fs/xfs/xfs_super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6
   7 #include "xfs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_sb.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_btree.h"
  16 #include "xfs_bmap.h"
  17 #include "xfs_alloc.h"
  18 #include "xfs_fsops.h"
  19 #include "xfs_trans.h"
  20 #include "xfs_buf_item.h"
  21 #include "xfs_log.h"
  22 #include "xfs_log_priv.h"
  23 #include "xfs_dir2.h"
  24 #include "xfs_extfree_item.h"
  25 #include "xfs_mru_cache.h"
  26 #include "xfs_inode_item.h"
  27 #include "xfs_icache.h"
  28 #include "xfs_trace.h"
  29 #include "xfs_icreate_item.h"
  30 #include "xfs_filestream.h"
  31 #include "xfs_quota.h"
  32 #include "xfs_sysfs.h"
  33 #include "xfs_ondisk.h"
  34 #include "xfs_rmap_item.h"
  35 #include "xfs_refcount_item.h"
  36 #include "xfs_bmap_item.h"
  37 #include "xfs_reflink.h"
  38 #include "xfs_pwork.h"
  39 #include "xfs_ag.h"
  40 #include "xfs_defer.h"
  41 #include "xfs_attr_item.h"
  42 #include "xfs_xattr.h"
  43 #include "xfs_iunlink_item.h"
  44
  45 #include <linux/magic.h>
  46 #include <linux/fs_context.h>
  47 #include <linux/fs_parser.h>
  48
  49 static const struct super_operations xfs_super_operations;
  50
  51 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  52 #ifdef DEBUG
  53 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  54 #endif
  55
  56 #ifdef CONFIG_HOTPLUG_CPU
  57 static LIST_HEAD(xfs_mount_list);
  58 static DEFINE_SPINLOCK(xfs_mount_list_lock);
  59
  60 static inline void xfs_mount_list_add(struct xfs_mount *mp)
  61 {
  62         spin_lock(&xfs_mount_list_lock);
  63         list_add(&mp->m_mount_list, &xfs_mount_list);
  64         spin_unlock(&xfs_mount_list_lock);
  65 }
  66
  67 static inline void xfs_mount_list_del(struct xfs_mount *mp)
  68 {
  69         spin_lock(&xfs_mount_list_lock);
  70         list_del(&mp->m_mount_list);
  71         spin_unlock(&xfs_mount_list_lock);
  72 }
  73 #else /* !CONFIG_HOTPLUG_CPU */
  74 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
  75 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
  76 #endif
  77
  78 enum xfs_dax_mode {
  79         XFS_DAX_INODE = 0,
  80         XFS_DAX_ALWAYS = 1,
  81         XFS_DAX_NEVER = 2,
  82 };
  83
  84 static void
  85 xfs_mount_set_dax_mode(
  86         struct xfs_mount        *mp,
  87         enum xfs_dax_mode       mode)
  88 {
  89         switch (mode) {
  90         case XFS_DAX_INODE:
  91                 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
  92                 break;
  93         case XFS_DAX_ALWAYS:
  94                 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
  95                 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
  96                 break;
  97         case XFS_DAX_NEVER:
  98                 mp->m_features |= XFS_FEAT_DAX_NEVER;
  99                 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
 100                 break;
 101         }
 102 }
 103
 104 static const struct constant_table dax_param_enums[] = {
 105         {"inode",       XFS_DAX_INODE },
 106         {"always",      XFS_DAX_ALWAYS },
 107         {"never",       XFS_DAX_NEVER },
 108         {}
 109 };
 110
 111 /*
 112  * Table driven mount option parser.
 113  */
 114 enum {
 115         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
 116         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
 117         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
 118         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
 119         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
 120         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
 121         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
 122         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
 123         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
 124 };
 125
 126 static const struct fs_parameter_spec xfs_fs_parameters[] = {
 127         fsparam_u32("logbufs",          Opt_logbufs),
 128         fsparam_string("logbsize",      Opt_logbsize),
 129         fsparam_string("logdev",        Opt_logdev),
 130         fsparam_string("rtdev",         Opt_rtdev),
 131         fsparam_flag("wsync",           Opt_wsync),
 132         fsparam_flag("noalign",         Opt_noalign),
 133         fsparam_flag("swalloc",         Opt_swalloc),
 134         fsparam_u32("sunit",            Opt_sunit),
 135         fsparam_u32("swidth",           Opt_swidth),
 136         fsparam_flag("nouuid",          Opt_nouuid),
 137         fsparam_flag("grpid",           Opt_grpid),
 138         fsparam_flag("nogrpid",         Opt_nogrpid),
 139         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 140         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 141         fsparam_string("allocsize",     Opt_allocsize),
 142         fsparam_flag("norecovery",      Opt_norecovery),
 143         fsparam_flag("inode64",         Opt_inode64),
 144         fsparam_flag("inode32",         Opt_inode32),
 145         fsparam_flag("ikeep",           Opt_ikeep),
 146         fsparam_flag("noikeep",         Opt_noikeep),
 147         fsparam_flag("largeio",         Opt_largeio),
 148         fsparam_flag("nolargeio",       Opt_nolargeio),
 149         fsparam_flag("attr2",           Opt_attr2),
 150         fsparam_flag("noattr2",         Opt_noattr2),
 151         fsparam_flag("filestreams",     Opt_filestreams),
 152         fsparam_flag("quota",           Opt_quota),
 153         fsparam_flag("noquota",         Opt_noquota),
 154         fsparam_flag("usrquota",        Opt_usrquota),
 155         fsparam_flag("grpquota",        Opt_grpquota),
 156         fsparam_flag("prjquota",        Opt_prjquota),
 157         fsparam_flag("uquota",          Opt_uquota),
 158         fsparam_flag("gquota",          Opt_gquota),
 159         fsparam_flag("pquota",          Opt_pquota),
 160         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 161         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 162         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 163         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 164         fsparam_flag("discard",         Opt_discard),
 165         fsparam_flag("nodiscard",       Opt_nodiscard),
 166         fsparam_flag("dax",             Opt_dax),
 167         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 168         {}
 169 };
 170
 171 struct proc_xfs_info {
 172         uint64_t        flag;
 173         char            *str;
 174 };
 175
 176 static int
 177 xfs_fs_show_options(
 178         struct seq_file         *m,
 179         struct dentry           *root)
 180 {
 181         static struct proc_xfs_info xfs_info_set[] = {
 182                 /* the few simple ones we can get from the mount struct */
 183                 { XFS_FEAT_IKEEP,               ",ikeep" },
 184                 { XFS_FEAT_WSYNC,               ",wsync" },
 185                 { XFS_FEAT_NOALIGN,             ",noalign" },
 186                 { XFS_FEAT_SWALLOC,             ",swalloc" },
 187                 { XFS_FEAT_NOUUID,              ",nouuid" },
 188                 { XFS_FEAT_NORECOVERY,          ",norecovery" },
 189                 { XFS_FEAT_ATTR2,               ",attr2" },
 190                 { XFS_FEAT_FILESTREAMS,         ",filestreams" },
 191                 { XFS_FEAT_GRPID,               ",grpid" },
 192                 { XFS_FEAT_DISCARD,             ",discard" },
 193                 { XFS_FEAT_LARGE_IOSIZE,        ",largeio" },
 194                 { XFS_FEAT_DAX_ALWAYS,          ",dax=always" },
 195                 { XFS_FEAT_DAX_NEVER,           ",dax=never" },
 196                 { 0, NULL }
 197         };
 198         struct xfs_mount        *mp = XFS_M(root->d_sb);
 199         struct proc_xfs_info    *xfs_infop;
 200
 201         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 202                 if (mp->m_features & xfs_infop->flag)
 203                         seq_puts(m, xfs_infop->str);
 204         }
 205
 206         seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
 207
 208         if (xfs_has_allocsize(mp))
 209                 seq_printf(m, ",allocsize=%dk",
 210                            (1 << mp->m_allocsize_log) >> 10);
 211
 212         if (mp->m_logbufs > 0)
 213                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 214         if (mp->m_logbsize > 0)
 215                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 216
 217         if (mp->m_logname)
 218                 seq_show_option(m, "logdev", mp->m_logname);
 219         if (mp->m_rtname)
 220                 seq_show_option(m, "rtdev", mp->m_rtname);
 221
 222         if (mp->m_dalign > 0)
 223                 seq_printf(m, ",sunit=%d",
 224                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 225         if (mp->m_swidth > 0)
 226                 seq_printf(m, ",swidth=%d",
 227                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 228
 229         if (mp->m_qflags & XFS_UQUOTA_ENFD)
 230                 seq_puts(m, ",usrquota");
 231         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 232                 seq_puts(m, ",uqnoenforce");
 233
 234         if (mp->m_qflags & XFS_PQUOTA_ENFD)
 235                 seq_puts(m, ",prjquota");
 236         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
 237                 seq_puts(m, ",pqnoenforce");
 238
 239         if (mp->m_qflags & XFS_GQUOTA_ENFD)
 240                 seq_puts(m, ",grpquota");
 241         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
 242                 seq_puts(m, ",gqnoenforce");
 243
 244         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 245                 seq_puts(m, ",noquota");
 246
 247         return 0;
 248 }
 249
 250 static bool
 251 xfs_set_inode_alloc_perag(
 252         struct xfs_perag        *pag,
 253         xfs_ino_t               ino,
 254         xfs_agnumber_t          max_metadata)
 255 {
 256         if (!xfs_is_inode32(pag->pag_mount)) {
 257                 set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
 258                 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 259                 return false;
 260         }
 261
 262         if (ino > XFS_MAXINUMBER_32) {
 263                 clear_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
 264                 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 265                 return false;
 266         }
 267
 268         set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
 269         if (pag->pag_agno < max_metadata)
 270                 set_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 271         else
 272                 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 273         return true;
 274 }
 275
 276 /*
 277  * Set parameters for inode allocation heuristics, taking into account
 278  * filesystem size and inode32/inode64 mount options; i.e. specifically
 279  * whether or not XFS_FEAT_SMALL_INUMS is set.
