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 #ifdef CONFIG_HOTPLUG_CPU
  53 static LIST_HEAD(xfs_mount_list);
  54 static DEFINE_SPINLOCK(xfs_mount_list_lock);
  55
  56 static inline void xfs_mount_list_add(struct xfs_mount *mp)
  57 {
  58         spin_lock(&xfs_mount_list_lock);
  59         list_add(&mp->m_mount_list, &xfs_mount_list);
  60         spin_unlock(&xfs_mount_list_lock);
  61 }
  62
  63 static inline void xfs_mount_list_del(struct xfs_mount *mp)
  64 {
  65         spin_lock(&xfs_mount_list_lock);
  66         list_del(&mp->m_mount_list);
  67         spin_unlock(&xfs_mount_list_lock);
  68 }
  69 #else /* !CONFIG_HOTPLUG_CPU */
  70 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
  71 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
  72 #endif
  73
  74 enum xfs_dax_mode {
  75         XFS_DAX_INODE = 0,
  76         XFS_DAX_ALWAYS = 1,
  77         XFS_DAX_NEVER = 2,
  78 };
  79
  80 static void
  81 xfs_mount_set_dax_mode(
  82         struct xfs_mount        *mp,
  83         enum xfs_dax_mode       mode)
  84 {
  85         switch (mode) {
  86         case XFS_DAX_INODE:
  87                 mp->m_flags &= ~(XFS_MOUNT_DAX_ALWAYS | XFS_MOUNT_DAX_NEVER);
  88                 break;
  89         case XFS_DAX_ALWAYS:
  90                 mp->m_flags |= XFS_MOUNT_DAX_ALWAYS;
  91                 mp->m_flags &= ~XFS_MOUNT_DAX_NEVER;
  92                 break;
  93         case XFS_DAX_NEVER:
  94                 mp->m_flags |= XFS_MOUNT_DAX_NEVER;
  95                 mp->m_flags &= ~XFS_MOUNT_DAX_ALWAYS;
  96                 break;
  97         }
  98 }
  99
 100 static const struct constant_table dax_param_enums[] = {
 101         {"inode",       XFS_DAX_INODE },
 102         {"always",      XFS_DAX_ALWAYS },
 103         {"never",       XFS_DAX_NEVER },
 104         {}
 105 };
 106
 107 /*
 108  * Table driven mount option parser.
 109  */
 110 enum {
 111         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
 112         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
 113         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
 114         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
 115         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
 116         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
 117         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
 118         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
 119         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
 120 };
 121
 122 static const struct fs_parameter_spec xfs_fs_parameters[] = {
 123         fsparam_u32("logbufs",          Opt_logbufs),
 124         fsparam_string("logbsize",      Opt_logbsize),
 125         fsparam_string("logdev",        Opt_logdev),
 126         fsparam_string("rtdev",         Opt_rtdev),
 127         fsparam_flag("wsync",           Opt_wsync),
 128         fsparam_flag("noalign",         Opt_noalign),
 129         fsparam_flag("swalloc",         Opt_swalloc),
 130         fsparam_u32("sunit",            Opt_sunit),
 131         fsparam_u32("swidth",           Opt_swidth),
 132         fsparam_flag("nouuid",          Opt_nouuid),
 133         fsparam_flag("grpid",           Opt_grpid),
 134         fsparam_flag("nogrpid",         Opt_nogrpid),
 135         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 136         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 137         fsparam_string("allocsize",     Opt_allocsize),
 138         fsparam_flag("norecovery",      Opt_norecovery),
 139         fsparam_flag("inode64",         Opt_inode64),
 140         fsparam_flag("inode32",         Opt_inode32),
 141         fsparam_flag("ikeep",           Opt_ikeep),
 142         fsparam_flag("noikeep",         Opt_noikeep),
 143         fsparam_flag("largeio",         Opt_largeio),
 144         fsparam_flag("nolargeio",       Opt_nolargeio),
 145         fsparam_flag("attr2",           Opt_attr2),
 146         fsparam_flag("noattr2",         Opt_noattr2),
 147         fsparam_flag("filestreams",     Opt_filestreams),
 148         fsparam_flag("quota",           Opt_quota),
 149         fsparam_flag("noquota",         Opt_noquota),
 150         fsparam_flag("usrquota",        Opt_usrquota),
 151         fsparam_flag("grpquota",        Opt_grpquota),
 152         fsparam_flag("prjquota",        Opt_prjquota),
 153         fsparam_flag("uquota",          Opt_uquota),
 154         fsparam_flag("gquota",          Opt_gquota),
 155         fsparam_flag("pquota",          Opt_pquota),
 156         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 157         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 158         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 159         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 160         fsparam_flag("discard",         Opt_discard),
 161         fsparam_flag("nodiscard",       Opt_nodiscard),
 162         fsparam_flag("dax",             Opt_dax),
 163         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 164         {}
 165 };
 166
 167 struct proc_xfs_info {
 168         uint64_t        flag;
 169         char            *str;
 170 };
 171
 172 static int
 173 xfs_fs_show_options(
 174         struct seq_file         *m,
 175         struct dentry           *root)
 176 {
 177         static struct proc_xfs_info xfs_info_set[] = {
 178                 /* the few simple ones we can get from the mount struct */
 179                 { XFS_MOUNT_IKEEP,              ",ikeep" },
 180                 { XFS_MOUNT_WSYNC,              ",wsync" },
 181                 { XFS_MOUNT_NOALIGN,            ",noalign" },
 182                 { XFS_MOUNT_SWALLOC,            ",swalloc" },
 183                 { XFS_MOUNT_NOUUID,             ",nouuid" },
 184                 { XFS_MOUNT_NORECOVERY,         ",norecovery" },
 185                 { XFS_MOUNT_ATTR2,              ",attr2" },
 186                 { XFS_MOUNT_FILESTREAMS,        ",filestreams" },
 187                 { XFS_MOUNT_GRPID,              ",grpid" },
 188                 { XFS_MOUNT_DISCARD,            ",discard" },
 189                 { XFS_MOUNT_LARGEIO,            ",largeio" },
 190                 { XFS_MOUNT_DAX_ALWAYS,         ",dax=always" },
 191                 { XFS_MOUNT_DAX_NEVER,          ",dax=never" },
 192                 { 0, NULL }
 193         };
 194         struct xfs_mount        *mp = XFS_M(root->d_sb);
 195         struct proc_xfs_info    *xfs_infop;
 196
 197         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 198                 if (mp->m_flags & xfs_infop->flag)
 199                         seq_puts(m, xfs_infop->str);
 200         }
 201
 202         seq_printf(m, ",inode%d",
 203                 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
 204
 205         if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
 206                 seq_printf(m, ",allocsize=%dk",
 207                            (1 << mp->m_allocsize_log) >> 10);
 208
 209         if (mp->m_logbufs > 0)
 210                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 211         if (mp->m_logbsize > 0)
 212                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 213
 214         if (mp->m_logname)
 215                 seq_show_option(m, "logdev", mp->m_logname);
 216         if (mp->m_rtname)
 217                 seq_show_option(m, "rtdev", mp->m_rtname);
 218
 219         if (mp->m_dalign > 0)
 220                 seq_printf(m, ",sunit=%d",
 221                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 222         if (mp->m_swidth > 0)
 223                 seq_printf(m, ",swidth=%d",
 224                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 225
 226         if (mp->m_qflags & XFS_UQUOTA_ENFD)
 227                 seq_puts(m, ",usrquota");
 228         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 229                 seq_puts(m, ",uqnoenforce");
 230
 231         if (mp->m_qflags & XFS_PQUOTA_ENFD)
 232                 seq_puts(m, ",prjquota");
 233         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
 234                 seq_puts(m, ",pqnoenforce");
 235
 236         if (mp->m_qflags & XFS_GQUOTA_ENFD)
 237                 seq_puts(m, ",grpquota");
 238         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
 239                 seq_puts(m, ",gqnoenforce");
 240
 241         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 242                 seq_puts(m, ",noquota");
 243
 244         return 0;
 245 }
 246
 247 /*
 248  * Set parameters for inode allocation heuristics, taking into account
 249  * filesystem size and inode32/inode64 mount options; i.e. specifically
 250  * whether or not XFS_MOUNT_SMALL_INUMS is set.
 251  *
 252  * Inode allocation patterns are altered only if inode32 is requested
 253  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
 254  * If altered, XFS_MOUNT_32BITINODES is set as well.
 255  *
 256  * An agcount independent of that in the mount structure is provided
 257  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 258  * to the potentially higher ag count.
 259  *
 260  * Returns the maximum AG index which may contain inodes.
 261  */
 262 xfs_agnumber_t
 263 xfs_set_inode_alloc(
 264         struct xfs_mount *mp,
 265         xfs_agnumber_t  agcount)
 266 {
 267         xfs_agnumber_t  index;
 268         xfs_agnumber_t  maxagi = 0;
 269         xfs_sb_t        *sbp = &mp->m_sb;
 270         xfs_agnumber_t  max_metadata;
 271         xfs_agino_t     agino;
 272         xfs_ino_t       ino;
 273
 274         /*
 275          * Calculate how much should be reserved for inodes to meet
 276          * the max inode percentage.  Used only for inode32.
