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 #include "xfs_defer.h"
  41 #include "xfs_attr_item.h"
  42 #include "xfs_xattr.h"
  43
  44 #include <linux/magic.h>
  45 #include <linux/fs_context.h>
  46 #include <linux/fs_parser.h>
  47
  48 static const struct super_operations xfs_super_operations;
  49
  50 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  51 #ifdef DEBUG
  52 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  53 #endif
  54
  55 #ifdef CONFIG_HOTPLUG_CPU
  56 static LIST_HEAD(xfs_mount_list);
  57 static DEFINE_SPINLOCK(xfs_mount_list_lock);
  58
  59 static inline void xfs_mount_list_add(struct xfs_mount *mp)
  60 {
  61         spin_lock(&xfs_mount_list_lock);
  62         list_add(&mp->m_mount_list, &xfs_mount_list);
  63         spin_unlock(&xfs_mount_list_lock);
  64 }
  65
  66 static inline void xfs_mount_list_del(struct xfs_mount *mp)
  67 {
  68         spin_lock(&xfs_mount_list_lock);
  69         list_del(&mp->m_mount_list);
  70         spin_unlock(&xfs_mount_list_lock);
  71 }
  72 #else /* !CONFIG_HOTPLUG_CPU */
  73 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
  74 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
  75 #endif
  76
  77 enum xfs_dax_mode {
  78         XFS_DAX_INODE = 0,
  79         XFS_DAX_ALWAYS = 1,
  80         XFS_DAX_NEVER = 2,
  81 };
  82
  83 static void
  84 xfs_mount_set_dax_mode(
  85         struct xfs_mount        *mp,
  86         enum xfs_dax_mode       mode)
  87 {
  88         switch (mode) {
  89         case XFS_DAX_INODE:
  90                 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
  91                 break;
  92         case XFS_DAX_ALWAYS:
  93                 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
  94                 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
  95                 break;
  96         case XFS_DAX_NEVER:
  97                 mp->m_features |= XFS_FEAT_DAX_NEVER;
  98                 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
  99                 break;
 100         }
 101 }
 102
 103 static const struct constant_table dax_param_enums[] = {
 104         {"inode",       XFS_DAX_INODE },
 105         {"always",      XFS_DAX_ALWAYS },
 106         {"never",       XFS_DAX_NEVER },
 107         {}
 108 };
 109
 110 /*
 111  * Table driven mount option parser.
 112  */
 113 enum {
 114         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
 115         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
 116         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
 117         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
 118         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
 119         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
 120         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
 121         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
 122         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
 123 };
 124
 125 static const struct fs_parameter_spec xfs_fs_parameters[] = {
 126         fsparam_u32("logbufs",          Opt_logbufs),
 127         fsparam_string("logbsize",      Opt_logbsize),
 128         fsparam_string("logdev",        Opt_logdev),
 129         fsparam_string("rtdev",         Opt_rtdev),
 130         fsparam_flag("wsync",           Opt_wsync),
 131         fsparam_flag("noalign",         Opt_noalign),
 132         fsparam_flag("swalloc",         Opt_swalloc),
 133         fsparam_u32("sunit",            Opt_sunit),
 134         fsparam_u32("swidth",           Opt_swidth),
 135         fsparam_flag("nouuid",          Opt_nouuid),
 136         fsparam_flag("grpid",           Opt_grpid),
 137         fsparam_flag("nogrpid",         Opt_nogrpid),
 138         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 139         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 140         fsparam_string("allocsize",     Opt_allocsize),
 141         fsparam_flag("norecovery",      Opt_norecovery),
 142         fsparam_flag("inode64",         Opt_inode64),
 143         fsparam_flag("inode32",         Opt_inode32),
 144         fsparam_flag("ikeep",           Opt_ikeep),
 145         fsparam_flag("noikeep",         Opt_noikeep),
 146         fsparam_flag("largeio",         Opt_largeio),
 147         fsparam_flag("nolargeio",       Opt_nolargeio),
 148         fsparam_flag("attr2",           Opt_attr2),
 149         fsparam_flag("noattr2",         Opt_noattr2),
 150         fsparam_flag("filestreams",     Opt_filestreams),
 151         fsparam_flag("quota",           Opt_quota),
 152         fsparam_flag("noquota",         Opt_noquota),
 153         fsparam_flag("usrquota",        Opt_usrquota),
 154         fsparam_flag("grpquota",        Opt_grpquota),
 155         fsparam_flag("prjquota",        Opt_prjquota),
 156         fsparam_flag("uquota",          Opt_uquota),
 157         fsparam_flag("gquota",          Opt_gquota),
 158         fsparam_flag("pquota",          Opt_pquota),
 159         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 160         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 161         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 162         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 163         fsparam_flag("discard",         Opt_discard),
 164         fsparam_flag("nodiscard",       Opt_nodiscard),
 165         fsparam_flag("dax",             Opt_dax),
 166         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 167         {}
 168 };
 169
 170 struct proc_xfs_info {
 171         uint64_t        flag;
 172         char            *str;
 173 };
 174
 175 static int
 176 xfs_fs_show_options(
 177         struct seq_file         *m,
 178         struct dentry           *root)
 179 {
 180         static struct proc_xfs_info xfs_info_set[] = {
 181                 /* the few simple ones we can get from the mount struct */
 182                 { XFS_FEAT_IKEEP,               ",ikeep" },
 183                 { XFS_FEAT_WSYNC,               ",wsync" },
 184                 { XFS_FEAT_NOALIGN,             ",noalign" },
 185                 { XFS_FEAT_SWALLOC,             ",swalloc" },
 186                 { XFS_FEAT_NOUUID,              ",nouuid" },
 187                 { XFS_FEAT_NORECOVERY,          ",norecovery" },
 188                 { XFS_FEAT_ATTR2,               ",attr2" },
 189                 { XFS_FEAT_FILESTREAMS,         ",filestreams" },
 190                 { XFS_FEAT_GRPID,               ",grpid" },
 191                 { XFS_FEAT_DISCARD,             ",discard" },
 192                 { XFS_FEAT_LARGE_IOSIZE,        ",largeio" },
 193                 { XFS_FEAT_DAX_ALWAYS,          ",dax=always" },
 194                 { XFS_FEAT_DAX_NEVER,           ",dax=never" },
 195                 { 0, NULL }
 196         };
 197         struct xfs_mount        *mp = XFS_M(root->d_sb);
 198         struct proc_xfs_info    *xfs_infop;
 199
 200         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 201                 if (mp->m_features & xfs_infop->flag)
 202                         seq_puts(m, xfs_infop->str);
 203         }
 204
 205         seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
 206
 207         if (xfs_has_allocsize(mp))
 208                 seq_printf(m, ",allocsize=%dk",
 209                            (1 << mp->m_allocsize_log) >> 10);
 210
 211         if (mp->m_logbufs > 0)
 212                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 213         if (mp->m_logbsize > 0)
 214                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 215
 216         if (mp->m_logname)
 217                 seq_show_option(m, "logdev", mp->m_logname);
 218         if (mp->m_rtname)
 219                 seq_show_option(m, "rtdev", mp->m_rtname);
 220
 221         if (mp->m_dalign > 0)
 222                 seq_printf(m, ",sunit=%d",
 223                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 224         if (mp->m_swidth > 0)
 225                 seq_printf(m, ",swidth=%d",
 226                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 227
 228         if (mp->m_qflags & XFS_UQUOTA_ENFD)
 229                 seq_puts(m, ",usrquota");
 230         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 231                 seq_puts(m, ",uqnoenforce");
 232
 233         if (mp->m_qflags & XFS_PQUOTA_ENFD)
 234                 seq_puts(m, ",prjquota");
 235         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
 236                 seq_puts(m, ",pqnoenforce");
 237
 238         if (mp->m_qflags & XFS_GQUOTA_ENFD)
 239                 seq_puts(m, ",grpquota");
 240         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
 241                 seq_puts(m, ",gqnoenforce");
 242
 243         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 244                 seq_puts(m, ",noquota");
 245
 246         return 0;
 247 }
 248
 249 /*
 250  * Set parameters for inode allocation heuristics, taking into account
 251  * filesystem size and inode32/inode64 mount options; i.e. specifically
 252  * whether or not XFS_FEAT_SMALL_INUMS is set.
 253  *
 254  * Inode allocation patterns are altered only if inode32 is requested
 255  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
 256  * If altered, XFS_OPSTATE_INODE32 is set as well.
 257  *
 258  * An agcount independent of that in the mount structure is provided
 259  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 260  * to the potentially higher ag count.
 261  *
 262  * Returns the maximum AG index which may contain inodes.
 263  */
 264 xfs_agnumber_t
 265 xfs_set_inode_alloc(
 266         struct xfs_mount *mp,
 267         xfs_agnumber_t  agcount)
 268 {
 269         xfs_agnumber_t  index;
 270         xfs_agnumber_t  maxagi = 0;
 271         xfs_sb_t        *sbp = &mp->m_sb;
 272         xfs_agnumber_t  max_metadata;
 273         xfs_agino_t     agino;
 274         xfs_ino_t       ino;
 275
 276         /*
 277          * Calculate how much should be reserved for inodes to meet
 278          * the max inode percentage.  Used only for inode32.
