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 enum xfs_dax_mode {
  51         XFS_DAX_INODE = 0,
  52         XFS_DAX_ALWAYS = 1,
  53         XFS_DAX_NEVER = 2,
  54 };
  55
  56 static void
  57 xfs_mount_set_dax_mode(
  58         struct xfs_mount        *mp,
  59         enum xfs_dax_mode       mode)
  60 {
  61         switch (mode) {
  62         case XFS_DAX_INODE:
  63                 mp->m_flags &= ~(XFS_MOUNT_DAX_ALWAYS | XFS_MOUNT_DAX_NEVER);
  64                 break;
  65         case XFS_DAX_ALWAYS:
  66                 mp->m_flags |= XFS_MOUNT_DAX_ALWAYS;
  67                 mp->m_flags &= ~XFS_MOUNT_DAX_NEVER;
  68                 break;
  69         case XFS_DAX_NEVER:
  70                 mp->m_flags |= XFS_MOUNT_DAX_NEVER;
  71                 mp->m_flags &= ~XFS_MOUNT_DAX_ALWAYS;
  72                 break;
  73         }
  74 }
  75
  76 static const struct constant_table dax_param_enums[] = {
  77         {"inode",       XFS_DAX_INODE },
  78         {"always",      XFS_DAX_ALWAYS },
  79         {"never",       XFS_DAX_NEVER },
  80         {}
  81 };
  82
  83 /*
  84  * Table driven mount option parser.
  85  */
  86 enum {
  87         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
  88         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
  89         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
  90         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
  91         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
  92         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
  93         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
  94         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
  95         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
  96 };
  97
  98 static const struct fs_parameter_spec xfs_fs_parameters[] = {
  99         fsparam_u32("logbufs",          Opt_logbufs),
 100         fsparam_string("logbsize",      Opt_logbsize),
 101         fsparam_string("logdev",        Opt_logdev),
 102         fsparam_string("rtdev",         Opt_rtdev),
 103         fsparam_flag("wsync",           Opt_wsync),
 104         fsparam_flag("noalign",         Opt_noalign),
 105         fsparam_flag("swalloc",         Opt_swalloc),
 106         fsparam_u32("sunit",            Opt_sunit),
 107         fsparam_u32("swidth",           Opt_swidth),
 108         fsparam_flag("nouuid",          Opt_nouuid),
 109         fsparam_flag("grpid",           Opt_grpid),
 110         fsparam_flag("nogrpid",         Opt_nogrpid),
 111         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 112         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 113         fsparam_string("allocsize",     Opt_allocsize),
 114         fsparam_flag("norecovery",      Opt_norecovery),
 115         fsparam_flag("inode64",         Opt_inode64),
 116         fsparam_flag("inode32",         Opt_inode32),
 117         fsparam_flag("ikeep",           Opt_ikeep),
 118         fsparam_flag("noikeep",         Opt_noikeep),
 119         fsparam_flag("largeio",         Opt_largeio),
 120         fsparam_flag("nolargeio",       Opt_nolargeio),
 121         fsparam_flag("attr2",           Opt_attr2),
 122         fsparam_flag("noattr2",         Opt_noattr2),
 123         fsparam_flag("filestreams",     Opt_filestreams),
 124         fsparam_flag("quota",           Opt_quota),
 125         fsparam_flag("noquota",         Opt_noquota),
 126         fsparam_flag("usrquota",        Opt_usrquota),
 127         fsparam_flag("grpquota",        Opt_grpquota),
 128         fsparam_flag("prjquota",        Opt_prjquota),
 129         fsparam_flag("uquota",          Opt_uquota),
 130         fsparam_flag("gquota",          Opt_gquota),
 131         fsparam_flag("pquota",          Opt_pquota),
 132         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 133         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 134         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 135         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 136         fsparam_flag("discard",         Opt_discard),
 137         fsparam_flag("nodiscard",       Opt_nodiscard),
 138         fsparam_flag("dax",             Opt_dax),
 139         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 140         {}
 141 };
 142
 143 struct proc_xfs_info {
 144         uint64_t        flag;
 145         char            *str;
 146 };
 147
 148 static int
 149 xfs_fs_show_options(
 150         struct seq_file         *m,
 151         struct dentry           *root)
 152 {
 153         static struct proc_xfs_info xfs_info_set[] = {
 154                 /* the few simple ones we can get from the mount struct */
 155                 { XFS_MOUNT_IKEEP,              ",ikeep" },
 156                 { XFS_MOUNT_WSYNC,              ",wsync" },
 157                 { XFS_MOUNT_NOALIGN,            ",noalign" },
 158                 { XFS_MOUNT_SWALLOC,            ",swalloc" },
 159                 { XFS_MOUNT_NOUUID,             ",nouuid" },
 160                 { XFS_MOUNT_NORECOVERY,         ",norecovery" },
 161                 { XFS_MOUNT_ATTR2,              ",attr2" },
 162                 { XFS_MOUNT_FILESTREAMS,        ",filestreams" },
 163                 { XFS_MOUNT_GRPID,              ",grpid" },
 164                 { XFS_MOUNT_DISCARD,            ",discard" },
 165                 { XFS_MOUNT_LARGEIO,            ",largeio" },
 166                 { XFS_MOUNT_DAX_ALWAYS,         ",dax=always" },
 167                 { XFS_MOUNT_DAX_NEVER,          ",dax=never" },
 168                 { 0, NULL }
 169         };
 170         struct xfs_mount        *mp = XFS_M(root->d_sb);
 171         struct proc_xfs_info    *xfs_infop;
 172
 173         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 174                 if (mp->m_flags & xfs_infop->flag)
 175                         seq_puts(m, xfs_infop->str);
 176         }
 177
 178         seq_printf(m, ",inode%d",
 179                 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
 180
 181         if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
 182                 seq_printf(m, ",allocsize=%dk",
 183                            (1 << mp->m_allocsize_log) >> 10);
 184
 185         if (mp->m_logbufs > 0)
 186                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 187         if (mp->m_logbsize > 0)
 188                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 189
 190         if (mp->m_logname)
 191                 seq_show_option(m, "logdev", mp->m_logname);
 192         if (mp->m_rtname)
 193                 seq_show_option(m, "rtdev", mp->m_rtname);
 194
 195         if (mp->m_dalign > 0)
 196                 seq_printf(m, ",sunit=%d",
 197                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 198         if (mp->m_swidth > 0)
 199                 seq_printf(m, ",swidth=%d",
 200                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 201
 202         if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
 203                 seq_puts(m, ",usrquota");
 204         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 205                 seq_puts(m, ",uqnoenforce");
 206
 207         if (mp->m_qflags & XFS_PQUOTA_ACCT) {
 208                 if (mp->m_qflags & XFS_PQUOTA_ENFD)
 209                         seq_puts(m, ",prjquota");
 210                 else
 211                         seq_puts(m, ",pqnoenforce");
 212         }
 213         if (mp->m_qflags & XFS_GQUOTA_ACCT) {
 214                 if (mp->m_qflags & XFS_GQUOTA_ENFD)
 215                         seq_puts(m, ",grpquota");
 216                 else
 217                         seq_puts(m, ",gqnoenforce");
 218         }
 219
 220         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 221                 seq_puts(m, ",noquota");
 222
 223         return 0;
 224 }
 225
 226 /*
 227  * Set parameters for inode allocation heuristics, taking into account
 228  * filesystem size and inode32/inode64 mount options; i.e. specifically
 229  * whether or not XFS_MOUNT_SMALL_INUMS is set.
 230  *
 231  * Inode allocation patterns are altered only if inode32 is requested
 232  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
 233  * If altered, XFS_MOUNT_32BITINODES is set as well.
 234  *
 235  * An agcount independent of that in the mount structure is provided
 236  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 237  * to the potentially higher ag count.
 238  *
 239  * Returns the maximum AG index which may contain inodes.
 240  */
 241 xfs_agnumber_t
 242 xfs_set_inode_alloc(
 243         struct xfs_mount *mp,
 244         xfs_agnumber_t  agcount)
 245 {
 246         xfs_agnumber_t  index;
 247         xfs_agnumber_t  maxagi = 0;
 248         xfs_sb_t        *sbp = &mp->m_sb;
 249         xfs_agnumber_t  max_metadata;
 250         xfs_agino_t     agino;
 251         xfs_ino_t       ino;
 252
 253         /*
 254          * Calculate how much should be reserved for inodes to meet
 255          * the max inode percentage.  Used only for inode32.
 256          */
 257         if (M_IGEO(mp)->maxicount) {
 258                 uint64_t        icount;
 259
 260                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 261                 do_div(icount, 100);
 262                 icount += sbp->sb_agblocks - 1;
 263                 do_div(icount, sbp->sb_agblocks);
 264                 max_metadata = icount;
 265         } else {
 266                 max_metadata = agcount;
 267         }
 268
 269         /* Get the last possible inode in the filesystem */
 270         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 271         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 272
 273         /*
 274          * If user asked for no more than 32-bit inodes, and the fs is
 275          * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
 276          * the allocator to accommodate the request.