 280  *
 281  * Inode allocation patterns are altered only if inode32 is requested
 282  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
 283  * If altered, XFS_OPSTATE_INODE32 is set as well.
 284  *
 285  * An agcount independent of that in the mount structure is provided
 286  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 287  * to the potentially higher ag count.
 288  *
 289  * Returns the maximum AG index which may contain inodes.
 290  */
 291 xfs_agnumber_t
 292 xfs_set_inode_alloc(
 293         struct xfs_mount *mp,
 294         xfs_agnumber_t  agcount)
 295 {
 296         xfs_agnumber_t  index;
 297         xfs_agnumber_t  maxagi = 0;
 298         xfs_sb_t        *sbp = &mp->m_sb;
 299         xfs_agnumber_t  max_metadata;
 300         xfs_agino_t     agino;
 301         xfs_ino_t       ino;
 302
 303         /*
 304          * Calculate how much should be reserved for inodes to meet
 305          * the max inode percentage.  Used only for inode32.
 306          */
 307         if (M_IGEO(mp)->maxicount) {
 308                 uint64_t        icount;
 309
 310                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 311                 do_div(icount, 100);
 312                 icount += sbp->sb_agblocks - 1;
 313                 do_div(icount, sbp->sb_agblocks);
 314                 max_metadata = icount;
 315         } else {
 316                 max_metadata = agcount;
 317         }
 318
 319         /* Get the last possible inode in the filesystem */
 320         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 321         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 322
 323         /*
 324          * If user asked for no more than 32-bit inodes, and the fs is
 325          * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
 326          * the allocator to accommodate the request.
 327          */
 328         if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
 329                 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 330         else
 331                 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 332
 333         for (index = 0; index < agcount; index++) {
 334                 struct xfs_perag        *pag;
 335
 336                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 337
 338                 pag = xfs_perag_get(mp, index);
 339                 if (xfs_set_inode_alloc_perag(pag, ino, max_metadata))
 340                         maxagi++;
 341                 xfs_perag_put(pag);
 342         }
 343
 344         return xfs_is_inode32(mp) ? maxagi : agcount;
 345 }
 346
 347 static int
 348 xfs_setup_dax_always(
 349         struct xfs_mount        *mp)
 350 {
 351         if (!mp->m_ddev_targp->bt_daxdev &&
 352             (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
 353                 xfs_alert(mp,
 354                         "DAX unsupported by block device. Turning off DAX.");
 355                 goto disable_dax;
 356         }
 357
 358         if (mp->m_super->s_blocksize != PAGE_SIZE) {
 359                 xfs_alert(mp,
 360                         "DAX not supported for blocksize. Turning off DAX.");
 361                 goto disable_dax;
 362         }
 363
 364         if (xfs_has_reflink(mp) &&
 365             bdev_is_partition(mp->m_ddev_targp->bt_bdev)) {
 366                 xfs_alert(mp,
 367                         "DAX and reflink cannot work with multi-partitions!");
 368                 return -EINVAL;
 369         }
 370
 371         xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
 372         return 0;
 373
 374 disable_dax:
 375         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
 376         return 0;
 377 }
 378
 379 STATIC int
 380 xfs_blkdev_get(
 381         xfs_mount_t             *mp,
 382         const char              *name,
 383         struct block_device     **bdevp)
 384 {
 385         int                     error = 0;
 386
 387         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 388                                     mp);
 389         if (IS_ERR(*bdevp)) {
 390                 error = PTR_ERR(*bdevp);
 391                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 392         }
 393
 394         return error;
 395 }
 396
 397 STATIC void
 398 xfs_blkdev_put(
 399         struct block_device     *bdev)
 400 {
 401         if (bdev)
 402                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 403 }
 404
 405 STATIC void
 406 xfs_close_devices(
 407         struct xfs_mount        *mp)
 408 {
 409         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 410                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 411
 412                 xfs_free_buftarg(mp->m_logdev_targp);
 413                 xfs_blkdev_put(logdev);
 414         }
 415         if (mp->m_rtdev_targp) {
 416                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 417
 418                 xfs_free_buftarg(mp->m_rtdev_targp);
 419                 xfs_blkdev_put(rtdev);
 420         }
 421         xfs_free_buftarg(mp->m_ddev_targp);
 422 }
 423
 424 /*
 425  * The file system configurations are:
 426  *      (1) device (partition) with data and internal log
 427  *      (2) logical volume with data and log subvolumes.
 428  *      (3) logical volume with data, log, and realtime subvolumes.
 429  *
 430  * We only have to handle opening the log and realtime volumes here if
 431  * they are present.  The data subvolume has already been opened by
 432  * get_sb_bdev() and is stored in sb->s_bdev.
 433  */
 434 STATIC int
 435 xfs_open_devices(
 436         struct xfs_mount        *mp)
 437 {
 438         struct block_device     *ddev = mp->m_super->s_bdev;
 439         struct block_device     *logdev = NULL, *rtdev = NULL;
 440         int                     error;
 441
 442         /*
 443          * Open real time and log devices - order is important.
 444          */
 445         if (mp->m_logname) {
 446                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 447                 if (error)
 448                         return error;
 449         }
 450
 451         if (mp->m_rtname) {
 452                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 453                 if (error)
 454                         goto out_close_logdev;
 455
 456                 if (rtdev == ddev || rtdev == logdev) {
 457                         xfs_warn(mp,
 458         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 459                         error = -EINVAL;
 460                         goto out_close_rtdev;
 461                 }
 462         }
 463
 464         /*
 465          * Setup xfs_mount buffer target pointers
 466          */
 467         error = -ENOMEM;
 468         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
 469         if (!mp->m_ddev_targp)
 470                 goto out_close_rtdev;
 471
 472         if (rtdev) {
 473                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
 474                 if (!mp->m_rtdev_targp)
 475                         goto out_free_ddev_targ;
 476         }
 477
 478         if (logdev && logdev != ddev) {
 479                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
 480                 if (!mp->m_logdev_targp)
 481                         goto out_free_rtdev_targ;
 482         } else {
 483                 mp->m_logdev_targp = mp->m_ddev_targp;
 484         }
 485
 486         return 0;
 487
 488  out_free_rtdev_targ:
 489         if (mp->m_rtdev_targp)
 490                 xfs_free_buftarg(mp->m_rtdev_targp);
 491  out_free_ddev_targ:
 492         xfs_free_buftarg(mp->m_ddev_targp);
 493  out_close_rtdev:
 494         xfs_blkdev_put(rtdev);
 495  out_close_logdev:
 496         if (logdev && logdev != ddev)
 497                 xfs_blkdev_put(logdev);
 498         return error;
 499 }
 500
 501 /*
 502  * Setup xfs_mount buffer target pointers based on superblock
 503  */
 504 STATIC int
 505 xfs_setup_devices(
 506         struct xfs_mount        *mp)
 507 {
 508         int                     error;
 509
 510         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 511         if (error)
 512                 return error;
 513
 514         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 515                 unsigned int    log_sector_size = BBSIZE;
 516
 517                 if (xfs_has_sector(mp))
 518                         log_sector_size = mp->m_sb.sb_logsectsize;
 519                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 520                                             log_sector_size);
 521                 if (error)
 522                         return error;
 523         }
 524         if (mp->m_rtdev_targp) {
 525                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 526                                             mp->m_sb.sb_sectsize);
 527                 if (error)
 528                         return error;
 529         }
 530
 531         return 0;
 532 }
 533
 534 STATIC int
 535 xfs_init_mount_workqueues(
 536         struct xfs_mount        *mp)
 537 {
 538         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 539                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 540                         1, mp->m_super->s_id);
 541         if (!mp->m_buf_workqueue)
 542                 goto out;
 543
 544         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 545                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 546                         0, mp->m_super->s_id);
 547         if (!mp->m_unwritten_workqueue)
 548                 goto out_destroy_buf;
 549
 550         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 551                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 552                         0, mp->m_super->s_id);
 553         if (!mp->m_reclaim_workqueue)
 554                 goto out_destroy_unwritten;
 555
 556         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
 557                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
 558                         0, mp->m_super->s_id);
 559         if (!mp->m_blockgc_wq)
 560                 goto out_destroy_reclaim;
 561
 562         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
 563                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 564                         1, mp->m_super->s_id);
 565         if (!mp->m_inodegc_wq)
 566                 goto out_destroy_blockgc;
 567
 568         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 569                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 570         if (!mp->m_sync_workqueue)
 571                 goto out_destroy_inodegc;
 572
 573         return 0;
 574
 575 out_destroy_inodegc:
 576         destroy_workqueue(mp->m_inodegc_wq);
 577 out_destroy_blockgc:
 578         destroy_workqueue(mp->m_blockgc_wq);
 579 out_destroy_reclaim:
 580         destroy_workqueue(mp->m_reclaim_workqueue);
 581 out_destroy_unwritten:
 582         destroy_workqueue(mp->m_unwritten_workqueue);
 583 out_destroy_buf:
 584         destroy_workqueue(mp->m_buf_workqueue);
 585 out:
 586         return -ENOMEM;
 587 }
 588
 589 STATIC void
 590 xfs_destroy_mount_workqueues(
 591         struct xfs_mount        *mp)
 592 {
 593         destroy_workqueue(mp->m_sync_workqueue);
 594         destroy_workqueue(mp->m_blockgc_wq);
 595         destroy_workqueue(mp->m_inodegc_wq);
 596         destroy_workqueue(mp->m_reclaim_workqueue);
 597         destroy_workqueue(mp->m_unwritten_workqueue);
 598         destroy_workqueue(mp->m_buf_workqueue);
 599 }
 600
 601 static void
 602 xfs_flush_inodes_worker(
 603         struct work_struct      *work)
 604 {
 605         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 606                                                    m_flush_inodes_work);
 607         struct super_block      *sb = mp->m_super;
 608
 609         if (down_read_trylock(&sb->s_umount)) {
 610                 sync_inodes_sb(sb);
 611                 up_read(&sb->s_umount);
 612         }
 613 }
 614
 615 /*
 616  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 617  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 618  * for IO to complete so that we effectively throttle multiple callers to the
 619  * rate at which IO is completing.
