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 #include "xfs_iunlink_item.h"
  44
  45 #include <linux/magic.h>
  46 #include <linux/fs_context.h>
  47 #include <linux/fs_parser.h>
  48
  49 static const struct super_operations xfs_super_operations;
  50
  51 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  52 #ifdef DEBUG
  53 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  54 #endif
  55
  56 #ifdef CONFIG_HOTPLUG_CPU
  57 static LIST_HEAD(xfs_mount_list);
  58 static DEFINE_SPINLOCK(xfs_mount_list_lock);
  59
  60 static inline void xfs_mount_list_add(struct xfs_mount *mp)
  61 {
  62         spin_lock(&xfs_mount_list_lock);
  63         list_add(&mp->m_mount_list, &xfs_mount_list);
  64         spin_unlock(&xfs_mount_list_lock);
  65 }
  66
  67 static inline void xfs_mount_list_del(struct xfs_mount *mp)
  68 {
  69         spin_lock(&xfs_mount_list_lock);
  70         list_del(&mp->m_mount_list);
  71         spin_unlock(&xfs_mount_list_lock);
  72 }
  73 #else /* !CONFIG_HOTPLUG_CPU */
  74 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
  75 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
  76 #endif
  77
  78 enum xfs_dax_mode {
  79         XFS_DAX_INODE = 0,
  80         XFS_DAX_ALWAYS = 1,
  81         XFS_DAX_NEVER = 2,
  82 };
  83
  84 static void
  85 xfs_mount_set_dax_mode(
  86         struct xfs_mount        *mp,
  87         enum xfs_dax_mode       mode)
  88 {
  89         switch (mode) {
  90         case XFS_DAX_INODE:
  91                 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
  92                 break;
  93         case XFS_DAX_ALWAYS:
  94                 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
  95                 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
  96                 break;
  97         case XFS_DAX_NEVER:
  98                 mp->m_features |= XFS_FEAT_DAX_NEVER;
  99                 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
 100                 break;
 101         }
 102 }
 103
 104 static const struct constant_table dax_param_enums[] = {
 105         {"inode",       XFS_DAX_INODE },
 106         {"always",      XFS_DAX_ALWAYS },
 107         {"never",       XFS_DAX_NEVER },
 108         {}
 109 };
 110
 111 /*
 112  * Table driven mount option parser.
 113  */
 114 enum {
 115         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
 116         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
 117         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
 118         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
 119         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
 120         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
 121         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
 122         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
 123         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
 124 };
 125
 126 static const struct fs_parameter_spec xfs_fs_parameters[] = {
 127         fsparam_u32("logbufs",          Opt_logbufs),
 128         fsparam_string("logbsize",      Opt_logbsize),
 129         fsparam_string("logdev",        Opt_logdev),
 130         fsparam_string("rtdev",         Opt_rtdev),
 131         fsparam_flag("wsync",           Opt_wsync),
 132         fsparam_flag("noalign",         Opt_noalign),
 133         fsparam_flag("swalloc",         Opt_swalloc),
 134         fsparam_u32("sunit",            Opt_sunit),
 135         fsparam_u32("swidth",           Opt_swidth),
 136         fsparam_flag("nouuid",          Opt_nouuid),
 137         fsparam_flag("grpid",           Opt_grpid),
 138         fsparam_flag("nogrpid",         Opt_nogrpid),
 139         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 140         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 141         fsparam_string("allocsize",     Opt_allocsize),
 142         fsparam_flag("norecovery",      Opt_norecovery),
 143         fsparam_flag("inode64",         Opt_inode64),
 144         fsparam_flag("inode32",         Opt_inode32),
 145         fsparam_flag("ikeep",           Opt_ikeep),
 146         fsparam_flag("noikeep",         Opt_noikeep),
 147         fsparam_flag("largeio",         Opt_largeio),
 148         fsparam_flag("nolargeio",       Opt_nolargeio),
 149         fsparam_flag("attr2",           Opt_attr2),
 150         fsparam_flag("noattr2",         Opt_noattr2),
 151         fsparam_flag("filestreams",     Opt_filestreams),
 152         fsparam_flag("quota",           Opt_quota),
 153         fsparam_flag("noquota",         Opt_noquota),
 154         fsparam_flag("usrquota",        Opt_usrquota),
 155         fsparam_flag("grpquota",        Opt_grpquota),
 156         fsparam_flag("prjquota",        Opt_prjquota),
 157         fsparam_flag("uquota",          Opt_uquota),
 158         fsparam_flag("gquota",          Opt_gquota),
 159         fsparam_flag("pquota",          Opt_pquota),
 160         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 161         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 162         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 163         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 164         fsparam_flag("discard",         Opt_discard),
 165         fsparam_flag("nodiscard",       Opt_nodiscard),
 166         fsparam_flag("dax",             Opt_dax),
 167         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 168         {}
 169 };
 170
 171 struct proc_xfs_info {
 172         uint64_t        flag;
 173         char            *str;
 174 };
 175
 176 static int
 177 xfs_fs_show_options(
 178         struct seq_file         *m,
 179         struct dentry           *root)
 180 {
 181         static struct proc_xfs_info xfs_info_set[] = {
 182                 /* the few simple ones we can get from the mount struct */
 183                 { XFS_FEAT_IKEEP,               ",ikeep" },
 184                 { XFS_FEAT_WSYNC,               ",wsync" },
 185                 { XFS_FEAT_NOALIGN,             ",noalign" },
 186                 { XFS_FEAT_SWALLOC,             ",swalloc" },
 187                 { XFS_FEAT_NOUUID,              ",nouuid" },
 188                 { XFS_FEAT_NORECOVERY,          ",norecovery" },
 189                 { XFS_FEAT_ATTR2,               ",attr2" },
 190                 { XFS_FEAT_FILESTREAMS,         ",filestreams" },
 191                 { XFS_FEAT_GRPID,               ",grpid" },
 192                 { XFS_FEAT_DISCARD,             ",discard" },
 193                 { XFS_FEAT_LARGE_IOSIZE,        ",largeio" },
 194                 { XFS_FEAT_DAX_ALWAYS,          ",dax=always" },
 195                 { XFS_FEAT_DAX_NEVER,           ",dax=never" },
 196                 { 0, NULL }
 197         };
 198         struct xfs_mount        *mp = XFS_M(root->d_sb);
 199         struct proc_xfs_info    *xfs_infop;
 200
 201         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 202                 if (mp->m_features & xfs_infop->flag)
 203                         seq_puts(m, xfs_infop->str);
 204         }
 205
 206         seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
 207
 208         if (xfs_has_allocsize(mp))
 209                 seq_printf(m, ",allocsize=%dk",
 210                            (1 << mp->m_allocsize_log) >> 10);
 211
 212         if (mp->m_logbufs > 0)
 213                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 214         if (mp->m_logbsize > 0)
 215                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 216
 217         if (mp->m_logname)
 218                 seq_show_option(m, "logdev", mp->m_logname);
 219         if (mp->m_rtname)
 220                 seq_show_option(m, "rtdev", mp->m_rtname);
 221
 222         if (mp->m_dalign > 0)
 223                 seq_printf(m, ",sunit=%d",
 224                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 225         if (mp->m_swidth > 0)
 226                 seq_printf(m, ",swidth=%d",
 227                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 228
 229         if (mp->m_qflags & XFS_UQUOTA_ENFD)
 230                 seq_puts(m, ",usrquota");
 231         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 232                 seq_puts(m, ",uqnoenforce");
 233
 234         if (mp->m_qflags & XFS_PQUOTA_ENFD)
 235                 seq_puts(m, ",prjquota");
 236         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
 237                 seq_puts(m, ",pqnoenforce");
 238
 239         if (mp->m_qflags & XFS_GQUOTA_ENFD)
 240                 seq_puts(m, ",grpquota");
 241         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
 242                 seq_puts(m, ",gqnoenforce");
 243
 244         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 245                 seq_puts(m, ",noquota");
 246
 247         return 0;
 248 }
 249
 250 /*
 251  * Set parameters for inode allocation heuristics, taking into account
 252  * filesystem size and inode32/inode64 mount options; i.e. specifically
 253  * whether or not XFS_FEAT_SMALL_INUMS is set.
 254  *
 255  * Inode allocation patterns are altered only if inode32 is requested
 256  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
 257  * If altered, XFS_OPSTATE_INODE32 is set as well.
 258  *
 259  * An agcount independent of that in the mount structure is provided
 260  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 261  * to the potentially higher ag count.
 262  *
 263  * Returns the maximum AG index which may contain inodes.
 264  */
 265 xfs_agnumber_t
 266 xfs_set_inode_alloc(
 267         struct xfs_mount *mp,
 268         xfs_agnumber_t  agcount)
 269 {
 270         xfs_agnumber_t  index;
 271         xfs_agnumber_t  maxagi = 0;
 272         xfs_sb_t        *sbp = &mp->m_sb;
 273         xfs_agnumber_t  max_metadata;
 274         xfs_agino_t     agino;
 275         xfs_ino_t       ino;
 276
 277         /*
 278          * Calculate how much should be reserved for inodes to meet
 279          * the max inode percentage.  Used only for inode32.