 277          */
 278         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
 279                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 280         else
 281                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
 282
 283         for (index = 0; index < agcount; index++) {
 284                 struct xfs_perag        *pag;
 285
 286                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 287
 288                 pag = xfs_perag_get(mp, index);
 289
 290                 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
 291                         if (ino > XFS_MAXINUMBER_32) {
 292                                 pag->pagi_inodeok = 0;
 293                                 pag->pagf_metadata = 0;
 294                         } else {
 295                                 pag->pagi_inodeok = 1;
 296                                 maxagi++;
 297                                 if (index < max_metadata)
 298                                         pag->pagf_metadata = 1;
 299                                 else
 300                                         pag->pagf_metadata = 0;
 301                         }
 302                 } else {
 303                         pag->pagi_inodeok = 1;
 304                         pag->pagf_metadata = 0;
 305                 }
 306
 307                 xfs_perag_put(pag);
 308         }
 309
 310         return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
 311 }
 312
 313 STATIC int
 314 xfs_blkdev_get(
 315         xfs_mount_t             *mp,
 316         const char              *name,
 317         struct block_device     **bdevp)
 318 {
 319         int                     error = 0;
 320
 321         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 322                                     mp);
 323         if (IS_ERR(*bdevp)) {
 324                 error = PTR_ERR(*bdevp);
 325                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 326         }
 327
 328         return error;
 329 }
 330
 331 STATIC void
 332 xfs_blkdev_put(
 333         struct block_device     *bdev)
 334 {
 335         if (bdev)
 336                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 337 }
 338
 339 void
 340 xfs_blkdev_issue_flush(
 341         xfs_buftarg_t           *buftarg)
 342 {
 343         blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS);
 344 }
 345
 346 STATIC void
 347 xfs_close_devices(
 348         struct xfs_mount        *mp)
 349 {
 350         struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
 351
 352         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 353                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 354                 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
 355
 356                 xfs_free_buftarg(mp->m_logdev_targp);
 357                 xfs_blkdev_put(logdev);
 358                 fs_put_dax(dax_logdev);
 359         }
 360         if (mp->m_rtdev_targp) {
 361                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 362                 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
 363
 364                 xfs_free_buftarg(mp->m_rtdev_targp);
 365                 xfs_blkdev_put(rtdev);
 366                 fs_put_dax(dax_rtdev);
 367         }
 368         xfs_free_buftarg(mp->m_ddev_targp);
 369         fs_put_dax(dax_ddev);
 370 }
 371
 372 /*
 373  * The file system configurations are:
 374  *      (1) device (partition) with data and internal log
 375  *      (2) logical volume with data and log subvolumes.
 376  *      (3) logical volume with data, log, and realtime subvolumes.
 377  *
 378  * We only have to handle opening the log and realtime volumes here if
 379  * they are present.  The data subvolume has already been opened by
 380  * get_sb_bdev() and is stored in sb->s_bdev.
 381  */
 382 STATIC int
 383 xfs_open_devices(
 384         struct xfs_mount        *mp)
 385 {
 386         struct block_device     *ddev = mp->m_super->s_bdev;
 387         struct dax_device       *dax_ddev = fs_dax_get_by_bdev(ddev);
 388         struct dax_device       *dax_logdev = NULL, *dax_rtdev = NULL;
 389         struct block_device     *logdev = NULL, *rtdev = NULL;
 390         int                     error;
 391
 392         /*
 393          * Open real time and log devices - order is important.
 394          */
 395         if (mp->m_logname) {
 396                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 397                 if (error)
 398                         goto out;
 399                 dax_logdev = fs_dax_get_by_bdev(logdev);
 400         }
 401
 402         if (mp->m_rtname) {
 403                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 404                 if (error)
 405                         goto out_close_logdev;
 406
 407                 if (rtdev == ddev || rtdev == logdev) {
 408                         xfs_warn(mp,
 409         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 410                         error = -EINVAL;
 411                         goto out_close_rtdev;
 412                 }
 413                 dax_rtdev = fs_dax_get_by_bdev(rtdev);
 414         }
 415
 416         /*
 417          * Setup xfs_mount buffer target pointers
 418          */
 419         error = -ENOMEM;
 420         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
 421         if (!mp->m_ddev_targp)
 422                 goto out_close_rtdev;
 423
 424         if (rtdev) {
 425                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
 426                 if (!mp->m_rtdev_targp)
 427                         goto out_free_ddev_targ;
 428         }
 429
 430         if (logdev && logdev != ddev) {
 431                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
 432                 if (!mp->m_logdev_targp)
 433                         goto out_free_rtdev_targ;
 434         } else {
 435                 mp->m_logdev_targp = mp->m_ddev_targp;
 436         }
 437
 438         return 0;
 439
 440  out_free_rtdev_targ:
 441         if (mp->m_rtdev_targp)
 442                 xfs_free_buftarg(mp->m_rtdev_targp);
 443  out_free_ddev_targ:
 444         xfs_free_buftarg(mp->m_ddev_targp);
 445  out_close_rtdev:
 446         xfs_blkdev_put(rtdev);
 447         fs_put_dax(dax_rtdev);
 448  out_close_logdev:
 449         if (logdev && logdev != ddev) {
 450                 xfs_blkdev_put(logdev);
 451                 fs_put_dax(dax_logdev);
 452         }
 453  out:
 454         fs_put_dax(dax_ddev);
 455         return error;
 456 }
 457
 458 /*
 459  * Setup xfs_mount buffer target pointers based on superblock
 460  */
 461 STATIC int
 462 xfs_setup_devices(
 463         struct xfs_mount        *mp)
 464 {
 465         int                     error;
 466
 467         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 468         if (error)
 469                 return error;
 470
 471         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 472                 unsigned int    log_sector_size = BBSIZE;
 473
 474                 if (xfs_sb_version_hassector(&mp->m_sb))
 475                         log_sector_size = mp->m_sb.sb_logsectsize;
 476                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 477                                             log_sector_size);
 478                 if (error)
 479                         return error;
 480         }
 481         if (mp->m_rtdev_targp) {
 482                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 483                                             mp->m_sb.sb_sectsize);
 484                 if (error)
 485                         return error;
 486         }
 487
 488         return 0;
 489 }
 490
 491 STATIC int
 492 xfs_init_mount_workqueues(
 493         struct xfs_mount        *mp)
 494 {
 495         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 496                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_super->s_id);
 497         if (!mp->m_buf_workqueue)
 498                 goto out;
 499
 500         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 501                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
 502         if (!mp->m_unwritten_workqueue)
 503                 goto out_destroy_buf;
 504
 505         mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
 506                         WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND,
 507                         0, mp->m_super->s_id);
 508         if (!mp->m_cil_workqueue)
 509                 goto out_destroy_unwritten;
 510
 511         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 512                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
 513         if (!mp->m_reclaim_workqueue)
 514                 goto out_destroy_cil;
 515
 516         mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
 517                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
 518         if (!mp->m_eofblocks_workqueue)
 519                 goto out_destroy_reclaim;
 520
 521         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
 522                                                mp->m_super->s_id);
 523         if (!mp->m_sync_workqueue)
 524                 goto out_destroy_eofb;
 525
 526         return 0;
 527
 528 out_destroy_eofb:
 529         destroy_workqueue(mp->m_eofblocks_workqueue);
 530 out_destroy_reclaim:
 531         destroy_workqueue(mp->m_reclaim_workqueue);
 532 out_destroy_cil:
 533         destroy_workqueue(mp->m_cil_workqueue);
 534 out_destroy_unwritten:
 535         destroy_workqueue(mp->m_unwritten_workqueue);
 536 out_destroy_buf:
 537         destroy_workqueue(mp->m_buf_workqueue);
 538 out:
 539         return -ENOMEM;
 540 }
 541
 542 STATIC void
 543 xfs_destroy_mount_workqueues(
 544         struct xfs_mount        *mp)
 545 {
 546         destroy_workqueue(mp->m_sync_workqueue);
 547         destroy_workqueue(mp->m_eofblocks_workqueue);
 548         destroy_workqueue(mp->m_reclaim_workqueue);
 549         destroy_workqueue(mp->m_cil_workqueue);
 550         destroy_workqueue(mp->m_unwritten_workqueue);
 551         destroy_workqueue(mp->m_buf_workqueue);
 552 }
 553
 554 static void
 555 xfs_flush_inodes_worker(
 556         struct work_struct      *work)
 557 {
 558         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 559                                                    m_flush_inodes_work);
 560         struct super_block      *sb = mp->m_super;
 561
 562         if (down_read_trylock(&sb->s_umount)) {
 563                 sync_inodes_sb(sb);
 564                 up_read(&sb->s_umount);
 565         }
 566 }
 567
 568 /*
 569  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 570  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 571  * for IO to complete so that we effectively throttle multiple callers to the
 572  * rate at which IO is completing.
