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