fs/xfs/xfs_super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6
   7 #include "xfs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_sb.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_btree.h"
  16 #include "xfs_bmap.h"
  17 #include "xfs_alloc.h"
  18 #include "xfs_fsops.h"
  19 #include "xfs_trans.h"
  20 #include "xfs_buf_item.h"
  21 #include "xfs_log.h"
  22 #include "xfs_log_priv.h"
  23 #include "xfs_dir2.h"
  24 #include "xfs_extfree_item.h"
  25 #include "xfs_mru_cache.h"
  26 #include "xfs_inode_item.h"
  27 #include "xfs_icache.h"
  28 #include "xfs_trace.h"
  29 #include "xfs_icreate_item.h"
  30 #include "xfs_filestream.h"
  31 #include "xfs_quota.h"
  32 #include "xfs_sysfs.h"
  33 #include "xfs_ondisk.h"
  34 #include "xfs_rmap_item.h"
  35 #include "xfs_refcount_item.h"
  36 #include "xfs_bmap_item.h"
  37 #include "xfs_reflink.h"
  38 #include "xfs_pwork.h"
  39 #include "xfs_ag.h"
  40
  41 #include <linux/magic.h>
  42 #include <linux/fs_context.h>
  43 #include <linux/fs_parser.h>
  44
  45 static const struct super_operations xfs_super_operations;
  46
  47 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  48 #ifdef DEBUG
  49 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  50 #endif
  51
  52 enum xfs_dax_mode {
  53         XFS_DAX_INODE = 0,
  54         XFS_DAX_ALWAYS = 1,
  55         XFS_DAX_NEVER = 2,
  56 };
  57
  58 static void
  59 xfs_mount_set_dax_mode(
  60         struct xfs_mount        *mp,
  61         enum xfs_dax_mode       mode)
  62 {
  63         switch (mode) {
  64         case XFS_DAX_INODE:
  65                 mp->m_flags &= ~(XFS_MOUNT_DAX_ALWAYS | XFS_MOUNT_DAX_NEVER);
  66                 break;
  67         case XFS_DAX_ALWAYS:
  68                 mp->m_flags |= XFS_MOUNT_DAX_ALWAYS;
  69                 mp->m_flags &= ~XFS_MOUNT_DAX_NEVER;
  70                 break;
  71         case XFS_DAX_NEVER:
  72                 mp->m_flags |= XFS_MOUNT_DAX_NEVER;
  73                 mp->m_flags &= ~XFS_MOUNT_DAX_ALWAYS;
  74                 break;
  75         }
  76 }
  77
  78 static const struct constant_table dax_param_enums[] = {
  79         {"inode",       XFS_DAX_INODE },
  80         {"always",      XFS_DAX_ALWAYS },
  81         {"never",       XFS_DAX_NEVER },
  82         {}
  83 };
  84
  85 /*
  86  * Table driven mount option parser.
  87  */
  88 enum {
  89         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
  90         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
  91         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
  92         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
  93         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
  94         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
  95         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
  96         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
  97         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
  98 };
  99
 100 static const struct fs_parameter_spec xfs_fs_parameters[] = {
 101         fsparam_u32("logbufs",          Opt_logbufs),
 102         fsparam_string("logbsize",      Opt_logbsize),
 103         fsparam_string("logdev",        Opt_logdev),
 104         fsparam_string("rtdev",         Opt_rtdev),
 105         fsparam_flag("wsync",           Opt_wsync),
 106         fsparam_flag("noalign",         Opt_noalign),
 107         fsparam_flag("swalloc",         Opt_swalloc),
 108         fsparam_u32("sunit",            Opt_sunit),
 109         fsparam_u32("swidth",           Opt_swidth),
 110         fsparam_flag("nouuid",          Opt_nouuid),
 111         fsparam_flag("grpid",           Opt_grpid),
 112         fsparam_flag("nogrpid",         Opt_nogrpid),
 113         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 114         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 115         fsparam_string("allocsize",     Opt_allocsize),
 116         fsparam_flag("norecovery",      Opt_norecovery),
 117         fsparam_flag("inode64",         Opt_inode64),
 118         fsparam_flag("inode32",         Opt_inode32),
 119         fsparam_flag("ikeep",           Opt_ikeep),
 120         fsparam_flag("noikeep",         Opt_noikeep),
 121         fsparam_flag("largeio",         Opt_largeio),
 122         fsparam_flag("nolargeio",       Opt_nolargeio),
 123         fsparam_flag("attr2",           Opt_attr2),
 124         fsparam_flag("noattr2",         Opt_noattr2),
 125         fsparam_flag("filestreams",     Opt_filestreams),
 126         fsparam_flag("quota",           Opt_quota),
 127         fsparam_flag("noquota",         Opt_noquota),
 128         fsparam_flag("usrquota",        Opt_usrquota),
 129         fsparam_flag("grpquota",        Opt_grpquota),
 130         fsparam_flag("prjquota",        Opt_prjquota),
 131         fsparam_flag("uquota",          Opt_uquota),
 132         fsparam_flag("gquota",          Opt_gquota),
 133         fsparam_flag("pquota",          Opt_pquota),
 134         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 135         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 136         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 137         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 138         fsparam_flag("discard",         Opt_discard),
 139         fsparam_flag("nodiscard",       Opt_nodiscard),
 140         fsparam_flag("dax",             Opt_dax),
 141         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 142         {}
 143 };
 144
 145 struct proc_xfs_info {
 146         uint64_t        flag;
 147         char            *str;
 148 };
 149
 150 static int
 151 xfs_fs_show_options(
 152         struct seq_file         *m,
 153         struct dentry           *root)
 154 {
 155         static struct proc_xfs_info xfs_info_set[] = {
 156                 /* the few simple ones we can get from the mount struct */
 157                 { XFS_MOUNT_IKEEP,              ",ikeep" },
 158                 { XFS_MOUNT_WSYNC,              ",wsync" },
 159                 { XFS_MOUNT_NOALIGN,            ",noalign" },
 160                 { XFS_MOUNT_SWALLOC,            ",swalloc" },
 161                 { XFS_MOUNT_NOUUID,             ",nouuid" },
 162                 { XFS_MOUNT_NORECOVERY,         ",norecovery" },
 163                 { XFS_MOUNT_ATTR2,              ",attr2" },
 164                 { XFS_MOUNT_FILESTREAMS,        ",filestreams" },
 165                 { XFS_MOUNT_GRPID,              ",grpid" },
 166                 { XFS_MOUNT_DISCARD,            ",discard" },
 167                 { XFS_MOUNT_LARGEIO,            ",largeio" },
 168                 { XFS_MOUNT_DAX_ALWAYS,         ",dax=always" },
 169                 { XFS_MOUNT_DAX_NEVER,          ",dax=never" },
 170                 { 0, NULL }
 171         };
 172         struct xfs_mount        *mp = XFS_M(root->d_sb);
 173         struct proc_xfs_info    *xfs_infop;
 174
 175         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 176                 if (mp->m_flags & xfs_infop->flag)
 177                         seq_puts(m, xfs_infop->str);
 178         }
 179
 180         seq_printf(m, ",inode%d",
 181                 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
 182
 183         if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
 184                 seq_printf(m, ",allocsize=%dk",
 185                            (1 << mp->m_allocsize_log) >> 10);
 186
 187         if (mp->m_logbufs > 0)
 188                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 189         if (mp->m_logbsize > 0)
 190                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 191
 192         if (mp->m_logname)
 193                 seq_show_option(m, "logdev", mp->m_logname);
 194         if (mp->m_rtname)
 195                 seq_show_option(m, "rtdev", mp->m_rtname);
 196
 197         if (mp->m_dalign > 0)
 198                 seq_printf(m, ",sunit=%d",
 199                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 200         if (mp->m_swidth > 0)
 201                 seq_printf(m, ",swidth=%d",
 202                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 203
 204         if (mp->m_qflags & XFS_UQUOTA_ACCT) {
 205                 if (mp->m_qflags & XFS_UQUOTA_ENFD)
 206                         seq_puts(m, ",usrquota");
 207                 else
 208                         seq_puts(m, ",uqnoenforce");
 209         }
 210
 211         if (mp->m_qflags & XFS_PQUOTA_ACCT) {
 212                 if (mp->m_qflags & XFS_PQUOTA_ENFD)
 213                         seq_puts(m, ",prjquota");
 214                 else
 215                         seq_puts(m, ",pqnoenforce");
 216         }
 217         if (mp->m_qflags & XFS_GQUOTA_ACCT) {
 218                 if (mp->m_qflags & XFS_GQUOTA_ENFD)
 219                         seq_puts(m, ",grpquota");
 220                 else
 221                         seq_puts(m, ",gqnoenforce");
 222         }
 223
 224         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 225                 seq_puts(m, ",noquota");
 226
 227         return 0;
 228 }
 229
 230 /*
 231  * Set parameters for inode allocation heuristics, taking into account
 232  * filesystem size and inode32/inode64 mount options; i.e. specifically
 233  * whether or not XFS_MOUNT_SMALL_INUMS is set.
 234  *
 235  * Inode allocation patterns are altered only if inode32 is requested
 236  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
 237  * If altered, XFS_MOUNT_32BITINODES is set as well.