 279          */
 280         if (M_IGEO(mp)->maxicount) {
 281                 uint64_t        icount;
 282
 283                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 284                 do_div(icount, 100);
 285                 icount += sbp->sb_agblocks - 1;
 286                 do_div(icount, sbp->sb_agblocks);
 287                 max_metadata = icount;
 288         } else {
 289                 max_metadata = agcount;
 290         }
 291
 292         /* Get the last possible inode in the filesystem */
 293         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 294         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 295
 296         /*
 297          * If user asked for no more than 32-bit inodes, and the fs is
 298          * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
 299          * the allocator to accommodate the request.
 300          */
 301         if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
 302                 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 303         else
 304                 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 305
 306         for (index = 0; index < agcount; index++) {
 307                 struct xfs_perag        *pag;
 308
 309                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 310
 311                 pag = xfs_perag_get(mp, index);
 312
 313                 if (xfs_is_inode32(mp)) {
 314                         if (ino > XFS_MAXINUMBER_32) {
 315                                 pag->pagi_inodeok = 0;
 316                                 pag->pagf_metadata = 0;
 317                         } else {
 318                                 pag->pagi_inodeok = 1;
 319                                 maxagi++;
 320                                 if (index < max_metadata)
 321                                         pag->pagf_metadata = 1;
 322                                 else
 323                                         pag->pagf_metadata = 0;
 324                         }
 325                 } else {
 326                         pag->pagi_inodeok = 1;
 327                         pag->pagf_metadata = 0;
 328                 }
 329
 330                 xfs_perag_put(pag);
 331         }
 332
 333         return xfs_is_inode32(mp) ? maxagi : agcount;
 334 }
 335
 336 static int
 337 xfs_setup_dax_always(
 338         struct xfs_mount        *mp)
 339 {
 340         if (!mp->m_ddev_targp->bt_daxdev &&
 341             (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
 342                 xfs_alert(mp,
 343                         "DAX unsupported by block device. Turning off DAX.");
 344                 goto disable_dax;
 345         }
 346
 347         if (mp->m_super->s_blocksize != PAGE_SIZE) {
 348                 xfs_alert(mp,
 349                         "DAX not supported for blocksize. Turning off DAX.");
 350                 goto disable_dax;
 351         }
 352
 353         if (xfs_has_reflink(mp)) {
 354                 xfs_alert(mp, "DAX and reflink cannot be used together!");
 355                 return -EINVAL;
 356         }
 357
 358         xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
 359         return 0;
 360
 361 disable_dax:
 362         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
 363         return 0;
 364 }
 365
 366 STATIC int
 367 xfs_blkdev_get(
 368         xfs_mount_t             *mp,
 369         const char              *name,
 370         struct block_device     **bdevp)
 371 {
 372         int                     error = 0;
 373
 374         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 375                                     mp);
 376         if (IS_ERR(*bdevp)) {
 377                 error = PTR_ERR(*bdevp);
 378                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 379         }
 380
 381         return error;
 382 }
 383
 384 STATIC void
 385 xfs_blkdev_put(
 386         struct block_device     *bdev)
 387 {
 388         if (bdev)
 389                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 390 }
 391
 392 STATIC void
 393 xfs_close_devices(
 394         struct xfs_mount        *mp)
 395 {
 396         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 397                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 398
 399                 xfs_free_buftarg(mp->m_logdev_targp);
 400                 xfs_blkdev_put(logdev);
 401         }
 402         if (mp->m_rtdev_targp) {
 403                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 404
 405                 xfs_free_buftarg(mp->m_rtdev_targp);
 406                 xfs_blkdev_put(rtdev);
 407         }
 408         xfs_free_buftarg(mp->m_ddev_targp);
 409 }
 410
 411 /*
 412  * The file system configurations are:
 413  *      (1) device (partition) with data and internal log
 414  *      (2) logical volume with data and log subvolumes.
 415  *      (3) logical volume with data, log, and realtime subvolumes.
 416  *
 417  * We only have to handle opening the log and realtime volumes here if
 418  * they are present.  The data subvolume has already been opened by
 419  * get_sb_bdev() and is stored in sb->s_bdev.
 420  */
 421 STATIC int
 422 xfs_open_devices(
 423         struct xfs_mount        *mp)
 424 {
 425         struct block_device     *ddev = mp->m_super->s_bdev;
 426         struct block_device     *logdev = NULL, *rtdev = NULL;
 427         int                     error;
 428
 429         /*
 430          * Open real time and log devices - order is important.
 431          */
 432         if (mp->m_logname) {
 433                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 434                 if (error)
 435                         return error;
 436         }
 437
 438         if (mp->m_rtname) {
 439                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 440                 if (error)
 441                         goto out_close_logdev;
 442
 443                 if (rtdev == ddev || rtdev == logdev) {
 444                         xfs_warn(mp,
 445         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 446                         error = -EINVAL;
 447                         goto out_close_rtdev;
 448                 }
 449         }
 450
 451         /*
 452          * Setup xfs_mount buffer target pointers
 453          */
 454         error = -ENOMEM;
 455         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
 456         if (!mp->m_ddev_targp)
 457                 goto out_close_rtdev;
 458
 459         if (rtdev) {
 460                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
 461                 if (!mp->m_rtdev_targp)
 462                         goto out_free_ddev_targ;
 463         }
 464
 465         if (logdev && logdev != ddev) {
 466                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
 467                 if (!mp->m_logdev_targp)
 468                         goto out_free_rtdev_targ;
 469         } else {
 470                 mp->m_logdev_targp = mp->m_ddev_targp;
 471         }
 472
 473         return 0;
 474
 475  out_free_rtdev_targ:
 476         if (mp->m_rtdev_targp)
 477                 xfs_free_buftarg(mp->m_rtdev_targp);
 478  out_free_ddev_targ:
 479         xfs_free_buftarg(mp->m_ddev_targp);
 480  out_close_rtdev:
 481         xfs_blkdev_put(rtdev);
 482  out_close_logdev:
 483         if (logdev && logdev != ddev)
 484                 xfs_blkdev_put(logdev);
 485         return error;
 486 }
 487
 488 /*
 489  * Setup xfs_mount buffer target pointers based on superblock
 490  */
 491 STATIC int
 492 xfs_setup_devices(
 493         struct xfs_mount        *mp)
 494 {
 495         int                     error;
 496
 497         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 498         if (error)
 499                 return error;
 500
 501         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 502                 unsigned int    log_sector_size = BBSIZE;
 503
 504                 if (xfs_has_sector(mp))
 505                         log_sector_size = mp->m_sb.sb_logsectsize;
 506                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 507                                             log_sector_size);
 508                 if (error)
 509                         return error;
 510         }
 511         if (mp->m_rtdev_targp) {
 512                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 513                                             mp->m_sb.sb_sectsize);
 514                 if (error)
 515                         return error;
 516         }
 517
 518         return 0;
 519 }
 520
 521 STATIC int
 522 xfs_init_mount_workqueues(
 523         struct xfs_mount        *mp)
 524 {
 525         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 526                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 527                         1, mp->m_super->s_id);
 528         if (!mp->m_buf_workqueue)
 529                 goto out;
 530
 531         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 532                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 533                         0, mp->m_super->s_id);
 534         if (!mp->m_unwritten_workqueue)
 535                 goto out_destroy_buf;
 536
 537         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 538                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 539                         0, mp->m_super->s_id);
 540         if (!mp->m_reclaim_workqueue)
 541                 goto out_destroy_unwritten;
 542
 543         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
 544                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
 545                         0, mp->m_super->s_id);
 546         if (!mp->m_blockgc_wq)
 547                 goto out_destroy_reclaim;
 548
 549         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
 550                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 551                         1, mp->m_super->s_id);
 552         if (!mp->m_inodegc_wq)
 553                 goto out_destroy_blockgc;
 554
 555         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 556                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 557         if (!mp->m_sync_workqueue)
 558                 goto out_destroy_inodegc;
 559
 560         return 0;
 561
 562 out_destroy_inodegc:
 563         destroy_workqueue(mp->m_inodegc_wq);
 564 out_destroy_blockgc:
 565         destroy_workqueue(mp->m_blockgc_wq);
 566 out_destroy_reclaim:
 567         destroy_workqueue(mp->m_reclaim_workqueue);
 568 out_destroy_unwritten:
 569         destroy_workqueue(mp->m_unwritten_workqueue);
 570 out_destroy_buf:
 571         destroy_workqueue(mp->m_buf_workqueue);
 572 out:
 573         return -ENOMEM;
 574 }
 575
 576 STATIC void
 577 xfs_destroy_mount_workqueues(
 578         struct xfs_mount        *mp)
 579 {
 580         destroy_workqueue(mp->m_sync_workqueue);
 581         destroy_workqueue(mp->m_blockgc_wq);
 582         destroy_workqueue(mp->m_inodegc_wq);
 583         destroy_workqueue(mp->m_reclaim_workqueue);
 584         destroy_workqueue(mp->m_unwritten_workqueue);
 585         destroy_workqueue(mp->m_buf_workqueue);
 586 }
 587
 588 static void
 589 xfs_flush_inodes_worker(
 590         struct work_struct      *work)
 591 {
 592         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 593                                                    m_flush_inodes_work);
 594         struct super_block      *sb = mp->m_super;
 595
 596         if (down_read_trylock(&sb->s_umount)) {
 597                 sync_inodes_sb(sb);
 598                 up_read(&sb->s_umount);
 599         }
 600 }
 601
 602 /*
 603  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 604  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 605  * for IO to complete so that we effectively throttle multiple callers to the
 606  * rate at which IO is completing.