 277          */
 278         if (M_IGEO(mp)->maxicount) {
 279                 uint64_t        icount;
 280
 281                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 282                 do_div(icount, 100);
 283                 icount += sbp->sb_agblocks - 1;
 284                 do_div(icount, sbp->sb_agblocks);
 285                 max_metadata = icount;
 286         } else {
 287                 max_metadata = agcount;
 288         }
 289
 290         /* Get the last possible inode in the filesystem */
 291         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 292         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 293
 294         /*
 295          * If user asked for no more than 32-bit inodes, and the fs is
 296          * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
 297          * the allocator to accommodate the request.
 298          */
 299         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
 300                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 301         else
 302                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
 303
 304         for (index = 0; index < agcount; index++) {
 305                 struct xfs_perag        *pag;
 306
 307                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 308
 309                 pag = xfs_perag_get(mp, index);
 310
 311                 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
 312                         if (ino > XFS_MAXINUMBER_32) {
 313                                 pag->pagi_inodeok = 0;
 314                                 pag->pagf_metadata = 0;
 315                         } else {
 316                                 pag->pagi_inodeok = 1;
 317                                 maxagi++;
 318                                 if (index < max_metadata)
 319                                         pag->pagf_metadata = 1;
 320                                 else
 321                                         pag->pagf_metadata = 0;
 322                         }
 323                 } else {
 324                         pag->pagi_inodeok = 1;
 325                         pag->pagf_metadata = 0;
 326                 }
 327
 328                 xfs_perag_put(pag);
 329         }
 330
 331         return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
 332 }
 333
 334 STATIC int
 335 xfs_blkdev_get(
 336         xfs_mount_t             *mp,
 337         const char              *name,
 338         struct block_device     **bdevp)
 339 {
 340         int                     error = 0;
 341
 342         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 343                                     mp);
 344         if (IS_ERR(*bdevp)) {
 345                 error = PTR_ERR(*bdevp);
 346                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 347         }
 348
 349         return error;
 350 }
 351
 352 STATIC void
 353 xfs_blkdev_put(
 354         struct block_device     *bdev)
 355 {
 356         if (bdev)
 357                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 358 }
 359
 360 STATIC void
 361 xfs_close_devices(
 362         struct xfs_mount        *mp)
 363 {
 364         struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
 365
 366         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 367                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 368                 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
 369
 370                 xfs_free_buftarg(mp->m_logdev_targp);
 371                 xfs_blkdev_put(logdev);
 372                 fs_put_dax(dax_logdev);
 373         }
 374         if (mp->m_rtdev_targp) {
 375                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 376                 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
 377
 378                 xfs_free_buftarg(mp->m_rtdev_targp);
 379                 xfs_blkdev_put(rtdev);
 380                 fs_put_dax(dax_rtdev);
 381         }
 382         xfs_free_buftarg(mp->m_ddev_targp);
 383         fs_put_dax(dax_ddev);
 384 }
 385
 386 /*
 387  * The file system configurations are:
 388  *      (1) device (partition) with data and internal log
 389  *      (2) logical volume with data and log subvolumes.
 390  *      (3) logical volume with data, log, and realtime subvolumes.
 391  *
 392  * We only have to handle opening the log and realtime volumes here if
 393  * they are present.  The data subvolume has already been opened by
 394  * get_sb_bdev() and is stored in sb->s_bdev.
 395  */
 396 STATIC int
 397 xfs_open_devices(
 398         struct xfs_mount        *mp)
 399 {
 400         struct block_device     *ddev = mp->m_super->s_bdev;
 401         struct dax_device       *dax_ddev = fs_dax_get_by_bdev(ddev);
 402         struct dax_device       *dax_logdev = NULL, *dax_rtdev = NULL;
 403         struct block_device     *logdev = NULL, *rtdev = NULL;
 404         int                     error;
 405
 406         /*
 407          * Open real time and log devices - order is important.
 408          */
 409         if (mp->m_logname) {
 410                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 411                 if (error)
 412                         goto out;
 413                 dax_logdev = fs_dax_get_by_bdev(logdev);
 414         }
 415
 416         if (mp->m_rtname) {
 417                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 418                 if (error)
 419                         goto out_close_logdev;
 420
 421                 if (rtdev == ddev || rtdev == logdev) {
 422                         xfs_warn(mp,
 423         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 424                         error = -EINVAL;
 425                         goto out_close_rtdev;
 426                 }
 427                 dax_rtdev = fs_dax_get_by_bdev(rtdev);
 428         }
 429
 430         /*
 431          * Setup xfs_mount buffer target pointers
 432          */
 433         error = -ENOMEM;
 434         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
 435         if (!mp->m_ddev_targp)
 436                 goto out_close_rtdev;
 437
 438         if (rtdev) {
 439                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
 440                 if (!mp->m_rtdev_targp)
 441                         goto out_free_ddev_targ;
 442         }
 443
 444         if (logdev && logdev != ddev) {
 445                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
 446                 if (!mp->m_logdev_targp)
 447                         goto out_free_rtdev_targ;
 448         } else {
 449                 mp->m_logdev_targp = mp->m_ddev_targp;
 450         }
 451
 452         return 0;
 453
 454  out_free_rtdev_targ:
 455         if (mp->m_rtdev_targp)
 456                 xfs_free_buftarg(mp->m_rtdev_targp);
 457  out_free_ddev_targ:
 458         xfs_free_buftarg(mp->m_ddev_targp);
 459  out_close_rtdev:
 460         xfs_blkdev_put(rtdev);
 461         fs_put_dax(dax_rtdev);
 462  out_close_logdev:
 463         if (logdev && logdev != ddev) {
 464                 xfs_blkdev_put(logdev);
 465                 fs_put_dax(dax_logdev);
 466         }
 467  out:
 468         fs_put_dax(dax_ddev);
 469         return error;
 470 }
 471
 472 /*
 473  * Setup xfs_mount buffer target pointers based on superblock
 474  */
 475 STATIC int
 476 xfs_setup_devices(
 477         struct xfs_mount        *mp)
 478 {
 479         int                     error;
 480
 481         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 482         if (error)
 483                 return error;
 484
 485         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 486                 unsigned int    log_sector_size = BBSIZE;
 487
 488                 if (xfs_sb_version_hassector(&mp->m_sb))
 489                         log_sector_size = mp->m_sb.sb_logsectsize;
 490                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 491                                             log_sector_size);
 492                 if (error)
 493                         return error;
 494         }
 495         if (mp->m_rtdev_targp) {
 496                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 497                                             mp->m_sb.sb_sectsize);
 498                 if (error)
 499                         return error;
 500         }
 501
 502         return 0;
 503 }
 504
 505 STATIC int
 506 xfs_init_mount_workqueues(
 507         struct xfs_mount        *mp)
 508 {
 509         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 510                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 511                         1, mp->m_super->s_id);
 512         if (!mp->m_buf_workqueue)
 513                 goto out;
 514
 515         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 516                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 517                         0, mp->m_super->s_id);
 518         if (!mp->m_unwritten_workqueue)
 519                 goto out_destroy_buf;
 520
 521         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 522                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 523                         0, mp->m_super->s_id);
 524         if (!mp->m_reclaim_workqueue)
 525                 goto out_destroy_unwritten;
 526
 527         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
 528                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
 529                         0, mp->m_super->s_id);
 530         if (!mp->m_blockgc_wq)
 531                 goto out_destroy_reclaim;
 532
 533         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
 534                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 535                         1, mp->m_super->s_id);
 536         if (!mp->m_inodegc_wq)
 537                 goto out_destroy_blockgc;
 538
 539         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 540                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 541         if (!mp->m_sync_workqueue)
 542                 goto out_destroy_inodegc;
 543
 544         return 0;
 545
 546 out_destroy_inodegc:
 547         destroy_workqueue(mp->m_inodegc_wq);
 548 out_destroy_blockgc:
 549         destroy_workqueue(mp->m_blockgc_wq);
 550 out_destroy_reclaim:
 551         destroy_workqueue(mp->m_reclaim_workqueue);
 552 out_destroy_unwritten:
 553         destroy_workqueue(mp->m_unwritten_workqueue);
 554 out_destroy_buf:
 555         destroy_workqueue(mp->m_buf_workqueue);
 556 out:
 557         return -ENOMEM;
 558 }
 559
 560 STATIC void
 561 xfs_destroy_mount_workqueues(
 562         struct xfs_mount        *mp)
 563 {
 564         destroy_workqueue(mp->m_sync_workqueue);
 565         destroy_workqueue(mp->m_blockgc_wq);
 566         destroy_workqueue(mp->m_inodegc_wq);
 567         destroy_workqueue(mp->m_reclaim_workqueue);
 568         destroy_workqueue(mp->m_unwritten_workqueue);
 569         destroy_workqueue(mp->m_buf_workqueue);
 570 }
 571
 572 static void
 573 xfs_flush_inodes_worker(
 574         struct work_struct      *work)
 575 {
 576         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 577                                                    m_flush_inodes_work);
 578         struct super_block      *sb = mp->m_super;
 579
 580         if (down_read_trylock(&sb->s_umount)) {
 581                 sync_inodes_sb(sb);
 582                 up_read(&sb->s_umount);
 583         }
 584 }
 585
 586 /*
 587  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 588  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 589  * for IO to complete so that we effectively throttle multiple callers to the
 590  * rate at which IO is completing.
 591  */
 592 void
 593 xfs_flush_inodes(
 594         struct xfs_mount        *mp)
 595 {
 596         /*
 597          * If flush_work() returns true then that means we waited for a flush
 598          * which was already in progress.  Don't bother running another scan.
