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_da_format.h"
  15 #include "xfs_inode.h"
  16 #include "xfs_btree.h"
  17 #include "xfs_bmap.h"
  18 #include "xfs_alloc.h"
  19 #include "xfs_error.h"
  20 #include "xfs_fsops.h"
  21 #include "xfs_trans.h"
  22 #include "xfs_buf_item.h"
  23 #include "xfs_log.h"
  24 #include "xfs_log_priv.h"
  25 #include "xfs_da_btree.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_extfree_item.h"
  28 #include "xfs_mru_cache.h"
  29 #include "xfs_inode_item.h"
  30 #include "xfs_icache.h"
  31 #include "xfs_trace.h"
  32 #include "xfs_icreate_item.h"
  33 #include "xfs_filestream.h"
  34 #include "xfs_quota.h"
  35 #include "xfs_sysfs.h"
  36 #include "xfs_ondisk.h"
  37 #include "xfs_rmap_item.h"
  38 #include "xfs_refcount_item.h"
  39 #include "xfs_bmap_item.h"
  40 #include "xfs_reflink.h"
  41
  42 #include <linux/namei.h>
  43 #include <linux/dax.h>
  44 #include <linux/init.h>
  45 #include <linux/slab.h>
  46 #include <linux/magic.h>
  47 #include <linux/mount.h>
  48 #include <linux/mempool.h>
  49 #include <linux/writeback.h>
  50 #include <linux/kthread.h>
  51 #include <linux/freezer.h>
  52 #include <linux/parser.h>
  53
  54 static const struct super_operations xfs_super_operations;
  55 struct bio_set xfs_ioend_bioset;
  56
  57 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  58 #ifdef DEBUG
  59 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  60 #endif
  61
  62 /*
  63  * Table driven mount option parser.
  64  */
  65 enum {
  66         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
  67         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
  68         Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
  69         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
  70         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
  71         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
  72         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
  73         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
  74         Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
  75 };
  76
  77 static const match_table_t tokens = {
  78         {Opt_logbufs,   "logbufs=%u"},  /* number of XFS log buffers */
  79         {Opt_logbsize,  "logbsize=%s"}, /* size of XFS log buffers */
  80         {Opt_logdev,    "logdev=%s"},   /* log device */
  81         {Opt_rtdev,     "rtdev=%s"},    /* realtime I/O device */
  82         {Opt_biosize,   "biosize=%u"},  /* log2 of preferred buffered io size */
  83         {Opt_wsync,     "wsync"},       /* safe-mode nfs compatible mount */
  84         {Opt_noalign,   "noalign"},     /* turn off stripe alignment */
  85         {Opt_swalloc,   "swalloc"},     /* turn on stripe width allocation */
  86         {Opt_sunit,     "sunit=%u"},    /* data volume stripe unit */
  87         {Opt_swidth,    "swidth=%u"},   /* data volume stripe width */
  88         {Opt_nouuid,    "nouuid"},      /* ignore filesystem UUID */
  89         {Opt_mtpt,      "mtpt"},        /* filesystem mount point */
  90         {Opt_grpid,     "grpid"},       /* group-ID from parent directory */
  91         {Opt_nogrpid,   "nogrpid"},     /* group-ID from current process */
  92         {Opt_bsdgroups, "bsdgroups"},   /* group-ID from parent directory */
  93         {Opt_sysvgroups,"sysvgroups"},  /* group-ID from current process */
  94         {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */
  95         {Opt_norecovery,"norecovery"},  /* don't run XFS recovery */
  96         {Opt_inode64,   "inode64"},     /* inodes can be allocated anywhere */
  97         {Opt_inode32,   "inode32"},     /* inode allocation limited to
  98                                          * XFS_MAXINUMBER_32 */
  99         {Opt_ikeep,     "ikeep"},       /* do not free empty inode clusters */
 100         {Opt_noikeep,   "noikeep"},     /* free empty inode clusters */
 101         {Opt_largeio,   "largeio"},     /* report large I/O sizes in stat() */
 102         {Opt_nolargeio, "nolargeio"},   /* do not report large I/O sizes
 103                                          * in stat(). */
 104         {Opt_attr2,     "attr2"},       /* do use attr2 attribute format */
 105         {Opt_noattr2,   "noattr2"},     /* do not use attr2 attribute format */
 106         {Opt_filestreams,"filestreams"},/* use filestreams allocator */
 107         {Opt_quota,     "quota"},       /* disk quotas (user) */
 108         {Opt_noquota,   "noquota"},     /* no quotas */
 109         {Opt_usrquota,  "usrquota"},    /* user quota enabled */
 110         {Opt_grpquota,  "grpquota"},    /* group quota enabled */
 111         {Opt_prjquota,  "prjquota"},    /* project quota enabled */
 112         {Opt_uquota,    "uquota"},      /* user quota (IRIX variant) */
 113         {Opt_gquota,    "gquota"},      /* group quota (IRIX variant) */
 114         {Opt_pquota,    "pquota"},      /* project quota (IRIX variant) */
 115         {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
 116         {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
 117         {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
 118         {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */
 119         {Opt_discard,   "discard"},     /* Discard unused blocks */
 120         {Opt_nodiscard, "nodiscard"},   /* Do not discard unused blocks */
 121         {Opt_dax,       "dax"},         /* Enable direct access to bdev pages */
 122         {Opt_err,       NULL},
 123 };
 124
 125
 126 STATIC int
 127 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
 128 {
 129         int     last, shift_left_factor = 0, _res;
 130         char    *value;
 131         int     ret = 0;
 132
 133         value = match_strdup(s);
 134         if (!value)
 135                 return -ENOMEM;
 136
 137         last = strlen(value) - 1;
 138         if (value[last] == 'K' || value[last] == 'k') {
 139                 shift_left_factor = 10;
 140                 value[last] = '\0';
 141         }
 142         if (value[last] == 'M' || value[last] == 'm') {
 143                 shift_left_factor = 20;
 144                 value[last] = '\0';
 145         }
 146         if (value[last] == 'G' || value[last] == 'g') {
 147                 shift_left_factor = 30;
 148                 value[last] = '\0';
 149         }
 150
 151         if (kstrtoint(value, base, &_res))
 152                 ret = -EINVAL;
 153         kfree(value);
 154         *res = _res << shift_left_factor;
 155         return ret;
 156 }
 157
 158 /*
 159  * This function fills in xfs_mount_t fields based on mount args.
 160  * Note: the superblock has _not_ yet been read in.
 161  *
 162  * Note that this function leaks the various device name allocations on
 163  * failure.  The caller takes care of them.
 164  *
 165  * *sb is const because this is also used to test options on the remount
 166  * path, and we don't want this to have any side effects at remount time.
 167  * Today this function does not change *sb, but just to future-proof...
 168  */
 169 STATIC int
 170 xfs_parseargs(
 171         struct xfs_mount        *mp,
 172         char                    *options)
 173 {
 174         const struct super_block *sb = mp->m_super;
 175         char                    *p;
 176         substring_t             args[MAX_OPT_ARGS];
 177         int                     dsunit = 0;
 178         int                     dswidth = 0;
 179         int                     iosize = 0;
 180         uint8_t                 iosizelog = 0;
 181
 182         /*
 183          * set up the mount name first so all the errors will refer to the
 184          * correct device.
 185          */
 186         mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
 187         if (!mp->m_fsname)
 188                 return -ENOMEM;
 189         mp->m_fsname_len = strlen(mp->m_fsname) + 1;
 190
 191         /*
 192          * Copy binary VFS mount flags we are interested in.
 193          */
 194         if (sb_rdonly(sb))
 195                 mp->m_flags |= XFS_MOUNT_RDONLY;
 196         if (sb->s_flags & SB_DIRSYNC)
 197                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 198         if (sb->s_flags & SB_SYNCHRONOUS)
 199                 mp->m_flags |= XFS_MOUNT_WSYNC;
 200
 201         /*
 202          * Set some default flags that could be cleared by the mount option
 203          * parsing.
 204          */
 205         mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 206
 207         /*
 208          * These can be overridden by the mount option parsing.