 280          */
 281         if (M_IGEO(mp)->maxicount) {
 282                 uint64_t        icount;
 283
 284                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 285                 do_div(icount, 100);
 286                 icount += sbp->sb_agblocks - 1;
 287                 do_div(icount, sbp->sb_agblocks);
 288                 max_metadata = icount;
 289         } else {
 290                 max_metadata = agcount;
 291         }
 292
 293         /* Get the last possible inode in the filesystem */
 294         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 295         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 296
 297         /*
 298          * If user asked for no more than 32-bit inodes, and the fs is
 299          * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
 300          * the allocator to accommodate the request.
 301          */
 302         if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
 303                 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 304         else
 305                 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 306
 307         for (index = 0; index < agcount; index++) {
 308                 struct xfs_perag        *pag;
 309
 310                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 311
 312                 pag = xfs_perag_get(mp, index);
 313
 314                 if (xfs_is_inode32(mp)) {
 315                         if (ino > XFS_MAXINUMBER_32) {
 316                                 pag->pagi_inodeok = 0;
 317                                 pag->pagf_metadata = 0;
 318                         } else {
 319                                 pag->pagi_inodeok = 1;
 320                                 maxagi++;
 321                                 if (index < max_metadata)
 322                                         pag->pagf_metadata = 1;
 323                                 else
 324                                         pag->pagf_metadata = 0;
 325                         }
 326                 } else {
 327                         pag->pagi_inodeok = 1;
 328                         pag->pagf_metadata = 0;
 329                 }
 330
 331                 xfs_perag_put(pag);
 332         }
 333
 334         return xfs_is_inode32(mp) ? maxagi : agcount;
 335 }
 336
 337 static int
 338 xfs_setup_dax_always(
 339         struct xfs_mount        *mp)
 340 {
 341         if (!mp->m_ddev_targp->bt_daxdev &&
 342             (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
 343                 xfs_alert(mp,
 344                         "DAX unsupported by block device. Turning off DAX.");
 345                 goto disable_dax;
 346         }
 347
 348         if (mp->m_super->s_blocksize != PAGE_SIZE) {
 349                 xfs_alert(mp,
 350                         "DAX not supported for blocksize. Turning off DAX.");
 351                 goto disable_dax;
 352         }
 353
 354         if (xfs_has_reflink(mp) &&
 355             bdev_is_partition(mp->m_ddev_targp->bt_bdev)) {
 356                 xfs_alert(mp,
 357                         "DAX and reflink cannot work with multi-partitions!");
 358                 return -EINVAL;
 359         }
 360
 361         xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
 362         return 0;
 363
 364 disable_dax:
 365         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
 366         return 0;
 367 }
 368
 369 STATIC int
 370 xfs_blkdev_get(
 371         xfs_mount_t             *mp,
 372         const char              *name,
 373         struct block_device     **bdevp)
 374 {
 375         int                     error = 0;
 376
 377         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 378                                     mp);
 379         if (IS_ERR(*bdevp)) {
 380                 error = PTR_ERR(*bdevp);
 381                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 382         }
 383
 384         return error;
 385 }
 386
 387 STATIC void
 388 xfs_blkdev_put(
 389         struct block_device     *bdev)
 390 {
 391         if (bdev)
 392                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 393 }
 394
 395 STATIC void
 396 xfs_close_devices(
 397         struct xfs_mount        *mp)
 398 {
 399         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 400                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 401
 402                 xfs_free_buftarg(mp->m_logdev_targp);
 403                 xfs_blkdev_put(logdev);
 404         }
 405         if (mp->m_rtdev_targp) {
 406                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 407
 408                 xfs_free_buftarg(mp->m_rtdev_targp);
 409                 xfs_blkdev_put(rtdev);
 410         }
 411         xfs_free_buftarg(mp->m_ddev_targp);
 412 }
 413
 414 /*
 415  * The file system configurations are:
 416  *      (1) device (partition) with data and internal log
 417  *      (2) logical volume with data and log subvolumes.
 418  *      (3) logical volume with data, log, and realtime subvolumes.
 419  *
 420  * We only have to handle opening the log and realtime volumes here if
 421  * they are present.  The data subvolume has already been opened by
 422  * get_sb_bdev() and is stored in sb->s_bdev.
 423  */
 424 STATIC int
 425 xfs_open_devices(
 426         struct xfs_mount        *mp)
 427 {
 428         struct block_device     *ddev = mp->m_super->s_bdev;
 429         struct block_device     *logdev = NULL, *rtdev = NULL;
 430         int                     error;
 431
 432         /*
 433          * Open real time and log devices - order is important.
 434          */
 435         if (mp->m_logname) {
 436                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 437                 if (error)
 438                         return error;
 439         }
 440
 441         if (mp->m_rtname) {
 442                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 443                 if (error)
 444                         goto out_close_logdev;
 445
 446                 if (rtdev == ddev || rtdev == logdev) {
 447                         xfs_warn(mp,
 448         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 449                         error = -EINVAL;
 450                         goto out_close_rtdev;
 451                 }
 452         }
 453
 454         /*
 455          * Setup xfs_mount buffer target pointers
 456          */
 457         error = -ENOMEM;
 458         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
 459         if (!mp->m_ddev_targp)
 460                 goto out_close_rtdev;
 461
 462         if (rtdev) {
 463                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
 464                 if (!mp->m_rtdev_targp)
 465                         goto out_free_ddev_targ;
 466         }
 467
 468         if (logdev && logdev != ddev) {
 469                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
 470                 if (!mp->m_logdev_targp)
 471                         goto out_free_rtdev_targ;
 472         } else {
 473                 mp->m_logdev_targp = mp->m_ddev_targp;
 474         }
 475
 476         return 0;
 477
 478  out_free_rtdev_targ:
 479         if (mp->m_rtdev_targp)
 480                 xfs_free_buftarg(mp->m_rtdev_targp);
 481  out_free_ddev_targ:
 482         xfs_free_buftarg(mp->m_ddev_targp);
 483  out_close_rtdev:
 484         xfs_blkdev_put(rtdev);
 485  out_close_logdev:
 486         if (logdev && logdev != ddev)
 487                 xfs_blkdev_put(logdev);
 488         return error;
 489 }
 490
 491 /*
 492  * Setup xfs_mount buffer target pointers based on superblock
 493  */
 494 STATIC int
 495 xfs_setup_devices(
 496         struct xfs_mount        *mp)
 497 {
 498         int                     error;
 499
 500         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 501         if (error)
 502                 return error;
 503
 504         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 505                 unsigned int    log_sector_size = BBSIZE;
 506
 507                 if (xfs_has_sector(mp))
 508                         log_sector_size = mp->m_sb.sb_logsectsize;
 509                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 510                                             log_sector_size);
 511                 if (error)
 512                         return error;
 513         }
 514         if (mp->m_rtdev_targp) {
 515                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 516                                             mp->m_sb.sb_sectsize);
 517                 if (error)
 518                         return error;
 519         }
 520
 521         return 0;
 522 }
 523
 524 STATIC int
 525 xfs_init_mount_workqueues(
 526         struct xfs_mount        *mp)
 527 {
 528         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 529                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 530                         1, mp->m_super->s_id);
 531         if (!mp->m_buf_workqueue)
 532                 goto out;
 533
 534         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 535                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 536                         0, mp->m_super->s_id);
 537         if (!mp->m_unwritten_workqueue)
 538                 goto out_destroy_buf;
 539
 540         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 541                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 542                         0, mp->m_super->s_id);
 543         if (!mp->m_reclaim_workqueue)
 544                 goto out_destroy_unwritten;
 545
 546         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
 547                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
 548                         0, mp->m_super->s_id);
 549         if (!mp->m_blockgc_wq)
 550                 goto out_destroy_reclaim;
 551
 552         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
 553                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 554                         1, mp->m_super->s_id);
 555         if (!mp->m_inodegc_wq)
 556                 goto out_destroy_blockgc;
 557
 558         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 559                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 560         if (!mp->m_sync_workqueue)
 561                 goto out_destroy_inodegc;
 562
 563         return 0;
 564
 565 out_destroy_inodegc:
 566         destroy_workqueue(mp->m_inodegc_wq);
 567 out_destroy_blockgc:
 568         destroy_workqueue(mp->m_blockgc_wq);
 569 out_destroy_reclaim:
 570         destroy_workqueue(mp->m_reclaim_workqueue);
 571 out_destroy_unwritten:
 572         destroy_workqueue(mp->m_unwritten_workqueue);
 573 out_destroy_buf:
 574         destroy_workqueue(mp->m_buf_workqueue);
 575 out:
 576         return -ENOMEM;
 577 }
 578
 579 STATIC void
 580 xfs_destroy_mount_workqueues(
 581         struct xfs_mount        *mp)
 582 {
 583         destroy_workqueue(mp->m_sync_workqueue);
 584         destroy_workqueue(mp->m_blockgc_wq);
 585         destroy_workqueue(mp->m_inodegc_wq);
 586         destroy_workqueue(mp->m_reclaim_workqueue);
 587         destroy_workqueue(mp->m_unwritten_workqueue);
 588         destroy_workqueue(mp->m_buf_workqueue);
 589 }
 590
 591 static void
 592 xfs_flush_inodes_worker(
 593         struct work_struct      *work)
 594 {
 595         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 596                                                    m_flush_inodes_work);
 597         struct super_block      *sb = mp->m_super;
 598
 599         if (down_read_trylock(&sb->s_umount)) {
 600                 sync_inodes_sb(sb);
 601                 up_read(&sb->s_umount);
 602         }
 603 }
 604
 605 /*
 606  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 607  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 608  * for IO to complete so that we effectively throttle multiple callers to the
 609  * rate at which IO is completing.