 573  */
 574 void
 575 xfs_flush_inodes(
 576         struct xfs_mount        *mp)
 577 {
 578         /*
 579          * If flush_work() returns true then that means we waited for a flush
 580          * which was already in progress.  Don't bother running another scan.
 581          */
 582         if (flush_work(&mp->m_flush_inodes_work))
 583                 return;
 584
 585         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 586         flush_work(&mp->m_flush_inodes_work);
 587 }
 588
 589 /* Catch misguided souls that try to use this interface on XFS */
 590 STATIC struct inode *
 591 xfs_fs_alloc_inode(
 592         struct super_block      *sb)
 593 {
 594         BUG();
 595         return NULL;
 596 }
 597
 598 #ifdef DEBUG
 599 static void
 600 xfs_check_delalloc(
 601         struct xfs_inode        *ip,
 602         int                     whichfork)
 603 {
 604         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 605         struct xfs_bmbt_irec    got;
 606         struct xfs_iext_cursor  icur;
 607
 608         if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
 609                 return;
 610         do {
 611                 if (isnullstartblock(got.br_startblock)) {
 612                         xfs_warn(ip->i_mount,
 613         "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
 614                                 ip->i_ino,
 615                                 whichfork == XFS_DATA_FORK ? "data" : "cow",
 616                                 got.br_startoff, got.br_blockcount);
 617                 }
 618         } while (xfs_iext_next_extent(ifp, &icur, &got));
 619 }
 620 #else
 621 #define xfs_check_delalloc(ip, whichfork)       do { } while (0)
 622 #endif
 623
 624 /*
 625  * Now that the generic code is guaranteed not to be accessing
 626  * the linux inode, we can inactivate and reclaim the inode.
 627  */
 628 STATIC void
 629 xfs_fs_destroy_inode(
 630         struct inode            *inode)
 631 {
 632         struct xfs_inode        *ip = XFS_I(inode);
 633
 634         trace_xfs_destroy_inode(ip);
 635
 636         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 637         XFS_STATS_INC(ip->i_mount, vn_rele);
 638         XFS_STATS_INC(ip->i_mount, vn_remove);
 639
 640         xfs_inactive(ip);
 641
 642         if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
 643                 xfs_check_delalloc(ip, XFS_DATA_FORK);
 644                 xfs_check_delalloc(ip, XFS_COW_FORK);
 645                 ASSERT(0);
 646         }
 647
 648         XFS_STATS_INC(ip->i_mount, vn_reclaim);
 649
 650         /*
 651          * We should never get here with one of the reclaim flags already set.
 652          */
 653         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
 654         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
 655
 656         /*
 657          * We always use background reclaim here because even if the
 658          * inode is clean, it still may be under IO and hence we have
 659          * to take the flush lock. The background reclaim path handles
 660          * this more efficiently than we can here, so simply let background
 661          * reclaim tear down all inodes.
 662          */
 663         xfs_inode_set_reclaim_tag(ip);
 664 }
 665
 666 static void
 667 xfs_fs_dirty_inode(
 668         struct inode                    *inode,
 669         int                             flag)
 670 {
 671         struct xfs_inode                *ip = XFS_I(inode);
 672         struct xfs_mount                *mp = ip->i_mount;
 673         struct xfs_trans                *tp;
 674
 675         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 676                 return;
 677         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 678                 return;
 679
 680         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 681                 return;
 682         xfs_ilock(ip, XFS_ILOCK_EXCL);
 683         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 684         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 685         xfs_trans_commit(tp);
 686 }
 687
 688 /*
 689  * Slab object creation initialisation for the XFS inode.
 690  * This covers only the idempotent fields in the XFS inode;
 691  * all other fields need to be initialised on allocation
 692  * from the slab. This avoids the need to repeatedly initialise
 693  * fields in the xfs inode that left in the initialise state
 694  * when freeing the inode.
 695  */
 696 STATIC void
 697 xfs_fs_inode_init_once(
 698         void                    *inode)
 699 {
 700         struct xfs_inode        *ip = inode;
 701
 702         memset(ip, 0, sizeof(struct xfs_inode));
 703
 704         /* vfs inode */
 705         inode_init_once(VFS_I(ip));
 706
 707         /* xfs inode */
 708         atomic_set(&ip->i_pincount, 0);
 709         spin_lock_init(&ip->i_flags_lock);
 710
 711         mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 712                      "xfsino", ip->i_ino);
 713         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 714                      "xfsino", ip->i_ino);
 715 }
 716
 717 /*
 718  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 719  * serialised against cache hits here via the inode->i_lock and igrab() in
 720  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 721  * racing with us, and it avoids needing to grab a spinlock here for every inode
 722  * we drop the final reference on.
 723  */
 724 STATIC int
 725 xfs_fs_drop_inode(
 726         struct inode            *inode)
 727 {
 728         struct xfs_inode        *ip = XFS_I(inode);
 729
 730         /*
 731          * If this unlinked inode is in the middle of recovery, don't
 732          * drop the inode just yet; log recovery will take care of
 733          * that.  See the comment for this inode flag.
 734          */
 735         if (ip->i_flags & XFS_IRECOVERY) {
 736                 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
 737                 return 0;
 738         }
 739
 740         return generic_drop_inode(inode);
 741 }
 742
 743 static void
 744 xfs_mount_free(
 745         struct xfs_mount        *mp)
 746 {
 747         kfree(mp->m_rtname);
 748         kfree(mp->m_logname);
 749         kmem_free(mp);
 750 }
 751
 752 STATIC int
 753 xfs_fs_sync_fs(
 754         struct super_block      *sb,
 755         int                     wait)
 756 {
 757         struct xfs_mount        *mp = XFS_M(sb);
 758
 759         /*
 760          * Doing anything during the async pass would be counterproductive.
 761          */
 762         if (!wait)
 763                 return 0;
 764
 765         xfs_log_force(mp, XFS_LOG_SYNC);
 766         if (laptop_mode) {
 767                 /*
 768                  * The disk must be active because we're syncing.
 769                  * We schedule log work now (now that the disk is
 770                  * active) instead of later (when it might not be).
 771                  */
 772                 flush_delayed_work(&mp->m_log->l_work);
 773         }
 774
 775         return 0;
 776 }
 777
 778 STATIC int
 779 xfs_fs_statfs(
 780         struct dentry           *dentry,
 781         struct kstatfs          *statp)
 782 {
 783         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 784         xfs_sb_t                *sbp = &mp->m_sb;
 785         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 786         uint64_t                fakeinos, id;
 787         uint64_t                icount;
 788         uint64_t                ifree;
 789         uint64_t                fdblocks;
 790         xfs_extlen_t            lsize;
 791         int64_t                 ffree;
 792
 793         statp->f_type = XFS_SUPER_MAGIC;
 794         statp->f_namelen = MAXNAMELEN - 1;
 795
 796         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 797         statp->f_fsid.val[0] = (u32)id;
 798         statp->f_fsid.val[1] = (u32)(id >> 32);
 799
 800         icount = percpu_counter_sum(&mp->m_icount);
 801         ifree = percpu_counter_sum(&mp->m_ifree);
 802         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 803
 804         spin_lock(&mp->m_sb_lock);
 805         statp->f_bsize = sbp->sb_blocksize;
 806         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 807         statp->f_blocks = sbp->sb_dblocks - lsize;
 808         spin_unlock(&mp->m_sb_lock);
 809
 810         /* make sure statp->f_bfree does not underflow */
 811         statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
 812         statp->f_bavail = statp->f_bfree;
 813
 814         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 815         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 816         if (M_IGEO(mp)->maxicount)
 817                 statp->f_files = min_t(typeof(statp->f_files),
 818                                         statp->f_files,
 819                                         M_IGEO(mp)->maxicount);
 820
 821         /* If sb_icount overshot maxicount, report actual allocation */
 822         statp->f_files = max_t(typeof(statp->f_files),
 823                                         statp->f_files,
 824                                         sbp->sb_icount);
 825
 826         /* make sure statp->f_ffree does not underflow */
 827         ffree = statp->f_files - (icount - ifree);
 828         statp->f_ffree = max_t(int64_t, ffree, 0);
 829
 830
 831         if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
 832             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 833                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 834                 xfs_qm_statvfs(ip, statp);
 835
 836         if (XFS_IS_REALTIME_MOUNT(mp) &&
 837             (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 838                 statp->f_blocks = sbp->sb_rblocks;
 839                 statp->f_bavail = statp->f_bfree =
 840                         sbp->sb_frextents * sbp->sb_rextsize;
 841         }
 842
 843         return 0;
 844 }
 845
 846 STATIC void
 847 xfs_save_resvblks(struct xfs_mount *mp)
 848 {
 849         uint64_t resblks = 0;
 850
 851         mp->m_resblks_save = mp->m_resblks;
 852         xfs_reserve_blocks(mp, &resblks, NULL);
 853 }
 854
 855 STATIC void
 856 xfs_restore_resvblks(struct xfs_mount *mp)
 857 {
 858         uint64_t resblks;
 859
 860         if (mp->m_resblks_save) {
 861                 resblks = mp->m_resblks_save;
 862                 mp->m_resblks_save = 0;
 863         } else
 864                 resblks = xfs_default_resblks(mp);
 865
 866         xfs_reserve_blocks(mp, &resblks, NULL);
 867 }
 868
 869 /*
 870  * Trigger writeback of all the dirty metadata in the file system.