 620  */
 621 void
 622 xfs_flush_inodes(
 623         struct xfs_mount        *mp)
 624 {
 625         /*
 626          * If flush_work() returns true then that means we waited for a flush
 627          * which was already in progress.  Don't bother running another scan.
 628          */
 629         if (flush_work(&mp->m_flush_inodes_work))
 630                 return;
 631
 632         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 633         flush_work(&mp->m_flush_inodes_work);
 634 }
 635
 636 /* Catch misguided souls that try to use this interface on XFS */
 637 STATIC struct inode *
 638 xfs_fs_alloc_inode(
 639         struct super_block      *sb)
 640 {
 641         BUG();
 642         return NULL;
 643 }
 644
 645 /*
 646  * Now that the generic code is guaranteed not to be accessing
 647  * the linux inode, we can inactivate and reclaim the inode.
 648  */
 649 STATIC void
 650 xfs_fs_destroy_inode(
 651         struct inode            *inode)
 652 {
 653         struct xfs_inode        *ip = XFS_I(inode);
 654
 655         trace_xfs_destroy_inode(ip);
 656
 657         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 658         XFS_STATS_INC(ip->i_mount, vn_rele);
 659         XFS_STATS_INC(ip->i_mount, vn_remove);
 660         xfs_inode_mark_reclaimable(ip);
 661 }
 662
 663 static void
 664 xfs_fs_dirty_inode(
 665         struct inode                    *inode,
 666         int                             flags)
 667 {
 668         struct xfs_inode                *ip = XFS_I(inode);
 669         struct xfs_mount                *mp = ip->i_mount;
 670         struct xfs_trans                *tp;
 671
 672         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 673                 return;
 674
 675         /*
 676          * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC)
 677          * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed
 678          * in flags possibly together with I_DIRTY_SYNC.
 679          */
 680         if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME))
 681                 return;
 682
 683         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 684                 return;
 685         xfs_ilock(ip, XFS_ILOCK_EXCL);
 686         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 687         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 688         xfs_trans_commit(tp);
 689 }
 690
 691 /*
 692  * Slab object creation initialisation for the XFS inode.
 693  * This covers only the idempotent fields in the XFS inode;
 694  * all other fields need to be initialised on allocation
 695  * from the slab. This avoids the need to repeatedly initialise
 696  * fields in the xfs inode that left in the initialise state
 697  * when freeing the inode.
 698  */
 699 STATIC void
 700 xfs_fs_inode_init_once(
 701         void                    *inode)
 702 {
 703         struct xfs_inode        *ip = inode;
 704
 705         memset(ip, 0, sizeof(struct xfs_inode));
 706
 707         /* vfs inode */
 708         inode_init_once(VFS_I(ip));
 709
 710         /* xfs inode */
 711         atomic_set(&ip->i_pincount, 0);
 712         spin_lock_init(&ip->i_flags_lock);
 713
 714         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 715                      "xfsino", ip->i_ino);
 716 }
 717
 718 /*
 719  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 720  * serialised against cache hits here via the inode->i_lock and igrab() in
 721  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 722  * racing with us, and it avoids needing to grab a spinlock here for every inode
 723  * we drop the final reference on.
 724  */
 725 STATIC int
 726 xfs_fs_drop_inode(
 727         struct inode            *inode)
 728 {
 729         struct xfs_inode        *ip = XFS_I(inode);
 730
 731         /*
 732          * If this unlinked inode is in the middle of recovery, don't
 733          * drop the inode just yet; log recovery will take care of
 734          * that.  See the comment for this inode flag.
 735          */
 736         if (ip->i_flags & XFS_IRECOVERY) {
 737                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
 738                 return 0;
 739         }
 740
 741         return generic_drop_inode(inode);
 742 }
 743
 744 static void
 745 xfs_mount_free(
 746         struct xfs_mount        *mp)
 747 {
 748         kfree(mp->m_rtname);
 749         kfree(mp->m_logname);
 750         kmem_free(mp);
 751 }
 752
 753 STATIC int
 754 xfs_fs_sync_fs(
 755         struct super_block      *sb,
 756         int                     wait)
 757 {
 758         struct xfs_mount        *mp = XFS_M(sb);
 759         int                     error;
 760
 761         trace_xfs_fs_sync_fs(mp, __return_address);
 762
 763         /*
 764          * Doing anything during the async pass would be counterproductive.
 765          */
 766         if (!wait)
 767                 return 0;
 768
 769         error = xfs_log_force(mp, XFS_LOG_SYNC);
 770         if (error)
 771                 return error;
 772
 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         /*
 783          * If we are called with page faults frozen out, it means we are about
 784          * to freeze the transaction subsystem. Take the opportunity to shut
 785          * down inodegc because once SB_FREEZE_FS is set it's too late to
 786          * prevent inactivation races with freeze. The fs doesn't get called
 787          * again by the freezing process until after SB_FREEZE_FS has been set,
 788          * so it's now or never.  Same logic applies to speculative allocation
 789          * garbage collection.
 790          *
 791          * We don't care if this is a normal syncfs call that does this or
 792          * freeze that does this - we can run this multiple times without issue
 793          * and we won't race with a restart because a restart can only occur
 794          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
 795          */
 796         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
 797                 xfs_inodegc_stop(mp);
 798                 xfs_blockgc_stop(mp);
 799         }
 800
 801         return 0;
 802 }
 803
 804 STATIC int
 805 xfs_fs_statfs(
 806         struct dentry           *dentry,
 807         struct kstatfs          *statp)
 808 {
 809         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 810         xfs_sb_t                *sbp = &mp->m_sb;
 811         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 812         uint64_t                fakeinos, id;
 813         uint64_t                icount;
 814         uint64_t                ifree;
 815         uint64_t                fdblocks;
 816         xfs_extlen_t            lsize;
 817         int64_t                 ffree;
 818
 819         /*
 820          * Expedite background inodegc but don't wait. We do not want to block
 821          * here waiting hours for a billion extent file to be truncated.
 822          */
 823         xfs_inodegc_push(mp);
 824
 825         statp->f_type = XFS_SUPER_MAGIC;
 826         statp->f_namelen = MAXNAMELEN - 1;
 827
 828         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 829         statp->f_fsid = u64_to_fsid(id);
 830
 831         icount = percpu_counter_sum(&mp->m_icount);
 832         ifree = percpu_counter_sum(&mp->m_ifree);
 833         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 834
 835         spin_lock(&mp->m_sb_lock);
 836         statp->f_bsize = sbp->sb_blocksize;
 837         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 838         statp->f_blocks = sbp->sb_dblocks - lsize;
 839         spin_unlock(&mp->m_sb_lock);
 840
 841         /* make sure statp->f_bfree does not underflow */
 842         statp->f_bfree = max_t(int64_t, 0,
 843                                 fdblocks - xfs_fdblocks_unavailable(mp));
 844         statp->f_bavail = statp->f_bfree;
 845
 846         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 847         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 848         if (M_IGEO(mp)->maxicount)
 849                 statp->f_files = min_t(typeof(statp->f_files),
 850                                         statp->f_files,
 851                                         M_IGEO(mp)->maxicount);
 852
 853         /* If sb_icount overshot maxicount, report actual allocation */
 854         statp->f_files = max_t(typeof(statp->f_files),
 855                                         statp->f_files,
 856                                         sbp->sb_icount);
 857
 858         /* make sure statp->f_ffree does not underflow */
 859         ffree = statp->f_files - (icount - ifree);
 860         statp->f_ffree = max_t(int64_t, ffree, 0);
 861
 862
 863         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 864             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 865                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 866                 xfs_qm_statvfs(ip, statp);
 867
 868         if (XFS_IS_REALTIME_MOUNT(mp) &&
 869             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 870                 s64     freertx;
 871
 872                 statp->f_blocks = sbp->sb_rblocks;
 873                 freertx = percpu_counter_sum_positive(&mp->m_frextents);
 874                 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
 875         }
 876
 877         return 0;
 878 }
 879
 880 STATIC void
 881 xfs_save_resvblks(struct xfs_mount *mp)
 882 {
 883         uint64_t resblks = 0;
 884
 885         mp->m_resblks_save = mp->m_resblks;
 886         xfs_reserve_blocks(mp, &resblks, NULL);
 887 }
 888
 889 STATIC void
 890 xfs_restore_resvblks(struct xfs_mount *mp)
 891 {
 892         uint64_t resblks;
 893
 894         if (mp->m_resblks_save) {
 895                 resblks = mp->m_resblks_save;
 896                 mp->m_resblks_save = 0;
 897         } else
 898                 resblks = xfs_default_resblks(mp);
 899
 900         xfs_reserve_blocks(mp, &resblks, NULL);
 901 }
 902
 903 /*
 904  * Second stage of a freeze. The data is already frozen so we only
 905  * need to take care of the metadata. Once that's done sync the superblock
 906  * to the log to dirty it in case of a crash while frozen. This ensures that we
 907  * will recover the unlinked inode lists on the next mount.
