
2129                 goto out_destroy_cud_cache;
2130
2131         xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2132                                          sizeof(struct xfs_bud_log_item),
2133                                          0, 0, NULL);
2134         if (!xfs_bud_cache)
2135                 goto out_destroy_cui_cache;
2136
2137         xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2138                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2139                         0, 0, NULL);
2140         if (!xfs_bui_cache)
2141                 goto out_destroy_bud_cache;
2142
2143         xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2144                                             sizeof(struct xfs_attrd_log_item),
2145                                             0, 0, NULL);
2146         if (!xfs_attrd_cache)
2147                 goto out_destroy_bui_cache;
2148
2149         xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2150                                             sizeof(struct xfs_attri_log_item),
2151                                             0, 0, NULL);
2152         if (!xfs_attri_cache)
2153                 goto out_destroy_attrd_cache;
2154
2155         xfs_iunlink_cache = kmem_cache_create("xfs_iul_item",
2156                                              sizeof(struct xfs_iunlink_item),
2157                                              0, 0, NULL);
2158         if (!xfs_iunlink_cache)
2159                 goto out_destroy_attri_cache;
2160
2161         return 0;
2162
2163  out_destroy_attri_cache:
2164         kmem_cache_destroy(xfs_attri_cache);
2165  out_destroy_attrd_cache:
2166         kmem_cache_destroy(xfs_attrd_cache);
2167  out_destroy_bui_cache:
2168         kmem_cache_destroy(xfs_bui_cache);
2169  out_destroy_bud_cache:
2170         kmem_cache_destroy(xfs_bud_cache);
2171  out_destroy_cui_cache:
2172         kmem_cache_destroy(xfs_cui_cache);
2173  out_destroy_cud_cache:
2174         kmem_cache_destroy(xfs_cud_cache);
2175  out_destroy_rui_cache:
2176         kmem_cache_destroy(xfs_rui_cache);
2177  out_destroy_rud_cache:
2178         kmem_cache_destroy(xfs_rud_cache);
2179  out_destroy_icreate_cache:
2180         kmem_cache_destroy(xfs_icreate_cache);
2181  out_destroy_ili_cache:
2182         kmem_cache_destroy(xfs_ili_cache);
2183  out_destroy_inode_cache:
2184         kmem_cache_destroy(xfs_inode_cache);
2185  out_destroy_efi_cache:
2186         kmem_cache_destroy(xfs_efi_cache);
2187  out_destroy_efd_cache:
2188         kmem_cache_destroy(xfs_efd_cache);
2189  out_destroy_buf_item_cache:
2190         kmem_cache_destroy(xfs_buf_item_cache);
2191  out_destroy_trans_cache:
2192         kmem_cache_destroy(xfs_trans_cache);
2193  out_destroy_ifork_cache:
2194         kmem_cache_destroy(xfs_ifork_cache);
2195  out_destroy_da_state_cache:
2196         kmem_cache_destroy(xfs_da_state_cache);
2197  out_destroy_defer_item_cache:
2198         xfs_defer_destroy_item_caches();
2199  out_destroy_btree_cur_cache:
2200         xfs_btree_destroy_cur_caches();
2201  out_destroy_log_ticket_cache:
2202         kmem_cache_destroy(xfs_log_ticket_cache);
2203  out_destroy_buf_cache:
2204         kmem_cache_destroy(xfs_buf_cache);
2205  out:
2206         return -ENOMEM;
2207 }
2208
2209 STATIC void
2210 xfs_destroy_caches(void)
2211 {
2212         /*
2213          * Make sure all delayed rcu free are flushed before we
2214          * destroy caches.
2215          */
2216         rcu_barrier();
2217         kmem_cache_destroy(xfs_iunlink_cache);
2218         kmem_cache_destroy(xfs_attri_cache);
2219         kmem_cache_destroy(xfs_attrd_cache);
2220         kmem_cache_destroy(xfs_bui_cache);
2221         kmem_cache_destroy(xfs_bud_cache);
2222         kmem_cache_destroy(xfs_cui_cache);
2223         kmem_cache_destroy(xfs_cud_cache);
2224         kmem_cache_destroy(xfs_rui_cache);
2225         kmem_cache_destroy(xfs_rud_cache);
2226         kmem_cache_destroy(xfs_icreate_cache);
2227         kmem_cache_destroy(xfs_ili_cache);
2228         kmem_cache_destroy(xfs_inode_cache);
2229         kmem_cache_destroy(xfs_efi_cache);
2230         kmem_cache_destroy(xfs_efd_cache);
2231         kmem_cache_destroy(xfs_buf_item_cache);
2232         kmem_cache_destroy(xfs_trans_cache);
2233         kmem_cache_destroy(xfs_ifork_cache);
2234         kmem_cache_destroy(xfs_da_state_cache);
2235         xfs_defer_destroy_item_caches();
2236         xfs_btree_destroy_cur_caches();
2237         kmem_cache_destroy(xfs_log_ticket_cache);
2238         kmem_cache_destroy(xfs_buf_cache);
2239 }
2240
2241 STATIC int __init
2242 xfs_init_workqueues(void)
2243 {
2244         /*
2245          * The allocation workqueue can be used in memory reclaim situations
2246          * (writepage path), and parallelism is only limited by the number of
2247          * AGs in all the filesystems mounted. Hence use the default large
2248          * max_active value for this workqueue.
