1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
  24  * Copyright 2015 Joyent, Inc.
  25  */
  26 
  27 #include <sys/types.h>
  28 #include <sys/param.h>
  29 #include <sys/sysmacros.h>
  30 #include <sys/kmem.h>
  31 #include <sys/time.h>
  32 #include <sys/pathname.h>
  33 #include <sys/vfs.h>
  34 #include <sys/vfs_opreg.h>
  35 #include <sys/vnode.h>
  36 #include <sys/stat.h>
  37 #include <sys/uio.h>
  38 #include <sys/stat.h>
  39 #include <sys/errno.h>
  40 #include <sys/cmn_err.h>
  41 #include <sys/cred.h>
  42 #include <sys/statvfs.h>
  43 #include <sys/mount.h>
  44 #include <sys/debug.h>
  45 #include <sys/systm.h>
  46 #include <sys/mntent.h>
  47 #include <fs/fs_subr.h>
  48 #include <vm/page.h>
  49 #include <vm/anon.h>
  50 #include <sys/model.h>
  51 #include <sys/policy.h>
  52 
  53 #include <sys/fs/swapnode.h>
  54 #include <sys/fs/tmp.h>
  55 #include <sys/fs/tmpnode.h>
  56 
  57 static int tmpfsfstype;
  58 
  59 /*
  60  * tmpfs vfs operations.
  61  */
  62 static int tmpfsinit(int, char *);
  63 static int tmp_mount(struct vfs *, struct vnode *,
  64         struct mounta *, struct cred *);
  65 static int tmp_unmount(struct vfs *, int, struct cred *);
  66 static int tmp_root(struct vfs *, struct vnode **);
  67 static int tmp_statvfs(struct vfs *, struct statvfs64 *);
  68 static int tmp_vget(struct vfs *, struct vnode **, struct fid *);
  69 
  70 /*
  71  * Loadable module wrapper
  72  */
  73 #include <sys/modctl.h>
  74 
  75 static mntopts_t tmpfs_proto_opttbl;
  76 
  77 static vfsdef_t vfw = {
  78         VFSDEF_VERSION,
  79         "tmpfs",
  80         tmpfsinit,
  81         VSW_HASPROTO|VSW_CANREMOUNT|VSW_STATS|VSW_ZMOUNT,
  82         &tmpfs_proto_opttbl
  83 };
  84 
  85 /*
  86  * in-kernel mnttab options
  87  */
  88 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
  89 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
  90 
  91 static mntopt_t tmpfs_options[] = {
  92         /* Option name          Cancel Opt      Arg     Flags           Data */
  93         { MNTOPT_XATTR,         xattr_cancel,   NULL,   MO_DEFAULT,     NULL},
  94         { MNTOPT_NOXATTR,       noxattr_cancel, NULL,   NULL,           NULL},
  95         { "size",               NULL,           "0",    MO_HASVALUE,    NULL},
  96         { "mode",               NULL,           NULL,   MO_HASVALUE,    NULL}
  97 };
  98 
  99 
 100 static mntopts_t tmpfs_proto_opttbl = {
 101         sizeof (tmpfs_options) / sizeof (mntopt_t),
 102         tmpfs_options
 103 };
 104 
 105 /*
 106  * Module linkage information
 107  */
 108 static struct modlfs modlfs = {
 109         &mod_fsops, "filesystem for tmpfs", &vfw
 110 };
 111 
 112 static struct modlinkage modlinkage = {
 113         MODREV_1, &modlfs, NULL
 114 };
 115 
 116 int
 117 _init()
 118 {
 119         return (mod_install(&modlinkage));
 120 }
 121 
 122 int
 123 _fini()
 124 {
 125         int error;
 126 
 127         error = mod_remove(&modlinkage);
 128         if (error)
 129                 return (error);
 130         /*
 131          * Tear down the operations vectors
 132          */
 133         (void) vfs_freevfsops_by_type(tmpfsfstype);
 134         vn_freevnodeops(tmp_vnodeops);
 135         return (0);
 136 }
 137 
 138 int
 139 _info(struct modinfo *modinfop)
 140 {
 141         return (mod_info(&modlinkage, modinfop));
 142 }
 143 
 144 /*
 145  * The following are patchable variables limiting the amount of system
 146  * resources tmpfs can use.
