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
  40 #include <linux/magic.h>
  41 #include <linux/fs_context.h>
  42 #include <linux/fs_parser.h>
  43
  44 static const struct super_operations xfs_super_operations;
  45
  46 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  47 #ifdef DEBUG
  48 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  49 #endif
  50
  51 enum xfs_dax_mode {
  52         XFS_DAX_INODE = 0,
  53         XFS_DAX_ALWAYS = 1,
  54         XFS_DAX_NEVER = 2,
  55 };
  56
  57 static void
  58 xfs_mount_set_dax_mode(
  59         struct xfs_mount        *mp,
  60         enum xfs_dax_mode       mode)
  61 {
  62         switch (mode) {
  63         case XFS_DAX_INODE:
  64                 mp->m_flags &= ~(XFS_MOUNT_DAX_ALWAYS | XFS_MOUNT_DAX_NEVER);
  65                 break;
  66         case XFS_DAX_ALWAYS:
  67                 mp->m_flags |= XFS_MOUNT_DAX_ALWAYS;
  68                 mp->m_flags &= ~XFS_MOUNT_DAX_NEVER;
  69                 break;
  70         case XFS_DAX_NEVER:
  71                 mp->m_flags |= XFS_MOUNT_DAX_NEVER;
  72                 mp->m_flags &= ~XFS_MOUNT_DAX_ALWAYS;
  73                 break;
  74         }
  75 }
  76
  77 static const struct constant_table dax_param_enums[] = {
  78         {"inode",       XFS_DAX_INODE },
  79         {"always",      XFS_DAX_ALWAYS },
  80         {"never",       XFS_DAX_NEVER },
  81         {}
  82 };
  83
  84 /*
  85  * Table driven mount option parser.
  86  */
  87 enum {
  88         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
  89         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
  90         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
  91         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
  92         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
  93         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
  94         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
  95         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
  96         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
  97 };
  98
  99 static const struct fs_parameter_spec xfs_fs_parameters[] = {
 100         fsparam_u32("logbufs",          Opt_logbufs),
 101         fsparam_string("logbsize",      Opt_logbsize),
 102         fsparam_string("logdev",        Opt_logdev),
 103         fsparam_string("rtdev",         Opt_rtdev),
 104         fsparam_flag("wsync",           Opt_wsync),
 105         fsparam_flag("noalign",         Opt_noalign),
 106         fsparam_flag("swalloc",         Opt_swalloc),
 107         fsparam_u32("sunit",            Opt_sunit),
 108         fsparam_u32("swidth",           Opt_swidth),
 109         fsparam_flag("nouuid",          Opt_nouuid),
 110         fsparam_flag("grpid",           Opt_grpid),
 111         fsparam_flag("nogrpid",         Opt_nogrpid),
 112         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 113         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 114         fsparam_string("allocsize",     Opt_allocsize),
 115         fsparam_flag("norecovery",      Opt_norecovery),
 116         fsparam_flag("inode64",         Opt_inode64),
 117         fsparam_flag("inode32",         Opt_inode32),
 118         fsparam_flag("ikeep",           Opt_ikeep),
 119         fsparam_flag("noikeep",         Opt_noikeep),
 120         fsparam_flag("largeio",         Opt_largeio),
 121         fsparam_flag("nolargeio",       Opt_nolargeio),
 122         fsparam_flag("attr2",           Opt_attr2),
 123         fsparam_flag("noattr2",         Opt_noattr2),
 124         fsparam_flag("filestreams",     Opt_filestreams),
 125         fsparam_flag("quota",           Opt_quota),
 126         fsparam_flag("noquota",         Opt_noquota),
 127         fsparam_flag("usrquota",        Opt_usrquota),
 128         fsparam_flag("grpquota",        Opt_grpquota),
 129         fsparam_flag("prjquota",        Opt_prjquota),
 130         fsparam_flag("uquota",          Opt_uquota),
 131         fsparam_flag("gquota",          Opt_gquota),
 132         fsparam_flag("pquota",          Opt_pquota),
 133         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 134         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 135         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 136         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 137         fsparam_flag("discard",         Opt_discard),
 138         fsparam_flag("nodiscard",       Opt_nodiscard),
 139         fsparam_flag("dax",             Opt_dax),
 140         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 141         {}
 142 };
 143
 144 struct proc_xfs_info {
 145         uint64_t        flag;
 146         char            *str;
 147 };
 148
 149 static int
 150 xfs_fs_show_options(
 151         struct seq_file         *m,
 152         struct dentry           *root)
 153 {
 154         static struct proc_xfs_info xfs_info_set[] = {
 155                 /* the few simple ones we can get from the mount struct */
 156                 { XFS_MOUNT_IKEEP,              ",ikeep" },
 157                 { XFS_MOUNT_WSYNC,              ",wsync" },
 158                 { XFS_MOUNT_NOALIGN,            ",noalign" },
 159                 { XFS_MOUNT_SWALLOC,            ",swalloc" },
 160                 { XFS_MOUNT_NOUUID,             ",nouuid" },
 161                 { XFS_MOUNT_NORECOVERY,         ",norecovery" },
 162                 { XFS_MOUNT_ATTR2,              ",attr2" },
 163                 { XFS_MOUNT_FILESTREAMS,        ",filestreams" },
 164                 { XFS_MOUNT_GRPID,              ",grpid" },
 165                 { XFS_MOUNT_DISCARD,            ",discard" },
 166                 { XFS_MOUNT_LARGEIO,            ",largeio" },
 167                 { XFS_MOUNT_DAX_ALWAYS,         ",dax=always" },
 168                 { XFS_MOUNT_DAX_NEVER,          ",dax=never" },
 169                 { 0, NULL }
 170         };
 171         struct xfs_mount        *mp = XFS_M(root->d_sb);
 172         struct proc_xfs_info    *xfs_infop;
 173
 174         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 175                 if (mp->m_flags & xfs_infop->flag)
 176                         seq_puts(m, xfs_infop->str);
 177         }
 178
 179         seq_printf(m, ",inode%d",
 180                 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
 181
 182         if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
 183                 seq_printf(m, ",allocsize=%dk",
 184                            (1 << mp->m_allocsize_log) >> 10);
 185
 186         if (mp->m_logbufs > 0)
 187                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 188         if (mp->m_logbsize > 0)
 189                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 190
 191         if (mp->m_logname)
 192                 seq_show_option(m, "logdev", mp->m_logname);
 193         if (mp->m_rtname)
 194                 seq_show_option(m, "rtdev", mp->m_rtname);
 195
 196         if (mp->m_dalign > 0)
 197                 seq_printf(m, ",sunit=%d",
 198                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 199         if (mp->m_swidth > 0)
 200                 seq_printf(m, ",swidth=%d",
 201                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 202
 203         if (mp->m_qflags & XFS_UQUOTA_ACCT) {
 204                 if (mp->m_qflags & XFS_UQUOTA_ENFD)
 205                         seq_puts(m, ",usrquota");
 206                 else
 207                         seq_puts(m, ",uqnoenforce");
 208         }
 209
 210         if (mp->m_qflags & XFS_PQUOTA_ACCT) {
 211                 if (mp->m_qflags & XFS_PQUOTA_ENFD)
 212                         seq_puts(m, ",prjquota");
 213                 else
 214                         seq_puts(m, ",pqnoenforce");
 215         }
 216         if (mp->m_qflags & XFS_GQUOTA_ACCT) {
 217                 if (mp->m_qflags & XFS_GQUOTA_ENFD)
 218                         seq_puts(m, ",grpquota");
 219                 else
 220                         seq_puts(m, ",gqnoenforce");
 221         }
 222
 223         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 224                 seq_puts(m, ",noquota");
 225
 226         return 0;
 227 }
 228
 229 /*
 230  * Set parameters for inode allocation heuristics, taking into account
 231  * filesystem size and inode32/inode64 mount options; i.e. specifically
 232  * whether or not XFS_MOUNT_SMALL_INUMS is set.
 233  *
 234  * Inode allocation patterns are altered only if inode32 is requested
 235  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
 236  * If altered, XFS_MOUNT_32BITINODES is set as well.
 237  *
 238  * An agcount independent of that in the mount structure is provided
 239  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 240  * to the potentially higher ag count.
 241  *
 242  * Returns the maximum AG index which may contain inodes.
 243  */
 244 xfs_agnumber_t
 245 xfs_set_inode_alloc(
 246         struct xfs_mount *mp,
 247         xfs_agnumber_t  agcount)
 248 {
 249         xfs_agnumber_t  index;
 250         xfs_agnumber_t  maxagi = 0;
 251         xfs_sb_t        *sbp = &mp->m_sb;
 252         xfs_agnumber_t  max_metadata;
 253         xfs_agino_t     agino;
 254         xfs_ino_t       ino;
 255
 256         /*
 257          * Calculate how much should be reserved for inodes to meet
 258          * the max inode percentage.  Used only for inode32.
