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_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
  88                 break;
  89         case XFS_DAX_ALWAYS:
  90                 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
  91                 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
  92                 break;
  93         case XFS_DAX_NEVER:
  94                 mp->m_features |= XFS_FEAT_DAX_NEVER;
  95                 mp->m_features &= ~XFS_FEAT_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_FEAT_IKEEP,               ",ikeep" },
 180                 { XFS_FEAT_WSYNC,               ",wsync" },
 181                 { XFS_FEAT_NOALIGN,             ",noalign" },
 182                 { XFS_FEAT_SWALLOC,             ",swalloc" },
 183                 { XFS_FEAT_NOUUID,              ",nouuid" },
 184                 { XFS_FEAT_NORECOVERY,          ",norecovery" },
 185                 { XFS_FEAT_ATTR2,               ",attr2" },
 186                 { XFS_FEAT_FILESTREAMS,         ",filestreams" },
 187                 { XFS_FEAT_GRPID,               ",grpid" },
 188                 { XFS_FEAT_DISCARD,             ",discard" },
 189                 { XFS_FEAT_LARGE_IOSIZE,        ",largeio" },
 190                 { XFS_FEAT_DAX_ALWAYS,          ",dax=always" },
 191                 { XFS_FEAT_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_features & xfs_infop->flag)
 199                         seq_puts(m, xfs_infop->str);
 200         }
 201
 202         seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
 203
 204         if (xfs_has_allocsize(mp))
 205                 seq_printf(m, ",allocsize=%dk",
 206                            (1 << mp->m_allocsize_log) >> 10);
 207
 208         if (mp->m_logbufs > 0)
 209                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 210         if (mp->m_logbsize > 0)
 211                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 212
 213         if (mp->m_logname)
 214                 seq_show_option(m, "logdev", mp->m_logname);
 215         if (mp->m_rtname)
 216                 seq_show_option(m, "rtdev", mp->m_rtname);
 217
 218         if (mp->m_dalign > 0)
 219                 seq_printf(m, ",sunit=%d",
 220                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 221         if (mp->m_swidth > 0)
 222                 seq_printf(m, ",swidth=%d",
 223                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 224
 225         if (mp->m_qflags & XFS_UQUOTA_ENFD)
 226                 seq_puts(m, ",usrquota");
 227         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 228                 seq_puts(m, ",uqnoenforce");
 229
 230         if (mp->m_qflags & XFS_PQUOTA_ENFD)
 231                 seq_puts(m, ",prjquota");
 232         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
 233                 seq_puts(m, ",pqnoenforce");
 234
 235         if (mp->m_qflags & XFS_GQUOTA_ENFD)
 236                 seq_puts(m, ",grpquota");
 237         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
 238                 seq_puts(m, ",gqnoenforce");
 239
 240         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 241                 seq_puts(m, ",noquota");
 242
 243         return 0;
 244 }
 245
 246 /*
 247  * Set parameters for inode allocation heuristics, taking into account
 248  * filesystem size and inode32/inode64 mount options; i.e. specifically
 249  * whether or not XFS_FEAT_SMALL_INUMS is set.
 250  *
 251  * Inode allocation patterns are altered only if inode32 is requested
 252  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
 253  * If altered, XFS_OPSTATE_INODE32 is set as well.
 254  *
 255  * An agcount independent of that in the mount structure is provided
 256  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 257  * to the potentially higher ag count.
 258  *
 259  * Returns the maximum AG index which may contain inodes.
 260  */
 261 xfs_agnumber_t
 262 xfs_set_inode_alloc(
 263         struct xfs_mount *mp,
 264         xfs_agnumber_t  agcount)
 265 {
 266         xfs_agnumber_t  index;
 267         xfs_agnumber_t  maxagi = 0;
 268         xfs_sb_t        *sbp = &mp->m_sb;
 269         xfs_agnumber_t  max_metadata;
 270         xfs_agino_t     agino;
 271         xfs_ino_t       ino;
 272
 273         /*
 274          * Calculate how much should be reserved for inodes to meet
 275          * the max inode percentage.  Used only for inode32.
 276          */
 277         if (M_IGEO(mp)->maxicount) {
 278                 uint64_t        icount;
 279
 280                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 281                 do_div(icount, 100);
 282                 icount += sbp->sb_agblocks - 1;
 283                 do_div(icount, sbp->sb_agblocks);
 284                 max_metadata = icount;
 285         } else {
 286                 max_metadata = agcount;
 287         }
 288
 289         /* Get the last possible inode in the filesystem */
 290         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 291         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 292
 293         /*
 294          * If user asked for no more than 32-bit inodes, and the fs is
 295          * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
 296          * the allocator to accommodate the request.
 297          */
 298         if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
 299                 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 300         else
 301                 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 302
 303         for (index = 0; index < agcount; index++) {
 304                 struct xfs_perag        *pag;
 305
 306                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 307
 308                 pag = xfs_perag_get(mp, index);
 309
 310                 if (xfs_is_inode32(mp)) {
 311                         if (ino > XFS_MAXINUMBER_32) {
 312                                 pag->pagi_inodeok = 0;
 313                                 pag->pagf_metadata = 0;
 314                         } else {
 315                                 pag->pagi_inodeok = 1;
 316                                 maxagi++;
 317                                 if (index < max_metadata)
 318                                         pag->pagf_metadata = 1;
 319                                 else
 320                                         pag->pagf_metadata = 0;
 321                         }
 322                 } else {
 323                         pag->pagi_inodeok = 1;
 324                         pag->pagf_metadata = 0;
 325                 }
 326
 327                 xfs_perag_put(pag);
 328         }
 329
 330         return xfs_is_inode32(mp) ? maxagi : agcount;
 331 }
 332
 333 STATIC int
 334 xfs_blkdev_get(
 335         xfs_mount_t             *mp,
 336         const char              *name,
 337         struct block_device     **bdevp)
 338 {
 339         int                     error = 0;
 340
 341         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 342                                     mp);
 343         if (IS_ERR(*bdevp)) {
 344                 error = PTR_ERR(*bdevp);
 345                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 346         }
 347
 348         return error;
 349 }
 350
 351 STATIC void
 352 xfs_blkdev_put(
 353         struct block_device     *bdev)
 354 {
 355         if (bdev)
 356                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 357 }
 358
 359 STATIC void
 360 xfs_close_devices(
 361         struct xfs_mount        *mp)
 362 {
 363         struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
 364
 365         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 366                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 367                 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
 368
 369                 xfs_free_buftarg(mp->m_logdev_targp);
 370                 xfs_blkdev_put(logdev);
 371                 fs_put_dax(dax_logdev);
 372         }
 373         if (mp->m_rtdev_targp) {
 374                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 375                 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
 376
 377                 xfs_free_buftarg(mp->m_rtdev_targp);
 378                 xfs_blkdev_put(rtdev);
 379                 fs_put_dax(dax_rtdev);
 380         }
 381         xfs_free_buftarg(mp->m_ddev_targp);
 382         fs_put_dax(dax_ddev);
 383 }
 384
 385 /*
 386  * The file system configurations are:
 387  *      (1) device (partition) with data and internal log
 388  *      (2) logical volume with data and log subvolumes.
 389  *      (3) logical volume with data, log, and realtime subvolumes.
 390  *
 391  * We only have to handle opening the log and realtime volumes here if
 392  * they are present.  The data subvolume has already been opened by
 393  * get_sb_bdev() and is stored in sb->s_bdev.
 394  */
 395 STATIC int
 396 xfs_open_devices(
 397         struct xfs_mount        *mp)
 398 {
 399         struct block_device     *ddev = mp->m_super->s_bdev;
 400         struct dax_device       *dax_ddev = fs_dax_get_by_bdev(ddev);
 401         struct dax_device       *dax_logdev = NULL, *dax_rtdev = NULL;
 402         struct block_device     *logdev = NULL, *rtdev = NULL;
 403         int                     error;
 404
 405         /*
 406          * Open real time and log devices - order is important.