 238  *
 239  * An agcount independent of that in the mount structure is provided
 240  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 241  * to the potentially higher ag count.
 242  *
 243  * Returns the maximum AG index which may contain inodes.
 244  */
 245 xfs_agnumber_t
 246 xfs_set_inode_alloc(
 247         struct xfs_mount *mp,
 248         xfs_agnumber_t  agcount)
 249 {
 250         xfs_agnumber_t  index;
 251         xfs_agnumber_t  maxagi = 0;
 252         xfs_sb_t        *sbp = &mp->m_sb;
 253         xfs_agnumber_t  max_metadata;
 254         xfs_agino_t     agino;
 255         xfs_ino_t       ino;
 256
 257         /*
 258          * Calculate how much should be reserved for inodes to meet
 259          * the max inode percentage.  Used only for inode32.
 260          */
 261         if (M_IGEO(mp)->maxicount) {
 262                 uint64_t        icount;
 263
 264                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 265                 do_div(icount, 100);
 266                 icount += sbp->sb_agblocks - 1;
 267                 do_div(icount, sbp->sb_agblocks);
 268                 max_metadata = icount;
 269         } else {
 270                 max_metadata = agcount;
 271         }
 272
 273         /* Get the last possible inode in the filesystem */
 274         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 275         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 276
 277         /*
 278          * If user asked for no more than 32-bit inodes, and the fs is
 279          * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
 280          * the allocator to accommodate the request.
 281          */
 282         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
 283                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 284         else
 285                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
 286
 287         for (index = 0; index < agcount; index++) {
 288                 struct xfs_perag        *pag;
 289
 290                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 291
 292                 pag = xfs_perag_get(mp, index);
 293
 294                 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
 295                         if (ino > XFS_MAXINUMBER_32) {
 296                                 pag->pagi_inodeok = 0;
 297                                 pag->pagf_metadata = 0;
 298                         } else {
 299                                 pag->pagi_inodeok = 1;
 300                                 maxagi++;
 301                                 if (index < max_metadata)
 302                                         pag->pagf_metadata = 1;
 303                                 else
 304                                         pag->pagf_metadata = 0;
 305                         }
 306                 } else {
 307                         pag->pagi_inodeok = 1;
 308                         pag->pagf_metadata = 0;
 309                 }
 310
 311                 xfs_perag_put(pag);
 312         }
 313
 314         return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
 315 }
 316
 317 static bool
 318 xfs_buftarg_is_dax(
 319         struct super_block      *sb,
 320         struct xfs_buftarg      *bt)
 321 {
 322         return dax_supported(bt->bt_daxdev, bt->bt_bdev, sb->s_blocksize, 0,
 323                         bdev_nr_sectors(bt->bt_bdev));
 324 }
 325
 326 STATIC int
 327 xfs_blkdev_get(
 328         xfs_mount_t             *mp,
 329         const char              *name,
 330         struct block_device     **bdevp)
 331 {
 332         int                     error = 0;
 333
 334         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 335                                     mp);
 336         if (IS_ERR(*bdevp)) {
 337                 error = PTR_ERR(*bdevp);
 338                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 339         }
 340
 341         return error;
 342 }
 343
 344 STATIC void
 345 xfs_blkdev_put(
 346         struct block_device     *bdev)
 347 {
 348         if (bdev)
 349                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 350 }
 351
 352 STATIC void
 353 xfs_close_devices(
 354         struct xfs_mount        *mp)
 355 {
 356         struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
 357
 358         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 359                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 360                 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
 361
 362                 xfs_free_buftarg(mp->m_logdev_targp);
 363                 xfs_blkdev_put(logdev);
 364                 fs_put_dax(dax_logdev);
 365         }
 366         if (mp->m_rtdev_targp) {
 367                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 368                 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
 369
 370                 xfs_free_buftarg(mp->m_rtdev_targp);
 371                 xfs_blkdev_put(rtdev);
 372                 fs_put_dax(dax_rtdev);
 373         }
 374         xfs_free_buftarg(mp->m_ddev_targp);
 375         fs_put_dax(dax_ddev);
 376 }
 377
 378 /*
 379  * The file system configurations are:
 380  *      (1) device (partition) with data and internal log
 381  *      (2) logical volume with data and log subvolumes.
 382  *      (3) logical volume with data, log, and realtime subvolumes.
 383  *
 384  * We only have to handle opening the log and realtime volumes here if
 385  * they are present.  The data subvolume has already been opened by
 386  * get_sb_bdev() and is stored in sb->s_bdev.
 387  */
 388 STATIC int
 389 xfs_open_devices(
 390         struct xfs_mount        *mp)
 391 {
 392         struct block_device     *ddev = mp->m_super->s_bdev;
 393         struct dax_device       *dax_ddev = fs_dax_get_by_bdev(ddev);
 394         struct dax_device       *dax_logdev = NULL, *dax_rtdev = NULL;
 395         struct block_device     *logdev = NULL, *rtdev = NULL;
 396         int                     error;
 397
 398         /*
 399          * Open real time and log devices - order is important.
 400          */
 401         if (mp->m_logname) {
 402                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 403                 if (error)
 404                         goto out;
 405                 dax_logdev = fs_dax_get_by_bdev(logdev);
 406         }
 407
 408         if (mp->m_rtname) {
 409                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 410                 if (error)
 411                         goto out_close_logdev;
 412
 413                 if (rtdev == ddev || rtdev == logdev) {
 414                         xfs_warn(mp,
 415         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 416                         error = -EINVAL;
 417                         goto out_close_rtdev;
 418                 }
 419                 dax_rtdev = fs_dax_get_by_bdev(rtdev);
 420         }
 421
 422         /*
 423          * Setup xfs_mount buffer target pointers
 424          */
 425         error = -ENOMEM;
 426         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
 427         if (!mp->m_ddev_targp)
 428                 goto out_close_rtdev;
 429
 430         if (rtdev) {
 431                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
 432                 if (!mp->m_rtdev_targp)
 433                         goto out_free_ddev_targ;
 434         }
 435
 436         if (logdev && logdev != ddev) {
 437                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
 438                 if (!mp->m_logdev_targp)
 439                         goto out_free_rtdev_targ;
 440         } else {
 441                 mp->m_logdev_targp = mp->m_ddev_targp;
 442         }
 443
 444         return 0;
 445
 446  out_free_rtdev_targ:
 447         if (mp->m_rtdev_targp)
 448                 xfs_free_buftarg(mp->m_rtdev_targp);
 449  out_free_ddev_targ:
 450         xfs_free_buftarg(mp->m_ddev_targp);
 451  out_close_rtdev:
 452         xfs_blkdev_put(rtdev);
 453         fs_put_dax(dax_rtdev);
 454  out_close_logdev:
 455         if (logdev && logdev != ddev) {
 456                 xfs_blkdev_put(logdev);
 457                 fs_put_dax(dax_logdev);
 458         }
 459  out:
 460         fs_put_dax(dax_ddev);
 461         return error;
 462 }
 463
 464 /*
 465  * Setup xfs_mount buffer target pointers based on superblock
 466  */
 467 STATIC int
 468 xfs_setup_devices(
 469         struct xfs_mount        *mp)
 470 {
 471         int                     error;
 472
 473         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 474         if (error)
 475                 return error;
 476
 477         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 478                 unsigned int    log_sector_size = BBSIZE;
 479
 480                 if (xfs_sb_version_hassector(&mp->m_sb))
 481                         log_sector_size = mp->m_sb.sb_logsectsize;
 482                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 483                                             log_sector_size);
 484                 if (error)
 485                         return error;
 486         }
 487         if (mp->m_rtdev_targp) {
 488                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 489                                             mp->m_sb.sb_sectsize);
 490                 if (error)
 491                         return error;
 492         }
 493
 494         return 0;
 495 }
 496
 497 STATIC int
 498 xfs_init_mount_workqueues(
 499         struct xfs_mount        *mp)
 500 {
 501         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 502                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 503                         1, mp->m_super->s_id);
 504         if (!mp->m_buf_workqueue)
 505                 goto out;
 506
 507         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 508                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 509                         0, mp->m_super->s_id);
 510         if (!mp->m_unwritten_workqueue)
 511                 goto out_destroy_buf;
 512
 513         mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
 514                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | WQ_UNBOUND),
 515                         0, mp->m_super->s_id);
 516         if (!mp->m_cil_workqueue)
 517                 goto out_destroy_unwritten;
 518
 519         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 520                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 521                         0, mp->m_super->s_id);
 522         if (!mp->m_reclaim_workqueue)
 523                 goto out_destroy_cil;
 524
 525         mp->m_gc_workqueue = alloc_workqueue("xfs-gc/%s",
 526                         WQ_SYSFS | WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM,
 527                         0, mp->m_super->s_id);
 528         if (!mp->m_gc_workqueue)
 529                 goto out_destroy_reclaim;
 530
 531         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 532                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 533         if (!mp->m_sync_workqueue)
 534                 goto out_destroy_eofb;
 535
 536         return 0;
 537
 538 out_destroy_eofb:
 539         destroy_workqueue(mp->m_gc_workqueue);
 540 out_destroy_reclaim:
 541         destroy_workqueue(mp->m_reclaim_workqueue);
 542 out_destroy_cil:
 543         destroy_workqueue(mp->m_cil_workqueue);
 544 out_destroy_unwritten:
 545         destroy_workqueue(mp->m_unwritten_workqueue);
 546 out_destroy_buf:
 547         destroy_workqueue(mp->m_buf_workqueue);
 548 out:
 549         return -ENOMEM;
 550 }
 551
 552 STATIC void
 553 xfs_destroy_mount_workqueues(
 554         struct xfs_mount        *mp)
 555 {
 556         destroy_workqueue(mp->m_sync_workqueue);
 557         destroy_workqueue(mp->m_gc_workqueue);
 558         destroy_workqueue(mp->m_reclaim_workqueue);
 559         destroy_workqueue(mp->m_cil_workqueue);
 560         destroy_workqueue(mp->m_unwritten_workqueue);
 561         destroy_workqueue(mp->m_buf_workqueue);
 562 }
 563
 564 static void
 565 xfs_flush_inodes_worker(
 566         struct work_struct      *work)
 567 {
 568         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 569                                                    m_flush_inodes_work);
 570         struct super_block      *sb = mp->m_super;
 571
 572         if (down_read_trylock(&sb->s_umount)) {
 573                 sync_inodes_sb(sb);
 574                 up_read(&sb->s_umount);
 575         }
 576 }
 577
 578 /*
 579  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 580  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 581  * for IO to complete so that we effectively throttle multiple callers to the
 582  * rate at which IO is completing.