 599          */
 600         if (flush_work(&mp->m_flush_inodes_work))
 601                 return;
 602
 603         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 604         flush_work(&mp->m_flush_inodes_work);
 605 }
 606
 607 /* Catch misguided souls that try to use this interface on XFS */
 608 STATIC struct inode *
 609 xfs_fs_alloc_inode(
 610         struct super_block      *sb)
 611 {
 612         BUG();
 613         return NULL;
 614 }
 615
 616 /*
 617  * Now that the generic code is guaranteed not to be accessing
 618  * the linux inode, we can inactivate and reclaim the inode.
 619  */
 620 STATIC void
 621 xfs_fs_destroy_inode(
 622         struct inode            *inode)
 623 {
 624         struct xfs_inode        *ip = XFS_I(inode);
 625
 626         trace_xfs_destroy_inode(ip);
 627
 628         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 629         XFS_STATS_INC(ip->i_mount, vn_rele);
 630         XFS_STATS_INC(ip->i_mount, vn_remove);
 631         xfs_inode_mark_reclaimable(ip);
 632 }
 633
 634 static void
 635 xfs_fs_dirty_inode(
 636         struct inode                    *inode,
 637         int                             flag)
 638 {
 639         struct xfs_inode                *ip = XFS_I(inode);
 640         struct xfs_mount                *mp = ip->i_mount;
 641         struct xfs_trans                *tp;
 642
 643         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 644                 return;
 645         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 646                 return;
 647
 648         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 649                 return;
 650         xfs_ilock(ip, XFS_ILOCK_EXCL);
 651         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 652         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 653         xfs_trans_commit(tp);
 654 }
 655
 656 /*
 657  * Slab object creation initialisation for the XFS inode.
 658  * This covers only the idempotent fields in the XFS inode;
 659  * all other fields need to be initialised on allocation
 660  * from the slab. This avoids the need to repeatedly initialise
 661  * fields in the xfs inode that left in the initialise state
 662  * when freeing the inode.
 663  */
 664 STATIC void
 665 xfs_fs_inode_init_once(
 666         void                    *inode)
 667 {
 668         struct xfs_inode        *ip = inode;
 669
 670         memset(ip, 0, sizeof(struct xfs_inode));
 671
 672         /* vfs inode */
 673         inode_init_once(VFS_I(ip));
 674
 675         /* xfs inode */
 676         atomic_set(&ip->i_pincount, 0);
 677         spin_lock_init(&ip->i_flags_lock);
 678
 679         mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 680                      "xfsino", ip->i_ino);
 681         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 682                      "xfsino", ip->i_ino);
 683 }
 684
 685 /*
 686  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 687  * serialised against cache hits here via the inode->i_lock and igrab() in
 688  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 689  * racing with us, and it avoids needing to grab a spinlock here for every inode
 690  * we drop the final reference on.
 691  */
 692 STATIC int
 693 xfs_fs_drop_inode(
 694         struct inode            *inode)
 695 {
 696         struct xfs_inode        *ip = XFS_I(inode);
 697
 698         /*
 699          * If this unlinked inode is in the middle of recovery, don't
 700          * drop the inode just yet; log recovery will take care of
 701          * that.  See the comment for this inode flag.
 702          */
 703         if (ip->i_flags & XFS_IRECOVERY) {
 704                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
 705                 return 0;
 706         }
 707
 708         return generic_drop_inode(inode);
 709 }
 710
 711 static void
 712 xfs_mount_free(
 713         struct xfs_mount        *mp)
 714 {
 715         kfree(mp->m_rtname);
 716         kfree(mp->m_logname);
 717         kmem_free(mp);
 718 }
 719
 720 STATIC int
 721 xfs_fs_sync_fs(
 722         struct super_block      *sb,
 723         int                     wait)
 724 {
 725         struct xfs_mount        *mp = XFS_M(sb);
 726
 727         trace_xfs_fs_sync_fs(mp, __return_address);
 728
 729         /*
 730          * Doing anything during the async pass would be counterproductive.
 731          */
 732         if (!wait)
 733                 return 0;
 734
 735         xfs_log_force(mp, XFS_LOG_SYNC);
 736         if (laptop_mode) {
 737                 /*
 738                  * The disk must be active because we're syncing.
 739                  * We schedule log work now (now that the disk is
 740                  * active) instead of later (when it might not be).
 741                  */
 742                 flush_delayed_work(&mp->m_log->l_work);
 743         }
 744
 745         /*
 746          * If we are called with page faults frozen out, it means we are about
 747          * to freeze the transaction subsystem. Take the opportunity to shut
 748          * down inodegc because once SB_FREEZE_FS is set it's too late to
 749          * prevent inactivation races with freeze. The fs doesn't get called
 750          * again by the freezing process until after SB_FREEZE_FS has been set,
 751          * so it's now or never.  Same logic applies to speculative allocation
 752          * garbage collection.
 753          *
 754          * We don't care if this is a normal syncfs call that does this or
 755          * freeze that does this - we can run this multiple times without issue
 756          * and we won't race with a restart because a restart can only occur
 757          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
 758          */
 759         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
 760                 xfs_inodegc_stop(mp);
 761                 xfs_blockgc_stop(mp);
 762         }
 763
 764         return 0;
 765 }
 766
 767 STATIC int
 768 xfs_fs_statfs(
 769         struct dentry           *dentry,
 770         struct kstatfs          *statp)
 771 {
 772         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 773         xfs_sb_t                *sbp = &mp->m_sb;
 774         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 775         uint64_t                fakeinos, id;
 776         uint64_t                icount;
 777         uint64_t                ifree;
 778         uint64_t                fdblocks;
 779         xfs_extlen_t            lsize;
 780         int64_t                 ffree;
 781
 782         /* Wait for whatever inactivations are in progress. */
 783         xfs_inodegc_flush(mp);
 784
 785         statp->f_type = XFS_SUPER_MAGIC;
 786         statp->f_namelen = MAXNAMELEN - 1;
 787
 788         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 789         statp->f_fsid = u64_to_fsid(id);
 790
 791         icount = percpu_counter_sum(&mp->m_icount);
 792         ifree = percpu_counter_sum(&mp->m_ifree);
 793         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 794
 795         spin_lock(&mp->m_sb_lock);
 796         statp->f_bsize = sbp->sb_blocksize;
 797         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 798         statp->f_blocks = sbp->sb_dblocks - lsize;
 799         spin_unlock(&mp->m_sb_lock);
 800
 801         /* make sure statp->f_bfree does not underflow */
 802         statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
 803         statp->f_bavail = statp->f_bfree;
 804
 805         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 806         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 807         if (M_IGEO(mp)->maxicount)
 808                 statp->f_files = min_t(typeof(statp->f_files),
 809                                         statp->f_files,
 810                                         M_IGEO(mp)->maxicount);
 811
 812         /* If sb_icount overshot maxicount, report actual allocation */
 813         statp->f_files = max_t(typeof(statp->f_files),
 814                                         statp->f_files,
 815                                         sbp->sb_icount);
 816
 817         /* make sure statp->f_ffree does not underflow */
 818         ffree = statp->f_files - (icount - ifree);
 819         statp->f_ffree = max_t(int64_t, ffree, 0);
 820
 821
 822         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 823             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 824                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 825                 xfs_qm_statvfs(ip, statp);
 826
 827         if (XFS_IS_REALTIME_MOUNT(mp) &&
 828             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 829                 statp->f_blocks = sbp->sb_rblocks;
 830                 statp->f_bavail = statp->f_bfree =
 831                         sbp->sb_frextents * sbp->sb_rextsize;
 832         }
 833
 834         return 0;
 835 }
 836
 837 STATIC void
 838 xfs_save_resvblks(struct xfs_mount *mp)
 839 {
 840         uint64_t resblks = 0;
 841
 842         mp->m_resblks_save = mp->m_resblks;
 843         xfs_reserve_blocks(mp, &resblks, NULL);
 844 }
 845
 846 STATIC void
 847 xfs_restore_resvblks(struct xfs_mount *mp)
 848 {
 849         uint64_t resblks;
 850
 851         if (mp->m_resblks_save) {
 852                 resblks = mp->m_resblks_save;
 853                 mp->m_resblks_save = 0;
 854         } else
 855                 resblks = xfs_default_resblks(mp);
 856
 857         xfs_reserve_blocks(mp, &resblks, NULL);
 858 }
 859
 860 /*
 861  * Second stage of a freeze. The data is already frozen so we only
 862  * need to take care of the metadata. Once that's done sync the superblock
 863  * to the log to dirty it in case of a crash while frozen. This ensures that we
 864  * will recover the unlinked inode lists on the next mount.
 865  */
 866 STATIC int
 867 xfs_fs_freeze(
 868         struct super_block      *sb)
 869 {
 870         struct xfs_mount        *mp = XFS_M(sb);
 871         unsigned int            flags;
 872         int                     ret;
 873
 874         /*
 875          * The filesystem is now frozen far enough that memory reclaim
 876          * cannot safely operate on the filesystem. Hence we need to
 877          * set a GFP_NOFS context here to avoid recursion deadlocks.
 878          */
 879         flags = memalloc_nofs_save();
 880         xfs_save_resvblks(mp);
 881         ret = xfs_log_quiesce(mp);
 882         memalloc_nofs_restore(flags);
 883
 884         /*
 885          * For read-write filesystems, we need to restart the inodegc on error
 886          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
 887          * going to be run to restart it now.  We are at SB_FREEZE_FS level
 888          * here, so we can restart safely without racing with a stop in
 889          * xfs_fs_sync_fs().
