fs/xfs/xfs_super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6
   7 #include "xfs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_sb.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_btree.h"
  16 #include "xfs_bmap.h"
  17 #include "xfs_alloc.h"
  18 #include "xfs_fsops.h"
  19 #include "xfs_trans.h"
  20 #include "xfs_buf_item.h"
  21 #include "xfs_log.h"
  22 #include "xfs_log_priv.h"
  23 #include "xfs_dir2.h"
  24 #include "xfs_extfree_item.h"
  25 #include "xfs_mru_cache.h"
  26 #include "xfs_inode_item.h"
  27 #include "xfs_icache.h"
  28 #include "xfs_trace.h"
  29 #include "xfs_icreate_item.h"
  30 #include "xfs_filestream.h"
  31 #include "xfs_quota.h"
  32 #include "xfs_sysfs.h"
  33 #include "xfs_ondisk.h"
  34 #include "xfs_rmap_item.h"
  35 #include "xfs_refcount_item.h"
  36 #include "xfs_bmap_item.h"
  37 #include "xfs_reflink.h"
  38 #include "xfs_pwork.h"
  39 #include "xfs_ag.h"
  40 #include "xfs_defer.h"
  41
  42 #include <linux/magic.h>
  43 #include <linux/fs_context.h>
  44 #include <linux/fs_parser.h>
  45
  46 static const struct super_operations xfs_super_operations;
  47
  48 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  49 #ifdef DEBUG
  50 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  51 #endif
  52
  53 #ifdef CONFIG_HOTPLUG_CPU
  54 static LIST_HEAD(xfs_mount_list);
  55 static DEFINE_SPINLOCK(xfs_mount_list_lock);
  56
  57 static inline void xfs_mount_list_add(struct xfs_mount *mp)
  58 {
  59         spin_lock(&xfs_mount_list_lock);
  60         list_add(&mp->m_mount_list, &xfs_mount_list);
  61         spin_unlock(&xfs_mount_list_lock);
  62 }
  63
  64 static inline void xfs_mount_list_del(struct xfs_mount *mp)
  65 {
  66         spin_lock(&xfs_mount_list_lock);
  67         list_del(&mp->m_mount_list);
  68         spin_unlock(&xfs_mount_list_lock);
  69 }
  70 #else /* !CONFIG_HOTPLUG_CPU */
  71 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
  72 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
  73 #endif
  74
  75 enum xfs_dax_mode {
  76         XFS_DAX_INODE = 0,
  77         XFS_DAX_ALWAYS = 1,
  78         XFS_DAX_NEVER = 2,
  79 };
  80
  81 static void
  82 xfs_mount_set_dax_mode(
  83         struct xfs_mount        *mp,
  84         enum xfs_dax_mode       mode)
  85 {
  86         switch (mode) {
  87         case XFS_DAX_INODE:
  88                 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
  89                 break;
  90         case XFS_DAX_ALWAYS:
  91                 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
  92                 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
  93                 break;
  94         case XFS_DAX_NEVER:
  95                 mp->m_features |= XFS_FEAT_DAX_NEVER;
  96                 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
  97                 break;
  98         }
  99 }
 100
 101 static const struct constant_table dax_param_enums[] = {
 102         {"inode",       XFS_DAX_INODE },
 103         {"always",      XFS_DAX_ALWAYS },
 104         {"never",       XFS_DAX_NEVER },
 105         {}
 106 };
 107
 108 /*
 109  * Table driven mount option parser.
 110  */
 111 enum {
 112         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
 113         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
 114         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
 115         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
 116         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
 117         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
 118         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
 119         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
 120         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
 121 };
 122
 123 static const struct fs_parameter_spec xfs_fs_parameters[] = {
 124         fsparam_u32("logbufs",          Opt_logbufs),
 125         fsparam_string("logbsize",      Opt_logbsize),
 126         fsparam_string("logdev",        Opt_logdev),
 127         fsparam_string("rtdev",         Opt_rtdev),
 128         fsparam_flag("wsync",           Opt_wsync),
 129         fsparam_flag("noalign",         Opt_noalign),
 130         fsparam_flag("swalloc",         Opt_swalloc),
 131         fsparam_u32("sunit",            Opt_sunit),
 132         fsparam_u32("swidth",           Opt_swidth),
 133         fsparam_flag("nouuid",          Opt_nouuid),
 134         fsparam_flag("grpid",           Opt_grpid),
 135         fsparam_flag("nogrpid",         Opt_nogrpid),
 136         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 137         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 138         fsparam_string("allocsize",     Opt_allocsize),
 139         fsparam_flag("norecovery",      Opt_norecovery),
 140         fsparam_flag("inode64",         Opt_inode64),
 141         fsparam_flag("inode32",         Opt_inode32),
 142         fsparam_flag("ikeep",           Opt_ikeep),
 143         fsparam_flag("noikeep",         Opt_noikeep),
 144         fsparam_flag("largeio",         Opt_largeio),
 145         fsparam_flag("nolargeio",       Opt_nolargeio),
 146         fsparam_flag("attr2",           Opt_attr2),
 147         fsparam_flag("noattr2",         Opt_noattr2),
 148         fsparam_flag("filestreams",     Opt_filestreams),
 149         fsparam_flag("quota",           Opt_quota),
 150         fsparam_flag("noquota",         Opt_noquota),
 151         fsparam_flag("usrquota",        Opt_usrquota),
 152         fsparam_flag("grpquota",        Opt_grpquota),
 153         fsparam_flag("prjquota",        Opt_prjquota),
 154         fsparam_flag("uquota",          Opt_uquota),
 155         fsparam_flag("gquota",          Opt_gquota),
 156         fsparam_flag("pquota",          Opt_pquota),
 157         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 158         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 159         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 160         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 161         fsparam_flag("discard",         Opt_discard),
 162         fsparam_flag("nodiscard",       Opt_nodiscard),
 163         fsparam_flag("dax",             Opt_dax),
 164         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 165         {}
 166 };
 167
 168 struct proc_xfs_info {
 169         uint64_t        flag;
 170         char            *str;
 171 };
 172
 173 static int
 174 xfs_fs_show_options(
 175         struct seq_file         *m,
 176         struct dentry           *root)
 177 {
 178         static struct proc_xfs_info xfs_info_set[] = {
 179                 /* the few simple ones we can get from the mount struct */
 180                 { XFS_FEAT_IKEEP,               ",ikeep" },
 181                 { XFS_FEAT_WSYNC,               ",wsync" },
 182                 { XFS_FEAT_NOALIGN,             ",noalign" },
 183                 { XFS_FEAT_SWALLOC,             ",swalloc" },
 184                 { XFS_FEAT_NOUUID,              ",nouuid" },
 185                 { XFS_FEAT_NORECOVERY,          ",norecovery" },
 186                 { XFS_FEAT_ATTR2,               ",attr2" },
 187                 { XFS_FEAT_FILESTREAMS,         ",filestreams" },
 188                 { XFS_FEAT_GRPID,               ",grpid" },
 189                 { XFS_FEAT_DISCARD,             ",discard" },
 190                 { XFS_FEAT_LARGE_IOSIZE,        ",largeio" },
 191                 { XFS_FEAT_DAX_ALWAYS,          ",dax=always" },
 192                 { XFS_FEAT_DAX_NEVER,           ",dax=never" },
 193                 { 0, NULL }
 194         };
 195         struct xfs_mount        *mp = XFS_M(root->d_sb);
 196         struct proc_xfs_info    *xfs_infop;
 197
 198         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 199                 if (mp->m_features & xfs_infop->flag)
 200                         seq_puts(m, xfs_infop->str);
 201         }
 202
 203         seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
 204
 205         if (xfs_has_allocsize(mp))
 206                 seq_printf(m, ",allocsize=%dk",
 207                            (1 << mp->m_allocsize_log) >> 10);
 208
 209         if (mp->m_logbufs > 0)
 210                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 211         if (mp->m_logbsize > 0)
 212                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 213
 214         if (mp->m_logname)
 215                 seq_show_option(m, "logdev", mp->m_logname);
 216         if (mp->m_rtname)
 217                 seq_show_option(m, "rtdev", mp->m_rtname);
 218
 219         if (mp->m_dalign > 0)
 220                 seq_printf(m, ",sunit=%d",
 221                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 222         if (mp->m_swidth > 0)
 223                 seq_printf(m, ",swidth=%d",
 224                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 225
 226         if (mp->m_qflags & XFS_UQUOTA_ENFD)
 227                 seq_puts(m, ",usrquota");
 228         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 229                 seq_puts(m, ",uqnoenforce");
 230
 231         if (mp->m_qflags & XFS_PQUOTA_ENFD)
 232                 seq_puts(m, ",prjquota");
 233         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
 234                 seq_puts(m, ",pqnoenforce");
 235
 236         if (mp->m_qflags & XFS_GQUOTA_ENFD)
 237                 seq_puts(m, ",grpquota");
 238         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
 239                 seq_puts(m, ",gqnoenforce");
 240
 241         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 242                 seq_puts(m, ",noquota");
 243
 244         return 0;
 245 }
 246
 247 /*
 248  * Set parameters for inode allocation heuristics, taking into account
 249  * filesystem size and inode32/inode64 mount options; i.e. specifically
 250  * whether or not XFS_FEAT_SMALL_INUMS is set.