 607  */
 608 void
 609 xfs_flush_inodes(
 610         struct xfs_mount        *mp)
 611 {
 612         /*
 613          * If flush_work() returns true then that means we waited for a flush
 614          * which was already in progress.  Don't bother running another scan.
 615          */
 616         if (flush_work(&mp->m_flush_inodes_work))
 617                 return;
 618
 619         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 620         flush_work(&mp->m_flush_inodes_work);
 621 }
 622
 623 /* Catch misguided souls that try to use this interface on XFS */
 624 STATIC struct inode *
 625 xfs_fs_alloc_inode(
 626         struct super_block      *sb)
 627 {
 628         BUG();
 629         return NULL;
 630 }
 631
 632 /*
 633  * Now that the generic code is guaranteed not to be accessing
 634  * the linux inode, we can inactivate and reclaim the inode.
 635  */
 636 STATIC void
 637 xfs_fs_destroy_inode(
 638         struct inode            *inode)
 639 {
 640         struct xfs_inode        *ip = XFS_I(inode);
 641
 642         trace_xfs_destroy_inode(ip);
 643
 644         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 645         XFS_STATS_INC(ip->i_mount, vn_rele);
 646         XFS_STATS_INC(ip->i_mount, vn_remove);
 647         xfs_inode_mark_reclaimable(ip);
 648 }
 649
 650 static void
 651 xfs_fs_dirty_inode(
 652         struct inode                    *inode,
 653         int                             flag)
 654 {
 655         struct xfs_inode                *ip = XFS_I(inode);
 656         struct xfs_mount                *mp = ip->i_mount;
 657         struct xfs_trans                *tp;
 658
 659         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 660                 return;
 661         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 662                 return;
 663
 664         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 665                 return;
 666         xfs_ilock(ip, XFS_ILOCK_EXCL);
 667         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 668         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 669         xfs_trans_commit(tp);
 670 }
 671
 672 /*
 673  * Slab object creation initialisation for the XFS inode.
 674  * This covers only the idempotent fields in the XFS inode;
 675  * all other fields need to be initialised on allocation
 676  * from the slab. This avoids the need to repeatedly initialise
 677  * fields in the xfs inode that left in the initialise state
 678  * when freeing the inode.
 679  */
 680 STATIC void
 681 xfs_fs_inode_init_once(
 682         void                    *inode)
 683 {
 684         struct xfs_inode        *ip = inode;
 685
 686         memset(ip, 0, sizeof(struct xfs_inode));
 687
 688         /* vfs inode */
 689         inode_init_once(VFS_I(ip));
 690
 691         /* xfs inode */
 692         atomic_set(&ip->i_pincount, 0);
 693         spin_lock_init(&ip->i_flags_lock);
 694
 695         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 696                      "xfsino", ip->i_ino);
 697 }
 698
 699 /*
 700  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 701  * serialised against cache hits here via the inode->i_lock and igrab() in
 702  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 703  * racing with us, and it avoids needing to grab a spinlock here for every inode
 704  * we drop the final reference on.
 705  */
 706 STATIC int
 707 xfs_fs_drop_inode(
 708         struct inode            *inode)
 709 {
 710         struct xfs_inode        *ip = XFS_I(inode);
 711
 712         /*
 713          * If this unlinked inode is in the middle of recovery, don't
 714          * drop the inode just yet; log recovery will take care of
 715          * that.  See the comment for this inode flag.
 716          */
 717         if (ip->i_flags & XFS_IRECOVERY) {
 718                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
 719                 return 0;
 720         }
 721
 722         return generic_drop_inode(inode);
 723 }
 724
 725 static void
 726 xfs_mount_free(
 727         struct xfs_mount        *mp)
 728 {
 729         kfree(mp->m_rtname);
 730         kfree(mp->m_logname);
 731         kmem_free(mp);
 732 }
 733
 734 STATIC int
 735 xfs_fs_sync_fs(
 736         struct super_block      *sb,
 737         int                     wait)
 738 {
 739         struct xfs_mount        *mp = XFS_M(sb);
 740         int                     error;
 741
 742         trace_xfs_fs_sync_fs(mp, __return_address);
 743
 744         /*
 745          * Doing anything during the async pass would be counterproductive.
 746          */
 747         if (!wait)
 748                 return 0;
 749
 750         error = xfs_log_force(mp, XFS_LOG_SYNC);
 751         if (error)
 752                 return error;
 753
 754         if (laptop_mode) {
 755                 /*
 756                  * The disk must be active because we're syncing.
 757                  * We schedule log work now (now that the disk is
 758                  * active) instead of later (when it might not be).
 759                  */
 760                 flush_delayed_work(&mp->m_log->l_work);
 761         }
 762
 763         /*
 764          * If we are called with page faults frozen out, it means we are about
 765          * to freeze the transaction subsystem. Take the opportunity to shut
 766          * down inodegc because once SB_FREEZE_FS is set it's too late to
 767          * prevent inactivation races with freeze. The fs doesn't get called
 768          * again by the freezing process until after SB_FREEZE_FS has been set,
 769          * so it's now or never.  Same logic applies to speculative allocation
 770          * garbage collection.
 771          *
 772          * We don't care if this is a normal syncfs call that does this or
 773          * freeze that does this - we can run this multiple times without issue
 774          * and we won't race with a restart because a restart can only occur
 775          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
 776          */
 777         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
 778                 xfs_inodegc_stop(mp);
 779                 xfs_blockgc_stop(mp);
 780         }
 781
 782         return 0;
 783 }
 784
 785 STATIC int
 786 xfs_fs_statfs(
 787         struct dentry           *dentry,
 788         struct kstatfs          *statp)
 789 {
 790         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 791         xfs_sb_t                *sbp = &mp->m_sb;
 792         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 793         uint64_t                fakeinos, id;
 794         uint64_t                icount;
 795         uint64_t                ifree;
 796         uint64_t                fdblocks;
 797         xfs_extlen_t            lsize;
 798         int64_t                 ffree;
 799
 800         /* Wait for whatever inactivations are in progress. */
 801         xfs_inodegc_flush(mp);
 802
 803         statp->f_type = XFS_SUPER_MAGIC;
 804         statp->f_namelen = MAXNAMELEN - 1;
 805
 806         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 807         statp->f_fsid = u64_to_fsid(id);
 808
 809         icount = percpu_counter_sum(&mp->m_icount);
 810         ifree = percpu_counter_sum(&mp->m_ifree);
 811         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 812
 813         spin_lock(&mp->m_sb_lock);
 814         statp->f_bsize = sbp->sb_blocksize;
 815         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 816         statp->f_blocks = sbp->sb_dblocks - lsize;
 817         spin_unlock(&mp->m_sb_lock);
 818
 819         /* make sure statp->f_bfree does not underflow */
 820         statp->f_bfree = max_t(int64_t, 0,
 821                                 fdblocks - xfs_fdblocks_unavailable(mp));
 822         statp->f_bavail = statp->f_bfree;
 823
 824         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 825         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 826         if (M_IGEO(mp)->maxicount)
 827                 statp->f_files = min_t(typeof(statp->f_files),
 828                                         statp->f_files,
 829                                         M_IGEO(mp)->maxicount);
 830
 831         /* If sb_icount overshot maxicount, report actual allocation */
 832         statp->f_files = max_t(typeof(statp->f_files),
 833                                         statp->f_files,
 834                                         sbp->sb_icount);
 835
 836         /* make sure statp->f_ffree does not underflow */
 837         ffree = statp->f_files - (icount - ifree);
 838         statp->f_ffree = max_t(int64_t, ffree, 0);
 839
 840
 841         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 842             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 843                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 844                 xfs_qm_statvfs(ip, statp);
 845
 846         if (XFS_IS_REALTIME_MOUNT(mp) &&
 847             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 848                 s64     freertx;
 849
 850                 statp->f_blocks = sbp->sb_rblocks;
 851                 freertx = percpu_counter_sum_positive(&mp->m_frextents);
 852                 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
 853         }
 854
 855         return 0;
 856 }
 857
 858 STATIC void
 859 xfs_save_resvblks(struct xfs_mount *mp)
 860 {
 861         uint64_t resblks = 0;
 862
 863         mp->m_resblks_save = mp->m_resblks;
 864         xfs_reserve_blocks(mp, &resblks, NULL);
 865 }
 866
 867 STATIC void
 868 xfs_restore_resvblks(struct xfs_mount *mp)
 869 {
 870         uint64_t resblks;
 871
 872         if (mp->m_resblks_save) {
 873                 resblks = mp->m_resblks_save;
 874                 mp->m_resblks_save = 0;
 875         } else
 876                 resblks = xfs_default_resblks(mp);
 877
 878         xfs_reserve_blocks(mp, &resblks, NULL);
 879 }
 880
 881 /*
 882  * Second stage of a freeze. The data is already frozen so we only
 883  * need to take care of the metadata. Once that's done sync the superblock
 884  * to the log to dirty it in case of a crash while frozen. This ensures that we
 885  * will recover the unlinked inode lists on the next mount.