 209          */
 210         mp->m_logbufs = -1;
 211         mp->m_logbsize = -1;
 212
 213         if (!options)
 214                 goto done;
 215
 216         while ((p = strsep(&options, ",")) != NULL) {
 217                 int             token;
 218
 219                 if (!*p)
 220                         continue;
 221
 222                 token = match_token(p, tokens, args);
 223                 switch (token) {
 224                 case Opt_logbufs:
 225                         if (match_int(args, &mp->m_logbufs))
 226                                 return -EINVAL;
 227                         break;
 228                 case Opt_logbsize:
 229                         if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
 230                                 return -EINVAL;
 231                         break;
 232                 case Opt_logdev:
 233                         kfree(mp->m_logname);
 234                         mp->m_logname = match_strdup(args);
 235                         if (!mp->m_logname)
 236                                 return -ENOMEM;
 237                         break;
 238                 case Opt_mtpt:
 239                         xfs_warn(mp, "%s option not allowed on this system", p);
 240                         return -EINVAL;
 241                 case Opt_rtdev:
 242                         kfree(mp->m_rtname);
 243                         mp->m_rtname = match_strdup(args);
 244                         if (!mp->m_rtname)
 245                                 return -ENOMEM;
 246                         break;
 247                 case Opt_allocsize:
 248                 case Opt_biosize:
 249                         if (suffix_kstrtoint(args, 10, &iosize))
 250                                 return -EINVAL;
 251                         iosizelog = ffs(iosize) - 1;
 252                         break;
 253                 case Opt_grpid:
 254                 case Opt_bsdgroups:
 255                         mp->m_flags |= XFS_MOUNT_GRPID;
 256                         break;
 257                 case Opt_nogrpid:
 258                 case Opt_sysvgroups:
 259                         mp->m_flags &= ~XFS_MOUNT_GRPID;
 260                         break;
 261                 case Opt_wsync:
 262                         mp->m_flags |= XFS_MOUNT_WSYNC;
 263                         break;
 264                 case Opt_norecovery:
 265                         mp->m_flags |= XFS_MOUNT_NORECOVERY;
 266                         break;
 267                 case Opt_noalign:
 268                         mp->m_flags |= XFS_MOUNT_NOALIGN;
 269                         break;
 270                 case Opt_swalloc:
 271                         mp->m_flags |= XFS_MOUNT_SWALLOC;
 272                         break;
 273                 case Opt_sunit:
 274                         if (match_int(args, &dsunit))
 275                                 return -EINVAL;
 276                         break;
 277                 case Opt_swidth:
 278                         if (match_int(args, &dswidth))
 279                                 return -EINVAL;
 280                         break;
 281                 case Opt_inode32:
 282                         mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
 283                         break;
 284                 case Opt_inode64:
 285                         mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
 286                         break;
 287                 case Opt_nouuid:
 288                         mp->m_flags |= XFS_MOUNT_NOUUID;
 289                         break;
 290                 case Opt_ikeep:
 291                         mp->m_flags |= XFS_MOUNT_IKEEP;
 292                         break;
 293                 case Opt_noikeep:
 294                         mp->m_flags &= ~XFS_MOUNT_IKEEP;
 295                         break;
 296                 case Opt_largeio:
 297                         mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
 298                         break;
 299                 case Opt_nolargeio:
 300                         mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 301                         break;
 302                 case Opt_attr2:
 303                         mp->m_flags |= XFS_MOUNT_ATTR2;
 304                         break;
 305                 case Opt_noattr2:
 306                         mp->m_flags &= ~XFS_MOUNT_ATTR2;
 307                         mp->m_flags |= XFS_MOUNT_NOATTR2;
 308                         break;
 309                 case Opt_filestreams:
 310                         mp->m_flags |= XFS_MOUNT_FILESTREAMS;
 311                         break;
 312                 case Opt_noquota:
 313                         mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
 314                         mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
 315                         mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
 316                         break;
 317                 case Opt_quota:
 318                 case Opt_uquota:
 319                 case Opt_usrquota:
 320                         mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
 321                                          XFS_UQUOTA_ENFD);
 322                         break;
 323                 case Opt_qnoenforce:
 324                 case Opt_uqnoenforce:
 325                         mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
 326                         mp->m_qflags &= ~XFS_UQUOTA_ENFD;
 327                         break;
 328                 case Opt_pquota:
 329                 case Opt_prjquota:
 330                         mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
 331                                          XFS_PQUOTA_ENFD);
 332                         break;
 333                 case Opt_pqnoenforce:
 334                         mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
 335                         mp->m_qflags &= ~XFS_PQUOTA_ENFD;
 336                         break;
 337                 case Opt_gquota:
 338                 case Opt_grpquota:
 339                         mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
 340                                          XFS_GQUOTA_ENFD);
 341                         break;
 342                 case Opt_gqnoenforce:
 343                         mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
 344                         mp->m_qflags &= ~XFS_GQUOTA_ENFD;
 345                         break;
 346                 case Opt_discard:
 347                         mp->m_flags |= XFS_MOUNT_DISCARD;
 348                         break;
 349                 case Opt_nodiscard:
 350                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
 351                         break;
 352 #ifdef CONFIG_FS_DAX
 353                 case Opt_dax:
 354                         mp->m_flags |= XFS_MOUNT_DAX;
 355                         break;
 356 #endif
 357                 default:
 358                         xfs_warn(mp, "unknown mount option [%s].", p);
 359                         return -EINVAL;
 360                 }
 361         }
 362
 363         /*
 364          * no recovery flag requires a read-only mount
 365          */
 366         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
 367             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
 368                 xfs_warn(mp, "no-recovery mounts must be read-only.");
 369                 return -EINVAL;
 370         }
 371
 372         if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
 373                 xfs_warn(mp,
 374         "sunit and swidth options incompatible with the noalign option");
 375                 return -EINVAL;
 376         }
 377
 378 #ifndef CONFIG_XFS_QUOTA
 379         if (XFS_IS_QUOTA_RUNNING(mp)) {
 380                 xfs_warn(mp, "quota support not available in this kernel.");
 381                 return -EINVAL;
 382         }
 383 #endif
 384
 385         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
 386                 xfs_warn(mp, "sunit and swidth must be specified together");
 387                 return -EINVAL;
 388         }
 389
 390         if (dsunit && (dswidth % dsunit != 0)) {
 391                 xfs_warn(mp,
 392         "stripe width (%d) must be a multiple of the stripe unit (%d)",
 393                         dswidth, dsunit);
 394                 return -EINVAL;
 395         }
 396
 397 done:
 398         if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
 399                 /*
 400                  * At this point the superblock has not been read
 401                  * in, therefore we do not know the block size.
 402                  * Before the mount call ends we will convert
 403                  * these to FSBs.
 404                  */
 405                 mp->m_dalign = dsunit;
 406                 mp->m_swidth = dswidth;
 407         }
 408
 409         if (mp->m_logbufs != -1 &&
 410             mp->m_logbufs != 0 &&
 411             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
 412              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
 413                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
 414                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 415                 return -EINVAL;
 416         }
 417         if (mp->m_logbsize != -1 &&
 418             mp->m_logbsize !=  0 &&
 419             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
 420              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
 421              !is_power_of_2(mp->m_logbsize))) {
 422                 xfs_warn(mp,
 423                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 424                         mp->m_logbsize);
 425                 return -EINVAL;
 426         }
 427
 428         if (iosizelog) {
 429                 if (iosizelog > XFS_MAX_IO_LOG ||
 430                     iosizelog < XFS_MIN_IO_LOG) {
 431                         xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
 432                                 iosizelog, XFS_MIN_IO_LOG,
 433                                 XFS_MAX_IO_LOG);
 434                         return -EINVAL;
 435                 }
 436
 437                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 438                 mp->m_readio_log = iosizelog;
 439                 mp->m_writeio_log = iosizelog;
 440         }
 441
 442         return 0;
 443 }
 444
 445 struct proc_xfs_info {
 446         uint64_t        flag;
 447         char            *str;
 448 };
 449
 450 STATIC int
 451 xfs_showargs(
 452         struct xfs_mount        *mp,
 453         struct seq_file         *m)
 454 {
 455         static struct proc_xfs_info xfs_info_set[] = {
 456                 /* the few simple ones we can get from the mount struct */
 457                 { XFS_MOUNT_IKEEP,              ",ikeep" },
 458                 { XFS_MOUNT_WSYNC,              ",wsync" },
 459                 { XFS_MOUNT_NOALIGN,            ",noalign" },
 460                 { XFS_MOUNT_SWALLOC,            ",swalloc" },
 461                 { XFS_MOUNT_NOUUID,             ",nouuid" },
 462                 { XFS_MOUNT_NORECOVERY,         ",norecovery" },
 463                 { XFS_MOUNT_ATTR2,              ",attr2" },
 464                 { XFS_MOUNT_FILESTREAMS,        ",filestreams" },
 465                 { XFS_MOUNT_GRPID,              ",grpid" },
 466                 { XFS_MOUNT_DISCARD,            ",discard" },
 467                 { XFS_MOUNT_SMALL_INUMS,        ",inode32" },
 468                 { XFS_MOUNT_DAX,                ",dax" },
 469                 { 0, NULL }
 470         };
 471         static struct proc_xfs_info xfs_info_unset[] = {
 472                 /* the few simple ones we can get from the mount struct */
 473                 { XFS_MOUNT_COMPAT_IOSIZE,      ",largeio" },
 474                 { XFS_MOUNT_SMALL_INUMS,        ",inode64" },
 475                 { 0, NULL }
 476         };
 477         struct proc_xfs_info    *xfs_infop;
 478
 479         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 480                 if (mp->m_flags & xfs_infop->flag)
 481                         seq_puts(m, xfs_infop->str);
 482         }
 483         for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
 484                 if (!(mp->m_flags & xfs_infop->flag))
 485                         seq_puts(m, xfs_infop->str);
 486         }
 487
 488         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
 489                 seq_printf(m, ",allocsize=%dk",
 490                                 (int)(1 << mp->m_writeio_log) >> 10);
 491
 492         if (mp->m_logbufs > 0)
 493                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 494         if (mp->m_logbsize > 0)
 495                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 496
 497         if (mp->m_logname)
 498                 seq_show_option(m, "logdev", mp->m_logname);
 499         if (mp->m_rtname)
 500                 seq_show_option(m, "rtdev", mp->m_rtname);
 501
 502         if (mp->m_dalign > 0)
 503                 seq_printf(m, ",sunit=%d",
 504                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 505         if (mp->m_swidth > 0)
 506                 seq_printf(m, ",swidth=%d",
 507                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 508
 509         if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
 510                 seq_puts(m, ",usrquota");
 511         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 512                 seq_puts(m, ",uqnoenforce");
 513
 514         if (mp->m_qflags & XFS_PQUOTA_ACCT) {
 515                 if (mp->m_qflags & XFS_PQUOTA_ENFD)
 516                         seq_puts(m, ",prjquota");
 517                 else
 518                         seq_puts(m, ",pqnoenforce");
 519         }
 520         if (mp->m_qflags & XFS_GQUOTA_ACCT) {
 521                 if (mp->m_qflags & XFS_GQUOTA_ENFD)
 522                         seq_puts(m, ",grpquota");
 523                 else
 524                         seq_puts(m, ",gqnoenforce");
 525         }
 526
 527         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 528                 seq_puts(m, ",noquota");
 529
 530         return 0;
 531 }
 532 static uint64_t
 533 xfs_max_file_offset(
 534         unsigned int            blockshift)
 535 {
 536         unsigned int            pagefactor = 1;
 537         unsigned int            bitshift = BITS_PER_LONG - 1;
 538
 539         /* Figure out maximum filesize, on Linux this can depend on
 540          * the filesystem blocksize (on 32 bit platforms).