 610  */
 611 void
 612 xfs_flush_inodes(
 613         struct xfs_mount        *mp)
 614 {
 615         /*
 616          * If flush_work() returns true then that means we waited for a flush
 617          * which was already in progress.  Don't bother running another scan.
 618          */
 619         if (flush_work(&mp->m_flush_inodes_work))
 620                 return;
 621
 622         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 623         flush_work(&mp->m_flush_inodes_work);
 624 }
 625
 626 /* Catch misguided souls that try to use this interface on XFS */
 627 STATIC struct inode *
 628 xfs_fs_alloc_inode(
 629         struct super_block      *sb)
 630 {
 631         BUG();
 632         return NULL;
 633 }
 634
 635 /*
 636  * Now that the generic code is guaranteed not to be accessing
 637  * the linux inode, we can inactivate and reclaim the inode.
 638  */
 639 STATIC void
 640 xfs_fs_destroy_inode(
 641         struct inode            *inode)
 642 {
 643         struct xfs_inode        *ip = XFS_I(inode);
 644
 645         trace_xfs_destroy_inode(ip);
 646
 647         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 648         XFS_STATS_INC(ip->i_mount, vn_rele);
 649         XFS_STATS_INC(ip->i_mount, vn_remove);
 650         xfs_inode_mark_reclaimable(ip);
 651 }
 652
 653 static void
 654 xfs_fs_dirty_inode(
 655         struct inode                    *inode,
 656         int                             flag)
 657 {
 658         struct xfs_inode                *ip = XFS_I(inode);
 659         struct xfs_mount                *mp = ip->i_mount;
 660         struct xfs_trans                *tp;
 661
 662         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 663                 return;
 664         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 665                 return;
 666
 667         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 668                 return;
 669         xfs_ilock(ip, XFS_ILOCK_EXCL);
 670         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 671         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 672         xfs_trans_commit(tp);
 673 }
 674
 675 /*
 676  * Slab object creation initialisation for the XFS inode.
 677  * This covers only the idempotent fields in the XFS inode;
 678  * all other fields need to be initialised on allocation
 679  * from the slab. This avoids the need to repeatedly initialise
 680  * fields in the xfs inode that left in the initialise state
 681  * when freeing the inode.
 682  */
 683 STATIC void
 684 xfs_fs_inode_init_once(
 685         void                    *inode)
 686 {
 687         struct xfs_inode        *ip = inode;
 688
 689         memset(ip, 0, sizeof(struct xfs_inode));
 690
 691         /* vfs inode */
 692         inode_init_once(VFS_I(ip));
 693
 694         /* xfs inode */
 695         atomic_set(&ip->i_pincount, 0);
 696         spin_lock_init(&ip->i_flags_lock);
 697
 698         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 699                      "xfsino", ip->i_ino);
 700 }
 701
 702 /*
 703  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 704  * serialised against cache hits here via the inode->i_lock and igrab() in
 705  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 706  * racing with us, and it avoids needing to grab a spinlock here for every inode
 707  * we drop the final reference on.
 708  */
 709 STATIC int
 710 xfs_fs_drop_inode(
 711         struct inode            *inode)
 712 {
 713         struct xfs_inode        *ip = XFS_I(inode);
 714
 715         /*
 716          * If this unlinked inode is in the middle of recovery, don't
 717          * drop the inode just yet; log recovery will take care of
 718          * that.  See the comment for this inode flag.
 719          */
 720         if (ip->i_flags & XFS_IRECOVERY) {
 721                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
 722                 return 0;
 723         }
 724
 725         return generic_drop_inode(inode);
 726 }
 727
 728 static void
 729 xfs_mount_free(
 730         struct xfs_mount        *mp)
 731 {
 732         kfree(mp->m_rtname);
 733         kfree(mp->m_logname);
 734         kmem_free(mp);
 735 }
 736
 737 STATIC int
 738 xfs_fs_sync_fs(
 739         struct super_block      *sb,
 740         int                     wait)
 741 {
 742         struct xfs_mount        *mp = XFS_M(sb);
 743         int                     error;
 744
 745         trace_xfs_fs_sync_fs(mp, __return_address);
 746
 747         /*
 748          * Doing anything during the async pass would be counterproductive.
 749          */
 750         if (!wait)
 751                 return 0;
 752
 753         error = xfs_log_force(mp, XFS_LOG_SYNC);
 754         if (error)
 755                 return error;
 756
 757         if (laptop_mode) {
 758                 /*
 759                  * The disk must be active because we're syncing.
 760                  * We schedule log work now (now that the disk is
 761                  * active) instead of later (when it might not be).
 762                  */
 763                 flush_delayed_work(&mp->m_log->l_work);
 764         }
 765
 766         /*
 767          * If we are called with page faults frozen out, it means we are about
 768          * to freeze the transaction subsystem. Take the opportunity to shut
 769          * down inodegc because once SB_FREEZE_FS is set it's too late to
 770          * prevent inactivation races with freeze. The fs doesn't get called
 771          * again by the freezing process until after SB_FREEZE_FS has been set,
 772          * so it's now or never.  Same logic applies to speculative allocation
 773          * garbage collection.
 774          *
 775          * We don't care if this is a normal syncfs call that does this or
 776          * freeze that does this - we can run this multiple times without issue
 777          * and we won't race with a restart because a restart can only occur
 778          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
 779          */
 780         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
 781                 xfs_inodegc_stop(mp);
 782                 xfs_blockgc_stop(mp);
 783         }
 784
 785         return 0;
 786 }
 787
 788 STATIC int
 789 xfs_fs_statfs(
 790         struct dentry           *dentry,
 791         struct kstatfs          *statp)
 792 {
 793         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 794         xfs_sb_t                *sbp = &mp->m_sb;
 795         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 796         uint64_t                fakeinos, id;
 797         uint64_t                icount;
 798         uint64_t                ifree;
 799         uint64_t                fdblocks;
 800         xfs_extlen_t            lsize;
 801         int64_t                 ffree;
 802
 803         /*
 804          * Expedite background inodegc but don't wait. We do not want to block
 805          * here waiting hours for a billion extent file to be truncated.
 806          */
 807         xfs_inodegc_push(mp);
 808
 809         statp->f_type = XFS_SUPER_MAGIC;
 810         statp->f_namelen = MAXNAMELEN - 1;
 811
 812         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 813         statp->f_fsid = u64_to_fsid(id);
 814
 815         icount = percpu_counter_sum(&mp->m_icount);
 816         ifree = percpu_counter_sum(&mp->m_ifree);
 817         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 818
 819         spin_lock(&mp->m_sb_lock);
 820         statp->f_bsize = sbp->sb_blocksize;
 821         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 822         statp->f_blocks = sbp->sb_dblocks - lsize;
 823         spin_unlock(&mp->m_sb_lock);
 824
 825         /* make sure statp->f_bfree does not underflow */
 826         statp->f_bfree = max_t(int64_t, 0,
 827                                 fdblocks - xfs_fdblocks_unavailable(mp));
 828         statp->f_bavail = statp->f_bfree;
 829
 830         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 831         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 832         if (M_IGEO(mp)->maxicount)
 833                 statp->f_files = min_t(typeof(statp->f_files),
 834                                         statp->f_files,
 835                                         M_IGEO(mp)->maxicount);
 836
 837         /* If sb_icount overshot maxicount, report actual allocation */
 838         statp->f_files = max_t(typeof(statp->f_files),
 839                                         statp->f_files,
 840                                         sbp->sb_icount);
 841
 842         /* make sure statp->f_ffree does not underflow */
 843         ffree = statp->f_files - (icount - ifree);
 844         statp->f_ffree = max_t(int64_t, ffree, 0);
 845
 846
 847         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 848             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 849                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 850                 xfs_qm_statvfs(ip, statp);
 851
 852         if (XFS_IS_REALTIME_MOUNT(mp) &&
 853             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 854                 s64     freertx;
 855
 856                 statp->f_blocks = sbp->sb_rblocks;
 857                 freertx = percpu_counter_sum_positive(&mp->m_frextents);
 858                 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
 859         }
 860
 861         return 0;
 862 }
 863
 864 STATIC void
 865 xfs_save_resvblks(struct xfs_mount *mp)
 866 {
 867         uint64_t resblks = 0;
 868
 869         mp->m_resblks_save = mp->m_resblks;
 870         xfs_reserve_blocks(mp, &resblks, NULL);
 871 }
 872
 873 STATIC void
 874 xfs_restore_resvblks(struct xfs_mount *mp)
 875 {
 876         uint64_t resblks;
 877
 878         if (mp->m_resblks_save) {
 879                 resblks = mp->m_resblks_save;
 880                 mp->m_resblks_save = 0;
 881         } else
 882                 resblks = xfs_default_resblks(mp);
 883
 884         xfs_reserve_blocks(mp, &resblks, NULL);
 885 }
 886
 887 /*
 888  * Second stage of a freeze. The data is already frozen so we only
 889  * need to take care of the metadata. Once that's done sync the superblock
 890  * to the log to dirty it in case of a crash while frozen. This ensures that we
 891  * will recover the unlinked inode lists on the next mount.
 892  */
 893 STATIC int
 894 xfs_fs_freeze(
 895         struct super_block      *sb)
 896 {
 897         struct xfs_mount        *mp = XFS_M(sb);
 898         unsigned int            flags;
 899         int                     ret;
 900
 901         /*
 902          * The filesystem is now frozen far enough that memory reclaim
 903          * cannot safely operate on the filesystem. Hence we need to
 904          * set a GFP_NOFS context here to avoid recursion deadlocks.