 259          */
 260         if (M_IGEO(mp)->maxicount) {
 261                 uint64_t        icount;
 262
 263                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 264                 do_div(icount, 100);
 265                 icount += sbp->sb_agblocks - 1;
 266                 do_div(icount, sbp->sb_agblocks);
 267                 max_metadata = icount;
 268         } else {
 269                 max_metadata = agcount;
 270         }
 271
 272         /* Get the last possible inode in the filesystem */
 273         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 274         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 275
 276         /*
 277          * If user asked for no more than 32-bit inodes, and the fs is
 278          * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
 279          * the allocator to accommodate the request.
 280          */
 281         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
 282                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 283         else
 284                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
 285
 286         for (index = 0; index < agcount; index++) {
 287                 struct xfs_perag        *pag;
 288
 289                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 290
 291                 pag = xfs_perag_get(mp, index);
 292
 293                 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
 294                         if (ino > XFS_MAXINUMBER_32) {
 295                                 pag->pagi_inodeok = 0;
 296                                 pag->pagf_metadata = 0;
 297                         } else {
 298                                 pag->pagi_inodeok = 1;
 299                                 maxagi++;
 300                                 if (index < max_metadata)
 301                                         pag->pagf_metadata = 1;
 302                                 else
 303                                         pag->pagf_metadata = 0;
 304                         }
 305                 } else {
 306                         pag->pagi_inodeok = 1;
 307                         pag->pagf_metadata = 0;
 308                 }
 309
 310                 xfs_perag_put(pag);
 311         }
 312
 313         return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
 314 }
 315
 316 STATIC int
 317 xfs_blkdev_get(
 318         xfs_mount_t             *mp,
 319         const char              *name,
 320         struct block_device     **bdevp)
 321 {
 322         int                     error = 0;
 323
 324         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 325                                     mp);
 326         if (IS_ERR(*bdevp)) {
 327                 error = PTR_ERR(*bdevp);
 328                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 329         }
 330
 331         return error;
 332 }
 333
 334 STATIC void
 335 xfs_blkdev_put(
 336         struct block_device     *bdev)
 337 {
 338         if (bdev)
 339                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 340 }
 341
 342 void
 343 xfs_blkdev_issue_flush(
 344         xfs_buftarg_t           *buftarg)
 345 {
 346         blkdev_issue_flush(buftarg->bt_bdev);
 347 }
 348
 349 STATIC void
 350 xfs_close_devices(
 351         struct xfs_mount        *mp)
 352 {
 353         struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
 354
 355         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 356                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 357                 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
 358
 359                 xfs_free_buftarg(mp->m_logdev_targp);
 360                 xfs_blkdev_put(logdev);
 361                 fs_put_dax(dax_logdev);
 362         }
 363         if (mp->m_rtdev_targp) {
 364                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 365                 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
 366
 367                 xfs_free_buftarg(mp->m_rtdev_targp);
 368                 xfs_blkdev_put(rtdev);
 369                 fs_put_dax(dax_rtdev);
 370         }
 371         xfs_free_buftarg(mp->m_ddev_targp);
 372         fs_put_dax(dax_ddev);
 373 }
 374
 375 /*
 376  * The file system configurations are:
 377  *      (1) device (partition) with data and internal log
 378  *      (2) logical volume with data and log subvolumes.
 379  *      (3) logical volume with data, log, and realtime subvolumes.
 380  *
 381  * We only have to handle opening the log and realtime volumes here if
 382  * they are present.  The data subvolume has already been opened by
 383  * get_sb_bdev() and is stored in sb->s_bdev.
 384  */
 385 STATIC int
 386 xfs_open_devices(
 387         struct xfs_mount        *mp)
 388 {
 389         struct block_device     *ddev = mp->m_super->s_bdev;
 390         struct dax_device       *dax_ddev = fs_dax_get_by_bdev(ddev);
 391         struct dax_device       *dax_logdev = NULL, *dax_rtdev = NULL;
 392         struct block_device     *logdev = NULL, *rtdev = NULL;
 393         int                     error;
 394
 395         /*
 396          * Open real time and log devices - order is important.
 397          */
 398         if (mp->m_logname) {
 399                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 400                 if (error)
 401                         goto out;
 402                 dax_logdev = fs_dax_get_by_bdev(logdev);
 403         }
 404
 405         if (mp->m_rtname) {
 406                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 407                 if (error)
 408                         goto out_close_logdev;
 409
 410                 if (rtdev == ddev || rtdev == logdev) {
 411                         xfs_warn(mp,
 412         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 413                         error = -EINVAL;
 414                         goto out_close_rtdev;
 415                 }
 416                 dax_rtdev = fs_dax_get_by_bdev(rtdev);
 417         }
 418
 419         /*
 420          * Setup xfs_mount buffer target pointers
 421          */
 422         error = -ENOMEM;
 423         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
 424         if (!mp->m_ddev_targp)
 425                 goto out_close_rtdev;
 426
 427         if (rtdev) {
 428                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
 429                 if (!mp->m_rtdev_targp)
 430                         goto out_free_ddev_targ;
 431         }
 432
 433         if (logdev && logdev != ddev) {
 434                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
 435                 if (!mp->m_logdev_targp)
 436                         goto out_free_rtdev_targ;
 437         } else {
 438                 mp->m_logdev_targp = mp->m_ddev_targp;
 439         }
 440
 441         return 0;
 442
 443  out_free_rtdev_targ:
 444         if (mp->m_rtdev_targp)
 445                 xfs_free_buftarg(mp->m_rtdev_targp);
 446  out_free_ddev_targ:
 447         xfs_free_buftarg(mp->m_ddev_targp);
 448  out_close_rtdev:
 449         xfs_blkdev_put(rtdev);
 450         fs_put_dax(dax_rtdev);
 451  out_close_logdev:
 452         if (logdev && logdev != ddev) {
 453                 xfs_blkdev_put(logdev);
 454                 fs_put_dax(dax_logdev);
 455         }
 456  out:
 457         fs_put_dax(dax_ddev);
 458         return error;
 459 }
 460
 461 /*
 462  * Setup xfs_mount buffer target pointers based on superblock
 463  */
 464 STATIC int
 465 xfs_setup_devices(
 466         struct xfs_mount        *mp)
 467 {
 468         int                     error;
 469
 470         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 471         if (error)
 472                 return error;
 473
 474         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 475                 unsigned int    log_sector_size = BBSIZE;
 476
 477                 if (xfs_sb_version_hassector(&mp->m_sb))
 478                         log_sector_size = mp->m_sb.sb_logsectsize;
 479                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 480                                             log_sector_size);
 481                 if (error)
 482                         return error;
 483         }
 484         if (mp->m_rtdev_targp) {
 485                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 486                                             mp->m_sb.sb_sectsize);
 487                 if (error)
 488                         return error;
 489         }
 490
 491         return 0;
 492 }
 493
 494 STATIC int
 495 xfs_init_mount_workqueues(
 496         struct xfs_mount        *mp)
 497 {
 498         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 499                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 500                         1, mp->m_super->s_id);
 501         if (!mp->m_buf_workqueue)
 502                 goto out;
 503
 504         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 505                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 506                         0, mp->m_super->s_id);
 507         if (!mp->m_unwritten_workqueue)
 508                 goto out_destroy_buf;
 509
 510         mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
 511                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | WQ_UNBOUND),
 512                         0, mp->m_super->s_id);
 513         if (!mp->m_cil_workqueue)
 514                 goto out_destroy_unwritten;
 515
 516         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 517                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 518                         0, mp->m_super->s_id);
 519         if (!mp->m_reclaim_workqueue)
 520                 goto out_destroy_cil;
 521
 522         mp->m_gc_workqueue = alloc_workqueue("xfs-gc/%s",
 523                         WQ_SYSFS | WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM,
 524                         0, mp->m_super->s_id);
 525         if (!mp->m_gc_workqueue)
 526                 goto out_destroy_reclaim;
 527
 528         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 529                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 530         if (!mp->m_sync_workqueue)
 531                 goto out_destroy_eofb;
 532
 533         return 0;
 534
 535 out_destroy_eofb:
 536         destroy_workqueue(mp->m_gc_workqueue);
 537 out_destroy_reclaim:
 538         destroy_workqueue(mp->m_reclaim_workqueue);
 539 out_destroy_cil:
 540         destroy_workqueue(mp->m_cil_workqueue);
 541 out_destroy_unwritten:
 542         destroy_workqueue(mp->m_unwritten_workqueue);
 543 out_destroy_buf:
 544         destroy_workqueue(mp->m_buf_workqueue);
 545 out:
 546         return -ENOMEM;
 547 }
 548
 549 STATIC void
 550 xfs_destroy_mount_workqueues(
 551         struct xfs_mount        *mp)
 552 {
 553         destroy_workqueue(mp->m_sync_workqueue);
 554         destroy_workqueue(mp->m_gc_workqueue);
 555         destroy_workqueue(mp->m_reclaim_workqueue);
 556         destroy_workqueue(mp->m_cil_workqueue);
 557         destroy_workqueue(mp->m_unwritten_workqueue);
 558         destroy_workqueue(mp->m_buf_workqueue);
 559 }
 560
 561 static void
 562 xfs_flush_inodes_worker(
 563         struct work_struct      *work)
 564 {
 565         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 566                                                    m_flush_inodes_work);
 567         struct super_block      *sb = mp->m_super;
 568
 569         if (down_read_trylock(&sb->s_umount)) {
 570                 sync_inodes_sb(sb);
 571                 up_read(&sb->s_umount);
 572         }
 573 }
 574
 575 /*
 576  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 577  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 578  * for IO to complete so that we effectively throttle multiple callers to the
 579  * rate at which IO is completing.