 541          * __block_write_begin does this in an [unsigned] long...
 542          *      page->index << (PAGE_SHIFT - bbits)
 543          * So, for page sized blocks (4K on 32 bit platforms),
 544          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 545          *      (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
 546          * but for smaller blocksizes it is less (bbits = log2 bsize).
 547          * Note1: get_block_t takes a long (implicit cast from above)
 548          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
 549          * can optionally convert the [unsigned] long from above into
 550          * an [unsigned] long long.
 551          */
 552
 553 #if BITS_PER_LONG == 32
 554 # if defined(CONFIG_LBDAF)
 555         ASSERT(sizeof(sector_t) == 8);
 556         pagefactor = PAGE_SIZE;
 557         bitshift = BITS_PER_LONG;
 558 # else
 559         pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
 560 # endif
 561 #endif
 562
 563         return (((uint64_t)pagefactor) << bitshift) - 1;
 564 }
 565
 566 /*
 567  * Set parameters for inode allocation heuristics, taking into account
 568  * filesystem size and inode32/inode64 mount options; i.e. specifically
 569  * whether or not XFS_MOUNT_SMALL_INUMS is set.
 570  *
 571  * Inode allocation patterns are altered only if inode32 is requested
 572  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
 573  * If altered, XFS_MOUNT_32BITINODES is set as well.
 574  *
 575  * An agcount independent of that in the mount structure is provided
 576  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 577  * to the potentially higher ag count.
 578  *
 579  * Returns the maximum AG index which may contain inodes.
 580  */
 581 xfs_agnumber_t
 582 xfs_set_inode_alloc(
 583         struct xfs_mount *mp,
 584         xfs_agnumber_t  agcount)
 585 {
 586         xfs_agnumber_t  index;
 587         xfs_agnumber_t  maxagi = 0;
 588         xfs_sb_t        *sbp = &mp->m_sb;
 589         xfs_agnumber_t  max_metadata;
 590         xfs_agino_t     agino;
 591         xfs_ino_t       ino;
 592
 593         /*
 594          * Calculate how much should be reserved for inodes to meet
 595          * the max inode percentage.  Used only for inode32.
 596          */
 597         if (mp->m_maxicount) {
 598                 uint64_t        icount;
 599
 600                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 601                 do_div(icount, 100);
 602                 icount += sbp->sb_agblocks - 1;
 603                 do_div(icount, sbp->sb_agblocks);
 604                 max_metadata = icount;
 605         } else {
 606                 max_metadata = agcount;
 607         }
 608
 609         /* Get the last possible inode in the filesystem */
 610         agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
 611         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 612
 613         /*
 614          * If user asked for no more than 32-bit inodes, and the fs is
 615          * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
 616          * the allocator to accommodate the request.
 617          */
 618         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
 619                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 620         else
 621                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
 622
 623         for (index = 0; index < agcount; index++) {
 624                 struct xfs_perag        *pag;
 625
 626                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 627
 628                 pag = xfs_perag_get(mp, index);
 629
 630                 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
 631                         if (ino > XFS_MAXINUMBER_32) {
 632                                 pag->pagi_inodeok = 0;
 633                                 pag->pagf_metadata = 0;
 634                         } else {
 635                                 pag->pagi_inodeok = 1;
 636                                 maxagi++;
 637                                 if (index < max_metadata)
 638                                         pag->pagf_metadata = 1;
 639                                 else
 640                                         pag->pagf_metadata = 0;
 641                         }
 642                 } else {
 643                         pag->pagi_inodeok = 1;
 644                         pag->pagf_metadata = 0;
 645                 }
 646
 647                 xfs_perag_put(pag);
 648         }
 649
 650         return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
 651 }
 652
 653 STATIC int
 654 xfs_blkdev_get(
 655         xfs_mount_t             *mp,
 656         const char              *name,
 657         struct block_device     **bdevp)
 658 {
 659         int                     error = 0;
 660
 661         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 662                                     mp);
 663         if (IS_ERR(*bdevp)) {
 664                 error = PTR_ERR(*bdevp);
 665                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 666         }
 667
 668         return error;
 669 }
 670
 671 STATIC void
 672 xfs_blkdev_put(
 673         struct block_device     *bdev)
 674 {
 675         if (bdev)
 676                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 677 }
 678
 679 void
 680 xfs_blkdev_issue_flush(
 681         xfs_buftarg_t           *buftarg)
 682 {
 683         blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
 684 }
 685
 686 STATIC void
 687 xfs_close_devices(
 688         struct xfs_mount        *mp)
 689 {
 690         struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
 691
 692         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 693                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 694                 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
 695
 696                 xfs_free_buftarg(mp->m_logdev_targp);
 697                 xfs_blkdev_put(logdev);
 698                 fs_put_dax(dax_logdev);
 699         }
 700         if (mp->m_rtdev_targp) {
 701                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 702                 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
 703
 704                 xfs_free_buftarg(mp->m_rtdev_targp);
 705                 xfs_blkdev_put(rtdev);
 706                 fs_put_dax(dax_rtdev);
 707         }
 708         xfs_free_buftarg(mp->m_ddev_targp);
 709         fs_put_dax(dax_ddev);
 710 }
 711
 712 /*
 713  * The file system configurations are:
 714  *      (1) device (partition) with data and internal log
 715  *      (2) logical volume with data and log subvolumes.
 716  *      (3) logical volume with data, log, and realtime subvolumes.
 717  *
 718  * We only have to handle opening the log and realtime volumes here if
 719  * they are present.  The data subvolume has already been opened by
 720  * get_sb_bdev() and is stored in sb->s_bdev.
 721  */
 722 STATIC int
 723 xfs_open_devices(
 724         struct xfs_mount        *mp)
 725 {
 726         struct block_device     *ddev = mp->m_super->s_bdev;
 727         struct dax_device       *dax_ddev = fs_dax_get_by_bdev(ddev);
 728         struct dax_device       *dax_logdev = NULL, *dax_rtdev = NULL;
 729         struct block_device     *logdev = NULL, *rtdev = NULL;
 730         int                     error;
 731
 732         /*
 733          * Open real time and log devices - order is important.