 905          */
 906         flags = memalloc_nofs_save();
 907         xfs_save_resvblks(mp);
 908         ret = xfs_log_quiesce(mp);
 909         memalloc_nofs_restore(flags);
 910
 911         /*
 912          * For read-write filesystems, we need to restart the inodegc on error
 913          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
 914          * going to be run to restart it now.  We are at SB_FREEZE_FS level
 915          * here, so we can restart safely without racing with a stop in
 916          * xfs_fs_sync_fs().
 917          */
 918         if (ret && !xfs_is_readonly(mp)) {
 919                 xfs_blockgc_start(mp);
 920                 xfs_inodegc_start(mp);
 921         }
 922
 923         return ret;
 924 }
 925
 926 STATIC int
 927 xfs_fs_unfreeze(
 928         struct super_block      *sb)
 929 {
 930         struct xfs_mount        *mp = XFS_M(sb);
 931
 932         xfs_restore_resvblks(mp);
 933         xfs_log_work_queue(mp);
 934
 935         /*
 936          * Don't reactivate the inodegc worker on a readonly filesystem because
 937          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
 938          * worker because there are no speculative preallocations on a readonly
 939          * filesystem.
 940          */
 941         if (!xfs_is_readonly(mp)) {
 942                 xfs_blockgc_start(mp);
 943                 xfs_inodegc_start(mp);
 944         }
 945
 946         return 0;
 947 }
 948
 949 /*
 950  * This function fills in xfs_mount_t fields based on mount args.
 951  * Note: the superblock _has_ now been read in.
 952  */
 953 STATIC int
 954 xfs_finish_flags(
 955         struct xfs_mount        *mp)
 956 {
 957         /* Fail a mount where the logbuf is smaller than the log stripe */
 958         if (xfs_has_logv2(mp)) {
 959                 if (mp->m_logbsize <= 0 &&
 960                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 961                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 962                 } else if (mp->m_logbsize > 0 &&
 963                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 964                         xfs_warn(mp,
 965                 "logbuf size must be greater than or equal to log stripe size");
 966                         return -EINVAL;
 967                 }
 968         } else {
 969                 /* Fail a mount if the logbuf is larger than 32K */
 970                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 971                         xfs_warn(mp,
 972                 "logbuf size for version 1 logs must be 16K or 32K");
 973                         return -EINVAL;
 974                 }
 975         }
 976
 977         /*
 978          * V5 filesystems always use attr2 format for attributes.
 979          */
 980         if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
 981                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 982                              "attr2 is always enabled for V5 filesystems.");
 983                 return -EINVAL;
 984         }
 985
 986         /*
 987          * prohibit r/w mounts of read-only filesystems
 988          */
 989         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
 990                 xfs_warn(mp,
 991                         "cannot mount a read-only filesystem as read-write");
 992                 return -EROFS;
 993         }
 994
 995         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
 996             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
 997             !xfs_has_pquotino(mp)) {
 998                 xfs_warn(mp,
 999                   "Super block does not support project and group quota together");
1000                 return -EINVAL;
1001         }
1002
1003         return 0;
1004 }
1005
1006 static int
1007 xfs_init_percpu_counters(
1008         struct xfs_mount        *mp)
1009 {
1010         int             error;
1011
1012         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1013         if (error)
1014                 return -ENOMEM;
1015
1016         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1017         if (error)
1018                 goto free_icount;
1019
1020         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1021         if (error)
1022                 goto free_ifree;
1023
1024         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1025         if (error)
1026                 goto free_fdblocks;
1027
1028         error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1029         if (error)
1030                 goto free_delalloc;
1031
1032         return 0;
1033
1034 free_delalloc:
1035         percpu_counter_destroy(&mp->m_delalloc_blks);
1036 free_fdblocks:
1037         percpu_counter_destroy(&mp->m_fdblocks);
1038 free_ifree:
1039         percpu_counter_destroy(&mp->m_ifree);
1040 free_icount:
1041         percpu_counter_destroy(&mp->m_icount);
1042         return -ENOMEM;
1043 }
1044
1045 void
1046 xfs_reinit_percpu_counters(
1047         struct xfs_mount        *mp)
1048 {
1049         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1050         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1051         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1052         percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1053 }
1054
1055 static void
1056 xfs_destroy_percpu_counters(
1057         struct xfs_mount        *mp)
1058 {
1059         percpu_counter_destroy(&mp->m_icount);
1060         percpu_counter_destroy(&mp->m_ifree);
1061         percpu_counter_destroy(&mp->m_fdblocks);
1062         ASSERT(xfs_is_shutdown(mp) ||
1063                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1064         percpu_counter_destroy(&mp->m_delalloc_blks);
1065         percpu_counter_destroy(&mp->m_frextents);
1066 }
1067
1068 static int
1069 xfs_inodegc_init_percpu(
1070         struct xfs_mount        *mp)
1071 {
1072         struct xfs_inodegc      *gc;
1073         int                     cpu;
1074
1075         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1076         if (!mp->m_inodegc)
1077                 return -ENOMEM;
1078
1079         for_each_possible_cpu(cpu) {
1080                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1081                 init_llist_head(&gc->list);
1082                 gc->items = 0;
1083                 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1084         }
1085         return 0;
1086 }
1087
1088 static void
1089 xfs_inodegc_free_percpu(
1090         struct xfs_mount        *mp)
1091 {
1092         if (!mp->m_inodegc)
1093                 return;
1094         free_percpu(mp->m_inodegc);
1095 }
1096
1097 static void
1098 xfs_fs_put_super(
1099         struct super_block      *sb)
1100 {
1101         struct xfs_mount        *mp = XFS_M(sb);
1102
1103         /* if ->fill_super failed, we have no mount to tear down */
1104         if (!sb->s_fs_info)
1105                 return;
1106
1107         xfs_notice(mp, "Unmounting Filesystem");
1108         xfs_filestream_unmount(mp);
1109         xfs_unmountfs(mp);
1110
1111         xfs_freesb(mp);
1112         free_percpu(mp->m_stats.xs_stats);
1113         xfs_mount_list_del(mp);
1114         xfs_inodegc_free_percpu(mp);
1115         xfs_destroy_percpu_counters(mp);
1116         xfs_destroy_mount_workqueues(mp);
1117         xfs_close_devices(mp);
1118
1119         sb->s_fs_info = NULL;
1120         xfs_mount_free(mp);
1121 }
1122
1123 static long
1124 xfs_fs_nr_cached_objects(
1125         struct super_block      *sb,
1126         struct shrink_control   *sc)
1127 {
1128         /* Paranoia: catch incorrect calls during mount setup or teardown */
1129         if (WARN_ON_ONCE(!sb->s_fs_info))
1130                 return 0;
1131         return xfs_reclaim_inodes_count(XFS_M(sb));
1132 }
1133
1134 static long
1135 xfs_fs_free_cached_objects(
1136         struct super_block      *sb,
1137         struct shrink_control   *sc)
1138 {
1139         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1140 }
1141
1142 static const struct super_operations xfs_super_operations = {
1143         .alloc_inode            = xfs_fs_alloc_inode,
1144         .destroy_inode          = xfs_fs_destroy_inode,
1145         .dirty_inode            = xfs_fs_dirty_inode,
1146         .drop_inode             = xfs_fs_drop_inode,
1147         .put_super              = xfs_fs_put_super,
1148         .sync_fs                = xfs_fs_sync_fs,
1149         .freeze_fs              = xfs_fs_freeze,
1150         .unfreeze_fs            = xfs_fs_unfreeze,
1151         .statfs                 = xfs_fs_statfs,
1152         .show_options           = xfs_fs_show_options,
1153         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1154         .free_cached_objects    = xfs_fs_free_cached_objects,
1155 };
1156
1157 static int
1158 suffix_kstrtoint(
1159         const char      *s,
1160         unsigned int    base,
1161         int             *res)
1162 {
1163         int             last, shift_left_factor = 0, _res;
1164         char            *value;
1165         int             ret = 0;
1166
1167         value = kstrdup(s, GFP_KERNEL);
1168         if (!value)
1169                 return -ENOMEM;
1170
1171         last = strlen(value) - 1;
1172         if (value[last] == 'K' || value[last] == 'k') {
1173                 shift_left_factor = 10;
1174                 value[last] = '\0';
1175         }
1176         if (value[last] == 'M' || value[last] == 'm') {
1177                 shift_left_factor = 20;
1178                 value[last] = '\0';
1179         }
1180         if (value[last] == 'G' || value[last] == 'g') {
1181                 shift_left_factor = 30;
1182                 value[last] = '\0';
1183         }
1184
1185         if (kstrtoint(value, base, &_res))
1186                 ret = -EINVAL;
1187         kfree(value);
1188         *res = _res << shift_left_factor;
1189         return ret;
1190 }
1191
1192 static inline void
1193 xfs_fs_warn_deprecated(
1194         struct fs_context       *fc,
1195         struct fs_parameter     *param,
1196         uint64_t                flag,
1197         bool                    value)
1198 {
1199         /* Don't print the warning if reconfiguring and current mount point
1200          * already had the flag set
1201          */
1202         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1203             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1204                 return;
1205         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1206 }
1207
1208 /*
1209  * Set mount state from a mount option.
1210  *
1211  * NOTE: mp->m_super is NULL here!