2249          */
2250         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2251                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2252         if (!xfs_alloc_wq)
2253                 return -ENOMEM;
2254
2255         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2256                         0);
2257         if (!xfs_discard_wq)
2258                 goto out_free_alloc_wq;
2259
2260         return 0;
2261 out_free_alloc_wq:
2262         destroy_workqueue(xfs_alloc_wq);
2263         return -ENOMEM;
2264 }
2265
2266 STATIC void
2267 xfs_destroy_workqueues(void)
2268 {
2269         destroy_workqueue(xfs_discard_wq);
2270         destroy_workqueue(xfs_alloc_wq);
2271 }
2272
2273 #ifdef CONFIG_HOTPLUG_CPU
2274 static int
2275 xfs_cpu_dead(
2276         unsigned int            cpu)
2277 {
2278         struct xfs_mount        *mp, *n;
2279
2280         spin_lock(&xfs_mount_list_lock);
2281         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2282                 spin_unlock(&xfs_mount_list_lock);
2283                 xfs_inodegc_cpu_dead(mp, cpu);
2284                 xlog_cil_pcp_dead(mp->m_log, cpu);
2285                 spin_lock(&xfs_mount_list_lock);
2286         }
2287         spin_unlock(&xfs_mount_list_lock);
2288         return 0;
2289 }
2290
2291 static int __init
2292 xfs_cpu_hotplug_init(void)
2293 {
2294         int     error;
2295
2296         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2297                         xfs_cpu_dead);
2298         if (error < 0)
2299                 xfs_alert(NULL,
2300 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2301                         error);
2302         return error;
2303 }
2304
2305 static void
2306 xfs_cpu_hotplug_destroy(void)
2307 {
2308         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2309 }
2310
2311 #else /* !CONFIG_HOTPLUG_CPU */
2312 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2313 static inline void xfs_cpu_hotplug_destroy(void) {}
2314 #endif
2315
2316 STATIC int __init
2317 init_xfs_fs(void)
2318 {
2319         int                     error;
2320
2321         xfs_check_ondisk_structs();
2322
2323         error = xfs_dahash_test();
2324         if (error)
2325                 return error;
2326
2327         printk(KERN_INFO XFS_VERSION_STRING " with "
2328                          XFS_BUILD_OPTIONS " enabled\n");
2329
2330         xfs_dir_startup();
2331
2332         error = xfs_cpu_hotplug_init();
2333         if (error)
2334                 goto out;
2335
2336         error = xfs_init_caches();
2337         if (error)
2338                 goto out_destroy_hp;
2339
2340         error = xfs_init_workqueues();
2341         if (error)
2342                 goto out_destroy_caches;
2343
2344         error = xfs_mru_cache_init();
2345         if (error)
2346                 goto out_destroy_wq;
2347
2348         error = xfs_init_procfs();
2349         if (error)
2350                 goto out_mru_cache_uninit;
2351
2352         error = xfs_sysctl_register();
2353         if (error)
2354                 goto out_cleanup_procfs;
2355
2356         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2357         if (!xfs_kset) {
2358                 error = -ENOMEM;
2359                 goto out_sysctl_unregister;
2360         }
2361
2362         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2363
2364         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2365         if (!xfsstats.xs_stats) {
2366                 error = -ENOMEM;
2367                 goto out_kset_unregister;
2368         }
2369
2370         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2371                                "stats");
2372         if (error)
2373                 goto out_free_stats;
2374
2375 #ifdef DEBUG
2376         xfs_dbg_kobj.kobject.kset = xfs_kset;
2377         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2378         if (error)
2379                 goto out_remove_stats_kobj;
2380 #endif
2381
2382         error = xfs_qm_init();
2383         if (error)
2384                 goto out_remove_dbg_kobj;
2385
2386         error = register_filesystem(&xfs_fs_type);
2387         if (error)
2388                 goto out_qm_exit;
2389         return 0;
2390
2391  out_qm_exit:
2392         xfs_qm_exit();
2393  out_remove_dbg_kobj:
2394 #ifdef DEBUG
2395         xfs_sysfs_del(&xfs_dbg_kobj);
2396  out_remove_stats_kobj:
2397 #endif
2398         xfs_sysfs_del(&xfsstats.xs_kobj);
2399  out_free_stats:
2400         free_percpu(xfsstats.xs_stats);
2401  out_kset_unregister:
2402         kset_unregister(xfs_kset);
2403  out_sysctl_unregister:
2404         xfs_sysctl_unregister();
2405  out_cleanup_procfs:
2406         xfs_cleanup_procfs();
2407  out_mru_cache_uninit:
2408         xfs_mru_cache_uninit();
2409  out_destroy_wq:
2410         xfs_destroy_workqueues();
2411  out_destroy_caches:
2412         xfs_destroy_caches();
2413  out_destroy_hp:
2414         xfs_cpu_hotplug_destroy();
2415  out:
2416         return error;
2417 }
2418
2419 STATIC void __exit
2420 exit_xfs_fs(void)
2421 {
2422         xfs_qm_exit();
2423         unregister_filesystem(&xfs_fs_type);
2424 #ifdef DEBUG
2425         xfs_sysfs_del(&xfs_dbg_kobj);
2426 #endif
2427         xfs_sysfs_del(&xfsstats.xs_kobj);
2428         free_percpu(xfsstats.xs_stats);
2429         kset_unregister(xfs_kset);
2430         xfs_sysctl_unregister();
2431         xfs_cleanup_procfs();
2432         xfs_mru_cache_uninit();
2433         xfs_destroy_workqueues();
2434         xfs_destroy_caches();
2435         xfs_uuid_table_free();
2436         xfs_cpu_hotplug_destroy();
2437 }
2438
2439 module_init(init_xfs_fs);
2440 module_exit(exit_xfs_fs);
2441
2442 MODULE_AUTHOR("Silicon Graphics, Inc.");
2443 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2444 MODULE_LICENSE("GPL");