 734          */
 735         if (mp->m_logname) {
 736                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 737                 if (error)
 738                         goto out;
 739                 dax_logdev = fs_dax_get_by_bdev(logdev);
 740         }
 741
 742         if (mp->m_rtname) {
 743                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 744                 if (error)
 745                         goto out_close_logdev;
 746
 747                 if (rtdev == ddev || rtdev == logdev) {
 748                         xfs_warn(mp,
 749         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 750                         error = -EINVAL;
 751                         goto out_close_rtdev;
 752                 }
 753                 dax_rtdev = fs_dax_get_by_bdev(rtdev);
 754         }
 755
 756         /*
 757          * Setup xfs_mount buffer target pointers
 758          */
 759         error = -ENOMEM;
 760         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
 761         if (!mp->m_ddev_targp)
 762                 goto out_close_rtdev;
 763
 764         if (rtdev) {
 765                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
 766                 if (!mp->m_rtdev_targp)
 767                         goto out_free_ddev_targ;
 768         }
 769
 770         if (logdev && logdev != ddev) {
 771                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
 772                 if (!mp->m_logdev_targp)
 773                         goto out_free_rtdev_targ;
 774         } else {
 775                 mp->m_logdev_targp = mp->m_ddev_targp;
 776         }
 777
 778         return 0;
 779
 780  out_free_rtdev_targ:
 781         if (mp->m_rtdev_targp)
 782                 xfs_free_buftarg(mp->m_rtdev_targp);
 783  out_free_ddev_targ:
 784         xfs_free_buftarg(mp->m_ddev_targp);
 785  out_close_rtdev:
 786         xfs_blkdev_put(rtdev);
 787         fs_put_dax(dax_rtdev);
 788  out_close_logdev:
 789         if (logdev && logdev != ddev) {
 790                 xfs_blkdev_put(logdev);
 791                 fs_put_dax(dax_logdev);
 792         }
 793  out:
 794         fs_put_dax(dax_ddev);
 795         return error;
 796 }
 797
 798 /*
 799  * Setup xfs_mount buffer target pointers based on superblock
 800  */
 801 STATIC int
 802 xfs_setup_devices(
 803         struct xfs_mount        *mp)
 804 {
 805         int                     error;
 806
 807         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 808         if (error)
 809                 return error;
 810
 811         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 812                 unsigned int    log_sector_size = BBSIZE;
 813
 814                 if (xfs_sb_version_hassector(&mp->m_sb))
 815                         log_sector_size = mp->m_sb.sb_logsectsize;
 816                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 817                                             log_sector_size);
 818                 if (error)
 819                         return error;
 820         }
 821         if (mp->m_rtdev_targp) {
 822                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 823                                             mp->m_sb.sb_sectsize);
 824                 if (error)
 825                         return error;
 826         }
 827
 828         return 0;
 829 }
 830
 831 STATIC int
 832 xfs_init_mount_workqueues(
 833         struct xfs_mount        *mp)
 834 {
 835         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 836                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
 837         if (!mp->m_buf_workqueue)
 838                 goto out;
 839
 840         mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
 841                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 842         if (!mp->m_data_workqueue)
 843                 goto out_destroy_buf;
 844
 845         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 846                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 847         if (!mp->m_unwritten_workqueue)
 848                 goto out_destroy_data_iodone_queue;
 849
 850         mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
 851                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 852         if (!mp->m_cil_workqueue)
 853                 goto out_destroy_unwritten;
 854
 855         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 856                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 857         if (!mp->m_reclaim_workqueue)
 858                 goto out_destroy_cil;
 859
 860         mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
 861                         WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
 862                         mp->m_fsname);
 863         if (!mp->m_log_workqueue)
 864                 goto out_destroy_reclaim;
 865
 866         mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
 867                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 868         if (!mp->m_eofblocks_workqueue)
 869                 goto out_destroy_log;
 870
 871         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
 872                                                mp->m_fsname);
 873         if (!mp->m_sync_workqueue)
 874                 goto out_destroy_eofb;
 875
 876         return 0;
 877
 878 out_destroy_eofb:
 879         destroy_workqueue(mp->m_eofblocks_workqueue);
 880 out_destroy_log:
 881         destroy_workqueue(mp->m_log_workqueue);
 882 out_destroy_reclaim:
 883         destroy_workqueue(mp->m_reclaim_workqueue);
 884 out_destroy_cil:
 885         destroy_workqueue(mp->m_cil_workqueue);
 886 out_destroy_unwritten:
 887         destroy_workqueue(mp->m_unwritten_workqueue);
 888 out_destroy_data_iodone_queue:
 889         destroy_workqueue(mp->m_data_workqueue);
 890 out_destroy_buf:
 891         destroy_workqueue(mp->m_buf_workqueue);
 892 out:
 893         return -ENOMEM;
 894 }
 895
 896 STATIC void
 897 xfs_destroy_mount_workqueues(
 898         struct xfs_mount        *mp)
 899 {
 900         destroy_workqueue(mp->m_sync_workqueue);
 901         destroy_workqueue(mp->m_eofblocks_workqueue);
 902         destroy_workqueue(mp->m_log_workqueue);
 903         destroy_workqueue(mp->m_reclaim_workqueue);
 904         destroy_workqueue(mp->m_cil_workqueue);
 905         destroy_workqueue(mp->m_data_workqueue);
 906         destroy_workqueue(mp->m_unwritten_workqueue);
 907         destroy_workqueue(mp->m_buf_workqueue);
 908 }
 909
 910 /*
 911  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 912  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 913  * for IO to complete so that we effectively throttle multiple callers to the
 914  * rate at which IO is completing.
 915  */
 916 void
 917 xfs_flush_inodes(
 918         struct xfs_mount        *mp)
 919 {
 920         struct super_block      *sb = mp->m_super;
 921
 922         if (down_read_trylock(&sb->s_umount)) {
 923                 sync_inodes_sb(sb);
 924                 up_read(&sb->s_umount);
 925         }
 926 }
 927
 928 /* Catch misguided souls that try to use this interface on XFS */
 929 STATIC struct inode *
 930 xfs_fs_alloc_inode(
 931         struct super_block      *sb)
 932 {
 933         BUG();
 934         return NULL;
 935 }
 936
 937 #ifdef DEBUG
 938 static void
 939 xfs_check_delalloc(
 940         struct xfs_inode        *ip,
 941         int                     whichfork)
 942 {
 943         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 944         struct xfs_bmbt_irec    got;
 945         struct xfs_iext_cursor  icur;
 946
 947         if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
 948                 return;
 949         do {
 950                 if (isnullstartblock(got.br_startblock)) {
 951                         xfs_warn(ip->i_mount,
 952         "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
 953                                 ip->i_ino,
 954                                 whichfork == XFS_DATA_FORK ? "data" : "cow",
 955                                 got.br_startoff, got.br_blockcount);
 956                 }
 957         } while (xfs_iext_next_extent(ifp, &icur, &got));
 958 }
 959 #else
 960 #define xfs_check_delalloc(ip, whichfork)       do { } while (0)
 961 #endif
 962
 963 /*
 964  * Now that the generic code is guaranteed not to be accessing
 965  * the linux inode, we can inactivate and reclaim the inode.
 966  */
 967 STATIC void
 968 xfs_fs_destroy_inode(
 969         struct inode            *inode)
 970 {
 971         struct xfs_inode        *ip = XFS_I(inode);
 972
 973         trace_xfs_destroy_inode(ip);
 974
 975         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 976         XFS_STATS_INC(ip->i_mount, vn_rele);
 977         XFS_STATS_INC(ip->i_mount, vn_remove);
 978
 979         xfs_inactive(ip);
 980
 981         if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
 982                 xfs_check_delalloc(ip, XFS_DATA_FORK);
 983                 xfs_check_delalloc(ip, XFS_COW_FORK);
 984                 ASSERT(0);
 985         }
 986
 987         XFS_STATS_INC(ip->i_mount, vn_reclaim);
 988
 989         /*
 990          * We should never get here with one of the reclaim flags already set.
 991          */
 992         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
 993         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
 994
 995         /*
 996          * We always use background reclaim here because even if the
 997          * inode is clean, it still may be under IO and hence we have
 998          * to take the flush lock. The background reclaim path handles
 999          * this more efficiently than we can here, so simply let background
1000          * reclaim tear down all inodes.
1001          */
1002         xfs_inode_set_reclaim_tag(ip);
1003 }
1004
1005 static void
1006 xfs_fs_dirty_inode(
1007         struct inode                    *inode,
1008         int                             flag)
1009 {
1010         struct xfs_inode                *ip = XFS_I(inode);
1011         struct xfs_mount                *mp = ip->i_mount;
1012         struct xfs_trans                *tp;
1013
1014         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
1015                 return;
1016         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
1017                 return;
1018
1019         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
1020                 return;
1021         xfs_ilock(ip, XFS_ILOCK_EXCL);
1022         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1023         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1024         xfs_trans_commit(tp);
1025 }
1026
1027 /*
1028  * Slab object creation initialisation for the XFS inode.
1029  * This covers only the idempotent fields in the XFS inode;
1030  * all other fields need to be initialised on allocation
1031  * from the slab. This avoids the need to repeatedly initialise
1032  * fields in the xfs inode that left in the initialise state
1033  * when freeing the inode.
1034  */
1035 STATIC void
1036 xfs_fs_inode_init_once(
1037         void                    *inode)
1038 {
1039         struct xfs_inode        *ip = inode;
1040
1041         memset(ip, 0, sizeof(struct xfs_inode));
1042
1043         /* vfs inode */
1044         inode_init_once(VFS_I(ip));
1045
1046         /* xfs inode */
1047         atomic_set(&ip->i_pincount, 0);
1048         spin_lock_init(&ip->i_flags_lock);
1049
1050         mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1051                      "xfsino", ip->i_ino);
1052         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1053                      "xfsino", ip->i_ino);
1054 }
1055
1056 /*
1057  * We do an unlocked check for XFS_IDONTCACHE here because we are already
1058  * serialised against cache hits here via the inode->i_lock and igrab() in
1059  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1060  * racing with us, and it avoids needing to grab a spinlock here for every inode
1061  * we drop the final reference on.
1062  */
1063 STATIC int
1064 xfs_fs_drop_inode(
1065         struct inode            *inode)
1066 {
1067         struct xfs_inode        *ip = XFS_I(inode);
1068
1069         /*
1070          * If this unlinked inode is in the middle of recovery, don't
1071          * drop the inode just yet; log recovery will take care of
1072          * that.  See the comment for this inode flag.