 147  *
 148  * tmpfs_maxkmem limits the amount of kernel kmem_alloc memory
 149  * tmpfs can use for it's data structures (e.g. tmpnodes, directory entries)
 150  * It is not determined by setting a hard limit but rather as a percentage of
 151  * physical memory which is determined when tmpfs is first used in the system.
 152  *
 153  * tmpfs_minfree is the minimum amount of swap space that tmpfs leaves for
 154  * the rest of the system.  In other words, if the amount of free swap space
 155  * in the system (i.e. anoninfo.ani_free) drops below tmpfs_minfree, tmpfs
 156  * anon allocations will fail.
 157  *
 158  * There is also a per mount limit on the amount of swap space
 159  * (tmount.tm_anonmax) settable via a mount option.
 160  */
 161 volatile size_t tmpfs_maxkmem = 0;
 162 volatile size_t tmpfs_minfree = 0;
 163 size_t tmp_kmemspace;           /* bytes of kernel heap used by all tmpfs */
 164 
 165 static major_t tmpfs_major;
 166 static minor_t tmpfs_minor;
 167 static kmutex_t tmpfs_minor_lock;
 168 
 169 /*
 170  * initialize global tmpfs locks and such
 171  * called when loading tmpfs module
 172  */
 173 static int
 174 tmpfsinit(int fstype, char *name)
 175 {
 176         static const fs_operation_def_t tmp_vfsops_template[] = {
 177                 VFSNAME_MOUNT,          { .vfs_mount = tmp_mount },
 178                 VFSNAME_UNMOUNT,        { .vfs_unmount = tmp_unmount },
 179                 VFSNAME_ROOT,           { .vfs_root = tmp_root },
 180                 VFSNAME_STATVFS,        { .vfs_statvfs = tmp_statvfs },
 181                 VFSNAME_VGET,           { .vfs_vget = tmp_vget },
 182                 NULL,                   NULL
 183         };
 184         int error;
 185         extern  void    tmpfs_hash_init();
 186 
 187         tmpfs_hash_init();
 188         tmpfsfstype = fstype;
 189         ASSERT(tmpfsfstype != 0);
 190 
 191         error = vfs_setfsops(fstype, tmp_vfsops_template, NULL);
 192         if (error != 0) {
 193                 cmn_err(CE_WARN, "tmpfsinit: bad vfs ops template");
 194                 return (error);
 195         }
 196 
 197         error = vn_make_ops(name, tmp_vnodeops_template, &tmp_vnodeops);
 198         if (error != 0) {
 199                 (void) vfs_freevfsops_by_type(fstype);
 200                 cmn_err(CE_WARN, "tmpfsinit: bad vnode ops template");
 201                 return (error);
 202         }
 203 
 204         /*
 205          * tmpfs_minfree doesn't need to be some function of configured
 206          * swap space since it really is an absolute limit of swap space
 207          * which still allows other processes to execute.