 583  */
 584 void
 585 xfs_flush_inodes(
 586         struct xfs_mount        *mp)
 587 {
 588         /*
 589          * If flush_work() returns true then that means we waited for a flush
 590          * which was already in progress.  Don't bother running another scan.
 591          */
 592         if (flush_work(&mp->m_flush_inodes_work))
 593                 return;
 594
 595         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 596         flush_work(&mp->m_flush_inodes_work);
 597 }
 598
 599 /* Catch misguided souls that try to use this interface on XFS */
 600 STATIC struct inode *
 601 xfs_fs_alloc_inode(
 602         struct super_block      *sb)
 603 {
 604         BUG();
 605         return NULL;
 606 }
 607
 608 #ifdef DEBUG
 609 static void
 610 xfs_check_delalloc(
 611         struct xfs_inode        *ip,
 612         int                     whichfork)
 613 {
 614         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 615         struct xfs_bmbt_irec    got;
 616         struct xfs_iext_cursor  icur;
 617
 618         if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
 619                 return;
 620         do {
 621                 if (isnullstartblock(got.br_startblock)) {
 622                         xfs_warn(ip->i_mount,
 623         "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
 624                                 ip->i_ino,
 625                                 whichfork == XFS_DATA_FORK ? "data" : "cow",
 626                                 got.br_startoff, got.br_blockcount);
 627                 }
 628         } while (xfs_iext_next_extent(ifp, &icur, &got));
 629 }
 630 #else
 631 #define xfs_check_delalloc(ip, whichfork)       do { } while (0)
 632 #endif
 633
 634 /*
 635  * Now that the generic code is guaranteed not to be accessing
 636  * the linux inode, we can inactivate and reclaim the inode.
 637  */
 638 STATIC void
 639 xfs_fs_destroy_inode(
 640         struct inode            *inode)
 641 {
 642         struct xfs_inode        *ip = XFS_I(inode);
 643
 644         trace_xfs_destroy_inode(ip);
 645
 646         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 647         XFS_STATS_INC(ip->i_mount, vn_rele);
 648         XFS_STATS_INC(ip->i_mount, vn_remove);
 649
 650         xfs_inactive(ip);
 651
 652         if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
 653                 xfs_check_delalloc(ip, XFS_DATA_FORK);
 654                 xfs_check_delalloc(ip, XFS_COW_FORK);
 655                 ASSERT(0);
 656         }
 657
 658         XFS_STATS_INC(ip->i_mount, vn_reclaim);
 659
 660         /*
 661          * We should never get here with one of the reclaim flags already set.
 662          */
 663         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
 664         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
 665
 666         /*
 667          * We always use background reclaim here because even if the inode is
 668          * clean, it still may be under IO and hence we have wait for IO
 669          * completion to occur before we can reclaim the inode. The background
 670          * reclaim path handles this more efficiently than we can here, so
 671          * simply let background reclaim tear down all inodes.
 672          */
 673         xfs_inode_mark_reclaimable(ip);
 674 }
 675
 676 static void
 677 xfs_fs_dirty_inode(
 678         struct inode                    *inode,
 679         int                             flag)
 680 {
 681         struct xfs_inode                *ip = XFS_I(inode);
 682         struct xfs_mount                *mp = ip->i_mount;
 683         struct xfs_trans                *tp;
 684
 685         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 686                 return;
 687         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 688                 return;
 689
 690         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 691                 return;
 692         xfs_ilock(ip, XFS_ILOCK_EXCL);
 693         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 694         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 695         xfs_trans_commit(tp);
 696 }
 697
 698 /*
 699  * Slab object creation initialisation for the XFS inode.
 700  * This covers only the idempotent fields in the XFS inode;
 701  * all other fields need to be initialised on allocation
 702  * from the slab. This avoids the need to repeatedly initialise
 703  * fields in the xfs inode that left in the initialise state
 704  * when freeing the inode.
 705  */
 706 STATIC void
 707 xfs_fs_inode_init_once(
 708         void                    *inode)
 709 {
 710         struct xfs_inode        *ip = inode;
 711
 712         memset(ip, 0, sizeof(struct xfs_inode));
 713
 714         /* vfs inode */
 715         inode_init_once(VFS_I(ip));
 716
 717         /* xfs inode */
 718         atomic_set(&ip->i_pincount, 0);
 719         spin_lock_init(&ip->i_flags_lock);
 720
 721         mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 722                      "xfsino", ip->i_ino);
 723         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 724                      "xfsino", ip->i_ino);
 725 }
 726
 727 /*
 728  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 729  * serialised against cache hits here via the inode->i_lock and igrab() in
 730  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 731  * racing with us, and it avoids needing to grab a spinlock here for every inode
 732  * we drop the final reference on.
 733  */
 734 STATIC int
 735 xfs_fs_drop_inode(
 736         struct inode            *inode)
 737 {
 738         struct xfs_inode        *ip = XFS_I(inode);
 739
 740         /*
 741          * If this unlinked inode is in the middle of recovery, don't
 742          * drop the inode just yet; log recovery will take care of
 743          * that.  See the comment for this inode flag.
 744          */
 745         if (ip->i_flags & XFS_IRECOVERY) {
 746                 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
 747                 return 0;
 748         }
 749
 750         return generic_drop_inode(inode);
 751 }
 752
 753 static void
 754 xfs_mount_free(
 755         struct xfs_mount        *mp)
 756 {
 757         kfree(mp->m_rtname);
 758         kfree(mp->m_logname);
 759         kmem_free(mp);
 760 }
 761
 762 STATIC int
 763 xfs_fs_sync_fs(
 764         struct super_block      *sb,
 765         int                     wait)
 766 {
 767         struct xfs_mount        *mp = XFS_M(sb);
 768
 769         /*
 770          * Doing anything during the async pass would be counterproductive.
 771          */
 772         if (!wait)
 773                 return 0;
 774
 775         xfs_log_force(mp, XFS_LOG_SYNC);
 776         if (laptop_mode) {
 777                 /*
 778                  * The disk must be active because we're syncing.
 779                  * We schedule log work now (now that the disk is
 780                  * active) instead of later (when it might not be).
 781                  */
 782                 flush_delayed_work(&mp->m_log->l_work);
 783         }
 784
 785         return 0;
 786 }
 787
 788 STATIC int
 789 xfs_fs_statfs(
 790         struct dentry           *dentry,
 791         struct kstatfs          *statp)
 792 {
 793         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 794         xfs_sb_t                *sbp = &mp->m_sb;
 795         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 796         uint64_t                fakeinos, id;
 797         uint64_t                icount;
 798         uint64_t                ifree;
 799         uint64_t                fdblocks;
 800         xfs_extlen_t            lsize;
 801         int64_t                 ffree;
 802
 803         statp->f_type = XFS_SUPER_MAGIC;
 804         statp->f_namelen = MAXNAMELEN - 1;
 805
 806         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 807         statp->f_fsid = u64_to_fsid(id);
 808
 809         icount = percpu_counter_sum(&mp->m_icount);
 810         ifree = percpu_counter_sum(&mp->m_ifree);
 811         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 812
 813         spin_lock(&mp->m_sb_lock);
 814         statp->f_bsize = sbp->sb_blocksize;
 815         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 816         statp->f_blocks = sbp->sb_dblocks - lsize;
 817         spin_unlock(&mp->m_sb_lock);
 818
 819         /* make sure statp->f_bfree does not underflow */
 820         statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
 821         statp->f_bavail = statp->f_bfree;
 822
 823         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 824         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 825         if (M_IGEO(mp)->maxicount)
 826                 statp->f_files = min_t(typeof(statp->f_files),
 827                                         statp->f_files,
 828                                         M_IGEO(mp)->maxicount);
 829
 830         /* If sb_icount overshot maxicount, report actual allocation */
 831         statp->f_files = max_t(typeof(statp->f_files),
 832                                         statp->f_files,
 833                                         sbp->sb_icount);
 834
 835         /* make sure statp->f_ffree does not underflow */
 836         ffree = statp->f_files - (icount - ifree);
 837         statp->f_ffree = max_t(int64_t, ffree, 0);
 838
 839
 840         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 841             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 842                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 843                 xfs_qm_statvfs(ip, statp);
 844
 845         if (XFS_IS_REALTIME_MOUNT(mp) &&
 846             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 847                 statp->f_blocks = sbp->sb_rblocks;
 848                 statp->f_bavail = statp->f_bfree =
 849                         sbp->sb_frextents * sbp->sb_rextsize;
 850         }
 851
 852         return 0;
 853 }
 854
 855 STATIC void
 856 xfs_save_resvblks(struct xfs_mount *mp)
 857 {
 858         uint64_t resblks = 0;
 859
 860         mp->m_resblks_save = mp->m_resblks;
 861         xfs_reserve_blocks(mp, &resblks, NULL);
 862 }
 863
 864 STATIC void
 865 xfs_restore_resvblks(struct xfs_mount *mp)
 866 {
 867         uint64_t resblks;
 868
 869         if (mp->m_resblks_save) {
 870                 resblks = mp->m_resblks_save;
 871                 mp->m_resblks_save = 0;
 872         } else
 873                 resblks = xfs_default_resblks(mp);
 874
 875         xfs_reserve_blocks(mp, &resblks, NULL);
 876 }
 877
 878 /*
 879  * Second stage of a freeze. The data is already frozen so we only
 880  * need to take care of the metadata. Once that's done sync the superblock
 881  * to the log to dirty it in case of a crash while frozen. This ensures that we
 882  * will recover the unlinked inode lists on the next mount.