 251  *
 252  * Inode allocation patterns are altered only if inode32 is requested
 253  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
 254  * If altered, XFS_OPSTATE_INODE32 is set as well.
 255  *
 256  * An agcount independent of that in the mount structure is provided
 257  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 258  * to the potentially higher ag count.
 259  *
 260  * Returns the maximum AG index which may contain inodes.
 261  */
 262 xfs_agnumber_t
 263 xfs_set_inode_alloc(
 264         struct xfs_mount *mp,
 265         xfs_agnumber_t  agcount)
 266 {
 267         xfs_agnumber_t  index;
 268         xfs_agnumber_t  maxagi = 0;
 269         xfs_sb_t        *sbp = &mp->m_sb;
 270         xfs_agnumber_t  max_metadata;
 271         xfs_agino_t     agino;
 272         xfs_ino_t       ino;
 273
 274         /*
 275          * Calculate how much should be reserved for inodes to meet
 276          * the max inode percentage.  Used only for inode32.
 277          */
 278         if (M_IGEO(mp)->maxicount) {
 279                 uint64_t        icount;
 280
 281                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 282                 do_div(icount, 100);
 283                 icount += sbp->sb_agblocks - 1;
 284                 do_div(icount, sbp->sb_agblocks);
 285                 max_metadata = icount;
 286         } else {
 287                 max_metadata = agcount;
 288         }
 289
 290         /* Get the last possible inode in the filesystem */
 291         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 292         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 293
 294         /*
 295          * If user asked for no more than 32-bit inodes, and the fs is
 296          * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
 297          * the allocator to accommodate the request.
 298          */
 299         if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
 300                 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 301         else
 302                 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 303
 304         for (index = 0; index < agcount; index++) {
 305                 struct xfs_perag        *pag;
 306
 307                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 308
 309                 pag = xfs_perag_get(mp, index);
 310
 311                 if (xfs_is_inode32(mp)) {
 312                         if (ino > XFS_MAXINUMBER_32) {
 313                                 pag->pagi_inodeok = 0;
 314                                 pag->pagf_metadata = 0;
 315                         } else {
 316                                 pag->pagi_inodeok = 1;
 317                                 maxagi++;
 318                                 if (index < max_metadata)
 319                                         pag->pagf_metadata = 1;
 320                                 else
 321                                         pag->pagf_metadata = 0;
 322                         }
 323                 } else {
 324                         pag->pagi_inodeok = 1;
 325                         pag->pagf_metadata = 0;
 326                 }
 327
 328                 xfs_perag_put(pag);
 329         }
 330
 331         return xfs_is_inode32(mp) ? maxagi : agcount;
 332 }
 333
 334 static int
 335 xfs_setup_dax_always(
 336         struct xfs_mount        *mp)
 337 {
 338         if (!mp->m_ddev_targp->bt_daxdev &&
 339             (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
 340                 xfs_alert(mp,
 341                         "DAX unsupported by block device. Turning off DAX.");
 342                 goto disable_dax;
 343         }
 344
 345         if (mp->m_super->s_blocksize != PAGE_SIZE) {
 346                 xfs_alert(mp,
 347                         "DAX not supported for blocksize. Turning off DAX.");
 348                 goto disable_dax;
 349         }
 350
 351         if (xfs_has_reflink(mp)) {
 352                 xfs_alert(mp, "DAX and reflink cannot be used together!");
 353                 return -EINVAL;
 354         }
 355
 356         xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
 357         return 0;
 358
 359 disable_dax:
 360         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
 361         return 0;
 362 }
 363
 364 STATIC int
 365 xfs_blkdev_get(
 366         xfs_mount_t             *mp,
 367         const char              *name,
 368         struct block_device     **bdevp)
 369 {
 370         int                     error = 0;
 371
 372         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 373                                     mp);
 374         if (IS_ERR(*bdevp)) {
 375                 error = PTR_ERR(*bdevp);
 376                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 377         }
 378
 379         return error;
 380 }
 381
 382 STATIC void
 383 xfs_blkdev_put(
 384         struct block_device     *bdev)
 385 {
 386         if (bdev)
 387                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 388 }
 389
 390 STATIC void
 391 xfs_close_devices(
 392         struct xfs_mount        *mp)
 393 {
 394         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 395                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 396
 397                 xfs_free_buftarg(mp->m_logdev_targp);
 398                 xfs_blkdev_put(logdev);
 399         }
 400         if (mp->m_rtdev_targp) {
 401                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 402
 403                 xfs_free_buftarg(mp->m_rtdev_targp);
 404                 xfs_blkdev_put(rtdev);
 405         }
 406         xfs_free_buftarg(mp->m_ddev_targp);
 407 }
 408
 409 /*
 410  * The file system configurations are:
 411  *      (1) device (partition) with data and internal log
 412  *      (2) logical volume with data and log subvolumes.
 413  *      (3) logical volume with data, log, and realtime subvolumes.
 414  *
 415  * We only have to handle opening the log and realtime volumes here if
 416  * they are present.  The data subvolume has already been opened by
 417  * get_sb_bdev() and is stored in sb->s_bdev.
 418  */
 419 STATIC int
 420 xfs_open_devices(
 421         struct xfs_mount        *mp)
 422 {
 423         struct block_device     *ddev = mp->m_super->s_bdev;
 424         struct block_device     *logdev = NULL, *rtdev = NULL;
 425         int                     error;
 426
 427         /*
 428          * Open real time and log devices - order is important.
 429          */
 430         if (mp->m_logname) {
 431                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 432                 if (error)
 433                         return error;
 434         }
 435
 436         if (mp->m_rtname) {
 437                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 438                 if (error)
 439                         goto out_close_logdev;
 440
 441                 if (rtdev == ddev || rtdev == logdev) {
 442                         xfs_warn(mp,
 443         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 444                         error = -EINVAL;
 445                         goto out_close_rtdev;
 446                 }
 447         }
 448
 449         /*
 450          * Setup xfs_mount buffer target pointers
 451          */
 452         error = -ENOMEM;
 453         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
 454         if (!mp->m_ddev_targp)
 455                 goto out_close_rtdev;
 456
 457         if (rtdev) {
 458                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
 459                 if (!mp->m_rtdev_targp)
 460                         goto out_free_ddev_targ;
 461         }
 462
 463         if (logdev && logdev != ddev) {
 464                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
 465                 if (!mp->m_logdev_targp)
 466                         goto out_free_rtdev_targ;
 467         } else {
 468                 mp->m_logdev_targp = mp->m_ddev_targp;
 469         }
 470
 471         return 0;
 472
 473  out_free_rtdev_targ:
 474         if (mp->m_rtdev_targp)
 475                 xfs_free_buftarg(mp->m_rtdev_targp);
 476  out_free_ddev_targ:
 477         xfs_free_buftarg(mp->m_ddev_targp);
 478  out_close_rtdev:
 479         xfs_blkdev_put(rtdev);
 480  out_close_logdev:
 481         if (logdev && logdev != ddev)
 482                 xfs_blkdev_put(logdev);
 483         return error;
 484 }
 485
 486 /*
 487  * Setup xfs_mount buffer target pointers based on superblock
 488  */
 489 STATIC int
 490 xfs_setup_devices(
 491         struct xfs_mount        *mp)
 492 {
 493         int                     error;
 494
 495         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 496         if (error)
 497                 return error;
 498
 499         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 500                 unsigned int    log_sector_size = BBSIZE;
 501
 502                 if (xfs_has_sector(mp))
 503                         log_sector_size = mp->m_sb.sb_logsectsize;
 504                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 505                                             log_sector_size);
 506                 if (error)
 507                         return error;
 508         }
 509         if (mp->m_rtdev_targp) {
 510                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 511                                             mp->m_sb.sb_sectsize);
 512                 if (error)
 513                         return error;
 514         }
 515
 516         return 0;
 517 }
 518
 519 STATIC int
 520 xfs_init_mount_workqueues(
 521         struct xfs_mount        *mp)
 522 {
 523         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 524                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 525                         1, mp->m_super->s_id);
 526         if (!mp->m_buf_workqueue)
 527                 goto out;
 528
 529         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 530                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 531                         0, mp->m_super->s_id);
 532         if (!mp->m_unwritten_workqueue)
 533                 goto out_destroy_buf;
 534
 535         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 536                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 537                         0, mp->m_super->s_id);
 538         if (!mp->m_reclaim_workqueue)
 539                 goto out_destroy_unwritten;
 540
 541         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
 542                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
 543                         0, mp->m_super->s_id);
 544         if (!mp->m_blockgc_wq)
 545                 goto out_destroy_reclaim;
 546
 547         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
 548                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 549                         1, mp->m_super->s_id);
 550         if (!mp->m_inodegc_wq)
 551                 goto out_destroy_blockgc;
 552
 553         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 554                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 555         if (!mp->m_sync_workqueue)
 556                 goto out_destroy_inodegc;
 557
 558         return 0;
 559
 560 out_destroy_inodegc:
 561         destroy_workqueue(mp->m_inodegc_wq);
 562 out_destroy_blockgc:
 563         destroy_workqueue(mp->m_blockgc_wq);
 564 out_destroy_reclaim:
 565         destroy_workqueue(mp->m_reclaim_workqueue);
 566 out_destroy_unwritten:
 567         destroy_workqueue(mp->m_unwritten_workqueue);
 568 out_destroy_buf:
 569         destroy_workqueue(mp->m_buf_workqueue);
 570 out:
 571         return -ENOMEM;
 572 }
 573
 574 STATIC void
 575 xfs_destroy_mount_workqueues(
 576         struct xfs_mount        *mp)
 577 {
 578         destroy_workqueue(mp->m_sync_workqueue);
 579         destroy_workqueue(mp->m_blockgc_wq);
 580         destroy_workqueue(mp->m_inodegc_wq);
 581         destroy_workqueue(mp->m_reclaim_workqueue);
 582         destroy_workqueue(mp->m_unwritten_workqueue);
 583         destroy_workqueue(mp->m_buf_workqueue);
 584 }
 585
 586 static void
 587 xfs_flush_inodes_worker(
 588         struct work_struct      *work)
 589 {
 590         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 591                                                    m_flush_inodes_work);
 592         struct super_block      *sb = mp->m_super;
 593
 594         if (down_read_trylock(&sb->s_umount)) {
 595                 sync_inodes_sb(sb);
 596                 up_read(&sb->s_umount);
 597         }
 598 }
 599
 600 /*
 601  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 602  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 603  * for IO to complete so that we effectively throttle multiple callers to the
 604  * rate at which IO is completing.