1212  */
1213 static int
1214 xfs_fs_parse_param(
1215         struct fs_context       *fc,
1216         struct fs_parameter     *param)
1217 {
1218         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1219         struct fs_parse_result  result;
1220         int                     size = 0;
1221         int                     opt;
1222
1223         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1224         if (opt < 0)
1225                 return opt;
1226
1227         switch (opt) {
1228         case Opt_logbufs:
1229                 parsing_mp->m_logbufs = result.uint_32;
1230                 return 0;
1231         case Opt_logbsize:
1232                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1233                         return -EINVAL;
1234                 return 0;
1235         case Opt_logdev:
1236                 kfree(parsing_mp->m_logname);
1237                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1238                 if (!parsing_mp->m_logname)
1239                         return -ENOMEM;
1240                 return 0;
1241         case Opt_rtdev:
1242                 kfree(parsing_mp->m_rtname);
1243                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1244                 if (!parsing_mp->m_rtname)
1245                         return -ENOMEM;
1246                 return 0;
1247         case Opt_allocsize:
1248                 if (suffix_kstrtoint(param->string, 10, &size))
1249                         return -EINVAL;
1250                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1251                 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1252                 return 0;
1253         case Opt_grpid:
1254         case Opt_bsdgroups:
1255                 parsing_mp->m_features |= XFS_FEAT_GRPID;
1256                 return 0;
1257         case Opt_nogrpid:
1258         case Opt_sysvgroups:
1259                 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1260                 return 0;
1261         case Opt_wsync:
1262                 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1263                 return 0;
1264         case Opt_norecovery:
1265                 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1266                 return 0;
1267         case Opt_noalign:
1268                 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1269                 return 0;
1270         case Opt_swalloc:
1271                 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1272                 return 0;
1273         case Opt_sunit:
1274                 parsing_mp->m_dalign = result.uint_32;
1275                 return 0;
1276         case Opt_swidth:
1277                 parsing_mp->m_swidth = result.uint_32;
1278                 return 0;
1279         case Opt_inode32:
1280                 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1281                 return 0;
1282         case Opt_inode64:
1283                 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1284                 return 0;
1285         case Opt_nouuid:
1286                 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1287                 return 0;
1288         case Opt_largeio:
1289                 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1290                 return 0;
1291         case Opt_nolargeio:
1292                 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1293                 return 0;
1294         case Opt_filestreams:
1295                 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1296                 return 0;
1297         case Opt_noquota:
1298                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1299                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1300                 return 0;
1301         case Opt_quota:
1302         case Opt_uquota:
1303         case Opt_usrquota:
1304                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1305                 return 0;
1306         case Opt_qnoenforce:
1307         case Opt_uqnoenforce:
1308                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1309                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1310                 return 0;
1311         case Opt_pquota:
1312         case Opt_prjquota:
1313                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1314                 return 0;
1315         case Opt_pqnoenforce:
1316                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1317                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1318                 return 0;
1319         case Opt_gquota:
1320         case Opt_grpquota:
1321                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1322                 return 0;
1323         case Opt_gqnoenforce:
1324                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1325                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1326                 return 0;
1327         case Opt_discard:
1328                 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1329                 return 0;
1330         case Opt_nodiscard:
1331                 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1332                 return 0;
1333 #ifdef CONFIG_FS_DAX
1334         case Opt_dax:
1335                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1336                 return 0;
1337         case Opt_dax_enum:
1338                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1339                 return 0;
1340 #endif
1341         /* Following mount options will be removed in September 2025 */
1342         case Opt_ikeep:
1343                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1344                 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1345                 return 0;
1346         case Opt_noikeep:
1347                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1348                 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1349                 return 0;
1350         case Opt_attr2:
1351                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1352                 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1353                 return 0;
1354         case Opt_noattr2:
1355                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1356                 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1357                 return 0;
1358         default:
1359                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1360                 return -EINVAL;
1361         }
1362
1363         return 0;
1364 }
1365
1366 static int
1367 xfs_fs_validate_params(
1368         struct xfs_mount        *mp)
1369 {
1370         /* No recovery flag requires a read-only mount */
1371         if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1372                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1373                 return -EINVAL;
1374         }
1375
1376         /*
1377          * We have not read the superblock at this point, so only the attr2
1378          * mount option can set the attr2 feature by this stage.
1379          */
1380         if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1381                 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1382                 return -EINVAL;
1383         }
1384
1385
1386         if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1387                 xfs_warn(mp,
1388         "sunit and swidth options incompatible with the noalign option");
1389                 return -EINVAL;
1390         }
1391
1392         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1393                 xfs_warn(mp, "quota support not available in this kernel.");
1394                 return -EINVAL;
1395         }
1396
1397         if ((mp->m_dalign && !mp->m_swidth) ||
1398             (!mp->m_dalign && mp->m_swidth)) {
1399                 xfs_warn(mp, "sunit and swidth must be specified together");
1400                 return -EINVAL;
1401         }
1402
1403         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1404                 xfs_warn(mp,
1405         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1406                         mp->m_swidth, mp->m_dalign);
1407                 return -EINVAL;
1408         }
1409
1410         if (mp->m_logbufs != -1 &&
1411             mp->m_logbufs != 0 &&
1412             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1413              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1414                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1415                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1416                 return -EINVAL;
1417         }
1418
1419         if (mp->m_logbsize != -1 &&
1420             mp->m_logbsize !=  0 &&
1421             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1422              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1423              !is_power_of_2(mp->m_logbsize))) {
1424                 xfs_warn(mp,
1425                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1426                         mp->m_logbsize);
1427                 return -EINVAL;
1428         }
1429
1430         if (xfs_has_allocsize(mp) &&
1431             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1432              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1433                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1434                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1435                 return -EINVAL;
1436         }
1437
1438         return 0;
1439 }
1440
1441 static int
1442 xfs_fs_fill_super(
1443         struct super_block      *sb,
1444         struct fs_context       *fc)
1445 {
1446         struct xfs_mount        *mp = sb->s_fs_info;
1447         struct inode            *root;
1448         int                     flags = 0, error;
1449
1450         mp->m_super = sb;
1451
1452         error = xfs_fs_validate_params(mp);
1453         if (error)
1454                 goto out_free_names;
1455
1456         sb_min_blocksize(sb, BBSIZE);
1457         sb->s_xattr = xfs_xattr_handlers;
1458         sb->s_export_op = &xfs_export_operations;
1459 #ifdef CONFIG_XFS_QUOTA
1460         sb->s_qcop = &xfs_quotactl_operations;
1461         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1462 #endif
1463         sb->s_op = &xfs_super_operations;
1464
1465         /*
1466          * Delay mount work if the debug hook is set. This is debug
1467          * instrumention to coordinate simulation of xfs mount failures with
1468          * VFS superblock operations
1469          */
1470         if (xfs_globals.mount_delay) {
1471                 xfs_notice(mp, "Delaying mount for %d seconds.",
1472                         xfs_globals.mount_delay);
1473                 msleep(xfs_globals.mount_delay * 1000);
1474         }
1475
1476         if (fc->sb_flags & SB_SILENT)
1477                 flags |= XFS_MFSI_QUIET;
1478
1479         error = xfs_open_devices(mp);
1480         if (error)
1481                 goto out_free_names;
1482
1483         error = xfs_init_mount_workqueues(mp);
1484         if (error)
1485                 goto out_close_devices;
1486
1487         error = xfs_init_percpu_counters(mp);
1488         if (error)
1489                 goto out_destroy_workqueues;
1490
1491         error = xfs_inodegc_init_percpu(mp);
1492         if (error)
1493                 goto out_destroy_counters;
1494
1495         /*
1496          * All percpu data structures requiring cleanup when a cpu goes offline
1497          * must be allocated before adding this @mp to the cpu-dead handler's
1498          * mount list.
1499          */
1500         xfs_mount_list_add(mp);
1501
1502         /* Allocate stats memory before we do operations that might use it */
1503         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1504         if (!mp->m_stats.xs_stats) {
1505                 error = -ENOMEM;
1506                 goto out_destroy_inodegc;
1507         }
1508
1509         error = xfs_readsb(mp, flags);
1510         if (error)
1511                 goto out_free_stats;
1512
1513         error = xfs_finish_flags(mp);
1514         if (error)
1515                 goto out_free_sb;
1516
1517         error = xfs_setup_devices(mp);
1518         if (error)
1519                 goto out_free_sb;
1520
1521         /* V4 support is undergoing deprecation. */
1522         if (!xfs_has_crc(mp)) {
1523 #ifdef CONFIG_XFS_SUPPORT_V4
1524                 xfs_warn_once(mp,
1525         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1526 #else
1527                 xfs_warn(mp,
1528         "Deprecated V4 format (crc=0) not supported by kernel.");
1529                 error = -EINVAL;
1530                 goto out_free_sb;
1531 #endif
1532         }
1533
1534         /* Filesystem claims it needs repair, so refuse the mount. */
1535         if (xfs_has_needsrepair(mp)) {
1536                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1537                 error = -EFSCORRUPTED;
1538                 goto out_free_sb;
1539         }
1540
1541         /*
1542          * Don't touch the filesystem if a user tool thinks it owns the primary
1543          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1544          * we don't check them at all.
1545          */
1546         if (mp->m_sb.sb_inprogress) {
1547                 xfs_warn(mp, "Offline file system operation in progress!");
1548                 error = -EFSCORRUPTED;
1549                 goto out_free_sb;
1550         }
1551
1552         /*
1553          * Until this is fixed only page-sized or smaller data blocks work.
