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