 886  */
 887 STATIC int
 888 xfs_fs_freeze(
 889         struct super_block      *sb)
 890 {
 891         struct xfs_mount        *mp = XFS_M(sb);
 892         unsigned int            flags;
 893         int                     ret;
 894
 895         /*
 896          * The filesystem is now frozen far enough that memory reclaim
 897          * cannot safely operate on the filesystem. Hence we need to
 898          * set a GFP_NOFS context here to avoid recursion deadlocks.
 899          */
 900         flags = memalloc_nofs_save();
 901         xfs_save_resvblks(mp);
 902         ret = xfs_log_quiesce(mp);
 903         memalloc_nofs_restore(flags);
 904
 905         /*
 906          * For read-write filesystems, we need to restart the inodegc on error
 907          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
 908          * going to be run to restart it now.  We are at SB_FREEZE_FS level
 909          * here, so we can restart safely without racing with a stop in
 910          * xfs_fs_sync_fs().
 911          */
 912         if (ret && !xfs_is_readonly(mp)) {
 913                 xfs_blockgc_start(mp);
 914                 xfs_inodegc_start(mp);
 915         }
 916
 917         return ret;
 918 }
 919
 920 STATIC int
 921 xfs_fs_unfreeze(
 922         struct super_block      *sb)
 923 {
 924         struct xfs_mount        *mp = XFS_M(sb);
 925
 926         xfs_restore_resvblks(mp);
 927         xfs_log_work_queue(mp);
 928
 929         /*
 930          * Don't reactivate the inodegc worker on a readonly filesystem because
 931          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
 932          * worker because there are no speculative preallocations on a readonly
 933          * filesystem.
 934          */
 935         if (!xfs_is_readonly(mp)) {
 936                 xfs_blockgc_start(mp);
 937                 xfs_inodegc_start(mp);
 938         }
 939
 940         return 0;
 941 }
 942
 943 /*
 944  * This function fills in xfs_mount_t fields based on mount args.
 945  * Note: the superblock _has_ now been read in.
 946  */
 947 STATIC int
 948 xfs_finish_flags(
 949         struct xfs_mount        *mp)
 950 {
 951         /* Fail a mount where the logbuf is smaller than the log stripe */
 952         if (xfs_has_logv2(mp)) {
 953                 if (mp->m_logbsize <= 0 &&
 954                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 955                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 956                 } else if (mp->m_logbsize > 0 &&
 957                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 958                         xfs_warn(mp,
 959                 "logbuf size must be greater than or equal to log stripe size");
 960                         return -EINVAL;
 961                 }
 962         } else {
 963                 /* Fail a mount if the logbuf is larger than 32K */
 964                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 965                         xfs_warn(mp,
 966                 "logbuf size for version 1 logs must be 16K or 32K");
 967                         return -EINVAL;
 968                 }
 969         }
 970
 971         /*
 972          * V5 filesystems always use attr2 format for attributes.
 973          */
 974         if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
 975                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 976                              "attr2 is always enabled for V5 filesystems.");
 977                 return -EINVAL;
 978         }
 979
 980         /*
 981          * prohibit r/w mounts of read-only filesystems
 982          */
 983         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
 984                 xfs_warn(mp,
 985                         "cannot mount a read-only filesystem as read-write");
 986                 return -EROFS;
 987         }
 988
 989         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
 990             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
 991             !xfs_has_pquotino(mp)) {
 992                 xfs_warn(mp,
 993                   "Super block does not support project and group quota together");
 994                 return -EINVAL;
 995         }
 996
 997         return 0;
 998 }
 999
1000 static int
1001 xfs_init_percpu_counters(
1002         struct xfs_mount        *mp)
1003 {
1004         int             error;
1005
1006         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1007         if (error)
1008                 return -ENOMEM;
1009
1010         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1011         if (error)
1012                 goto free_icount;
1013
1014         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1015         if (error)
1016                 goto free_ifree;
1017
1018         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1019         if (error)
1020                 goto free_fdblocks;
1021
1022         error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1023         if (error)
1024                 goto free_delalloc;
1025
1026         return 0;
1027
1028 free_delalloc:
1029         percpu_counter_destroy(&mp->m_delalloc_blks);
1030 free_fdblocks:
1031         percpu_counter_destroy(&mp->m_fdblocks);
1032 free_ifree:
1033         percpu_counter_destroy(&mp->m_ifree);
1034 free_icount:
1035         percpu_counter_destroy(&mp->m_icount);
1036         return -ENOMEM;
1037 }
1038
1039 void
1040 xfs_reinit_percpu_counters(
1041         struct xfs_mount        *mp)
1042 {
1043         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1044         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1045         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1046         percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1047 }
1048
1049 static void
1050 xfs_destroy_percpu_counters(
1051         struct xfs_mount        *mp)
1052 {
1053         percpu_counter_destroy(&mp->m_icount);
1054         percpu_counter_destroy(&mp->m_ifree);
1055         percpu_counter_destroy(&mp->m_fdblocks);
1056         ASSERT(xfs_is_shutdown(mp) ||
1057                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1058         percpu_counter_destroy(&mp->m_delalloc_blks);
1059         percpu_counter_destroy(&mp->m_frextents);
1060 }
1061
1062 static int
1063 xfs_inodegc_init_percpu(
1064         struct xfs_mount        *mp)
1065 {
1066         struct xfs_inodegc      *gc;
1067         int                     cpu;
1068
1069         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1070         if (!mp->m_inodegc)
1071                 return -ENOMEM;
1072
1073         for_each_possible_cpu(cpu) {
1074                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1075                 init_llist_head(&gc->list);
1076                 gc->items = 0;
1077                 INIT_WORK(&gc->work, xfs_inodegc_worker);
1078         }
1079         return 0;
1080 }
1081
1082 static void
1083 xfs_inodegc_free_percpu(
1084         struct xfs_mount        *mp)
1085 {
1086         if (!mp->m_inodegc)
1087                 return;
1088         free_percpu(mp->m_inodegc);
1089 }
1090
1091 static void
1092 xfs_fs_put_super(
1093         struct super_block      *sb)
1094 {
1095         struct xfs_mount        *mp = XFS_M(sb);
1096
1097         /* if ->fill_super failed, we have no mount to tear down */
1098         if (!sb->s_fs_info)
1099                 return;
1100
1101         xfs_notice(mp, "Unmounting Filesystem");
1102         xfs_filestream_unmount(mp);
1103         xfs_unmountfs(mp);
1104
1105         xfs_freesb(mp);
1106         free_percpu(mp->m_stats.xs_stats);
1107         xfs_mount_list_del(mp);
1108         xfs_inodegc_free_percpu(mp);
1109         xfs_destroy_percpu_counters(mp);
1110         xfs_destroy_mount_workqueues(mp);
1111         xfs_close_devices(mp);
1112
1113         sb->s_fs_info = NULL;
1114         xfs_mount_free(mp);
1115 }
1116
1117 static long
1118 xfs_fs_nr_cached_objects(
1119         struct super_block      *sb,
1120         struct shrink_control   *sc)
1121 {
1122         /* Paranoia: catch incorrect calls during mount setup or teardown */
1123         if (WARN_ON_ONCE(!sb->s_fs_info))
1124                 return 0;
1125         return xfs_reclaim_inodes_count(XFS_M(sb));
1126 }
1127
1128 static long
1129 xfs_fs_free_cached_objects(
1130         struct super_block      *sb,
1131         struct shrink_control   *sc)
1132 {
1133         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1134 }
1135
1136 static const struct super_operations xfs_super_operations = {
1137         .alloc_inode            = xfs_fs_alloc_inode,
1138         .destroy_inode          = xfs_fs_destroy_inode,
1139         .dirty_inode            = xfs_fs_dirty_inode,
1140         .drop_inode             = xfs_fs_drop_inode,
1141         .put_super              = xfs_fs_put_super,
1142         .sync_fs                = xfs_fs_sync_fs,
1143         .freeze_fs              = xfs_fs_freeze,
1144         .unfreeze_fs            = xfs_fs_unfreeze,
1145         .statfs                 = xfs_fs_statfs,
1146         .show_options           = xfs_fs_show_options,
1147         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1148         .free_cached_objects    = xfs_fs_free_cached_objects,
1149 };
1150
1151 static int
1152 suffix_kstrtoint(
1153         const char      *s,
1154         unsigned int    base,
1155         int             *res)
1156 {
1157         int             last, shift_left_factor = 0, _res;
1158         char            *value;
1159         int             ret = 0;
1160
1161         value = kstrdup(s, GFP_KERNEL);
1162         if (!value)
1163                 return -ENOMEM;
1164
1165         last = strlen(value) - 1;
1166         if (value[last] == 'K' || value[last] == 'k') {
1167                 shift_left_factor = 10;
1168                 value[last] = '\0';
1169         }
1170         if (value[last] == 'M' || value[last] == 'm') {
1171                 shift_left_factor = 20;
1172                 value[last] = '\0';
1173         }
1174         if (value[last] == 'G' || value[last] == 'g') {
1175                 shift_left_factor = 30;
1176                 value[last] = '\0';
1177         }
1178
1179         if (kstrtoint(value, base, &_res))
1180                 ret = -EINVAL;
1181         kfree(value);
1182         *res = _res << shift_left_factor;
1183         return ret;
1184 }
1185
1186 static inline void
1187 xfs_fs_warn_deprecated(
1188         struct fs_context       *fc,
1189         struct fs_parameter     *param,
1190         uint64_t                flag,
1191         bool                    value)
1192 {
1193         /* Don't print the warning if reconfiguring and current mount point
1194          * already had the flag set
1195          */
1196         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1197             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1198                 return;
1199         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1200 }
1201
1202 /*
1203  * Set mount state from a mount option.