 883  */
 884 STATIC int
 885 xfs_fs_freeze(
 886         struct super_block      *sb)
 887 {
 888         struct xfs_mount        *mp = XFS_M(sb);
 889         unsigned int            flags;
 890         int                     ret;
 891
 892         /*
 893          * The filesystem is now frozen far enough that memory reclaim
 894          * cannot safely operate on the filesystem. Hence we need to
 895          * set a GFP_NOFS context here to avoid recursion deadlocks.
 896          */
 897         flags = memalloc_nofs_save();
 898         xfs_blockgc_stop(mp);
 899         xfs_save_resvblks(mp);
 900         ret = xfs_log_quiesce(mp);
 901         memalloc_nofs_restore(flags);
 902         return ret;
 903 }
 904
 905 STATIC int
 906 xfs_fs_unfreeze(
 907         struct super_block      *sb)
 908 {
 909         struct xfs_mount        *mp = XFS_M(sb);
 910
 911         xfs_restore_resvblks(mp);
 912         xfs_log_work_queue(mp);
 913         xfs_blockgc_start(mp);
 914         return 0;
 915 }
 916
 917 /*
 918  * This function fills in xfs_mount_t fields based on mount args.
 919  * Note: the superblock _has_ now been read in.
 920  */
 921 STATIC int
 922 xfs_finish_flags(
 923         struct xfs_mount        *mp)
 924 {
 925         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
 926
 927         /* Fail a mount where the logbuf is smaller than the log stripe */
 928         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
 929                 if (mp->m_logbsize <= 0 &&
 930                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 931                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 932                 } else if (mp->m_logbsize > 0 &&
 933                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 934                         xfs_warn(mp,
 935                 "logbuf size must be greater than or equal to log stripe size");
 936                         return -EINVAL;
 937                 }
 938         } else {
 939                 /* Fail a mount if the logbuf is larger than 32K */
 940                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 941                         xfs_warn(mp,
 942                 "logbuf size for version 1 logs must be 16K or 32K");
 943                         return -EINVAL;
 944                 }
 945         }
 946
 947         /*
 948          * V5 filesystems always use attr2 format for attributes.
 949          */
 950         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 951             (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 952                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 953                              "attr2 is always enabled for V5 filesystems.");
 954                 return -EINVAL;
 955         }
 956
 957         /*
 958          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
 959          * told by noattr2 to turn it off
 960          */
 961         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 962             !(mp->m_flags & XFS_MOUNT_NOATTR2))
 963                 mp->m_flags |= XFS_MOUNT_ATTR2;
 964
 965         /*
 966          * prohibit r/w mounts of read-only filesystems
 967          */
 968         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 969                 xfs_warn(mp,
 970                         "cannot mount a read-only filesystem as read-write");
 971                 return -EROFS;
 972         }
 973
 974         if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
 975             (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
 976             !xfs_sb_version_has_pquotino(&mp->m_sb)) {
 977                 xfs_warn(mp,
 978                   "Super block does not support project and group quota together");
 979                 return -EINVAL;
 980         }
 981
 982         return 0;
 983 }
 984
 985 static int
 986 xfs_init_percpu_counters(
 987         struct xfs_mount        *mp)
 988 {
 989         int             error;
 990
 991         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
 992         if (error)
 993                 return -ENOMEM;
 994
 995         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
 996         if (error)
 997                 goto free_icount;
 998
 999         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1000         if (error)
1001                 goto free_ifree;
1002
1003         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1004         if (error)
1005                 goto free_fdblocks;
1006
1007         return 0;
1008
1009 free_fdblocks:
1010         percpu_counter_destroy(&mp->m_fdblocks);
1011 free_ifree:
1012         percpu_counter_destroy(&mp->m_ifree);
1013 free_icount:
1014         percpu_counter_destroy(&mp->m_icount);
1015         return -ENOMEM;
1016 }
1017
1018 void
1019 xfs_reinit_percpu_counters(
1020         struct xfs_mount        *mp)
1021 {
1022         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1023         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1024         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1025 }
1026
1027 static void
1028 xfs_destroy_percpu_counters(
1029         struct xfs_mount        *mp)
1030 {
1031         percpu_counter_destroy(&mp->m_icount);
1032         percpu_counter_destroy(&mp->m_ifree);
1033         percpu_counter_destroy(&mp->m_fdblocks);
1034         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1035                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1036         percpu_counter_destroy(&mp->m_delalloc_blks);
1037 }
1038
1039 static void
1040 xfs_fs_put_super(
1041         struct super_block      *sb)
1042 {
1043         struct xfs_mount        *mp = XFS_M(sb);
1044
1045         /* if ->fill_super failed, we have no mount to tear down */
1046         if (!sb->s_fs_info)
1047                 return;
1048
1049         xfs_notice(mp, "Unmounting Filesystem");
1050         xfs_filestream_unmount(mp);
1051         xfs_unmountfs(mp);
1052
1053         xfs_freesb(mp);
1054         free_percpu(mp->m_stats.xs_stats);
1055         xfs_destroy_percpu_counters(mp);
1056         xfs_destroy_mount_workqueues(mp);
1057         xfs_close_devices(mp);
1058
1059         sb->s_fs_info = NULL;
1060         xfs_mount_free(mp);
1061 }
1062
1063 static long
1064 xfs_fs_nr_cached_objects(
1065         struct super_block      *sb,
1066         struct shrink_control   *sc)
1067 {
1068         /* Paranoia: catch incorrect calls during mount setup or teardown */
1069         if (WARN_ON_ONCE(!sb->s_fs_info))
1070                 return 0;
1071         return xfs_reclaim_inodes_count(XFS_M(sb));
1072 }
1073
1074 static long
1075 xfs_fs_free_cached_objects(
1076         struct super_block      *sb,
1077         struct shrink_control   *sc)
1078 {
1079         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1080 }
1081
1082 static const struct super_operations xfs_super_operations = {
1083         .alloc_inode            = xfs_fs_alloc_inode,
1084         .destroy_inode          = xfs_fs_destroy_inode,
1085         .dirty_inode            = xfs_fs_dirty_inode,
1086         .drop_inode             = xfs_fs_drop_inode,
1087         .put_super              = xfs_fs_put_super,
1088         .sync_fs                = xfs_fs_sync_fs,
1089         .freeze_fs              = xfs_fs_freeze,
1090         .unfreeze_fs            = xfs_fs_unfreeze,
1091         .statfs                 = xfs_fs_statfs,
1092         .show_options           = xfs_fs_show_options,
1093         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1094         .free_cached_objects    = xfs_fs_free_cached_objects,
1095 };
1096
1097 static int
1098 suffix_kstrtoint(
1099         const char      *s,
1100         unsigned int    base,
1101         int             *res)
1102 {
1103         int             last, shift_left_factor = 0, _res;
1104         char            *value;
1105         int             ret = 0;
1106
1107         value = kstrdup(s, GFP_KERNEL);
1108         if (!value)
1109                 return -ENOMEM;
1110
1111         last = strlen(value) - 1;
1112         if (value[last] == 'K' || value[last] == 'k') {
1113                 shift_left_factor = 10;
1114                 value[last] = '\0';
1115         }
1116         if (value[last] == 'M' || value[last] == 'm') {
1117                 shift_left_factor = 20;
1118                 value[last] = '\0';
1119         }
1120         if (value[last] == 'G' || value[last] == 'g') {
1121                 shift_left_factor = 30;
1122                 value[last] = '\0';
1123         }
1124
1125         if (kstrtoint(value, base, &_res))
1126                 ret = -EINVAL;
1127         kfree(value);
1128         *res = _res << shift_left_factor;
1129         return ret;
1130 }
1131
1132 static inline void
1133 xfs_fs_warn_deprecated(
1134         struct fs_context       *fc,
1135         struct fs_parameter     *param,
1136         uint64_t                flag,
1137         bool                    value)
1138 {
1139         /* Don't print the warning if reconfiguring and current mount point
1140          * already had the flag set
1141          */
1142         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1143                         !!(XFS_M(fc->root->d_sb)->m_flags & flag) == value)
1144                 return;
1145         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1146 }
1147
1148 /*
1149  * Set mount state from a mount option.