 605  */
 606 void
 607 xfs_flush_inodes(
 608         struct xfs_mount        *mp)
 609 {
 610         /*
 611          * If flush_work() returns true then that means we waited for a flush
 612          * which was already in progress.  Don't bother running another scan.
 613          */
 614         if (flush_work(&mp->m_flush_inodes_work))
 615                 return;
 616
 617         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 618         flush_work(&mp->m_flush_inodes_work);
 619 }
 620
 621 /* Catch misguided souls that try to use this interface on XFS */
 622 STATIC struct inode *
 623 xfs_fs_alloc_inode(
 624         struct super_block      *sb)
 625 {
 626         BUG();
 627         return NULL;
 628 }
 629
 630 /*
 631  * Now that the generic code is guaranteed not to be accessing
 632  * the linux inode, we can inactivate and reclaim the inode.
 633  */
 634 STATIC void
 635 xfs_fs_destroy_inode(
 636         struct inode            *inode)
 637 {
 638         struct xfs_inode        *ip = XFS_I(inode);
 639
 640         trace_xfs_destroy_inode(ip);
 641
 642         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 643         XFS_STATS_INC(ip->i_mount, vn_rele);
 644         XFS_STATS_INC(ip->i_mount, vn_remove);
 645         xfs_inode_mark_reclaimable(ip);
 646 }
 647
 648 static void
 649 xfs_fs_dirty_inode(
 650         struct inode                    *inode,
 651         int                             flag)
 652 {
 653         struct xfs_inode                *ip = XFS_I(inode);
 654         struct xfs_mount                *mp = ip->i_mount;
 655         struct xfs_trans                *tp;
 656
 657         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 658                 return;
 659         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 660                 return;
 661
 662         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 663                 return;
 664         xfs_ilock(ip, XFS_ILOCK_EXCL);
 665         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 666         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 667         xfs_trans_commit(tp);
 668 }
 669
 670 /*
 671  * Slab object creation initialisation for the XFS inode.
 672  * This covers only the idempotent fields in the XFS inode;
 673  * all other fields need to be initialised on allocation
 674  * from the slab. This avoids the need to repeatedly initialise
 675  * fields in the xfs inode that left in the initialise state
 676  * when freeing the inode.
 677  */
 678 STATIC void
 679 xfs_fs_inode_init_once(
 680         void                    *inode)
 681 {
 682         struct xfs_inode        *ip = inode;
 683
 684         memset(ip, 0, sizeof(struct xfs_inode));
 685
 686         /* vfs inode */
 687         inode_init_once(VFS_I(ip));
 688
 689         /* xfs inode */
 690         atomic_set(&ip->i_pincount, 0);
 691         spin_lock_init(&ip->i_flags_lock);
 692
 693         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 694                      "xfsino", ip->i_ino);
 695 }
 696
 697 /*
 698  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 699  * serialised against cache hits here via the inode->i_lock and igrab() in
 700  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 701  * racing with us, and it avoids needing to grab a spinlock here for every inode
 702  * we drop the final reference on.
 703  */
 704 STATIC int
 705 xfs_fs_drop_inode(
 706         struct inode            *inode)
 707 {
 708         struct xfs_inode        *ip = XFS_I(inode);
 709
 710         /*
 711          * If this unlinked inode is in the middle of recovery, don't
 712          * drop the inode just yet; log recovery will take care of
 713          * that.  See the comment for this inode flag.
 714          */
 715         if (ip->i_flags & XFS_IRECOVERY) {
 716                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
 717                 return 0;
 718         }
 719
 720         return generic_drop_inode(inode);
 721 }
 722
 723 static void
 724 xfs_mount_free(
 725         struct xfs_mount        *mp)
 726 {
 727         kfree(mp->m_rtname);
 728         kfree(mp->m_logname);
 729         kmem_free(mp);
 730 }
 731
 732 STATIC int
 733 xfs_fs_sync_fs(
 734         struct super_block      *sb,
 735         int                     wait)
 736 {
 737         struct xfs_mount        *mp = XFS_M(sb);
 738         int                     error;
 739
 740         trace_xfs_fs_sync_fs(mp, __return_address);
 741
 742         /*
 743          * Doing anything during the async pass would be counterproductive.
 744          */
 745         if (!wait)
 746                 return 0;
 747
 748         error = xfs_log_force(mp, XFS_LOG_SYNC);
 749         if (error)
 750                 return error;
 751
 752         if (laptop_mode) {
 753                 /*
 754                  * The disk must be active because we're syncing.
 755                  * We schedule log work now (now that the disk is
 756                  * active) instead of later (when it might not be).
 757                  */
 758                 flush_delayed_work(&mp->m_log->l_work);
 759         }
 760
 761         /*
 762          * If we are called with page faults frozen out, it means we are about
 763          * to freeze the transaction subsystem. Take the opportunity to shut
 764          * down inodegc because once SB_FREEZE_FS is set it's too late to
 765          * prevent inactivation races with freeze. The fs doesn't get called
 766          * again by the freezing process until after SB_FREEZE_FS has been set,
 767          * so it's now or never.  Same logic applies to speculative allocation
 768          * garbage collection.
 769          *
 770          * We don't care if this is a normal syncfs call that does this or
 771          * freeze that does this - we can run this multiple times without issue
 772          * and we won't race with a restart because a restart can only occur
 773          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
 774          */
 775         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
 776                 xfs_inodegc_stop(mp);
 777                 xfs_blockgc_stop(mp);
 778         }
 779
 780         return 0;
 781 }
 782
 783 STATIC int
 784 xfs_fs_statfs(
 785         struct dentry           *dentry,
 786         struct kstatfs          *statp)
 787 {
 788         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 789         xfs_sb_t                *sbp = &mp->m_sb;
 790         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 791         uint64_t                fakeinos, id;
 792         uint64_t                icount;
 793         uint64_t                ifree;
 794         uint64_t                fdblocks;
 795         xfs_extlen_t            lsize;
 796         int64_t                 ffree;
 797
 798         /* Wait for whatever inactivations are in progress. */
 799         xfs_inodegc_flush(mp);
 800
 801         statp->f_type = XFS_SUPER_MAGIC;
 802         statp->f_namelen = MAXNAMELEN - 1;
 803
 804         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 805         statp->f_fsid = u64_to_fsid(id);
 806
 807         icount = percpu_counter_sum(&mp->m_icount);
 808         ifree = percpu_counter_sum(&mp->m_ifree);
 809         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 810
 811         spin_lock(&mp->m_sb_lock);
 812         statp->f_bsize = sbp->sb_blocksize;
 813         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 814         statp->f_blocks = sbp->sb_dblocks - lsize;
 815         spin_unlock(&mp->m_sb_lock);
 816
 817         /* make sure statp->f_bfree does not underflow */
 818         statp->f_bfree = max_t(int64_t, 0,
 819                                 fdblocks - xfs_fdblocks_unavailable(mp));
 820         statp->f_bavail = statp->f_bfree;
 821
 822         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 823         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 824         if (M_IGEO(mp)->maxicount)
 825                 statp->f_files = min_t(typeof(statp->f_files),
 826                                         statp->f_files,
 827                                         M_IGEO(mp)->maxicount);
 828
 829         /* If sb_icount overshot maxicount, report actual allocation */
 830         statp->f_files = max_t(typeof(statp->f_files),
 831                                         statp->f_files,
 832                                         sbp->sb_icount);
 833
 834         /* make sure statp->f_ffree does not underflow */
 835         ffree = statp->f_files - (icount - ifree);
 836         statp->f_ffree = max_t(int64_t, ffree, 0);
 837
 838
 839         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 840             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 841                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 842                 xfs_qm_statvfs(ip, statp);
 843
 844         if (XFS_IS_REALTIME_MOUNT(mp) &&
 845             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 846                 s64     freertx;
 847
 848                 statp->f_blocks = sbp->sb_rblocks;
 849                 freertx = percpu_counter_sum_positive(&mp->m_frextents);
 850                 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
 851         }
 852
 853         return 0;
 854 }
 855
 856 STATIC void
 857 xfs_save_resvblks(struct xfs_mount *mp)
 858 {
 859         uint64_t resblks = 0;
 860
 861         mp->m_resblks_save = mp->m_resblks;
 862         xfs_reserve_blocks(mp, &resblks, NULL);
 863 }
 864
 865 STATIC void
 866 xfs_restore_resvblks(struct xfs_mount *mp)
 867 {
 868         uint64_t resblks;
 869
 870         if (mp->m_resblks_save) {
 871                 resblks = mp->m_resblks_save;
 872                 mp->m_resblks_save = 0;
 873         } else
 874                 resblks = xfs_default_resblks(mp);
 875
 876         xfs_reserve_blocks(mp, &resblks, NULL);
 877 }
 878
 879 /*
 880  * Second stage of a freeze. The data is already frozen so we only
 881  * need to take care of the metadata. Once that's done sync the superblock
 882  * to the log to dirty it in case of a crash while frozen. This ensures that we
 883  * will recover the unlinked inode lists on the next mount.