 871  *
 872  * This ensures that the metadata is written to their location on disk rather
 873  * than just existing in transactions in the log. This means after a quiesce
 874  * there is no log replay required to write the inodes to disk - this is the
 875  * primary difference between a sync and a quiesce.
 876  *
 877  * We cancel log work early here to ensure all transactions the log worker may
 878  * run have finished before we clean up and log the superblock and write an
 879  * unmount record. The unfreeze process is responsible for restarting the log
 880  * worker correctly.
 881  */
 882 void
 883 xfs_quiesce_attr(
 884         struct xfs_mount        *mp)
 885 {
 886         int     error = 0;
 887
 888         cancel_delayed_work_sync(&mp->m_log->l_work);
 889
 890         /* force the log to unpin objects from the now complete transactions */
 891         xfs_log_force(mp, XFS_LOG_SYNC);
 892
 893         /* reclaim inodes to do any IO before the freeze completes */
 894         xfs_reclaim_inodes(mp, 0);
 895         xfs_reclaim_inodes(mp, SYNC_WAIT);
 896
 897         /* Push the superblock and write an unmount record */
 898         error = xfs_log_sbcount(mp);
 899         if (error)
 900                 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
 901                                 "Frozen image may not be consistent.");
 902         xfs_log_quiesce(mp);
 903 }
 904
 905 /*
 906  * Second stage of a freeze. The data is already frozen so we only
 907  * need to take care of the metadata. Once that's done sync the superblock
 908  * to the log to dirty it in case of a crash while frozen. This ensures that we
 909  * will recover the unlinked inode lists on the next mount.
 910  */
 911 STATIC int
 912 xfs_fs_freeze(
 913         struct super_block      *sb)
 914 {
 915         struct xfs_mount        *mp = XFS_M(sb);
 916
 917         xfs_stop_block_reaping(mp);
 918         xfs_save_resvblks(mp);
 919         xfs_quiesce_attr(mp);
 920         return xfs_sync_sb(mp, true);
 921 }
 922
 923 STATIC int
 924 xfs_fs_unfreeze(
 925         struct super_block      *sb)
 926 {
 927         struct xfs_mount        *mp = XFS_M(sb);
 928
 929         xfs_restore_resvblks(mp);
 930         xfs_log_work_queue(mp);
 931         xfs_start_block_reaping(mp);
 932         return 0;
 933 }
 934
 935 /*
 936  * This function fills in xfs_mount_t fields based on mount args.
 937  * Note: the superblock _has_ now been read in.
 938  */
 939 STATIC int
 940 xfs_finish_flags(
 941         struct xfs_mount        *mp)
 942 {
 943         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
 944
 945         /* Fail a mount where the logbuf is smaller than the log stripe */
 946         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
 947                 if (mp->m_logbsize <= 0 &&
 948                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 949                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 950                 } else if (mp->m_logbsize > 0 &&
 951                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 952                         xfs_warn(mp,
 953                 "logbuf size must be greater than or equal to log stripe size");
 954                         return -EINVAL;
 955                 }
 956         } else {
 957                 /* Fail a mount if the logbuf is larger than 32K */
 958                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 959                         xfs_warn(mp,
 960                 "logbuf size for version 1 logs must be 16K or 32K");
 961                         return -EINVAL;
 962                 }
 963         }
 964
 965         /*
 966          * V5 filesystems always use attr2 format for attributes.
 967          */
 968         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 969             (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 970                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 971                              "attr2 is always enabled for V5 filesystems.");
 972                 return -EINVAL;
 973         }
 974
 975         /*
 976          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
 977          * told by noattr2 to turn it off
 978          */
 979         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 980             !(mp->m_flags & XFS_MOUNT_NOATTR2))
 981                 mp->m_flags |= XFS_MOUNT_ATTR2;
 982
 983         /*
 984          * prohibit r/w mounts of read-only filesystems
 985          */
 986         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 987                 xfs_warn(mp,
 988                         "cannot mount a read-only filesystem as read-write");
 989                 return -EROFS;
 990         }
 991
 992         if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
 993             (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
 994             !xfs_sb_version_has_pquotino(&mp->m_sb)) {
 995                 xfs_warn(mp,
 996                   "Super block does not support project and group quota together");
 997                 return -EINVAL;
 998         }
 999
1000         return 0;
1001 }
1002
1003 static int
1004 xfs_init_percpu_counters(
1005         struct xfs_mount        *mp)
1006 {
1007         int             error;
1008
1009         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1010         if (error)
1011                 return -ENOMEM;
1012
1013         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1014         if (error)
1015                 goto free_icount;
1016
1017         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1018         if (error)
1019                 goto free_ifree;
1020
1021         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1022         if (error)
1023                 goto free_fdblocks;
1024
1025         return 0;
1026
1027 free_fdblocks:
1028         percpu_counter_destroy(&mp->m_fdblocks);
1029 free_ifree:
1030         percpu_counter_destroy(&mp->m_ifree);
1031 free_icount:
1032         percpu_counter_destroy(&mp->m_icount);
1033         return -ENOMEM;
1034 }
1035
1036 void
1037 xfs_reinit_percpu_counters(
1038         struct xfs_mount        *mp)
1039 {
1040         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1041         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1042         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1043 }
1044
1045 static void
1046 xfs_destroy_percpu_counters(
1047         struct xfs_mount        *mp)
1048 {
1049         percpu_counter_destroy(&mp->m_icount);
1050         percpu_counter_destroy(&mp->m_ifree);
1051         percpu_counter_destroy(&mp->m_fdblocks);
1052         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1053                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1054         percpu_counter_destroy(&mp->m_delalloc_blks);
1055 }
1056
1057 static void
1058 xfs_fs_put_super(
1059         struct super_block      *sb)
1060 {
1061         struct xfs_mount        *mp = XFS_M(sb);
1062
1063         /* if ->fill_super failed, we have no mount to tear down */
1064         if (!sb->s_fs_info)
1065                 return;
1066
1067         xfs_notice(mp, "Unmounting Filesystem");
1068         xfs_filestream_unmount(mp);
1069         xfs_unmountfs(mp);
1070
1071         xfs_freesb(mp);
1072         free_percpu(mp->m_stats.xs_stats);
1073         xfs_destroy_percpu_counters(mp);
1074         xfs_destroy_mount_workqueues(mp);
1075         xfs_close_devices(mp);
1076
1077         sb->s_fs_info = NULL;
1078         xfs_mount_free(mp);
1079 }
1080
1081 static long
1082 xfs_fs_nr_cached_objects(
1083         struct super_block      *sb,
1084         struct shrink_control   *sc)
1085 {
1086         /* Paranoia: catch incorrect calls during mount setup or teardown */
1087         if (WARN_ON_ONCE(!sb->s_fs_info))
1088                 return 0;
1089         return xfs_reclaim_inodes_count(XFS_M(sb));
1090 }
1091
1092 static long
1093 xfs_fs_free_cached_objects(
1094         struct super_block      *sb,
1095         struct shrink_control   *sc)
1096 {
1097         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1098 }
1099
1100 static const struct super_operations xfs_super_operations = {
1101         .alloc_inode            = xfs_fs_alloc_inode,
1102         .destroy_inode          = xfs_fs_destroy_inode,
1103         .dirty_inode            = xfs_fs_dirty_inode,
1104         .drop_inode             = xfs_fs_drop_inode,
1105         .put_super              = xfs_fs_put_super,
1106         .sync_fs                = xfs_fs_sync_fs,
1107         .freeze_fs              = xfs_fs_freeze,
1108         .unfreeze_fs            = xfs_fs_unfreeze,
1109         .statfs                 = xfs_fs_statfs,
1110         .show_options           = xfs_fs_show_options,
1111         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1112         .free_cached_objects    = xfs_fs_free_cached_objects,
1113 };
1114
1115 static int
1116 suffix_kstrtoint(
1117         const char      *s,
1118         unsigned int    base,
1119         int             *res)
1120 {
1121         int             last, shift_left_factor = 0, _res;
1122         char            *value;
1123         int             ret = 0;
1124
1125         value = kstrdup(s, GFP_KERNEL);
1126         if (!value)
1127                 return -ENOMEM;
1128
1129         last = strlen(value) - 1;
1130         if (value[last] == 'K' || value[last] == 'k') {
1131                 shift_left_factor = 10;
1132                 value[last] = '\0';
1133         }
1134         if (value[last] == 'M' || value[last] == 'm') {
1135                 shift_left_factor = 20;
1136                 value[last] = '\0';
1137         }
1138         if (value[last] == 'G' || value[last] == 'g') {
1139                 shift_left_factor = 30;
1140                 value[last] = '\0';
1141         }
1142
1143         if (kstrtoint(value, base, &_res))
1144                 ret = -EINVAL;
1145         kfree(value);
1146         *res = _res << shift_left_factor;
1147         return ret;
1148 }
1149
1150 /*
1151  * Set mount state from a mount option.