 208          */
 209         if (tmpfs_minfree == 0) {
 210                 /*
 211                  * Set if not patched
 212                  */
 213                 tmpfs_minfree = btopr(TMPMINFREE);
 214         }
 215 
 216         /*
 217          * The maximum amount of space tmpfs can allocate is
 218          * TMPMAXPROCKMEM percent of kernel memory
 219          */
 220         if (tmpfs_maxkmem == 0)
 221                 tmpfs_maxkmem = MAX(PAGESIZE, kmem_maxavail() / TMPMAXFRACKMEM);
 222 
 223         if ((tmpfs_major = getudev()) == (major_t)-1) {
 224                 cmn_err(CE_WARN, "tmpfsinit: Can't get unique device number.");
 225                 tmpfs_major = 0;
 226         }
 227         mutex_init(&tmpfs_minor_lock, NULL, MUTEX_DEFAULT, NULL);
 228         return (0);
 229 }
 230 
 231 static int
 232 tmp_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
 233 {
 234         struct tmount *tm = NULL;
 235         struct tmpnode *tp;
 236         struct pathname dpn;
 237         int error;
 238         pgcnt_t anonmax;
 239         struct vattr rattr;
 240         int got_attrs;
 241         boolean_t mode_arg = B_FALSE;
 242         mode_t root_mode = 0777;
 243         char *argstr;
 244 
 245         if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
 246                 return (error);
 247 
 248         if (mvp->v_type != VDIR)
 249                 return (ENOTDIR);
 250 
 251         mutex_enter(&mvp->v_lock);
 252         if ((uap->flags & MS_REMOUNT) == 0 && (uap->flags & MS_OVERLAY) == 0 &&
 253             (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
 254                 mutex_exit(&mvp->v_lock);
 255                 return (EBUSY);
 256         }
 257         mutex_exit(&mvp->v_lock);
 258 
 259         /*
 260          * Having the resource be anything but "swap" doesn't make sense.
 261          */
 262         vfs_setresource(vfsp, "swap", 0);
 263 
 264         /*
 265          * now look for options we understand...
 266          */
 267 
 268         /* tmpfs doesn't support read-only mounts */
 269         if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) {
 270                 error = EINVAL;
 271                 goto out;
 272         }
 273 
 274         /*
 275          * tm_anonmax is set according to the mount arguments
 276          * if any.  Otherwise, it is set to a maximum value.
 277          */
 278         if (vfs_optionisset(vfsp, "size", &argstr)) {
 279                 if ((error = tmp_convnum(argstr, &anonmax)) != 0)
 280                         goto out;
 281         } else {
 282                 anonmax = ULONG_MAX;
 283         }
 284 
 285         /*
 286          * The "mode" mount argument allows the operator to override the
 287          * permissions of the root of the tmpfs mount.
 288          */
 289         if (vfs_optionisset(vfsp, "mode", &argstr)) {
 290                 if ((error = tmp_convmode(argstr, &root_mode)) != 0) {
 291                         goto out;
 292                 }
 293                 mode_arg = B_TRUE;
 294         }
 295 
 296         if (error = pn_get(uap->dir,
 297             (uap->flags & MS_SYSSPACE) ? UIO_SYSSPACE : UIO_USERSPACE, &dpn))
 298                 goto out;
 299 
 300         if (uap->flags & MS_REMOUNT) {
 301                 tm = (struct tmount *)VFSTOTM(vfsp);
 302 
 303                 /*
 304                  * If we change the size so its less than what is currently
 305                  * being used, we allow that. The file system will simply be
 306                  * full until enough files have been removed to get below the
 307                  * new max.