 908  */
 909 STATIC int
 910 xfs_fs_freeze(
 911         struct super_block      *sb)
 912 {
 913         struct xfs_mount        *mp = XFS_M(sb);
 914         unsigned int            flags;
 915         int                     ret;
 916
 917         /*
 918          * The filesystem is now frozen far enough that memory reclaim
 919          * cannot safely operate on the filesystem. Hence we need to
 920          * set a GFP_NOFS context here to avoid recursion deadlocks.
 921          */
 922         flags = memalloc_nofs_save();
 923         xfs_save_resvblks(mp);
 924         ret = xfs_log_quiesce(mp);
 925         memalloc_nofs_restore(flags);
 926
 927         /*
 928          * For read-write filesystems, we need to restart the inodegc on error
 929          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
 930          * going to be run to restart it now.  We are at SB_FREEZE_FS level
 931          * here, so we can restart safely without racing with a stop in
 932          * xfs_fs_sync_fs().
 933          */
 934         if (ret && !xfs_is_readonly(mp)) {
 935                 xfs_blockgc_start(mp);
 936                 xfs_inodegc_start(mp);
 937         }
 938
 939         return ret;
 940 }
 941
 942 STATIC int
 943 xfs_fs_unfreeze(
 944         struct super_block      *sb)
 945 {
 946         struct xfs_mount        *mp = XFS_M(sb);
 947
 948         xfs_restore_resvblks(mp);
 949         xfs_log_work_queue(mp);
 950
 951         /*
 952          * Don't reactivate the inodegc worker on a readonly filesystem because
 953          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
 954          * worker because there are no speculative preallocations on a readonly
 955          * filesystem.
 956          */
 957         if (!xfs_is_readonly(mp)) {
 958                 xfs_blockgc_start(mp);
 959                 xfs_inodegc_start(mp);
 960         }
 961
 962         return 0;
 963 }
 964
 965 /*
 966  * This function fills in xfs_mount_t fields based on mount args.
 967  * Note: the superblock _has_ now been read in.
 968  */
 969 STATIC int
 970 xfs_finish_flags(
 971         struct xfs_mount        *mp)
 972 {
 973         /* Fail a mount where the logbuf is smaller than the log stripe */
 974         if (xfs_has_logv2(mp)) {
 975                 if (mp->m_logbsize <= 0 &&
 976                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 977                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 978                 } else if (mp->m_logbsize > 0 &&
 979                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 980                         xfs_warn(mp,
 981                 "logbuf size must be greater than or equal to log stripe size");
 982                         return -EINVAL;
 983                 }
 984         } else {
 985                 /* Fail a mount if the logbuf is larger than 32K */
 986                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 987                         xfs_warn(mp,
 988                 "logbuf size for version 1 logs must be 16K or 32K");
 989                         return -EINVAL;
 990                 }
 991         }
 992
 993         /*
 994          * V5 filesystems always use attr2 format for attributes.
 995          */
 996         if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
 997                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 998                              "attr2 is always enabled for V5 filesystems.");
 999                 return -EINVAL;
1000         }
1001
1002         /*
1003          * prohibit r/w mounts of read-only filesystems
1004          */
1005         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
1006                 xfs_warn(mp,
1007                         "cannot mount a read-only filesystem as read-write");
1008                 return -EROFS;
1009         }
1010
1011         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
1012             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
1013             !xfs_has_pquotino(mp)) {
1014                 xfs_warn(mp,
1015                   "Super block does not support project and group quota together");
1016                 return -EINVAL;
1017         }
1018
1019         return 0;
1020 }
1021
1022 static int
1023 xfs_init_percpu_counters(
1024         struct xfs_mount        *mp)
1025 {
1026         int             error;
1027
1028         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1029         if (error)
1030                 return -ENOMEM;
1031
1032         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1033         if (error)
1034                 goto free_icount;
1035
1036         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1037         if (error)
1038                 goto free_ifree;
1039
1040         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1041         if (error)
1042                 goto free_fdblocks;
1043
1044         error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1045         if (error)
1046                 goto free_delalloc;
1047
1048         return 0;
1049
1050 free_delalloc:
1051         percpu_counter_destroy(&mp->m_delalloc_blks);
1052 free_fdblocks:
1053         percpu_counter_destroy(&mp->m_fdblocks);
1054 free_ifree:
1055         percpu_counter_destroy(&mp->m_ifree);
1056 free_icount:
1057         percpu_counter_destroy(&mp->m_icount);
1058         return -ENOMEM;
1059 }
1060
1061 void
1062 xfs_reinit_percpu_counters(
1063         struct xfs_mount        *mp)
1064 {
1065         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1066         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1067         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1068         percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1069 }
1070
1071 static void
1072 xfs_destroy_percpu_counters(
1073         struct xfs_mount        *mp)
1074 {
1075         percpu_counter_destroy(&mp->m_icount);
1076         percpu_counter_destroy(&mp->m_ifree);
1077         percpu_counter_destroy(&mp->m_fdblocks);
1078         ASSERT(xfs_is_shutdown(mp) ||
1079                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1080         percpu_counter_destroy(&mp->m_delalloc_blks);
1081         percpu_counter_destroy(&mp->m_frextents);
1082 }
1083
1084 static int
1085 xfs_inodegc_init_percpu(
1086         struct xfs_mount        *mp)
1087 {
1088         struct xfs_inodegc      *gc;
1089         int                     cpu;
1090
1091         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1092         if (!mp->m_inodegc)
1093                 return -ENOMEM;
1094
1095         for_each_possible_cpu(cpu) {
1096                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1097                 init_llist_head(&gc->list);
1098                 gc->items = 0;
1099                 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1100         }
1101         return 0;
1102 }
1103
1104 static void
1105 xfs_inodegc_free_percpu(
1106         struct xfs_mount        *mp)
1107 {
1108         if (!mp->m_inodegc)
1109                 return;
1110         free_percpu(mp->m_inodegc);
1111 }
1112
1113 static void
1114 xfs_fs_put_super(
1115         struct super_block      *sb)
1116 {
1117         struct xfs_mount        *mp = XFS_M(sb);
1118
1119         /* if ->fill_super failed, we have no mount to tear down */
1120         if (!sb->s_fs_info)
1121                 return;
1122
1123         xfs_notice(mp, "Unmounting Filesystem %pU", &mp->m_sb.sb_uuid);
1124         xfs_filestream_unmount(mp);
1125         xfs_unmountfs(mp);
1126
1127         xfs_freesb(mp);
1128         free_percpu(mp->m_stats.xs_stats);
1129         xfs_mount_list_del(mp);
1130         xfs_inodegc_free_percpu(mp);
1131         xfs_destroy_percpu_counters(mp);
1132         xfs_destroy_mount_workqueues(mp);
1133         xfs_close_devices(mp);
1134
1135         sb->s_fs_info = NULL;
1136         xfs_mount_free(mp);
1137 }
1138
1139 static long
1140 xfs_fs_nr_cached_objects(
1141         struct super_block      *sb,
1142         struct shrink_control   *sc)
1143 {
1144         /* Paranoia: catch incorrect calls during mount setup or teardown */
1145         if (WARN_ON_ONCE(!sb->s_fs_info))
1146                 return 0;
1147         return xfs_reclaim_inodes_count(XFS_M(sb));
1148 }
1149
1150 static long
1151 xfs_fs_free_cached_objects(
1152         struct super_block      *sb,
1153         struct shrink_control   *sc)
1154 {
1155         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1156 }
1157
1158 static const struct super_operations xfs_super_operations = {
1159         .alloc_inode            = xfs_fs_alloc_inode,
1160         .destroy_inode          = xfs_fs_destroy_inode,
1161         .dirty_inode            = xfs_fs_dirty_inode,
1162         .drop_inode             = xfs_fs_drop_inode,
1163         .put_super              = xfs_fs_put_super,
1164         .sync_fs                = xfs_fs_sync_fs,
1165         .freeze_fs              = xfs_fs_freeze,
1166         .unfreeze_fs            = xfs_fs_unfreeze,
1167         .statfs                 = xfs_fs_statfs,
1168         .show_options           = xfs_fs_show_options,
1169         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1170         .free_cached_objects    = xfs_fs_free_cached_objects,
1171 };
1172
1173 static int
1174 suffix_kstrtoint(
1175         const char      *s,
1176         unsigned int    base,
1177         int             *res)
1178 {
1179         int             last, shift_left_factor = 0, _res;
1180         char            *value;
1181         int             ret = 0;
1182
1183         value = kstrdup(s, GFP_KERNEL);
1184         if (!value)
1185                 return -ENOMEM;
1186
1187         last = strlen(value) - 1;
1188         if (value[last] == 'K' || value[last] == 'k') {
1189                 shift_left_factor = 10;
1190                 value[last] = '\0';
1191         }
1192         if (value[last] == 'M' || value[last] == 'm') {
1193                 shift_left_factor = 20;
1194                 value[last] = '\0';
1195         }
1196         if (value[last] == 'G' || value[last] == 'g') {
1197                 shift_left_factor = 30;
1198                 value[last] = '\0';
1199         }
1200
1201         if (kstrtoint(value, base, &_res))
1202                 ret = -EINVAL;
1203         kfree(value);
1204         *res = _res << shift_left_factor;
1205         return ret;
1206 }
1207
1208 static inline void
1209 xfs_fs_warn_deprecated(
1210         struct fs_context       *fc,
1211         struct fs_parameter     *param,
1212         uint64_t                flag,
1213         bool                    value)
1214 {
1215         /* Don't print the warning if reconfiguring and current mount point
1216          * already had the flag set
1217          */
1218         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1219             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1220                 return;
1221         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1222 }
1223
1224 /*
1225  * Set mount state from a mount option.