 407          */
 408         if (mp->m_logname) {
 409                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 410                 if (error)
 411                         goto out;
 412                 dax_logdev = fs_dax_get_by_bdev(logdev);
 413         }
 414
 415         if (mp->m_rtname) {
 416                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 417                 if (error)
 418                         goto out_close_logdev;
 419
 420                 if (rtdev == ddev || rtdev == logdev) {
 421                         xfs_warn(mp,
 422         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 423                         error = -EINVAL;
 424                         goto out_close_rtdev;
 425                 }
 426                 dax_rtdev = fs_dax_get_by_bdev(rtdev);
 427         }
 428
 429         /*
 430          * Setup xfs_mount buffer target pointers
 431          */
 432         error = -ENOMEM;
 433         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
 434         if (!mp->m_ddev_targp)
 435                 goto out_close_rtdev;
 436
 437         if (rtdev) {
 438                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
 439                 if (!mp->m_rtdev_targp)
 440                         goto out_free_ddev_targ;
 441         }
 442
 443         if (logdev && logdev != ddev) {
 444                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
 445                 if (!mp->m_logdev_targp)
 446                         goto out_free_rtdev_targ;
 447         } else {
 448                 mp->m_logdev_targp = mp->m_ddev_targp;
 449         }
 450
 451         return 0;
 452
 453  out_free_rtdev_targ:
 454         if (mp->m_rtdev_targp)
 455                 xfs_free_buftarg(mp->m_rtdev_targp);
 456  out_free_ddev_targ:
 457         xfs_free_buftarg(mp->m_ddev_targp);
 458  out_close_rtdev:
 459         xfs_blkdev_put(rtdev);
 460         fs_put_dax(dax_rtdev);
 461  out_close_logdev:
 462         if (logdev && logdev != ddev) {
 463                 xfs_blkdev_put(logdev);
 464                 fs_put_dax(dax_logdev);
 465         }
 466  out:
 467         fs_put_dax(dax_ddev);
 468         return error;
 469 }
 470
 471 /*
 472  * Setup xfs_mount buffer target pointers based on superblock
 473  */
 474 STATIC int
 475 xfs_setup_devices(
 476         struct xfs_mount        *mp)
 477 {
 478         int                     error;
 479
 480         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 481         if (error)
 482                 return error;
 483
 484         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 485                 unsigned int    log_sector_size = BBSIZE;
 486
 487                 if (xfs_has_sector(mp))
 488                         log_sector_size = mp->m_sb.sb_logsectsize;
 489                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 490                                             log_sector_size);
 491                 if (error)
 492                         return error;
 493         }
 494         if (mp->m_rtdev_targp) {
 495                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 496                                             mp->m_sb.sb_sectsize);
 497                 if (error)
 498                         return error;
 499         }
 500
 501         return 0;
 502 }
 503
 504 STATIC int
 505 xfs_init_mount_workqueues(
 506         struct xfs_mount        *mp)
 507 {
 508         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 509                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 510                         1, mp->m_super->s_id);
 511         if (!mp->m_buf_workqueue)
 512                 goto out;
 513
 514         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 515                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 516                         0, mp->m_super->s_id);
 517         if (!mp->m_unwritten_workqueue)
 518                 goto out_destroy_buf;
 519
 520         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 521                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 522                         0, mp->m_super->s_id);
 523         if (!mp->m_reclaim_workqueue)
 524                 goto out_destroy_unwritten;
 525
 526         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
 527                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
 528                         0, mp->m_super->s_id);
 529         if (!mp->m_blockgc_wq)
 530                 goto out_destroy_reclaim;
 531
 532         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
 533                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 534                         1, mp->m_super->s_id);
 535         if (!mp->m_inodegc_wq)
 536                 goto out_destroy_blockgc;
 537
 538         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 539                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 540         if (!mp->m_sync_workqueue)
 541                 goto out_destroy_inodegc;
 542
 543         return 0;
 544
 545 out_destroy_inodegc:
 546         destroy_workqueue(mp->m_inodegc_wq);
 547 out_destroy_blockgc:
 548         destroy_workqueue(mp->m_blockgc_wq);
 549 out_destroy_reclaim:
 550         destroy_workqueue(mp->m_reclaim_workqueue);
 551 out_destroy_unwritten:
 552         destroy_workqueue(mp->m_unwritten_workqueue);
 553 out_destroy_buf:
 554         destroy_workqueue(mp->m_buf_workqueue);
 555 out:
 556         return -ENOMEM;
 557 }
 558
 559 STATIC void
 560 xfs_destroy_mount_workqueues(
 561         struct xfs_mount        *mp)
 562 {
 563         destroy_workqueue(mp->m_sync_workqueue);
 564         destroy_workqueue(mp->m_blockgc_wq);
 565         destroy_workqueue(mp->m_inodegc_wq);
 566         destroy_workqueue(mp->m_reclaim_workqueue);
 567         destroy_workqueue(mp->m_unwritten_workqueue);
 568         destroy_workqueue(mp->m_buf_workqueue);
 569 }
 570
 571 static void
 572 xfs_flush_inodes_worker(
 573         struct work_struct      *work)
 574 {
 575         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 576                                                    m_flush_inodes_work);
 577         struct super_block      *sb = mp->m_super;
 578
 579         if (down_read_trylock(&sb->s_umount)) {
 580                 sync_inodes_sb(sb);
 581                 up_read(&sb->s_umount);
 582         }
 583 }
 584
 585 /*
 586  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 587  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 588  * for IO to complete so that we effectively throttle multiple callers to the
 589  * rate at which IO is completing.
 590  */
 591 void
 592 xfs_flush_inodes(
 593         struct xfs_mount        *mp)
 594 {
 595         /*
 596          * If flush_work() returns true then that means we waited for a flush
 597          * which was already in progress.  Don't bother running another scan.
 598          */
 599         if (flush_work(&mp->m_flush_inodes_work))
 600                 return;
 601
 602         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 603         flush_work(&mp->m_flush_inodes_work);
 604 }
 605
 606 /* Catch misguided souls that try to use this interface on XFS */
 607 STATIC struct inode *
 608 xfs_fs_alloc_inode(
 609         struct super_block      *sb)
 610 {
 611         BUG();
 612         return NULL;
 613 }
 614
 615 /*
 616  * Now that the generic code is guaranteed not to be accessing
 617  * the linux inode, we can inactivate and reclaim the inode.
 618  */
 619 STATIC void
 620 xfs_fs_destroy_inode(
 621         struct inode            *inode)
 622 {
 623         struct xfs_inode        *ip = XFS_I(inode);
 624
 625         trace_xfs_destroy_inode(ip);
 626
 627         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 628         XFS_STATS_INC(ip->i_mount, vn_rele);
 629         XFS_STATS_INC(ip->i_mount, vn_remove);
 630         xfs_inode_mark_reclaimable(ip);
 631 }
 632
 633 static void
 634 xfs_fs_dirty_inode(
 635         struct inode                    *inode,
 636         int                             flag)
 637 {
 638         struct xfs_inode                *ip = XFS_I(inode);
 639         struct xfs_mount                *mp = ip->i_mount;
 640         struct xfs_trans                *tp;
 641
 642         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 643                 return;
 644         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 645                 return;
 646
 647         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 648                 return;
 649         xfs_ilock(ip, XFS_ILOCK_EXCL);
 650         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 651         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 652         xfs_trans_commit(tp);
 653 }
 654
 655 /*
 656  * Slab object creation initialisation for the XFS inode.
 657  * This covers only the idempotent fields in the XFS inode;
 658  * all other fields need to be initialised on allocation
 659  * from the slab. This avoids the need to repeatedly initialise
 660  * fields in the xfs inode that left in the initialise state
 661  * when freeing the inode.
 662  */
 663 STATIC void
 664 xfs_fs_inode_init_once(
 665         void                    *inode)
 666 {
 667         struct xfs_inode        *ip = inode;
 668
 669         memset(ip, 0, sizeof(struct xfs_inode));
 670
 671         /* vfs inode */
 672         inode_init_once(VFS_I(ip));
 673
 674         /* xfs inode */
 675         atomic_set(&ip->i_pincount, 0);
 676         spin_lock_init(&ip->i_flags_lock);
 677
 678         mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 679                      "xfsino", ip->i_ino);
 680         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 681                      "xfsino", ip->i_ino);
 682 }
 683
 684 /*
 685  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 686  * serialised against cache hits here via the inode->i_lock and igrab() in
 687  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 688  * racing with us, and it avoids needing to grab a spinlock here for every inode
 689  * we drop the final reference on.
 690  */
 691 STATIC int
 692 xfs_fs_drop_inode(
 693         struct inode            *inode)
 694 {
 695         struct xfs_inode        *ip = XFS_I(inode);
 696
 697         /*
 698          * If this unlinked inode is in the middle of recovery, don't
 699          * drop the inode just yet; log recovery will take care of
 700          * that.  See the comment for this inode flag.
 701          */
 702         if (ip->i_flags & XFS_IRECOVERY) {
 703                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
 704                 return 0;
 705         }
 706
 707         return generic_drop_inode(inode);
 708 }
 709
 710 static void
 711 xfs_mount_free(
 712         struct xfs_mount        *mp)
 713 {
 714         kfree(mp->m_rtname);
 715         kfree(mp->m_logname);
 716         kmem_free(mp);
 717 }
 718
 719 STATIC int
 720 xfs_fs_sync_fs(
 721         struct super_block      *sb,
 722         int                     wait)
 723 {
 724         struct xfs_mount        *mp = XFS_M(sb);
 725
 726         trace_xfs_fs_sync_fs(mp, __return_address);
 727
 728         /*
 729          * Doing anything during the async pass would be counterproductive.