1554          */
1555         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1556                 xfs_warn(mp,
1557                 "File system with blocksize %d bytes. "
1558                 "Only pagesize (%ld) or less will currently work.",
1559                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1560                 error = -ENOSYS;
1561                 goto out_free_sb;
1562         }
1563
1564         /* Ensure this filesystem fits in the page cache limits */
1565         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1566             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1567                 xfs_warn(mp,
1568                 "file system too large to be mounted on this system.");
1569                 error = -EFBIG;
1570                 goto out_free_sb;
1571         }
1572
1573         /*
1574          * XFS block mappings use 54 bits to store the logical block offset.
1575          * This should suffice to handle the maximum file size that the VFS
1576          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1577          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1578          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1579          * to check this assertion.
1580          *
1581          * Avoid integer overflow by comparing the maximum bmbt offset to the
1582          * maximum pagecache offset in units of fs blocks.
1583          */
1584         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1585                 xfs_warn(mp,
1586 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1587                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1588                          XFS_MAX_FILEOFF);
1589                 error = -EINVAL;
1590                 goto out_free_sb;
1591         }
1592
1593         error = xfs_filestream_mount(mp);
1594         if (error)
1595                 goto out_free_sb;
1596
1597         /*
1598          * we must configure the block size in the superblock before we run the
1599          * full mount process as the mount process can lookup and cache inodes.
1600          */
1601         sb->s_magic = XFS_SUPER_MAGIC;
1602         sb->s_blocksize = mp->m_sb.sb_blocksize;
1603         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1604         sb->s_maxbytes = MAX_LFS_FILESIZE;
1605         sb->s_max_links = XFS_MAXLINK;
1606         sb->s_time_gran = 1;
1607         if (xfs_has_bigtime(mp)) {
1608                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1609                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1610         } else {
1611                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1612                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1613         }
1614         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1615         sb->s_iflags |= SB_I_CGROUPWB;
1616
1617         set_posix_acl_flag(sb);
1618
1619         /* version 5 superblocks support inode version counters. */
1620         if (xfs_has_crc(mp))
1621                 sb->s_flags |= SB_I_VERSION;
1622
1623         if (xfs_has_dax_always(mp)) {
1624                 error = xfs_setup_dax_always(mp);
1625                 if (error)
1626                         goto out_filestream_unmount;
1627         }
1628
1629         if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1630                 xfs_warn(mp,
1631         "mounting with \"discard\" option, but the device does not support discard");
1632                 mp->m_features &= ~XFS_FEAT_DISCARD;
1633         }
1634
1635         if (xfs_has_reflink(mp)) {
1636                 if (mp->m_sb.sb_rblocks) {
1637                         xfs_alert(mp,
1638         "reflink not compatible with realtime device!");
1639                         error = -EINVAL;
1640                         goto out_filestream_unmount;
1641                 }
1642
1643                 if (xfs_globals.always_cow) {
1644                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1645                         mp->m_always_cow = true;
1646                 }
1647         }
1648
1649         if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1650                 xfs_alert(mp,
1651         "reverse mapping btree not compatible with realtime device!");
1652                 error = -EINVAL;
1653                 goto out_filestream_unmount;
1654         }
1655
1656         if (xfs_has_large_extent_counts(mp))
1657                 xfs_warn(mp,
1658         "EXPERIMENTAL Large extent counts feature in use. Use at your own risk!");
1659
1660         error = xfs_mountfs(mp);
1661         if (error)
1662                 goto out_filestream_unmount;
1663
1664         root = igrab(VFS_I(mp->m_rootip));
1665         if (!root) {
1666                 error = -ENOENT;
1667                 goto out_unmount;
1668         }
1669         sb->s_root = d_make_root(root);
1670         if (!sb->s_root) {
1671                 error = -ENOMEM;
1672                 goto out_unmount;
1673         }
1674
1675         return 0;
1676
1677  out_filestream_unmount:
1678         xfs_filestream_unmount(mp);
1679  out_free_sb:
1680         xfs_freesb(mp);
1681  out_free_stats:
1682         free_percpu(mp->m_stats.xs_stats);
1683  out_destroy_inodegc:
1684         xfs_mount_list_del(mp);
1685         xfs_inodegc_free_percpu(mp);
1686  out_destroy_counters:
1687         xfs_destroy_percpu_counters(mp);
1688  out_destroy_workqueues:
1689         xfs_destroy_mount_workqueues(mp);
1690  out_close_devices:
1691         xfs_close_devices(mp);
1692  out_free_names:
1693         sb->s_fs_info = NULL;
1694         xfs_mount_free(mp);
1695         return error;
1696
1697  out_unmount:
1698         xfs_filestream_unmount(mp);
1699         xfs_unmountfs(mp);
1700         goto out_free_sb;
1701 }
1702
1703 static int
1704 xfs_fs_get_tree(
1705         struct fs_context       *fc)
1706 {
1707         return get_tree_bdev(fc, xfs_fs_fill_super);
1708 }
1709
1710 static int
1711 xfs_remount_rw(
1712         struct xfs_mount        *mp)
1713 {
1714         struct xfs_sb           *sbp = &mp->m_sb;
1715         int error;
1716
1717         if (xfs_has_norecovery(mp)) {
1718                 xfs_warn(mp,
1719                         "ro->rw transition prohibited on norecovery mount");
1720                 return -EINVAL;
1721         }
1722
1723         if (xfs_sb_is_v5(sbp) &&
1724             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1725                 xfs_warn(mp,
1726         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1727                         (sbp->sb_features_ro_compat &
1728                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1729                 return -EINVAL;
1730         }
1731
1732         clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1733
1734         /*
1735          * If this is the first remount to writeable state we might have some
1736          * superblock changes to update.
1737          */
1738         if (mp->m_update_sb) {
1739                 error = xfs_sync_sb(mp, false);
1740                 if (error) {
1741                         xfs_warn(mp, "failed to write sb changes");
1742                         return error;
1743                 }
1744                 mp->m_update_sb = false;
1745         }
1746
1747         /*
1748          * Fill out the reserve pool if it is empty. Use the stashed value if
1749          * it is non-zero, otherwise go with the default.
1750          */
1751         xfs_restore_resvblks(mp);
1752         xfs_log_work_queue(mp);
1753         xfs_blockgc_start(mp);
1754
1755         /* Create the per-AG metadata reservation pool .*/
1756         error = xfs_fs_reserve_ag_blocks(mp);
1757         if (error && error != -ENOSPC)
1758                 return error;
1759
1760         /* Re-enable the background inode inactivation worker. */
1761         xfs_inodegc_start(mp);
1762
1763         return 0;
1764 }
1765
1766 static int
1767 xfs_remount_ro(
1768         struct xfs_mount        *mp)
1769 {
1770         struct xfs_icwalk       icw = {
1771                 .icw_flags      = XFS_ICWALK_FLAG_SYNC,
1772         };
1773         int                     error;
1774
1775         /* Flush all the dirty data to disk. */
1776         error = sync_filesystem(mp->m_super);
1777         if (error)
1778                 return error;
1779
1780         /*
1781          * Cancel background eofb scanning so it cannot race with the final
1782          * log force+buftarg wait and deadlock the remount.
1783          */
1784         xfs_blockgc_stop(mp);
1785
1786         /*
1787          * Clear out all remaining COW staging extents and speculative post-EOF
1788          * preallocations so that we don't leave inodes requiring inactivation
1789          * cleanups during reclaim on a read-only mount.  We must process every
1790          * cached inode, so this requires a synchronous cache scan.
1791          */
1792         error = xfs_blockgc_free_space(mp, &icw);
1793         if (error) {
1794                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1795                 return error;
1796         }
1797
1798         /*
1799          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1800          * flushed all pending inodegc work when it sync'd the filesystem.
1801          * The VFS holds s_umount, so we know that inodes cannot enter
1802          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1803          * we send inodes straight to reclaim, so no inodes will be queued.
1804          */
1805         xfs_inodegc_stop(mp);
1806
1807         /* Free the per-AG metadata reservation pool. */
1808         error = xfs_fs_unreserve_ag_blocks(mp);
1809         if (error) {
1810                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1811                 return error;
1812         }
1813
1814         /*
1815          * Before we sync the metadata, we need to free up the reserve block
1816          * pool so that the used block count in the superblock on disk is
1817          * correct at the end of the remount. Stash the current* reserve pool
1818          * size so that if we get remounted rw, we can return it to the same
1819          * size.
1820          */
1821         xfs_save_resvblks(mp);
1822
1823         xfs_log_clean(mp);
1824         set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1825
1826         return 0;
1827 }
1828
1829 /*
1830  * Logically we would return an error here to prevent users from believing
1831  * they might have changed mount options using remount which can't be changed.
1832  *
1833  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1834  * arguments in some cases so we can't blindly reject options, but have to
1835  * check for each specified option if it actually differs from the currently
1836  * set option and only reject it if that's the case.
1837  *
1838  * Until that is implemented we return success for every remount request, and
1839  * silently ignore all options that we can't actually change.