1150  *
1151  * NOTE: mp->m_super is NULL here!
1152  */
1153 static int
1154 xfs_fs_parse_param(
1155         struct fs_context       *fc,
1156         struct fs_parameter     *param)
1157 {
1158         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1159         struct fs_parse_result  result;
1160         int                     size = 0;
1161         int                     opt;
1162
1163         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1164         if (opt < 0)
1165                 return opt;
1166
1167         switch (opt) {
1168         case Opt_logbufs:
1169                 parsing_mp->m_logbufs = result.uint_32;
1170                 return 0;
1171         case Opt_logbsize:
1172                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1173                         return -EINVAL;
1174                 return 0;
1175         case Opt_logdev:
1176                 kfree(parsing_mp->m_logname);
1177                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1178                 if (!parsing_mp->m_logname)
1179                         return -ENOMEM;
1180                 return 0;
1181         case Opt_rtdev:
1182                 kfree(parsing_mp->m_rtname);
1183                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1184                 if (!parsing_mp->m_rtname)
1185                         return -ENOMEM;
1186                 return 0;
1187         case Opt_allocsize:
1188                 if (suffix_kstrtoint(param->string, 10, &size))
1189                         return -EINVAL;
1190                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1191                 parsing_mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1192                 return 0;
1193         case Opt_grpid:
1194         case Opt_bsdgroups:
1195                 parsing_mp->m_flags |= XFS_MOUNT_GRPID;
1196                 return 0;
1197         case Opt_nogrpid:
1198         case Opt_sysvgroups:
1199                 parsing_mp->m_flags &= ~XFS_MOUNT_GRPID;
1200                 return 0;
1201         case Opt_wsync:
1202                 parsing_mp->m_flags |= XFS_MOUNT_WSYNC;
1203                 return 0;
1204         case Opt_norecovery:
1205                 parsing_mp->m_flags |= XFS_MOUNT_NORECOVERY;
1206                 return 0;
1207         case Opt_noalign:
1208                 parsing_mp->m_flags |= XFS_MOUNT_NOALIGN;
1209                 return 0;
1210         case Opt_swalloc:
1211                 parsing_mp->m_flags |= XFS_MOUNT_SWALLOC;
1212                 return 0;
1213         case Opt_sunit:
1214                 parsing_mp->m_dalign = result.uint_32;
1215                 return 0;
1216         case Opt_swidth:
1217                 parsing_mp->m_swidth = result.uint_32;
1218                 return 0;
1219         case Opt_inode32:
1220                 parsing_mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1221                 return 0;
1222         case Opt_inode64:
1223                 parsing_mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1224                 return 0;
1225         case Opt_nouuid:
1226                 parsing_mp->m_flags |= XFS_MOUNT_NOUUID;
1227                 return 0;
1228         case Opt_largeio:
1229                 parsing_mp->m_flags |= XFS_MOUNT_LARGEIO;
1230                 return 0;
1231         case Opt_nolargeio:
1232                 parsing_mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1233                 return 0;
1234         case Opt_filestreams:
1235                 parsing_mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1236                 return 0;
1237         case Opt_noquota:
1238                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1239                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1240                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1241                 return 0;
1242         case Opt_quota:
1243         case Opt_uquota:
1244         case Opt_usrquota:
1245                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1246                                  XFS_UQUOTA_ENFD);
1247                 return 0;
1248         case Opt_qnoenforce:
1249         case Opt_uqnoenforce:
1250                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1251                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1252                 return 0;
1253         case Opt_pquota:
1254         case Opt_prjquota:
1255                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1256                                  XFS_PQUOTA_ENFD);
1257                 return 0;
1258         case Opt_pqnoenforce:
1259                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1260                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1261                 return 0;
1262         case Opt_gquota:
1263         case Opt_grpquota:
1264                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1265                                  XFS_GQUOTA_ENFD);
1266                 return 0;
1267         case Opt_gqnoenforce:
1268                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1269                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1270                 return 0;
1271         case Opt_discard:
1272                 parsing_mp->m_flags |= XFS_MOUNT_DISCARD;
1273                 return 0;
1274         case Opt_nodiscard:
1275                 parsing_mp->m_flags &= ~XFS_MOUNT_DISCARD;
1276                 return 0;
1277 #ifdef CONFIG_FS_DAX
1278         case Opt_dax:
1279                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1280                 return 0;
1281         case Opt_dax_enum:
1282                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1283                 return 0;
1284 #endif
1285         /* Following mount options will be removed in September 2025 */
1286         case Opt_ikeep:
1287                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, true);
1288                 parsing_mp->m_flags |= XFS_MOUNT_IKEEP;
1289                 return 0;
1290         case Opt_noikeep:
1291                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, false);
1292                 parsing_mp->m_flags &= ~XFS_MOUNT_IKEEP;
1293                 return 0;
1294         case Opt_attr2:
1295                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_ATTR2, true);
1296                 parsing_mp->m_flags |= XFS_MOUNT_ATTR2;
1297                 return 0;
1298         case Opt_noattr2:
1299                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_NOATTR2, true);
1300                 parsing_mp->m_flags &= ~XFS_MOUNT_ATTR2;
1301                 parsing_mp->m_flags |= XFS_MOUNT_NOATTR2;
1302                 return 0;
1303         default:
1304                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1305                 return -EINVAL;
1306         }
1307
1308         return 0;
1309 }
1310
1311 static int
1312 xfs_fs_validate_params(
1313         struct xfs_mount        *mp)
1314 {
1315         /*
1316          * no recovery flag requires a read-only mount
1317          */
1318         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1319             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1320                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1321                 return -EINVAL;
1322         }
1323
1324         if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1325             (mp->m_dalign || mp->m_swidth)) {
1326                 xfs_warn(mp,
1327         "sunit and swidth options incompatible with the noalign option");
1328                 return -EINVAL;
1329         }
1330
1331         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1332                 xfs_warn(mp, "quota support not available in this kernel.");
1333                 return -EINVAL;
1334         }
1335
1336         if ((mp->m_dalign && !mp->m_swidth) ||
1337             (!mp->m_dalign && mp->m_swidth)) {
1338                 xfs_warn(mp, "sunit and swidth must be specified together");
1339                 return -EINVAL;
1340         }
1341
1342         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1343                 xfs_warn(mp,
1344         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1345                         mp->m_swidth, mp->m_dalign);
1346                 return -EINVAL;
1347         }
1348
1349         if (mp->m_logbufs != -1 &&
1350             mp->m_logbufs != 0 &&
1351             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1352              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1353                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1354                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1355                 return -EINVAL;
1356         }
1357
1358         if (mp->m_logbsize != -1 &&
1359             mp->m_logbsize !=  0 &&
1360             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1361              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1362              !is_power_of_2(mp->m_logbsize))) {
1363                 xfs_warn(mp,
1364                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1365                         mp->m_logbsize);
1366                 return -EINVAL;
1367         }
1368
1369         if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1370             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1371              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1372                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1373                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1374                 return -EINVAL;
1375         }
1376
1377         return 0;
1378 }
1379
1380 static int
1381 xfs_fs_fill_super(
1382         struct super_block      *sb,
1383         struct fs_context       *fc)
1384 {
1385         struct xfs_mount        *mp = sb->s_fs_info;
1386         struct inode            *root;
1387         int                     flags = 0, error;
1388
1389         mp->m_super = sb;
1390
1391         error = xfs_fs_validate_params(mp);
1392         if (error)
1393                 goto out_free_names;
1394
1395         sb_min_blocksize(sb, BBSIZE);
1396         sb->s_xattr = xfs_xattr_handlers;
1397         sb->s_export_op = &xfs_export_operations;
1398 #ifdef CONFIG_XFS_QUOTA
1399         sb->s_qcop = &xfs_quotactl_operations;
1400         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1401 #endif
1402         sb->s_op = &xfs_super_operations;
1403
1404         /*
1405          * Delay mount work if the debug hook is set. This is debug
1406          * instrumention to coordinate simulation of xfs mount failures with
1407          * VFS superblock operations
1408          */
1409         if (xfs_globals.mount_delay) {
1410                 xfs_notice(mp, "Delaying mount for %d seconds.",
1411                         xfs_globals.mount_delay);
1412                 msleep(xfs_globals.mount_delay * 1000);
1413         }
1414
1415         if (fc->sb_flags & SB_SILENT)
1416                 flags |= XFS_MFSI_QUIET;
1417
1418         error = xfs_open_devices(mp);
1419         if (error)
1420                 goto out_free_names;
1421
1422         error = xfs_init_mount_workqueues(mp);
1423         if (error)
1424                 goto out_close_devices;
1425
1426         error = xfs_init_percpu_counters(mp);
1427         if (error)
1428                 goto out_destroy_workqueues;
1429
1430         /* Allocate stats memory before we do operations that might use it */
1431         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1432         if (!mp->m_stats.xs_stats) {
1433                 error = -ENOMEM;
1434                 goto out_destroy_counters;
1435         }
1436
1437         error = xfs_readsb(mp, flags);
1438         if (error)
1439                 goto out_free_stats;
1440
1441         error = xfs_finish_flags(mp);
1442         if (error)
1443                 goto out_free_sb;
1444
1445         error = xfs_setup_devices(mp);
1446         if (error)
1447                 goto out_free_sb;
1448
1449         /* V4 support is undergoing deprecation. */
1450         if (!xfs_sb_version_hascrc(&mp->m_sb)) {
1451 #ifdef CONFIG_XFS_SUPPORT_V4
1452                 xfs_warn_once(mp,
1453         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1454 #else
1455                 xfs_warn(mp,
1456         "Deprecated V4 format (crc=0) not supported by kernel.");
1457                 error = -EINVAL;
1458                 goto out_free_sb;
1459 #endif
1460         }
1461
1462         /* Filesystem claims it needs repair, so refuse the mount. */
1463         if (xfs_sb_version_needsrepair(&mp->m_sb)) {
1464                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1465                 error = -EFSCORRUPTED;
1466                 goto out_free_sb;
1467         }
1468
1469         /*
1470          * Don't touch the filesystem if a user tool thinks it owns the primary
1471          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1472          * we don't check them at all.
