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 /*
  23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  24  * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  25  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
  26  */
  27 
  28 /*      Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
  29 /*        All Rights Reserved   */
  30 
  31 /*
  32  * University Copyright- Copyright (c) 1982, 1986, 1988
  33  * The Regents of the University of California
  34  * All Rights Reserved
  35  *
  36  * University Acknowledgment- Portions of this document are derived from
  37  * software developed by the University of California, Berkeley, and its
  38  * contributors.
  39  */
  40 
  41 #include <sys/types.h>
  42 #include <sys/param.h>
  43 #include <sys/systm.h>
  44 #include <sys/cpuvar.h>
  45 #include <sys/errno.h>
  46 #include <sys/cred.h>
  47 #include <sys/user.h>
  48 #include <sys/uio.h>
  49 #include <sys/vfs.h>
  50 #include <sys/vnode.h>
  51 #include <sys/pathname.h>
  52 #include <sys/proc.h>
  53 #include <sys/vtrace.h>
  54 #include <sys/sysmacros.h>
  55 #include <sys/debug.h>
  56 #include <sys/dirent.h>
  57 #include <c2/audit.h>
  58 #include <sys/zone.h>
  59 #include <sys/dnlc.h>
  60 #include <sys/fs/snode.h>
  61 #include <sys/brand.h>
  62 
  63 /* Controls whether paths are stored with vnodes. */
  64 int vfs_vnode_path = 1;
  65 
  66 int
  67 lookupname(
  68         char *fnamep,
  69         enum uio_seg seg,
  70         int followlink,
  71         vnode_t **dirvpp,
  72         vnode_t **compvpp)
  73 {
  74         return (lookupnameatcred(fnamep, seg, followlink, dirvpp, compvpp, NULL,
  75             CRED()));
  76 }
  77 
  78 /*
  79  * Lookup the user file name,
  80  * Handle allocation and freeing of pathname buffer, return error.
  81  */
  82 int
  83 lookupnameatcred(
  84         char *fnamep,                   /* user pathname */
  85         enum uio_seg seg,               /* addr space that name is in */
  86         int followlink,                 /* follow sym links */
  87         vnode_t **dirvpp,               /* ret for ptr to parent dir vnode */
  88         vnode_t **compvpp,              /* ret for ptr to component vnode */
  89         vnode_t *startvp,               /* start path search from vp */
  90         cred_t *cr)                     /* credential */
  91 {
  92         char namebuf[TYPICALMAXPATHLEN];
  93         struct pathname lookpn;
  94         int error;
  95 
  96         error = pn_get_buf(fnamep, seg, &lookpn, namebuf, sizeof (namebuf));
  97         if (error == 0) {
  98                 error = lookuppnatcred(&lookpn, NULL, followlink,
  99                     dirvpp, compvpp, startvp, cr);
 100         }
 101         if (error == ENAMETOOLONG) {
 102                 /*
 103                  * This thread used a pathname > TYPICALMAXPATHLEN bytes long.
 104                  */
 105                 if (error = pn_get(fnamep, seg, &lookpn))
 106                         return (error);
 107                 error = lookuppnatcred(&lookpn, NULL, followlink,
 108                     dirvpp, compvpp, startvp, cr);
 109                 pn_free(&lookpn);
 110         }
 111 
 112         return (error);
 113 }
 114 
 115 int
 116 lookupnameat(char *fnamep, enum uio_seg seg, int followlink,
 117     vnode_t **dirvpp, vnode_t **compvpp, vnode_t *startvp)
 118 {
 119         return (lookupnameatcred(fnamep, seg, followlink, dirvpp, compvpp,
 120             startvp, CRED()));
 121 }
 122 
 123 int
 124 lookuppn(
 125         struct pathname *pnp,
 126         struct pathname *rpnp,
 127         int followlink,
 128         vnode_t **dirvpp,
 129         vnode_t **compvpp)
 130 {
 131         return (lookuppnatcred(pnp, rpnp, followlink, dirvpp, compvpp, NULL,
 132             CRED()));
 133 }
 134 
 135 /*
 136  * Lookup the user file name from a given vp, using a specific credential.