1204  *
1205  * NOTE: mp->m_super is NULL here!
1206  */
1207 static int
1208 xfs_fs_parse_param(
1209         struct fs_context       *fc,
1210         struct fs_parameter     *param)
1211 {
1212         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1213         struct fs_parse_result  result;
1214         int                     size = 0;
1215         int                     opt;
1216
1217         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1218         if (opt < 0)
1219                 return opt;
1220
1221         switch (opt) {
1222         case Opt_logbufs:
1223                 parsing_mp->m_logbufs = result.uint_32;
1224                 return 0;
1225         case Opt_logbsize:
1226                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1227                         return -EINVAL;
1228                 return 0;
1229         case Opt_logdev:
1230                 kfree(parsing_mp->m_logname);
1231                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1232                 if (!parsing_mp->m_logname)
1233                         return -ENOMEM;
1234                 return 0;
1235         case Opt_rtdev:
1236                 kfree(parsing_mp->m_rtname);
1237                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1238                 if (!parsing_mp->m_rtname)
1239                         return -ENOMEM;
1240                 return 0;
1241         case Opt_allocsize:
1242                 if (suffix_kstrtoint(param->string, 10, &size))
1243                         return -EINVAL;
1244                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1245                 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1246                 return 0;
1247         case Opt_grpid:
1248         case Opt_bsdgroups:
1249                 parsing_mp->m_features |= XFS_FEAT_GRPID;
1250                 return 0;
1251         case Opt_nogrpid:
1252         case Opt_sysvgroups:
1253                 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1254                 return 0;
1255         case Opt_wsync:
1256                 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1257                 return 0;
1258         case Opt_norecovery:
1259                 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1260                 return 0;
1261         case Opt_noalign:
1262                 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1263                 return 0;
1264         case Opt_swalloc:
1265                 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1266                 return 0;
1267         case Opt_sunit:
1268                 parsing_mp->m_dalign = result.uint_32;
1269                 return 0;
1270         case Opt_swidth:
1271                 parsing_mp->m_swidth = result.uint_32;
1272                 return 0;
1273         case Opt_inode32:
1274                 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1275                 return 0;
1276         case Opt_inode64:
1277                 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1278                 return 0;
1279         case Opt_nouuid:
1280                 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1281                 return 0;
1282         case Opt_largeio:
1283                 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1284                 return 0;
1285         case Opt_nolargeio:
1286                 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1287                 return 0;
1288         case Opt_filestreams:
1289                 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1290                 return 0;
1291         case Opt_noquota:
1292                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1293                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1294                 return 0;
1295         case Opt_quota:
1296         case Opt_uquota:
1297         case Opt_usrquota:
1298                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1299                 return 0;
1300         case Opt_qnoenforce:
1301         case Opt_uqnoenforce:
1302                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1303                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1304                 return 0;
1305         case Opt_pquota:
1306         case Opt_prjquota:
1307                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1308                 return 0;
1309         case Opt_pqnoenforce:
1310                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1311                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1312                 return 0;
1313         case Opt_gquota:
1314         case Opt_grpquota:
1315                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1316                 return 0;
1317         case Opt_gqnoenforce:
1318                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1319                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1320                 return 0;
1321         case Opt_discard:
1322                 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1323                 return 0;
1324         case Opt_nodiscard:
1325                 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1326                 return 0;
1327 #ifdef CONFIG_FS_DAX
1328         case Opt_dax:
1329                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1330                 return 0;
1331         case Opt_dax_enum:
1332                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1333                 return 0;
1334 #endif
1335         /* Following mount options will be removed in September 2025 */
1336         case Opt_ikeep:
1337                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1338                 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1339                 return 0;
1340         case Opt_noikeep:
1341                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1342                 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1343                 return 0;
1344         case Opt_attr2:
1345                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1346                 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1347                 return 0;
1348         case Opt_noattr2:
1349                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1350                 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1351                 return 0;
1352         default:
1353                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1354                 return -EINVAL;
1355         }
1356
1357         return 0;
1358 }
1359
1360 static int
1361 xfs_fs_validate_params(
1362         struct xfs_mount        *mp)
1363 {
1364         /* No recovery flag requires a read-only mount */
1365         if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1366                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1367                 return -EINVAL;
1368         }
1369
1370         /*
1371          * We have not read the superblock at this point, so only the attr2
1372          * mount option can set the attr2 feature by this stage.
1373          */
1374         if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1375                 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1376                 return -EINVAL;
1377         }
1378
1379
1380         if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1381                 xfs_warn(mp,
1382         "sunit and swidth options incompatible with the noalign option");
1383                 return -EINVAL;
1384         }
1385
1386         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1387                 xfs_warn(mp, "quota support not available in this kernel.");
1388                 return -EINVAL;
1389         }
1390
1391         if ((mp->m_dalign && !mp->m_swidth) ||
1392             (!mp->m_dalign && mp->m_swidth)) {
1393                 xfs_warn(mp, "sunit and swidth must be specified together");
1394                 return -EINVAL;
1395         }
1396
1397         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1398                 xfs_warn(mp,
1399         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1400                         mp->m_swidth, mp->m_dalign);
1401                 return -EINVAL;
1402         }
1403
1404         if (mp->m_logbufs != -1 &&
1405             mp->m_logbufs != 0 &&
1406             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1407              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1408                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1409                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1410                 return -EINVAL;
1411         }
1412
1413         if (mp->m_logbsize != -1 &&
1414             mp->m_logbsize !=  0 &&
1415             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1416              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1417              !is_power_of_2(mp->m_logbsize))) {
1418                 xfs_warn(mp,
1419                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1420                         mp->m_logbsize);
1421                 return -EINVAL;
1422         }
1423
1424         if (xfs_has_allocsize(mp) &&
1425             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1426              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1427                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1428                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1429                 return -EINVAL;
1430         }
1431
1432         return 0;
1433 }
1434
1435 static int
1436 xfs_fs_fill_super(
1437         struct super_block      *sb,
1438         struct fs_context       *fc)
1439 {
1440         struct xfs_mount        *mp = sb->s_fs_info;
1441         struct inode            *root;
1442         int                     flags = 0, error;
1443
1444         mp->m_super = sb;
1445
1446         error = xfs_fs_validate_params(mp);
1447         if (error)
1448                 goto out_free_names;
1449
1450         sb_min_blocksize(sb, BBSIZE);
1451         sb->s_xattr = xfs_xattr_handlers;
1452         sb->s_export_op = &xfs_export_operations;
1453 #ifdef CONFIG_XFS_QUOTA
1454         sb->s_qcop = &xfs_quotactl_operations;
1455         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1456 #endif
1457         sb->s_op = &xfs_super_operations;
1458
1459         /*
1460          * Delay mount work if the debug hook is set. This is debug
1461          * instrumention to coordinate simulation of xfs mount failures with
1462          * VFS superblock operations
1463          */
1464         if (xfs_globals.mount_delay) {
1465                 xfs_notice(mp, "Delaying mount for %d seconds.",
1466                         xfs_globals.mount_delay);
1467                 msleep(xfs_globals.mount_delay * 1000);
1468         }
1469
1470         if (fc->sb_flags & SB_SILENT)
1471                 flags |= XFS_MFSI_QUIET;
1472
1473         error = xfs_open_devices(mp);
1474         if (error)
1475                 goto out_free_names;
1476
1477         error = xfs_init_mount_workqueues(mp);
1478         if (error)
1479                 goto out_close_devices;
1480
1481         error = xfs_init_percpu_counters(mp);
1482         if (error)
1483                 goto out_destroy_workqueues;
1484
1485         error = xfs_inodegc_init_percpu(mp);
1486         if (error)
1487                 goto out_destroy_counters;
1488
1489         /*
1490          * All percpu data structures requiring cleanup when a cpu goes offline
1491          * must be allocated before adding this @mp to the cpu-dead handler's
1492          * mount list.
1493          */
1494         xfs_mount_list_add(mp);
1495
1496         /* Allocate stats memory before we do operations that might use it */
1497         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1498         if (!mp->m_stats.xs_stats) {
1499                 error = -ENOMEM;
1500                 goto out_destroy_inodegc;
1501         }
1502
1503         error = xfs_readsb(mp, flags);
1504         if (error)
1505                 goto out_free_stats;
1506
1507         error = xfs_finish_flags(mp);
1508         if (error)
1509                 goto out_free_sb;
1510
1511         error = xfs_setup_devices(mp);
1512         if (error)
1513                 goto out_free_sb;
1514
1515         /* V4 support is undergoing deprecation. */
1516         if (!xfs_has_crc(mp)) {
1517 #ifdef CONFIG_XFS_SUPPORT_V4
1518                 xfs_warn_once(mp,
1519         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1520 #else
1521                 xfs_warn(mp,
1522         "Deprecated V4 format (crc=0) not supported by kernel.");
1523                 error = -EINVAL;
1524                 goto out_free_sb;
1525 #endif
1526         }
1527
1528         /* Filesystem claims it needs repair, so refuse the mount. */
1529         if (xfs_has_needsrepair(mp)) {
1530                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1531                 error = -EFSCORRUPTED;
1532                 goto out_free_sb;
1533         }
1534
1535         /*
1536          * Don't touch the filesystem if a user tool thinks it owns the primary
1537          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1538          * we don't check them at all.