1473          */
1474         if (mp->m_sb.sb_inprogress) {
1475                 xfs_warn(mp, "Offline file system operation in progress!");
1476                 error = -EFSCORRUPTED;
1477                 goto out_free_sb;
1478         }
1479
1480         /*
1481          * Until this is fixed only page-sized or smaller data blocks work.
1482          */
1483         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1484                 xfs_warn(mp,
1485                 "File system with blocksize %d bytes. "
1486                 "Only pagesize (%ld) or less will currently work.",
1487                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1488                 error = -ENOSYS;
1489                 goto out_free_sb;
1490         }
1491
1492         /* Ensure this filesystem fits in the page cache limits */
1493         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1494             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1495                 xfs_warn(mp,
1496                 "file system too large to be mounted on this system.");
1497                 error = -EFBIG;
1498                 goto out_free_sb;
1499         }
1500
1501         /*
1502          * XFS block mappings use 54 bits to store the logical block offset.
1503          * This should suffice to handle the maximum file size that the VFS
1504          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1505          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1506          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1507          * to check this assertion.
1508          *
1509          * Avoid integer overflow by comparing the maximum bmbt offset to the
1510          * maximum pagecache offset in units of fs blocks.
1511          */
1512         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1513                 xfs_warn(mp,
1514 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1515                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1516                          XFS_MAX_FILEOFF);
1517                 error = -EINVAL;
1518                 goto out_free_sb;
1519         }
1520
1521         error = xfs_filestream_mount(mp);
1522         if (error)
1523                 goto out_free_sb;
1524
1525         /*
1526          * we must configure the block size in the superblock before we run the
1527          * full mount process as the mount process can lookup and cache inodes.
1528          */
1529         sb->s_magic = XFS_SUPER_MAGIC;
1530         sb->s_blocksize = mp->m_sb.sb_blocksize;
1531         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1532         sb->s_maxbytes = MAX_LFS_FILESIZE;
1533         sb->s_max_links = XFS_MAXLINK;
1534         sb->s_time_gran = 1;
1535         if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
1536                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1537                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1538         } else {
1539                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1540                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1541         }
1542         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1543         sb->s_iflags |= SB_I_CGROUPWB;
1544
1545         set_posix_acl_flag(sb);
1546
1547         /* version 5 superblocks support inode version counters. */
1548         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1549                 sb->s_flags |= SB_I_VERSION;
1550
1551         if (xfs_sb_version_hasbigtime(&mp->m_sb))
1552                 xfs_warn(mp,
1553  "EXPERIMENTAL big timestamp feature in use. Use at your own risk!");
1554
1555         if (mp->m_flags & XFS_MOUNT_DAX_ALWAYS) {
1556                 bool rtdev_is_dax = false, datadev_is_dax;
1557
1558                 xfs_warn(mp,
1559                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1560
1561                 datadev_is_dax = xfs_buftarg_is_dax(sb, mp->m_ddev_targp);
1562                 if (mp->m_rtdev_targp)
1563                         rtdev_is_dax = xfs_buftarg_is_dax(sb,
1564                                                 mp->m_rtdev_targp);
1565                 if (!rtdev_is_dax && !datadev_is_dax) {
1566                         xfs_alert(mp,
1567                         "DAX unsupported by block device. Turning off DAX.");
1568                         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1569                 }
1570                 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1571                         xfs_alert(mp,
1572                 "DAX and reflink cannot be used together!");
1573                         error = -EINVAL;
1574                         goto out_filestream_unmount;
1575                 }
1576         }
1577
1578         if (mp->m_flags & XFS_MOUNT_DISCARD) {
1579                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1580
1581                 if (!blk_queue_discard(q)) {
1582                         xfs_warn(mp, "mounting with \"discard\" option, but "
1583                                         "the device does not support discard");
1584                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
1585                 }
1586         }
1587
1588         if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1589                 if (mp->m_sb.sb_rblocks) {
1590                         xfs_alert(mp,
1591         "reflink not compatible with realtime device!");
1592                         error = -EINVAL;
1593                         goto out_filestream_unmount;
1594                 }
1595
1596                 if (xfs_globals.always_cow) {
1597                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1598                         mp->m_always_cow = true;
1599                 }
1600         }
1601
1602         if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1603                 xfs_alert(mp,
1604         "reverse mapping btree not compatible with realtime device!");
1605                 error = -EINVAL;
1606                 goto out_filestream_unmount;
1607         }
1608
1609         if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
1610                 xfs_warn(mp,
1611  "EXPERIMENTAL inode btree counters feature in use. Use at your own risk!");
1612
1613         error = xfs_mountfs(mp);
1614         if (error)
1615                 goto out_filestream_unmount;
1616
1617         root = igrab(VFS_I(mp->m_rootip));
1618         if (!root) {
1619                 error = -ENOENT;
1620                 goto out_unmount;
1621         }
1622         sb->s_root = d_make_root(root);
1623         if (!sb->s_root) {
1624                 error = -ENOMEM;
1625                 goto out_unmount;
1626         }
1627
1628         return 0;
1629
1630  out_filestream_unmount:
1631         xfs_filestream_unmount(mp);
1632  out_free_sb:
1633         xfs_freesb(mp);
1634  out_free_stats:
1635         free_percpu(mp->m_stats.xs_stats);
1636  out_destroy_counters:
1637         xfs_destroy_percpu_counters(mp);
1638  out_destroy_workqueues:
1639         xfs_destroy_mount_workqueues(mp);
1640  out_close_devices:
1641         xfs_close_devices(mp);
1642  out_free_names:
1643         sb->s_fs_info = NULL;
1644         xfs_mount_free(mp);
1645         return error;
1646
1647  out_unmount:
1648         xfs_filestream_unmount(mp);
1649         xfs_unmountfs(mp);
1650         goto out_free_sb;
1651 }
1652
1653 static int
1654 xfs_fs_get_tree(
1655         struct fs_context       *fc)
1656 {
1657         return get_tree_bdev(fc, xfs_fs_fill_super);
1658 }
1659
1660 static int
1661 xfs_remount_rw(
1662         struct xfs_mount        *mp)
1663 {
1664         struct xfs_sb           *sbp = &mp->m_sb;
1665         int error;
1666
1667         if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1668                 xfs_warn(mp,
1669                         "ro->rw transition prohibited on norecovery mount");
1670                 return -EINVAL;
1671         }
1672
1673         if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1674             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1675                 xfs_warn(mp,
1676         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1677                         (sbp->sb_features_ro_compat &
1678                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1679                 return -EINVAL;
1680         }
1681
1682         mp->m_flags &= ~XFS_MOUNT_RDONLY;
1683
1684         /*
1685          * If this is the first remount to writeable state we might have some
1686          * superblock changes to update.
1687          */
1688         if (mp->m_update_sb) {
1689                 error = xfs_sync_sb(mp, false);
1690                 if (error) {
1691                         xfs_warn(mp, "failed to write sb changes");
1692                         return error;
1693                 }
1694                 mp->m_update_sb = false;
1695         }
1696
1697         /*
1698          * Fill out the reserve pool if it is empty. Use the stashed value if
1699          * it is non-zero, otherwise go with the default.
1700          */
1701         xfs_restore_resvblks(mp);
1702         xfs_log_work_queue(mp);
1703
1704         /* Recover any CoW blocks that never got remapped. */
1705         error = xfs_reflink_recover_cow(mp);
1706         if (error) {
1707                 xfs_err(mp,
1708                         "Error %d recovering leftover CoW allocations.", error);
1709                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1710                 return error;
1711         }
1712         xfs_blockgc_start(mp);
1713
1714         /* Create the per-AG metadata reservation pool .*/
1715         error = xfs_fs_reserve_ag_blocks(mp);
1716         if (error && error != -ENOSPC)
1717                 return error;
1718
1719         return 0;
1720 }
1721
1722 static int
1723 xfs_remount_ro(
1724         struct xfs_mount        *mp)
1725 {
1726         int error;
1727
1728         /*
1729          * Cancel background eofb scanning so it cannot race with the final
1730          * log force+buftarg wait and deadlock the remount.