 137  */
 138 int
 139 lookuppnatcred(
 140         struct pathname *pnp,           /* pathname to lookup */
 141         struct pathname *rpnp,          /* if non-NULL, return resolved path */
 142         int followlink,                 /* (don't) follow sym links */
 143         vnode_t **dirvpp,               /* ptr for parent vnode */
 144         vnode_t **compvpp,              /* ptr for entry vnode */
 145         vnode_t *startvp,               /* start search from this vp */
 146         cred_t *cr)                     /* user credential */
 147 {
 148         vnode_t *vp;    /* current directory vp */
 149         vnode_t *rootvp;
 150         proc_t *p = curproc;
 151 
 152         if (pnp->pn_pathlen == 0)
 153                 return (ENOENT);
 154 
 155         mutex_enter(&p->p_lock); /* for u_rdir and u_cdir */
 156         if ((rootvp = PTOU(p)->u_rdir) == NULL)
 157                 rootvp = rootdir;
 158         else if (rootvp != rootdir)     /* no need to VN_HOLD rootdir */
 159                 VN_HOLD(rootvp);
 160 
 161         if (pnp->pn_path[0] == '/') {
 162                 vp = rootvp;
 163         } else {
 164                 vp = (startvp == NULL) ? PTOU(p)->u_cdir : startvp;
 165         }
 166         VN_HOLD(vp);
 167         mutex_exit(&p->p_lock);
 168 
 169         /*
 170          * Skip over leading slashes
 171          */
 172         if (pnp->pn_path[0] == '/') {
 173                 do {
 174                         pnp->pn_path++;
 175                         pnp->pn_pathlen--;
 176                 } while (pnp->pn_path[0] == '/');
 177         }
 178 
 179         return (lookuppnvp(pnp, rpnp, followlink, dirvpp,
 180             compvpp, rootvp, vp, cr));
 181 }
 182 
 183 int
 184 lookuppnat(struct pathname *pnp, struct pathname *rpnp,
 185     int followlink, vnode_t **dirvpp, vnode_t **compvpp,
 186     vnode_t *startvp)
 187 {
 188         return (lookuppnatcred(pnp, rpnp, followlink, dirvpp, compvpp, startvp,
 189             CRED()));
 190 }
 191 
 192 /* Private flag to do our getcwd() dirty work */
 193 #define LOOKUP_CHECKREAD        0x10
 194 #define LOOKUP_MASK             (~LOOKUP_CHECKREAD)
 195 
 196 /*
 197  * Starting at current directory, translate pathname pnp to end.
 198  * Leave pathname of final component in pnp, return the vnode
 199  * for the final component in *compvpp, and return the vnode
 200  * for the parent of the final component in dirvpp.
 201  *
 202  * This is the central routine in pathname translation and handles
 203  * multiple components in pathnames, separating them at /'s.  It also
 204  * implements mounted file systems and processes symbolic links.
 205  *
 206  * vp is the vnode where the directory search should start.
 207  *
 208  * Reference counts: vp must be held prior to calling this function.  rootvp
 209  * should only be held if rootvp != rootdir.
 210  */
 211 int
 212 lookuppnvp(
 213         struct pathname *pnp,           /* pathname to lookup */
 214         struct pathname *rpnp,          /* if non-NULL, return resolved path */
 215         int flags,                      /* follow symlinks */
 216         vnode_t **dirvpp,               /* ptr for parent vnode */
 217         vnode_t **compvpp,              /* ptr for entry vnode */
 218         vnode_t *rootvp,                /* rootvp */
 219         vnode_t *vp,                    /* directory to start search at */
 220         cred_t *cr)                     /* user's credential */
 221 {
 222         vnode_t *cvp;   /* current component vp */
 223         char component[MAXNAMELEN];     /* buffer for component (incl null) */
 224         int error;
 225         int nlink;
 226         int lookup_flags;
 227         struct pathname presrvd; /* case preserved name */
 228         struct pathname *pp = NULL;
 229         vnode_t *startvp;
 230         vnode_t *zonevp = curproc->p_zone->zone_rootvp;           /* zone root */
 231         int must_be_directory = 0;
 232         boolean_t retry_with_kcred;
 233         uint32_t auditing = AU_AUDITING();
 234 
 235         CPU_STATS_ADDQ(CPU, sys, namei, 1);
 236         nlink = 0;
 237         cvp = NULL;
 238         if (rpnp)
 239                 rpnp->pn_pathlen = 0;
 240 
 241         lookup_flags = dirvpp ? LOOKUP_DIR : 0;
 242         if (flags & FIGNORECASE) {
 243                 lookup_flags |= FIGNORECASE;
 244                 pn_alloc(&presrvd);
 245                 pp = &presrvd;
 246         }
 247 
 248         if (auditing)
 249                 audit_anchorpath(pnp, vp == rootvp);
 250 
 251         /*
 252          * Eliminate any trailing slashes in the pathname.