 730          */
 731         if (!wait)
 732                 return 0;
 733
 734         xfs_log_force(mp, XFS_LOG_SYNC);
 735         if (laptop_mode) {
 736                 /*
 737                  * The disk must be active because we're syncing.
 738                  * We schedule log work now (now that the disk is
 739                  * active) instead of later (when it might not be).
 740                  */
 741                 flush_delayed_work(&mp->m_log->l_work);
 742         }
 743
 744         /*
 745          * If we are called with page faults frozen out, it means we are about
 746          * to freeze the transaction subsystem. Take the opportunity to shut
 747          * down inodegc because once SB_FREEZE_FS is set it's too late to
 748          * prevent inactivation races with freeze. The fs doesn't get called
 749          * again by the freezing process until after SB_FREEZE_FS has been set,
 750          * so it's now or never.  Same logic applies to speculative allocation
 751          * garbage collection.
 752          *
 753          * We don't care if this is a normal syncfs call that does this or
 754          * freeze that does this - we can run this multiple times without issue
 755          * and we won't race with a restart because a restart can only occur
 756          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
 757          */
 758         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
 759                 xfs_inodegc_stop(mp);
 760                 xfs_blockgc_stop(mp);
 761         }
 762
 763         return 0;
 764 }
 765
 766 STATIC int
 767 xfs_fs_statfs(
 768         struct dentry           *dentry,
 769         struct kstatfs          *statp)
 770 {
 771         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 772         xfs_sb_t                *sbp = &mp->m_sb;
 773         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 774         uint64_t                fakeinos, id;
 775         uint64_t                icount;
 776         uint64_t                ifree;
 777         uint64_t                fdblocks;
 778         xfs_extlen_t            lsize;
 779         int64_t                 ffree;
 780
 781         /* Wait for whatever inactivations are in progress. */
 782         xfs_inodegc_flush(mp);
 783
 784         statp->f_type = XFS_SUPER_MAGIC;
 785         statp->f_namelen = MAXNAMELEN - 1;
 786
 787         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 788         statp->f_fsid = u64_to_fsid(id);
 789
 790         icount = percpu_counter_sum(&mp->m_icount);
 791         ifree = percpu_counter_sum(&mp->m_ifree);
 792         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 793
 794         spin_lock(&mp->m_sb_lock);
 795         statp->f_bsize = sbp->sb_blocksize;
 796         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 797         statp->f_blocks = sbp->sb_dblocks - lsize;
 798         spin_unlock(&mp->m_sb_lock);
 799
 800         /* make sure statp->f_bfree does not underflow */
 801         statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
 802         statp->f_bavail = statp->f_bfree;
 803
 804         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 805         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 806         if (M_IGEO(mp)->maxicount)
 807                 statp->f_files = min_t(typeof(statp->f_files),
 808                                         statp->f_files,
 809                                         M_IGEO(mp)->maxicount);
 810
 811         /* If sb_icount overshot maxicount, report actual allocation */
 812         statp->f_files = max_t(typeof(statp->f_files),
 813                                         statp->f_files,
 814                                         sbp->sb_icount);
 815
 816         /* make sure statp->f_ffree does not underflow */
 817         ffree = statp->f_files - (icount - ifree);
 818         statp->f_ffree = max_t(int64_t, ffree, 0);
 819
 820
 821         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 822             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 823                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 824                 xfs_qm_statvfs(ip, statp);
 825
 826         if (XFS_IS_REALTIME_MOUNT(mp) &&
 827             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 828                 statp->f_blocks = sbp->sb_rblocks;
 829                 statp->f_bavail = statp->f_bfree =
 830                         sbp->sb_frextents * sbp->sb_rextsize;
 831         }
 832
 833         return 0;
 834 }
 835
 836 STATIC void
 837 xfs_save_resvblks(struct xfs_mount *mp)
 838 {
 839         uint64_t resblks = 0;
 840
 841         mp->m_resblks_save = mp->m_resblks;
 842         xfs_reserve_blocks(mp, &resblks, NULL);
 843 }
 844
 845 STATIC void
 846 xfs_restore_resvblks(struct xfs_mount *mp)
 847 {
 848         uint64_t resblks;
 849
 850         if (mp->m_resblks_save) {
 851                 resblks = mp->m_resblks_save;
 852                 mp->m_resblks_save = 0;
 853         } else
 854                 resblks = xfs_default_resblks(mp);
 855
 856         xfs_reserve_blocks(mp, &resblks, NULL);
 857 }
 858
 859 /*
 860  * Second stage of a freeze. The data is already frozen so we only
 861  * need to take care of the metadata. Once that's done sync the superblock
 862  * to the log to dirty it in case of a crash while frozen. This ensures that we
 863  * will recover the unlinked inode lists on the next mount.
 864  */
 865 STATIC int
 866 xfs_fs_freeze(
 867         struct super_block      *sb)
 868 {
 869         struct xfs_mount        *mp = XFS_M(sb);
 870         unsigned int            flags;
 871         int                     ret;
 872
 873         /*
 874          * The filesystem is now frozen far enough that memory reclaim
 875          * cannot safely operate on the filesystem. Hence we need to
 876          * set a GFP_NOFS context here to avoid recursion deadlocks.
 877          */
 878         flags = memalloc_nofs_save();
 879         xfs_save_resvblks(mp);
 880         ret = xfs_log_quiesce(mp);
 881         memalloc_nofs_restore(flags);
 882
 883         /*
 884          * For read-write filesystems, we need to restart the inodegc on error
 885          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
 886          * going to be run to restart it now.  We are at SB_FREEZE_FS level
 887          * here, so we can restart safely without racing with a stop in
 888          * xfs_fs_sync_fs().
 889          */
 890         if (ret && !xfs_is_readonly(mp)) {
 891                 xfs_blockgc_start(mp);
 892                 xfs_inodegc_start(mp);
 893         }
 894
 895         return ret;
 896 }
 897
 898 STATIC int
 899 xfs_fs_unfreeze(
 900         struct super_block      *sb)
 901 {
 902         struct xfs_mount        *mp = XFS_M(sb);
 903
 904         xfs_restore_resvblks(mp);
 905         xfs_log_work_queue(mp);
 906
 907         /*
 908          * Don't reactivate the inodegc worker on a readonly filesystem because
 909          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
 910          * worker because there are no speculative preallocations on a readonly
 911          * filesystem.
 912          */
 913         if (!xfs_is_readonly(mp)) {
 914                 xfs_blockgc_start(mp);
 915                 xfs_inodegc_start(mp);
 916         }
 917
 918         return 0;
 919 }
 920
 921 /*
 922  * This function fills in xfs_mount_t fields based on mount args.
 923  * Note: the superblock _has_ now been read in.
 924  */
 925 STATIC int
 926 xfs_finish_flags(
 927         struct xfs_mount        *mp)
 928 {
 929         /* Fail a mount where the logbuf is smaller than the log stripe */
 930         if (xfs_has_logv2(mp)) {
 931                 if (mp->m_logbsize <= 0 &&
 932                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 933                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 934                 } else if (mp->m_logbsize > 0 &&
 935                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 936                         xfs_warn(mp,
 937                 "logbuf size must be greater than or equal to log stripe size");
 938                         return -EINVAL;
 939                 }
 940         } else {
 941                 /* Fail a mount if the logbuf is larger than 32K */
 942                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 943                         xfs_warn(mp,
 944                 "logbuf size for version 1 logs must be 16K or 32K");
 945                         return -EINVAL;
 946                 }
 947         }
 948
 949         /*
 950          * V5 filesystems always use attr2 format for attributes.