 308                  */
 309                 mutex_enter(&tm->tm_contents);
 310                 tm->tm_anonmax = anonmax;
 311                 mutex_exit(&tm->tm_contents);
 312                 goto out;
 313         }
 314 
 315         if ((tm = tmp_memalloc(sizeof (struct tmount), 0)) == NULL) {
 316                 pn_free(&dpn);
 317                 error = ENOMEM;
 318                 goto out;
 319         }
 320 
 321         /*
 322          * find an available minor device number for this mount
 323          */
 324         mutex_enter(&tmpfs_minor_lock);
 325         do {
 326                 tmpfs_minor = (tmpfs_minor + 1) & L_MAXMIN32;
 327                 tm->tm_dev = makedevice(tmpfs_major, tmpfs_minor);
 328         } while (vfs_devismounted(tm->tm_dev));
 329         mutex_exit(&tmpfs_minor_lock);
 330 
 331         /*
 332          * Set but don't bother entering the mutex
 333          * (tmount not on mount list yet)
 334          */
 335         mutex_init(&tm->tm_contents, NULL, MUTEX_DEFAULT, NULL);
 336         mutex_init(&tm->tm_renamelck, NULL, MUTEX_DEFAULT, NULL);
 337 
 338         tm->tm_vfsp = vfsp;
 339         tm->tm_anonmax = anonmax;
 340 
 341         vfsp->vfs_data = (caddr_t)tm;
 342         vfsp->vfs_fstype = tmpfsfstype;
 343         vfsp->vfs_dev = tm->tm_dev;
 344         vfsp->vfs_bsize = PAGESIZE;
 345         vfsp->vfs_flag |= VFS_NOTRUNC;
 346         vfs_make_fsid(&vfsp->vfs_fsid, tm->tm_dev, tmpfsfstype);
 347         tm->tm_mntpath = tmp_memalloc(dpn.pn_pathlen + 1, TMP_MUSTHAVE);
 348         (void) strcpy(tm->tm_mntpath, dpn.pn_path);
 349 
 350         /*
 351          * allocate and initialize root tmpnode structure
 352          */
 353         bzero(&rattr, sizeof (struct vattr));
 354         rattr.va_mode = (mode_t)(S_IFDIR | root_mode);
 355         rattr.va_type = VDIR;
 356         rattr.va_rdev = 0;
 357         tp = tmp_memalloc(sizeof (struct tmpnode), TMP_MUSTHAVE);
 358         tmpnode_init(tm, tp, &rattr, cr);
 359 
 360         /*
 361          * Get the mode, uid, and gid from the underlying mount point.
 362          */
 363         rattr.va_mask = AT_MODE|AT_UID|AT_GID;  /* Hint to getattr */
 364         got_attrs = VOP_GETATTR(mvp, &rattr, 0, cr, NULL);
 365 
 366         rw_enter(&tp->tn_rwlock, RW_WRITER);
 367         TNTOV(tp)->v_flag |= VROOT;
 368 
 369         /*
 370          * If the getattr succeeded, use its results.  Otherwise allow
 371          * the previously set hardwired defaults to prevail.
 372          */
 373         if (got_attrs == 0) {
 374                 if (!mode_arg) {
 375                         /*
 376                          * Only use the underlying mount point for the
 377                          * mode if the "mode" mount argument was not
 378                          * provided.
 379                          */
 380                         tp->tn_mode = rattr.va_mode;
 381                 }
 382                 tp->tn_uid = rattr.va_uid;
 383                 tp->tn_gid = rattr.va_gid;
 384         }
 385 
 386         /*
 387          * initialize linked list of tmpnodes so that the back pointer of
 388          * the root tmpnode always points to the last one on the list
 389          * and the forward pointer of the last node is null
 390          */
 391         tp->tn_back = tp;
 392         tp->tn_forw = NULL;
 393         tp->tn_nlink = 0;
 394         tm->tm_rootnode = tp;
 395 
 396         tdirinit(tp, tp);
 397 
 398         rw_exit(&tp->tn_rwlock);
 399 
 400         pn_free(&dpn);
 401         error = 0;
 402 
 403 out:
 404         if (error == 0)
 405                 vfs_set_feature(vfsp, VFSFT_SYSATTR_VIEWS);
 406 
 407         return (error);
 408 }
 409 
 410 static int
 411 tmp_unmount(struct vfs *vfsp, int flag, struct cred *cr)
 412 {
 413         struct tmount *tm = (struct tmount *)VFSTOTM(vfsp);
 414         struct tmpnode *tnp, *cancel;
 415         struct vnode    *vp;
 416         int error;
 417 
 418         if ((error = secpolicy_fs_unmount(cr, vfsp)) != 0)
 419                 return (error);
 420 
 421         /*
 422          * forced unmount is not supported by this file system
 423          * and thus, ENOTSUP, is being returned.
 424          */
 425         if (flag & MS_FORCE)
 426                 return (ENOTSUP);
 427 
 428         mutex_enter(&tm->tm_contents);
 429 
 430         /*
 431          * If there are no open files, only the root node should have
 432          * a reference count.
 433          * With tm_contents held, nothing can be added or removed.