1073          */
1074         if (ip->i_flags & XFS_IRECOVERY) {
1075                 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1076                 return 0;
1077         }
1078
1079         return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1080 }
1081
1082 STATIC void
1083 xfs_free_fsname(
1084         struct xfs_mount        *mp)
1085 {
1086         kfree(mp->m_fsname);
1087         kfree(mp->m_rtname);
1088         kfree(mp->m_logname);
1089 }
1090
1091 STATIC int
1092 xfs_fs_sync_fs(
1093         struct super_block      *sb,
1094         int                     wait)
1095 {
1096         struct xfs_mount        *mp = XFS_M(sb);
1097
1098         /*
1099          * Doing anything during the async pass would be counterproductive.
1100          */
1101         if (!wait)
1102                 return 0;
1103
1104         xfs_log_force(mp, XFS_LOG_SYNC);
1105         if (laptop_mode) {
1106                 /*
1107                  * The disk must be active because we're syncing.
1108                  * We schedule log work now (now that the disk is
1109                  * active) instead of later (when it might not be).
1110                  */
1111                 flush_delayed_work(&mp->m_log->l_work);
1112         }
1113
1114         return 0;
1115 }
1116
1117 STATIC int
1118 xfs_fs_statfs(
1119         struct dentry           *dentry,
1120         struct kstatfs          *statp)
1121 {
1122         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
1123         xfs_sb_t                *sbp = &mp->m_sb;
1124         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
1125         uint64_t                fakeinos, id;
1126         uint64_t                icount;
1127         uint64_t                ifree;
1128         uint64_t                fdblocks;
1129         xfs_extlen_t            lsize;
1130         int64_t                 ffree;
1131
1132         statp->f_type = XFS_SUPER_MAGIC;
1133         statp->f_namelen = MAXNAMELEN - 1;
1134
1135         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1136         statp->f_fsid.val[0] = (u32)id;
1137         statp->f_fsid.val[1] = (u32)(id >> 32);
1138
1139         icount = percpu_counter_sum(&mp->m_icount);
1140         ifree = percpu_counter_sum(&mp->m_ifree);
1141         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1142
1143         spin_lock(&mp->m_sb_lock);
1144         statp->f_bsize = sbp->sb_blocksize;
1145         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1146         statp->f_blocks = sbp->sb_dblocks - lsize;
1147         spin_unlock(&mp->m_sb_lock);
1148
1149         statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1150         statp->f_bavail = statp->f_bfree;
1151
1152         fakeinos = statp->f_bfree << sbp->sb_inopblog;
1153         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
1154         if (mp->m_maxicount)
1155                 statp->f_files = min_t(typeof(statp->f_files),
1156                                         statp->f_files,
1157                                         mp->m_maxicount);
1158
1159         /* If sb_icount overshot maxicount, report actual allocation */
1160         statp->f_files = max_t(typeof(statp->f_files),
1161                                         statp->f_files,
1162                                         sbp->sb_icount);
1163
1164         /* make sure statp->f_ffree does not underflow */
1165         ffree = statp->f_files - (icount - ifree);
1166         statp->f_ffree = max_t(int64_t, ffree, 0);
1167
1168
1169         if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1170             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1171                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1172                 xfs_qm_statvfs(ip, statp);
1173
1174         if (XFS_IS_REALTIME_MOUNT(mp) &&
1175             (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
1176                 statp->f_blocks = sbp->sb_rblocks;
1177                 statp->f_bavail = statp->f_bfree =
1178                         sbp->sb_frextents * sbp->sb_rextsize;
1179         }
1180
1181         return 0;
1182 }
1183
1184 STATIC void
1185 xfs_save_resvblks(struct xfs_mount *mp)
1186 {
1187         uint64_t resblks = 0;
1188
1189         mp->m_resblks_save = mp->m_resblks;
1190         xfs_reserve_blocks(mp, &resblks, NULL);
1191 }
1192
1193 STATIC void
1194 xfs_restore_resvblks(struct xfs_mount *mp)
1195 {
1196         uint64_t resblks;
1197
1198         if (mp->m_resblks_save) {
1199                 resblks = mp->m_resblks_save;
1200                 mp->m_resblks_save = 0;
1201         } else
1202                 resblks = xfs_default_resblks(mp);
1203
1204         xfs_reserve_blocks(mp, &resblks, NULL);
1205 }
1206
1207 /*
1208  * Trigger writeback of all the dirty metadata in the file system.
1209  *
1210  * This ensures that the metadata is written to their location on disk rather
1211  * than just existing in transactions in the log. This means after a quiesce
1212  * there is no log replay required to write the inodes to disk - this is the
1213  * primary difference between a sync and a quiesce.
1214  *
1215  * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1216  * it is started again when appropriate.
1217  */
1218 void
1219 xfs_quiesce_attr(
1220         struct xfs_mount        *mp)
1221 {
1222         int     error = 0;
1223
1224         /* wait for all modifications to complete */
1225         while (atomic_read(&mp->m_active_trans) > 0)
1226                 delay(100);
1227
1228         /* force the log to unpin objects from the now complete transactions */
1229         xfs_log_force(mp, XFS_LOG_SYNC);
1230
1231         /* reclaim inodes to do any IO before the freeze completes */
1232         xfs_reclaim_inodes(mp, 0);
1233         xfs_reclaim_inodes(mp, SYNC_WAIT);
1234
1235         /* Push the superblock and write an unmount record */
1236         error = xfs_log_sbcount(mp);
1237         if (error)
1238                 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1239                                 "Frozen image may not be consistent.");
1240         /*
1241          * Just warn here till VFS can correctly support
1242          * read-only remount without racing.
1243          */
1244         WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1245
1246         xfs_log_quiesce(mp);
1247 }
1248
1249 STATIC int
1250 xfs_test_remount_options(
1251         struct super_block      *sb,
1252         char                    *options)
1253 {
1254         int                     error = 0;
1255         struct xfs_mount        *tmp_mp;
1256
1257         tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1258         if (!tmp_mp)
1259                 return -ENOMEM;
1260
1261         tmp_mp->m_super = sb;
1262         error = xfs_parseargs(tmp_mp, options);
1263         xfs_free_fsname(tmp_mp);
1264         kmem_free(tmp_mp);
1265
1266         return error;
1267 }
1268
1269 STATIC int
1270 xfs_fs_remount(
1271         struct super_block      *sb,
1272         int                     *flags,
1273         char                    *options)
1274 {
1275         struct xfs_mount        *mp = XFS_M(sb);
1276         xfs_sb_t                *sbp = &mp->m_sb;
1277         substring_t             args[MAX_OPT_ARGS];
1278         char                    *p;
1279         int                     error;
1280
1281         /* First, check for complete junk; i.e. invalid options */
1282         error = xfs_test_remount_options(sb, options);
1283         if (error)
1284                 return error;
1285
1286         sync_filesystem(sb);
1287         while ((p = strsep(&options, ",")) != NULL) {
1288                 int token;
1289
1290                 if (!*p)
1291                         continue;
1292
1293                 token = match_token(p, tokens, args);
1294                 switch (token) {
1295                 case Opt_inode64:
1296                         mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1297                         mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1298                         break;
1299                 case Opt_inode32:
1300                         mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1301                         mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1302                         break;
1303                 default:
1304                         /*
1305                          * Logically we would return an error here to prevent
1306                          * users from believing they might have changed
1307                          * mount options using remount which can't be changed.
1308                          *
1309                          * But unfortunately mount(8) adds all options from
1310                          * mtab and fstab to the mount arguments in some cases
1311                          * so we can't blindly reject options, but have to
1312                          * check for each specified option if it actually
1313                          * differs from the currently set option and only
1314                          * reject it if that's the case.
1315                          *
1316                          * Until that is implemented we return success for
1317                          * every remount request, and silently ignore all
1318                          * options that we can't actually change.
1319                          */
1320 #if 0
1321                         xfs_info(mp,
1322                 "mount option \"%s\" not supported for remount", p);
1323                         return -EINVAL;
1324 #else
1325                         break;
1326 #endif
1327                 }
1328         }
1329
1330         /* ro -> rw */
1331         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & SB_RDONLY)) {
1332                 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1333                         xfs_warn(mp,
1334                 "ro->rw transition prohibited on norecovery mount");
1335                         return -EINVAL;
1336                 }
1337
1338                 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1339                     xfs_sb_has_ro_compat_feature(sbp,
1340                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1341                         xfs_warn(mp,
1342 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1343                                 (sbp->sb_features_ro_compat &
1344                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1345                         return -EINVAL;
1346                 }
1347
1348                 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1349
1350                 /*
1351                  * If this is the first remount to writeable state we
1352                  * might have some superblock changes to update.
1353                  */
1354                 if (mp->m_update_sb) {
1355                         error = xfs_sync_sb(mp, false);
1356                         if (error) {
1357                                 xfs_warn(mp, "failed to write sb changes");
1358                                 return error;
1359                         }
1360                         mp->m_update_sb = false;
1361                 }
1362
1363                 /*
1364                  * Fill out the reserve pool if it is empty. Use the stashed
1365                  * value if it is non-zero, otherwise go with the default.