 890          */
 891         if (ret && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
 892                 xfs_blockgc_start(mp);
 893                 xfs_inodegc_start(mp);
 894         }
 895
 896         return ret;
 897 }
 898
 899 STATIC int
 900 xfs_fs_unfreeze(
 901         struct super_block      *sb)
 902 {
 903         struct xfs_mount        *mp = XFS_M(sb);
 904
 905         xfs_restore_resvblks(mp);
 906         xfs_log_work_queue(mp);
 907
 908         /*
 909          * Don't reactivate the inodegc worker on a readonly filesystem because
 910          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
 911          * worker because there are no speculative preallocations on a readonly
 912          * filesystem.
 913          */
 914         if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 915                 xfs_blockgc_start(mp);
 916                 xfs_inodegc_start(mp);
 917         }
 918
 919         return 0;
 920 }
 921
 922 /*
 923  * This function fills in xfs_mount_t fields based on mount args.
 924  * Note: the superblock _has_ now been read in.
 925  */
 926 STATIC int
 927 xfs_finish_flags(
 928         struct xfs_mount        *mp)
 929 {
 930         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
 931
 932         /* Fail a mount where the logbuf is smaller than the log stripe */
 933         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
 934                 if (mp->m_logbsize <= 0 &&
 935                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 936                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 937                 } else if (mp->m_logbsize > 0 &&
 938                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 939                         xfs_warn(mp,
 940                 "logbuf size must be greater than or equal to log stripe size");
 941                         return -EINVAL;
 942                 }
 943         } else {
 944                 /* Fail a mount if the logbuf is larger than 32K */
 945                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 946                         xfs_warn(mp,
 947                 "logbuf size for version 1 logs must be 16K or 32K");
 948                         return -EINVAL;
 949                 }
 950         }
 951
 952         /*
 953          * V5 filesystems always use attr2 format for attributes.
 954          */
 955         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 956             (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 957                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 958                              "attr2 is always enabled for V5 filesystems.");
 959                 return -EINVAL;
 960         }
 961
 962         /*
 963          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
 964          * told by noattr2 to turn it off
 965          */
 966         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 967             !(mp->m_flags & XFS_MOUNT_NOATTR2))
 968                 mp->m_flags |= XFS_MOUNT_ATTR2;
 969
 970         /*
 971          * prohibit r/w mounts of read-only filesystems
 972          */
 973         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 974                 xfs_warn(mp,
 975                         "cannot mount a read-only filesystem as read-write");
 976                 return -EROFS;
 977         }
 978
 979         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
 980             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
 981             !xfs_sb_version_has_pquotino(&mp->m_sb)) {
 982                 xfs_warn(mp,
 983                   "Super block does not support project and group quota together");
 984                 return -EINVAL;
 985         }
 986
 987         return 0;
 988 }
 989
 990 static int
 991 xfs_init_percpu_counters(
 992         struct xfs_mount        *mp)
 993 {
 994         int             error;
 995
 996         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
 997         if (error)
 998                 return -ENOMEM;
 999
1000         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1001         if (error)
1002                 goto free_icount;
1003
1004         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1005         if (error)
1006                 goto free_ifree;
1007
1008         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1009         if (error)
1010                 goto free_fdblocks;
1011
1012         return 0;
1013
1014 free_fdblocks:
1015         percpu_counter_destroy(&mp->m_fdblocks);
1016 free_ifree:
1017         percpu_counter_destroy(&mp->m_ifree);
1018 free_icount:
1019         percpu_counter_destroy(&mp->m_icount);
1020         return -ENOMEM;
1021 }
1022
1023 void
1024 xfs_reinit_percpu_counters(
1025         struct xfs_mount        *mp)
1026 {
1027         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1028         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1029         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1030 }
1031
1032 static void
1033 xfs_destroy_percpu_counters(
1034         struct xfs_mount        *mp)
1035 {
1036         percpu_counter_destroy(&mp->m_icount);
1037         percpu_counter_destroy(&mp->m_ifree);
1038         percpu_counter_destroy(&mp->m_fdblocks);
1039         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1040                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1041         percpu_counter_destroy(&mp->m_delalloc_blks);
1042 }
1043
1044 static int
1045 xfs_inodegc_init_percpu(
1046         struct xfs_mount        *mp)
1047 {
1048         struct xfs_inodegc      *gc;
1049         int                     cpu;
1050
1051         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1052         if (!mp->m_inodegc)
1053                 return -ENOMEM;
1054
1055         for_each_possible_cpu(cpu) {
1056                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1057                 init_llist_head(&gc->list);
1058                 gc->items = 0;
1059                 INIT_WORK(&gc->work, xfs_inodegc_worker);
1060         }
1061         return 0;
1062 }
1063
1064 static void
1065 xfs_inodegc_free_percpu(
1066         struct xfs_mount        *mp)
1067 {
1068         if (!mp->m_inodegc)
1069                 return;
1070         free_percpu(mp->m_inodegc);
1071 }
1072
1073 static void
1074 xfs_fs_put_super(
1075         struct super_block      *sb)
1076 {
1077         struct xfs_mount        *mp = XFS_M(sb);
1078
1079         /* if ->fill_super failed, we have no mount to tear down */
1080         if (!sb->s_fs_info)
1081                 return;
1082
1083         xfs_notice(mp, "Unmounting Filesystem");
1084         xfs_filestream_unmount(mp);
1085         xfs_unmountfs(mp);
1086
1087         xfs_freesb(mp);
1088         free_percpu(mp->m_stats.xs_stats);
1089         xfs_mount_list_del(mp);
1090         xfs_inodegc_free_percpu(mp);
1091         xfs_destroy_percpu_counters(mp);
1092         xfs_destroy_mount_workqueues(mp);
1093         xfs_close_devices(mp);
1094
1095         sb->s_fs_info = NULL;
1096         xfs_mount_free(mp);
1097 }
1098
1099 static long
1100 xfs_fs_nr_cached_objects(
1101         struct super_block      *sb,
1102         struct shrink_control   *sc)
1103 {
1104         /* Paranoia: catch incorrect calls during mount setup or teardown */
1105         if (WARN_ON_ONCE(!sb->s_fs_info))
1106                 return 0;
1107         return xfs_reclaim_inodes_count(XFS_M(sb));
1108 }
1109
1110 static long
1111 xfs_fs_free_cached_objects(
1112         struct super_block      *sb,
1113         struct shrink_control   *sc)
1114 {
1115         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1116 }
1117
1118 static const struct super_operations xfs_super_operations = {
1119         .alloc_inode            = xfs_fs_alloc_inode,
1120         .destroy_inode          = xfs_fs_destroy_inode,
1121         .dirty_inode            = xfs_fs_dirty_inode,
1122         .drop_inode             = xfs_fs_drop_inode,
1123         .put_super              = xfs_fs_put_super,
1124         .sync_fs                = xfs_fs_sync_fs,
1125         .freeze_fs              = xfs_fs_freeze,
1126         .unfreeze_fs            = xfs_fs_unfreeze,
1127         .statfs                 = xfs_fs_statfs,
1128         .show_options           = xfs_fs_show_options,
1129         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1130         .free_cached_objects    = xfs_fs_free_cached_objects,
1131 };
1132
1133 static int
1134 suffix_kstrtoint(
1135         const char      *s,
1136         unsigned int    base,
1137         int             *res)
1138 {
1139         int             last, shift_left_factor = 0, _res;
1140         char            *value;
1141         int             ret = 0;
1142
1143         value = kstrdup(s, GFP_KERNEL);
1144         if (!value)
1145                 return -ENOMEM;
1146
1147         last = strlen(value) - 1;
1148         if (value[last] == 'K' || value[last] == 'k') {
1149                 shift_left_factor = 10;
1150                 value[last] = '\0';
1151         }
1152         if (value[last] == 'M' || value[last] == 'm') {
1153                 shift_left_factor = 20;
1154                 value[last] = '\0';
1155         }
1156         if (value[last] == 'G' || value[last] == 'g') {
1157                 shift_left_factor = 30;
1158                 value[last] = '\0';
1159         }
1160
1161         if (kstrtoint(value, base, &_res))
1162                 ret = -EINVAL;
1163         kfree(value);
1164         *res = _res << shift_left_factor;
1165         return ret;
1166 }
1167
1168 static inline void
1169 xfs_fs_warn_deprecated(
1170         struct fs_context       *fc,
1171         struct fs_parameter     *param,
1172         uint64_t                flag,
1173         bool                    value)
1174 {
1175         /* Don't print the warning if reconfiguring and current mount point
1176          * already had the flag set
1177          */
1178         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1179                         !!(XFS_M(fc->root->d_sb)->m_flags & flag) == value)
1180                 return;
1181         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1182 }
1183
1184 /*
1185  * Set mount state from a mount option.
1186  *
1187  * NOTE: mp->m_super is NULL here!