1539          */
1540         if (mp->m_sb.sb_inprogress) {
1541                 xfs_warn(mp, "Offline file system operation in progress!");
1542                 error = -EFSCORRUPTED;
1543                 goto out_free_sb;
1544         }
1545
1546         /*
1547          * Until this is fixed only page-sized or smaller data blocks work.
1548          */
1549         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1550                 xfs_warn(mp,
1551                 "File system with blocksize %d bytes. "
1552                 "Only pagesize (%ld) or less will currently work.",
1553                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1554                 error = -ENOSYS;
1555                 goto out_free_sb;
1556         }
1557
1558         /* Ensure this filesystem fits in the page cache limits */
1559         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1560             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1561                 xfs_warn(mp,
1562                 "file system too large to be mounted on this system.");
1563                 error = -EFBIG;
1564                 goto out_free_sb;
1565         }
1566
1567         /*
1568          * XFS block mappings use 54 bits to store the logical block offset.
1569          * This should suffice to handle the maximum file size that the VFS
1570          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1571          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1572          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1573          * to check this assertion.
1574          *
1575          * Avoid integer overflow by comparing the maximum bmbt offset to the
1576          * maximum pagecache offset in units of fs blocks.
1577          */
1578         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1579                 xfs_warn(mp,
1580 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1581                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1582                          XFS_MAX_FILEOFF);
1583                 error = -EINVAL;
1584                 goto out_free_sb;
1585         }
1586
1587         error = xfs_filestream_mount(mp);
1588         if (error)
1589                 goto out_free_sb;
1590
1591         /*
1592          * we must configure the block size in the superblock before we run the
1593          * full mount process as the mount process can lookup and cache inodes.
1594          */
1595         sb->s_magic = XFS_SUPER_MAGIC;
1596         sb->s_blocksize = mp->m_sb.sb_blocksize;
1597         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1598         sb->s_maxbytes = MAX_LFS_FILESIZE;
1599         sb->s_max_links = XFS_MAXLINK;
1600         sb->s_time_gran = 1;
1601         if (xfs_has_bigtime(mp)) {
1602                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1603                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1604         } else {
1605                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1606                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1607         }
1608         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1609         sb->s_iflags |= SB_I_CGROUPWB;
1610
1611         set_posix_acl_flag(sb);
1612
1613         /* version 5 superblocks support inode version counters. */
1614         if (xfs_has_crc(mp))
1615                 sb->s_flags |= SB_I_VERSION;
1616
1617         if (xfs_has_dax_always(mp)) {
1618                 error = xfs_setup_dax_always(mp);
1619                 if (error)
1620                         goto out_filestream_unmount;
1621         }
1622
1623         if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1624                 xfs_warn(mp,
1625         "mounting with \"discard\" option, but the device does not support discard");
1626                 mp->m_features &= ~XFS_FEAT_DISCARD;
1627         }
1628
1629         if (xfs_has_reflink(mp)) {
1630                 if (mp->m_sb.sb_rblocks) {
1631                         xfs_alert(mp,
1632         "reflink not compatible with realtime device!");
1633                         error = -EINVAL;
1634                         goto out_filestream_unmount;
1635                 }
1636
1637                 if (xfs_globals.always_cow) {
1638                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1639                         mp->m_always_cow = true;
1640                 }
1641         }
1642
1643         if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1644                 xfs_alert(mp,
1645         "reverse mapping btree not compatible with realtime device!");
1646                 error = -EINVAL;
1647                 goto out_filestream_unmount;
1648         }
1649
1650         if (xfs_has_large_extent_counts(mp))
1651                 xfs_warn(mp,
1652         "EXPERIMENTAL Large extent counts feature in use. Use at your own risk!");
1653
1654         error = xfs_mountfs(mp);
1655         if (error)
1656                 goto out_filestream_unmount;
1657
1658         root = igrab(VFS_I(mp->m_rootip));
1659         if (!root) {
1660                 error = -ENOENT;
1661                 goto out_unmount;
1662         }
1663         sb->s_root = d_make_root(root);
1664         if (!sb->s_root) {
1665                 error = -ENOMEM;
1666                 goto out_unmount;
1667         }
1668
1669         return 0;
1670
1671  out_filestream_unmount:
1672         xfs_filestream_unmount(mp);
1673  out_free_sb:
1674         xfs_freesb(mp);
1675  out_free_stats:
1676         free_percpu(mp->m_stats.xs_stats);
1677  out_destroy_inodegc:
1678         xfs_mount_list_del(mp);
1679         xfs_inodegc_free_percpu(mp);
1680  out_destroy_counters:
1681         xfs_destroy_percpu_counters(mp);
1682  out_destroy_workqueues:
1683         xfs_destroy_mount_workqueues(mp);
1684  out_close_devices:
1685         xfs_close_devices(mp);
1686  out_free_names:
1687         sb->s_fs_info = NULL;
1688         xfs_mount_free(mp);
1689         return error;
1690
1691  out_unmount:
1692         xfs_filestream_unmount(mp);
1693         xfs_unmountfs(mp);
1694         goto out_free_sb;
1695 }
1696
1697 static int
1698 xfs_fs_get_tree(
1699         struct fs_context       *fc)
1700 {
1701         return get_tree_bdev(fc, xfs_fs_fill_super);
1702 }
1703
1704 static int
1705 xfs_remount_rw(
1706         struct xfs_mount        *mp)
1707 {
1708         struct xfs_sb           *sbp = &mp->m_sb;
1709         int error;
1710
1711         if (xfs_has_norecovery(mp)) {
1712                 xfs_warn(mp,
1713                         "ro->rw transition prohibited on norecovery mount");
1714                 return -EINVAL;
1715         }
1716
1717         if (xfs_sb_is_v5(sbp) &&
1718             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1719                 xfs_warn(mp,
1720         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1721                         (sbp->sb_features_ro_compat &
1722                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1723                 return -EINVAL;
1724         }
1725
1726         clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1727
1728         /*
1729          * If this is the first remount to writeable state we might have some
1730          * superblock changes to update.
1731          */
1732         if (mp->m_update_sb) {
1733                 error = xfs_sync_sb(mp, false);
1734                 if (error) {
1735                         xfs_warn(mp, "failed to write sb changes");
1736                         return error;
1737                 }
1738                 mp->m_update_sb = false;
1739         }
1740
1741         /*
1742          * Fill out the reserve pool if it is empty. Use the stashed value if
1743          * it is non-zero, otherwise go with the default.
1744          */
1745         xfs_restore_resvblks(mp);
1746         xfs_log_work_queue(mp);
1747         xfs_blockgc_start(mp);
1748
1749         /* Create the per-AG metadata reservation pool .*/
1750         error = xfs_fs_reserve_ag_blocks(mp);
1751         if (error && error != -ENOSPC)
1752                 return error;
1753
1754         /* Re-enable the background inode inactivation worker. */
1755         xfs_inodegc_start(mp);
1756
1757         return 0;
1758 }
1759
1760 static int
1761 xfs_remount_ro(
1762         struct xfs_mount        *mp)
1763 {
1764         struct xfs_icwalk       icw = {
1765                 .icw_flags      = XFS_ICWALK_FLAG_SYNC,
1766         };
1767         int                     error;
1768
1769         /* Flush all the dirty data to disk. */
1770         error = sync_filesystem(mp->m_super);
1771         if (error)
1772                 return error;
1773
1774         /*
1775          * Cancel background eofb scanning so it cannot race with the final
1776          * log force+buftarg wait and deadlock the remount.
1777          */
1778         xfs_blockgc_stop(mp);
1779
1780         /*
1781          * Clear out all remaining COW staging extents and speculative post-EOF
1782          * preallocations so that we don't leave inodes requiring inactivation
1783          * cleanups during reclaim on a read-only mount.  We must process every
1784          * cached inode, so this requires a synchronous cache scan.
1785          */
1786         error = xfs_blockgc_free_space(mp, &icw);
1787         if (error) {
1788                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1789                 return error;
1790         }
1791
1792         /*
1793          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1794          * flushed all pending inodegc work when it sync'd the filesystem.
1795          * The VFS holds s_umount, so we know that inodes cannot enter
1796          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1797          * we send inodes straight to reclaim, so no inodes will be queued.
1798          */
1799         xfs_inodegc_stop(mp);
1800
1801         /* Free the per-AG metadata reservation pool. */
1802         error = xfs_fs_unreserve_ag_blocks(mp);
1803         if (error) {
1804                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1805                 return error;
1806         }
1807
1808         /*
1809          * Before we sync the metadata, we need to free up the reserve block
1810          * pool so that the used block count in the superblock on disk is
1811          * correct at the end of the remount. Stash the current* reserve pool
1812          * size so that if we get remounted rw, we can return it to the same
1813          * size.
1814          */
1815         xfs_save_resvblks(mp);
1816
1817         xfs_log_clean(mp);
1818         set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1819
1820         return 0;
1821 }
1822
1823 /*
1824  * Logically we would return an error here to prevent users from believing
1825  * they might have changed mount options using remount which can't be changed.