1152  *
1153  * NOTE: mp->m_super is NULL here!
1154  */
1155 static int
1156 xfs_fc_parse_param(
1157         struct fs_context       *fc,
1158         struct fs_parameter     *param)
1159 {
1160         struct xfs_mount        *mp = fc->s_fs_info;
1161         struct fs_parse_result  result;
1162         int                     size = 0;
1163         int                     opt;
1164
1165         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1166         if (opt < 0)
1167                 return opt;
1168
1169         switch (opt) {
1170         case Opt_logbufs:
1171                 mp->m_logbufs = result.uint_32;
1172                 return 0;
1173         case Opt_logbsize:
1174                 if (suffix_kstrtoint(param->string, 10, &mp->m_logbsize))
1175                         return -EINVAL;
1176                 return 0;
1177         case Opt_logdev:
1178                 kfree(mp->m_logname);
1179                 mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1180                 if (!mp->m_logname)
1181                         return -ENOMEM;
1182                 return 0;
1183         case Opt_rtdev:
1184                 kfree(mp->m_rtname);
1185                 mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1186                 if (!mp->m_rtname)
1187                         return -ENOMEM;
1188                 return 0;
1189         case Opt_allocsize:
1190                 if (suffix_kstrtoint(param->string, 10, &size))
1191                         return -EINVAL;
1192                 mp->m_allocsize_log = ffs(size) - 1;
1193                 mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1194                 return 0;
1195         case Opt_grpid:
1196         case Opt_bsdgroups:
1197                 mp->m_flags |= XFS_MOUNT_GRPID;
1198                 return 0;
1199         case Opt_nogrpid:
1200         case Opt_sysvgroups:
1201                 mp->m_flags &= ~XFS_MOUNT_GRPID;
1202                 return 0;
1203         case Opt_wsync:
1204                 mp->m_flags |= XFS_MOUNT_WSYNC;
1205                 return 0;
1206         case Opt_norecovery:
1207                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
1208                 return 0;
1209         case Opt_noalign:
1210                 mp->m_flags |= XFS_MOUNT_NOALIGN;
1211                 return 0;
1212         case Opt_swalloc:
1213                 mp->m_flags |= XFS_MOUNT_SWALLOC;
1214                 return 0;
1215         case Opt_sunit:
1216                 mp->m_dalign = result.uint_32;
1217                 return 0;
1218         case Opt_swidth:
1219                 mp->m_swidth = result.uint_32;
1220                 return 0;
1221         case Opt_inode32:
1222                 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1223                 return 0;
1224         case Opt_inode64:
1225                 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1226                 return 0;
1227         case Opt_nouuid:
1228                 mp->m_flags |= XFS_MOUNT_NOUUID;
1229                 return 0;
1230         case Opt_ikeep:
1231                 mp->m_flags |= XFS_MOUNT_IKEEP;
1232                 return 0;
1233         case Opt_noikeep:
1234                 mp->m_flags &= ~XFS_MOUNT_IKEEP;
1235                 return 0;
1236         case Opt_largeio:
1237                 mp->m_flags |= XFS_MOUNT_LARGEIO;
1238                 return 0;
1239         case Opt_nolargeio:
1240                 mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1241                 return 0;
1242         case Opt_attr2:
1243                 mp->m_flags |= XFS_MOUNT_ATTR2;
1244                 return 0;
1245         case Opt_noattr2:
1246                 mp->m_flags &= ~XFS_MOUNT_ATTR2;
1247                 mp->m_flags |= XFS_MOUNT_NOATTR2;
1248                 return 0;
1249         case Opt_filestreams:
1250                 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1251                 return 0;
1252         case Opt_noquota:
1253                 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1254                 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1255                 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1256                 return 0;
1257         case Opt_quota:
1258         case Opt_uquota:
1259         case Opt_usrquota:
1260                 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1261                                  XFS_UQUOTA_ENFD);
1262                 return 0;
1263         case Opt_qnoenforce:
1264         case Opt_uqnoenforce:
1265                 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1266                 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1267                 return 0;
1268         case Opt_pquota:
1269         case Opt_prjquota:
1270                 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1271                                  XFS_PQUOTA_ENFD);
1272                 return 0;
1273         case Opt_pqnoenforce:
1274                 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1275                 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1276                 return 0;
1277         case Opt_gquota:
1278         case Opt_grpquota:
1279                 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1280                                  XFS_GQUOTA_ENFD);
1281                 return 0;
1282         case Opt_gqnoenforce:
1283                 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1284                 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1285                 return 0;
1286         case Opt_discard:
1287                 mp->m_flags |= XFS_MOUNT_DISCARD;
1288                 return 0;
1289         case Opt_nodiscard:
1290                 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1291                 return 0;
1292 #ifdef CONFIG_FS_DAX
1293         case Opt_dax:
1294                 xfs_mount_set_dax_mode(mp, XFS_DAX_ALWAYS);
1295                 return 0;
1296         case Opt_dax_enum:
1297                 xfs_mount_set_dax_mode(mp, result.uint_32);
1298                 return 0;
1299 #endif
1300         default:
1301                 xfs_warn(mp, "unknown mount option [%s].", param->key);
1302                 return -EINVAL;
1303         }
1304
1305         return 0;
1306 }
1307
1308 static int
1309 xfs_fc_validate_params(
1310         struct xfs_mount        *mp)
1311 {
1312         /*
1313          * no recovery flag requires a read-only mount
1314          */
1315         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1316             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1317                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1318                 return -EINVAL;
1319         }
1320
1321         if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1322             (mp->m_dalign || mp->m_swidth)) {
1323                 xfs_warn(mp,
1324         "sunit and swidth options incompatible with the noalign option");
1325                 return -EINVAL;
1326         }
1327
1328         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1329                 xfs_warn(mp, "quota support not available in this kernel.");
1330                 return -EINVAL;
1331         }
1332
1333         if ((mp->m_dalign && !mp->m_swidth) ||
1334             (!mp->m_dalign && mp->m_swidth)) {
1335                 xfs_warn(mp, "sunit and swidth must be specified together");
1336                 return -EINVAL;
1337         }
1338
1339         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1340                 xfs_warn(mp,
1341         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1342                         mp->m_swidth, mp->m_dalign);
1343                 return -EINVAL;
1344         }
1345
1346         if (mp->m_logbufs != -1 &&
1347             mp->m_logbufs != 0 &&
1348             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1349              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1350                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1351                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1352                 return -EINVAL;
1353         }
1354
1355         if (mp->m_logbsize != -1 &&
1356             mp->m_logbsize !=  0 &&
1357             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1358              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1359              !is_power_of_2(mp->m_logbsize))) {
1360                 xfs_warn(mp,
1361                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1362                         mp->m_logbsize);
1363                 return -EINVAL;
1364         }
1365
1366         if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1367             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1368              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1369                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1370                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1371                 return -EINVAL;
1372         }
1373
1374         return 0;
1375 }
1376
1377 static int
1378 xfs_fc_fill_super(
1379         struct super_block      *sb,
1380         struct fs_context       *fc)
1381 {
1382         struct xfs_mount        *mp = sb->s_fs_info;
1383         struct inode            *root;
1384         int                     flags = 0, error;
1385
1386         mp->m_super = sb;
1387
1388         error = xfs_fc_validate_params(mp);
1389         if (error)
1390                 goto out_free_names;
1391
1392         sb_min_blocksize(sb, BBSIZE);
1393         sb->s_xattr = xfs_xattr_handlers;
1394         sb->s_export_op = &xfs_export_operations;
1395 #ifdef CONFIG_XFS_QUOTA
1396         sb->s_qcop = &xfs_quotactl_operations;
1397         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1398 #endif
1399         sb->s_op = &xfs_super_operations;
1400
1401         /*
1402          * Delay mount work if the debug hook is set. This is debug
1403          * instrumention to coordinate simulation of xfs mount failures with
1404          * VFS superblock operations
1405          */
1406         if (xfs_globals.mount_delay) {
1407                 xfs_notice(mp, "Delaying mount for %d seconds.",
1408                         xfs_globals.mount_delay);
1409                 msleep(xfs_globals.mount_delay * 1000);
1410         }
1411
1412         if (fc->sb_flags & SB_SILENT)
1413                 flags |= XFS_MFSI_QUIET;
1414
1415         error = xfs_open_devices(mp);
1416         if (error)
1417                 goto out_free_names;
1418
1419         error = xfs_init_mount_workqueues(mp);
1420         if (error)
1421                 goto out_close_devices;
1422
1423         error = xfs_init_percpu_counters(mp);
1424         if (error)
1425                 goto out_destroy_workqueues;
1426
1427         /* Allocate stats memory before we do operations that might use it */
1428         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1429         if (!mp->m_stats.xs_stats) {
1430                 error = -ENOMEM;
1431                 goto out_destroy_counters;
1432         }
1433
1434         error = xfs_readsb(mp, flags);
1435         if (error)
1436                 goto out_free_stats;
1437
1438         error = xfs_finish_flags(mp);
1439         if (error)
1440                 goto out_free_sb;
1441
1442         error = xfs_setup_devices(mp);
1443         if (error)
1444                 goto out_free_sb;
1445
1446         /*
1447          * XFS block mappings use 54 bits to store the logical block offset.