 951          */
 952         if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
 953                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 954                              "attr2 is always enabled for V5 filesystems.");
 955                 return -EINVAL;
 956         }
 957
 958         /*
 959          * prohibit r/w mounts of read-only filesystems
 960          */
 961         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
 962                 xfs_warn(mp,
 963                         "cannot mount a read-only filesystem as read-write");
 964                 return -EROFS;
 965         }
 966
 967         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
 968             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
 969             !xfs_has_pquotino(mp)) {
 970                 xfs_warn(mp,
 971                   "Super block does not support project and group quota together");
 972                 return -EINVAL;
 973         }
 974
 975         return 0;
 976 }
 977
 978 static int
 979 xfs_init_percpu_counters(
 980         struct xfs_mount        *mp)
 981 {
 982         int             error;
 983
 984         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
 985         if (error)
 986                 return -ENOMEM;
 987
 988         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
 989         if (error)
 990                 goto free_icount;
 991
 992         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
 993         if (error)
 994                 goto free_ifree;
 995
 996         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
 997         if (error)
 998                 goto free_fdblocks;
 999
1000         return 0;
1001
1002 free_fdblocks:
1003         percpu_counter_destroy(&mp->m_fdblocks);
1004 free_ifree:
1005         percpu_counter_destroy(&mp->m_ifree);
1006 free_icount:
1007         percpu_counter_destroy(&mp->m_icount);
1008         return -ENOMEM;
1009 }
1010
1011 void
1012 xfs_reinit_percpu_counters(
1013         struct xfs_mount        *mp)
1014 {
1015         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1016         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1017         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1018 }
1019
1020 static void
1021 xfs_destroy_percpu_counters(
1022         struct xfs_mount        *mp)
1023 {
1024         percpu_counter_destroy(&mp->m_icount);
1025         percpu_counter_destroy(&mp->m_ifree);
1026         percpu_counter_destroy(&mp->m_fdblocks);
1027         ASSERT(xfs_is_shutdown(mp) ||
1028                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1029         percpu_counter_destroy(&mp->m_delalloc_blks);
1030 }
1031
1032 static int
1033 xfs_inodegc_init_percpu(
1034         struct xfs_mount        *mp)
1035 {
1036         struct xfs_inodegc      *gc;
1037         int                     cpu;
1038
1039         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1040         if (!mp->m_inodegc)
1041                 return -ENOMEM;
1042
1043         for_each_possible_cpu(cpu) {
1044                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1045                 init_llist_head(&gc->list);
1046                 gc->items = 0;
1047                 INIT_WORK(&gc->work, xfs_inodegc_worker);
1048         }
1049         return 0;
1050 }
1051
1052 static void
1053 xfs_inodegc_free_percpu(
1054         struct xfs_mount        *mp)
1055 {
1056         if (!mp->m_inodegc)
1057                 return;
1058         free_percpu(mp->m_inodegc);
1059 }
1060
1061 static void
1062 xfs_fs_put_super(
1063         struct super_block      *sb)
1064 {
1065         struct xfs_mount        *mp = XFS_M(sb);
1066
1067         /* if ->fill_super failed, we have no mount to tear down */
1068         if (!sb->s_fs_info)
1069                 return;
1070
1071         xfs_notice(mp, "Unmounting Filesystem");
1072         xfs_filestream_unmount(mp);
1073         xfs_unmountfs(mp);
1074
1075         xfs_freesb(mp);
1076         free_percpu(mp->m_stats.xs_stats);
1077         xfs_mount_list_del(mp);
1078         xfs_inodegc_free_percpu(mp);
1079         xfs_destroy_percpu_counters(mp);
1080         xfs_destroy_mount_workqueues(mp);
1081         xfs_close_devices(mp);
1082
1083         sb->s_fs_info = NULL;
1084         xfs_mount_free(mp);
1085 }
1086
1087 static long
1088 xfs_fs_nr_cached_objects(
1089         struct super_block      *sb,
1090         struct shrink_control   *sc)
1091 {
1092         /* Paranoia: catch incorrect calls during mount setup or teardown */
1093         if (WARN_ON_ONCE(!sb->s_fs_info))
1094                 return 0;
1095         return xfs_reclaim_inodes_count(XFS_M(sb));
1096 }
1097
1098 static long
1099 xfs_fs_free_cached_objects(
1100         struct super_block      *sb,
1101         struct shrink_control   *sc)
1102 {
1103         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1104 }
1105
1106 static const struct super_operations xfs_super_operations = {
1107         .alloc_inode            = xfs_fs_alloc_inode,
1108         .destroy_inode          = xfs_fs_destroy_inode,
1109         .dirty_inode            = xfs_fs_dirty_inode,
1110         .drop_inode             = xfs_fs_drop_inode,
1111         .put_super              = xfs_fs_put_super,
1112         .sync_fs                = xfs_fs_sync_fs,
1113         .freeze_fs              = xfs_fs_freeze,
1114         .unfreeze_fs            = xfs_fs_unfreeze,
1115         .statfs                 = xfs_fs_statfs,
1116         .show_options           = xfs_fs_show_options,
1117         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1118         .free_cached_objects    = xfs_fs_free_cached_objects,
1119 };
1120
1121 static int
1122 suffix_kstrtoint(
1123         const char      *s,
1124         unsigned int    base,
1125         int             *res)
1126 {
1127         int             last, shift_left_factor = 0, _res;
1128         char            *value;
1129         int             ret = 0;
1130
1131         value = kstrdup(s, GFP_KERNEL);
1132         if (!value)
1133                 return -ENOMEM;
1134
1135         last = strlen(value) - 1;
1136         if (value[last] == 'K' || value[last] == 'k') {
1137                 shift_left_factor = 10;
1138                 value[last] = '\0';
1139         }
1140         if (value[last] == 'M' || value[last] == 'm') {
1141                 shift_left_factor = 20;
1142                 value[last] = '\0';
1143         }
1144         if (value[last] == 'G' || value[last] == 'g') {
1145                 shift_left_factor = 30;
1146                 value[last] = '\0';
1147         }
1148
1149         if (kstrtoint(value, base, &_res))
1150                 ret = -EINVAL;
1151         kfree(value);
1152         *res = _res << shift_left_factor;
1153         return ret;
1154 }
1155
1156 static inline void
1157 xfs_fs_warn_deprecated(
1158         struct fs_context       *fc,
1159         struct fs_parameter     *param,
1160         uint64_t                flag,
1161         bool                    value)
1162 {
1163         /* Don't print the warning if reconfiguring and current mount point
1164          * already had the flag set
1165          */
1166         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1167             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1168                 return;
1169         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1170 }
1171
1172 /*
1173  * Set mount state from a mount option.
1174  *
1175  * NOTE: mp->m_super is NULL here!
1176  */
1177 static int
1178 xfs_fs_parse_param(
1179         struct fs_context       *fc,
1180         struct fs_parameter     *param)
1181 {
1182         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1183         struct fs_parse_result  result;
1184         int                     size = 0;
1185         int                     opt;
1186
1187         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1188         if (opt < 0)
1189                 return opt;
1190
1191         switch (opt) {
1192         case Opt_logbufs:
1193                 parsing_mp->m_logbufs = result.uint_32;
1194                 return 0;
1195         case Opt_logbsize:
1196                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1197                         return -EINVAL;
1198                 return 0;
1199         case Opt_logdev:
1200                 kfree(parsing_mp->m_logname);
1201                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1202                 if (!parsing_mp->m_logname)
1203                         return -ENOMEM;
1204                 return 0;
1205         case Opt_rtdev:
1206                 kfree(parsing_mp->m_rtname);
1207                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1208                 if (!parsing_mp->m_rtname)
1209                         return -ENOMEM;
1210                 return 0;
1211         case Opt_allocsize:
1212                 if (suffix_kstrtoint(param->string, 10, &size))
1213                         return -EINVAL;
1214                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1215                 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1216                 return 0;
1217         case Opt_grpid:
1218         case Opt_bsdgroups:
1219                 parsing_mp->m_features |= XFS_FEAT_GRPID;
1220                 return 0;
1221         case Opt_nogrpid:
1222         case Opt_sysvgroups:
1223                 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1224                 return 0;
1225         case Opt_wsync:
1226                 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1227                 return 0;
1228         case Opt_norecovery:
1229                 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1230                 return 0;
1231         case Opt_noalign:
1232                 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1233                 return 0;
1234         case Opt_swalloc:
1235                 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1236                 return 0;
1237         case Opt_sunit:
1238                 parsing_mp->m_dalign = result.uint_32;
1239                 return 0;
1240         case Opt_swidth:
1241                 parsing_mp->m_swidth = result.uint_32;
1242                 return 0;
1243         case Opt_inode32:
1244                 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1245                 return 0;
1246         case Opt_inode64:
1247                 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1248                 return 0;
1249         case Opt_nouuid:
1250                 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1251                 return 0;
1252         case Opt_largeio:
1253                 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1254                 return 0;
1255         case Opt_nolargeio:
1256                 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1257                 return 0;
1258         case Opt_filestreams:
1259                 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1260                 return 0;
1261         case Opt_noquota:
1262                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1263                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1264                 return 0;
1265         case Opt_quota:
1266         case Opt_uquota:
1267         case Opt_usrquota:
1268                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1269                 return 0;
1270         case Opt_qnoenforce:
1271         case Opt_uqnoenforce:
1272                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1273                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1274                 return 0;
1275         case Opt_pquota:
1276         case Opt_prjquota:
1277                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1278                 return 0;
1279         case Opt_pqnoenforce:
1280                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1281                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1282                 return 0;
1283         case Opt_gquota:
1284         case Opt_grpquota:
1285                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1286                 return 0;
1287         case Opt_gqnoenforce:
1288                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1289                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1290                 return 0;
1291         case Opt_discard:
1292                 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1293                 return 0;
1294         case Opt_nodiscard:
1295                 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1296                 return 0;
1297 #ifdef CONFIG_FS_DAX
1298         case Opt_dax:
1299                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1300                 return 0;
1301         case Opt_dax_enum:
1302                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1303                 return 0;
1304 #endif
1305         /* Following mount options will be removed in September 2025 */
1306         case Opt_ikeep:
1307                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1308                 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1309                 return 0;
1310         case Opt_noikeep:
1311                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1312                 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1313                 return 0;
1314         case Opt_attr2:
1315                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1316                 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1317                 return 0;
1318         case Opt_noattr2:
1319                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1320                 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1321                 return 0;
1322         default:
1323                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1324                 return -EINVAL;
1325         }
1326
1327         return 0;
1328 }
1329
1330 static int
1331 xfs_fs_validate_params(
1332         struct xfs_mount        *mp)
1333 {
1334         /* No recovery flag requires a read-only mount */
1335         if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1336                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1337                 return -EINVAL;
1338         }
1339
1340         /*
1341          * We have not read the superblock at this point, so only the attr2
1342          * mount option can set the attr2 feature by this stage.