 434          * There may be some dirty pages.  To prevent fsflush from
 435          * disrupting the unmount, put a hold on each node while scanning.
 436          * If we find a previously referenced node, undo the holds we have
 437          * placed and fail EBUSY.
 438          */
 439         tnp = tm->tm_rootnode;
 440         if (TNTOV(tnp)->v_count > 1) {
 441                 mutex_exit(&tm->tm_contents);
 442                 return (EBUSY);
 443         }
 444 
 445         for (tnp = tnp->tn_forw; tnp; tnp = tnp->tn_forw) {
 446                 if ((vp = TNTOV(tnp))->v_count > 0) {
 447                         cancel = tm->tm_rootnode->tn_forw;
 448                         while (cancel != tnp) {
 449                                 vp = TNTOV(cancel);
 450                                 ASSERT(vp->v_count > 0);
 451                                 VN_RELE(vp);
 452                                 cancel = cancel->tn_forw;
 453                         }
 454                         mutex_exit(&tm->tm_contents);
 455                         return (EBUSY);
 456                 }
 457                 VN_HOLD(vp);
 458         }
 459 
 460         /*
 461          * We can drop the mutex now because no one can find this mount
 462          */
 463         mutex_exit(&tm->tm_contents);
 464 
 465         /*
 466          * Free all kmemalloc'd and anonalloc'd memory associated with
 467          * this filesystem.  To do this, we go through the file list twice,
 468          * once to remove all the directory entries, and then to remove
 469          * all the files.  We do this because there is useful code in
 470          * tmpnode_free which assumes that the directory entry has been
 471          * removed before the file.
 472          */
 473         /*
 474          * Remove all directory entries
 475          */
 476         for (tnp = tm->tm_rootnode; tnp; tnp = tnp->tn_forw) {
 477                 rw_enter(&tnp->tn_rwlock, RW_WRITER);
 478                 if (tnp->tn_type == VDIR)
 479                         tdirtrunc(tnp);
 480                 if (tnp->tn_vnode->v_flag & V_XATTRDIR) {
 481                         /*
 482                          * Account for implicit attrdir reference.
 483                          */
 484                         ASSERT(tnp->tn_nlink > 0);
 485                         DECR_COUNT(&tnp->tn_nlink, &tnp->tn_tlock);
 486                 }
 487                 rw_exit(&tnp->tn_rwlock);
 488         }
 489 
 490         ASSERT(tm->tm_rootnode);
 491 
 492         /*
 493          * All links are gone, v_count is keeping nodes in place.
 494          * VN_RELE should make the node disappear, unless somebody
 495          * is holding pages against it.  Nap and retry until it disappears.
 496          *
 497          * We re-acquire the lock to prevent others who have a HOLD on
 498          * a tmpnode via its pages or anon slots from blowing it away
 499          * (in tmp_inactive) while we're trying to get to it here. Once
 500          * we have a HOLD on it we know it'll stick around.
 501          *
 502          */
 503         mutex_enter(&tm->tm_contents);
 504         /*
 505          * Remove all the files (except the rootnode) backwards.
 506          */
 507         while ((tnp = tm->tm_rootnode->tn_back) != tm->tm_rootnode) {
 508                 mutex_exit(&tm->tm_contents);
 509                 /*
 510                  * Inhibit tmp_inactive from touching attribute directory
 511                  * as all nodes will be released here.
 512                  * Note we handled the link count in pass 2 above.
 513                  */
 514                 rw_enter(&tnp->tn_rwlock, RW_WRITER);
 515                 tnp->tn_xattrdp = NULL;
 516                 rw_exit(&tnp->tn_rwlock);
 517                 vp = TNTOV(tnp);
 518                 VN_RELE(vp);
 519                 mutex_enter(&tm->tm_contents);
 520                 /*
 521                  * It's still there after the RELE. Someone else like pageout
 522                  * has a hold on it so wait a bit and then try again - we know
 523                  * they'll give it up soon.