1826  *
1827  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1828  * arguments in some cases so we can't blindly reject options, but have to
1829  * check for each specified option if it actually differs from the currently
1830  * set option and only reject it if that's the case.
1831  *
1832  * Until that is implemented we return success for every remount request, and
1833  * silently ignore all options that we can't actually change.
1834  */
1835 static int
1836 xfs_fs_reconfigure(
1837         struct fs_context *fc)
1838 {
1839         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1840         struct xfs_mount        *new_mp = fc->s_fs_info;
1841         int                     flags = fc->sb_flags;
1842         int                     error;
1843
1844         /* version 5 superblocks always support version counters. */
1845         if (xfs_has_crc(mp))
1846                 fc->sb_flags |= SB_I_VERSION;
1847
1848         error = xfs_fs_validate_params(new_mp);
1849         if (error)
1850                 return error;
1851
1852         /* inode32 -> inode64 */
1853         if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1854                 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1855                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1856         }
1857
1858         /* inode64 -> inode32 */
1859         if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1860                 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1861                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1862         }
1863
1864         /* ro -> rw */
1865         if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1866                 error = xfs_remount_rw(mp);
1867                 if (error)
1868                         return error;
1869         }
1870
1871         /* rw -> ro */
1872         if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1873                 error = xfs_remount_ro(mp);
1874                 if (error)
1875                         return error;
1876         }
1877
1878         return 0;
1879 }
1880
1881 static void xfs_fs_free(
1882         struct fs_context       *fc)
1883 {
1884         struct xfs_mount        *mp = fc->s_fs_info;
1885
1886         /*
1887          * mp is stored in the fs_context when it is initialized.
1888          * mp is transferred to the superblock on a successful mount,
1889          * but if an error occurs before the transfer we have to free
1890          * it here.
1891          */
1892         if (mp)
1893                 xfs_mount_free(mp);
1894 }
1895
1896 static const struct fs_context_operations xfs_context_ops = {
1897         .parse_param = xfs_fs_parse_param,
1898         .get_tree    = xfs_fs_get_tree,
1899         .reconfigure = xfs_fs_reconfigure,
1900         .free        = xfs_fs_free,
1901 };
1902
1903 static int xfs_init_fs_context(
1904         struct fs_context       *fc)
1905 {
1906         struct xfs_mount        *mp;
1907
1908         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1909         if (!mp)
1910                 return -ENOMEM;
1911
1912         spin_lock_init(&mp->m_sb_lock);
1913         spin_lock_init(&mp->m_agirotor_lock);
1914         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1915         spin_lock_init(&mp->m_perag_lock);
1916         mutex_init(&mp->m_growlock);
1917         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1918         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1919         mp->m_kobj.kobject.kset = xfs_kset;
1920         /*
1921          * We don't create the finobt per-ag space reservation until after log
1922          * recovery, so we must set this to true so that an ifree transaction
1923          * started during log recovery will not depend on space reservations
1924          * for finobt expansion.
1925          */
1926         mp->m_finobt_nores = true;
1927
1928         /*
1929          * These can be overridden by the mount option parsing.
1930          */
1931         mp->m_logbufs = -1;
1932         mp->m_logbsize = -1;
1933         mp->m_allocsize_log = 16; /* 64k */
1934
1935         /*
1936          * Copy binary VFS mount flags we are interested in.
1937          */
1938         if (fc->sb_flags & SB_RDONLY)
1939                 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1940         if (fc->sb_flags & SB_DIRSYNC)
1941                 mp->m_features |= XFS_FEAT_DIRSYNC;
1942         if (fc->sb_flags & SB_SYNCHRONOUS)
1943                 mp->m_features |= XFS_FEAT_WSYNC;
1944
1945         fc->s_fs_info = mp;
1946         fc->ops = &xfs_context_ops;
1947
1948         return 0;
1949 }
1950
1951 static struct file_system_type xfs_fs_type = {
1952         .owner                  = THIS_MODULE,
1953         .name                   = "xfs",
1954         .init_fs_context        = xfs_init_fs_context,
1955         .parameters             = xfs_fs_parameters,
1956         .kill_sb                = kill_block_super,
1957         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1958 };
1959 MODULE_ALIAS_FS("xfs");
1960
1961 STATIC int __init
1962 xfs_init_caches(void)
1963 {
1964         int             error;
1965
1966         xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
1967                                                 sizeof(struct xlog_ticket),
1968                                                 0, 0, NULL);
1969         if (!xfs_log_ticket_cache)
1970                 goto out;
1971
1972         error = xfs_btree_init_cur_caches();
1973         if (error)
1974                 goto out_destroy_log_ticket_cache;
1975
1976         error = xfs_defer_init_item_caches();
1977         if (error)
1978                 goto out_destroy_btree_cur_cache;
1979
1980         xfs_da_state_cache = kmem_cache_create("xfs_da_state",
1981                                               sizeof(struct xfs_da_state),
1982                                               0, 0, NULL);
1983         if (!xfs_da_state_cache)
1984                 goto out_destroy_defer_item_cache;
1985
1986         xfs_ifork_cache = kmem_cache_create("xfs_ifork",
1987                                            sizeof(struct xfs_ifork),
1988                                            0, 0, NULL);
1989         if (!xfs_ifork_cache)
1990                 goto out_destroy_da_state_cache;
1991
1992         xfs_trans_cache = kmem_cache_create("xfs_trans",
1993                                            sizeof(struct xfs_trans),
1994                                            0, 0, NULL);
1995         if (!xfs_trans_cache)
1996                 goto out_destroy_ifork_cache;
1997
1998
1999         /*
2000          * The size of the cache-allocated buf log item is the maximum
2001          * size possible under XFS.  This wastes a little bit of memory,
2002          * but it is much faster.
2003          */
2004         xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2005                                               sizeof(struct xfs_buf_log_item),
2006                                               0, 0, NULL);
2007         if (!xfs_buf_item_cache)
2008                 goto out_destroy_trans_cache;
2009
2010         xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2011                                         (sizeof(struct xfs_efd_log_item) +
2012                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
2013                                         sizeof(struct xfs_extent)),
2014                                         0, 0, NULL);
2015         if (!xfs_efd_cache)
2016                 goto out_destroy_buf_item_cache;
2017
2018         xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2019                                          (sizeof(struct xfs_efi_log_item) +
2020                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2021                                          sizeof(struct xfs_extent)),
2022                                          0, 0, NULL);
2023         if (!xfs_efi_cache)
2024                 goto out_destroy_efd_cache;
2025
2026         xfs_inode_cache = kmem_cache_create("xfs_inode",
2027                                            sizeof(struct xfs_inode), 0,
2028                                            (SLAB_HWCACHE_ALIGN |
2029                                             SLAB_RECLAIM_ACCOUNT |
2030                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2031                                            xfs_fs_inode_init_once);
2032         if (!xfs_inode_cache)
2033                 goto out_destroy_efi_cache;
2034
2035         xfs_ili_cache = kmem_cache_create("xfs_ili",
2036                                          sizeof(struct xfs_inode_log_item), 0,
2037                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2038                                          NULL);
2039         if (!xfs_ili_cache)
2040                 goto out_destroy_inode_cache;
2041
2042         xfs_icreate_cache = kmem_cache_create("xfs_icr",
2043                                              sizeof(struct xfs_icreate_item),
2044                                              0, 0, NULL);
2045         if (!xfs_icreate_cache)
2046                 goto out_destroy_ili_cache;
2047
2048         xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2049                                          sizeof(struct xfs_rud_log_item),
2050                                          0, 0, NULL);
2051         if (!xfs_rud_cache)
2052                 goto out_destroy_icreate_cache;
2053
2054         xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2055                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2056                         0, 0, NULL);
2057         if (!xfs_rui_cache)
2058                 goto out_destroy_rud_cache;
2059
2060         xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2061                                          sizeof(struct xfs_cud_log_item),
2062                                          0, 0, NULL);
2063         if (!xfs_cud_cache)
2064                 goto out_destroy_rui_cache;
2065
2066         xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2067                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2068                         0, 0, NULL);
2069         if (!xfs_cui_cache)
2070                 goto out_destroy_cud_cache;
2071
2072         xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2073                                          sizeof(struct xfs_bud_log_item),
2074                                          0, 0, NULL);
2075         if (!xfs_bud_cache)
2076                 goto out_destroy_cui_cache;
2077
2078         xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2079                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2080                         0, 0, NULL);
2081         if (!xfs_bui_cache)
2082                 goto out_destroy_bud_cache;
2083
2084         xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2085                                             sizeof(struct xfs_attrd_log_item),
2086                                             0, 0, NULL);
2087         if (!xfs_attrd_cache)
2088                 goto out_destroy_bui_cache;
2089
2090         xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2091                                             sizeof(struct xfs_attri_log_item),
2092                                             0, 0, NULL);
2093         if (!xfs_attri_cache)
2094                 goto out_destroy_attrd_cache;
2095
2096         return 0;
2097
2098  out_destroy_attrd_cache:
2099         kmem_cache_destroy(xfs_attrd_cache);
2100  out_destroy_bui_cache:
2101         kmem_cache_destroy(xfs_bui_cache);
2102  out_destroy_bud_cache:
2103         kmem_cache_destroy(xfs_bud_cache);
2104  out_destroy_cui_cache:
2105         kmem_cache_destroy(xfs_cui_cache);
2106  out_destroy_cud_cache:
2107         kmem_cache_destroy(xfs_cud_cache);
2108  out_destroy_rui_cache:
2109         kmem_cache_destroy(xfs_rui_cache);
2110  out_destroy_rud_cache:
2111         kmem_cache_destroy(xfs_rud_cache);
2112  out_destroy_icreate_cache:
2113         kmem_cache_destroy(xfs_icreate_cache);
2114  out_destroy_ili_cache:
2115         kmem_cache_destroy(xfs_ili_cache);
2116  out_destroy_inode_cache:
2117         kmem_cache_destroy(xfs_inode_cache);
2118  out_destroy_efi_cache:
2119         kmem_cache_destroy(xfs_efi_cache);
2120  out_destroy_efd_cache:
2121         kmem_cache_destroy(xfs_efd_cache);
2122  out_destroy_buf_item_cache:
2123         kmem_cache_destroy(xfs_buf_item_cache);
2124  out_destroy_trans_cache:
2125         kmem_cache_destroy(xfs_trans_cache);
2126  out_destroy_ifork_cache:
2127         kmem_cache_destroy(xfs_ifork_cache);
2128  out_destroy_da_state_cache:
2129         kmem_cache_destroy(xfs_da_state_cache);
2130  out_destroy_defer_item_cache:
2131         xfs_defer_destroy_item_caches();
2132  out_destroy_btree_cur_cache:
2133         xfs_btree_destroy_cur_caches();
2134  out_destroy_log_ticket_cache:
2135         kmem_cache_destroy(xfs_log_ticket_cache);
2136  out:
2137         return -ENOMEM;
2138 }
2139
2140 STATIC void
2141 xfs_destroy_caches(void)
2142 {
2143         /*
2144          * Make sure all delayed rcu free are flushed before we
2145          * destroy caches.