1366                  */
1367                 xfs_restore_resvblks(mp);
1368                 xfs_log_work_queue(mp);
1369
1370                 /* Recover any CoW blocks that never got remapped. */
1371                 error = xfs_reflink_recover_cow(mp);
1372                 if (error) {
1373                         xfs_err(mp,
1374         "Error %d recovering leftover CoW allocations.", error);
1375                         xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1376                         return error;
1377                 }
1378                 xfs_icache_enable_reclaim(mp);
1379
1380                 /* Create the per-AG metadata reservation pool .*/
1381                 error = xfs_fs_reserve_ag_blocks(mp);
1382                 if (error && error != -ENOSPC)
1383                         return error;
1384         }
1385
1386         /* rw -> ro */
1387         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & SB_RDONLY)) {
1388                 /*
1389                  * Cancel background eofb scanning so it cannot race with the
1390                  * final log force+buftarg wait and deadlock the remount.
1391                  */
1392                 xfs_icache_disable_reclaim(mp);
1393
1394                 /* Get rid of any leftover CoW reservations... */
1395                 error = xfs_icache_free_cowblocks(mp, NULL);
1396                 if (error) {
1397                         xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1398                         return error;
1399                 }
1400
1401                 /* Free the per-AG metadata reservation pool. */
1402                 error = xfs_fs_unreserve_ag_blocks(mp);
1403                 if (error) {
1404                         xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1405                         return error;
1406                 }
1407
1408                 /*
1409                  * Before we sync the metadata, we need to free up the reserve
1410                  * block pool so that the used block count in the superblock on
1411                  * disk is correct at the end of the remount. Stash the current
1412                  * reserve pool size so that if we get remounted rw, we can
1413                  * return it to the same size.
1414                  */
1415                 xfs_save_resvblks(mp);
1416
1417                 xfs_quiesce_attr(mp);
1418                 mp->m_flags |= XFS_MOUNT_RDONLY;
1419         }
1420
1421         return 0;
1422 }
1423
1424 /*
1425  * Second stage of a freeze. The data is already frozen so we only
1426  * need to take care of the metadata. Once that's done sync the superblock
1427  * to the log to dirty it in case of a crash while frozen. This ensures that we
1428  * will recover the unlinked inode lists on the next mount.
1429  */
1430 STATIC int
1431 xfs_fs_freeze(
1432         struct super_block      *sb)
1433 {
1434         struct xfs_mount        *mp = XFS_M(sb);
1435
1436         xfs_icache_disable_reclaim(mp);
1437         xfs_save_resvblks(mp);
1438         xfs_quiesce_attr(mp);
1439         return xfs_sync_sb(mp, true);
1440 }
1441
1442 STATIC int
1443 xfs_fs_unfreeze(
1444         struct super_block      *sb)
1445 {
1446         struct xfs_mount        *mp = XFS_M(sb);
1447
1448         xfs_restore_resvblks(mp);
1449         xfs_log_work_queue(mp);
1450         xfs_icache_enable_reclaim(mp);
1451         return 0;
1452 }
1453
1454 STATIC int
1455 xfs_fs_show_options(
1456         struct seq_file         *m,
1457         struct dentry           *root)
1458 {
1459         return xfs_showargs(XFS_M(root->d_sb), m);
1460 }
1461
1462 /*
1463  * This function fills in xfs_mount_t fields based on mount args.
1464  * Note: the superblock _has_ now been read in.
1465  */
1466 STATIC int
1467 xfs_finish_flags(
1468         struct xfs_mount        *mp)
1469 {
1470         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1471
1472         /* Fail a mount where the logbuf is smaller than the log stripe */
1473         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1474                 if (mp->m_logbsize <= 0 &&
1475                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1476                         mp->m_logbsize = mp->m_sb.sb_logsunit;
1477                 } else if (mp->m_logbsize > 0 &&
1478                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
1479                         xfs_warn(mp,
1480                 "logbuf size must be greater than or equal to log stripe size");
1481                         return -EINVAL;
1482                 }
1483         } else {
1484                 /* Fail a mount if the logbuf is larger than 32K */
1485                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1486                         xfs_warn(mp,
1487                 "logbuf size for version 1 logs must be 16K or 32K");
1488                         return -EINVAL;
1489                 }
1490         }
1491
1492         /*
1493          * V5 filesystems always use attr2 format for attributes.
1494          */
1495         if (xfs_sb_version_hascrc(&mp->m_sb) &&
1496             (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1497                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1498                              "attr2 is always enabled for V5 filesystems.");
1499                 return -EINVAL;
1500         }
1501
1502         /*
1503          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1504          * told by noattr2 to turn it off
1505          */
1506         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1507             !(mp->m_flags & XFS_MOUNT_NOATTR2))
1508                 mp->m_flags |= XFS_MOUNT_ATTR2;
1509
1510         /*
1511          * prohibit r/w mounts of read-only filesystems
1512          */
1513         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1514                 xfs_warn(mp,
1515                         "cannot mount a read-only filesystem as read-write");
1516                 return -EROFS;
1517         }
1518
1519         if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1520             (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1521             !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1522                 xfs_warn(mp,
1523                   "Super block does not support project and group quota together");
1524                 return -EINVAL;
1525         }
1526
1527         return 0;
1528 }
1529
1530 static int
1531 xfs_init_percpu_counters(
1532         struct xfs_mount        *mp)
1533 {
1534         int             error;
1535
1536         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1537         if (error)
1538                 return -ENOMEM;
1539
1540         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1541         if (error)
1542                 goto free_icount;
1543
1544         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1545         if (error)
1546                 goto free_ifree;
1547
1548         return 0;
1549
1550 free_ifree:
1551         percpu_counter_destroy(&mp->m_ifree);
1552 free_icount:
1553         percpu_counter_destroy(&mp->m_icount);
1554         return -ENOMEM;
1555 }
1556
1557 void
1558 xfs_reinit_percpu_counters(
1559         struct xfs_mount        *mp)
1560 {
1561         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1562         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1563         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1564 }
1565
1566 static void
1567 xfs_destroy_percpu_counters(
1568         struct xfs_mount        *mp)
1569 {
1570         percpu_counter_destroy(&mp->m_icount);
1571         percpu_counter_destroy(&mp->m_ifree);
1572         percpu_counter_destroy(&mp->m_fdblocks);
1573 }
1574
1575 static struct xfs_mount *
1576 xfs_mount_alloc(
1577         struct super_block      *sb)
1578 {
1579         struct xfs_mount        *mp;
1580
1581         mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1582         if (!mp)
1583                 return NULL;
1584
1585         mp->m_super = sb;
1586         spin_lock_init(&mp->m_sb_lock);
1587         spin_lock_init(&mp->m_agirotor_lock);
1588         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1589         spin_lock_init(&mp->m_perag_lock);
1590         mutex_init(&mp->m_growlock);
1591         atomic_set(&mp->m_active_trans, 0);
1592         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1593         INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1594         INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1595         mp->m_kobj.kobject.kset = xfs_kset;
1596         return mp;
1597 }
1598
1599
1600 STATIC int
1601 xfs_fs_fill_super(
1602         struct super_block      *sb,
1603         void                    *data,
1604         int                     silent)
1605 {
1606         struct inode            *root;
1607         struct xfs_mount        *mp = NULL;
1608         int                     flags = 0, error = -ENOMEM;
1609
1610         /*
1611          * allocate mp and do all low-level struct initializations before we
1612          * attach it to the super
1613          */
1614         mp = xfs_mount_alloc(sb);
1615         if (!mp)
1616                 goto out;
1617         sb->s_fs_info = mp;
1618
1619         error = xfs_parseargs(mp, (char *)data);
1620         if (error)
1621                 goto out_free_fsname;
1622
1623         sb_min_blocksize(sb, BBSIZE);
1624         sb->s_xattr = xfs_xattr_handlers;
1625         sb->s_export_op = &xfs_export_operations;
1626 #ifdef CONFIG_XFS_QUOTA
1627         sb->s_qcop = &xfs_quotactl_operations;
1628         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1629 #endif
1630         sb->s_op = &xfs_super_operations;
1631
1632         /*
1633          * Delay mount work if the debug hook is set. This is debug
1634          * instrumention to coordinate simulation of xfs mount failures with
1635          * VFS superblock operations
1636          */
1637         if (xfs_globals.mount_delay) {
1638                 xfs_notice(mp, "Delaying mount for %d seconds.",
1639                         xfs_globals.mount_delay);
1640                 msleep(xfs_globals.mount_delay * 1000);
1641         }
1642
1643         if (silent)
1644                 flags |= XFS_MFSI_QUIET;
1645
1646         error = xfs_open_devices(mp);
1647         if (error)
1648                 goto out_free_fsname;
1649
1650         error = xfs_init_mount_workqueues(mp);
1651         if (error)
1652                 goto out_close_devices;
1653
1654         error = xfs_init_percpu_counters(mp);
1655         if (error)
1656                 goto out_destroy_workqueues;
1657
1658         /* Allocate stats memory before we do operations that might use it */
1659         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1660         if (!mp->m_stats.xs_stats) {
1661                 error = -ENOMEM;
1662                 goto out_destroy_counters;
1663         }
1664
1665         error = xfs_readsb(mp, flags);
1666         if (error)
1667                 goto out_free_stats;
1668
1669         error = xfs_finish_flags(mp);
1670         if (error)
1671                 goto out_free_sb;
1672
1673         error = xfs_setup_devices(mp);
1674         if (error)
1675                 goto out_free_sb;
1676
1677         error = xfs_filestream_mount(mp);
1678         if (error)
1679                 goto out_free_sb;
1680
1681         /*
1682          * we must configure the block size in the superblock before we run the
1683          * full mount process as the mount process can lookup and cache inodes.