 253          * If there are any, we must follow all symlinks.
 254          * Also, we must guarantee that the last component is a directory.
 255          */
 256         if (pn_fixslash(pnp)) {
 257                 flags |= FOLLOW;
 258                 must_be_directory = 1;
 259         }
 260 
 261         startvp = vp;
 262 next:
 263         retry_with_kcred = B_FALSE;
 264 
 265         /*
 266          * Make sure we have a directory.
 267          */
 268         if (vp->v_type != VDIR) {
 269                 error = ENOTDIR;
 270                 goto bad;
 271         }
 272 
 273         if (rpnp && VN_CMP(vp, rootvp))
 274                 (void) pn_set(rpnp, "/");
 275 
 276         /*
 277          * Process the next component of the pathname.
 278          */
 279         if (error = pn_getcomponent(pnp, component)) {
 280                 goto bad;
 281         }
 282 
 283         /*
 284          * Handle "..": two special cases.
 285          * 1. If we're at the root directory (e.g. after chroot or
 286          *    zone_enter) then change ".." to "." so we can't get
 287          *    out of this subtree.
 288          * 2. If this vnode is the root of a mounted file system,
 289          *    then replace it with the vnode that was mounted on
 290          *    so that we take the ".." in the other file system.
 291          */
 292         if (component[0] == '.' && component[1] == '.' && component[2] == 0) {
 293 checkforroot:
 294                 if (VN_CMP(vp, rootvp) || VN_CMP(vp, zonevp)) {
 295                         component[1] = '\0';
 296                 } else if (vp->v_flag & VROOT) {
 297                         vfs_t *vfsp;
 298                         cvp = vp;
 299 
 300                         /*
 301                          * While we deal with the vfs pointer from the vnode
 302                          * the filesystem could have been forcefully unmounted
 303                          * and the vnode's v_vfsp could have been invalidated
 304                          * by VFS_UNMOUNT. Hence, we cache v_vfsp and use it
 305                          * with vfs_rlock_wait/vfs_unlock.
 306                          * It is safe to use the v_vfsp even it is freed by
 307                          * VFS_UNMOUNT because vfs_rlock_wait/vfs_unlock
 308                          * do not dereference v_vfsp. It is just used as a
 309                          * magic cookie.
 310                          * One more corner case here is the memory getting
 311                          * reused for another vfs structure. In this case
 312                          * lookuppnvp's vfs_rlock_wait will succeed, domount's
 313                          * vfs_lock will fail and domount will bail out with an
 314                          * error (EBUSY).
 315                          */
 316                         vfsp = cvp->v_vfsp;
 317 
 318                         /*
 319                          * This lock is used to synchronize
 320                          * mounts/unmounts and lookups.
 321                          * Threads doing mounts/unmounts hold the
 322                          * writers version vfs_lock_wait().
 323                          */
 324 
 325                         vfs_rlock_wait(vfsp);
 326 
 327                         /*
 328                          * If this vnode is on a file system that
 329                          * has been forcibly unmounted,
 330                          * we can't proceed. Cancel this operation
 331                          * and return EIO.
 332                          *
 333                          * vfs_vnodecovered is NULL if unmounted.
 334                          * Currently, nfs uses VFS_UNMOUNTED to
 335                          * check if it's a forced-umount. Keep the
 336                          * same checking here as well even though it
 337                          * may not be needed.