1188  */
1189 static int
1190 xfs_fs_parse_param(
1191         struct fs_context       *fc,
1192         struct fs_parameter     *param)
1193 {
1194         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1195         struct fs_parse_result  result;
1196         int                     size = 0;
1197         int                     opt;
1198
1199         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1200         if (opt < 0)
1201                 return opt;
1202
1203         switch (opt) {
1204         case Opt_logbufs:
1205                 parsing_mp->m_logbufs = result.uint_32;
1206                 return 0;
1207         case Opt_logbsize:
1208                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1209                         return -EINVAL;
1210                 return 0;
1211         case Opt_logdev:
1212                 kfree(parsing_mp->m_logname);
1213                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1214                 if (!parsing_mp->m_logname)
1215                         return -ENOMEM;
1216                 return 0;
1217         case Opt_rtdev:
1218                 kfree(parsing_mp->m_rtname);
1219                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1220                 if (!parsing_mp->m_rtname)
1221                         return -ENOMEM;
1222                 return 0;
1223         case Opt_allocsize:
1224                 if (suffix_kstrtoint(param->string, 10, &size))
1225                         return -EINVAL;
1226                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1227                 parsing_mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1228                 return 0;
1229         case Opt_grpid:
1230         case Opt_bsdgroups:
1231                 parsing_mp->m_flags |= XFS_MOUNT_GRPID;
1232                 return 0;
1233         case Opt_nogrpid:
1234         case Opt_sysvgroups:
1235                 parsing_mp->m_flags &= ~XFS_MOUNT_GRPID;
1236                 return 0;
1237         case Opt_wsync:
1238                 parsing_mp->m_flags |= XFS_MOUNT_WSYNC;
1239                 return 0;
1240         case Opt_norecovery:
1241                 parsing_mp->m_flags |= XFS_MOUNT_NORECOVERY;
1242                 return 0;
1243         case Opt_noalign:
1244                 parsing_mp->m_flags |= XFS_MOUNT_NOALIGN;
1245                 return 0;
1246         case Opt_swalloc:
1247                 parsing_mp->m_flags |= XFS_MOUNT_SWALLOC;
1248                 return 0;
1249         case Opt_sunit:
1250                 parsing_mp->m_dalign = result.uint_32;
1251                 return 0;
1252         case Opt_swidth:
1253                 parsing_mp->m_swidth = result.uint_32;
1254                 return 0;
1255         case Opt_inode32:
1256                 parsing_mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1257                 return 0;
1258         case Opt_inode64:
1259                 parsing_mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1260                 return 0;
1261         case Opt_nouuid:
1262                 parsing_mp->m_flags |= XFS_MOUNT_NOUUID;
1263                 return 0;
1264         case Opt_largeio:
1265                 parsing_mp->m_flags |= XFS_MOUNT_LARGEIO;
1266                 return 0;
1267         case Opt_nolargeio:
1268                 parsing_mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1269                 return 0;
1270         case Opt_filestreams:
1271                 parsing_mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1272                 return 0;
1273         case Opt_noquota:
1274                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1275                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1276                 return 0;
1277         case Opt_quota:
1278         case Opt_uquota:
1279         case Opt_usrquota:
1280                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1281                 return 0;
1282         case Opt_qnoenforce:
1283         case Opt_uqnoenforce:
1284                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1285                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1286                 return 0;
1287         case Opt_pquota:
1288         case Opt_prjquota:
1289                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1290                 return 0;
1291         case Opt_pqnoenforce:
1292                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1293                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1294                 return 0;
1295         case Opt_gquota:
1296         case Opt_grpquota:
1297                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1298                 return 0;
1299         case Opt_gqnoenforce:
1300                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1301                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1302                 return 0;
1303         case Opt_discard:
1304                 parsing_mp->m_flags |= XFS_MOUNT_DISCARD;
1305                 return 0;
1306         case Opt_nodiscard:
1307                 parsing_mp->m_flags &= ~XFS_MOUNT_DISCARD;
1308                 return 0;
1309 #ifdef CONFIG_FS_DAX
1310         case Opt_dax:
1311                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1312                 return 0;
1313         case Opt_dax_enum:
1314                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1315                 return 0;
1316 #endif
1317         /* Following mount options will be removed in September 2025 */
1318         case Opt_ikeep:
1319                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, true);
1320                 parsing_mp->m_flags |= XFS_MOUNT_IKEEP;
1321                 return 0;
1322         case Opt_noikeep:
1323                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, false);
1324                 parsing_mp->m_flags &= ~XFS_MOUNT_IKEEP;
1325                 return 0;
1326         case Opt_attr2:
1327                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_ATTR2, true);
1328                 parsing_mp->m_flags |= XFS_MOUNT_ATTR2;
1329                 return 0;
1330         case Opt_noattr2:
1331                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_NOATTR2, true);
1332                 parsing_mp->m_flags &= ~XFS_MOUNT_ATTR2;
1333                 parsing_mp->m_flags |= XFS_MOUNT_NOATTR2;
1334                 return 0;
1335         default:
1336                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1337                 return -EINVAL;
1338         }
1339
1340         return 0;
1341 }
1342
1343 static int
1344 xfs_fs_validate_params(
1345         struct xfs_mount        *mp)
1346 {
1347         /*
1348          * no recovery flag requires a read-only mount
1349          */
1350         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1351             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1352                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1353                 return -EINVAL;
1354         }
1355
1356         if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1357             (mp->m_dalign || mp->m_swidth)) {
1358                 xfs_warn(mp,
1359         "sunit and swidth options incompatible with the noalign option");
1360                 return -EINVAL;
1361         }
1362
1363         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1364                 xfs_warn(mp, "quota support not available in this kernel.");
1365                 return -EINVAL;
1366         }
1367
1368         if ((mp->m_dalign && !mp->m_swidth) ||
1369             (!mp->m_dalign && mp->m_swidth)) {
1370                 xfs_warn(mp, "sunit and swidth must be specified together");
1371                 return -EINVAL;
1372         }
1373
1374         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1375                 xfs_warn(mp,
1376         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1377                         mp->m_swidth, mp->m_dalign);
1378                 return -EINVAL;
1379         }
1380
1381         if (mp->m_logbufs != -1 &&
1382             mp->m_logbufs != 0 &&
1383             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1384              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1385                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1386                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1387                 return -EINVAL;
1388         }
1389
1390         if (mp->m_logbsize != -1 &&
1391             mp->m_logbsize !=  0 &&
1392             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1393              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1394              !is_power_of_2(mp->m_logbsize))) {
1395                 xfs_warn(mp,
1396                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1397                         mp->m_logbsize);
1398                 return -EINVAL;
1399         }
1400
1401         if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1402             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1403              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1404                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1405                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1406                 return -EINVAL;
1407         }
1408
1409         return 0;
1410 }
1411
1412 static int
1413 xfs_fs_fill_super(
1414         struct super_block      *sb,
1415         struct fs_context       *fc)
1416 {
1417         struct xfs_mount        *mp = sb->s_fs_info;
1418         struct inode            *root;
1419         int                     flags = 0, error;
1420
1421         mp->m_super = sb;
1422
1423         error = xfs_fs_validate_params(mp);
1424         if (error)
1425                 goto out_free_names;
1426
1427         sb_min_blocksize(sb, BBSIZE);
1428         sb->s_xattr = xfs_xattr_handlers;
1429         sb->s_export_op = &xfs_export_operations;
1430 #ifdef CONFIG_XFS_QUOTA
1431         sb->s_qcop = &xfs_quotactl_operations;
1432         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1433 #endif
1434         sb->s_op = &xfs_super_operations;
1435
1436         /*
1437          * Delay mount work if the debug hook is set. This is debug
1438          * instrumention to coordinate simulation of xfs mount failures with
1439          * VFS superblock operations
1440          */
1441         if (xfs_globals.mount_delay) {
1442                 xfs_notice(mp, "Delaying mount for %d seconds.",
1443                         xfs_globals.mount_delay);
1444                 msleep(xfs_globals.mount_delay * 1000);
1445         }
1446
1447         if (fc->sb_flags & SB_SILENT)
1448                 flags |= XFS_MFSI_QUIET;
1449
1450         error = xfs_open_devices(mp);
1451         if (error)
1452                 goto out_free_names;
1453
1454         error = xfs_init_mount_workqueues(mp);
1455         if (error)
1456                 goto out_close_devices;
1457
1458         error = xfs_init_percpu_counters(mp);
1459         if (error)
1460                 goto out_destroy_workqueues;
1461
1462         error = xfs_inodegc_init_percpu(mp);
1463         if (error)
1464                 goto out_destroy_counters;
1465
1466         /*
1467          * All percpu data structures requiring cleanup when a cpu goes offline
1468          * must be allocated before adding this @mp to the cpu-dead handler's
1469          * mount list.
1470          */
1471         xfs_mount_list_add(mp);
1472
1473         /* Allocate stats memory before we do operations that might use it */
1474         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1475         if (!mp->m_stats.xs_stats) {
1476                 error = -ENOMEM;
1477                 goto out_destroy_inodegc;
1478         }
1479
1480         error = xfs_readsb(mp, flags);
1481         if (error)
1482                 goto out_free_stats;
1483
1484         error = xfs_finish_flags(mp);
1485         if (error)
1486                 goto out_free_sb;
1487
1488         error = xfs_setup_devices(mp);
1489         if (error)
1490                 goto out_free_sb;
1491
1492         /* V4 support is undergoing deprecation. */
1493         if (!xfs_sb_version_hascrc(&mp->m_sb)) {
1494 #ifdef CONFIG_XFS_SUPPORT_V4
1495                 xfs_warn_once(mp,
1496         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1497 #else
1498                 xfs_warn(mp,
1499         "Deprecated V4 format (crc=0) not supported by kernel.");
1500                 error = -EINVAL;
1501                 goto out_free_sb;
1502 #endif
1503         }
1504
1505         /* Filesystem claims it needs repair, so refuse the mount. */
1506         if (xfs_sb_version_needsrepair(&mp->m_sb)) {
1507                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1508                 error = -EFSCORRUPTED;
1509                 goto out_free_sb;
1510         }
1511
1512         /*
1513          * Don't touch the filesystem if a user tool thinks it owns the primary
1514          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1515          * we don't check them at all.