 580  */
 581 void
 582 xfs_flush_inodes(
 583         struct xfs_mount        *mp)
 584 {
 585         /*
 586          * If flush_work() returns true then that means we waited for a flush
 587          * which was already in progress.  Don't bother running another scan.
 588          */
 589         if (flush_work(&mp->m_flush_inodes_work))
 590                 return;
 591
 592         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 593         flush_work(&mp->m_flush_inodes_work);
 594 }
 595
 596 /* Catch misguided souls that try to use this interface on XFS */
 597 STATIC struct inode *
 598 xfs_fs_alloc_inode(
 599         struct super_block      *sb)
 600 {
 601         BUG();
 602         return NULL;
 603 }
 604
 605 #ifdef DEBUG
 606 static void
 607 xfs_check_delalloc(
 608         struct xfs_inode        *ip,
 609         int                     whichfork)
 610 {
 611         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 612         struct xfs_bmbt_irec    got;
 613         struct xfs_iext_cursor  icur;
 614
 615         if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
 616                 return;
 617         do {
 618                 if (isnullstartblock(got.br_startblock)) {
 619                         xfs_warn(ip->i_mount,
 620         "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
 621                                 ip->i_ino,
 622                                 whichfork == XFS_DATA_FORK ? "data" : "cow",
 623                                 got.br_startoff, got.br_blockcount);
 624                 }
 625         } while (xfs_iext_next_extent(ifp, &icur, &got));
 626 }
 627 #else
 628 #define xfs_check_delalloc(ip, whichfork)       do { } while (0)
 629 #endif
 630
 631 /*
 632  * Now that the generic code is guaranteed not to be accessing
 633  * the linux inode, we can inactivate and reclaim the inode.
 634  */
 635 STATIC void
 636 xfs_fs_destroy_inode(
 637         struct inode            *inode)
 638 {
 639         struct xfs_inode        *ip = XFS_I(inode);
 640
 641         trace_xfs_destroy_inode(ip);
 642
 643         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 644         XFS_STATS_INC(ip->i_mount, vn_rele);
 645         XFS_STATS_INC(ip->i_mount, vn_remove);
 646
 647         xfs_inactive(ip);
 648
 649         if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
 650                 xfs_check_delalloc(ip, XFS_DATA_FORK);
 651                 xfs_check_delalloc(ip, XFS_COW_FORK);
 652                 ASSERT(0);
 653         }
 654
 655         XFS_STATS_INC(ip->i_mount, vn_reclaim);
 656
 657         /*
 658          * We should never get here with one of the reclaim flags already set.
 659          */
 660         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
 661         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
 662
 663         /*
 664          * We always use background reclaim here because even if the inode is
 665          * clean, it still may be under IO and hence we have wait for IO
 666          * completion to occur before we can reclaim the inode. The background
 667          * reclaim path handles this more efficiently than we can here, so
 668          * simply let background reclaim tear down all inodes.
 669          */
 670         xfs_inode_set_reclaim_tag(ip);
 671 }
 672
 673 static void
 674 xfs_fs_dirty_inode(
 675         struct inode                    *inode,
 676         int                             flag)
 677 {
 678         struct xfs_inode                *ip = XFS_I(inode);
 679         struct xfs_mount                *mp = ip->i_mount;
 680         struct xfs_trans                *tp;
 681
 682         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 683                 return;
 684         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 685                 return;
 686
 687         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 688                 return;
 689         xfs_ilock(ip, XFS_ILOCK_EXCL);
 690         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 691         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 692         xfs_trans_commit(tp);
 693 }
 694
 695 /*
 696  * Slab object creation initialisation for the XFS inode.
 697  * This covers only the idempotent fields in the XFS inode;
 698  * all other fields need to be initialised on allocation
 699  * from the slab. This avoids the need to repeatedly initialise
 700  * fields in the xfs inode that left in the initialise state
 701  * when freeing the inode.
 702  */
 703 STATIC void
 704 xfs_fs_inode_init_once(
 705         void                    *inode)
 706 {
 707         struct xfs_inode        *ip = inode;
 708
 709         memset(ip, 0, sizeof(struct xfs_inode));
 710
 711         /* vfs inode */
 712         inode_init_once(VFS_I(ip));
 713
 714         /* xfs inode */
 715         atomic_set(&ip->i_pincount, 0);
 716         spin_lock_init(&ip->i_flags_lock);
 717
 718         mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 719                      "xfsino", ip->i_ino);
 720         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 721                      "xfsino", ip->i_ino);
 722 }
 723
 724 /*
 725  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 726  * serialised against cache hits here via the inode->i_lock and igrab() in
 727  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 728  * racing with us, and it avoids needing to grab a spinlock here for every inode
 729  * we drop the final reference on.
 730  */
 731 STATIC int
 732 xfs_fs_drop_inode(
 733         struct inode            *inode)
 734 {
 735         struct xfs_inode        *ip = XFS_I(inode);
 736
 737         /*
 738          * If this unlinked inode is in the middle of recovery, don't
 739          * drop the inode just yet; log recovery will take care of
 740          * that.  See the comment for this inode flag.
 741          */
 742         if (ip->i_flags & XFS_IRECOVERY) {
 743                 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
 744                 return 0;
 745         }
 746
 747         return generic_drop_inode(inode);
 748 }
 749
 750 static void
 751 xfs_mount_free(
 752         struct xfs_mount        *mp)
 753 {
 754         kfree(mp->m_rtname);
 755         kfree(mp->m_logname);
 756         kmem_free(mp);
 757 }
 758
 759 STATIC int
 760 xfs_fs_sync_fs(
 761         struct super_block      *sb,
 762         int                     wait)
 763 {
 764         struct xfs_mount        *mp = XFS_M(sb);
 765
 766         /*
 767          * Doing anything during the async pass would be counterproductive.
 768          */
 769         if (!wait)
 770                 return 0;
 771
 772         xfs_log_force(mp, XFS_LOG_SYNC);
 773         if (laptop_mode) {
 774                 /*
 775                  * The disk must be active because we're syncing.
 776                  * We schedule log work now (now that the disk is
 777                  * active) instead of later (when it might not be).
 778                  */
 779                 flush_delayed_work(&mp->m_log->l_work);
 780         }
 781
 782         return 0;
 783 }
 784
 785 STATIC int
 786 xfs_fs_statfs(
 787         struct dentry           *dentry,
 788         struct kstatfs          *statp)
 789 {
 790         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 791         xfs_sb_t                *sbp = &mp->m_sb;
 792         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 793         uint64_t                fakeinos, id;
 794         uint64_t                icount;
 795         uint64_t                ifree;
 796         uint64_t                fdblocks;
 797         xfs_extlen_t            lsize;
 798         int64_t                 ffree;
 799
 800         statp->f_type = XFS_SUPER_MAGIC;
 801         statp->f_namelen = MAXNAMELEN - 1;
 802
 803         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 804         statp->f_fsid = u64_to_fsid(id);
 805
 806         icount = percpu_counter_sum(&mp->m_icount);
 807         ifree = percpu_counter_sum(&mp->m_ifree);
 808         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 809
 810         spin_lock(&mp->m_sb_lock);
 811         statp->f_bsize = sbp->sb_blocksize;
 812         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 813         statp->f_blocks = sbp->sb_dblocks - lsize;
 814         spin_unlock(&mp->m_sb_lock);
 815
 816         /* make sure statp->f_bfree does not underflow */
 817         statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
 818         statp->f_bavail = statp->f_bfree;
 819
 820         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 821         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 822         if (M_IGEO(mp)->maxicount)
 823                 statp->f_files = min_t(typeof(statp->f_files),
 824                                         statp->f_files,
 825                                         M_IGEO(mp)->maxicount);
 826
 827         /* If sb_icount overshot maxicount, report actual allocation */
 828         statp->f_files = max_t(typeof(statp->f_files),
 829                                         statp->f_files,
 830                                         sbp->sb_icount);
 831
 832         /* make sure statp->f_ffree does not underflow */
 833         ffree = statp->f_files - (icount - ifree);
 834         statp->f_ffree = max_t(int64_t, ffree, 0);
 835
 836
 837         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 838             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 839                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 840                 xfs_qm_statvfs(ip, statp);
 841
 842         if (XFS_IS_REALTIME_MOUNT(mp) &&
 843             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 844                 statp->f_blocks = sbp->sb_rblocks;
 845                 statp->f_bavail = statp->f_bfree =
 846                         sbp->sb_frextents * sbp->sb_rextsize;
 847         }
 848
 849         return 0;
 850 }
 851
 852 STATIC void
 853 xfs_save_resvblks(struct xfs_mount *mp)
 854 {
 855         uint64_t resblks = 0;
 856
 857         mp->m_resblks_save = mp->m_resblks;
 858         xfs_reserve_blocks(mp, &resblks, NULL);
 859 }
 860
 861 STATIC void
 862 xfs_restore_resvblks(struct xfs_mount *mp)
 863 {
 864         uint64_t resblks;
 865
 866         if (mp->m_resblks_save) {
 867                 resblks = mp->m_resblks_save;
 868                 mp->m_resblks_save = 0;
 869         } else
 870                 resblks = xfs_default_resblks(mp);
 871
 872         xfs_reserve_blocks(mp, &resblks, NULL);
 873 }
 874
 875 /*
 876  * Second stage of a freeze. The data is already frozen so we only
 877  * need to take care of the metadata. Once that's done sync the superblock
 878  * to the log to dirty it in case of a crash while frozen. This ensures that we
 879  * will recover the unlinked inode lists on the next mount.