 338                          */
 339                         if (((vp = cvp->v_vfsp->vfs_vnodecovered) == NULL) ||
 340                             (cvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)) {
 341                                 vfs_unlock(vfsp);
 342                                 VN_RELE(cvp);
 343                                 if (pp)
 344                                         pn_free(pp);
 345                                 return (EIO);
 346                         }
 347                         VN_HOLD(vp);
 348                         vfs_unlock(vfsp);
 349                         VN_RELE(cvp);
 350                         cvp = NULL;
 351                         /*
 352                          * Crossing mount points. For eg: We are doing
 353                          * a lookup of ".." for file systems root vnode
 354                          * mounted here, and VOP_LOOKUP() (with covered vnode)
 355                          * will be on underlying file systems mount point
 356                          * vnode. Set retry_with_kcred flag as we might end
 357                          * up doing VOP_LOOKUP() with kcred if required.
 358                          */
 359                         retry_with_kcred = B_TRUE;
 360                         goto checkforroot;
 361                 }
 362         }
 363 
 364         /*
 365          * LOOKUP_CHECKREAD is a private flag used by vnodetopath() to indicate
 366          * that we need to have read permission on every directory in the entire
 367          * path.  This is used to ensure that a forward-lookup of a cached value
 368          * has the same effect as a reverse-lookup when the cached value cannot
 369          * be found.
 370          */
 371         if ((flags & LOOKUP_CHECKREAD) &&
 372             (error = VOP_ACCESS(vp, VREAD, 0, cr, NULL)) != 0)
 373                 goto bad;
 374 
 375         /*
 376          * Perform a lookup in the current directory.
 377          */
 378         error = VOP_LOOKUP(vp, component, &cvp, pnp, lookup_flags,
 379             rootvp, cr, NULL, NULL, pp);
 380 
 381         /*
 382          * Retry with kcred - If crossing mount points & error is EACCES.
 383          *
 384          * If we are crossing mount points here and doing ".." lookup,
 385          * VOP_LOOKUP() might fail if the underlying file systems
 386          * mount point has no execute permission. In cases like these,
 387          * we retry VOP_LOOKUP() by giving as much privilage as possible
 388          * by passing kcred credentials.
 389          *
 390          * In case of hierarchical file systems, passing kcred still may
 391          * or may not work.
 392          * For eg: UFS FS --> Mount NFS FS --> Again mount UFS on some
 393          *                      directory inside NFS FS.
 394          */
 395         if ((error == EACCES) && retry_with_kcred)
 396                 error = VOP_LOOKUP(vp, component, &cvp, pnp, lookup_flags,
 397                     rootvp, zone_kcred(), NULL, NULL, pp);
 398 
 399         if (error) {
 400                 cvp = NULL;
 401                 /*
 402                  * On error, return hard error if
 403                  * (a) we're not at the end of the pathname yet, or
 404                  * (b) the caller didn't want the parent directory, or
 405                  * (c) we failed for some reason other than a missing entry.
 406                  */
 407                 if (pn_pathleft(pnp) || dirvpp == NULL || error != ENOENT)
 408                         goto bad;
 409                 if (auditing) { /* directory access */
 410                         if (error = audit_savepath(pnp, vp, vp, error, cr))
 411                                 goto bad_noaudit;
 412                 }
 413 
 414                 pn_setlast(pnp);
 415                 /*
 416                  * We inform the caller that the desired entry must be
 417                  * a directory by adding a '/' to the component name.
 418                  */
 419                 if (must_be_directory && (error = pn_addslash(pnp)) != 0)
 420                         goto bad;
 421                 *dirvpp = vp;
 422                 if (compvpp != NULL)
 423                         *compvpp = NULL;
 424                 if (rootvp != rootdir)
 425                         VN_RELE(rootvp);
 426                 if (pp)
 427                         pn_free(pp);
 428                 return (0);
 429         }
 430 
 431         /*
 432          * Traverse mount points.
 433          * XXX why don't we need to hold a read lock here (call vn_vfsrlock)?
 434          * What prevents a concurrent update to v_vfsmountedhere?