1684          */
1685         sb->s_magic = XFS_SUPER_MAGIC;
1686         sb->s_blocksize = mp->m_sb.sb_blocksize;
1687         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1688         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1689         sb->s_max_links = XFS_MAXLINK;
1690         sb->s_time_gran = 1;
1691         set_posix_acl_flag(sb);
1692
1693         /* version 5 superblocks support inode version counters. */
1694         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1695                 sb->s_flags |= SB_I_VERSION;
1696
1697         if (mp->m_flags & XFS_MOUNT_DAX) {
1698                 bool rtdev_is_dax = false, datadev_is_dax;
1699
1700                 xfs_warn(mp,
1701                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1702
1703                 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1704                         sb->s_blocksize);
1705                 if (mp->m_rtdev_targp)
1706                         rtdev_is_dax = bdev_dax_supported(
1707                                 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1708                 if (!rtdev_is_dax && !datadev_is_dax) {
1709                         xfs_alert(mp,
1710                         "DAX unsupported by block device. Turning off DAX.");
1711                         mp->m_flags &= ~XFS_MOUNT_DAX;
1712                 }
1713                 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1714                         xfs_alert(mp,
1715                 "DAX and reflink cannot be used together!");
1716                         error = -EINVAL;
1717                         goto out_filestream_unmount;
1718                 }
1719         }
1720
1721         if (mp->m_flags & XFS_MOUNT_DISCARD) {
1722                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1723
1724                 if (!blk_queue_discard(q)) {
1725                         xfs_warn(mp, "mounting with \"discard\" option, but "
1726                                         "the device does not support discard");
1727                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
1728                 }
1729         }
1730
1731         if (xfs_sb_version_hasreflink(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1732                 xfs_alert(mp,
1733         "reflink not compatible with realtime device!");
1734                 error = -EINVAL;
1735                 goto out_filestream_unmount;
1736         }
1737
1738         if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1739                 xfs_alert(mp,
1740         "reverse mapping btree not compatible with realtime device!");
1741                 error = -EINVAL;
1742                 goto out_filestream_unmount;
1743         }
1744
1745         error = xfs_mountfs(mp);
1746         if (error)
1747                 goto out_filestream_unmount;
1748
1749         root = igrab(VFS_I(mp->m_rootip));
1750         if (!root) {
1751                 error = -ENOENT;
1752                 goto out_unmount;
1753         }
1754         sb->s_root = d_make_root(root);
1755         if (!sb->s_root) {
1756                 error = -ENOMEM;
1757                 goto out_unmount;
1758         }
1759
1760         return 0;
1761
1762  out_filestream_unmount:
1763         xfs_filestream_unmount(mp);
1764  out_free_sb:
1765         xfs_freesb(mp);
1766  out_free_stats:
1767         free_percpu(mp->m_stats.xs_stats);
1768  out_destroy_counters:
1769         xfs_destroy_percpu_counters(mp);
1770  out_destroy_workqueues:
1771         xfs_destroy_mount_workqueues(mp);
1772  out_close_devices:
1773         xfs_close_devices(mp);
1774  out_free_fsname:
1775         sb->s_fs_info = NULL;
1776         xfs_free_fsname(mp);
1777         kfree(mp);
1778  out:
1779         return error;
1780
1781  out_unmount:
1782         xfs_filestream_unmount(mp);
1783         xfs_unmountfs(mp);
1784         goto out_free_sb;
1785 }
1786
1787 STATIC void
1788 xfs_fs_put_super(
1789         struct super_block      *sb)
1790 {
1791         struct xfs_mount        *mp = XFS_M(sb);
1792
1793         /* if ->fill_super failed, we have no mount to tear down */
1794         if (!sb->s_fs_info)
1795                 return;
1796
1797         xfs_notice(mp, "Unmounting Filesystem");
1798         xfs_filestream_unmount(mp);
1799         xfs_unmountfs(mp);
1800
1801         xfs_freesb(mp);
1802         free_percpu(mp->m_stats.xs_stats);
1803         xfs_destroy_percpu_counters(mp);
1804         xfs_destroy_mount_workqueues(mp);
1805         xfs_close_devices(mp);
1806
1807         sb->s_fs_info = NULL;
1808         xfs_free_fsname(mp);
1809         kfree(mp);
1810 }
1811
1812 STATIC struct dentry *
1813 xfs_fs_mount(
1814         struct file_system_type *fs_type,
1815         int                     flags,
1816         const char              *dev_name,
1817         void                    *data)
1818 {
1819         return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1820 }
1821
1822 static long
1823 xfs_fs_nr_cached_objects(
1824         struct super_block      *sb,
1825         struct shrink_control   *sc)
1826 {
1827         /* Paranoia: catch incorrect calls during mount setup or teardown */
1828         if (WARN_ON_ONCE(!sb->s_fs_info))
1829                 return 0;
1830         return xfs_reclaim_inodes_count(XFS_M(sb));
1831 }
1832
1833 static long
1834 xfs_fs_free_cached_objects(
1835         struct super_block      *sb,
1836         struct shrink_control   *sc)
1837 {
1838         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1839 }
1840
1841 static const struct super_operations xfs_super_operations = {
1842         .alloc_inode            = xfs_fs_alloc_inode,
1843         .destroy_inode          = xfs_fs_destroy_inode,
1844         .dirty_inode            = xfs_fs_dirty_inode,
1845         .drop_inode             = xfs_fs_drop_inode,
1846         .put_super              = xfs_fs_put_super,
1847         .sync_fs                = xfs_fs_sync_fs,
1848         .freeze_fs              = xfs_fs_freeze,
1849         .unfreeze_fs            = xfs_fs_unfreeze,
1850         .statfs                 = xfs_fs_statfs,
1851         .remount_fs             = xfs_fs_remount,
1852         .show_options           = xfs_fs_show_options,
1853         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1854         .free_cached_objects    = xfs_fs_free_cached_objects,
1855 };
1856
1857 static struct file_system_type xfs_fs_type = {
1858         .owner                  = THIS_MODULE,
1859         .name                   = "xfs",
1860         .mount                  = xfs_fs_mount,
1861         .kill_sb                = kill_block_super,
1862         .fs_flags               = FS_REQUIRES_DEV,
1863 };
1864 MODULE_ALIAS_FS("xfs");
1865
1866 STATIC int __init
1867 xfs_init_zones(void)
1868 {
1869         if (bioset_init(&xfs_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE),
1870                         offsetof(struct xfs_ioend, io_inline_bio),
1871                         BIOSET_NEED_BVECS))
1872                 goto out;
1873
1874         xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1875                                                 "xfs_log_ticket");
1876         if (!xfs_log_ticket_zone)
1877                 goto out_free_ioend_bioset;
1878
1879         xfs_bmap_free_item_zone = kmem_zone_init(
1880                         sizeof(struct xfs_extent_free_item),
1881                         "xfs_bmap_free_item");
1882         if (!xfs_bmap_free_item_zone)
1883                 goto out_destroy_log_ticket_zone;
1884
1885         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1886                                                 "xfs_btree_cur");
1887         if (!xfs_btree_cur_zone)
1888                 goto out_destroy_bmap_free_item_zone;
1889
1890         xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1891                                                 "xfs_da_state");
1892         if (!xfs_da_state_zone)
1893                 goto out_destroy_btree_cur_zone;
1894
1895         xfs_ifork_zone = kmem_zone_init(sizeof(struct xfs_ifork), "xfs_ifork");
1896         if (!xfs_ifork_zone)
1897                 goto out_destroy_da_state_zone;
1898
1899         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1900         if (!xfs_trans_zone)
1901                 goto out_destroy_ifork_zone;
1902
1903
1904         /*
1905          * The size of the zone allocated buf log item is the maximum
1906          * size possible under XFS.  This wastes a little bit of memory,
1907          * but it is much faster.