 880  */
 881 STATIC int
 882 xfs_fs_freeze(
 883         struct super_block      *sb)
 884 {
 885         struct xfs_mount        *mp = XFS_M(sb);
 886         unsigned int            flags;
 887         int                     ret;
 888
 889         /*
 890          * The filesystem is now frozen far enough that memory reclaim
 891          * cannot safely operate on the filesystem. Hence we need to
 892          * set a GFP_NOFS context here to avoid recursion deadlocks.
 893          */
 894         flags = memalloc_nofs_save();
 895         xfs_blockgc_stop(mp);
 896         xfs_save_resvblks(mp);
 897         ret = xfs_log_quiesce(mp);
 898         memalloc_nofs_restore(flags);
 899         return ret;
 900 }
 901
 902 STATIC int
 903 xfs_fs_unfreeze(
 904         struct super_block      *sb)
 905 {
 906         struct xfs_mount        *mp = XFS_M(sb);
 907
 908         xfs_restore_resvblks(mp);
 909         xfs_log_work_queue(mp);
 910         xfs_blockgc_start(mp);
 911         return 0;
 912 }
 913
 914 /*
 915  * This function fills in xfs_mount_t fields based on mount args.
 916  * Note: the superblock _has_ now been read in.
 917  */
 918 STATIC int
 919 xfs_finish_flags(
 920         struct xfs_mount        *mp)
 921 {
 922         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
 923
 924         /* Fail a mount where the logbuf is smaller than the log stripe */
 925         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
 926                 if (mp->m_logbsize <= 0 &&
 927                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 928                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 929                 } else if (mp->m_logbsize > 0 &&
 930                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 931                         xfs_warn(mp,
 932                 "logbuf size must be greater than or equal to log stripe size");
 933                         return -EINVAL;
 934                 }
 935         } else {
 936                 /* Fail a mount if the logbuf is larger than 32K */
 937                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 938                         xfs_warn(mp,
 939                 "logbuf size for version 1 logs must be 16K or 32K");
 940                         return -EINVAL;
 941                 }
 942         }
 943
 944         /*
 945          * V5 filesystems always use attr2 format for attributes.
 946          */
 947         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 948             (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 949                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 950                              "attr2 is always enabled for V5 filesystems.");
 951                 return -EINVAL;
 952         }
 953
 954         /*
 955          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
 956          * told by noattr2 to turn it off
 957          */
 958         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 959             !(mp->m_flags & XFS_MOUNT_NOATTR2))
 960                 mp->m_flags |= XFS_MOUNT_ATTR2;
 961
 962         /*
 963          * prohibit r/w mounts of read-only filesystems
 964          */
 965         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 966                 xfs_warn(mp,
 967                         "cannot mount a read-only filesystem as read-write");
 968                 return -EROFS;
 969         }
 970
 971         if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
 972             (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
 973             !xfs_sb_version_has_pquotino(&mp->m_sb)) {
 974                 xfs_warn(mp,
 975                   "Super block does not support project and group quota together");
 976                 return -EINVAL;
 977         }
 978
 979         return 0;
 980 }
 981
 982 static int
 983 xfs_init_percpu_counters(
 984         struct xfs_mount        *mp)
 985 {
 986         int             error;
 987
 988         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
 989         if (error)
 990                 return -ENOMEM;
 991
 992         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
 993         if (error)
 994                 goto free_icount;
 995
 996         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
 997         if (error)
 998                 goto free_ifree;
 999
1000         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1001         if (error)
1002                 goto free_fdblocks;
1003
1004         return 0;
1005
1006 free_fdblocks:
1007         percpu_counter_destroy(&mp->m_fdblocks);
1008 free_ifree:
1009         percpu_counter_destroy(&mp->m_ifree);
1010 free_icount:
1011         percpu_counter_destroy(&mp->m_icount);
1012         return -ENOMEM;
1013 }
1014
1015 void
1016 xfs_reinit_percpu_counters(
1017         struct xfs_mount        *mp)
1018 {
1019         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1020         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1021         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1022 }
1023
1024 static void
1025 xfs_destroy_percpu_counters(
1026         struct xfs_mount        *mp)
1027 {
1028         percpu_counter_destroy(&mp->m_icount);
1029         percpu_counter_destroy(&mp->m_ifree);
1030         percpu_counter_destroy(&mp->m_fdblocks);
1031         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1032                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1033         percpu_counter_destroy(&mp->m_delalloc_blks);
1034 }
1035
1036 static void
1037 xfs_fs_put_super(
1038         struct super_block      *sb)
1039 {
1040         struct xfs_mount        *mp = XFS_M(sb);
1041
1042         /* if ->fill_super failed, we have no mount to tear down */
1043         if (!sb->s_fs_info)
1044                 return;
1045
1046         xfs_notice(mp, "Unmounting Filesystem");
1047         xfs_filestream_unmount(mp);
1048         xfs_unmountfs(mp);
1049
1050         xfs_freesb(mp);
1051         free_percpu(mp->m_stats.xs_stats);
1052         xfs_destroy_percpu_counters(mp);
1053         xfs_destroy_mount_workqueues(mp);
1054         xfs_close_devices(mp);
1055
1056         sb->s_fs_info = NULL;
1057         xfs_mount_free(mp);
1058 }
1059
1060 static long
1061 xfs_fs_nr_cached_objects(
1062         struct super_block      *sb,
1063         struct shrink_control   *sc)
1064 {
1065         /* Paranoia: catch incorrect calls during mount setup or teardown */
1066         if (WARN_ON_ONCE(!sb->s_fs_info))
1067                 return 0;
1068         return xfs_reclaim_inodes_count(XFS_M(sb));
1069 }
1070
1071 static long
1072 xfs_fs_free_cached_objects(
1073         struct super_block      *sb,
1074         struct shrink_control   *sc)
1075 {
1076         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1077 }
1078
1079 static const struct super_operations xfs_super_operations = {
1080         .alloc_inode            = xfs_fs_alloc_inode,
1081         .destroy_inode          = xfs_fs_destroy_inode,
1082         .dirty_inode            = xfs_fs_dirty_inode,
1083         .drop_inode             = xfs_fs_drop_inode,
1084         .put_super              = xfs_fs_put_super,
1085         .sync_fs                = xfs_fs_sync_fs,
1086         .freeze_fs              = xfs_fs_freeze,
1087         .unfreeze_fs            = xfs_fs_unfreeze,
1088         .statfs                 = xfs_fs_statfs,
1089         .show_options           = xfs_fs_show_options,
1090         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1091         .free_cached_objects    = xfs_fs_free_cached_objects,
1092 };
1093
1094 static int
1095 suffix_kstrtoint(
1096         const char      *s,
1097         unsigned int    base,
1098         int             *res)
1099 {
1100         int             last, shift_left_factor = 0, _res;
1101         char            *value;
1102         int             ret = 0;
1103
1104         value = kstrdup(s, GFP_KERNEL);
1105         if (!value)
1106                 return -ENOMEM;
1107
1108         last = strlen(value) - 1;
1109         if (value[last] == 'K' || value[last] == 'k') {
1110                 shift_left_factor = 10;
1111                 value[last] = '\0';
1112         }
1113         if (value[last] == 'M' || value[last] == 'm') {
1114                 shift_left_factor = 20;
1115                 value[last] = '\0';
1116         }
1117         if (value[last] == 'G' || value[last] == 'g') {
1118                 shift_left_factor = 30;
1119                 value[last] = '\0';
1120         }
1121
1122         if (kstrtoint(value, base, &_res))
1123                 ret = -EINVAL;
1124         kfree(value);
1125         *res = _res << shift_left_factor;
1126         return ret;
1127 }
1128
1129 static inline void
1130 xfs_fs_warn_deprecated(
1131         struct fs_context       *fc,
1132         struct fs_parameter     *param,
1133         uint64_t                flag,
1134         bool                    value)
1135 {
1136         /* Don't print the warning if reconfiguring and current mount point
1137          * already had the flag set
1138          */
1139         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1140                         !!(XFS_M(fc->root->d_sb)->m_flags & flag) == value)
1141                 return;
1142         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1143 }
1144
1145 /*
1146  * Set mount state from a mount option.