1226  *
1227  * NOTE: mp->m_super is NULL here!
1228  */
1229 static int
1230 xfs_fs_parse_param(
1231         struct fs_context       *fc,
1232         struct fs_parameter     *param)
1233 {
1234         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1235         struct fs_parse_result  result;
1236         int                     size = 0;
1237         int                     opt;
1238
1239         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1240         if (opt < 0)
1241                 return opt;
1242
1243         switch (opt) {
1244         case Opt_logbufs:
1245                 parsing_mp->m_logbufs = result.uint_32;
1246                 return 0;
1247         case Opt_logbsize:
1248                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1249                         return -EINVAL;
1250                 return 0;
1251         case Opt_logdev:
1252                 kfree(parsing_mp->m_logname);
1253                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1254                 if (!parsing_mp->m_logname)
1255                         return -ENOMEM;
1256                 return 0;
1257         case Opt_rtdev:
1258                 kfree(parsing_mp->m_rtname);
1259                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1260                 if (!parsing_mp->m_rtname)
1261                         return -ENOMEM;
1262                 return 0;
1263         case Opt_allocsize:
1264                 if (suffix_kstrtoint(param->string, 10, &size))
1265                         return -EINVAL;
1266                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1267                 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1268                 return 0;
1269         case Opt_grpid:
1270         case Opt_bsdgroups:
1271                 parsing_mp->m_features |= XFS_FEAT_GRPID;
1272                 return 0;
1273         case Opt_nogrpid:
1274         case Opt_sysvgroups:
1275                 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1276                 return 0;
1277         case Opt_wsync:
1278                 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1279                 return 0;
1280         case Opt_norecovery:
1281                 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1282                 return 0;
1283         case Opt_noalign:
1284                 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1285                 return 0;
1286         case Opt_swalloc:
1287                 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1288                 return 0;
1289         case Opt_sunit:
1290                 parsing_mp->m_dalign = result.uint_32;
1291                 return 0;
1292         case Opt_swidth:
1293                 parsing_mp->m_swidth = result.uint_32;
1294                 return 0;
1295         case Opt_inode32:
1296                 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1297                 return 0;
1298         case Opt_inode64:
1299                 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1300                 return 0;
1301         case Opt_nouuid:
1302                 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1303                 return 0;
1304         case Opt_largeio:
1305                 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1306                 return 0;
1307         case Opt_nolargeio:
1308                 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1309                 return 0;
1310         case Opt_filestreams:
1311                 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1312                 return 0;
1313         case Opt_noquota:
1314                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1315                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1316                 return 0;
1317         case Opt_quota:
1318         case Opt_uquota:
1319         case Opt_usrquota:
1320                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1321                 return 0;
1322         case Opt_qnoenforce:
1323         case Opt_uqnoenforce:
1324                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1325                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1326                 return 0;
1327         case Opt_pquota:
1328         case Opt_prjquota:
1329                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1330                 return 0;
1331         case Opt_pqnoenforce:
1332                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1333                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1334                 return 0;
1335         case Opt_gquota:
1336         case Opt_grpquota:
1337                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1338                 return 0;
1339         case Opt_gqnoenforce:
1340                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1341                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1342                 return 0;
1343         case Opt_discard:
1344                 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1345                 return 0;
1346         case Opt_nodiscard:
1347                 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1348                 return 0;
1349 #ifdef CONFIG_FS_DAX
1350         case Opt_dax:
1351                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1352                 return 0;
1353         case Opt_dax_enum:
1354                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1355                 return 0;
1356 #endif
1357         /* Following mount options will be removed in September 2025 */
1358         case Opt_ikeep:
1359                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1360                 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1361                 return 0;
1362         case Opt_noikeep:
1363                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1364                 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1365                 return 0;
1366         case Opt_attr2:
1367                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1368                 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1369                 return 0;
1370         case Opt_noattr2:
1371                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1372                 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1373                 return 0;
1374         default:
1375                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1376                 return -EINVAL;
1377         }
1378
1379         return 0;
1380 }
1381
1382 static int
1383 xfs_fs_validate_params(
1384         struct xfs_mount        *mp)
1385 {
1386         /* No recovery flag requires a read-only mount */
1387         if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1388                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1389                 return -EINVAL;
1390         }
1391
1392         /*
1393          * We have not read the superblock at this point, so only the attr2
1394          * mount option can set the attr2 feature by this stage.
1395          */
1396         if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1397                 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1398                 return -EINVAL;
1399         }
1400
1401
1402         if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1403                 xfs_warn(mp,
1404         "sunit and swidth options incompatible with the noalign option");
1405                 return -EINVAL;
1406         }
1407
1408         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1409                 xfs_warn(mp, "quota support not available in this kernel.");
1410                 return -EINVAL;
1411         }
1412
1413         if ((mp->m_dalign && !mp->m_swidth) ||
1414             (!mp->m_dalign && mp->m_swidth)) {
1415                 xfs_warn(mp, "sunit and swidth must be specified together");
1416                 return -EINVAL;
1417         }
1418
1419         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1420                 xfs_warn(mp,
1421         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1422                         mp->m_swidth, mp->m_dalign);
1423                 return -EINVAL;
1424         }
1425
1426         if (mp->m_logbufs != -1 &&
1427             mp->m_logbufs != 0 &&
1428             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1429              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1430                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1431                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1432                 return -EINVAL;
1433         }
1434
1435         if (mp->m_logbsize != -1 &&
1436             mp->m_logbsize !=  0 &&
1437             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1438              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1439              !is_power_of_2(mp->m_logbsize))) {
1440                 xfs_warn(mp,
1441                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1442                         mp->m_logbsize);
1443                 return -EINVAL;
1444         }
1445
1446         if (xfs_has_allocsize(mp) &&
1447             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1448              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1449                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1450                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1451                 return -EINVAL;
1452         }
1453
1454         return 0;
1455 }
1456
1457 static int
1458 xfs_fs_fill_super(
1459         struct super_block      *sb,
1460         struct fs_context       *fc)
1461 {
1462         struct xfs_mount        *mp = sb->s_fs_info;
1463         struct inode            *root;
1464         int                     flags = 0, error;
1465
1466         mp->m_super = sb;
1467
1468         error = xfs_fs_validate_params(mp);
1469         if (error)
1470                 goto out_free_names;
1471
1472         sb_min_blocksize(sb, BBSIZE);
1473         sb->s_xattr = xfs_xattr_handlers;
1474         sb->s_export_op = &xfs_export_operations;
1475 #ifdef CONFIG_XFS_QUOTA
1476         sb->s_qcop = &xfs_quotactl_operations;
1477         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1478 #endif
1479         sb->s_op = &xfs_super_operations;
1480
1481         /*
1482          * Delay mount work if the debug hook is set. This is debug
1483          * instrumention to coordinate simulation of xfs mount failures with
1484          * VFS superblock operations
1485          */
1486         if (xfs_globals.mount_delay) {
1487                 xfs_notice(mp, "Delaying mount for %d seconds.",
1488                         xfs_globals.mount_delay);
1489                 msleep(xfs_globals.mount_delay * 1000);
1490         }
1491
1492         if (fc->sb_flags & SB_SILENT)
1493                 flags |= XFS_MFSI_QUIET;
1494
1495         error = xfs_open_devices(mp);
1496         if (error)
1497                 goto out_free_names;
1498
1499         error = xfs_init_mount_workqueues(mp);
1500         if (error)
1501                 goto out_close_devices;
1502
1503         error = xfs_init_percpu_counters(mp);
1504         if (error)
1505                 goto out_destroy_workqueues;
1506
1507         error = xfs_inodegc_init_percpu(mp);
1508         if (error)
1509                 goto out_destroy_counters;
1510
1511         /*
1512          * All percpu data structures requiring cleanup when a cpu goes offline
1513          * must be allocated before adding this @mp to the cpu-dead handler's
1514          * mount list.
1515          */
1516         xfs_mount_list_add(mp);
1517
1518         /* Allocate stats memory before we do operations that might use it */
1519         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1520         if (!mp->m_stats.xs_stats) {
1521                 error = -ENOMEM;
1522                 goto out_destroy_inodegc;
1523         }
1524
1525         error = xfs_readsb(mp, flags);
1526         if (error)
1527                 goto out_free_stats;
1528
1529         error = xfs_finish_flags(mp);
1530         if (error)
1531                 goto out_free_sb;
1532
1533         error = xfs_setup_devices(mp);
1534         if (error)
1535                 goto out_free_sb;
1536
1537         /* V4 support is undergoing deprecation. */
1538         if (!xfs_has_crc(mp)) {
1539 #ifdef CONFIG_XFS_SUPPORT_V4
1540                 xfs_warn_once(mp,
1541         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1542 #else
1543                 xfs_warn(mp,
1544         "Deprecated V4 format (crc=0) not supported by kernel.");
1545                 error = -EINVAL;
1546                 goto out_free_sb;
1547 #endif
1548         }
1549
1550         /* Filesystem claims it needs repair, so refuse the mount. */
1551         if (xfs_has_needsrepair(mp)) {
1552                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1553                 error = -EFSCORRUPTED;
1554                 goto out_free_sb;
1555         }
1556
1557         /*
1558          * Don't touch the filesystem if a user tool thinks it owns the primary
1559          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1560          * we don't check them at all.
