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