1516          */
1517         if (mp->m_sb.sb_inprogress) {
1518                 xfs_warn(mp, "Offline file system operation in progress!");
1519                 error = -EFSCORRUPTED;
1520                 goto out_free_sb;
1521         }
1522
1523         /*
1524          * Until this is fixed only page-sized or smaller data blocks work.
1525          */
1526         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1527                 xfs_warn(mp,
1528                 "File system with blocksize %d bytes. "
1529                 "Only pagesize (%ld) or less will currently work.",
1530                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1531                 error = -ENOSYS;
1532                 goto out_free_sb;
1533         }
1534
1535         /* Ensure this filesystem fits in the page cache limits */
1536         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1537             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1538                 xfs_warn(mp,
1539                 "file system too large to be mounted on this system.");
1540                 error = -EFBIG;
1541                 goto out_free_sb;
1542         }
1543
1544         /*
1545          * XFS block mappings use 54 bits to store the logical block offset.
1546          * This should suffice to handle the maximum file size that the VFS
1547          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1548          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1549          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1550          * to check this assertion.
1551          *
1552          * Avoid integer overflow by comparing the maximum bmbt offset to the
1553          * maximum pagecache offset in units of fs blocks.
1554          */
1555         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1556                 xfs_warn(mp,
1557 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1558                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1559                          XFS_MAX_FILEOFF);
1560                 error = -EINVAL;
1561                 goto out_free_sb;
1562         }
1563
1564         error = xfs_filestream_mount(mp);
1565         if (error)
1566                 goto out_free_sb;
1567
1568         /*
1569          * we must configure the block size in the superblock before we run the
1570          * full mount process as the mount process can lookup and cache inodes.
1571          */
1572         sb->s_magic = XFS_SUPER_MAGIC;
1573         sb->s_blocksize = mp->m_sb.sb_blocksize;
1574         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1575         sb->s_maxbytes = MAX_LFS_FILESIZE;
1576         sb->s_max_links = XFS_MAXLINK;
1577         sb->s_time_gran = 1;
1578         if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
1579                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1580                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1581         } else {
1582                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1583                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1584         }
1585         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1586         sb->s_iflags |= SB_I_CGROUPWB;
1587
1588         set_posix_acl_flag(sb);
1589
1590         /* version 5 superblocks support inode version counters. */
1591         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1592                 sb->s_flags |= SB_I_VERSION;
1593
1594         if (mp->m_flags & XFS_MOUNT_DAX_ALWAYS) {
1595                 bool rtdev_is_dax = false, datadev_is_dax;
1596
1597                 xfs_warn(mp,
1598                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1599
1600                 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1601                         sb->s_blocksize);
1602                 if (mp->m_rtdev_targp)
1603                         rtdev_is_dax = bdev_dax_supported(
1604                                 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1605                 if (!rtdev_is_dax && !datadev_is_dax) {
1606                         xfs_alert(mp,
1607                         "DAX unsupported by block device. Turning off DAX.");
1608                         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1609                 }
1610                 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1611                         xfs_alert(mp,
1612                 "DAX and reflink cannot be used together!");
1613                         error = -EINVAL;
1614                         goto out_filestream_unmount;
1615                 }
1616         }
1617
1618         if (mp->m_flags & XFS_MOUNT_DISCARD) {
1619                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1620
1621                 if (!blk_queue_discard(q)) {
1622                         xfs_warn(mp, "mounting with \"discard\" option, but "
1623                                         "the device does not support discard");
1624                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
1625                 }
1626         }
1627
1628         if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1629                 if (mp->m_sb.sb_rblocks) {
1630                         xfs_alert(mp,
1631         "reflink not compatible with realtime device!");
1632                         error = -EINVAL;
1633                         goto out_filestream_unmount;
1634                 }
1635
1636                 if (xfs_globals.always_cow) {
1637                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1638                         mp->m_always_cow = true;
1639                 }
1640         }
1641
1642         if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1643                 xfs_alert(mp,
1644         "reverse mapping btree not compatible with realtime device!");
1645                 error = -EINVAL;
1646                 goto out_filestream_unmount;
1647         }
1648
1649         error = xfs_mountfs(mp);
1650         if (error)
1651                 goto out_filestream_unmount;
1652
1653         root = igrab(VFS_I(mp->m_rootip));
1654         if (!root) {
1655                 error = -ENOENT;
1656                 goto out_unmount;
1657         }
1658         sb->s_root = d_make_root(root);
1659         if (!sb->s_root) {
1660                 error = -ENOMEM;
1661                 goto out_unmount;
1662         }
1663
1664         return 0;
1665
1666  out_filestream_unmount:
1667         xfs_filestream_unmount(mp);
1668  out_free_sb:
1669         xfs_freesb(mp);
1670  out_free_stats:
1671         free_percpu(mp->m_stats.xs_stats);
1672  out_destroy_inodegc:
1673         xfs_mount_list_del(mp);
1674         xfs_inodegc_free_percpu(mp);
1675  out_destroy_counters:
1676         xfs_destroy_percpu_counters(mp);
1677  out_destroy_workqueues:
1678         xfs_destroy_mount_workqueues(mp);
1679  out_close_devices:
1680         xfs_close_devices(mp);
1681  out_free_names:
1682         sb->s_fs_info = NULL;
1683         xfs_mount_free(mp);
1684         return error;
1685
1686  out_unmount:
1687         xfs_filestream_unmount(mp);
1688         xfs_unmountfs(mp);
1689         goto out_free_sb;
1690 }
1691
1692 static int
1693 xfs_fs_get_tree(
1694         struct fs_context       *fc)
1695 {
1696         return get_tree_bdev(fc, xfs_fs_fill_super);
1697 }
1698
1699 static int
1700 xfs_remount_rw(
1701         struct xfs_mount        *mp)
1702 {
1703         struct xfs_sb           *sbp = &mp->m_sb;
1704         int error;
1705
1706         if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1707                 xfs_warn(mp,
1708                         "ro->rw transition prohibited on norecovery mount");
1709                 return -EINVAL;
1710         }
1711
1712         if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1713             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1714                 xfs_warn(mp,
1715         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1716                         (sbp->sb_features_ro_compat &
1717                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1718                 return -EINVAL;
1719         }
1720
1721         mp->m_flags &= ~XFS_MOUNT_RDONLY;
1722
1723         /*
1724          * If this is the first remount to writeable state we might have some
1725          * superblock changes to update.
1726          */
1727         if (mp->m_update_sb) {
1728                 error = xfs_sync_sb(mp, false);
1729                 if (error) {
1730                         xfs_warn(mp, "failed to write sb changes");
1731                         return error;
1732                 }
1733                 mp->m_update_sb = false;
1734         }
1735
1736         /*
1737          * Fill out the reserve pool if it is empty. Use the stashed value if
1738          * it is non-zero, otherwise go with the default.
1739          */
1740         xfs_restore_resvblks(mp);
1741         xfs_log_work_queue(mp);
1742
1743         /* Recover any CoW blocks that never got remapped. */
1744         error = xfs_reflink_recover_cow(mp);
1745         if (error) {
1746                 xfs_err(mp,
1747                         "Error %d recovering leftover CoW allocations.", error);
1748                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1749                 return error;
1750         }
1751         xfs_blockgc_start(mp);
1752
1753         /* Create the per-AG metadata reservation pool .*/
1754         error = xfs_fs_reserve_ag_blocks(mp);
1755         if (error && error != -ENOSPC)
1756                 return error;
1757
1758         /* Re-enable the background inode inactivation worker. */
1759         xfs_inodegc_start(mp);
1760
1761         return 0;
1762 }
1763
1764 static int
1765 xfs_remount_ro(
1766         struct xfs_mount        *mp)
1767 {
1768         int error;
1769
1770         /*
1771          * Cancel background eofb scanning so it cannot race with the final
1772          * log force+buftarg wait and deadlock the remount.
1773          */
1774         xfs_blockgc_stop(mp);
1775
1776         /* Get rid of any leftover CoW reservations... */
1777         error = xfs_blockgc_free_space(mp, NULL);
1778         if (error) {
1779                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1780                 return error;
1781         }
1782
1783         /*
1784          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1785          * flushed all pending inodegc work when it sync'd the filesystem.
1786          * The VFS holds s_umount, so we know that inodes cannot enter
1787          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1788          * we send inodes straight to reclaim, so no inodes will be queued.
1789          */
1790         xfs_inodegc_stop(mp);
1791
1792         /* Free the per-AG metadata reservation pool. */
1793         error = xfs_fs_unreserve_ag_blocks(mp);
1794         if (error) {
1795                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1796                 return error;
1797         }
1798
1799         /*
1800          * Before we sync the metadata, we need to free up the reserve block
1801          * pool so that the used block count in the superblock on disk is
1802          * correct at the end of the remount. Stash the current* reserve pool
1803          * size so that if we get remounted rw, we can return it to the same
1804          * size.