 884  */
 885 STATIC int
 886 xfs_fs_freeze(
 887         struct super_block      *sb)
 888 {
 889         struct xfs_mount        *mp = XFS_M(sb);
 890         unsigned int            flags;
 891         int                     ret;
 892
 893         /*
 894          * The filesystem is now frozen far enough that memory reclaim
 895          * cannot safely operate on the filesystem. Hence we need to
 896          * set a GFP_NOFS context here to avoid recursion deadlocks.
 897          */
 898         flags = memalloc_nofs_save();
 899         xfs_save_resvblks(mp);
 900         ret = xfs_log_quiesce(mp);
 901         memalloc_nofs_restore(flags);
 902
 903         /*
 904          * For read-write filesystems, we need to restart the inodegc on error
 905          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
 906          * going to be run to restart it now.  We are at SB_FREEZE_FS level
 907          * here, so we can restart safely without racing with a stop in
 908          * xfs_fs_sync_fs().
 909          */
 910         if (ret && !xfs_is_readonly(mp)) {
 911                 xfs_blockgc_start(mp);
 912                 xfs_inodegc_start(mp);
 913         }
 914
 915         return ret;
 916 }
 917
 918 STATIC int
 919 xfs_fs_unfreeze(
 920         struct super_block      *sb)
 921 {
 922         struct xfs_mount        *mp = XFS_M(sb);
 923
 924         xfs_restore_resvblks(mp);
 925         xfs_log_work_queue(mp);
 926
 927         /*
 928          * Don't reactivate the inodegc worker on a readonly filesystem because
 929          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
 930          * worker because there are no speculative preallocations on a readonly
 931          * filesystem.
 932          */
 933         if (!xfs_is_readonly(mp)) {
 934                 xfs_blockgc_start(mp);
 935                 xfs_inodegc_start(mp);
 936         }
 937
 938         return 0;
 939 }
 940
 941 /*
 942  * This function fills in xfs_mount_t fields based on mount args.
 943  * Note: the superblock _has_ now been read in.
 944  */
 945 STATIC int
 946 xfs_finish_flags(
 947         struct xfs_mount        *mp)
 948 {
 949         /* Fail a mount where the logbuf is smaller than the log stripe */
 950         if (xfs_has_logv2(mp)) {
 951                 if (mp->m_logbsize <= 0 &&
 952                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 953                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 954                 } else if (mp->m_logbsize > 0 &&
 955                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 956                         xfs_warn(mp,
 957                 "logbuf size must be greater than or equal to log stripe size");
 958                         return -EINVAL;
 959                 }
 960         } else {
 961                 /* Fail a mount if the logbuf is larger than 32K */
 962                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 963                         xfs_warn(mp,
 964                 "logbuf size for version 1 logs must be 16K or 32K");
 965                         return -EINVAL;
 966                 }
 967         }
 968
 969         /*
 970          * V5 filesystems always use attr2 format for attributes.
 971          */
 972         if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
 973                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 974                              "attr2 is always enabled for V5 filesystems.");
 975                 return -EINVAL;
 976         }
 977
 978         /*
 979          * prohibit r/w mounts of read-only filesystems
 980          */
 981         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
 982                 xfs_warn(mp,
 983                         "cannot mount a read-only filesystem as read-write");
 984                 return -EROFS;
 985         }
 986
 987         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
 988             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
 989             !xfs_has_pquotino(mp)) {
 990                 xfs_warn(mp,
 991                   "Super block does not support project and group quota together");
 992                 return -EINVAL;
 993         }
 994
 995         return 0;
 996 }
 997
 998 static int
 999 xfs_init_percpu_counters(
1000         struct xfs_mount        *mp)
1001 {
1002         int             error;
1003
1004         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1005         if (error)
1006                 return -ENOMEM;
1007
1008         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1009         if (error)
1010                 goto free_icount;
1011
1012         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1013         if (error)
1014                 goto free_ifree;
1015
1016         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1017         if (error)
1018                 goto free_fdblocks;
1019
1020         error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1021         if (error)
1022                 goto free_delalloc;
1023
1024         return 0;
1025
1026 free_delalloc:
1027         percpu_counter_destroy(&mp->m_delalloc_blks);
1028 free_fdblocks:
1029         percpu_counter_destroy(&mp->m_fdblocks);
1030 free_ifree:
1031         percpu_counter_destroy(&mp->m_ifree);
1032 free_icount:
1033         percpu_counter_destroy(&mp->m_icount);
1034         return -ENOMEM;
1035 }
1036
1037 void
1038 xfs_reinit_percpu_counters(
1039         struct xfs_mount        *mp)
1040 {
1041         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1042         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1043         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1044         percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1045 }
1046
1047 static void
1048 xfs_destroy_percpu_counters(
1049         struct xfs_mount        *mp)
1050 {
1051         percpu_counter_destroy(&mp->m_icount);
1052         percpu_counter_destroy(&mp->m_ifree);
1053         percpu_counter_destroy(&mp->m_fdblocks);
1054         ASSERT(xfs_is_shutdown(mp) ||
1055                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1056         percpu_counter_destroy(&mp->m_delalloc_blks);
1057         percpu_counter_destroy(&mp->m_frextents);
1058 }
1059
1060 static int
1061 xfs_inodegc_init_percpu(
1062         struct xfs_mount        *mp)
1063 {
1064         struct xfs_inodegc      *gc;
1065         int                     cpu;
1066
1067         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1068         if (!mp->m_inodegc)
1069                 return -ENOMEM;
1070
1071         for_each_possible_cpu(cpu) {
1072                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1073                 init_llist_head(&gc->list);
1074                 gc->items = 0;
1075                 INIT_WORK(&gc->work, xfs_inodegc_worker);
1076         }
1077         return 0;
1078 }
1079
1080 static void
1081 xfs_inodegc_free_percpu(
1082         struct xfs_mount        *mp)
1083 {
1084         if (!mp->m_inodegc)
1085                 return;
1086         free_percpu(mp->m_inodegc);
1087 }
1088
1089 static void
1090 xfs_fs_put_super(
1091         struct super_block      *sb)
1092 {
1093         struct xfs_mount        *mp = XFS_M(sb);
1094
1095         /* if ->fill_super failed, we have no mount to tear down */
1096         if (!sb->s_fs_info)
1097                 return;
1098
1099         xfs_notice(mp, "Unmounting Filesystem");
1100         xfs_filestream_unmount(mp);
1101         xfs_unmountfs(mp);
1102
1103         xfs_freesb(mp);
1104         free_percpu(mp->m_stats.xs_stats);
1105         xfs_mount_list_del(mp);
1106         xfs_inodegc_free_percpu(mp);
1107         xfs_destroy_percpu_counters(mp);
1108         xfs_destroy_mount_workqueues(mp);
1109         xfs_close_devices(mp);
1110
1111         sb->s_fs_info = NULL;
1112         xfs_mount_free(mp);
1113 }
1114
1115 static long
1116 xfs_fs_nr_cached_objects(
1117         struct super_block      *sb,
1118         struct shrink_control   *sc)
1119 {
1120         /* Paranoia: catch incorrect calls during mount setup or teardown */
1121         if (WARN_ON_ONCE(!sb->s_fs_info))
1122                 return 0;
1123         return xfs_reclaim_inodes_count(XFS_M(sb));
1124 }
1125
1126 static long
1127 xfs_fs_free_cached_objects(
1128         struct super_block      *sb,
1129         struct shrink_control   *sc)
1130 {
1131         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1132 }
1133
1134 static const struct super_operations xfs_super_operations = {
1135         .alloc_inode            = xfs_fs_alloc_inode,
1136         .destroy_inode          = xfs_fs_destroy_inode,
1137         .dirty_inode            = xfs_fs_dirty_inode,
1138         .drop_inode             = xfs_fs_drop_inode,
1139         .put_super              = xfs_fs_put_super,
1140         .sync_fs                = xfs_fs_sync_fs,
1141         .freeze_fs              = xfs_fs_freeze,
1142         .unfreeze_fs            = xfs_fs_unfreeze,
1143         .statfs                 = xfs_fs_statfs,
1144         .show_options           = xfs_fs_show_options,
1145         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1146         .free_cached_objects    = xfs_fs_free_cached_objects,
1147 };
1148
1149 static int
1150 suffix_kstrtoint(
1151         const char      *s,
1152         unsigned int    base,
1153         int             *res)
1154 {
1155         int             last, shift_left_factor = 0, _res;
1156         char            *value;
1157         int             ret = 0;
1158
1159         value = kstrdup(s, GFP_KERNEL);
1160         if (!value)
1161                 return -ENOMEM;
1162
1163         last = strlen(value) - 1;
1164         if (value[last] == 'K' || value[last] == 'k') {
1165                 shift_left_factor = 10;
1166                 value[last] = '\0';
1167         }
1168         if (value[last] == 'M' || value[last] == 'm') {
1169                 shift_left_factor = 20;
1170                 value[last] = '\0';
1171         }
1172         if (value[last] == 'G' || value[last] == 'g') {
1173                 shift_left_factor = 30;
1174                 value[last] = '\0';
1175         }
1176
1177         if (kstrtoint(value, base, &_res))
1178                 ret = -EINVAL;
1179         kfree(value);
1180         *res = _res << shift_left_factor;
1181         return ret;
1182 }
1183
1184 static inline void
1185 xfs_fs_warn_deprecated(
1186         struct fs_context       *fc,
1187         struct fs_parameter     *param,
1188         uint64_t                flag,
1189         bool                    value)
1190 {
1191         /* Don't print the warning if reconfiguring and current mount point
1192          * already had the flag set
1193          */
1194         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1195             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1196                 return;
1197         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1198 }
1199
1200 /*
1201  * Set mount state from a mount option.
