fs/xfs/xfs_super.c

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