2146          */
2147         rcu_barrier();
2148         kmem_cache_destroy(xfs_attri_cache);
2149         kmem_cache_destroy(xfs_attrd_cache);
2150         kmem_cache_destroy(xfs_bui_cache);
2151         kmem_cache_destroy(xfs_bud_cache);
2152         kmem_cache_destroy(xfs_cui_cache);
2153         kmem_cache_destroy(xfs_cud_cache);
2154         kmem_cache_destroy(xfs_rui_cache);
2155         kmem_cache_destroy(xfs_rud_cache);
2156         kmem_cache_destroy(xfs_icreate_cache);
2157         kmem_cache_destroy(xfs_ili_cache);
2158         kmem_cache_destroy(xfs_inode_cache);
2159         kmem_cache_destroy(xfs_efi_cache);
2160         kmem_cache_destroy(xfs_efd_cache);
2161         kmem_cache_destroy(xfs_buf_item_cache);
2162         kmem_cache_destroy(xfs_trans_cache);
2163         kmem_cache_destroy(xfs_ifork_cache);
2164         kmem_cache_destroy(xfs_da_state_cache);
2165         xfs_defer_destroy_item_caches();
2166         xfs_btree_destroy_cur_caches();
2167         kmem_cache_destroy(xfs_log_ticket_cache);
2168 }
2169
2170 STATIC int __init
2171 xfs_init_workqueues(void)
2172 {
2173         /*
2174          * The allocation workqueue can be used in memory reclaim situations
2175          * (writepage path), and parallelism is only limited by the number of
2176          * AGs in all the filesystems mounted. Hence use the default large
2177          * max_active value for this workqueue.
2178          */
2179         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2180                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2181         if (!xfs_alloc_wq)
2182                 return -ENOMEM;
2183
2184         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2185                         0);
2186         if (!xfs_discard_wq)
2187                 goto out_free_alloc_wq;
2188
2189         return 0;
2190 out_free_alloc_wq:
2191         destroy_workqueue(xfs_alloc_wq);
2192         return -ENOMEM;
2193 }
2194
2195 STATIC void
2196 xfs_destroy_workqueues(void)
2197 {
2198         destroy_workqueue(xfs_discard_wq);
2199         destroy_workqueue(xfs_alloc_wq);
2200 }
2201
2202 #ifdef CONFIG_HOTPLUG_CPU
2203 static int
2204 xfs_cpu_dead(
2205         unsigned int            cpu)
2206 {
2207         struct xfs_mount        *mp, *n;
2208
2209         spin_lock(&xfs_mount_list_lock);
2210         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2211                 spin_unlock(&xfs_mount_list_lock);
2212                 xfs_inodegc_cpu_dead(mp, cpu);
2213                 spin_lock(&xfs_mount_list_lock);
2214         }
2215         spin_unlock(&xfs_mount_list_lock);
2216         return 0;
2217 }
2218
2219 static int __init
2220 xfs_cpu_hotplug_init(void)
2221 {
2222         int     error;
2223
2224         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2225                         xfs_cpu_dead);
2226         if (error < 0)
2227                 xfs_alert(NULL,
2228 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2229                         error);
2230         return error;
2231 }
2232
2233 static void
2234 xfs_cpu_hotplug_destroy(void)
2235 {
2236         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2237 }
2238
2239 #else /* !CONFIG_HOTPLUG_CPU */
2240 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2241 static inline void xfs_cpu_hotplug_destroy(void) {}
2242 #endif
2243
2244 STATIC int __init
2245 init_xfs_fs(void)
2246 {
2247         int                     error;
2248
2249         xfs_check_ondisk_structs();
2250
2251         printk(KERN_INFO XFS_VERSION_STRING " with "
2252                          XFS_BUILD_OPTIONS " enabled\n");
2253
2254         xfs_dir_startup();
2255
2256         error = xfs_cpu_hotplug_init();
2257         if (error)
2258                 goto out;
2259
2260         error = xfs_init_caches();
2261         if (error)
2262                 goto out_destroy_hp;
2263
2264         error = xfs_init_workqueues();
2265         if (error)
2266                 goto out_destroy_caches;
2267
2268         error = xfs_mru_cache_init();
2269         if (error)
2270                 goto out_destroy_wq;
2271
2272         error = xfs_buf_init();
2273         if (error)
2274                 goto out_mru_cache_uninit;
2275
2276         error = xfs_init_procfs();
2277         if (error)
2278                 goto out_buf_terminate;
2279
2280         error = xfs_sysctl_register();
2281         if (error)
2282                 goto out_cleanup_procfs;
2283
2284         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2285         if (!xfs_kset) {
2286                 error = -ENOMEM;
2287                 goto out_sysctl_unregister;
2288         }
2289
2290         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2291
2292         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2293         if (!xfsstats.xs_stats) {
2294                 error = -ENOMEM;
2295                 goto out_kset_unregister;
2296         }
2297
2298         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2299                                "stats");
2300         if (error)
2301                 goto out_free_stats;
2302
2303 #ifdef DEBUG
2304         xfs_dbg_kobj.kobject.kset = xfs_kset;
2305         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2306         if (error)
2307                 goto out_remove_stats_kobj;
2308 #endif
2309
2310         error = xfs_qm_init();
2311         if (error)
2312                 goto out_remove_dbg_kobj;
2313
2314         error = register_filesystem(&xfs_fs_type);
2315         if (error)
2316                 goto out_qm_exit;
2317         return 0;
2318
2319  out_qm_exit:
2320         xfs_qm_exit();
2321  out_remove_dbg_kobj:
2322 #ifdef DEBUG
2323         xfs_sysfs_del(&xfs_dbg_kobj);
2324  out_remove_stats_kobj:
2325 #endif
2326         xfs_sysfs_del(&xfsstats.xs_kobj);
2327  out_free_stats:
2328         free_percpu(xfsstats.xs_stats);
2329  out_kset_unregister:
2330         kset_unregister(xfs_kset);
2331  out_sysctl_unregister:
2332         xfs_sysctl_unregister();
2333  out_cleanup_procfs:
2334         xfs_cleanup_procfs();
2335  out_buf_terminate:
2336         xfs_buf_terminate();
2337  out_mru_cache_uninit:
2338         xfs_mru_cache_uninit();
2339  out_destroy_wq:
2340         xfs_destroy_workqueues();
2341  out_destroy_caches:
2342         xfs_destroy_caches();
2343  out_destroy_hp:
2344         xfs_cpu_hotplug_destroy();
2345  out:
2346         return error;
2347 }
2348
2349 STATIC void __exit
2350 exit_xfs_fs(void)
2351 {
2352         xfs_qm_exit();
2353         unregister_filesystem(&xfs_fs_type);
2354 #ifdef DEBUG
2355         xfs_sysfs_del(&xfs_dbg_kobj);
2356 #endif
2357         xfs_sysfs_del(&xfsstats.xs_kobj);
2358         free_percpu(xfsstats.xs_stats);
2359         kset_unregister(xfs_kset);
2360         xfs_sysctl_unregister();
2361         xfs_cleanup_procfs();
2362         xfs_buf_terminate();
2363         xfs_mru_cache_uninit();
2364         xfs_destroy_workqueues();
2365         xfs_destroy_caches();
2366         xfs_uuid_table_free();
2367         xfs_cpu_hotplug_destroy();
2368 }
2369
2370 module_init(init_xfs_fs);
2371 module_exit(exit_xfs_fs);
2372
2373 MODULE_AUTHOR("Silicon Graphics, Inc.");
2374 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2375 MODULE_LICENSE("GPL");