1343          */
1344         if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1345                 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1346                 return -EINVAL;
1347         }
1348
1349
1350         if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1351                 xfs_warn(mp,
1352         "sunit and swidth options incompatible with the noalign option");
1353                 return -EINVAL;
1354         }
1355
1356         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1357                 xfs_warn(mp, "quota support not available in this kernel.");
1358                 return -EINVAL;
1359         }
1360
1361         if ((mp->m_dalign && !mp->m_swidth) ||
1362             (!mp->m_dalign && mp->m_swidth)) {
1363                 xfs_warn(mp, "sunit and swidth must be specified together");
1364                 return -EINVAL;
1365         }
1366
1367         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1368                 xfs_warn(mp,
1369         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1370                         mp->m_swidth, mp->m_dalign);
1371                 return -EINVAL;
1372         }
1373
1374         if (mp->m_logbufs != -1 &&
1375             mp->m_logbufs != 0 &&
1376             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1377              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1378                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1379                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1380                 return -EINVAL;
1381         }
1382
1383         if (mp->m_logbsize != -1 &&
1384             mp->m_logbsize !=  0 &&
1385             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1386              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1387              !is_power_of_2(mp->m_logbsize))) {
1388                 xfs_warn(mp,
1389                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1390                         mp->m_logbsize);
1391                 return -EINVAL;
1392         }
1393
1394         if (xfs_has_allocsize(mp) &&
1395             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1396              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1397                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1398                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1399                 return -EINVAL;
1400         }
1401
1402         return 0;
1403 }
1404
1405 static int
1406 xfs_fs_fill_super(
1407         struct super_block      *sb,
1408         struct fs_context       *fc)
1409 {
1410         struct xfs_mount        *mp = sb->s_fs_info;
1411         struct inode            *root;
1412         int                     flags = 0, error;
1413
1414         mp->m_super = sb;
1415
1416         error = xfs_fs_validate_params(mp);
1417         if (error)
1418                 goto out_free_names;
1419
1420         sb_min_blocksize(sb, BBSIZE);
1421         sb->s_xattr = xfs_xattr_handlers;
1422         sb->s_export_op = &xfs_export_operations;
1423 #ifdef CONFIG_XFS_QUOTA
1424         sb->s_qcop = &xfs_quotactl_operations;
1425         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1426 #endif
1427         sb->s_op = &xfs_super_operations;
1428
1429         /*
1430          * Delay mount work if the debug hook is set. This is debug
1431          * instrumention to coordinate simulation of xfs mount failures with
1432          * VFS superblock operations
1433          */
1434         if (xfs_globals.mount_delay) {
1435                 xfs_notice(mp, "Delaying mount for %d seconds.",
1436                         xfs_globals.mount_delay);
1437                 msleep(xfs_globals.mount_delay * 1000);
1438         }
1439
1440         if (fc->sb_flags & SB_SILENT)
1441                 flags |= XFS_MFSI_QUIET;
1442
1443         error = xfs_open_devices(mp);
1444         if (error)
1445                 goto out_free_names;
1446
1447         error = xfs_init_mount_workqueues(mp);
1448         if (error)
1449                 goto out_close_devices;
1450
1451         error = xfs_init_percpu_counters(mp);
1452         if (error)
1453                 goto out_destroy_workqueues;
1454
1455         error = xfs_inodegc_init_percpu(mp);
1456         if (error)
1457                 goto out_destroy_counters;
1458
1459         /*
1460          * All percpu data structures requiring cleanup when a cpu goes offline
1461          * must be allocated before adding this @mp to the cpu-dead handler's
1462          * mount list.
1463          */
1464         xfs_mount_list_add(mp);
1465
1466         /* Allocate stats memory before we do operations that might use it */
1467         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1468         if (!mp->m_stats.xs_stats) {
1469                 error = -ENOMEM;
1470                 goto out_destroy_inodegc;
1471         }
1472
1473         error = xfs_readsb(mp, flags);
1474         if (error)
1475                 goto out_free_stats;
1476
1477         error = xfs_finish_flags(mp);
1478         if (error)
1479                 goto out_free_sb;
1480
1481         error = xfs_setup_devices(mp);
1482         if (error)
1483                 goto out_free_sb;
1484
1485         /* V4 support is undergoing deprecation. */
1486         if (!xfs_has_crc(mp)) {
1487 #ifdef CONFIG_XFS_SUPPORT_V4
1488                 xfs_warn_once(mp,
1489         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1490 #else
1491                 xfs_warn(mp,
1492         "Deprecated V4 format (crc=0) not supported by kernel.");
1493                 error = -EINVAL;
1494                 goto out_free_sb;
1495 #endif
1496         }
1497
1498         /* Filesystem claims it needs repair, so refuse the mount. */
1499         if (xfs_has_needsrepair(mp)) {
1500                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1501                 error = -EFSCORRUPTED;
1502                 goto out_free_sb;
1503         }
1504
1505         /*
1506          * Don't touch the filesystem if a user tool thinks it owns the primary
1507          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1508          * we don't check them at all.
1509          */
1510         if (mp->m_sb.sb_inprogress) {
1511                 xfs_warn(mp, "Offline file system operation in progress!");
1512                 error = -EFSCORRUPTED;
1513                 goto out_free_sb;
1514         }
1515
1516         /*
1517          * Until this is fixed only page-sized or smaller data blocks work.
1518          */
1519         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1520                 xfs_warn(mp,
1521                 "File system with blocksize %d bytes. "
1522                 "Only pagesize (%ld) or less will currently work.",
1523                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1524                 error = -ENOSYS;
1525                 goto out_free_sb;
1526         }
1527
1528         /* Ensure this filesystem fits in the page cache limits */
1529         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1530             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1531                 xfs_warn(mp,
1532                 "file system too large to be mounted on this system.");
1533                 error = -EFBIG;
1534                 goto out_free_sb;
1535         }
1536
1537         /*
1538          * XFS block mappings use 54 bits to store the logical block offset.
1539          * This should suffice to handle the maximum file size that the VFS
1540          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1541          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1542          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1543          * to check this assertion.
1544          *
1545          * Avoid integer overflow by comparing the maximum bmbt offset to the
1546          * maximum pagecache offset in units of fs blocks.
1547          */
1548         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1549                 xfs_warn(mp,
1550 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1551                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1552                          XFS_MAX_FILEOFF);
1553                 error = -EINVAL;
1554                 goto out_free_sb;
1555         }
1556
1557         error = xfs_filestream_mount(mp);
1558         if (error)
1559                 goto out_free_sb;
1560
1561         /*
1562          * we must configure the block size in the superblock before we run the
1563          * full mount process as the mount process can lookup and cache inodes.