1805          */
1806         xfs_save_resvblks(mp);
1807
1808         xfs_log_clean(mp);
1809         mp->m_flags |= XFS_MOUNT_RDONLY;
1810
1811         return 0;
1812 }
1813
1814 /*
1815  * Logically we would return an error here to prevent users from believing
1816  * they might have changed mount options using remount which can't be changed.
1817  *
1818  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1819  * arguments in some cases so we can't blindly reject options, but have to
1820  * check for each specified option if it actually differs from the currently
1821  * set option and only reject it if that's the case.
1822  *
1823  * Until that is implemented we return success for every remount request, and
1824  * silently ignore all options that we can't actually change.
1825  */
1826 static int
1827 xfs_fs_reconfigure(
1828         struct fs_context *fc)
1829 {
1830         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1831         struct xfs_mount        *new_mp = fc->s_fs_info;
1832         xfs_sb_t                *sbp = &mp->m_sb;
1833         int                     flags = fc->sb_flags;
1834         int                     error;
1835
1836         /* version 5 superblocks always support version counters. */
1837         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1838                 fc->sb_flags |= SB_I_VERSION;
1839
1840         error = xfs_fs_validate_params(new_mp);
1841         if (error)
1842                 return error;
1843
1844         sync_filesystem(mp->m_super);
1845
1846         /* inode32 -> inode64 */
1847         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1848             !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1849                 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1850                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1851         }
1852
1853         /* inode64 -> inode32 */
1854         if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1855             (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1856                 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1857                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1858         }
1859
1860         /* ro -> rw */
1861         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1862                 error = xfs_remount_rw(mp);
1863                 if (error)
1864                         return error;
1865         }
1866
1867         /* rw -> ro */
1868         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1869                 error = xfs_remount_ro(mp);
1870                 if (error)
1871                         return error;
1872         }
1873
1874         return 0;
1875 }
1876
1877 static void xfs_fs_free(
1878         struct fs_context       *fc)
1879 {
1880         struct xfs_mount        *mp = fc->s_fs_info;
1881
1882         /*
1883          * mp is stored in the fs_context when it is initialized.
1884          * mp is transferred to the superblock on a successful mount,
1885          * but if an error occurs before the transfer we have to free
1886          * it here.
1887          */
1888         if (mp)
1889                 xfs_mount_free(mp);
1890 }
1891
1892 static const struct fs_context_operations xfs_context_ops = {
1893         .parse_param = xfs_fs_parse_param,
1894         .get_tree    = xfs_fs_get_tree,
1895         .reconfigure = xfs_fs_reconfigure,
1896         .free        = xfs_fs_free,
1897 };
1898
1899 static int xfs_init_fs_context(
1900         struct fs_context       *fc)
1901 {
1902         struct xfs_mount        *mp;
1903
1904         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1905         if (!mp)
1906                 return -ENOMEM;
1907
1908         spin_lock_init(&mp->m_sb_lock);
1909         spin_lock_init(&mp->m_agirotor_lock);
1910         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1911         spin_lock_init(&mp->m_perag_lock);
1912         mutex_init(&mp->m_growlock);
1913         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1914         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1915         mp->m_kobj.kobject.kset = xfs_kset;
1916         /*
1917          * We don't create the finobt per-ag space reservation until after log
1918          * recovery, so we must set this to true so that an ifree transaction
1919          * started during log recovery will not depend on space reservations
1920          * for finobt expansion.
1921          */
1922         mp->m_finobt_nores = true;
1923
1924         /*
1925          * These can be overridden by the mount option parsing.
1926          */
1927         mp->m_logbufs = -1;
1928         mp->m_logbsize = -1;
1929         mp->m_allocsize_log = 16; /* 64k */
1930
1931         /*
1932          * Copy binary VFS mount flags we are interested in.
1933          */
1934         if (fc->sb_flags & SB_RDONLY)
1935                 mp->m_flags |= XFS_MOUNT_RDONLY;
1936         if (fc->sb_flags & SB_DIRSYNC)
1937                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1938         if (fc->sb_flags & SB_SYNCHRONOUS)
1939                 mp->m_flags |= XFS_MOUNT_WSYNC;
1940
1941         fc->s_fs_info = mp;
1942         fc->ops = &xfs_context_ops;
1943
1944         return 0;
1945 }
1946
1947 static struct file_system_type xfs_fs_type = {
1948         .owner                  = THIS_MODULE,
1949         .name                   = "xfs",
1950         .init_fs_context        = xfs_init_fs_context,
1951         .parameters             = xfs_fs_parameters,
1952         .kill_sb                = kill_block_super,
1953         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1954 };
1955 MODULE_ALIAS_FS("xfs");
1956
1957 STATIC int __init
1958 xfs_init_zones(void)
1959 {
1960         xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1961                                                 sizeof(struct xlog_ticket),
1962                                                 0, 0, NULL);
1963         if (!xfs_log_ticket_zone)
1964                 goto out;
1965
1966         xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1967                                         sizeof(struct xfs_extent_free_item),
1968                                         0, 0, NULL);
1969         if (!xfs_bmap_free_item_zone)
1970                 goto out_destroy_log_ticket_zone;
1971
1972         xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1973                                                sizeof(struct xfs_btree_cur),
1974                                                0, 0, NULL);
1975         if (!xfs_btree_cur_zone)
1976                 goto out_destroy_bmap_free_item_zone;
1977
1978         xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1979                                               sizeof(struct xfs_da_state),
1980                                               0, 0, NULL);
1981         if (!xfs_da_state_zone)
1982                 goto out_destroy_btree_cur_zone;
1983
1984         xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1985                                            sizeof(struct xfs_ifork),
1986                                            0, 0, NULL);
1987         if (!xfs_ifork_zone)
1988                 goto out_destroy_da_state_zone;
1989
1990         xfs_trans_zone = kmem_cache_create("xfs_trans",
1991                                            sizeof(struct xfs_trans),
1992                                            0, 0, NULL);
1993         if (!xfs_trans_zone)
1994                 goto out_destroy_ifork_zone;
1995
1996
1997         /*
1998          * The size of the zone allocated buf log item is the maximum
1999          * size possible under XFS.  This wastes a little bit of memory,
2000          * but it is much faster.
2001          */
2002         xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
2003                                               sizeof(struct xfs_buf_log_item),
2004                                               0, 0, NULL);
2005         if (!xfs_buf_item_zone)
2006                 goto out_destroy_trans_zone;
2007
2008         xfs_efd_zone = kmem_cache_create("xfs_efd_item",
2009                                         (sizeof(struct xfs_efd_log_item) +
2010                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
2011                                         sizeof(struct xfs_extent)),
2012                                         0, 0, NULL);
2013         if (!xfs_efd_zone)
2014                 goto out_destroy_buf_item_zone;
2015
2016         xfs_efi_zone = kmem_cache_create("xfs_efi_item",
2017                                          (sizeof(struct xfs_efi_log_item) +
2018                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2019                                          sizeof(struct xfs_extent)),
2020                                          0, 0, NULL);
2021         if (!xfs_efi_zone)
2022                 goto out_destroy_efd_zone;
2023
2024         xfs_inode_zone = kmem_cache_create("xfs_inode",
2025                                            sizeof(struct xfs_inode), 0,
2026                                            (SLAB_HWCACHE_ALIGN |
2027                                             SLAB_RECLAIM_ACCOUNT |
2028                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2029                                            xfs_fs_inode_init_once);
2030         if (!xfs_inode_zone)
2031                 goto out_destroy_efi_zone;
2032
2033         xfs_ili_zone = kmem_cache_create("xfs_ili",
2034                                          sizeof(struct xfs_inode_log_item), 0,
2035                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2036                                          NULL);
2037         if (!xfs_ili_zone)
2038                 goto out_destroy_inode_zone;
2039
2040         xfs_icreate_zone = kmem_cache_create("xfs_icr",
2041                                              sizeof(struct xfs_icreate_item),
2042                                              0, 0, NULL);
2043         if (!xfs_icreate_zone)
2044                 goto out_destroy_ili_zone;
2045
2046         xfs_rud_zone = kmem_cache_create("xfs_rud_item",
2047                                          sizeof(struct xfs_rud_log_item),
2048                                          0, 0, NULL);
2049         if (!xfs_rud_zone)
2050                 goto out_destroy_icreate_zone;
2051
2052         xfs_rui_zone = kmem_cache_create("xfs_rui_item",
2053                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2054                         0, 0, NULL);
2055         if (!xfs_rui_zone)
2056                 goto out_destroy_rud_zone;
2057
2058         xfs_cud_zone = kmem_cache_create("xfs_cud_item",
2059                                          sizeof(struct xfs_cud_log_item),
2060                                          0, 0, NULL);
2061         if (!xfs_cud_zone)
2062                 goto out_destroy_rui_zone;
2063
2064         xfs_cui_zone = kmem_cache_create("xfs_cui_item",
2065                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2066                         0, 0, NULL);
2067         if (!xfs_cui_zone)
2068                 goto out_destroy_cud_zone;
2069
2070         xfs_bud_zone = kmem_cache_create("xfs_bud_item",
2071                                          sizeof(struct xfs_bud_log_item),
2072                                          0, 0, NULL);
2073         if (!xfs_bud_zone)
2074                 goto out_destroy_cui_zone;
2075
2076         xfs_bui_zone = kmem_cache_create("xfs_bui_item",
2077                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2078                         0, 0, NULL);
2079         if (!xfs_bui_zone)
2080                 goto out_destroy_bud_zone;
2081
2082         return 0;
2083
2084  out_destroy_bud_zone:
2085         kmem_cache_destroy(xfs_bud_zone);
2086  out_destroy_cui_zone:
2087         kmem_cache_destroy(xfs_cui_zone);
2088  out_destroy_cud_zone:
2089         kmem_cache_destroy(xfs_cud_zone);
2090  out_destroy_rui_zone:
2091         kmem_cache_destroy(xfs_rui_zone);
2092  out_destroy_rud_zone:
2093         kmem_cache_destroy(xfs_rud_zone);
2094  out_destroy_icreate_zone:
2095         kmem_cache_destroy(xfs_icreate_zone);
2096  out_destroy_ili_zone:
2097         kmem_cache_destroy(xfs_ili_zone);
2098  out_destroy_inode_zone:
2099         kmem_cache_destroy(xfs_inode_zone);
2100  out_destroy_efi_zone:
2101         kmem_cache_destroy(xfs_efi_zone);
2102  out_destroy_efd_zone:
2103         kmem_cache_destroy(xfs_efd_zone);
2104  out_destroy_buf_item_zone:
2105         kmem_cache_destroy(xfs_buf_item_zone);
2106  out_destroy_trans_zone:
2107         kmem_cache_destroy(xfs_trans_zone);
2108  out_destroy_ifork_zone:
2109         kmem_cache_destroy(xfs_ifork_zone);
2110  out_destroy_da_state_zone:
2111         kmem_cache_destroy(xfs_da_state_zone);
2112  out_destroy_btree_cur_zone:
2113         kmem_cache_destroy(xfs_btree_cur_zone);
2114  out_destroy_bmap_free_item_zone:
2115         kmem_cache_destroy(xfs_bmap_free_item_zone);
2116  out_destroy_log_ticket_zone:
2117         kmem_cache_destroy(xfs_log_ticket_zone);
2118  out:
2119         return -ENOMEM;
2120 }
2121
2122 STATIC void
2123 xfs_destroy_zones(void)
2124 {
2125         /*
2126          * Make sure all delayed rcu free are flushed before we
2127          * destroy caches.