1561          */
1562         if (mp->m_sb.sb_inprogress) {
1563                 xfs_warn(mp, "Offline file system operation in progress!");
1564                 error = -EFSCORRUPTED;
1565                 goto out_free_sb;
1566         }
1567
1568         /*
1569          * Until this is fixed only page-sized or smaller data blocks work.
1570          */
1571         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1572                 xfs_warn(mp,
1573                 "File system with blocksize %d bytes. "
1574                 "Only pagesize (%ld) or less will currently work.",
1575                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1576                 error = -ENOSYS;
1577                 goto out_free_sb;
1578         }
1579
1580         /* Ensure this filesystem fits in the page cache limits */
1581         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1582             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1583                 xfs_warn(mp,
1584                 "file system too large to be mounted on this system.");
1585                 error = -EFBIG;
1586                 goto out_free_sb;
1587         }
1588
1589         /*
1590          * XFS block mappings use 54 bits to store the logical block offset.
1591          * This should suffice to handle the maximum file size that the VFS
1592          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1593          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1594          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1595          * to check this assertion.
1596          *
1597          * Avoid integer overflow by comparing the maximum bmbt offset to the
1598          * maximum pagecache offset in units of fs blocks.
1599          */
1600         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1601                 xfs_warn(mp,
1602 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1603                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1604                          XFS_MAX_FILEOFF);
1605                 error = -EINVAL;
1606                 goto out_free_sb;
1607         }
1608
1609         error = xfs_filestream_mount(mp);
1610         if (error)
1611                 goto out_free_sb;
1612
1613         /*
1614          * we must configure the block size in the superblock before we run the
1615          * full mount process as the mount process can lookup and cache inodes.
1616          */
1617         sb->s_magic = XFS_SUPER_MAGIC;
1618         sb->s_blocksize = mp->m_sb.sb_blocksize;
1619         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1620         sb->s_maxbytes = MAX_LFS_FILESIZE;
1621         sb->s_max_links = XFS_MAXLINK;
1622         sb->s_time_gran = 1;
1623         if (xfs_has_bigtime(mp)) {
1624                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1625                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1626         } else {
1627                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1628                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1629         }
1630         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1631         sb->s_iflags |= SB_I_CGROUPWB;
1632
1633         set_posix_acl_flag(sb);
1634
1635         /* version 5 superblocks support inode version counters. */
1636         if (xfs_has_crc(mp))
1637                 sb->s_flags |= SB_I_VERSION;
1638
1639         if (xfs_has_dax_always(mp)) {
1640                 error = xfs_setup_dax_always(mp);
1641                 if (error)
1642                         goto out_filestream_unmount;
1643         }
1644
1645         if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1646                 xfs_warn(mp,
1647         "mounting with \"discard\" option, but the device does not support discard");
1648                 mp->m_features &= ~XFS_FEAT_DISCARD;
1649         }
1650
1651         if (xfs_has_reflink(mp)) {
1652                 if (mp->m_sb.sb_rblocks) {
1653                         xfs_alert(mp,
1654         "reflink not compatible with realtime device!");
1655                         error = -EINVAL;
1656                         goto out_filestream_unmount;
1657                 }
1658
1659                 if (xfs_globals.always_cow) {
1660                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1661                         mp->m_always_cow = true;
1662                 }
1663         }
1664
1665         if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1666                 xfs_alert(mp,
1667         "reverse mapping btree not compatible with realtime device!");
1668                 error = -EINVAL;
1669                 goto out_filestream_unmount;
1670         }
1671
1672         if (xfs_has_large_extent_counts(mp))
1673                 xfs_warn(mp,
1674         "EXPERIMENTAL Large extent counts feature in use. Use at your own risk!");
1675
1676         error = xfs_mountfs(mp);
1677         if (error)
1678                 goto out_filestream_unmount;
1679
1680         root = igrab(VFS_I(mp->m_rootip));
1681         if (!root) {
1682                 error = -ENOENT;
1683                 goto out_unmount;
1684         }
1685         sb->s_root = d_make_root(root);
1686         if (!sb->s_root) {
1687                 error = -ENOMEM;
1688                 goto out_unmount;
1689         }
1690
1691         return 0;
1692
1693  out_filestream_unmount:
1694         xfs_filestream_unmount(mp);
1695  out_free_sb:
1696         xfs_freesb(mp);
1697  out_free_stats:
1698         free_percpu(mp->m_stats.xs_stats);
1699  out_destroy_inodegc:
1700         xfs_mount_list_del(mp);
1701         xfs_inodegc_free_percpu(mp);
1702  out_destroy_counters:
1703         xfs_destroy_percpu_counters(mp);
1704  out_destroy_workqueues:
1705         xfs_destroy_mount_workqueues(mp);
1706  out_close_devices:
1707         xfs_close_devices(mp);
1708  out_free_names:
1709         sb->s_fs_info = NULL;
1710         xfs_mount_free(mp);
1711         return error;
1712
1713  out_unmount:
1714         xfs_filestream_unmount(mp);
1715         xfs_unmountfs(mp);
1716         goto out_free_sb;
1717 }
1718
1719 static int
1720 xfs_fs_get_tree(
1721         struct fs_context       *fc)
1722 {
1723         return get_tree_bdev(fc, xfs_fs_fill_super);
1724 }
1725
1726 static int
1727 xfs_remount_rw(
1728         struct xfs_mount        *mp)
1729 {
1730         struct xfs_sb           *sbp = &mp->m_sb;
1731         int error;
1732
1733         if (xfs_has_norecovery(mp)) {
1734                 xfs_warn(mp,
1735                         "ro->rw transition prohibited on norecovery mount");
1736                 return -EINVAL;
1737         }
1738
1739         if (xfs_sb_is_v5(sbp) &&
1740             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1741                 xfs_warn(mp,
1742         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1743                         (sbp->sb_features_ro_compat &
1744                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1745                 return -EINVAL;
1746         }
1747
1748         clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1749
1750         /*
1751          * If this is the first remount to writeable state we might have some
1752          * superblock changes to update.
1753          */
1754         if (mp->m_update_sb) {
1755                 error = xfs_sync_sb(mp, false);
1756                 if (error) {
1757                         xfs_warn(mp, "failed to write sb changes");
1758                         return error;
1759                 }
1760                 mp->m_update_sb = false;
1761         }
1762
1763         /*
1764          * Fill out the reserve pool if it is empty. Use the stashed value if
1765          * it is non-zero, otherwise go with the default.
1766          */
1767         xfs_restore_resvblks(mp);
1768         xfs_log_work_queue(mp);
1769         xfs_blockgc_start(mp);
1770
1771         /* Create the per-AG metadata reservation pool .*/
1772         error = xfs_fs_reserve_ag_blocks(mp);
1773         if (error && error != -ENOSPC)
1774                 return error;
1775
1776         /* Re-enable the background inode inactivation worker. */
1777         xfs_inodegc_start(mp);
1778
1779         return 0;
1780 }
1781
1782 static int
1783 xfs_remount_ro(
1784         struct xfs_mount        *mp)
1785 {
1786         struct xfs_icwalk       icw = {
1787                 .icw_flags      = XFS_ICWALK_FLAG_SYNC,
1788         };
1789         int                     error;
1790
1791         /* Flush all the dirty data to disk. */
1792         error = sync_filesystem(mp->m_super);
1793         if (error)
1794                 return error;
1795
1796         /*
1797          * Cancel background eofb scanning so it cannot race with the final
1798          * log force+buftarg wait and deadlock the remount.
1799          */
1800         xfs_blockgc_stop(mp);
1801
1802         /*
1803          * Clear out all remaining COW staging extents and speculative post-EOF
1804          * preallocations so that we don't leave inodes requiring inactivation
1805          * cleanups during reclaim on a read-only mount.  We must process every
1806          * cached inode, so this requires a synchronous cache scan.
1807          */
1808         error = xfs_blockgc_free_space(mp, &icw);
1809         if (error) {
1810                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1811                 return error;
1812         }
1813
1814         /*
1815          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1816          * flushed all pending inodegc work when it sync'd the filesystem.
1817          * The VFS holds s_umount, so we know that inodes cannot enter
1818          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1819          * we send inodes straight to reclaim, so no inodes will be queued.
1820          */
1821         xfs_inodegc_stop(mp);
1822
1823         /* Free the per-AG metadata reservation pool. */
1824         error = xfs_fs_unreserve_ag_blocks(mp);
1825         if (error) {
1826                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1827                 return error;
1828         }
1829
1830         /*
1831          * Before we sync the metadata, we need to free up the reserve block
1832          * pool so that the used block count in the superblock on disk is
1833          * correct at the end of the remount. Stash the current* reserve pool
1834          * size so that if we get remounted rw, we can return it to the same
1835          * size.
1836          */
1837         xfs_save_resvblks(mp);
1838
1839         xfs_log_clean(mp);
1840         set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1841
1842         return 0;
1843 }
1844
1845 /*
1846  * Logically we would return an error here to prevent users from believing
1847  * they might have changed mount options using remount which can't be changed.
1848  *
1849  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1850  * arguments in some cases so we can't blindly reject options, but have to
1851  * check for each specified option if it actually differs from the currently
1852  * set option and only reject it if that's the case.