1564          */
1565         sb->s_magic = XFS_SUPER_MAGIC;
1566         sb->s_blocksize = mp->m_sb.sb_blocksize;
1567         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1568         sb->s_maxbytes = MAX_LFS_FILESIZE;
1569         sb->s_max_links = XFS_MAXLINK;
1570         sb->s_time_gran = 1;
1571         if (xfs_has_bigtime(mp)) {
1572                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1573                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1574         } else {
1575                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1576                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1577         }
1578         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1579         sb->s_iflags |= SB_I_CGROUPWB;
1580
1581         set_posix_acl_flag(sb);
1582
1583         /* version 5 superblocks support inode version counters. */
1584         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1585                 sb->s_flags |= SB_I_VERSION;
1586
1587         if (xfs_has_dax_always(mp)) {
1588                 bool rtdev_is_dax = false, datadev_is_dax;
1589
1590                 xfs_warn(mp,
1591                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1592
1593                 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1594                         sb->s_blocksize);
1595                 if (mp->m_rtdev_targp)
1596                         rtdev_is_dax = bdev_dax_supported(
1597                                 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1598                 if (!rtdev_is_dax && !datadev_is_dax) {
1599                         xfs_alert(mp,
1600                         "DAX unsupported by block device. Turning off DAX.");
1601                         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1602                 }
1603                 if (xfs_has_reflink(mp)) {
1604                         xfs_alert(mp,
1605                 "DAX and reflink cannot be used together!");
1606                         error = -EINVAL;
1607                         goto out_filestream_unmount;
1608                 }
1609         }
1610
1611         if (xfs_has_discard(mp)) {
1612                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1613
1614                 if (!blk_queue_discard(q)) {
1615                         xfs_warn(mp, "mounting with \"discard\" option, but "
1616                                         "the device does not support discard");
1617                         mp->m_features &= ~XFS_FEAT_DISCARD;
1618                 }
1619         }
1620
1621         if (xfs_has_reflink(mp)) {
1622                 if (mp->m_sb.sb_rblocks) {
1623                         xfs_alert(mp,
1624         "reflink not compatible with realtime device!");
1625                         error = -EINVAL;
1626                         goto out_filestream_unmount;
1627                 }
1628
1629                 if (xfs_globals.always_cow) {
1630                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1631                         mp->m_always_cow = true;
1632                 }
1633         }
1634
1635         if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1636                 xfs_alert(mp,
1637         "reverse mapping btree not compatible with realtime device!");
1638                 error = -EINVAL;
1639                 goto out_filestream_unmount;
1640         }
1641
1642         error = xfs_mountfs(mp);
1643         if (error)
1644                 goto out_filestream_unmount;
1645
1646         root = igrab(VFS_I(mp->m_rootip));
1647         if (!root) {
1648                 error = -ENOENT;
1649                 goto out_unmount;
1650         }
1651         sb->s_root = d_make_root(root);
1652         if (!sb->s_root) {
1653                 error = -ENOMEM;
1654                 goto out_unmount;
1655         }
1656
1657         return 0;
1658
1659  out_filestream_unmount:
1660         xfs_filestream_unmount(mp);
1661  out_free_sb:
1662         xfs_freesb(mp);
1663  out_free_stats:
1664         free_percpu(mp->m_stats.xs_stats);
1665  out_destroy_inodegc:
1666         xfs_mount_list_del(mp);
1667         xfs_inodegc_free_percpu(mp);
1668  out_destroy_counters:
1669         xfs_destroy_percpu_counters(mp);
1670  out_destroy_workqueues:
1671         xfs_destroy_mount_workqueues(mp);
1672  out_close_devices:
1673         xfs_close_devices(mp);
1674  out_free_names:
1675         sb->s_fs_info = NULL;
1676         xfs_mount_free(mp);
1677         return error;
1678
1679  out_unmount:
1680         xfs_filestream_unmount(mp);
1681         xfs_unmountfs(mp);
1682         goto out_free_sb;
1683 }
1684
1685 static int
1686 xfs_fs_get_tree(
1687         struct fs_context       *fc)
1688 {
1689         return get_tree_bdev(fc, xfs_fs_fill_super);
1690 }
1691
1692 static int
1693 xfs_remount_rw(
1694         struct xfs_mount        *mp)
1695 {
1696         struct xfs_sb           *sbp = &mp->m_sb;
1697         int error;
1698
1699         if (xfs_has_norecovery(mp)) {
1700                 xfs_warn(mp,
1701                         "ro->rw transition prohibited on norecovery mount");
1702                 return -EINVAL;
1703         }
1704
1705         if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1706             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1707                 xfs_warn(mp,
1708         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1709                         (sbp->sb_features_ro_compat &
1710                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1711                 return -EINVAL;
1712         }
1713
1714         clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1715
1716         /*
1717          * If this is the first remount to writeable state we might have some
1718          * superblock changes to update.
1719          */
1720         if (mp->m_update_sb) {
1721                 error = xfs_sync_sb(mp, false);
1722                 if (error) {
1723                         xfs_warn(mp, "failed to write sb changes");
1724                         return error;
1725                 }
1726                 mp->m_update_sb = false;
1727         }
1728
1729         /*
1730          * Fill out the reserve pool if it is empty. Use the stashed value if
1731          * it is non-zero, otherwise go with the default.
1732          */
1733         xfs_restore_resvblks(mp);
1734         xfs_log_work_queue(mp);
1735
1736         /* Recover any CoW blocks that never got remapped. */
1737         error = xfs_reflink_recover_cow(mp);
1738         if (error) {
1739                 xfs_err(mp,
1740                         "Error %d recovering leftover CoW allocations.", error);
1741                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1742                 return error;
1743         }
1744         xfs_blockgc_start(mp);
1745
1746         /* Create the per-AG metadata reservation pool .*/
1747         error = xfs_fs_reserve_ag_blocks(mp);
1748         if (error && error != -ENOSPC)
1749                 return error;
1750
1751         /* Re-enable the background inode inactivation worker. */
1752         xfs_inodegc_start(mp);
1753
1754         return 0;
1755 }
1756
1757 static int
1758 xfs_remount_ro(
1759         struct xfs_mount        *mp)
1760 {
1761         int error;
1762
1763         /*
1764          * Cancel background eofb scanning so it cannot race with the final
1765          * log force+buftarg wait and deadlock the remount.
1766          */
1767         xfs_blockgc_stop(mp);
1768
1769         /* Get rid of any leftover CoW reservations... */
1770         error = xfs_blockgc_free_space(mp, NULL);
1771         if (error) {
1772                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1773                 return error;
1774         }
1775
1776         /*
1777          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1778          * flushed all pending inodegc work when it sync'd the filesystem.
1779          * The VFS holds s_umount, so we know that inodes cannot enter
1780          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1781          * we send inodes straight to reclaim, so no inodes will be queued.
1782          */
1783         xfs_inodegc_stop(mp);
1784
1785         /* Free the per-AG metadata reservation pool. */
1786         error = xfs_fs_unreserve_ag_blocks(mp);
1787         if (error) {
1788                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1789                 return error;
1790         }
1791
1792         /*
1793          * Before we sync the metadata, we need to free up the reserve block
1794          * pool so that the used block count in the superblock on disk is
1795          * correct at the end of the remount. Stash the current* reserve pool
1796          * size so that if we get remounted rw, we can return it to the same
1797          * size.
1798          */
1799         xfs_save_resvblks(mp);
1800
1801         xfs_log_clean(mp);
1802         set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1803
1804         return 0;
1805 }
1806
1807 /*
1808  * Logically we would return an error here to prevent users from believing
1809  * they might have changed mount options using remount which can't be changed.
1810  *
1811  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1812  * arguments in some cases so we can't blindly reject options, but have to
1813  * check for each specified option if it actually differs from the currently
1814  * set option and only reject it if that's the case.
1815  *
1816  * Until that is implemented we return success for every remount request, and
1817  * silently ignore all options that we can't actually change.
1818  */
1819 static int
1820 xfs_fs_reconfigure(
1821         struct fs_context *fc)
1822 {
1823         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1824         struct xfs_mount        *new_mp = fc->s_fs_info;
1825         xfs_sb_t                *sbp = &mp->m_sb;
1826         int                     flags = fc->sb_flags;
1827         int                     error;
1828
1829         /* version 5 superblocks always support version counters. */
1830         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1831                 fc->sb_flags |= SB_I_VERSION;
1832
1833         error = xfs_fs_validate_params(new_mp);
1834         if (error)
1835                 return error;
1836
1837         sync_filesystem(mp->m_super);
1838
1839         /* inode32 -> inode64 */
1840         if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1841                 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1842                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1843         }
1844
1845         /* inode64 -> inode32 */
1846         if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1847                 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1848                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1849         }
1850
1851         /* ro -> rw */
1852         if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1853                 error = xfs_remount_rw(mp);
1854                 if (error)
1855                         return error;
1856         }
1857
1858         /* rw -> ro */
1859         if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1860                 error = xfs_remount_ro(mp);
1861                 if (error)
1862                         return error;
1863         }
1864
1865         return 0;
1866 }
1867
1868 static void xfs_fs_free(
1869         struct fs_context       *fc)
1870 {
1871         struct xfs_mount        *mp = fc->s_fs_info;
1872
1873         /*
1874          * mp is stored in the fs_context when it is initialized.