1202  *
1203  * NOTE: mp->m_super is NULL here!
1204  */
1205 static int
1206 xfs_fs_parse_param(
1207         struct fs_context       *fc,
1208         struct fs_parameter     *param)
1209 {
1210         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1211         struct fs_parse_result  result;
1212         int                     size = 0;
1213         int                     opt;
1214
1215         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1216         if (opt < 0)
1217                 return opt;
1218
1219         switch (opt) {
1220         case Opt_logbufs:
1221                 parsing_mp->m_logbufs = result.uint_32;
1222                 return 0;
1223         case Opt_logbsize:
1224                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1225                         return -EINVAL;
1226                 return 0;
1227         case Opt_logdev:
1228                 kfree(parsing_mp->m_logname);
1229                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1230                 if (!parsing_mp->m_logname)
1231                         return -ENOMEM;
1232                 return 0;
1233         case Opt_rtdev:
1234                 kfree(parsing_mp->m_rtname);
1235                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1236                 if (!parsing_mp->m_rtname)
1237                         return -ENOMEM;
1238                 return 0;
1239         case Opt_allocsize:
1240                 if (suffix_kstrtoint(param->string, 10, &size))
1241                         return -EINVAL;
1242                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1243                 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1244                 return 0;
1245         case Opt_grpid:
1246         case Opt_bsdgroups:
1247                 parsing_mp->m_features |= XFS_FEAT_GRPID;
1248                 return 0;
1249         case Opt_nogrpid:
1250         case Opt_sysvgroups:
1251                 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1252                 return 0;
1253         case Opt_wsync:
1254                 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1255                 return 0;
1256         case Opt_norecovery:
1257                 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1258                 return 0;
1259         case Opt_noalign:
1260                 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1261                 return 0;
1262         case Opt_swalloc:
1263                 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1264                 return 0;
1265         case Opt_sunit:
1266                 parsing_mp->m_dalign = result.uint_32;
1267                 return 0;
1268         case Opt_swidth:
1269                 parsing_mp->m_swidth = result.uint_32;
1270                 return 0;
1271         case Opt_inode32:
1272                 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1273                 return 0;
1274         case Opt_inode64:
1275                 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1276                 return 0;
1277         case Opt_nouuid:
1278                 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1279                 return 0;
1280         case Opt_largeio:
1281                 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1282                 return 0;
1283         case Opt_nolargeio:
1284                 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1285                 return 0;
1286         case Opt_filestreams:
1287                 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1288                 return 0;
1289         case Opt_noquota:
1290                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1291                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1292                 return 0;
1293         case Opt_quota:
1294         case Opt_uquota:
1295         case Opt_usrquota:
1296                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1297                 return 0;
1298         case Opt_qnoenforce:
1299         case Opt_uqnoenforce:
1300                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1301                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1302                 return 0;
1303         case Opt_pquota:
1304         case Opt_prjquota:
1305                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1306                 return 0;
1307         case Opt_pqnoenforce:
1308                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1309                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1310                 return 0;
1311         case Opt_gquota:
1312         case Opt_grpquota:
1313                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1314                 return 0;
1315         case Opt_gqnoenforce:
1316                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1317                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1318                 return 0;
1319         case Opt_discard:
1320                 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1321                 return 0;
1322         case Opt_nodiscard:
1323                 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1324                 return 0;
1325 #ifdef CONFIG_FS_DAX
1326         case Opt_dax:
1327                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1328                 return 0;
1329         case Opt_dax_enum:
1330                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1331                 return 0;
1332 #endif
1333         /* Following mount options will be removed in September 2025 */
1334         case Opt_ikeep:
1335                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1336                 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1337                 return 0;
1338         case Opt_noikeep:
1339                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1340                 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1341                 return 0;
1342         case Opt_attr2:
1343                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1344                 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1345                 return 0;
1346         case Opt_noattr2:
1347                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1348                 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1349                 return 0;
1350         default:
1351                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1352                 return -EINVAL;
1353         }
1354
1355         return 0;
1356 }
1357
1358 static int
1359 xfs_fs_validate_params(
1360         struct xfs_mount        *mp)
1361 {
1362         /* No recovery flag requires a read-only mount */
1363         if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1364                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1365                 return -EINVAL;
1366         }
1367
1368         /*
1369          * We have not read the superblock at this point, so only the attr2
1370          * mount option can set the attr2 feature by this stage.
1371          */
1372         if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1373                 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1374                 return -EINVAL;
1375         }
1376
1377
1378         if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1379                 xfs_warn(mp,
1380         "sunit and swidth options incompatible with the noalign option");
1381                 return -EINVAL;
1382         }
1383
1384         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1385                 xfs_warn(mp, "quota support not available in this kernel.");
1386                 return -EINVAL;
1387         }
1388
1389         if ((mp->m_dalign && !mp->m_swidth) ||
1390             (!mp->m_dalign && mp->m_swidth)) {
1391                 xfs_warn(mp, "sunit and swidth must be specified together");
1392                 return -EINVAL;
1393         }
1394
1395         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1396                 xfs_warn(mp,
1397         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1398                         mp->m_swidth, mp->m_dalign);
1399                 return -EINVAL;
1400         }
1401
1402         if (mp->m_logbufs != -1 &&
1403             mp->m_logbufs != 0 &&
1404             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1405              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1406                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1407                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1408                 return -EINVAL;
1409         }
1410
1411         if (mp->m_logbsize != -1 &&
1412             mp->m_logbsize !=  0 &&
1413             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1414              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1415              !is_power_of_2(mp->m_logbsize))) {
1416                 xfs_warn(mp,
1417                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1418                         mp->m_logbsize);
1419                 return -EINVAL;
1420         }
1421
1422         if (xfs_has_allocsize(mp) &&
1423             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1424              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1425                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1426                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1427                 return -EINVAL;
1428         }
1429
1430         return 0;
1431 }
1432
1433 static int
1434 xfs_fs_fill_super(
1435         struct super_block      *sb,
1436         struct fs_context       *fc)
1437 {
1438         struct xfs_mount        *mp = sb->s_fs_info;
1439         struct inode            *root;
1440         int                     flags = 0, error;
1441
1442         mp->m_super = sb;
1443
1444         error = xfs_fs_validate_params(mp);
1445         if (error)
1446                 goto out_free_names;
1447
1448         sb_min_blocksize(sb, BBSIZE);
1449         sb->s_xattr = xfs_xattr_handlers;
1450         sb->s_export_op = &xfs_export_operations;
1451 #ifdef CONFIG_XFS_QUOTA
1452         sb->s_qcop = &xfs_quotactl_operations;
1453         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1454 #endif
1455         sb->s_op = &xfs_super_operations;
1456
1457         /*
1458          * Delay mount work if the debug hook is set. This is debug
1459          * instrumention to coordinate simulation of xfs mount failures with
1460          * VFS superblock operations
1461          */
1462         if (xfs_globals.mount_delay) {
1463                 xfs_notice(mp, "Delaying mount for %d seconds.",
1464                         xfs_globals.mount_delay);
1465                 msleep(xfs_globals.mount_delay * 1000);
1466         }
1467
1468         if (fc->sb_flags & SB_SILENT)
1469                 flags |= XFS_MFSI_QUIET;
1470
1471         error = xfs_open_devices(mp);
1472         if (error)
1473                 goto out_free_names;
1474
1475         error = xfs_init_mount_workqueues(mp);
1476         if (error)
1477                 goto out_close_devices;
1478
1479         error = xfs_init_percpu_counters(mp);
1480         if (error)
1481                 goto out_destroy_workqueues;
1482
1483         error = xfs_inodegc_init_percpu(mp);
1484         if (error)
1485                 goto out_destroy_counters;
1486
1487         /*
1488          * All percpu data structures requiring cleanup when a cpu goes offline
1489          * must be allocated before adding this @mp to the cpu-dead handler's
1490          * mount list.