1908          */
1909         xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1910                                            "xfs_buf_item");
1911         if (!xfs_buf_item_zone)
1912                 goto out_destroy_trans_zone;
1913
1914         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1915                         ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1916                                  sizeof(xfs_extent_t))), "xfs_efd_item");
1917         if (!xfs_efd_zone)
1918                 goto out_destroy_buf_item_zone;
1919
1920         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1921                         ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1922                                 sizeof(xfs_extent_t))), "xfs_efi_item");
1923         if (!xfs_efi_zone)
1924                 goto out_destroy_efd_zone;
1925
1926         xfs_inode_zone =
1927                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1928                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1929                         KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1930         if (!xfs_inode_zone)
1931                 goto out_destroy_efi_zone;
1932
1933         xfs_ili_zone =
1934                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1935                                         KM_ZONE_SPREAD, NULL);
1936         if (!xfs_ili_zone)
1937                 goto out_destroy_inode_zone;
1938         xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1939                                         "xfs_icr");
1940         if (!xfs_icreate_zone)
1941                 goto out_destroy_ili_zone;
1942
1943         xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1944                         "xfs_rud_item");
1945         if (!xfs_rud_zone)
1946                 goto out_destroy_icreate_zone;
1947
1948         xfs_rui_zone = kmem_zone_init(
1949                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1950                         "xfs_rui_item");
1951         if (!xfs_rui_zone)
1952                 goto out_destroy_rud_zone;
1953
1954         xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1955                         "xfs_cud_item");
1956         if (!xfs_cud_zone)
1957                 goto out_destroy_rui_zone;
1958
1959         xfs_cui_zone = kmem_zone_init(
1960                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1961                         "xfs_cui_item");
1962         if (!xfs_cui_zone)
1963                 goto out_destroy_cud_zone;
1964
1965         xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1966                         "xfs_bud_item");
1967         if (!xfs_bud_zone)
1968                 goto out_destroy_cui_zone;
1969
1970         xfs_bui_zone = kmem_zone_init(
1971                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1972                         "xfs_bui_item");
1973         if (!xfs_bui_zone)
1974                 goto out_destroy_bud_zone;
1975
1976         return 0;
1977
1978  out_destroy_bud_zone:
1979         kmem_zone_destroy(xfs_bud_zone);
1980  out_destroy_cui_zone:
1981         kmem_zone_destroy(xfs_cui_zone);
1982  out_destroy_cud_zone:
1983         kmem_zone_destroy(xfs_cud_zone);
1984  out_destroy_rui_zone:
1985         kmem_zone_destroy(xfs_rui_zone);
1986  out_destroy_rud_zone:
1987         kmem_zone_destroy(xfs_rud_zone);
1988  out_destroy_icreate_zone:
1989         kmem_zone_destroy(xfs_icreate_zone);
1990  out_destroy_ili_zone:
1991         kmem_zone_destroy(xfs_ili_zone);
1992  out_destroy_inode_zone:
1993         kmem_zone_destroy(xfs_inode_zone);
1994  out_destroy_efi_zone:
1995         kmem_zone_destroy(xfs_efi_zone);
1996  out_destroy_efd_zone:
1997         kmem_zone_destroy(xfs_efd_zone);
1998  out_destroy_buf_item_zone:
1999         kmem_zone_destroy(xfs_buf_item_zone);
2000  out_destroy_trans_zone:
2001         kmem_zone_destroy(xfs_trans_zone);
2002  out_destroy_ifork_zone:
2003         kmem_zone_destroy(xfs_ifork_zone);
2004  out_destroy_da_state_zone:
2005         kmem_zone_destroy(xfs_da_state_zone);
2006  out_destroy_btree_cur_zone:
2007         kmem_zone_destroy(xfs_btree_cur_zone);
2008  out_destroy_bmap_free_item_zone:
2009         kmem_zone_destroy(xfs_bmap_free_item_zone);
2010  out_destroy_log_ticket_zone:
2011         kmem_zone_destroy(xfs_log_ticket_zone);
2012  out_free_ioend_bioset:
2013         bioset_exit(&xfs_ioend_bioset);
2014  out:
2015         return -ENOMEM;
2016 }
2017
2018 STATIC void
2019 xfs_destroy_zones(void)
2020 {
2021         /*
2022          * Make sure all delayed rcu free are flushed before we
2023          * destroy caches.
2024          */
2025         rcu_barrier();
2026         kmem_zone_destroy(xfs_bui_zone);
2027         kmem_zone_destroy(xfs_bud_zone);
2028         kmem_zone_destroy(xfs_cui_zone);
2029         kmem_zone_destroy(xfs_cud_zone);
2030         kmem_zone_destroy(xfs_rui_zone);
2031         kmem_zone_destroy(xfs_rud_zone);
2032         kmem_zone_destroy(xfs_icreate_zone);
2033         kmem_zone_destroy(xfs_ili_zone);
2034         kmem_zone_destroy(xfs_inode_zone);
2035         kmem_zone_destroy(xfs_efi_zone);
2036         kmem_zone_destroy(xfs_efd_zone);
2037         kmem_zone_destroy(xfs_buf_item_zone);
2038         kmem_zone_destroy(xfs_trans_zone);
2039         kmem_zone_destroy(xfs_ifork_zone);
2040         kmem_zone_destroy(xfs_da_state_zone);
2041         kmem_zone_destroy(xfs_btree_cur_zone);
2042         kmem_zone_destroy(xfs_bmap_free_item_zone);
2043         kmem_zone_destroy(xfs_log_ticket_zone);
2044         bioset_exit(&xfs_ioend_bioset);
2045 }
2046
2047 STATIC int __init
2048 xfs_init_workqueues(void)
2049 {
2050         /*
2051          * The allocation workqueue can be used in memory reclaim situations
2052          * (writepage path), and parallelism is only limited by the number of
2053          * AGs in all the filesystems mounted. Hence use the default large
2054          * max_active value for this workqueue.
2055          */
2056         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2057                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2058         if (!xfs_alloc_wq)
2059                 return -ENOMEM;
2060
2061         xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2062         if (!xfs_discard_wq)
2063                 goto out_free_alloc_wq;
2064
2065         return 0;
2066 out_free_alloc_wq:
2067         destroy_workqueue(xfs_alloc_wq);
2068         return -ENOMEM;
2069 }
2070
2071 STATIC void
2072 xfs_destroy_workqueues(void)
2073 {
2074         destroy_workqueue(xfs_discard_wq);
2075         destroy_workqueue(xfs_alloc_wq);
2076 }
2077
2078 STATIC int __init
2079 init_xfs_fs(void)
2080 {
2081         int                     error;
2082
2083         xfs_check_ondisk_structs();
2084
2085         printk(KERN_INFO XFS_VERSION_STRING " with "
2086                          XFS_BUILD_OPTIONS " enabled\n");
2087
2088         xfs_extent_free_init_defer_op();
2089         xfs_rmap_update_init_defer_op();
2090         xfs_refcount_update_init_defer_op();
2091         xfs_bmap_update_init_defer_op();
2092
2093         xfs_dir_startup();
2094
2095         error = xfs_init_zones();
2096         if (error)
2097                 goto out;
2098
2099         error = xfs_init_workqueues();
2100         if (error)
2101                 goto out_destroy_zones;
2102
2103         error = xfs_mru_cache_init();
2104         if (error)
2105                 goto out_destroy_wq;
2106
2107         error = xfs_buf_init();
2108         if (error)
2109                 goto out_mru_cache_uninit;
2110
2111         error = xfs_init_procfs();
2112         if (error)
2113                 goto out_buf_terminate;
2114
2115         error = xfs_sysctl_register();
2116         if (error)
2117                 goto out_cleanup_procfs;
2118
2119         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2120         if (!xfs_kset) {
2121                 error = -ENOMEM;
2122                 goto out_sysctl_unregister;
2123         }
2124
2125         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2126
2127         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2128         if (!xfsstats.xs_stats) {
2129                 error = -ENOMEM;
2130                 goto out_kset_unregister;
2131         }
2132
2133         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2134                                "stats");
2135         if (error)
2136                 goto out_free_stats;
2137
2138 #ifdef DEBUG
2139         xfs_dbg_kobj.kobject.kset = xfs_kset;
2140         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2141         if (error)
2142                 goto out_remove_stats_kobj;
2143 #endif
2144
2145         error = xfs_qm_init();
2146         if (error)
2147                 goto out_remove_dbg_kobj;
2148
2149         error = register_filesystem(&xfs_fs_type);
2150         if (error)
2151                 goto out_qm_exit;
2152         return 0;
2153
2154  out_qm_exit:
2155         xfs_qm_exit();
2156  out_remove_dbg_kobj:
2157 #ifdef DEBUG
2158         xfs_sysfs_del(&xfs_dbg_kobj);
2159  out_remove_stats_kobj:
2160 #endif
2161         xfs_sysfs_del(&xfsstats.xs_kobj);
2162  out_free_stats:
2163         free_percpu(xfsstats.xs_stats);
2164  out_kset_unregister:
2165         kset_unregister(xfs_kset);
2166  out_sysctl_unregister:
2167         xfs_sysctl_unregister();
2168  out_cleanup_procfs:
2169         xfs_cleanup_procfs();
2170  out_buf_terminate:
2171         xfs_buf_terminate();
2172  out_mru_cache_uninit:
2173         xfs_mru_cache_uninit();
2174  out_destroy_wq:
2175         xfs_destroy_workqueues();
2176  out_destroy_zones:
2177         xfs_destroy_zones();
2178  out:
2179         return error;
2180 }
2181
2182 STATIC void __exit
2183 exit_xfs_fs(void)
2184 {
2185         xfs_qm_exit();
2186         unregister_filesystem(&xfs_fs_type);
2187 #ifdef DEBUG
2188         xfs_sysfs_del(&xfs_dbg_kobj);
2189 #endif
2190         xfs_sysfs_del(&xfsstats.xs_kobj);
2191         free_percpu(xfsstats.xs_stats);
2192         kset_unregister(xfs_kset);
2193         xfs_sysctl_unregister();
2194         xfs_cleanup_procfs();
2195         xfs_buf_terminate();
2196         xfs_mru_cache_uninit();
2197         xfs_destroy_workqueues();
2198         xfs_destroy_zones();
2199         xfs_uuid_table_free();
2200 }
2201
2202 module_init(init_xfs_fs);
2203 module_exit(exit_xfs_fs);
2204
2205 MODULE_AUTHOR("Silicon Graphics, Inc.");
2206 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2207 MODULE_LICENSE("GPL");