1731          */
1732         xfs_blockgc_stop(mp);
1733
1734         /* Get rid of any leftover CoW reservations... */
1735         error = xfs_blockgc_free_space(mp, NULL);
1736         if (error) {
1737                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1738                 return error;
1739         }
1740
1741         /* Free the per-AG metadata reservation pool. */
1742         error = xfs_fs_unreserve_ag_blocks(mp);
1743         if (error) {
1744                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1745                 return error;
1746         }
1747
1748         /*
1749          * Before we sync the metadata, we need to free up the reserve block
1750          * pool so that the used block count in the superblock on disk is
1751          * correct at the end of the remount. Stash the current* reserve pool
1752          * size so that if we get remounted rw, we can return it to the same
1753          * size.
1754          */
1755         xfs_save_resvblks(mp);
1756
1757         xfs_log_clean(mp);
1758         mp->m_flags |= XFS_MOUNT_RDONLY;
1759
1760         return 0;
1761 }
1762
1763 /*
1764  * Logically we would return an error here to prevent users from believing
1765  * they might have changed mount options using remount which can't be changed.
1766  *
1767  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1768  * arguments in some cases so we can't blindly reject options, but have to
1769  * check for each specified option if it actually differs from the currently
1770  * set option and only reject it if that's the case.
1771  *
1772  * Until that is implemented we return success for every remount request, and
1773  * silently ignore all options that we can't actually change.
1774  */
1775 static int
1776 xfs_fs_reconfigure(
1777         struct fs_context *fc)
1778 {
1779         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1780         struct xfs_mount        *new_mp = fc->s_fs_info;
1781         xfs_sb_t                *sbp = &mp->m_sb;
1782         int                     flags = fc->sb_flags;
1783         int                     error;
1784
1785         /* version 5 superblocks always support version counters. */
1786         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1787                 fc->sb_flags |= SB_I_VERSION;
1788
1789         error = xfs_fs_validate_params(new_mp);
1790         if (error)
1791                 return error;
1792
1793         sync_filesystem(mp->m_super);
1794
1795         /* inode32 -> inode64 */
1796         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1797             !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1798                 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1799                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1800         }
1801
1802         /* inode64 -> inode32 */
1803         if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1804             (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1805                 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1806                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1807         }
1808
1809         /* ro -> rw */
1810         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1811                 error = xfs_remount_rw(mp);
1812                 if (error)
1813                         return error;
1814         }
1815
1816         /* rw -> ro */
1817         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1818                 error = xfs_remount_ro(mp);
1819                 if (error)
1820                         return error;
1821         }
1822
1823         return 0;
1824 }
1825
1826 static void xfs_fs_free(
1827         struct fs_context       *fc)
1828 {
1829         struct xfs_mount        *mp = fc->s_fs_info;
1830
1831         /*
1832          * mp is stored in the fs_context when it is initialized.
1833          * mp is transferred to the superblock on a successful mount,
1834          * but if an error occurs before the transfer we have to free
1835          * it here.
1836          */
1837         if (mp)
1838                 xfs_mount_free(mp);
1839 }
1840
1841 static const struct fs_context_operations xfs_context_ops = {
1842         .parse_param = xfs_fs_parse_param,
1843         .get_tree    = xfs_fs_get_tree,
1844         .reconfigure = xfs_fs_reconfigure,
1845         .free        = xfs_fs_free,
1846 };
1847
1848 static int xfs_init_fs_context(
1849         struct fs_context       *fc)
1850 {
1851         struct xfs_mount        *mp;
1852
1853         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1854         if (!mp)
1855                 return -ENOMEM;
1856
1857         spin_lock_init(&mp->m_sb_lock);
1858         spin_lock_init(&mp->m_agirotor_lock);
1859         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1860         spin_lock_init(&mp->m_perag_lock);
1861         mutex_init(&mp->m_growlock);
1862         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1863         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1864         mp->m_kobj.kobject.kset = xfs_kset;
1865         /*
1866          * We don't create the finobt per-ag space reservation until after log
1867          * recovery, so we must set this to true so that an ifree transaction
1868          * started during log recovery will not depend on space reservations
1869          * for finobt expansion.
1870          */
1871         mp->m_finobt_nores = true;
1872
1873         /*
1874          * These can be overridden by the mount option parsing.
1875          */
1876         mp->m_logbufs = -1;
1877         mp->m_logbsize = -1;
1878         mp->m_allocsize_log = 16; /* 64k */
1879
1880         /*
1881          * Copy binary VFS mount flags we are interested in.
1882          */
1883         if (fc->sb_flags & SB_RDONLY)
1884                 mp->m_flags |= XFS_MOUNT_RDONLY;
1885         if (fc->sb_flags & SB_DIRSYNC)
1886                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1887         if (fc->sb_flags & SB_SYNCHRONOUS)
1888                 mp->m_flags |= XFS_MOUNT_WSYNC;
1889
1890         fc->s_fs_info = mp;
1891         fc->ops = &xfs_context_ops;
1892
1893         return 0;
1894 }
1895
1896 static struct file_system_type xfs_fs_type = {
1897         .owner                  = THIS_MODULE,
1898         .name                   = "xfs",
1899         .init_fs_context        = xfs_init_fs_context,
1900         .parameters             = xfs_fs_parameters,
1901         .kill_sb                = kill_block_super,
1902         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1903 };
1904 MODULE_ALIAS_FS("xfs");
1905
1906 STATIC int __init
1907 xfs_init_zones(void)
1908 {
1909         xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1910                                                 sizeof(struct xlog_ticket),
1911                                                 0, 0, NULL);
1912         if (!xfs_log_ticket_zone)
1913                 goto out;
1914
1915         xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1916                                         sizeof(struct xfs_extent_free_item),
1917                                         0, 0, NULL);
1918         if (!xfs_bmap_free_item_zone)
1919                 goto out_destroy_log_ticket_zone;
1920
1921         xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1922                                                sizeof(struct xfs_btree_cur),
1923                                                0, 0, NULL);
1924         if (!xfs_btree_cur_zone)
1925                 goto out_destroy_bmap_free_item_zone;
1926
1927         xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1928                                               sizeof(struct xfs_da_state),
1929                                               0, 0, NULL);
1930         if (!xfs_da_state_zone)
1931                 goto out_destroy_btree_cur_zone;
1932
1933         xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1934                                            sizeof(struct xfs_ifork),
1935                                            0, 0, NULL);
1936         if (!xfs_ifork_zone)
1937                 goto out_destroy_da_state_zone;
1938
1939         xfs_trans_zone = kmem_cache_create("xfs_trans",
1940                                            sizeof(struct xfs_trans),
1941                                            0, 0, NULL);
1942         if (!xfs_trans_zone)
1943                 goto out_destroy_ifork_zone;
1944
1945
1946         /*
1947          * The size of the zone allocated buf log item is the maximum
1948          * size possible under XFS.  This wastes a little bit of memory,
1949          * but it is much faster.