 524                  */
 525                 if (tnp == tm->tm_rootnode->tn_back) {
 526                         VN_HOLD(vp);
 527                         mutex_exit(&tm->tm_contents);
 528                         delay(hz / 4);
 529                         mutex_enter(&tm->tm_contents);
 530                 }
 531         }
 532         mutex_exit(&tm->tm_contents);
 533 
 534         tm->tm_rootnode->tn_xattrdp = NULL;
 535         VN_RELE(TNTOV(tm->tm_rootnode));
 536 
 537         ASSERT(tm->tm_mntpath);
 538 
 539         tmp_memfree(tm->tm_mntpath, strlen(tm->tm_mntpath) + 1);
 540 
 541         ASSERT(tm->tm_anonmem == 0);
 542 
 543         mutex_destroy(&tm->tm_contents);
 544         mutex_destroy(&tm->tm_renamelck);
 545         tmp_memfree(tm, sizeof (struct tmount));
 546 
 547         return (0);
 548 }
 549 
 550 /*
 551  * return root tmpnode for given vnode
 552  */
 553 static int
 554 tmp_root(struct vfs *vfsp, struct vnode **vpp)
 555 {
 556         struct tmount *tm = (struct tmount *)VFSTOTM(vfsp);
 557         struct tmpnode *tp = tm->tm_rootnode;
 558         struct vnode *vp;
 559 
 560         ASSERT(tp);
 561 
 562         vp = TNTOV(tp);
 563         VN_HOLD(vp);
 564         *vpp = vp;
 565         return (0);
 566 }
 567 
 568 static int
 569 tmp_statvfs(struct vfs *vfsp, struct statvfs64 *sbp)
 570 {
 571         struct tmount   *tm = (struct tmount *)VFSTOTM(vfsp);
 572         ulong_t blocks;
 573         dev32_t d32;
 574         zoneid_t eff_zid;
 575         struct zone *zp;
 576 
 577         /*
 578          * The file system may have been mounted by the global zone on
 579          * behalf of the non-global zone.  In that case, the tmount zone_id
 580          * will be the global zone.  We still want to show the swap cap inside
 581          * the zone in this case, even though the file system was mounted by
 582          * the global zone.
 583          */
 584         if (curproc->p_zone->zone_id != GLOBAL_ZONEUNIQID)
 585                 zp = curproc->p_zone;
 586         else
 587                 zp = tm->tm_vfsp->vfs_zone;
 588 
 589         if (zp == NULL)
 590                 eff_zid = GLOBAL_ZONEUNIQID;
 591         else
 592                 eff_zid = zp->zone_id;
 593 
 594         sbp->f_bsize = PAGESIZE;
 595         sbp->f_frsize = PAGESIZE;
 596 
 597         /*
 598          * Find the amount of available physical and memory swap
 599          */
 600         mutex_enter(&anoninfo_lock);
 601         ASSERT(k_anoninfo.ani_max >= k_anoninfo.ani_phys_resv);
 602         blocks = (ulong_t)CURRENT_TOTAL_AVAILABLE_SWAP;
 603         mutex_exit(&anoninfo_lock);
 604 
 605         /*
 606          * If tm_anonmax for this mount is less than the available swap space
 607          * (minus the amount tmpfs can't use), use that instead
 608          */
 609         if (blocks > tmpfs_minfree)
 610                 sbp->f_bfree = MIN(blocks - tmpfs_minfree,
 611                     tm->tm_anonmax - tm->tm_anonmem);
 612         else
 613                 sbp->f_bfree = 0;
 614 
 615         sbp->f_bavail = sbp->f_bfree;
 616 
 617         /*
 618          * Total number of blocks is what's available plus what's been used
 619          */
 620         sbp->f_blocks = (fsblkcnt64_t)(sbp->f_bfree + tm->tm_anonmem);
 621 
 622         if (eff_zid != GLOBAL_ZONEUNIQID &&
 623             zp->zone_max_swap_ctl != UINT64_MAX) {
 624                 /*
 625                  * If the fs is used by a non-global zone with a swap cap,
 626                  * then report the capped size.