2128          */
2129         rcu_barrier();
2130         kmem_cache_destroy(xfs_bui_zone);
2131         kmem_cache_destroy(xfs_bud_zone);
2132         kmem_cache_destroy(xfs_cui_zone);
2133         kmem_cache_destroy(xfs_cud_zone);
2134         kmem_cache_destroy(xfs_rui_zone);
2135         kmem_cache_destroy(xfs_rud_zone);
2136         kmem_cache_destroy(xfs_icreate_zone);
2137         kmem_cache_destroy(xfs_ili_zone);
2138         kmem_cache_destroy(xfs_inode_zone);
2139         kmem_cache_destroy(xfs_efi_zone);
2140         kmem_cache_destroy(xfs_efd_zone);
2141         kmem_cache_destroy(xfs_buf_item_zone);
2142         kmem_cache_destroy(xfs_trans_zone);
2143         kmem_cache_destroy(xfs_ifork_zone);
2144         kmem_cache_destroy(xfs_da_state_zone);
2145         kmem_cache_destroy(xfs_btree_cur_zone);
2146         kmem_cache_destroy(xfs_bmap_free_item_zone);
2147         kmem_cache_destroy(xfs_log_ticket_zone);
2148 }
2149
2150 STATIC int __init
2151 xfs_init_workqueues(void)
2152 {
2153         /*
2154          * The allocation workqueue can be used in memory reclaim situations
2155          * (writepage path), and parallelism is only limited by the number of
2156          * AGs in all the filesystems mounted. Hence use the default large
2157          * max_active value for this workqueue.
2158          */
2159         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2160                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2161         if (!xfs_alloc_wq)
2162                 return -ENOMEM;
2163
2164         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2165                         0);
2166         if (!xfs_discard_wq)
2167                 goto out_free_alloc_wq;
2168
2169         return 0;
2170 out_free_alloc_wq:
2171         destroy_workqueue(xfs_alloc_wq);
2172         return -ENOMEM;
2173 }
2174
2175 STATIC void
2176 xfs_destroy_workqueues(void)
2177 {
2178         destroy_workqueue(xfs_discard_wq);
2179         destroy_workqueue(xfs_alloc_wq);
2180 }
2181
2182 #ifdef CONFIG_HOTPLUG_CPU
2183 static int
2184 xfs_cpu_dead(
2185         unsigned int            cpu)
2186 {
2187         struct xfs_mount        *mp, *n;
2188
2189         spin_lock(&xfs_mount_list_lock);
2190         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2191                 spin_unlock(&xfs_mount_list_lock);
2192                 xfs_inodegc_cpu_dead(mp, cpu);
2193                 spin_lock(&xfs_mount_list_lock);
2194         }
2195         spin_unlock(&xfs_mount_list_lock);
2196         return 0;
2197 }
2198
2199 static int __init
2200 xfs_cpu_hotplug_init(void)
2201 {
2202         int     error;
2203
2204         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2205                         xfs_cpu_dead);
2206         if (error < 0)
2207                 xfs_alert(NULL,
2208 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2209                         error);
2210         return error;
2211 }
2212
2213 static void
2214 xfs_cpu_hotplug_destroy(void)
2215 {
2216         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2217 }
2218
2219 #else /* !CONFIG_HOTPLUG_CPU */
2220 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2221 static inline void xfs_cpu_hotplug_destroy(void) {}
2222 #endif
2223
2224 STATIC int __init
2225 init_xfs_fs(void)
2226 {
2227         int                     error;
2228
2229         xfs_check_ondisk_structs();
2230
2231         printk(KERN_INFO XFS_VERSION_STRING " with "
2232                          XFS_BUILD_OPTIONS " enabled\n");
2233
2234         xfs_dir_startup();
2235
2236         error = xfs_cpu_hotplug_init();
2237         if (error)
2238                 goto out;
2239
2240         error = xfs_init_zones();
2241         if (error)
2242                 goto out_destroy_hp;
2243
2244         error = xfs_init_workqueues();
2245         if (error)
2246                 goto out_destroy_zones;
2247
2248         error = xfs_mru_cache_init();
2249         if (error)
2250                 goto out_destroy_wq;
2251
2252         error = xfs_buf_init();
2253         if (error)
2254                 goto out_mru_cache_uninit;
2255
2256         error = xfs_init_procfs();
2257         if (error)
2258                 goto out_buf_terminate;
2259
2260         error = xfs_sysctl_register();
2261         if (error)
2262                 goto out_cleanup_procfs;
2263
2264         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2265         if (!xfs_kset) {
2266                 error = -ENOMEM;
2267                 goto out_sysctl_unregister;
2268         }
2269
2270         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2271
2272         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2273         if (!xfsstats.xs_stats) {
2274                 error = -ENOMEM;
2275                 goto out_kset_unregister;
2276         }
2277
2278         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2279                                "stats");
2280         if (error)
2281                 goto out_free_stats;
2282
2283 #ifdef DEBUG
2284         xfs_dbg_kobj.kobject.kset = xfs_kset;
2285         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2286         if (error)
2287                 goto out_remove_stats_kobj;
2288 #endif
2289
2290         error = xfs_qm_init();
2291         if (error)
2292                 goto out_remove_dbg_kobj;
2293
2294         error = register_filesystem(&xfs_fs_type);
2295         if (error)
2296                 goto out_qm_exit;
2297         return 0;
2298
2299  out_qm_exit:
2300         xfs_qm_exit();
2301  out_remove_dbg_kobj:
2302 #ifdef DEBUG
2303         xfs_sysfs_del(&xfs_dbg_kobj);
2304  out_remove_stats_kobj:
2305 #endif
2306         xfs_sysfs_del(&xfsstats.xs_kobj);
2307  out_free_stats:
2308         free_percpu(xfsstats.xs_stats);
2309  out_kset_unregister:
2310         kset_unregister(xfs_kset);
2311  out_sysctl_unregister:
2312         xfs_sysctl_unregister();
2313  out_cleanup_procfs:
2314         xfs_cleanup_procfs();
2315  out_buf_terminate:
2316         xfs_buf_terminate();
2317  out_mru_cache_uninit:
2318         xfs_mru_cache_uninit();
2319  out_destroy_wq:
2320         xfs_destroy_workqueues();
2321  out_destroy_zones:
2322         xfs_destroy_zones();
2323  out_destroy_hp:
2324         xfs_cpu_hotplug_destroy();
2325  out:
2326         return error;
2327 }
2328
2329 STATIC void __exit
2330 exit_xfs_fs(void)
2331 {
2332         xfs_qm_exit();
2333         unregister_filesystem(&xfs_fs_type);
2334 #ifdef DEBUG
2335         xfs_sysfs_del(&xfs_dbg_kobj);
2336 #endif
2337         xfs_sysfs_del(&xfsstats.xs_kobj);
2338         free_percpu(xfsstats.xs_stats);
2339         kset_unregister(xfs_kset);
2340         xfs_sysctl_unregister();
2341         xfs_cleanup_procfs();
2342         xfs_buf_terminate();
2343         xfs_mru_cache_uninit();
2344         xfs_destroy_workqueues();
2345         xfs_destroy_zones();
2346         xfs_uuid_table_free();
2347         xfs_cpu_hotplug_destroy();
2348 }
2349
2350 module_init(init_xfs_fs);
2351 module_exit(exit_xfs_fs);
2352
2353 MODULE_AUTHOR("Silicon Graphics, Inc.");
2354 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2355 MODULE_LICENSE("GPL");