1840  */
1841 static int
1842 xfs_fs_reconfigure(
1843         struct fs_context *fc)
1844 {
1845         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1846         struct xfs_mount        *new_mp = fc->s_fs_info;
1847         int                     flags = fc->sb_flags;
1848         int                     error;
1849
1850         /* version 5 superblocks always support version counters. */
1851         if (xfs_has_crc(mp))
1852                 fc->sb_flags |= SB_I_VERSION;
1853
1854         error = xfs_fs_validate_params(new_mp);
1855         if (error)
1856                 return error;
1857
1858         /* inode32 -> inode64 */
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         /* inode64 -> inode32 */
1865         if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1866                 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1867                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1868         }
1869
1870         /* ro -> rw */
1871         if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1872                 error = xfs_remount_rw(mp);
1873                 if (error)
1874                         return error;
1875         }
1876
1877         /* rw -> ro */
1878         if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1879                 error = xfs_remount_ro(mp);
1880                 if (error)
1881                         return error;
1882         }
1883
1884         return 0;
1885 }
1886
1887 static void xfs_fs_free(
1888         struct fs_context       *fc)
1889 {
1890         struct xfs_mount        *mp = fc->s_fs_info;
1891
1892         /*
1893          * mp is stored in the fs_context when it is initialized.
1894          * mp is transferred to the superblock on a successful mount,
1895          * but if an error occurs before the transfer we have to free
1896          * it here.
1897          */
1898         if (mp)
1899                 xfs_mount_free(mp);
1900 }
1901
1902 static const struct fs_context_operations xfs_context_ops = {
1903         .parse_param = xfs_fs_parse_param,
1904         .get_tree    = xfs_fs_get_tree,
1905         .reconfigure = xfs_fs_reconfigure,
1906         .free        = xfs_fs_free,
1907 };
1908
1909 static int xfs_init_fs_context(
1910         struct fs_context       *fc)
1911 {
1912         struct xfs_mount        *mp;
1913
1914         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1915         if (!mp)
1916                 return -ENOMEM;
1917
1918         spin_lock_init(&mp->m_sb_lock);
1919         spin_lock_init(&mp->m_agirotor_lock);
1920         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1921         spin_lock_init(&mp->m_perag_lock);
1922         mutex_init(&mp->m_growlock);
1923         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1924         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1925         mp->m_kobj.kobject.kset = xfs_kset;
1926         /*
1927          * We don't create the finobt per-ag space reservation until after log
1928          * recovery, so we must set this to true so that an ifree transaction
1929          * started during log recovery will not depend on space reservations
1930          * for finobt expansion.
1931          */
1932         mp->m_finobt_nores = true;
1933
1934         /*
1935          * These can be overridden by the mount option parsing.
1936          */
1937         mp->m_logbufs = -1;
1938         mp->m_logbsize = -1;
1939         mp->m_allocsize_log = 16; /* 64k */
1940
1941         /*
1942          * Copy binary VFS mount flags we are interested in.
1943          */
1944         if (fc->sb_flags & SB_RDONLY)
1945                 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1946         if (fc->sb_flags & SB_DIRSYNC)
1947                 mp->m_features |= XFS_FEAT_DIRSYNC;
1948         if (fc->sb_flags & SB_SYNCHRONOUS)
1949                 mp->m_features |= XFS_FEAT_WSYNC;
1950
1951         fc->s_fs_info = mp;
1952         fc->ops = &xfs_context_ops;
1953
1954         return 0;
1955 }
1956
1957 static struct file_system_type xfs_fs_type = {
1958         .owner                  = THIS_MODULE,
1959         .name                   = "xfs",
1960         .init_fs_context        = xfs_init_fs_context,
1961         .parameters             = xfs_fs_parameters,
1962         .kill_sb                = kill_block_super,
1963         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1964 };
1965 MODULE_ALIAS_FS("xfs");
1966
1967 STATIC int __init
1968 xfs_init_caches(void)
1969 {
1970         int             error;
1971
1972         xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
1973                                          SLAB_HWCACHE_ALIGN |
1974                                          SLAB_RECLAIM_ACCOUNT |
1975                                          SLAB_MEM_SPREAD,
1976                                          NULL);
1977         if (!xfs_buf_cache)
1978                 goto out;
1979
1980         xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
1981                                                 sizeof(struct xlog_ticket),
1982                                                 0, 0, NULL);
1983         if (!xfs_log_ticket_cache)
1984                 goto out_destroy_buf_cache;
1985
1986         error = xfs_btree_init_cur_caches();
1987         if (error)
1988                 goto out_destroy_log_ticket_cache;
1989
1990         error = xfs_defer_init_item_caches();
1991         if (error)
1992                 goto out_destroy_btree_cur_cache;
1993
1994         xfs_da_state_cache = kmem_cache_create("xfs_da_state",
1995                                               sizeof(struct xfs_da_state),
1996                                               0, 0, NULL);
1997         if (!xfs_da_state_cache)
1998                 goto out_destroy_defer_item_cache;
1999
2000         xfs_ifork_cache = kmem_cache_create("xfs_ifork",
2001                                            sizeof(struct xfs_ifork),
2002                                            0, 0, NULL);
2003         if (!xfs_ifork_cache)
2004                 goto out_destroy_da_state_cache;
2005
2006         xfs_trans_cache = kmem_cache_create("xfs_trans",
2007                                            sizeof(struct xfs_trans),
2008                                            0, 0, NULL);
2009         if (!xfs_trans_cache)
2010                 goto out_destroy_ifork_cache;
2011
2012
2013         /*
2014          * The size of the cache-allocated buf log item is the maximum
2015          * size possible under XFS.  This wastes a little bit of memory,
2016          * but it is much faster.
2017          */
2018         xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2019                                               sizeof(struct xfs_buf_log_item),
2020                                               0, 0, NULL);
2021         if (!xfs_buf_item_cache)
2022                 goto out_destroy_trans_cache;
2023
2024         xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2025                                         (sizeof(struct xfs_efd_log_item) +
2026                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
2027                                         sizeof(struct xfs_extent)),
2028                                         0, 0, NULL);
2029         if (!xfs_efd_cache)
2030                 goto out_destroy_buf_item_cache;
2031
2032         xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2033                                          (sizeof(struct xfs_efi_log_item) +
2034                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2035                                          sizeof(struct xfs_extent)),
2036                                          0, 0, NULL);
2037         if (!xfs_efi_cache)
2038                 goto out_destroy_efd_cache;
2039
2040         xfs_inode_cache = kmem_cache_create("xfs_inode",
2041                                            sizeof(struct xfs_inode), 0,
2042                                            (SLAB_HWCACHE_ALIGN |
2043                                             SLAB_RECLAIM_ACCOUNT |
2044                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2045                                            xfs_fs_inode_init_once);
2046         if (!xfs_inode_cache)
2047                 goto out_destroy_efi_cache;
2048
2049         xfs_ili_cache = kmem_cache_create("xfs_ili",
2050                                          sizeof(struct xfs_inode_log_item), 0,
2051                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2052                                          NULL);
2053         if (!xfs_ili_cache)
2054                 goto out_destroy_inode_cache;
2055
2056         xfs_icreate_cache = kmem_cache_create("xfs_icr",
2057                                              sizeof(struct xfs_icreate_item),
2058                                              0, 0, NULL);
2059         if (!xfs_icreate_cache)
2060                 goto out_destroy_ili_cache;
2061
2062         xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2063                                          sizeof(struct xfs_rud_log_item),
2064                                          0, 0, NULL);
2065         if (!xfs_rud_cache)
2066                 goto out_destroy_icreate_cache;
2067
2068         xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2069                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2070                         0, 0, NULL);
2071         if (!xfs_rui_cache)
2072                 goto out_destroy_rud_cache;
2073
2074         xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2075                                          sizeof(struct xfs_cud_log_item),
2076                                          0, 0, NULL);
2077         if (!xfs_cud_cache)
2078                 goto out_destroy_rui_cache;
2079
2080         xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2081                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2082                         0, 0, NULL);
2083         if (!xfs_cui_cache)
2084                 goto out_destroy_cud_cache;
2085
2086         xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2087                                          sizeof(struct xfs_bud_log_item),
2088                                          0, 0, NULL);
2089         if (!xfs_bud_cache)
2090                 goto out_destroy_cui_cache;
2091
2092         xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2093                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2094                         0, 0, NULL);
2095         if (!xfs_bui_cache)
2096                 goto out_destroy_bud_cache;
2097
2098         xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2099                                             sizeof(struct xfs_attrd_log_item),
2100                                             0, 0, NULL);
2101         if (!xfs_attrd_cache)
2102                 goto out_destroy_bui_cache;
2103
2104         xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2105                                             sizeof(struct xfs_attri_log_item),
2106                                             0, 0, NULL);
2107         if (!xfs_attri_cache)
2108                 goto out_destroy_attrd_cache;
2109
2110         xfs_iunlink_cache = kmem_cache_create("xfs_iul_item",
2111                                              sizeof(struct xfs_iunlink_item),
2112                                              0, 0, NULL);
2113         if (!xfs_iunlink_cache)
2114                 goto out_destroy_attri_cache;
2115
2116         return 0;
2117
2118  out_destroy_attri_cache:
2119         kmem_cache_destroy(xfs_attri_cache);
2120  out_destroy_attrd_cache:
2121         kmem_cache_destroy(xfs_attrd_cache);
2122  out_destroy_bui_cache:
2123         kmem_cache_destroy(xfs_bui_cache);
2124  out_destroy_bud_cache:
2125         kmem_cache_destroy(xfs_bud_cache);
2126  out_destroy_cui_cache:
2127         kmem_cache_destroy(xfs_cui_cache);
2128  out_destroy_cud_cache:
2129         kmem_cache_destroy(xfs_cud_cache);
2130  out_destroy_rui_cache:
2131         kmem_cache_destroy(xfs_rui_cache);
2132  out_destroy_rud_cache:
2133         kmem_cache_destroy(xfs_rud_cache);
2134  out_destroy_icreate_cache:
2135         kmem_cache_destroy(xfs_icreate_cache);
2136  out_destroy_ili_cache:
2137         kmem_cache_destroy(xfs_ili_cache);
2138  out_destroy_inode_cache:
2139         kmem_cache_destroy(xfs_inode_cache);
2140  out_destroy_efi_cache:
2141         kmem_cache_destroy(xfs_efi_cache);
2142  out_destroy_efd_cache:
2143         kmem_cache_destroy(xfs_efd_cache);
2144  out_destroy_buf_item_cache:
2145         kmem_cache_destroy(xfs_buf_item_cache);
2146  out_destroy_trans_cache:
2147         kmem_cache_destroy(xfs_trans_cache);
2148  out_destroy_ifork_cache:
2149         kmem_cache_destroy(xfs_ifork_cache);
2150  out_destroy_da_state_cache:
2151         kmem_cache_destroy(xfs_da_state_cache);
2152  out_destroy_defer_item_cache:
2153         xfs_defer_destroy_item_caches();
2154  out_destroy_btree_cur_cache:
2155         xfs_btree_destroy_cur_caches();
2156  out_destroy_log_ticket_cache:
2157         kmem_cache_destroy(xfs_log_ticket_cache);
2158  out_destroy_buf_cache:
2159         kmem_cache_destroy(xfs_buf_cache);
2160  out:
2161         return -ENOMEM;
2162 }
2163
2164 STATIC void
2165 xfs_destroy_caches(void)
2166 {
2167         /*
2168          * Make sure all delayed rcu free are flushed before we
2169          * destroy caches.