 627                  */
 628                 rctl_qty_t cap, used;
 629                 pgcnt_t pgcap, pgused;
 630 
 631                 mutex_enter(&zp->zone_mem_lock);
 632                 cap = zp->zone_max_swap_ctl;
 633                 used = zp->zone_max_swap;
 634                 mutex_exit(&zp->zone_mem_lock);
 635 
 636                 pgcap = btop(cap);
 637                 pgused = btop(used);
 638 
 639                 sbp->f_bfree = MIN(pgcap - pgused, sbp->f_bfree);
 640                 sbp->f_bavail = sbp->f_bfree;
 641                 sbp->f_blocks = MIN(pgcap, sbp->f_blocks);
 642         }
 643 
 644         /*
 645          * The maximum number of files available is approximately the number
 646          * of tmpnodes we can allocate from the remaining kernel memory
 647          * available to tmpfs.  This is fairly inaccurate since it doesn't
 648          * take into account the names stored in the directory entries.
 649          */
 650         if (tmpfs_maxkmem > tmp_kmemspace)
 651                 sbp->f_ffree = (tmpfs_maxkmem - tmp_kmemspace) /
 652                     (sizeof (struct tmpnode) + sizeof (struct tdirent));
 653         else
 654                 sbp->f_ffree = 0;
 655 
 656         sbp->f_files = tmpfs_maxkmem /
 657             (sizeof (struct tmpnode) + sizeof (struct tdirent));
 658         sbp->f_favail = (fsfilcnt64_t)(sbp->f_ffree);
 659         (void) cmpldev(&d32, vfsp->vfs_dev);
 660         sbp->f_fsid = d32;
 661         (void) strcpy(sbp->f_basetype, vfssw[tmpfsfstype].vsw_name);
 662         (void) strncpy(sbp->f_fstr, tm->tm_mntpath, sizeof (sbp->f_fstr));
 663         /*
 664          * ensure null termination
 665          */
 666         sbp->f_fstr[sizeof (sbp->f_fstr) - 1] = '\0';
 667         sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
 668         sbp->f_namemax = MAXNAMELEN - 1;
 669         return (0);
 670 }
 671 
 672 static int
 673 tmp_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp)
 674 {
 675         struct tfid *tfid;
 676         struct tmount *tm = (struct tmount *)VFSTOTM(vfsp);
 677         struct tmpnode *tp = NULL;
 678 
 679         tfid = (struct tfid *)fidp;
 680         *vpp = NULL;
 681 
 682         mutex_enter(&tm->tm_contents);
 683         for (tp = tm->tm_rootnode; tp; tp = tp->tn_forw) {
 684                 mutex_enter(&tp->tn_tlock);
 685                 if (tp->tn_nodeid == tfid->tfid_ino) {
 686                         /*
 687                          * If the gen numbers don't match we know the
 688                          * file won't be found since only one tmpnode
 689                          * can have this number at a time.
 690                          */
 691                         if (tp->tn_gen != tfid->tfid_gen || tp->tn_nlink == 0) {
 692                                 mutex_exit(&tp->tn_tlock);
 693                                 mutex_exit(&tm->tm_contents);
 694                                 return (0);
 695                         }
 696                         *vpp = (struct vnode *)TNTOV(tp);
 697 
 698                         VN_HOLD(*vpp);
 699 
 700                         if ((tp->tn_mode & S_ISVTX) &&
 701                             !(tp->tn_mode & (S_IXUSR | S_IFDIR))) {
 702                                 mutex_enter(&(*vpp)->v_lock);
 703                                 (*vpp)->v_flag |= VISSWAP;
 704                                 mutex_exit(&(*vpp)->v_lock);
 705                         }
 706                         mutex_exit(&tp->tn_tlock);
 707                         mutex_exit(&tm->tm_contents);
 708                         return (0);
 709                 }
 710                 mutex_exit(&tp->tn_tlock);
 711         }
 712         mutex_exit(&tm->tm_contents);
 713         return (0);
 714 }