1950          */
1951         xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1952                                               sizeof(struct xfs_buf_log_item),
1953                                               0, 0, NULL);
1954         if (!xfs_buf_item_zone)
1955                 goto out_destroy_trans_zone;
1956
1957         xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1958                                         (sizeof(struct xfs_efd_log_item) +
1959                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1960                                         sizeof(struct xfs_extent)),
1961                                         0, 0, NULL);
1962         if (!xfs_efd_zone)
1963                 goto out_destroy_buf_item_zone;
1964
1965         xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1966                                          (sizeof(struct xfs_efi_log_item) +
1967                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1968                                          sizeof(struct xfs_extent)),
1969                                          0, 0, NULL);
1970         if (!xfs_efi_zone)
1971                 goto out_destroy_efd_zone;
1972
1973         xfs_inode_zone = kmem_cache_create("xfs_inode",
1974                                            sizeof(struct xfs_inode), 0,
1975                                            (SLAB_HWCACHE_ALIGN |
1976                                             SLAB_RECLAIM_ACCOUNT |
1977                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1978                                            xfs_fs_inode_init_once);
1979         if (!xfs_inode_zone)
1980                 goto out_destroy_efi_zone;
1981
1982         xfs_ili_zone = kmem_cache_create("xfs_ili",
1983                                          sizeof(struct xfs_inode_log_item), 0,
1984                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1985                                          NULL);
1986         if (!xfs_ili_zone)
1987                 goto out_destroy_inode_zone;
1988
1989         xfs_icreate_zone = kmem_cache_create("xfs_icr",
1990                                              sizeof(struct xfs_icreate_item),
1991                                              0, 0, NULL);
1992         if (!xfs_icreate_zone)
1993                 goto out_destroy_ili_zone;
1994
1995         xfs_rud_zone = kmem_cache_create("xfs_rud_item",
1996                                          sizeof(struct xfs_rud_log_item),
1997                                          0, 0, NULL);
1998         if (!xfs_rud_zone)
1999                 goto out_destroy_icreate_zone;
2000
2001         xfs_rui_zone = kmem_cache_create("xfs_rui_item",
2002                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2003                         0, 0, NULL);
2004         if (!xfs_rui_zone)
2005                 goto out_destroy_rud_zone;
2006
2007         xfs_cud_zone = kmem_cache_create("xfs_cud_item",
2008                                          sizeof(struct xfs_cud_log_item),
2009                                          0, 0, NULL);
2010         if (!xfs_cud_zone)
2011                 goto out_destroy_rui_zone;
2012
2013         xfs_cui_zone = kmem_cache_create("xfs_cui_item",
2014                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2015                         0, 0, NULL);
2016         if (!xfs_cui_zone)
2017                 goto out_destroy_cud_zone;
2018
2019         xfs_bud_zone = kmem_cache_create("xfs_bud_item",
2020                                          sizeof(struct xfs_bud_log_item),
2021                                          0, 0, NULL);
2022         if (!xfs_bud_zone)
2023                 goto out_destroy_cui_zone;
2024
2025         xfs_bui_zone = kmem_cache_create("xfs_bui_item",
2026                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2027                         0, 0, NULL);
2028         if (!xfs_bui_zone)
2029                 goto out_destroy_bud_zone;
2030
2031         return 0;
2032
2033  out_destroy_bud_zone:
2034         kmem_cache_destroy(xfs_bud_zone);
2035  out_destroy_cui_zone:
2036         kmem_cache_destroy(xfs_cui_zone);
2037  out_destroy_cud_zone:
2038         kmem_cache_destroy(xfs_cud_zone);
2039  out_destroy_rui_zone:
2040         kmem_cache_destroy(xfs_rui_zone);
2041  out_destroy_rud_zone:
2042         kmem_cache_destroy(xfs_rud_zone);
2043  out_destroy_icreate_zone:
2044         kmem_cache_destroy(xfs_icreate_zone);
2045  out_destroy_ili_zone:
2046         kmem_cache_destroy(xfs_ili_zone);
2047  out_destroy_inode_zone:
2048         kmem_cache_destroy(xfs_inode_zone);
2049  out_destroy_efi_zone:
2050         kmem_cache_destroy(xfs_efi_zone);
2051  out_destroy_efd_zone:
2052         kmem_cache_destroy(xfs_efd_zone);
2053  out_destroy_buf_item_zone:
2054         kmem_cache_destroy(xfs_buf_item_zone);
2055  out_destroy_trans_zone:
2056         kmem_cache_destroy(xfs_trans_zone);
2057  out_destroy_ifork_zone:
2058         kmem_cache_destroy(xfs_ifork_zone);
2059  out_destroy_da_state_zone:
2060         kmem_cache_destroy(xfs_da_state_zone);
2061  out_destroy_btree_cur_zone:
2062         kmem_cache_destroy(xfs_btree_cur_zone);
2063  out_destroy_bmap_free_item_zone:
2064         kmem_cache_destroy(xfs_bmap_free_item_zone);
2065  out_destroy_log_ticket_zone:
2066         kmem_cache_destroy(xfs_log_ticket_zone);
2067  out:
2068         return -ENOMEM;
2069 }
2070
2071 STATIC void
2072 xfs_destroy_zones(void)
2073 {
2074         /*
2075          * Make sure all delayed rcu free are flushed before we
2076          * destroy caches.
2077          */
2078         rcu_barrier();
2079         kmem_cache_destroy(xfs_bui_zone);
2080         kmem_cache_destroy(xfs_bud_zone);
2081         kmem_cache_destroy(xfs_cui_zone);
2082         kmem_cache_destroy(xfs_cud_zone);
2083         kmem_cache_destroy(xfs_rui_zone);
2084         kmem_cache_destroy(xfs_rud_zone);
2085         kmem_cache_destroy(xfs_icreate_zone);
2086         kmem_cache_destroy(xfs_ili_zone);
2087         kmem_cache_destroy(xfs_inode_zone);
2088         kmem_cache_destroy(xfs_efi_zone);
2089         kmem_cache_destroy(xfs_efd_zone);
2090         kmem_cache_destroy(xfs_buf_item_zone);
2091         kmem_cache_destroy(xfs_trans_zone);
2092         kmem_cache_destroy(xfs_ifork_zone);
2093         kmem_cache_destroy(xfs_da_state_zone);
2094         kmem_cache_destroy(xfs_btree_cur_zone);
2095         kmem_cache_destroy(xfs_bmap_free_item_zone);
2096         kmem_cache_destroy(xfs_log_ticket_zone);
2097 }
2098
2099 STATIC int __init
2100 xfs_init_workqueues(void)
2101 {
2102         /*
2103          * The allocation workqueue can be used in memory reclaim situations
2104          * (writepage path), and parallelism is only limited by the number of
2105          * AGs in all the filesystems mounted. Hence use the default large
2106          * max_active value for this workqueue.
2107          */
2108         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2109                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2110         if (!xfs_alloc_wq)
2111                 return -ENOMEM;
2112
2113         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2114                         0);
2115         if (!xfs_discard_wq)
2116                 goto out_free_alloc_wq;
2117
2118         return 0;
2119 out_free_alloc_wq:
2120         destroy_workqueue(xfs_alloc_wq);
2121         return -ENOMEM;
2122 }
2123
2124 STATIC void
2125 xfs_destroy_workqueues(void)
2126 {
2127         destroy_workqueue(xfs_discard_wq);
2128         destroy_workqueue(xfs_alloc_wq);
2129 }
2130
2131 STATIC int __init
2132 init_xfs_fs(void)
2133 {
2134         int                     error;
2135
2136         xfs_check_ondisk_structs();
2137
2138         printk(KERN_INFO XFS_VERSION_STRING " with "
2139                          XFS_BUILD_OPTIONS " enabled\n");
2140
2141         xfs_dir_startup();
2142
2143         error = xfs_init_zones();
2144         if (error)
2145                 goto out;
2146
2147         error = xfs_init_workqueues();
2148         if (error)
2149                 goto out_destroy_zones;
2150
2151         error = xfs_mru_cache_init();
2152         if (error)
2153                 goto out_destroy_wq;
2154
2155         error = xfs_buf_init();
2156         if (error)
2157                 goto out_mru_cache_uninit;
2158
2159         error = xfs_init_procfs();
2160         if (error)
2161                 goto out_buf_terminate;
2162
2163         error = xfs_sysctl_register();
2164         if (error)
2165                 goto out_cleanup_procfs;
2166
2167         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2168         if (!xfs_kset) {
2169                 error = -ENOMEM;
2170                 goto out_sysctl_unregister;
2171         }
2172
2173         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2174
2175         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2176         if (!xfsstats.xs_stats) {
2177                 error = -ENOMEM;
2178                 goto out_kset_unregister;
2179         }
2180
2181         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2182                                "stats");
2183         if (error)
2184                 goto out_free_stats;
2185
2186 #ifdef DEBUG
2187         xfs_dbg_kobj.kobject.kset = xfs_kset;
2188         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2189         if (error)
2190                 goto out_remove_stats_kobj;
2191 #endif
2192
2193         error = xfs_qm_init();
2194         if (error)
2195                 goto out_remove_dbg_kobj;
2196
2197         error = register_filesystem(&xfs_fs_type);
2198         if (error)
2199                 goto out_qm_exit;
2200         return 0;
2201
2202  out_qm_exit:
2203         xfs_qm_exit();
2204  out_remove_dbg_kobj:
2205 #ifdef DEBUG
2206         xfs_sysfs_del(&xfs_dbg_kobj);
2207  out_remove_stats_kobj:
2208 #endif
2209         xfs_sysfs_del(&xfsstats.xs_kobj);
2210  out_free_stats:
2211         free_percpu(xfsstats.xs_stats);
2212  out_kset_unregister:
2213         kset_unregister(xfs_kset);
2214  out_sysctl_unregister:
2215         xfs_sysctl_unregister();
2216  out_cleanup_procfs:
2217         xfs_cleanup_procfs();
2218  out_buf_terminate:
2219         xfs_buf_terminate();
2220  out_mru_cache_uninit:
2221         xfs_mru_cache_uninit();
2222  out_destroy_wq:
2223         xfs_destroy_workqueues();
2224  out_destroy_zones:
2225         xfs_destroy_zones();
2226  out:
2227         return error;
2228 }
2229
2230 STATIC void __exit
2231 exit_xfs_fs(void)
2232 {
2233         xfs_qm_exit();
2234         unregister_filesystem(&xfs_fs_type);
2235 #ifdef DEBUG
2236         xfs_sysfs_del(&xfs_dbg_kobj);
2237 #endif
2238         xfs_sysfs_del(&xfsstats.xs_kobj);
2239         free_percpu(xfsstats.xs_stats);
2240         kset_unregister(xfs_kset);
2241         xfs_sysctl_unregister();
2242         xfs_cleanup_procfs();
2243         xfs_buf_terminate();
2244         xfs_mru_cache_uninit();
2245         xfs_destroy_workqueues();
2246         xfs_destroy_zones();
2247         xfs_uuid_table_free();
2248 }
2249
2250 module_init(init_xfs_fs);
2251 module_exit(exit_xfs_fs);
2252
2253 MODULE_AUTHOR("Silicon Graphics, Inc.");
2254 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2255 MODULE_LICENSE("GPL");