1147  *
1148  * NOTE: mp->m_super is NULL here!
1149  */
1150 static int
1151 xfs_fs_parse_param(
1152         struct fs_context       *fc,
1153         struct fs_parameter     *param)
1154 {
1155         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1156         struct fs_parse_result  result;
1157         int                     size = 0;
1158         int                     opt;
1159
1160         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1161         if (opt < 0)
1162                 return opt;
1163
1164         switch (opt) {
1165         case Opt_logbufs:
1166                 parsing_mp->m_logbufs = result.uint_32;
1167                 return 0;
1168         case Opt_logbsize:
1169                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1170                         return -EINVAL;
1171                 return 0;
1172         case Opt_logdev:
1173                 kfree(parsing_mp->m_logname);
1174                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1175                 if (!parsing_mp->m_logname)
1176                         return -ENOMEM;
1177                 return 0;
1178         case Opt_rtdev:
1179                 kfree(parsing_mp->m_rtname);
1180                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1181                 if (!parsing_mp->m_rtname)
1182                         return -ENOMEM;
1183                 return 0;
1184         case Opt_allocsize:
1185                 if (suffix_kstrtoint(param->string, 10, &size))
1186                         return -EINVAL;
1187                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1188                 parsing_mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1189                 return 0;
1190         case Opt_grpid:
1191         case Opt_bsdgroups:
1192                 parsing_mp->m_flags |= XFS_MOUNT_GRPID;
1193                 return 0;
1194         case Opt_nogrpid:
1195         case Opt_sysvgroups:
1196                 parsing_mp->m_flags &= ~XFS_MOUNT_GRPID;
1197                 return 0;
1198         case Opt_wsync:
1199                 parsing_mp->m_flags |= XFS_MOUNT_WSYNC;
1200                 return 0;
1201         case Opt_norecovery:
1202                 parsing_mp->m_flags |= XFS_MOUNT_NORECOVERY;
1203                 return 0;
1204         case Opt_noalign:
1205                 parsing_mp->m_flags |= XFS_MOUNT_NOALIGN;
1206                 return 0;
1207         case Opt_swalloc:
1208                 parsing_mp->m_flags |= XFS_MOUNT_SWALLOC;
1209                 return 0;
1210         case Opt_sunit:
1211                 parsing_mp->m_dalign = result.uint_32;
1212                 return 0;
1213         case Opt_swidth:
1214                 parsing_mp->m_swidth = result.uint_32;
1215                 return 0;
1216         case Opt_inode32:
1217                 parsing_mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1218                 return 0;
1219         case Opt_inode64:
1220                 parsing_mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1221                 return 0;
1222         case Opt_nouuid:
1223                 parsing_mp->m_flags |= XFS_MOUNT_NOUUID;
1224                 return 0;
1225         case Opt_largeio:
1226                 parsing_mp->m_flags |= XFS_MOUNT_LARGEIO;
1227                 return 0;
1228         case Opt_nolargeio:
1229                 parsing_mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1230                 return 0;
1231         case Opt_filestreams:
1232                 parsing_mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1233                 return 0;
1234         case Opt_noquota:
1235                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1236                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1237                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1238                 return 0;
1239         case Opt_quota:
1240         case Opt_uquota:
1241         case Opt_usrquota:
1242                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1243                                  XFS_UQUOTA_ENFD);
1244                 return 0;
1245         case Opt_qnoenforce:
1246         case Opt_uqnoenforce:
1247                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1248                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1249                 return 0;
1250         case Opt_pquota:
1251         case Opt_prjquota:
1252                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1253                                  XFS_PQUOTA_ENFD);
1254                 return 0;
1255         case Opt_pqnoenforce:
1256                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1257                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1258                 return 0;
1259         case Opt_gquota:
1260         case Opt_grpquota:
1261                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1262                                  XFS_GQUOTA_ENFD);
1263                 return 0;
1264         case Opt_gqnoenforce:
1265                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1266                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1267                 return 0;
1268         case Opt_discard:
1269                 parsing_mp->m_flags |= XFS_MOUNT_DISCARD;
1270                 return 0;
1271         case Opt_nodiscard:
1272                 parsing_mp->m_flags &= ~XFS_MOUNT_DISCARD;
1273                 return 0;
1274 #ifdef CONFIG_FS_DAX
1275         case Opt_dax:
1276                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1277                 return 0;
1278         case Opt_dax_enum:
1279                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1280                 return 0;
1281 #endif
1282         /* Following mount options will be removed in September 2025 */
1283         case Opt_ikeep:
1284                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, true);
1285                 parsing_mp->m_flags |= XFS_MOUNT_IKEEP;
1286                 return 0;
1287         case Opt_noikeep:
1288                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, false);
1289                 parsing_mp->m_flags &= ~XFS_MOUNT_IKEEP;
1290                 return 0;
1291         case Opt_attr2:
1292                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_ATTR2, true);
1293                 parsing_mp->m_flags |= XFS_MOUNT_ATTR2;
1294                 return 0;
1295         case Opt_noattr2:
1296                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_NOATTR2, true);
1297                 parsing_mp->m_flags &= ~XFS_MOUNT_ATTR2;
1298                 parsing_mp->m_flags |= XFS_MOUNT_NOATTR2;
1299                 return 0;
1300         default:
1301                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1302                 return -EINVAL;
1303         }
1304
1305         return 0;
1306 }
1307
1308 static int
1309 xfs_fs_validate_params(
1310         struct xfs_mount        *mp)
1311 {
1312         /*
1313          * no recovery flag requires a read-only mount
1314          */
1315         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1316             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1317                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1318                 return -EINVAL;
1319         }
1320
1321         if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1322             (mp->m_dalign || mp->m_swidth)) {
1323                 xfs_warn(mp,
1324         "sunit and swidth options incompatible with the noalign option");
1325                 return -EINVAL;
1326         }
1327
1328         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1329                 xfs_warn(mp, "quota support not available in this kernel.");
1330                 return -EINVAL;
1331         }
1332
1333         if ((mp->m_dalign && !mp->m_swidth) ||
1334             (!mp->m_dalign && mp->m_swidth)) {
1335                 xfs_warn(mp, "sunit and swidth must be specified together");
1336                 return -EINVAL;
1337         }
1338
1339         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1340                 xfs_warn(mp,
1341         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1342                         mp->m_swidth, mp->m_dalign);
1343                 return -EINVAL;
1344         }
1345
1346         if (mp->m_logbufs != -1 &&
1347             mp->m_logbufs != 0 &&
1348             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1349              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1350                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1351                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1352                 return -EINVAL;
1353         }
1354
1355         if (mp->m_logbsize != -1 &&
1356             mp->m_logbsize !=  0 &&
1357             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1358              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1359              !is_power_of_2(mp->m_logbsize))) {
1360                 xfs_warn(mp,
1361                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1362                         mp->m_logbsize);
1363                 return -EINVAL;
1364         }
1365
1366         if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1367             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1368              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1369                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1370                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1371                 return -EINVAL;
1372         }
1373
1374         return 0;
1375 }
1376
1377 static int
1378 xfs_fs_fill_super(
1379         struct super_block      *sb,
1380         struct fs_context       *fc)
1381 {
1382         struct xfs_mount        *mp = sb->s_fs_info;
1383         struct inode            *root;
1384         int                     flags = 0, error;
1385
1386         mp->m_super = sb;
1387
1388         error = xfs_fs_validate_params(mp);
1389         if (error)
1390                 goto out_free_names;
1391
1392         sb_min_blocksize(sb, BBSIZE);
1393         sb->s_xattr = xfs_xattr_handlers;
1394         sb->s_export_op = &xfs_export_operations;
1395 #ifdef CONFIG_XFS_QUOTA
1396         sb->s_qcop = &xfs_quotactl_operations;
1397         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1398 #endif
1399         sb->s_op = &xfs_super_operations;
1400
1401         /*
1402          * Delay mount work if the debug hook is set. This is debug
1403          * instrumention to coordinate simulation of xfs mount failures with
1404          * VFS superblock operations
1405          */
1406         if (xfs_globals.mount_delay) {
1407                 xfs_notice(mp, "Delaying mount for %d seconds.",
1408                         xfs_globals.mount_delay);
1409                 msleep(xfs_globals.mount_delay * 1000);
1410         }
1411
1412         if (fc->sb_flags & SB_SILENT)
1413                 flags |= XFS_MFSI_QUIET;
1414
1415         error = xfs_open_devices(mp);
1416         if (error)
1417                 goto out_free_names;
1418
1419         error = xfs_init_mount_workqueues(mp);
1420         if (error)
1421                 goto out_close_devices;
1422
1423         error = xfs_init_percpu_counters(mp);
1424         if (error)
1425                 goto out_destroy_workqueues;
1426
1427         /* Allocate stats memory before we do operations that might use it */
1428         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1429         if (!mp->m_stats.xs_stats) {
1430                 error = -ENOMEM;
1431                 goto out_destroy_counters;
1432         }
1433
1434         error = xfs_readsb(mp, flags);
1435         if (error)
1436                 goto out_free_stats;
1437
1438         error = xfs_finish_flags(mp);
1439         if (error)
1440                 goto out_free_sb;
1441
1442         error = xfs_setup_devices(mp);
1443         if (error)
1444                 goto out_free_sb;
1445
1446         /* V4 support is undergoing deprecation. */
1447         if (!xfs_sb_version_hascrc(&mp->m_sb)) {
1448 #ifdef CONFIG_XFS_SUPPORT_V4
1449                 xfs_warn_once(mp,
1450         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1451 #else
1452                 xfs_warn(mp,
1453         "Deprecated V4 format (crc=0) not supported by kernel.");
1454                 error = -EINVAL;
1455                 goto out_free_sb;
1456 #endif
1457         }
1458
1459         /* Filesystem claims it needs repair, so refuse the mount. */
1460         if (xfs_sb_version_needsrepair(&mp->m_sb)) {
1461                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1462                 error = -EFSCORRUPTED;
1463                 goto out_free_sb;
1464         }
1465
1466         /*
1467          * Don't touch the filesystem if a user tool thinks it owns the primary
1468          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1469          * we don't check them at all.