1875          * mp is transferred to the superblock on a successful mount,
1876          * but if an error occurs before the transfer we have to free
1877          * it here.
1878          */
1879         if (mp)
1880                 xfs_mount_free(mp);
1881 }
1882
1883 static const struct fs_context_operations xfs_context_ops = {
1884         .parse_param = xfs_fs_parse_param,
1885         .get_tree    = xfs_fs_get_tree,
1886         .reconfigure = xfs_fs_reconfigure,
1887         .free        = xfs_fs_free,
1888 };
1889
1890 static int xfs_init_fs_context(
1891         struct fs_context       *fc)
1892 {
1893         struct xfs_mount        *mp;
1894
1895         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1896         if (!mp)
1897                 return -ENOMEM;
1898
1899         spin_lock_init(&mp->m_sb_lock);
1900         spin_lock_init(&mp->m_agirotor_lock);
1901         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1902         spin_lock_init(&mp->m_perag_lock);
1903         mutex_init(&mp->m_growlock);
1904         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1905         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1906         mp->m_kobj.kobject.kset = xfs_kset;
1907         /*
1908          * We don't create the finobt per-ag space reservation until after log
1909          * recovery, so we must set this to true so that an ifree transaction
1910          * started during log recovery will not depend on space reservations
1911          * for finobt expansion.
1912          */
1913         mp->m_finobt_nores = true;
1914
1915         /*
1916          * These can be overridden by the mount option parsing.
1917          */
1918         mp->m_logbufs = -1;
1919         mp->m_logbsize = -1;
1920         mp->m_allocsize_log = 16; /* 64k */
1921
1922         /*
1923          * Copy binary VFS mount flags we are interested in.
1924          */
1925         if (fc->sb_flags & SB_RDONLY)
1926                 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1927         if (fc->sb_flags & SB_DIRSYNC)
1928                 mp->m_features |= XFS_FEAT_DIRSYNC;
1929         if (fc->sb_flags & SB_SYNCHRONOUS)
1930                 mp->m_features |= XFS_FEAT_WSYNC;
1931
1932         fc->s_fs_info = mp;
1933         fc->ops = &xfs_context_ops;
1934
1935         return 0;
1936 }
1937
1938 static struct file_system_type xfs_fs_type = {
1939         .owner                  = THIS_MODULE,
1940         .name                   = "xfs",
1941         .init_fs_context        = xfs_init_fs_context,
1942         .parameters             = xfs_fs_parameters,
1943         .kill_sb                = kill_block_super,
1944         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1945 };
1946 MODULE_ALIAS_FS("xfs");
1947
1948 STATIC int __init
1949 xfs_init_zones(void)
1950 {
1951         xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1952                                                 sizeof(struct xlog_ticket),
1953                                                 0, 0, NULL);
1954         if (!xfs_log_ticket_zone)
1955                 goto out;
1956
1957         xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1958                                         sizeof(struct xfs_extent_free_item),
1959                                         0, 0, NULL);
1960         if (!xfs_bmap_free_item_zone)
1961                 goto out_destroy_log_ticket_zone;
1962
1963         xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1964                                                sizeof(struct xfs_btree_cur),
1965                                                0, 0, NULL);
1966         if (!xfs_btree_cur_zone)
1967                 goto out_destroy_bmap_free_item_zone;
1968
1969         xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1970                                               sizeof(struct xfs_da_state),
1971                                               0, 0, NULL);
1972         if (!xfs_da_state_zone)
1973                 goto out_destroy_btree_cur_zone;
1974
1975         xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1976                                            sizeof(struct xfs_ifork),
1977                                            0, 0, NULL);
1978         if (!xfs_ifork_zone)
1979                 goto out_destroy_da_state_zone;
1980
1981         xfs_trans_zone = kmem_cache_create("xfs_trans",
1982                                            sizeof(struct xfs_trans),
1983                                            0, 0, NULL);
1984         if (!xfs_trans_zone)
1985                 goto out_destroy_ifork_zone;
1986
1987
1988         /*
1989          * The size of the zone allocated buf log item is the maximum
1990          * size possible under XFS.  This wastes a little bit of memory,
1991          * but it is much faster.
1992          */
1993         xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1994                                               sizeof(struct xfs_buf_log_item),
1995                                               0, 0, NULL);
1996         if (!xfs_buf_item_zone)
1997                 goto out_destroy_trans_zone;
1998
1999         xfs_efd_zone = kmem_cache_create("xfs_efd_item",
2000                                         (sizeof(struct xfs_efd_log_item) +
2001                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
2002                                         sizeof(struct xfs_extent)),
2003                                         0, 0, NULL);
2004         if (!xfs_efd_zone)
2005                 goto out_destroy_buf_item_zone;
2006
2007         xfs_efi_zone = kmem_cache_create("xfs_efi_item",
2008                                          (sizeof(struct xfs_efi_log_item) +
2009                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2010                                          sizeof(struct xfs_extent)),
2011                                          0, 0, NULL);
2012         if (!xfs_efi_zone)
2013                 goto out_destroy_efd_zone;
2014
2015         xfs_inode_zone = kmem_cache_create("xfs_inode",
2016                                            sizeof(struct xfs_inode), 0,
2017                                            (SLAB_HWCACHE_ALIGN |
2018                                             SLAB_RECLAIM_ACCOUNT |
2019                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2020                                            xfs_fs_inode_init_once);
2021         if (!xfs_inode_zone)
2022                 goto out_destroy_efi_zone;
2023
2024         xfs_ili_zone = kmem_cache_create("xfs_ili",
2025                                          sizeof(struct xfs_inode_log_item), 0,
2026                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2027                                          NULL);
2028         if (!xfs_ili_zone)
2029                 goto out_destroy_inode_zone;
2030
2031         xfs_icreate_zone = kmem_cache_create("xfs_icr",
2032                                              sizeof(struct xfs_icreate_item),
2033                                              0, 0, NULL);
2034         if (!xfs_icreate_zone)
2035                 goto out_destroy_ili_zone;
2036
2037         xfs_rud_zone = kmem_cache_create("xfs_rud_item",
2038                                          sizeof(struct xfs_rud_log_item),
2039                                          0, 0, NULL);
2040         if (!xfs_rud_zone)
2041                 goto out_destroy_icreate_zone;
2042
2043         xfs_rui_zone = kmem_cache_create("xfs_rui_item",
2044                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2045                         0, 0, NULL);
2046         if (!xfs_rui_zone)
2047                 goto out_destroy_rud_zone;
2048
2049         xfs_cud_zone = kmem_cache_create("xfs_cud_item",
2050                                          sizeof(struct xfs_cud_log_item),
2051                                          0, 0, NULL);
2052         if (!xfs_cud_zone)
2053                 goto out_destroy_rui_zone;
2054
2055         xfs_cui_zone = kmem_cache_create("xfs_cui_item",
2056                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2057                         0, 0, NULL);
2058         if (!xfs_cui_zone)
2059                 goto out_destroy_cud_zone;
2060
2061         xfs_bud_zone = kmem_cache_create("xfs_bud_item",
2062                                          sizeof(struct xfs_bud_log_item),
2063                                          0, 0, NULL);
2064         if (!xfs_bud_zone)
2065                 goto out_destroy_cui_zone;
2066
2067         xfs_bui_zone = kmem_cache_create("xfs_bui_item",
2068                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2069                         0, 0, NULL);
2070         if (!xfs_bui_zone)
2071                 goto out_destroy_bud_zone;
2072
2073         return 0;
2074
2075  out_destroy_bud_zone:
2076         kmem_cache_destroy(xfs_bud_zone);
2077  out_destroy_cui_zone:
2078         kmem_cache_destroy(xfs_cui_zone);
2079  out_destroy_cud_zone:
2080         kmem_cache_destroy(xfs_cud_zone);
2081  out_destroy_rui_zone:
2082         kmem_cache_destroy(xfs_rui_zone);
2083  out_destroy_rud_zone:
2084         kmem_cache_destroy(xfs_rud_zone);
2085  out_destroy_icreate_zone:
2086         kmem_cache_destroy(xfs_icreate_zone);
2087  out_destroy_ili_zone:
2088         kmem_cache_destroy(xfs_ili_zone);
2089  out_destroy_inode_zone:
2090         kmem_cache_destroy(xfs_inode_zone);
2091  out_destroy_efi_zone:
2092         kmem_cache_destroy(xfs_efi_zone);
2093  out_destroy_efd_zone:
2094         kmem_cache_destroy(xfs_efd_zone);
2095  out_destroy_buf_item_zone:
2096         kmem_cache_destroy(xfs_buf_item_zone);
2097  out_destroy_trans_zone:
2098         kmem_cache_destroy(xfs_trans_zone);
2099  out_destroy_ifork_zone:
2100         kmem_cache_destroy(xfs_ifork_zone);
2101  out_destroy_da_state_zone:
2102         kmem_cache_destroy(xfs_da_state_zone);
2103  out_destroy_btree_cur_zone:
2104         kmem_cache_destroy(xfs_btree_cur_zone);
2105  out_destroy_bmap_free_item_zone:
2106         kmem_cache_destroy(xfs_bmap_free_item_zone);
2107  out_destroy_log_ticket_zone:
2108         kmem_cache_destroy(xfs_log_ticket_zone);
2109  out:
2110         return -ENOMEM;
2111 }
2112
2113 STATIC void
2114 xfs_destroy_zones(void)
2115 {
2116         /*
2117          * Make sure all delayed rcu free are flushed before we
2118          * destroy caches.