1853  *
1854  * Until that is implemented we return success for every remount request, and
1855  * silently ignore all options that we can't actually change.
1856  */
1857 static int
1858 xfs_fs_reconfigure(
1859         struct fs_context *fc)
1860 {
1861         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1862         struct xfs_mount        *new_mp = fc->s_fs_info;
1863         int                     flags = fc->sb_flags;
1864         int                     error;
1865
1866         /* version 5 superblocks always support version counters. */
1867         if (xfs_has_crc(mp))
1868                 fc->sb_flags |= SB_I_VERSION;
1869
1870         error = xfs_fs_validate_params(new_mp);
1871         if (error)
1872                 return error;
1873
1874         /* inode32 -> inode64 */
1875         if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1876                 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1877                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1878         }
1879
1880         /* inode64 -> inode32 */
1881         if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1882                 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1883                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1884         }
1885
1886         /* ro -> rw */
1887         if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1888                 error = xfs_remount_rw(mp);
1889                 if (error)
1890                         return error;
1891         }
1892
1893         /* rw -> ro */
1894         if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1895                 error = xfs_remount_ro(mp);
1896                 if (error)
1897                         return error;
1898         }
1899
1900         return 0;
1901 }
1902
1903 static void xfs_fs_free(
1904         struct fs_context       *fc)
1905 {
1906         struct xfs_mount        *mp = fc->s_fs_info;
1907
1908         /*
1909          * mp is stored in the fs_context when it is initialized.
1910          * mp is transferred to the superblock on a successful mount,
1911          * but if an error occurs before the transfer we have to free
1912          * it here.
1913          */
1914         if (mp)
1915                 xfs_mount_free(mp);
1916 }
1917
1918 static const struct fs_context_operations xfs_context_ops = {
1919         .parse_param = xfs_fs_parse_param,
1920         .get_tree    = xfs_fs_get_tree,
1921         .reconfigure = xfs_fs_reconfigure,
1922         .free        = xfs_fs_free,
1923 };
1924
1925 static int xfs_init_fs_context(
1926         struct fs_context       *fc)
1927 {
1928         struct xfs_mount        *mp;
1929
1930         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1931         if (!mp)
1932                 return -ENOMEM;
1933
1934         spin_lock_init(&mp->m_sb_lock);
1935         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1936         spin_lock_init(&mp->m_perag_lock);
1937         mutex_init(&mp->m_growlock);
1938         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1939         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1940         mp->m_kobj.kobject.kset = xfs_kset;
1941         /*
1942          * We don't create the finobt per-ag space reservation until after log
1943          * recovery, so we must set this to true so that an ifree transaction
1944          * started during log recovery will not depend on space reservations
1945          * for finobt expansion.
1946          */
1947         mp->m_finobt_nores = true;
1948
1949         /*
1950          * These can be overridden by the mount option parsing.
1951          */
1952         mp->m_logbufs = -1;
1953         mp->m_logbsize = -1;
1954         mp->m_allocsize_log = 16; /* 64k */
1955
1956         /*
1957          * Copy binary VFS mount flags we are interested in.
1958          */
1959         if (fc->sb_flags & SB_RDONLY)
1960                 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1961         if (fc->sb_flags & SB_DIRSYNC)
1962                 mp->m_features |= XFS_FEAT_DIRSYNC;
1963         if (fc->sb_flags & SB_SYNCHRONOUS)
1964                 mp->m_features |= XFS_FEAT_WSYNC;
1965
1966         fc->s_fs_info = mp;
1967         fc->ops = &xfs_context_ops;
1968
1969         return 0;
1970 }
1971
1972 static struct file_system_type xfs_fs_type = {
1973         .owner                  = THIS_MODULE,
1974         .name                   = "xfs",
1975         .init_fs_context        = xfs_init_fs_context,
1976         .parameters             = xfs_fs_parameters,
1977         .kill_sb                = kill_block_super,
1978         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1979 };
1980 MODULE_ALIAS_FS("xfs");
1981
1982 STATIC int __init
1983 xfs_init_caches(void)
1984 {
1985         int             error;
1986
1987         xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
1988                                          SLAB_HWCACHE_ALIGN |
1989                                          SLAB_RECLAIM_ACCOUNT |
1990                                          SLAB_MEM_SPREAD,
1991                                          NULL);
1992         if (!xfs_buf_cache)
1993                 goto out;
1994
1995         xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
1996                                                 sizeof(struct xlog_ticket),
1997                                                 0, 0, NULL);
1998         if (!xfs_log_ticket_cache)
1999                 goto out_destroy_buf_cache;
2000
2001         error = xfs_btree_init_cur_caches();
2002         if (error)
2003                 goto out_destroy_log_ticket_cache;
2004
2005         error = xfs_defer_init_item_caches();
2006         if (error)
2007                 goto out_destroy_btree_cur_cache;
2008
2009         xfs_da_state_cache = kmem_cache_create("xfs_da_state",
2010                                               sizeof(struct xfs_da_state),
2011                                               0, 0, NULL);
2012         if (!xfs_da_state_cache)
2013                 goto out_destroy_defer_item_cache;
2014
2015         xfs_ifork_cache = kmem_cache_create("xfs_ifork",
2016                                            sizeof(struct xfs_ifork),
2017                                            0, 0, NULL);
2018         if (!xfs_ifork_cache)
2019                 goto out_destroy_da_state_cache;
2020
2021         xfs_trans_cache = kmem_cache_create("xfs_trans",
2022                                            sizeof(struct xfs_trans),
2023                                            0, 0, NULL);
2024         if (!xfs_trans_cache)
2025                 goto out_destroy_ifork_cache;
2026
2027
2028         /*
2029          * The size of the cache-allocated buf log item is the maximum
2030          * size possible under XFS.  This wastes a little bit of memory,
2031          * but it is much faster.
2032          */
2033         xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2034                                               sizeof(struct xfs_buf_log_item),
2035                                               0, 0, NULL);
2036         if (!xfs_buf_item_cache)
2037                 goto out_destroy_trans_cache;
2038
2039         xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2040                         xfs_efd_log_item_sizeof(XFS_EFD_MAX_FAST_EXTENTS),
2041                         0, 0, NULL);
2042         if (!xfs_efd_cache)
2043                 goto out_destroy_buf_item_cache;
2044
2045         xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2046                         xfs_efi_log_item_sizeof(XFS_EFI_MAX_FAST_EXTENTS),
2047                         0, 0, NULL);
2048         if (!xfs_efi_cache)
2049                 goto out_destroy_efd_cache;
2050
2051         xfs_inode_cache = kmem_cache_create("xfs_inode",
2052                                            sizeof(struct xfs_inode), 0,
2053                                            (SLAB_HWCACHE_ALIGN |
2054                                             SLAB_RECLAIM_ACCOUNT |
2055                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2056                                            xfs_fs_inode_init_once);
2057         if (!xfs_inode_cache)
2058                 goto out_destroy_efi_cache;
2059
2060         xfs_ili_cache = kmem_cache_create("xfs_ili",
2061                                          sizeof(struct xfs_inode_log_item), 0,
2062                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2063                                          NULL);
2064         if (!xfs_ili_cache)
2065                 goto out_destroy_inode_cache;
2066
2067         xfs_icreate_cache = kmem_cache_create("xfs_icr",
2068                                              sizeof(struct xfs_icreate_item),
2069                                              0, 0, NULL);
2070         if (!xfs_icreate_cache)
2071                 goto out_destroy_ili_cache;
2072
2073         xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2074                                          sizeof(struct xfs_rud_log_item),
2075                                          0, 0, NULL);
2076         if (!xfs_rud_cache)
2077                 goto out_destroy_icreate_cache;
2078
2079         xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2080                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2081                         0, 0, NULL);
2082         if (!xfs_rui_cache)
2083                 goto out_destroy_rud_cache;
2084
2085         xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2086                                          sizeof(struct xfs_cud_log_item),
2087                                          0, 0, NULL);
2088         if (!xfs_cud_cache)
2089                 goto out_destroy_rui_cache;
2090
2091         xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2092                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2093                         0, 0, NULL);
2094         if (!xfs_cui_cache)
2095                 goto out_destroy_cud_cache;
2096
2097         xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2098                                          sizeof(struct xfs_bud_log_item),
2099                                          0, 0, NULL);
2100         if (!xfs_bud_cache)
2101                 goto out_destroy_cui_cache;
2102
2103         xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2104                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2105                         0, 0, NULL);
2106         if (!xfs_bui_cache)
2107                 goto out_destroy_bud_cache;
2108
2109         xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2110                                             sizeof(struct xfs_attrd_log_item),
2111                                             0, 0, NULL);
2112         if (!xfs_attrd_cache)
2113                 goto out_destroy_bui_cache;
2114
2115         xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2116                                             sizeof(struct xfs_attri_log_item),
2117                                             0, 0, NULL);
2118         if (!xfs_attri_cache)
2119                 goto out_destroy_attrd_cache;
2120
2121         xfs_iunlink_cache = kmem_cache_create("xfs_iul_item",
2122                                              sizeof(struct xfs_iunlink_item),
2123                                              0, 0, NULL);
2124         if (!xfs_iunlink_cache)
2125                 goto out_destroy_attri_cache;
2126
2127         return 0;
2128
2129  out_destroy_attri_cache:
2130         kmem_cache_destroy(xfs_attri_cache);
2131  out_destroy_attrd_cache:
2132         kmem_cache_destroy(xfs_attrd_cache);
2133  out_destroy_bui_cache:
2134         kmem_cache_destroy(xfs_bui_cache);
2135  out_destroy_bud_cache:
2136         kmem_cache_destroy(xfs_bud_cache);
2137  out_destroy_cui_cache:
2138         kmem_cache_destroy(xfs_cui_cache);
2139  out_destroy_cud_cache:
2140         kmem_cache_destroy(xfs_cud_cache);
2141  out_destroy_rui_cache:
2142         kmem_cache_destroy(xfs_rui_cache);
2143  out_destroy_rud_cache:
2144         kmem_cache_destroy(xfs_rud_cache);
2145  out_destroy_icreate_cache:
2146         kmem_cache_destroy(xfs_icreate_cache);
2147  out_destroy_ili_cache:
2148         kmem_cache_destroy(xfs_ili_cache);
2149  out_destroy_inode_cache:
2150         kmem_cache_destroy(xfs_inode_cache);
2151  out_destroy_efi_cache:
2152         kmem_cache_destroy(xfs_efi_cache);
2153  out_destroy_efd_cache:
2154         kmem_cache_destroy(xfs_efd_cache);
2155  out_destroy_buf_item_cache:
2156         kmem_cache_destroy(xfs_buf_item_cache);
2157  out_destroy_trans_cache:
2158         kmem_cache_destroy(xfs_trans_cache);
2159  out_destroy_ifork_cache:
2160         kmem_cache_destroy(xfs_ifork_cache);
2161  out_destroy_da_state_cache:
2162         kmem_cache_destroy(xfs_da_state_cache);
2163  out_destroy_defer_item_cache:
2164         xfs_defer_destroy_item_caches();
2165  out_destroy_btree_cur_cache:
2166         xfs_btree_destroy_cur_caches();
2167  out_destroy_log_ticket_cache:
2168         kmem_cache_destroy(xfs_log_ticket_cache);
2169  out_destroy_buf_cache:
2170         kmem_cache_destroy(xfs_buf_cache);
2171  out:
2172         return -ENOMEM;
2173 }
2174
2175 STATIC void
2176 xfs_destroy_caches(void)
2177 {
2178         /*
2179          * Make sure all delayed rcu free are flushed before we
2180          * destroy caches.