1491          */
1492         xfs_mount_list_add(mp);
1493
1494         /* Allocate stats memory before we do operations that might use it */
1495         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1496         if (!mp->m_stats.xs_stats) {
1497                 error = -ENOMEM;
1498                 goto out_destroy_inodegc;
1499         }
1500
1501         error = xfs_readsb(mp, flags);
1502         if (error)
1503                 goto out_free_stats;
1504
1505         error = xfs_finish_flags(mp);
1506         if (error)
1507                 goto out_free_sb;
1508
1509         error = xfs_setup_devices(mp);
1510         if (error)
1511                 goto out_free_sb;
1512
1513         /* V4 support is undergoing deprecation. */
1514         if (!xfs_has_crc(mp)) {
1515 #ifdef CONFIG_XFS_SUPPORT_V4
1516                 xfs_warn_once(mp,
1517         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1518 #else
1519                 xfs_warn(mp,
1520         "Deprecated V4 format (crc=0) not supported by kernel.");
1521                 error = -EINVAL;
1522                 goto out_free_sb;
1523 #endif
1524         }
1525
1526         /* Filesystem claims it needs repair, so refuse the mount. */
1527         if (xfs_has_needsrepair(mp)) {
1528                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1529                 error = -EFSCORRUPTED;
1530                 goto out_free_sb;
1531         }
1532
1533         /*
1534          * Don't touch the filesystem if a user tool thinks it owns the primary
1535          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1536          * we don't check them at all.
1537          */
1538         if (mp->m_sb.sb_inprogress) {
1539                 xfs_warn(mp, "Offline file system operation in progress!");
1540                 error = -EFSCORRUPTED;
1541                 goto out_free_sb;
1542         }
1543
1544         /*
1545          * Until this is fixed only page-sized or smaller data blocks work.
1546          */
1547         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1548                 xfs_warn(mp,
1549                 "File system with blocksize %d bytes. "
1550                 "Only pagesize (%ld) or less will currently work.",
1551                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1552                 error = -ENOSYS;
1553                 goto out_free_sb;
1554         }
1555
1556         /* Ensure this filesystem fits in the page cache limits */
1557         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1558             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1559                 xfs_warn(mp,
1560                 "file system too large to be mounted on this system.");
1561                 error = -EFBIG;
1562                 goto out_free_sb;
1563         }
1564
1565         /*
1566          * XFS block mappings use 54 bits to store the logical block offset.
1567          * This should suffice to handle the maximum file size that the VFS
1568          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1569          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1570          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1571          * to check this assertion.
1572          *
1573          * Avoid integer overflow by comparing the maximum bmbt offset to the
1574          * maximum pagecache offset in units of fs blocks.
1575          */
1576         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1577                 xfs_warn(mp,
1578 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1579                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1580                          XFS_MAX_FILEOFF);
1581                 error = -EINVAL;
1582                 goto out_free_sb;
1583         }
1584
1585         error = xfs_filestream_mount(mp);
1586         if (error)
1587                 goto out_free_sb;
1588
1589         /*
1590          * we must configure the block size in the superblock before we run the
1591          * full mount process as the mount process can lookup and cache inodes.
1592          */
1593         sb->s_magic = XFS_SUPER_MAGIC;
1594         sb->s_blocksize = mp->m_sb.sb_blocksize;
1595         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1596         sb->s_maxbytes = MAX_LFS_FILESIZE;
1597         sb->s_max_links = XFS_MAXLINK;
1598         sb->s_time_gran = 1;
1599         if (xfs_has_bigtime(mp)) {
1600                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1601                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1602         } else {
1603                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1604                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1605         }
1606         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1607         sb->s_iflags |= SB_I_CGROUPWB;
1608
1609         set_posix_acl_flag(sb);
1610
1611         /* version 5 superblocks support inode version counters. */
1612         if (xfs_has_crc(mp))
1613                 sb->s_flags |= SB_I_VERSION;
1614
1615         if (xfs_has_dax_always(mp)) {
1616                 error = xfs_setup_dax_always(mp);
1617                 if (error)
1618                         goto out_filestream_unmount;
1619         }
1620
1621         if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1622                 xfs_warn(mp,
1623         "mounting with \"discard\" option, but the device does not support discard");
1624                 mp->m_features &= ~XFS_FEAT_DISCARD;
1625         }
1626
1627         if (xfs_has_reflink(mp)) {
1628                 if (mp->m_sb.sb_rblocks) {
1629                         xfs_alert(mp,
1630         "reflink not compatible with realtime device!");
1631                         error = -EINVAL;
1632                         goto out_filestream_unmount;
1633                 }
1634
1635                 if (xfs_globals.always_cow) {
1636                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1637                         mp->m_always_cow = true;
1638                 }
1639         }
1640
1641         if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1642                 xfs_alert(mp,
1643         "reverse mapping btree not compatible with realtime device!");
1644                 error = -EINVAL;
1645                 goto out_filestream_unmount;
1646         }
1647
1648         if (xfs_has_large_extent_counts(mp))
1649                 xfs_warn(mp,
1650         "EXPERIMENTAL Large extent counts feature in use. Use at your own risk!");
1651
1652         error = xfs_mountfs(mp);
1653         if (error)
1654                 goto out_filestream_unmount;
1655
1656         root = igrab(VFS_I(mp->m_rootip));
1657         if (!root) {
1658                 error = -ENOENT;
1659                 goto out_unmount;
1660         }
1661         sb->s_root = d_make_root(root);
1662         if (!sb->s_root) {
1663                 error = -ENOMEM;
1664                 goto out_unmount;
1665         }
1666
1667         return 0;
1668
1669  out_filestream_unmount:
1670         xfs_filestream_unmount(mp);
1671  out_free_sb:
1672         xfs_freesb(mp);
1673  out_free_stats:
1674         free_percpu(mp->m_stats.xs_stats);
1675  out_destroy_inodegc:
1676         xfs_mount_list_del(mp);
1677         xfs_inodegc_free_percpu(mp);
1678  out_destroy_counters:
1679         xfs_destroy_percpu_counters(mp);
1680  out_destroy_workqueues:
1681         xfs_destroy_mount_workqueues(mp);
1682  out_close_devices:
1683         xfs_close_devices(mp);
1684  out_free_names:
1685         sb->s_fs_info = NULL;
1686         xfs_mount_free(mp);
1687         return error;
1688
1689  out_unmount:
1690         xfs_filestream_unmount(mp);
1691         xfs_unmountfs(mp);
1692         goto out_free_sb;
1693 }
1694
1695 static int
1696 xfs_fs_get_tree(
1697         struct fs_context       *fc)
1698 {
1699         return get_tree_bdev(fc, xfs_fs_fill_super);
1700 }
1701
1702 static int
1703 xfs_remount_rw(
1704         struct xfs_mount        *mp)
1705 {
1706         struct xfs_sb           *sbp = &mp->m_sb;
1707         int error;
1708
1709         if (xfs_has_norecovery(mp)) {
1710                 xfs_warn(mp,
1711                         "ro->rw transition prohibited on norecovery mount");
1712                 return -EINVAL;
1713         }
1714
1715         if (xfs_sb_is_v5(sbp) &&
1716             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1717                 xfs_warn(mp,
1718         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1719                         (sbp->sb_features_ro_compat &
1720                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1721                 return -EINVAL;
1722         }
1723
1724         clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1725
1726         /*
1727          * If this is the first remount to writeable state we might have some
1728          * superblock changes to update.
1729          */
1730         if (mp->m_update_sb) {
1731                 error = xfs_sync_sb(mp, false);
1732                 if (error) {
1733                         xfs_warn(mp, "failed to write sb changes");
1734                         return error;
1735                 }
1736                 mp->m_update_sb = false;
1737         }
1738
1739         /*
1740          * Fill out the reserve pool if it is empty. Use the stashed value if
1741          * it is non-zero, otherwise go with the default.
1742          */
1743         xfs_restore_resvblks(mp);
1744         xfs_log_work_queue(mp);
1745         xfs_blockgc_start(mp);
1746
1747         /* Create the per-AG metadata reservation pool .*/
1748         error = xfs_fs_reserve_ag_blocks(mp);
1749         if (error && error != -ENOSPC)
1750                 return error;
1751
1752         /* Re-enable the background inode inactivation worker. */
1753         xfs_inodegc_start(mp);
1754
1755         return 0;
1756 }
1757
1758 static int
1759 xfs_remount_ro(
1760         struct xfs_mount        *mp)
1761 {
1762         struct xfs_icwalk       icw = {
1763                 .icw_flags      = XFS_ICWALK_FLAG_SYNC,
1764         };
1765         int                     error;
1766
1767         /* Flush all the dirty data to disk. */
1768         error = sync_filesystem(mp->m_super);
1769         if (error)
1770                 return error;
1771
1772         /*
1773          * Cancel background eofb scanning so it cannot race with the final
1774          * log force+buftarg wait and deadlock the remount.
1775          */
1776         xfs_blockgc_stop(mp);
1777
1778         /*
1779          * Clear out all remaining COW staging extents and speculative post-EOF
1780          * preallocations so that we don't leave inodes requiring inactivation
1781          * cleanups during reclaim on a read-only mount.  We must process every
1782          * cached inode, so this requires a synchronous cache scan.
1783          */
1784         error = xfs_blockgc_free_space(mp, &icw);
1785         if (error) {
1786                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1787                 return error;
1788         }
1789
1790         /*
1791          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1792          * flushed all pending inodegc work when it sync'd the filesystem.
1793          * The VFS holds s_umount, so we know that inodes cannot enter
1794          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1795          * we send inodes straight to reclaim, so no inodes will be queued.