1470          */
1471         if (mp->m_sb.sb_inprogress) {
1472                 xfs_warn(mp, "Offline file system operation in progress!");
1473                 error = -EFSCORRUPTED;
1474                 goto out_free_sb;
1475         }
1476
1477         /*
1478          * Until this is fixed only page-sized or smaller data blocks work.
1479          */
1480         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1481                 xfs_warn(mp,
1482                 "File system with blocksize %d bytes. "
1483                 "Only pagesize (%ld) or less will currently work.",
1484                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1485                 error = -ENOSYS;
1486                 goto out_free_sb;
1487         }
1488
1489         /* Ensure this filesystem fits in the page cache limits */
1490         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1491             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1492                 xfs_warn(mp,
1493                 "file system too large to be mounted on this system.");
1494                 error = -EFBIG;
1495                 goto out_free_sb;
1496         }
1497
1498         /*
1499          * XFS block mappings use 54 bits to store the logical block offset.
1500          * This should suffice to handle the maximum file size that the VFS
1501          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1502          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1503          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1504          * to check this assertion.
1505          *
1506          * Avoid integer overflow by comparing the maximum bmbt offset to the
1507          * maximum pagecache offset in units of fs blocks.
1508          */
1509         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1510                 xfs_warn(mp,
1511 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1512                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1513                          XFS_MAX_FILEOFF);
1514                 error = -EINVAL;
1515                 goto out_free_sb;
1516         }
1517
1518         error = xfs_filestream_mount(mp);
1519         if (error)
1520                 goto out_free_sb;
1521
1522         /*
1523          * we must configure the block size in the superblock before we run the
1524          * full mount process as the mount process can lookup and cache inodes.
1525          */
1526         sb->s_magic = XFS_SUPER_MAGIC;
1527         sb->s_blocksize = mp->m_sb.sb_blocksize;
1528         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1529         sb->s_maxbytes = MAX_LFS_FILESIZE;
1530         sb->s_max_links = XFS_MAXLINK;
1531         sb->s_time_gran = 1;
1532         if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
1533                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1534                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1535         } else {
1536                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1537                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1538         }
1539         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1540         sb->s_iflags |= SB_I_CGROUPWB;
1541
1542         set_posix_acl_flag(sb);
1543
1544         /* version 5 superblocks support inode version counters. */
1545         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1546                 sb->s_flags |= SB_I_VERSION;
1547
1548         if (xfs_sb_version_hasbigtime(&mp->m_sb))
1549                 xfs_warn(mp,
1550  "EXPERIMENTAL big timestamp feature in use. Use at your own risk!");
1551
1552         if (mp->m_flags & XFS_MOUNT_DAX_ALWAYS) {
1553                 bool rtdev_is_dax = false, datadev_is_dax;
1554
1555                 xfs_warn(mp,
1556                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1557
1558                 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1559                         sb->s_blocksize);
1560                 if (mp->m_rtdev_targp)
1561                         rtdev_is_dax = bdev_dax_supported(
1562                                 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1563                 if (!rtdev_is_dax && !datadev_is_dax) {
1564                         xfs_alert(mp,
1565                         "DAX unsupported by block device. Turning off DAX.");
1566                         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1567                 }
1568                 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1569                         xfs_alert(mp,
1570                 "DAX and reflink cannot be used together!");
1571                         error = -EINVAL;
1572                         goto out_filestream_unmount;
1573                 }
1574         }
1575
1576         if (mp->m_flags & XFS_MOUNT_DISCARD) {
1577                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1578
1579                 if (!blk_queue_discard(q)) {
1580                         xfs_warn(mp, "mounting with \"discard\" option, but "
1581                                         "the device does not support discard");
1582                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
1583                 }
1584         }
1585
1586         if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1587                 if (mp->m_sb.sb_rblocks) {
1588                         xfs_alert(mp,
1589         "reflink not compatible with realtime device!");
1590                         error = -EINVAL;
1591                         goto out_filestream_unmount;
1592                 }
1593
1594                 if (xfs_globals.always_cow) {
1595                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1596                         mp->m_always_cow = true;
1597                 }
1598         }
1599
1600         if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1601                 xfs_alert(mp,
1602         "reverse mapping btree not compatible with realtime device!");
1603                 error = -EINVAL;
1604                 goto out_filestream_unmount;
1605         }
1606
1607         if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
1608                 xfs_warn(mp,
1609  "EXPERIMENTAL inode btree counters feature in use. Use at your own risk!");
1610
1611         error = xfs_mountfs(mp);
1612         if (error)
1613                 goto out_filestream_unmount;
1614
1615         root = igrab(VFS_I(mp->m_rootip));
1616         if (!root) {
1617                 error = -ENOENT;
1618                 goto out_unmount;
1619         }
1620         sb->s_root = d_make_root(root);
1621         if (!sb->s_root) {
1622                 error = -ENOMEM;
1623                 goto out_unmount;
1624         }
1625
1626         return 0;
1627
1628  out_filestream_unmount:
1629         xfs_filestream_unmount(mp);
1630  out_free_sb:
1631         xfs_freesb(mp);
1632  out_free_stats:
1633         free_percpu(mp->m_stats.xs_stats);
1634  out_destroy_counters:
1635         xfs_destroy_percpu_counters(mp);
1636  out_destroy_workqueues:
1637         xfs_destroy_mount_workqueues(mp);
1638  out_close_devices:
1639         xfs_close_devices(mp);
1640  out_free_names:
1641         sb->s_fs_info = NULL;
1642         xfs_mount_free(mp);
1643         return error;
1644
1645  out_unmount:
1646         xfs_filestream_unmount(mp);
1647         xfs_unmountfs(mp);
1648         goto out_free_sb;
1649 }
1650
1651 static int
1652 xfs_fs_get_tree(
1653         struct fs_context       *fc)
1654 {
1655         return get_tree_bdev(fc, xfs_fs_fill_super);
1656 }
1657
1658 static int
1659 xfs_remount_rw(
1660         struct xfs_mount        *mp)
1661 {
1662         struct xfs_sb           *sbp = &mp->m_sb;
1663         int error;
1664
1665         if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1666                 xfs_warn(mp,
1667                         "ro->rw transition prohibited on norecovery mount");
1668                 return -EINVAL;
1669         }
1670
1671         if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1672             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1673                 xfs_warn(mp,
1674         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1675                         (sbp->sb_features_ro_compat &
1676                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1677                 return -EINVAL;
1678         }
1679
1680         mp->m_flags &= ~XFS_MOUNT_RDONLY;
1681
1682         /*
1683          * If this is the first remount to writeable state we might have some
1684          * superblock changes to update.
1685          */
1686         if (mp->m_update_sb) {
1687                 error = xfs_sync_sb(mp, false);
1688                 if (error) {
1689                         xfs_warn(mp, "failed to write sb changes");
1690                         return error;
1691                 }
1692                 mp->m_update_sb = false;
1693         }
1694
1695         /*
1696          * Fill out the reserve pool if it is empty. Use the stashed value if
1697          * it is non-zero, otherwise go with the default.
1698          */
1699         xfs_restore_resvblks(mp);
1700         xfs_log_work_queue(mp);
1701
1702         /* Recover any CoW blocks that never got remapped. */
1703         error = xfs_reflink_recover_cow(mp);
1704         if (error) {
1705                 xfs_err(mp,
1706                         "Error %d recovering leftover CoW allocations.", error);
1707                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1708                 return error;
1709         }
1710         xfs_blockgc_start(mp);
1711
1712         /* Create the per-AG metadata reservation pool .*/
1713         error = xfs_fs_reserve_ag_blocks(mp);
1714         if (error && error != -ENOSPC)
1715                 return error;
1716
1717         return 0;
1718 }
1719
1720 static int
1721 xfs_remount_ro(
1722         struct xfs_mount        *mp)
1723 {
1724         int error;
1725
1726         /*
1727          * Cancel background eofb scanning so it cannot race with the final
1728          * log force+buftarg wait and deadlock the remount.