2119          */
2120         rcu_barrier();
2121         kmem_cache_destroy(xfs_bui_zone);
2122         kmem_cache_destroy(xfs_bud_zone);
2123         kmem_cache_destroy(xfs_cui_zone);
2124         kmem_cache_destroy(xfs_cud_zone);
2125         kmem_cache_destroy(xfs_rui_zone);
2126         kmem_cache_destroy(xfs_rud_zone);
2127         kmem_cache_destroy(xfs_icreate_zone);
2128         kmem_cache_destroy(xfs_ili_zone);
2129         kmem_cache_destroy(xfs_inode_zone);
2130         kmem_cache_destroy(xfs_efi_zone);
2131         kmem_cache_destroy(xfs_efd_zone);
2132         kmem_cache_destroy(xfs_buf_item_zone);
2133         kmem_cache_destroy(xfs_trans_zone);
2134         kmem_cache_destroy(xfs_ifork_zone);
2135         kmem_cache_destroy(xfs_da_state_zone);
2136         kmem_cache_destroy(xfs_btree_cur_zone);
2137         kmem_cache_destroy(xfs_bmap_free_item_zone);
2138         kmem_cache_destroy(xfs_log_ticket_zone);
2139 }
2140
2141 STATIC int __init
2142 xfs_init_workqueues(void)
2143 {
2144         /*
2145          * The allocation workqueue can be used in memory reclaim situations
2146          * (writepage path), and parallelism is only limited by the number of
2147          * AGs in all the filesystems mounted. Hence use the default large
2148          * max_active value for this workqueue.
2149          */
2150         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2151                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2152         if (!xfs_alloc_wq)
2153                 return -ENOMEM;
2154
2155         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2156                         0);
2157         if (!xfs_discard_wq)
2158                 goto out_free_alloc_wq;
2159
2160         return 0;
2161 out_free_alloc_wq:
2162         destroy_workqueue(xfs_alloc_wq);
2163         return -ENOMEM;
2164 }
2165
2166 STATIC void
2167 xfs_destroy_workqueues(void)
2168 {
2169         destroy_workqueue(xfs_discard_wq);
2170         destroy_workqueue(xfs_alloc_wq);
2171 }
2172
2173 #ifdef CONFIG_HOTPLUG_CPU
2174 static int
2175 xfs_cpu_dead(
2176         unsigned int            cpu)
2177 {
2178         struct xfs_mount        *mp, *n;
2179
2180         spin_lock(&xfs_mount_list_lock);
2181         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2182                 spin_unlock(&xfs_mount_list_lock);
2183                 xfs_inodegc_cpu_dead(mp, cpu);
2184                 spin_lock(&xfs_mount_list_lock);
2185         }
2186         spin_unlock(&xfs_mount_list_lock);
2187         return 0;
2188 }
2189
2190 static int __init
2191 xfs_cpu_hotplug_init(void)
2192 {
2193         int     error;
2194
2195         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2196                         xfs_cpu_dead);
2197         if (error < 0)
2198                 xfs_alert(NULL,
2199 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2200                         error);
2201         return error;
2202 }
2203
2204 static void
2205 xfs_cpu_hotplug_destroy(void)
2206 {
2207         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2208 }
2209
2210 #else /* !CONFIG_HOTPLUG_CPU */
2211 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2212 static inline void xfs_cpu_hotplug_destroy(void) {}
2213 #endif
2214
2215 STATIC int __init
2216 init_xfs_fs(void)
2217 {
2218         int                     error;
2219
2220         xfs_check_ondisk_structs();
2221
2222         printk(KERN_INFO XFS_VERSION_STRING " with "
2223                          XFS_BUILD_OPTIONS " enabled\n");
2224
2225         xfs_dir_startup();
2226
2227         error = xfs_cpu_hotplug_init();
2228         if (error)
2229                 goto out;
2230
2231         error = xfs_init_zones();
2232         if (error)
2233                 goto out_destroy_hp;
2234
2235         error = xfs_init_workqueues();
2236         if (error)
2237                 goto out_destroy_zones;
2238
2239         error = xfs_mru_cache_init();
2240         if (error)
2241                 goto out_destroy_wq;
2242
2243         error = xfs_buf_init();
2244         if (error)
2245                 goto out_mru_cache_uninit;
2246
2247         error = xfs_init_procfs();
2248         if (error)
2249                 goto out_buf_terminate;
2250
2251         error = xfs_sysctl_register();
2252         if (error)
2253                 goto out_cleanup_procfs;
2254
2255         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2256         if (!xfs_kset) {
2257                 error = -ENOMEM;
2258                 goto out_sysctl_unregister;
2259         }
2260
2261         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2262
2263         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2264         if (!xfsstats.xs_stats) {
2265                 error = -ENOMEM;
2266                 goto out_kset_unregister;
2267         }
2268
2269         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2270                                "stats");
2271         if (error)
2272                 goto out_free_stats;
2273
2274 #ifdef DEBUG
2275         xfs_dbg_kobj.kobject.kset = xfs_kset;
2276         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2277         if (error)
2278                 goto out_remove_stats_kobj;
2279 #endif
2280
2281         error = xfs_qm_init();
2282         if (error)
2283                 goto out_remove_dbg_kobj;
2284
2285         error = register_filesystem(&xfs_fs_type);
2286         if (error)
2287                 goto out_qm_exit;
2288         return 0;
2289
2290  out_qm_exit:
2291         xfs_qm_exit();
2292  out_remove_dbg_kobj:
2293 #ifdef DEBUG
2294         xfs_sysfs_del(&xfs_dbg_kobj);
2295  out_remove_stats_kobj:
2296 #endif
2297         xfs_sysfs_del(&xfsstats.xs_kobj);
2298  out_free_stats:
2299         free_percpu(xfsstats.xs_stats);
2300  out_kset_unregister:
2301         kset_unregister(xfs_kset);
2302  out_sysctl_unregister:
2303         xfs_sysctl_unregister();
2304  out_cleanup_procfs:
2305         xfs_cleanup_procfs();
2306  out_buf_terminate:
2307         xfs_buf_terminate();
2308  out_mru_cache_uninit:
2309         xfs_mru_cache_uninit();
2310  out_destroy_wq:
2311         xfs_destroy_workqueues();
2312  out_destroy_zones:
2313         xfs_destroy_zones();
2314  out_destroy_hp:
2315         xfs_cpu_hotplug_destroy();
2316  out:
2317         return error;
2318 }
2319
2320 STATIC void __exit
2321 exit_xfs_fs(void)
2322 {
2323         xfs_qm_exit();
2324         unregister_filesystem(&xfs_fs_type);
2325 #ifdef DEBUG
2326         xfs_sysfs_del(&xfs_dbg_kobj);
2327 #endif
2328         xfs_sysfs_del(&xfsstats.xs_kobj);
2329         free_percpu(xfsstats.xs_stats);
2330         kset_unregister(xfs_kset);
2331         xfs_sysctl_unregister();
2332         xfs_cleanup_procfs();
2333         xfs_buf_terminate();
2334         xfs_mru_cache_uninit();
2335         xfs_destroy_workqueues();
2336         xfs_destroy_zones();
2337         xfs_uuid_table_free();
2338         xfs_cpu_hotplug_destroy();
2339 }
2340
2341 module_init(init_xfs_fs);
2342 module_exit(exit_xfs_fs);
2343
2344 MODULE_AUTHOR("Silicon Graphics, Inc.");
2345 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2346 MODULE_LICENSE("GPL");