1448          * This should suffice to handle the maximum file size that the VFS
1449          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1450          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1451          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1452          * to check this assertion.
1453          *
1454          * Avoid integer overflow by comparing the maximum bmbt offset to the
1455          * maximum pagecache offset in units of fs blocks.
1456          */
1457         if (XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE) > XFS_MAX_FILEOFF) {
1458                 xfs_warn(mp,
1459 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1460                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1461                          XFS_MAX_FILEOFF);
1462                 error = -EINVAL;
1463                 goto out_free_sb;
1464         }
1465
1466         error = xfs_filestream_mount(mp);
1467         if (error)
1468                 goto out_free_sb;
1469
1470         /*
1471          * we must configure the block size in the superblock before we run the
1472          * full mount process as the mount process can lookup and cache inodes.
1473          */
1474         sb->s_magic = XFS_SUPER_MAGIC;
1475         sb->s_blocksize = mp->m_sb.sb_blocksize;
1476         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1477         sb->s_maxbytes = MAX_LFS_FILESIZE;
1478         sb->s_max_links = XFS_MAXLINK;
1479         sb->s_time_gran = 1;
1480         sb->s_time_min = S32_MIN;
1481         sb->s_time_max = S32_MAX;
1482         sb->s_iflags |= SB_I_CGROUPWB;
1483
1484         set_posix_acl_flag(sb);
1485
1486         /* version 5 superblocks support inode version counters. */
1487         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1488                 sb->s_flags |= SB_I_VERSION;
1489
1490         if (mp->m_flags & XFS_MOUNT_DAX_ALWAYS) {
1491                 bool rtdev_is_dax = false, datadev_is_dax;
1492
1493                 xfs_warn(mp,
1494                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1495
1496                 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1497                         sb->s_blocksize);
1498                 if (mp->m_rtdev_targp)
1499                         rtdev_is_dax = bdev_dax_supported(
1500                                 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1501                 if (!rtdev_is_dax && !datadev_is_dax) {
1502                         xfs_alert(mp,
1503                         "DAX unsupported by block device. Turning off DAX.");
1504                         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1505                 }
1506                 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1507                         xfs_alert(mp,
1508                 "DAX and reflink cannot be used together!");
1509                         error = -EINVAL;
1510                         goto out_filestream_unmount;
1511                 }
1512         }
1513
1514         if (mp->m_flags & XFS_MOUNT_DISCARD) {
1515                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1516
1517                 if (!blk_queue_discard(q)) {
1518                         xfs_warn(mp, "mounting with \"discard\" option, but "
1519                                         "the device does not support discard");
1520                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
1521                 }
1522         }
1523
1524         if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1525                 if (mp->m_sb.sb_rblocks) {
1526                         xfs_alert(mp,
1527         "reflink not compatible with realtime device!");
1528                         error = -EINVAL;
1529                         goto out_filestream_unmount;
1530                 }
1531
1532                 if (xfs_globals.always_cow) {
1533                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1534                         mp->m_always_cow = true;
1535                 }
1536         }
1537
1538         if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1539                 xfs_alert(mp,
1540         "reverse mapping btree not compatible with realtime device!");
1541                 error = -EINVAL;
1542                 goto out_filestream_unmount;
1543         }
1544
1545         error = xfs_mountfs(mp);
1546         if (error)
1547                 goto out_filestream_unmount;
1548
1549         root = igrab(VFS_I(mp->m_rootip));
1550         if (!root) {
1551                 error = -ENOENT;
1552                 goto out_unmount;
1553         }
1554         sb->s_root = d_make_root(root);
1555         if (!sb->s_root) {
1556                 error = -ENOMEM;
1557                 goto out_unmount;
1558         }
1559
1560         return 0;
1561
1562  out_filestream_unmount:
1563         xfs_filestream_unmount(mp);
1564  out_free_sb:
1565         xfs_freesb(mp);
1566  out_free_stats:
1567         free_percpu(mp->m_stats.xs_stats);
1568  out_destroy_counters:
1569         xfs_destroy_percpu_counters(mp);
1570  out_destroy_workqueues:
1571         xfs_destroy_mount_workqueues(mp);
1572  out_close_devices:
1573         xfs_close_devices(mp);
1574  out_free_names:
1575         sb->s_fs_info = NULL;
1576         xfs_mount_free(mp);
1577         return error;
1578
1579  out_unmount:
1580         xfs_filestream_unmount(mp);
1581         xfs_unmountfs(mp);
1582         goto out_free_sb;
1583 }
1584
1585 static int
1586 xfs_fc_get_tree(
1587         struct fs_context       *fc)
1588 {
1589         return get_tree_bdev(fc, xfs_fc_fill_super);
1590 }
1591
1592 static int
1593 xfs_remount_rw(
1594         struct xfs_mount        *mp)
1595 {
1596         struct xfs_sb           *sbp = &mp->m_sb;
1597         int error;
1598
1599         if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1600                 xfs_warn(mp,
1601                         "ro->rw transition prohibited on norecovery mount");
1602                 return -EINVAL;
1603         }
1604
1605         if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1606             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1607                 xfs_warn(mp,
1608         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1609                         (sbp->sb_features_ro_compat &
1610                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1611                 return -EINVAL;
1612         }
1613
1614         mp->m_flags &= ~XFS_MOUNT_RDONLY;
1615
1616         /*
1617          * If this is the first remount to writeable state we might have some
1618          * superblock changes to update.
1619          */
1620         if (mp->m_update_sb) {
1621                 error = xfs_sync_sb(mp, false);
1622                 if (error) {
1623                         xfs_warn(mp, "failed to write sb changes");
1624                         return error;
1625                 }
1626                 mp->m_update_sb = false;
1627         }
1628
1629         /*
1630          * Fill out the reserve pool if it is empty. Use the stashed value if
1631          * it is non-zero, otherwise go with the default.
1632          */
1633         xfs_restore_resvblks(mp);
1634         xfs_log_work_queue(mp);
1635
1636         /* Recover any CoW blocks that never got remapped. */
1637         error = xfs_reflink_recover_cow(mp);
1638         if (error) {
1639                 xfs_err(mp,
1640                         "Error %d recovering leftover CoW allocations.", error);
1641                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1642                 return error;
1643         }
1644         xfs_start_block_reaping(mp);
1645
1646         /* Create the per-AG metadata reservation pool .*/
1647         error = xfs_fs_reserve_ag_blocks(mp);
1648         if (error && error != -ENOSPC)
1649                 return error;
1650
1651         return 0;
1652 }
1653
1654 static int
1655 xfs_remount_ro(
1656         struct xfs_mount        *mp)
1657 {
1658         int error;
1659
1660         /*
1661          * Cancel background eofb scanning so it cannot race with the final
1662          * log force+buftarg wait and deadlock the remount.
1663          */
1664         xfs_stop_block_reaping(mp);
1665
1666         /* Get rid of any leftover CoW reservations... */
1667         error = xfs_icache_free_cowblocks(mp, NULL);
1668         if (error) {
1669                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1670                 return error;
1671         }
1672
1673         /* Free the per-AG metadata reservation pool. */
1674         error = xfs_fs_unreserve_ag_blocks(mp);
1675         if (error) {
1676                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1677                 return error;
1678         }
1679
1680         /*
1681          * Before we sync the metadata, we need to free up the reserve block
1682          * pool so that the used block count in the superblock on disk is
1683          * correct at the end of the remount. Stash the current* reserve pool
1684          * size so that if we get remounted rw, we can return it to the same
1685          * size.