1729          */
1730         xfs_blockgc_stop(mp);
1731
1732         /* Get rid of any leftover CoW reservations... */
1733         error = xfs_blockgc_free_space(mp, NULL);
1734         if (error) {
1735                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1736                 return error;
1737         }
1738
1739         /* Free the per-AG metadata reservation pool. */
1740         error = xfs_fs_unreserve_ag_blocks(mp);
1741         if (error) {
1742                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1743                 return error;
1744         }
1745
1746         /*
1747          * Before we sync the metadata, we need to free up the reserve block
1748          * pool so that the used block count in the superblock on disk is
1749          * correct at the end of the remount. Stash the current* reserve pool
1750          * size so that if we get remounted rw, we can return it to the same
1751          * size.
1752          */
1753         xfs_save_resvblks(mp);
1754
1755         xfs_log_clean(mp);
1756         mp->m_flags |= XFS_MOUNT_RDONLY;
1757
1758         return 0;
1759 }
1760
1761 /*
1762  * Logically we would return an error here to prevent users from believing
1763  * they might have changed mount options using remount which can't be changed.
1764  *
1765  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1766  * arguments in some cases so we can't blindly reject options, but have to
1767  * check for each specified option if it actually differs from the currently
1768  * set option and only reject it if that's the case.
1769  *
1770  * Until that is implemented we return success for every remount request, and
1771  * silently ignore all options that we can't actually change.
1772  */
1773 static int
1774 xfs_fs_reconfigure(
1775         struct fs_context *fc)
1776 {
1777         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1778         struct xfs_mount        *new_mp = fc->s_fs_info;
1779         xfs_sb_t                *sbp = &mp->m_sb;
1780         int                     flags = fc->sb_flags;
1781         int                     error;
1782
1783         /* version 5 superblocks always support version counters. */
1784         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1785                 fc->sb_flags |= SB_I_VERSION;
1786
1787         error = xfs_fs_validate_params(new_mp);
1788         if (error)
1789                 return error;
1790
1791         sync_filesystem(mp->m_super);
1792
1793         /* inode32 -> inode64 */
1794         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1795             !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1796                 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1797                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1798         }
1799
1800         /* inode64 -> inode32 */
1801         if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1802             (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1803                 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1804                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1805         }
1806
1807         /* ro -> rw */
1808         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1809                 error = xfs_remount_rw(mp);
1810                 if (error)
1811                         return error;
1812         }
1813
1814         /* rw -> ro */
1815         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1816                 error = xfs_remount_ro(mp);
1817                 if (error)
1818                         return error;
1819         }
1820
1821         return 0;
1822 }
1823
1824 static void xfs_fs_free(
1825         struct fs_context       *fc)
1826 {
1827         struct xfs_mount        *mp = fc->s_fs_info;
1828
1829         /*
1830          * mp is stored in the fs_context when it is initialized.
1831          * mp is transferred to the superblock on a successful mount,
1832          * but if an error occurs before the transfer we have to free
1833          * it here.
1834          */
1835         if (mp)
1836                 xfs_mount_free(mp);
1837 }
1838
1839 static const struct fs_context_operations xfs_context_ops = {
1840         .parse_param = xfs_fs_parse_param,
1841         .get_tree    = xfs_fs_get_tree,
1842         .reconfigure = xfs_fs_reconfigure,
1843         .free        = xfs_fs_free,
1844 };
1845
1846 static int xfs_init_fs_context(
1847         struct fs_context       *fc)
1848 {
1849         struct xfs_mount        *mp;
1850
1851         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1852         if (!mp)
1853                 return -ENOMEM;
1854
1855         spin_lock_init(&mp->m_sb_lock);
1856         spin_lock_init(&mp->m_agirotor_lock);
1857         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1858         spin_lock_init(&mp->m_perag_lock);
1859         mutex_init(&mp->m_growlock);
1860         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1861         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1862         mp->m_kobj.kobject.kset = xfs_kset;
1863         /*
1864          * We don't create the finobt per-ag space reservation until after log
1865          * recovery, so we must set this to true so that an ifree transaction
1866          * started during log recovery will not depend on space reservations
1867          * for finobt expansion.
1868          */
1869         mp->m_finobt_nores = true;
1870
1871         /*
1872          * These can be overridden by the mount option parsing.
1873          */
1874         mp->m_logbufs = -1;
1875         mp->m_logbsize = -1;
1876         mp->m_allocsize_log = 16; /* 64k */
1877
1878         /*
1879          * Copy binary VFS mount flags we are interested in.
1880          */
1881         if (fc->sb_flags & SB_RDONLY)
1882                 mp->m_flags |= XFS_MOUNT_RDONLY;
1883         if (fc->sb_flags & SB_DIRSYNC)
1884                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1885         if (fc->sb_flags & SB_SYNCHRONOUS)
1886                 mp->m_flags |= XFS_MOUNT_WSYNC;
1887
1888         fc->s_fs_info = mp;
1889         fc->ops = &xfs_context_ops;
1890
1891         return 0;
1892 }
1893
1894 static struct file_system_type xfs_fs_type = {
1895         .owner                  = THIS_MODULE,
1896         .name                   = "xfs",
1897         .init_fs_context        = xfs_init_fs_context,
1898         .parameters             = xfs_fs_parameters,
1899         .kill_sb                = kill_block_super,
1900         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1901 };
1902 MODULE_ALIAS_FS("xfs");
1903
1904 STATIC int __init
1905 xfs_init_zones(void)
1906 {
1907         xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1908                                                 sizeof(struct xlog_ticket),
1909                                                 0, 0, NULL);
1910         if (!xfs_log_ticket_zone)
1911                 goto out;
1912
1913         xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1914                                         sizeof(struct xfs_extent_free_item),
1915                                         0, 0, NULL);
1916         if (!xfs_bmap_free_item_zone)
1917                 goto out_destroy_log_ticket_zone;
1918
1919         xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1920                                                sizeof(struct xfs_btree_cur),
1921                                                0, 0, NULL);
1922         if (!xfs_btree_cur_zone)
1923                 goto out_destroy_bmap_free_item_zone;
1924
1925         xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1926                                               sizeof(struct xfs_da_state),
1927                                               0, 0, NULL);
1928         if (!xfs_da_state_zone)
1929                 goto out_destroy_btree_cur_zone;
1930
1931         xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1932                                            sizeof(struct xfs_ifork),
1933                                            0, 0, NULL);
1934         if (!xfs_ifork_zone)
1935                 goto out_destroy_da_state_zone;
1936
1937         xfs_trans_zone = kmem_cache_create("xfs_trans",
1938                                            sizeof(struct xfs_trans),
1939                                            0, 0, NULL);
1940         if (!xfs_trans_zone)
1941                 goto out_destroy_ifork_zone;
1942
1943
1944         /*
1945          * The size of the zone allocated buf log item is the maximum
1946          * size possible under XFS.  This wastes a little bit of memory,
1947          * but it is much faster.