2181          */
2182         rcu_barrier();
2183         kmem_cache_destroy(xfs_iunlink_cache);
2184         kmem_cache_destroy(xfs_attri_cache);
2185         kmem_cache_destroy(xfs_attrd_cache);
2186         kmem_cache_destroy(xfs_bui_cache);
2187         kmem_cache_destroy(xfs_bud_cache);
2188         kmem_cache_destroy(xfs_cui_cache);
2189         kmem_cache_destroy(xfs_cud_cache);
2190         kmem_cache_destroy(xfs_rui_cache);
2191         kmem_cache_destroy(xfs_rud_cache);
2192         kmem_cache_destroy(xfs_icreate_cache);
2193         kmem_cache_destroy(xfs_ili_cache);
2194         kmem_cache_destroy(xfs_inode_cache);
2195         kmem_cache_destroy(xfs_efi_cache);
2196         kmem_cache_destroy(xfs_efd_cache);
2197         kmem_cache_destroy(xfs_buf_item_cache);
2198         kmem_cache_destroy(xfs_trans_cache);
2199         kmem_cache_destroy(xfs_ifork_cache);
2200         kmem_cache_destroy(xfs_da_state_cache);
2201         xfs_defer_destroy_item_caches();
2202         xfs_btree_destroy_cur_caches();
2203         kmem_cache_destroy(xfs_log_ticket_cache);
2204         kmem_cache_destroy(xfs_buf_cache);
2205 }
2206
2207 STATIC int __init
2208 xfs_init_workqueues(void)
2209 {
2210         /*
2211          * The allocation workqueue can be used in memory reclaim situations
2212          * (writepage path), and parallelism is only limited by the number of
2213          * AGs in all the filesystems mounted. Hence use the default large
2214          * max_active value for this workqueue.
2215          */
2216         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2217                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2218         if (!xfs_alloc_wq)
2219                 return -ENOMEM;
2220
2221         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2222                         0);
2223         if (!xfs_discard_wq)
2224                 goto out_free_alloc_wq;
2225
2226         return 0;
2227 out_free_alloc_wq:
2228         destroy_workqueue(xfs_alloc_wq);
2229         return -ENOMEM;
2230 }
2231
2232 STATIC void
2233 xfs_destroy_workqueues(void)
2234 {
2235         destroy_workqueue(xfs_discard_wq);
2236         destroy_workqueue(xfs_alloc_wq);
2237 }
2238
2239 #ifdef CONFIG_HOTPLUG_CPU
2240 static int
2241 xfs_cpu_dead(
2242         unsigned int            cpu)
2243 {
2244         struct xfs_mount        *mp, *n;
2245
2246         spin_lock(&xfs_mount_list_lock);
2247         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2248                 spin_unlock(&xfs_mount_list_lock);
2249                 xfs_inodegc_cpu_dead(mp, cpu);
2250                 xlog_cil_pcp_dead(mp->m_log, cpu);
2251                 spin_lock(&xfs_mount_list_lock);
2252         }
2253         spin_unlock(&xfs_mount_list_lock);
2254         return 0;
2255 }
2256
2257 static int __init
2258 xfs_cpu_hotplug_init(void)
2259 {
2260         int     error;
2261
2262         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2263                         xfs_cpu_dead);
2264         if (error < 0)
2265                 xfs_alert(NULL,
2266 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2267                         error);
2268         return error;
2269 }
2270
2271 static void
2272 xfs_cpu_hotplug_destroy(void)
2273 {
2274         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2275 }
2276
2277 #else /* !CONFIG_HOTPLUG_CPU */
2278 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2279 static inline void xfs_cpu_hotplug_destroy(void) {}
2280 #endif
2281
2282 STATIC int __init
2283 init_xfs_fs(void)
2284 {
2285         int                     error;
2286
2287         xfs_check_ondisk_structs();
2288
2289         printk(KERN_INFO XFS_VERSION_STRING " with "
2290                          XFS_BUILD_OPTIONS " enabled\n");
2291
2292         xfs_dir_startup();
2293
2294         error = xfs_cpu_hotplug_init();
2295         if (error)
2296                 goto out;
2297
2298         error = xfs_init_caches();
2299         if (error)
2300                 goto out_destroy_hp;
2301
2302         error = xfs_init_workqueues();
2303         if (error)
2304                 goto out_destroy_caches;
2305
2306         error = xfs_mru_cache_init();
2307         if (error)
2308                 goto out_destroy_wq;
2309
2310         error = xfs_init_procfs();
2311         if (error)
2312                 goto out_mru_cache_uninit;
2313
2314         error = xfs_sysctl_register();
2315         if (error)
2316                 goto out_cleanup_procfs;
2317
2318         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2319         if (!xfs_kset) {
2320                 error = -ENOMEM;
2321                 goto out_sysctl_unregister;
2322         }
2323
2324         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2325
2326         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2327         if (!xfsstats.xs_stats) {
2328                 error = -ENOMEM;
2329                 goto out_kset_unregister;
2330         }
2331
2332         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2333                                "stats");
2334         if (error)
2335                 goto out_free_stats;
2336
2337 #ifdef DEBUG
2338         xfs_dbg_kobj.kobject.kset = xfs_kset;
2339         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2340         if (error)
2341                 goto out_remove_stats_kobj;
2342 #endif
2343
2344         error = xfs_qm_init();
2345         if (error)
2346                 goto out_remove_dbg_kobj;
2347
2348         error = register_filesystem(&xfs_fs_type);
2349         if (error)
2350                 goto out_qm_exit;
2351         return 0;
2352
2353  out_qm_exit:
2354         xfs_qm_exit();
2355  out_remove_dbg_kobj:
2356 #ifdef DEBUG
2357         xfs_sysfs_del(&xfs_dbg_kobj);
2358  out_remove_stats_kobj:
2359 #endif
2360         xfs_sysfs_del(&xfsstats.xs_kobj);
2361  out_free_stats:
2362         free_percpu(xfsstats.xs_stats);
2363  out_kset_unregister:
2364         kset_unregister(xfs_kset);
2365  out_sysctl_unregister:
2366         xfs_sysctl_unregister();
2367  out_cleanup_procfs:
2368         xfs_cleanup_procfs();
2369  out_mru_cache_uninit:
2370         xfs_mru_cache_uninit();
2371  out_destroy_wq:
2372         xfs_destroy_workqueues();
2373  out_destroy_caches:
2374         xfs_destroy_caches();
2375  out_destroy_hp:
2376         xfs_cpu_hotplug_destroy();
2377  out:
2378         return error;
2379 }
2380
2381 STATIC void __exit
2382 exit_xfs_fs(void)
2383 {
2384         xfs_qm_exit();
2385         unregister_filesystem(&xfs_fs_type);
2386 #ifdef DEBUG
2387         xfs_sysfs_del(&xfs_dbg_kobj);
2388 #endif
2389         xfs_sysfs_del(&xfsstats.xs_kobj);
2390         free_percpu(xfsstats.xs_stats);
2391         kset_unregister(xfs_kset);
2392         xfs_sysctl_unregister();
2393         xfs_cleanup_procfs();
2394         xfs_mru_cache_uninit();
2395         xfs_destroy_workqueues();
2396         xfs_destroy_caches();
2397         xfs_uuid_table_free();
2398         xfs_cpu_hotplug_destroy();
2399 }
2400
2401 module_init(init_xfs_fs);
2402 module_exit(exit_xfs_fs);
2403
2404 MODULE_AUTHOR("Silicon Graphics, Inc.");
2405 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2406 MODULE_LICENSE("GPL");