1796          */
1797         xfs_inodegc_stop(mp);
1798
1799         /* Free the per-AG metadata reservation pool. */
1800         error = xfs_fs_unreserve_ag_blocks(mp);
1801         if (error) {
1802                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1803                 return error;
1804         }
1805
1806         /*
1807          * Before we sync the metadata, we need to free up the reserve block
1808          * pool so that the used block count in the superblock on disk is
1809          * correct at the end of the remount. Stash the current* reserve pool
1810          * size so that if we get remounted rw, we can return it to the same
1811          * size.
1812          */
1813         xfs_save_resvblks(mp);
1814
1815         xfs_log_clean(mp);
1816         set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1817
1818         return 0;
1819 }
1820
1821 /*
1822  * Logically we would return an error here to prevent users from believing
1823  * they might have changed mount options using remount which can't be changed.
1824  *
1825  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1826  * arguments in some cases so we can't blindly reject options, but have to
1827  * check for each specified option if it actually differs from the currently
1828  * set option and only reject it if that's the case.
1829  *
1830  * Until that is implemented we return success for every remount request, and
1831  * silently ignore all options that we can't actually change.
1832  */
1833 static int
1834 xfs_fs_reconfigure(
1835         struct fs_context *fc)
1836 {
1837         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1838         struct xfs_mount        *new_mp = fc->s_fs_info;
1839         int                     flags = fc->sb_flags;
1840         int                     error;
1841
1842         /* version 5 superblocks always support version counters. */
1843         if (xfs_has_crc(mp))
1844                 fc->sb_flags |= SB_I_VERSION;
1845
1846         error = xfs_fs_validate_params(new_mp);
1847         if (error)
1848                 return error;
1849
1850         /* inode32 -> inode64 */
1851         if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1852                 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1853                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1854         }
1855
1856         /* inode64 -> inode32 */
1857         if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1858                 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1859                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1860         }
1861
1862         /* ro -> rw */
1863         if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1864                 error = xfs_remount_rw(mp);
1865                 if (error)
1866                         return error;
1867         }
1868
1869         /* rw -> ro */
1870         if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1871                 error = xfs_remount_ro(mp);
1872                 if (error)
1873                         return error;
1874         }
1875
1876         return 0;
1877 }
1878
1879 static void xfs_fs_free(
1880         struct fs_context       *fc)
1881 {
1882         struct xfs_mount        *mp = fc->s_fs_info;
1883
1884         /*
1885          * mp is stored in the fs_context when it is initialized.
1886          * mp is transferred to the superblock on a successful mount,
1887          * but if an error occurs before the transfer we have to free
1888          * it here.
1889          */
1890         if (mp)
1891                 xfs_mount_free(mp);
1892 }
1893
1894 static const struct fs_context_operations xfs_context_ops = {
1895         .parse_param = xfs_fs_parse_param,
1896         .get_tree    = xfs_fs_get_tree,
1897         .reconfigure = xfs_fs_reconfigure,
1898         .free        = xfs_fs_free,
1899 };
1900
1901 static int xfs_init_fs_context(
1902         struct fs_context       *fc)
1903 {
1904         struct xfs_mount        *mp;
1905
1906         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1907         if (!mp)
1908                 return -ENOMEM;
1909
1910         spin_lock_init(&mp->m_sb_lock);
1911         spin_lock_init(&mp->m_agirotor_lock);
1912         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1913         spin_lock_init(&mp->m_perag_lock);
1914         mutex_init(&mp->m_growlock);
1915         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1916         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1917         mp->m_kobj.kobject.kset = xfs_kset;
1918         /*
1919          * We don't create the finobt per-ag space reservation until after log
1920          * recovery, so we must set this to true so that an ifree transaction
1921          * started during log recovery will not depend on space reservations
1922          * for finobt expansion.
1923          */
1924         mp->m_finobt_nores = true;
1925
1926         /*
1927          * These can be overridden by the mount option parsing.
1928          */
1929         mp->m_logbufs = -1;
1930         mp->m_logbsize = -1;
1931         mp->m_allocsize_log = 16; /* 64k */
1932
1933         /*
1934          * Copy binary VFS mount flags we are interested in.
1935          */
1936         if (fc->sb_flags & SB_RDONLY)
1937                 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1938         if (fc->sb_flags & SB_DIRSYNC)
1939                 mp->m_features |= XFS_FEAT_DIRSYNC;
1940         if (fc->sb_flags & SB_SYNCHRONOUS)
1941                 mp->m_features |= XFS_FEAT_WSYNC;
1942
1943         fc->s_fs_info = mp;
1944         fc->ops = &xfs_context_ops;
1945
1946         return 0;
1947 }
1948
1949 static struct file_system_type xfs_fs_type = {
1950         .owner                  = THIS_MODULE,
1951         .name                   = "xfs",
1952         .init_fs_context        = xfs_init_fs_context,
1953         .parameters             = xfs_fs_parameters,
1954         .kill_sb                = kill_block_super,
1955         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1956 };
1957 MODULE_ALIAS_FS("xfs");
1958
1959 STATIC int __init
1960 xfs_init_caches(void)
1961 {
1962         int             error;
1963
1964         xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
1965                                                 sizeof(struct xlog_ticket),
1966                                                 0, 0, NULL);
1967         if (!xfs_log_ticket_cache)
1968                 goto out;
1969
1970         error = xfs_btree_init_cur_caches();
1971         if (error)
1972                 goto out_destroy_log_ticket_cache;
1973
1974         error = xfs_defer_init_item_caches();
1975         if (error)
1976                 goto out_destroy_btree_cur_cache;
1977
1978         xfs_da_state_cache = kmem_cache_create("xfs_da_state",
1979                                               sizeof(struct xfs_da_state),
1980                                               0, 0, NULL);
1981         if (!xfs_da_state_cache)
1982                 goto out_destroy_defer_item_cache;
1983
1984         xfs_ifork_cache = kmem_cache_create("xfs_ifork",
1985                                            sizeof(struct xfs_ifork),
1986                                            0, 0, NULL);
1987         if (!xfs_ifork_cache)
1988                 goto out_destroy_da_state_cache;
1989
1990         xfs_trans_cache = kmem_cache_create("xfs_trans",
1991                                            sizeof(struct xfs_trans),
1992                                            0, 0, NULL);
1993         if (!xfs_trans_cache)
1994                 goto out_destroy_ifork_cache;
1995
1996
1997         /*
1998          * The size of the cache-allocated buf log item is the maximum
1999          * size possible under XFS.  This wastes a little bit of memory,
2000          * but it is much faster.
2001          */
2002         xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2003                                               sizeof(struct xfs_buf_log_item),
2004                                               0, 0, NULL);
2005         if (!xfs_buf_item_cache)
2006                 goto out_destroy_trans_cache;
2007
2008         xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2009                                         (sizeof(struct xfs_efd_log_item) +
2010                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
2011                                         sizeof(struct xfs_extent)),
2012                                         0, 0, NULL);
2013         if (!xfs_efd_cache)
2014                 goto out_destroy_buf_item_cache;
2015
2016         xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2017                                          (sizeof(struct xfs_efi_log_item) +
2018                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2019                                          sizeof(struct xfs_extent)),
2020                                          0, 0, NULL);
2021         if (!xfs_efi_cache)
2022                 goto out_destroy_efd_cache;
2023
2024         xfs_inode_cache = kmem_cache_create("xfs_inode",
2025                                            sizeof(struct xfs_inode), 0,
2026                                            (SLAB_HWCACHE_ALIGN |
2027                                             SLAB_RECLAIM_ACCOUNT |
2028                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2029                                            xfs_fs_inode_init_once);
2030         if (!xfs_inode_cache)
2031                 goto out_destroy_efi_cache;
2032
2033         xfs_ili_cache = kmem_cache_create("xfs_ili",
2034                                          sizeof(struct xfs_inode_log_item), 0,
2035                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2036                                          NULL);
2037         if (!xfs_ili_cache)
2038                 goto out_destroy_inode_cache;
2039
2040         xfs_icreate_cache = kmem_cache_create("xfs_icr",
2041                                              sizeof(struct xfs_icreate_item),
2042                                              0, 0, NULL);
2043         if (!xfs_icreate_cache)
2044                 goto out_destroy_ili_cache;
2045
2046         xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2047                                          sizeof(struct xfs_rud_log_item),
2048                                          0, 0, NULL);
2049         if (!xfs_rud_cache)
2050                 goto out_destroy_icreate_cache;
2051
2052         xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2053                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2054                         0, 0, NULL);
2055         if (!xfs_rui_cache)
2056                 goto out_destroy_rud_cache;
2057
2058         xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2059                                          sizeof(struct xfs_cud_log_item),
2060                                          0, 0, NULL);
2061         if (!xfs_cud_cache)
2062                 goto out_destroy_rui_cache;
2063
2064         xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2065                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2066                         0, 0, NULL);
2067         if (!xfs_cui_cache)
2068                 goto out_destroy_cud_cache;
2069
2070         xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2071                                          sizeof(struct xfs_bud_log_item),
2072                                          0, 0, NULL);
2073         if (!xfs_bud_cache)
2074                 goto out_destroy_cui_cache;
2075
2076         xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2077                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2078                         0, 0, NULL);
2079         if (!xfs_bui_cache)
2080                 goto out_destroy_bud_cache;
2081
2082         return 0;
2083
2084  out_destroy_bud_cache:
2085         kmem_cache_destroy(xfs_bud_cache);
2086  out_destroy_cui_cache:
2087         kmem_cache_destroy(xfs_cui_cache);
2088  out_destroy_cud_cache:
2089         kmem_cache_destroy(xfs_cud_cache);
2090  out_destroy_rui_cache:
2091         kmem_cache_destroy(xfs_rui_cache);
2092  out_destroy_rud_cache:
2093         kmem_cache_destroy(xfs_rud_cache);
2094  out_destroy_icreate_cache:
2095         kmem_cache_destroy(xfs_icreate_cache);
2096  out_destroy_ili_cache:
2097         kmem_cache_destroy(xfs_ili_cache);
2098  out_destroy_inode_cache:
2099         kmem_cache_destroy(xfs_inode_cache);
2100  out_destroy_efi_cache:
2101         kmem_cache_destroy(xfs_efi_cache);
2102  out_destroy_efd_cache:
2103         kmem_cache_destroy(xfs_efd_cache);
2104  out_destroy_buf_item_cache:
2105         kmem_cache_destroy(xfs_buf_item_cache);
2106  out_destroy_trans_cache:
2107         kmem_cache_destroy(xfs_trans_cache);
2108  out_destroy_ifork_cache:
2109         kmem_cache_destroy(xfs_ifork_cache);
2110  out_destroy_da_state_cache:
2111         kmem_cache_destroy(xfs_da_state_cache);
2112  out_destroy_defer_item_cache:
2113         xfs_defer_destroy_item_caches();
2114  out_destroy_btree_cur_cache:
2115         xfs_btree_destroy_cur_caches();
2116  out_destroy_log_ticket_cache:
2117         kmem_cache_destroy(xfs_log_ticket_cache);
2118  out:
2119         return -ENOMEM;
2120 }
2121
2122 STATIC void
2123 xfs_destroy_caches(void)
2124 {
2125         /*
2126          * Make sure all delayed rcu free are flushed before we
2127          * destroy caches.