1686          */
1687         xfs_save_resvblks(mp);
1688
1689         xfs_quiesce_attr(mp);
1690         mp->m_flags |= XFS_MOUNT_RDONLY;
1691
1692         return 0;
1693 }
1694
1695 /*
1696  * Logically we would return an error here to prevent users from believing
1697  * they might have changed mount options using remount which can't be changed.
1698  *
1699  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1700  * arguments in some cases so we can't blindly reject options, but have to
1701  * check for each specified option if it actually differs from the currently
1702  * set option and only reject it if that's the case.
1703  *
1704  * Until that is implemented we return success for every remount request, and
1705  * silently ignore all options that we can't actually change.
1706  */
1707 static int
1708 xfs_fc_reconfigure(
1709         struct fs_context *fc)
1710 {
1711         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1712         struct xfs_mount        *new_mp = fc->s_fs_info;
1713         xfs_sb_t                *sbp = &mp->m_sb;
1714         int                     flags = fc->sb_flags;
1715         int                     error;
1716
1717         error = xfs_fc_validate_params(new_mp);
1718         if (error)
1719                 return error;
1720
1721         sync_filesystem(mp->m_super);
1722
1723         /* inode32 -> inode64 */
1724         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1725             !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1726                 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1727                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1728         }
1729
1730         /* inode64 -> inode32 */
1731         if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1732             (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1733                 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1734                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1735         }
1736
1737         /* ro -> rw */
1738         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1739                 error = xfs_remount_rw(mp);
1740                 if (error)
1741                         return error;
1742         }
1743
1744         /* rw -> ro */
1745         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1746                 error = xfs_remount_ro(mp);
1747                 if (error)
1748                         return error;
1749         }
1750
1751         return 0;
1752 }
1753
1754 static void xfs_fc_free(
1755         struct fs_context       *fc)
1756 {
1757         struct xfs_mount        *mp = fc->s_fs_info;
1758
1759         /*
1760          * mp is stored in the fs_context when it is initialized.
1761          * mp is transferred to the superblock on a successful mount,
1762          * but if an error occurs before the transfer we have to free
1763          * it here.
1764          */
1765         if (mp)
1766                 xfs_mount_free(mp);
1767 }
1768
1769 static const struct fs_context_operations xfs_context_ops = {
1770         .parse_param = xfs_fc_parse_param,
1771         .get_tree    = xfs_fc_get_tree,
1772         .reconfigure = xfs_fc_reconfigure,
1773         .free        = xfs_fc_free,
1774 };
1775
1776 static int xfs_init_fs_context(
1777         struct fs_context       *fc)
1778 {
1779         struct xfs_mount        *mp;
1780
1781         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1782         if (!mp)
1783                 return -ENOMEM;
1784
1785         spin_lock_init(&mp->m_sb_lock);
1786         spin_lock_init(&mp->m_agirotor_lock);
1787         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1788         spin_lock_init(&mp->m_perag_lock);
1789         mutex_init(&mp->m_growlock);
1790         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1791         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1792         INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1793         INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1794         mp->m_kobj.kobject.kset = xfs_kset;
1795         /*
1796          * We don't create the finobt per-ag space reservation until after log
1797          * recovery, so we must set this to true so that an ifree transaction
1798          * started during log recovery will not depend on space reservations
1799          * for finobt expansion.
1800          */
1801         mp->m_finobt_nores = true;
1802
1803         /*
1804          * These can be overridden by the mount option parsing.
1805          */
1806         mp->m_logbufs = -1;
1807         mp->m_logbsize = -1;
1808         mp->m_allocsize_log = 16; /* 64k */
1809
1810         /*
1811          * Copy binary VFS mount flags we are interested in.
1812          */
1813         if (fc->sb_flags & SB_RDONLY)
1814                 mp->m_flags |= XFS_MOUNT_RDONLY;
1815         if (fc->sb_flags & SB_DIRSYNC)
1816                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1817         if (fc->sb_flags & SB_SYNCHRONOUS)
1818                 mp->m_flags |= XFS_MOUNT_WSYNC;
1819
1820         fc->s_fs_info = mp;
1821         fc->ops = &xfs_context_ops;
1822
1823         return 0;
1824 }
1825
1826 static struct file_system_type xfs_fs_type = {
1827         .owner                  = THIS_MODULE,
1828         .name                   = "xfs",
1829         .init_fs_context        = xfs_init_fs_context,
1830         .parameters             = xfs_fs_parameters,
1831         .kill_sb                = kill_block_super,
1832         .fs_flags               = FS_REQUIRES_DEV,
1833 };
1834 MODULE_ALIAS_FS("xfs");
1835
1836 STATIC int __init
1837 xfs_init_zones(void)
1838 {
1839         xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1840                                                 sizeof(struct xlog_ticket),
1841                                                 0, 0, NULL);
1842         if (!xfs_log_ticket_zone)
1843                 goto out;
1844
1845         xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1846                                         sizeof(struct xfs_extent_free_item),
1847                                         0, 0, NULL);
1848         if (!xfs_bmap_free_item_zone)
1849                 goto out_destroy_log_ticket_zone;
1850
1851         xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1852                                                sizeof(struct xfs_btree_cur),
1853                                                0, 0, NULL);
1854         if (!xfs_btree_cur_zone)
1855                 goto out_destroy_bmap_free_item_zone;
1856
1857         xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1858                                               sizeof(struct xfs_da_state),
1859                                               0, 0, NULL);
1860         if (!xfs_da_state_zone)
1861                 goto out_destroy_btree_cur_zone;
1862
1863         xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1864                                            sizeof(struct xfs_ifork),
1865                                            0, 0, NULL);
1866         if (!xfs_ifork_zone)
1867                 goto out_destroy_da_state_zone;
1868
1869         xfs_trans_zone = kmem_cache_create("xf_trans",
1870                                            sizeof(struct xfs_trans),
1871                                            0, 0, NULL);
1872         if (!xfs_trans_zone)
1873                 goto out_destroy_ifork_zone;
1874
1875
1876         /*
1877          * The size of the zone allocated buf log item is the maximum
1878          * size possible under XFS.  This wastes a little bit of memory,
1879          * but it is much faster.