1948          */
1949         xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1950                                               sizeof(struct xfs_buf_log_item),
1951                                               0, 0, NULL);
1952         if (!xfs_buf_item_zone)
1953                 goto out_destroy_trans_zone;
1954
1955         xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1956                                         (sizeof(struct xfs_efd_log_item) +
1957                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1958                                         sizeof(struct xfs_extent)),
1959                                         0, 0, NULL);
1960         if (!xfs_efd_zone)
1961                 goto out_destroy_buf_item_zone;
1962
1963         xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1964                                          (sizeof(struct xfs_efi_log_item) +
1965                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1966                                          sizeof(struct xfs_extent)),
1967                                          0, 0, NULL);
1968         if (!xfs_efi_zone)
1969                 goto out_destroy_efd_zone;
1970
1971         xfs_inode_zone = kmem_cache_create("xfs_inode",
1972                                            sizeof(struct xfs_inode), 0,
1973                                            (SLAB_HWCACHE_ALIGN |
1974                                             SLAB_RECLAIM_ACCOUNT |
1975                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1976                                            xfs_fs_inode_init_once);
1977         if (!xfs_inode_zone)
1978                 goto out_destroy_efi_zone;
1979
1980         xfs_ili_zone = kmem_cache_create("xfs_ili",
1981                                          sizeof(struct xfs_inode_log_item), 0,
1982                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1983                                          NULL);
1984         if (!xfs_ili_zone)
1985                 goto out_destroy_inode_zone;
1986
1987         xfs_icreate_zone = kmem_cache_create("xfs_icr",
1988                                              sizeof(struct xfs_icreate_item),
1989                                              0, 0, NULL);
1990         if (!xfs_icreate_zone)
1991                 goto out_destroy_ili_zone;
1992
1993         xfs_rud_zone = kmem_cache_create("xfs_rud_item",
1994                                          sizeof(struct xfs_rud_log_item),
1995                                          0, 0, NULL);
1996         if (!xfs_rud_zone)
1997                 goto out_destroy_icreate_zone;
1998
1999         xfs_rui_zone = kmem_cache_create("xfs_rui_item",
2000                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2001                         0, 0, NULL);
2002         if (!xfs_rui_zone)
2003                 goto out_destroy_rud_zone;
2004
2005         xfs_cud_zone = kmem_cache_create("xfs_cud_item",
2006                                          sizeof(struct xfs_cud_log_item),
2007                                          0, 0, NULL);
2008         if (!xfs_cud_zone)
2009                 goto out_destroy_rui_zone;
2010
2011         xfs_cui_zone = kmem_cache_create("xfs_cui_item",
2012                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2013                         0, 0, NULL);
2014         if (!xfs_cui_zone)
2015                 goto out_destroy_cud_zone;
2016
2017         xfs_bud_zone = kmem_cache_create("xfs_bud_item",
2018                                          sizeof(struct xfs_bud_log_item),
2019                                          0, 0, NULL);
2020         if (!xfs_bud_zone)
2021                 goto out_destroy_cui_zone;
2022
2023         xfs_bui_zone = kmem_cache_create("xfs_bui_item",
2024                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2025                         0, 0, NULL);
2026         if (!xfs_bui_zone)
2027                 goto out_destroy_bud_zone;
2028
2029         return 0;
2030
2031  out_destroy_bud_zone:
2032         kmem_cache_destroy(xfs_bud_zone);
2033  out_destroy_cui_zone:
2034         kmem_cache_destroy(xfs_cui_zone);
2035  out_destroy_cud_zone:
2036         kmem_cache_destroy(xfs_cud_zone);
2037  out_destroy_rui_zone:
2038         kmem_cache_destroy(xfs_rui_zone);
2039  out_destroy_rud_zone:
2040         kmem_cache_destroy(xfs_rud_zone);
2041  out_destroy_icreate_zone:
2042         kmem_cache_destroy(xfs_icreate_zone);
2043  out_destroy_ili_zone:
2044         kmem_cache_destroy(xfs_ili_zone);
2045  out_destroy_inode_zone:
2046         kmem_cache_destroy(xfs_inode_zone);
2047  out_destroy_efi_zone:
2048         kmem_cache_destroy(xfs_efi_zone);
2049  out_destroy_efd_zone:
2050         kmem_cache_destroy(xfs_efd_zone);
2051  out_destroy_buf_item_zone:
2052         kmem_cache_destroy(xfs_buf_item_zone);
2053  out_destroy_trans_zone:
2054         kmem_cache_destroy(xfs_trans_zone);
2055  out_destroy_ifork_zone:
2056         kmem_cache_destroy(xfs_ifork_zone);
2057  out_destroy_da_state_zone:
2058         kmem_cache_destroy(xfs_da_state_zone);
2059  out_destroy_btree_cur_zone:
2060         kmem_cache_destroy(xfs_btree_cur_zone);
2061  out_destroy_bmap_free_item_zone:
2062         kmem_cache_destroy(xfs_bmap_free_item_zone);
2063  out_destroy_log_ticket_zone:
2064         kmem_cache_destroy(xfs_log_ticket_zone);
2065  out:
2066         return -ENOMEM;
2067 }
2068
2069 STATIC void
2070 xfs_destroy_zones(void)
2071 {
2072         /*
2073          * Make sure all delayed rcu free are flushed before we
2074          * destroy caches.
2075          */
2076         rcu_barrier();
2077         kmem_cache_destroy(xfs_bui_zone);
2078         kmem_cache_destroy(xfs_bud_zone);
2079         kmem_cache_destroy(xfs_cui_zone);
2080         kmem_cache_destroy(xfs_cud_zone);
2081         kmem_cache_destroy(xfs_rui_zone);
2082         kmem_cache_destroy(xfs_rud_zone);
2083         kmem_cache_destroy(xfs_icreate_zone);
2084         kmem_cache_destroy(xfs_ili_zone);
2085         kmem_cache_destroy(xfs_inode_zone);
2086         kmem_cache_destroy(xfs_efi_zone);
2087         kmem_cache_destroy(xfs_efd_zone);
2088         kmem_cache_destroy(xfs_buf_item_zone);
2089         kmem_cache_destroy(xfs_trans_zone);
2090         kmem_cache_destroy(xfs_ifork_zone);
2091         kmem_cache_destroy(xfs_da_state_zone);
2092         kmem_cache_destroy(xfs_btree_cur_zone);
2093         kmem_cache_destroy(xfs_bmap_free_item_zone);
2094         kmem_cache_destroy(xfs_log_ticket_zone);
2095 }
2096
2097 STATIC int __init
2098 xfs_init_workqueues(void)
2099 {
2100         /*
2101          * The allocation workqueue can be used in memory reclaim situations
2102          * (writepage path), and parallelism is only limited by the number of
2103          * AGs in all the filesystems mounted. Hence use the default large
2104          * max_active value for this workqueue.
2105          */
2106         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2107                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2108         if (!xfs_alloc_wq)
2109                 return -ENOMEM;
2110
2111         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2112                         0);
2113         if (!xfs_discard_wq)
2114                 goto out_free_alloc_wq;
2115
2116         return 0;
2117 out_free_alloc_wq:
2118         destroy_workqueue(xfs_alloc_wq);
2119         return -ENOMEM;
2120 }
2121
2122 STATIC void
2123 xfs_destroy_workqueues(void)
2124 {
2125         destroy_workqueue(xfs_discard_wq);
2126         destroy_workqueue(xfs_alloc_wq);
2127 }
2128
2129 STATIC int __init
2130 init_xfs_fs(void)
2131 {
2132         int                     error;
2133
2134         xfs_check_ondisk_structs();
2135
2136         printk(KERN_INFO XFS_VERSION_STRING " with "
2137                          XFS_BUILD_OPTIONS " enabled\n");
2138
2139         xfs_dir_startup();
2140
2141         error = xfs_init_zones();
2142         if (error)
2143                 goto out;
2144
2145         error = xfs_init_workqueues();
2146         if (error)
2147                 goto out_destroy_zones;
2148
2149         error = xfs_mru_cache_init();
2150         if (error)
2151                 goto out_destroy_wq;
2152
2153         error = xfs_buf_init();
2154         if (error)
2155                 goto out_mru_cache_uninit;
2156
2157         error = xfs_init_procfs();
2158         if (error)
2159                 goto out_buf_terminate;
2160
2161         error = xfs_sysctl_register();
2162         if (error)
2163                 goto out_cleanup_procfs;
2164
2165         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2166         if (!xfs_kset) {
2167                 error = -ENOMEM;
2168                 goto out_sysctl_unregister;
2169         }
2170
2171         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2172
2173         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2174         if (!xfsstats.xs_stats) {
2175                 error = -ENOMEM;
2176                 goto out_kset_unregister;
2177         }
2178
2179         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2180                                "stats");
2181         if (error)
2182                 goto out_free_stats;
2183
2184 #ifdef DEBUG
2185         xfs_dbg_kobj.kobject.kset = xfs_kset;
2186         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2187         if (error)
2188                 goto out_remove_stats_kobj;
2189 #endif
2190
2191         error = xfs_qm_init();
2192         if (error)
2193                 goto out_remove_dbg_kobj;
2194
2195         error = register_filesystem(&xfs_fs_type);
2196         if (error)
2197                 goto out_qm_exit;
2198         return 0;
2199
2200  out_qm_exit:
2201         xfs_qm_exit();
2202  out_remove_dbg_kobj:
2203 #ifdef DEBUG
2204         xfs_sysfs_del(&xfs_dbg_kobj);
2205  out_remove_stats_kobj:
2206 #endif
2207         xfs_sysfs_del(&xfsstats.xs_kobj);
2208  out_free_stats:
2209         free_percpu(xfsstats.xs_stats);
2210  out_kset_unregister:
2211         kset_unregister(xfs_kset);
2212  out_sysctl_unregister:
2213         xfs_sysctl_unregister();
2214  out_cleanup_procfs:
2215         xfs_cleanup_procfs();
2216  out_buf_terminate:
2217         xfs_buf_terminate();
2218  out_mru_cache_uninit:
2219         xfs_mru_cache_uninit();
2220  out_destroy_wq:
2221         xfs_destroy_workqueues();
2222  out_destroy_zones:
2223         xfs_destroy_zones();
2224  out:
2225         return error;
2226 }
2227
2228 STATIC void __exit
2229 exit_xfs_fs(void)
2230 {
2231         xfs_qm_exit();
2232         unregister_filesystem(&xfs_fs_type);
2233 #ifdef DEBUG
2234         xfs_sysfs_del(&xfs_dbg_kobj);
2235 #endif
2236         xfs_sysfs_del(&xfsstats.xs_kobj);
2237         free_percpu(xfsstats.xs_stats);
2238         kset_unregister(xfs_kset);
2239         xfs_sysctl_unregister();
2240         xfs_cleanup_procfs();
2241         xfs_buf_terminate();
2242         xfs_mru_cache_uninit();
2243         xfs_destroy_workqueues();
2244         xfs_destroy_zones();
2245         xfs_uuid_table_free();
2246 }
2247
2248 module_init(init_xfs_fs);
2249 module_exit(exit_xfs_fs);
2250
2251 MODULE_AUTHOR("Silicon Graphics, Inc.");
2252 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2253 MODULE_LICENSE("GPL");