2128          */
2129         rcu_barrier();
2130         kmem_cache_destroy(xfs_bui_cache);
2131         kmem_cache_destroy(xfs_bud_cache);
2132         kmem_cache_destroy(xfs_cui_cache);
2133         kmem_cache_destroy(xfs_cud_cache);
2134         kmem_cache_destroy(xfs_rui_cache);
2135         kmem_cache_destroy(xfs_rud_cache);
2136         kmem_cache_destroy(xfs_icreate_cache);
2137         kmem_cache_destroy(xfs_ili_cache);
2138         kmem_cache_destroy(xfs_inode_cache);
2139         kmem_cache_destroy(xfs_efi_cache);
2140         kmem_cache_destroy(xfs_efd_cache);
2141         kmem_cache_destroy(xfs_buf_item_cache);
2142         kmem_cache_destroy(xfs_trans_cache);
2143         kmem_cache_destroy(xfs_ifork_cache);
2144         kmem_cache_destroy(xfs_da_state_cache);
2145         xfs_defer_destroy_item_caches();
2146         xfs_btree_destroy_cur_caches();
2147         kmem_cache_destroy(xfs_log_ticket_cache);
2148 }
2149
2150 STATIC int __init
2151 xfs_init_workqueues(void)
2152 {
2153         /*
2154          * The allocation workqueue can be used in memory reclaim situations
2155          * (writepage path), and parallelism is only limited by the number of
2156          * AGs in all the filesystems mounted. Hence use the default large
2157          * max_active value for this workqueue.
2158          */
2159         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2160                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2161         if (!xfs_alloc_wq)
2162                 return -ENOMEM;
2163
2164         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2165                         0);
2166         if (!xfs_discard_wq)
2167                 goto out_free_alloc_wq;
2168
2169         return 0;
2170 out_free_alloc_wq:
2171         destroy_workqueue(xfs_alloc_wq);
2172         return -ENOMEM;
2173 }
2174
2175 STATIC void
2176 xfs_destroy_workqueues(void)
2177 {
2178         destroy_workqueue(xfs_discard_wq);
2179         destroy_workqueue(xfs_alloc_wq);
2180 }
2181
2182 #ifdef CONFIG_HOTPLUG_CPU
2183 static int
2184 xfs_cpu_dead(
2185         unsigned int            cpu)
2186 {
2187         struct xfs_mount        *mp, *n;
2188
2189         spin_lock(&xfs_mount_list_lock);
2190         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2191                 spin_unlock(&xfs_mount_list_lock);
2192                 xfs_inodegc_cpu_dead(mp, cpu);
2193                 spin_lock(&xfs_mount_list_lock);
2194         }
2195         spin_unlock(&xfs_mount_list_lock);
2196         return 0;
2197 }
2198
2199 static int __init
2200 xfs_cpu_hotplug_init(void)
2201 {
2202         int     error;
2203
2204         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2205                         xfs_cpu_dead);
2206         if (error < 0)
2207                 xfs_alert(NULL,
2208 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2209                         error);
2210         return error;
2211 }
2212
2213 static void
2214 xfs_cpu_hotplug_destroy(void)
2215 {
2216         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2217 }
2218
2219 #else /* !CONFIG_HOTPLUG_CPU */
2220 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2221 static inline void xfs_cpu_hotplug_destroy(void) {}
2222 #endif
2223
2224 STATIC int __init
2225 init_xfs_fs(void)
2226 {
2227         int                     error;
2228
2229         xfs_check_ondisk_structs();
2230
2231         printk(KERN_INFO XFS_VERSION_STRING " with "
2232                          XFS_BUILD_OPTIONS " enabled\n");
2233
2234         xfs_dir_startup();
2235
2236         error = xfs_cpu_hotplug_init();
2237         if (error)
2238                 goto out;
2239
2240         error = xfs_init_caches();
2241         if (error)
2242                 goto out_destroy_hp;
2243
2244         error = xfs_init_workqueues();
2245         if (error)
2246                 goto out_destroy_caches;
2247
2248         error = xfs_mru_cache_init();
2249         if (error)
2250                 goto out_destroy_wq;
2251
2252         error = xfs_buf_init();
2253         if (error)
2254                 goto out_mru_cache_uninit;
2255
2256         error = xfs_init_procfs();
2257         if (error)
2258                 goto out_buf_terminate;
2259
2260         error = xfs_sysctl_register();
2261         if (error)
2262                 goto out_cleanup_procfs;
2263
2264         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2265         if (!xfs_kset) {
2266                 error = -ENOMEM;
2267                 goto out_sysctl_unregister;
2268         }
2269
2270         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2271
2272         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2273         if (!xfsstats.xs_stats) {
2274                 error = -ENOMEM;
2275                 goto out_kset_unregister;
2276         }
2277
2278         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2279                                "stats");
2280         if (error)
2281                 goto out_free_stats;
2282
2283 #ifdef DEBUG
2284         xfs_dbg_kobj.kobject.kset = xfs_kset;
2285         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2286         if (error)
2287                 goto out_remove_stats_kobj;
2288 #endif
2289
2290         error = xfs_qm_init();
2291         if (error)
2292                 goto out_remove_dbg_kobj;
2293
2294         error = register_filesystem(&xfs_fs_type);
2295         if (error)
2296                 goto out_qm_exit;
2297         return 0;
2298
2299  out_qm_exit:
2300         xfs_qm_exit();
2301  out_remove_dbg_kobj:
2302 #ifdef DEBUG
2303         xfs_sysfs_del(&xfs_dbg_kobj);
2304  out_remove_stats_kobj:
2305 #endif
2306         xfs_sysfs_del(&xfsstats.xs_kobj);
2307  out_free_stats:
2308         free_percpu(xfsstats.xs_stats);
2309  out_kset_unregister:
2310         kset_unregister(xfs_kset);
2311  out_sysctl_unregister:
2312         xfs_sysctl_unregister();
2313  out_cleanup_procfs:
2314         xfs_cleanup_procfs();
2315  out_buf_terminate:
2316         xfs_buf_terminate();
2317  out_mru_cache_uninit:
2318         xfs_mru_cache_uninit();
2319  out_destroy_wq:
2320         xfs_destroy_workqueues();
2321  out_destroy_caches:
2322         xfs_destroy_caches();
2323  out_destroy_hp:
2324         xfs_cpu_hotplug_destroy();
2325  out:
2326         return error;
2327 }
2328
2329 STATIC void __exit
2330 exit_xfs_fs(void)
2331 {
2332         xfs_qm_exit();
2333         unregister_filesystem(&xfs_fs_type);
2334 #ifdef DEBUG
2335         xfs_sysfs_del(&xfs_dbg_kobj);
2336 #endif
2337         xfs_sysfs_del(&xfsstats.xs_kobj);
2338         free_percpu(xfsstats.xs_stats);
2339         kset_unregister(xfs_kset);
2340         xfs_sysctl_unregister();
2341         xfs_cleanup_procfs();
2342         xfs_buf_terminate();
2343         xfs_mru_cache_uninit();
2344         xfs_destroy_workqueues();
2345         xfs_destroy_caches();
2346         xfs_uuid_table_free();
2347         xfs_cpu_hotplug_destroy();
2348 }
2349
2350 module_init(init_xfs_fs);
2351 module_exit(exit_xfs_fs);
2352
2353 MODULE_AUTHOR("Silicon Graphics, Inc.");
2354 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2355 MODULE_LICENSE("GPL");