1880          */
1881         xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1882                                               sizeof(struct xfs_buf_log_item),
1883                                               0, 0, NULL);
1884         if (!xfs_buf_item_zone)
1885                 goto out_destroy_trans_zone;
1886
1887         xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1888                                         (sizeof(struct xfs_efd_log_item) +
1889                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1890                                         sizeof(struct xfs_extent)),
1891                                         0, 0, NULL);
1892         if (!xfs_efd_zone)
1893                 goto out_destroy_buf_item_zone;
1894
1895         xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1896                                          (sizeof(struct xfs_efi_log_item) +
1897                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1898                                          sizeof(struct xfs_extent)),
1899                                          0, 0, NULL);
1900         if (!xfs_efi_zone)
1901                 goto out_destroy_efd_zone;
1902
1903         xfs_inode_zone = kmem_cache_create("xfs_inode",
1904                                            sizeof(struct xfs_inode), 0,
1905                                            (SLAB_HWCACHE_ALIGN |
1906                                             SLAB_RECLAIM_ACCOUNT |
1907                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1908                                            xfs_fs_inode_init_once);
1909         if (!xfs_inode_zone)
1910                 goto out_destroy_efi_zone;
1911
1912         xfs_ili_zone = kmem_cache_create("xfs_ili",
1913                                          sizeof(struct xfs_inode_log_item), 0,
1914                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1915                                          NULL);
1916         if (!xfs_ili_zone)
1917                 goto out_destroy_inode_zone;
1918
1919         xfs_icreate_zone = kmem_cache_create("xfs_icr",
1920                                              sizeof(struct xfs_icreate_item),
1921                                              0, 0, NULL);
1922         if (!xfs_icreate_zone)
1923                 goto out_destroy_ili_zone;
1924
1925         xfs_rud_zone = kmem_cache_create("xfs_rud_item",
1926                                          sizeof(struct xfs_rud_log_item),
1927                                          0, 0, NULL);
1928         if (!xfs_rud_zone)
1929                 goto out_destroy_icreate_zone;
1930
1931         xfs_rui_zone = kmem_cache_create("xfs_rui_item",
1932                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1933                         0, 0, NULL);
1934         if (!xfs_rui_zone)
1935                 goto out_destroy_rud_zone;
1936
1937         xfs_cud_zone = kmem_cache_create("xfs_cud_item",
1938                                          sizeof(struct xfs_cud_log_item),
1939                                          0, 0, NULL);
1940         if (!xfs_cud_zone)
1941                 goto out_destroy_rui_zone;
1942
1943         xfs_cui_zone = kmem_cache_create("xfs_cui_item",
1944                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1945                         0, 0, NULL);
1946         if (!xfs_cui_zone)
1947                 goto out_destroy_cud_zone;
1948
1949         xfs_bud_zone = kmem_cache_create("xfs_bud_item",
1950                                          sizeof(struct xfs_bud_log_item),
1951                                          0, 0, NULL);
1952         if (!xfs_bud_zone)
1953                 goto out_destroy_cui_zone;
1954
1955         xfs_bui_zone = kmem_cache_create("xfs_bui_item",
1956                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1957                         0, 0, NULL);
1958         if (!xfs_bui_zone)
1959                 goto out_destroy_bud_zone;
1960
1961         return 0;
1962
1963  out_destroy_bud_zone:
1964         kmem_cache_destroy(xfs_bud_zone);
1965  out_destroy_cui_zone:
1966         kmem_cache_destroy(xfs_cui_zone);
1967  out_destroy_cud_zone:
1968         kmem_cache_destroy(xfs_cud_zone);
1969  out_destroy_rui_zone:
1970         kmem_cache_destroy(xfs_rui_zone);
1971  out_destroy_rud_zone:
1972         kmem_cache_destroy(xfs_rud_zone);
1973  out_destroy_icreate_zone:
1974         kmem_cache_destroy(xfs_icreate_zone);
1975  out_destroy_ili_zone:
1976         kmem_cache_destroy(xfs_ili_zone);
1977  out_destroy_inode_zone:
1978         kmem_cache_destroy(xfs_inode_zone);
1979  out_destroy_efi_zone:
1980         kmem_cache_destroy(xfs_efi_zone);
1981  out_destroy_efd_zone:
1982         kmem_cache_destroy(xfs_efd_zone);
1983  out_destroy_buf_item_zone:
1984         kmem_cache_destroy(xfs_buf_item_zone);
1985  out_destroy_trans_zone:
1986         kmem_cache_destroy(xfs_trans_zone);
1987  out_destroy_ifork_zone:
1988         kmem_cache_destroy(xfs_ifork_zone);
1989  out_destroy_da_state_zone:
1990         kmem_cache_destroy(xfs_da_state_zone);
1991  out_destroy_btree_cur_zone:
1992         kmem_cache_destroy(xfs_btree_cur_zone);
1993  out_destroy_bmap_free_item_zone:
1994         kmem_cache_destroy(xfs_bmap_free_item_zone);
1995  out_destroy_log_ticket_zone:
1996         kmem_cache_destroy(xfs_log_ticket_zone);
1997  out:
1998         return -ENOMEM;
1999 }
2000
2001 STATIC void
2002 xfs_destroy_zones(void)
2003 {
2004         /*
2005          * Make sure all delayed rcu free are flushed before we
2006          * destroy caches.
2007          */
2008         rcu_barrier();
2009         kmem_cache_destroy(xfs_bui_zone);
2010         kmem_cache_destroy(xfs_bud_zone);
2011         kmem_cache_destroy(xfs_cui_zone);
2012         kmem_cache_destroy(xfs_cud_zone);
2013         kmem_cache_destroy(xfs_rui_zone);
2014         kmem_cache_destroy(xfs_rud_zone);
2015         kmem_cache_destroy(xfs_icreate_zone);
2016         kmem_cache_destroy(xfs_ili_zone);
2017         kmem_cache_destroy(xfs_inode_zone);
2018         kmem_cache_destroy(xfs_efi_zone);
2019         kmem_cache_destroy(xfs_efd_zone);
2020         kmem_cache_destroy(xfs_buf_item_zone);
2021         kmem_cache_destroy(xfs_trans_zone);
2022         kmem_cache_destroy(xfs_ifork_zone);
2023         kmem_cache_destroy(xfs_da_state_zone);
2024         kmem_cache_destroy(xfs_btree_cur_zone);
2025         kmem_cache_destroy(xfs_bmap_free_item_zone);
2026         kmem_cache_destroy(xfs_log_ticket_zone);
2027 }
2028
2029 STATIC int __init
2030 xfs_init_workqueues(void)
2031 {
2032         /*
2033          * The allocation workqueue can be used in memory reclaim situations
2034          * (writepage path), and parallelism is only limited by the number of
2035          * AGs in all the filesystems mounted. Hence use the default large
2036          * max_active value for this workqueue.
2037          */
2038         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2039                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2040         if (!xfs_alloc_wq)
2041                 return -ENOMEM;
2042
2043         xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2044         if (!xfs_discard_wq)
2045                 goto out_free_alloc_wq;
2046
2047         return 0;
2048 out_free_alloc_wq:
2049         destroy_workqueue(xfs_alloc_wq);
2050         return -ENOMEM;
2051 }
2052
2053 STATIC void
2054 xfs_destroy_workqueues(void)
2055 {
2056         destroy_workqueue(xfs_discard_wq);
2057         destroy_workqueue(xfs_alloc_wq);
2058 }
2059
2060 STATIC int __init
2061 init_xfs_fs(void)
2062 {
2063         int                     error;
2064
2065         xfs_check_ondisk_structs();
2066
2067         printk(KERN_INFO XFS_VERSION_STRING " with "
2068                          XFS_BUILD_OPTIONS " enabled\n");
2069
2070         xfs_dir_startup();
2071
2072         error = xfs_init_zones();
2073         if (error)
2074                 goto out;
2075
2076         error = xfs_init_workqueues();
2077         if (error)
2078                 goto out_destroy_zones;
2079
2080         error = xfs_mru_cache_init();
2081         if (error)
2082                 goto out_destroy_wq;
2083
2084         error = xfs_buf_init();
2085         if (error)
2086                 goto out_mru_cache_uninit;
2087
2088         error = xfs_init_procfs();
2089         if (error)
2090                 goto out_buf_terminate;
2091
2092         error = xfs_sysctl_register();
2093         if (error)
2094                 goto out_cleanup_procfs;
2095
2096         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2097         if (!xfs_kset) {
2098                 error = -ENOMEM;
2099                 goto out_sysctl_unregister;
2100         }
2101
2102         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2103
2104         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2105         if (!xfsstats.xs_stats) {
2106                 error = -ENOMEM;
2107                 goto out_kset_unregister;
2108         }
2109
2110         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2111                                "stats");
2112         if (error)
2113                 goto out_free_stats;
2114
2115 #ifdef DEBUG
2116         xfs_dbg_kobj.kobject.kset = xfs_kset;
2117         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2118         if (error)
2119                 goto out_remove_stats_kobj;
2120 #endif
2121
2122         error = xfs_qm_init();
2123         if (error)
2124                 goto out_remove_dbg_kobj;
2125
2126         error = register_filesystem(&xfs_fs_type);
2127         if (error)
2128                 goto out_qm_exit;
2129         return 0;
2130
2131  out_qm_exit:
2132         xfs_qm_exit();
2133  out_remove_dbg_kobj:
2134 #ifdef DEBUG
2135         xfs_sysfs_del(&xfs_dbg_kobj);
2136  out_remove_stats_kobj:
2137 #endif
2138         xfs_sysfs_del(&xfsstats.xs_kobj);
2139  out_free_stats:
2140         free_percpu(xfsstats.xs_stats);
2141  out_kset_unregister:
2142         kset_unregister(xfs_kset);
2143  out_sysctl_unregister:
2144         xfs_sysctl_unregister();
2145  out_cleanup_procfs:
2146         xfs_cleanup_procfs();
2147  out_buf_terminate:
2148         xfs_buf_terminate();
2149  out_mru_cache_uninit:
2150         xfs_mru_cache_uninit();
2151  out_destroy_wq:
2152         xfs_destroy_workqueues();
2153  out_destroy_zones:
2154         xfs_destroy_zones();
2155  out:
2156         return error;
2157 }
2158
2159 STATIC void __exit
2160 exit_xfs_fs(void)
2161 {
2162         xfs_qm_exit();
2163         unregister_filesystem(&xfs_fs_type);
2164 #ifdef DEBUG
2165         xfs_sysfs_del(&xfs_dbg_kobj);
2166 #endif
2167         xfs_sysfs_del(&xfsstats.xs_kobj);
2168         free_percpu(xfsstats.xs_stats);
2169         kset_unregister(xfs_kset);
2170         xfs_sysctl_unregister();
2171         xfs_cleanup_procfs();
2172         xfs_buf_terminate();
2173         xfs_mru_cache_uninit();
2174         xfs_destroy_workqueues();
2175         xfs_destroy_zones();
2176         xfs_uuid_table_free();
2177 }
2178
2179 module_init(init_xfs_fs);
2180 module_exit(exit_xfs_fs);
2181
2182 MODULE_AUTHOR("Silicon Graphics, Inc.");
2183 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2184 MODULE_LICENSE("GPL");