2170          */
2171         rcu_barrier();
2172         kmem_cache_destroy(xfs_iunlink_cache);
2173         kmem_cache_destroy(xfs_attri_cache);
2174         kmem_cache_destroy(xfs_attrd_cache);
2175         kmem_cache_destroy(xfs_bui_cache);
2176         kmem_cache_destroy(xfs_bud_cache);
2177         kmem_cache_destroy(xfs_cui_cache);
2178         kmem_cache_destroy(xfs_cud_cache);
2179         kmem_cache_destroy(xfs_rui_cache);
2180         kmem_cache_destroy(xfs_rud_cache);
2181         kmem_cache_destroy(xfs_icreate_cache);
2182         kmem_cache_destroy(xfs_ili_cache);
2183         kmem_cache_destroy(xfs_inode_cache);
2184         kmem_cache_destroy(xfs_efi_cache);
2185         kmem_cache_destroy(xfs_efd_cache);
2186         kmem_cache_destroy(xfs_buf_item_cache);
2187         kmem_cache_destroy(xfs_trans_cache);
2188         kmem_cache_destroy(xfs_ifork_cache);
2189         kmem_cache_destroy(xfs_da_state_cache);
2190         xfs_defer_destroy_item_caches();
2191         xfs_btree_destroy_cur_caches();
2192         kmem_cache_destroy(xfs_log_ticket_cache);
2193         kmem_cache_destroy(xfs_buf_cache);
2194 }
2195
2196 STATIC int __init
2197 xfs_init_workqueues(void)
2198 {
2199         /*
2200          * The allocation workqueue can be used in memory reclaim situations
2201          * (writepage path), and parallelism is only limited by the number of
2202          * AGs in all the filesystems mounted. Hence use the default large
2203          * max_active value for this workqueue.
2204          */
2205         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2206                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2207         if (!xfs_alloc_wq)
2208                 return -ENOMEM;
2209
2210         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2211                         0);
2212         if (!xfs_discard_wq)
2213                 goto out_free_alloc_wq;
2214
2215         return 0;
2216 out_free_alloc_wq:
2217         destroy_workqueue(xfs_alloc_wq);
2218         return -ENOMEM;
2219 }
2220
2221 STATIC void
2222 xfs_destroy_workqueues(void)
2223 {
2224         destroy_workqueue(xfs_discard_wq);
2225         destroy_workqueue(xfs_alloc_wq);
2226 }
2227
2228 #ifdef CONFIG_HOTPLUG_CPU
2229 static int
2230 xfs_cpu_dead(
2231         unsigned int            cpu)
2232 {
2233         struct xfs_mount        *mp, *n;
2234
2235         spin_lock(&xfs_mount_list_lock);
2236         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2237                 spin_unlock(&xfs_mount_list_lock);
2238                 xfs_inodegc_cpu_dead(mp, cpu);
2239                 xlog_cil_pcp_dead(mp->m_log, cpu);
2240                 spin_lock(&xfs_mount_list_lock);
2241         }
2242         spin_unlock(&xfs_mount_list_lock);
2243         return 0;
2244 }
2245
2246 static int __init
2247 xfs_cpu_hotplug_init(void)
2248 {
2249         int     error;
2250
2251         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2252                         xfs_cpu_dead);
2253         if (error < 0)
2254                 xfs_alert(NULL,
2255 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2256                         error);
2257         return error;
2258 }
2259
2260 static void
2261 xfs_cpu_hotplug_destroy(void)
2262 {
2263         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2264 }
2265
2266 #else /* !CONFIG_HOTPLUG_CPU */
2267 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2268 static inline void xfs_cpu_hotplug_destroy(void) {}
2269 #endif
2270
2271 STATIC int __init
2272 init_xfs_fs(void)
2273 {
2274         int                     error;
2275
2276         xfs_check_ondisk_structs();
2277
2278         printk(KERN_INFO XFS_VERSION_STRING " with "
2279                          XFS_BUILD_OPTIONS " enabled\n");
2280
2281         xfs_dir_startup();
2282
2283         error = xfs_cpu_hotplug_init();
2284         if (error)
2285                 goto out;
2286
2287         error = xfs_init_caches();
2288         if (error)
2289                 goto out_destroy_hp;
2290
2291         error = xfs_init_workqueues();
2292         if (error)
2293                 goto out_destroy_caches;
2294
2295         error = xfs_mru_cache_init();
2296         if (error)
2297                 goto out_destroy_wq;
2298
2299         error = xfs_init_procfs();
2300         if (error)
2301                 goto out_mru_cache_uninit;
2302
2303         error = xfs_sysctl_register();
2304         if (error)
2305                 goto out_cleanup_procfs;
2306
2307         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2308         if (!xfs_kset) {
2309                 error = -ENOMEM;
2310                 goto out_sysctl_unregister;
2311         }
2312
2313         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2314
2315         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2316         if (!xfsstats.xs_stats) {
2317                 error = -ENOMEM;
2318                 goto out_kset_unregister;
2319         }
2320
2321         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2322                                "stats");
2323         if (error)
2324                 goto out_free_stats;
2325
2326 #ifdef DEBUG
2327         xfs_dbg_kobj.kobject.kset = xfs_kset;
2328         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2329         if (error)
2330                 goto out_remove_stats_kobj;
2331 #endif
2332
2333         error = xfs_qm_init();
2334         if (error)
2335                 goto out_remove_dbg_kobj;
2336
2337         error = register_filesystem(&xfs_fs_type);
2338         if (error)
2339                 goto out_qm_exit;
2340         return 0;
2341
2342  out_qm_exit:
2343         xfs_qm_exit();
2344  out_remove_dbg_kobj:
2345 #ifdef DEBUG
2346         xfs_sysfs_del(&xfs_dbg_kobj);
2347  out_remove_stats_kobj:
2348 #endif
2349         xfs_sysfs_del(&xfsstats.xs_kobj);
2350  out_free_stats:
2351         free_percpu(xfsstats.xs_stats);
2352  out_kset_unregister:
2353         kset_unregister(xfs_kset);
2354  out_sysctl_unregister:
2355         xfs_sysctl_unregister();
2356  out_cleanup_procfs:
2357         xfs_cleanup_procfs();
2358  out_mru_cache_uninit:
2359         xfs_mru_cache_uninit();
2360  out_destroy_wq:
2361         xfs_destroy_workqueues();
2362  out_destroy_caches:
2363         xfs_destroy_caches();
2364  out_destroy_hp:
2365         xfs_cpu_hotplug_destroy();
2366  out:
2367         return error;
2368 }
2369
2370 STATIC void __exit
2371 exit_xfs_fs(void)
2372 {
2373         xfs_qm_exit();
2374         unregister_filesystem(&xfs_fs_type);
2375 #ifdef DEBUG
2376         xfs_sysfs_del(&xfs_dbg_kobj);
2377 #endif
2378         xfs_sysfs_del(&xfsstats.xs_kobj);
2379         free_percpu(xfsstats.xs_stats);
2380         kset_unregister(xfs_kset);
2381         xfs_sysctl_unregister();
2382         xfs_cleanup_procfs();
2383         xfs_mru_cache_uninit();
2384         xfs_destroy_workqueues();
2385         xfs_destroy_caches();
2386         xfs_uuid_table_free();
2387         xfs_cpu_hotplug_destroy();
2388 }
2389
2390 module_init(init_xfs_fs);
2391 module_exit(exit_xfs_fs);
2392
2393 MODULE_AUTHOR("Silicon Graphics, Inc.");
2394 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2395 MODULE_LICENSE("GPL");