 435          *      Possible answer: if mounting, we might not see the mount
 436          *      if it is concurrently coming into existence, but that's
 437          *      really not much different from the thread running a bit slower.
 438          *      If unmounting, we may get into traverse() when we shouldn't,
 439          *      but traverse() will catch this case for us.
 440          *      (For this to work, fetching v_vfsmountedhere had better
 441          *      be atomic!)
 442          */
 443         if (vn_mountedvfs(cvp) != NULL) {
 444                 if ((error = traverse(&cvp)) != 0)
 445                         goto bad;
 446         }
 447 
 448         /*
 449          * If we hit a symbolic link and there is more path to be
 450          * translated or this operation does not wish to apply
 451          * to a link, then place the contents of the link at the
 452          * front of the remaining pathname.
 453          */
 454         if (cvp->v_type == VLNK && ((flags & FOLLOW) || pn_pathleft(pnp))) {
 455                 struct pathname linkpath;
 456 
 457                 if (++nlink > MAXSYMLINKS) {
 458                         error = ELOOP;
 459                         goto bad;
 460                 }
 461                 pn_alloc(&linkpath);
 462                 if (error = pn_getsymlink(cvp, &linkpath, cr)) {
 463                         pn_free(&linkpath);
 464                         goto bad;
 465                 }
 466 
 467                 if (auditing)
 468                         audit_symlink(pnp, &linkpath);
 469 
 470                 if (pn_pathleft(&linkpath) == 0)
 471                         (void) pn_set(&linkpath, ".");
 472                 error = pn_insert(pnp, &linkpath, strlen(component));
 473                 pn_free(&linkpath);
 474                 if (error)
 475                         goto bad;
 476                 VN_RELE(cvp);
 477                 cvp = NULL;
 478                 if (pnp->pn_pathlen == 0) {
 479                         error = ENOENT;
 480                         goto bad;
 481                 }
 482                 if (pnp->pn_path[0] == '/') {
 483                         do {
 484                                 pnp->pn_path++;
 485                                 pnp->pn_pathlen--;
 486                         } while (pnp->pn_path[0] == '/');
 487                         VN_RELE(vp);
 488                         vp = rootvp;
 489                         VN_HOLD(vp);
 490                 }
 491                 if (auditing)
 492                         audit_anchorpath(pnp, vp == rootvp);
 493                 if (pn_fixslash(pnp)) {
 494                         flags |= FOLLOW;
 495                         must_be_directory = 1;
 496                 }
 497                 goto next;
 498         }
 499 
 500         /*
 501          * If rpnp is non-NULL, remember the resolved path name therein.
 502          * Do not include "." components.  Collapse occurrences of
 503          * "previous/..", so long as "previous" is not itself "..".
 504          * Exhausting rpnp results in error ENAMETOOLONG.
 505          */
 506         if (rpnp && strcmp(component, ".") != 0) {
 507                 size_t len;
 508 
 509                 if (strcmp(component, "..") == 0 &&
 510                     rpnp->pn_pathlen != 0 &&
 511                     !((rpnp->pn_pathlen > 2 &&
 512                     strncmp(rpnp->pn_path+rpnp->pn_pathlen-3, "/..", 3) == 0) ||
 513                     (rpnp->pn_pathlen == 2 &&
 514                     strncmp(rpnp->pn_path, "..", 2) == 0))) {
 515                         while (rpnp->pn_pathlen &&
 516                             rpnp->pn_path[rpnp->pn_pathlen-1] != '/')
 517                                 rpnp->pn_pathlen--;
 518                         if (rpnp->pn_pathlen > 1)
 519                                 rpnp->pn_pathlen--;
 520                         rpnp->pn_path[rpnp->pn_pathlen] = '\0';
 521                 } else {
 522                         if (rpnp->pn_pathlen != 0 &&
 523                             rpnp->pn_path[rpnp->pn_pathlen-1] != '/')
 524                                 rpnp->pn_path[rpnp->pn_pathlen++] = '/';
 525                         if (flags & FIGNORECASE) {
 526                                 /*
 527                                  * Return the case-preserved name
 528                                  * within the resolved path.
 529                                  */
 530                                 error = copystr(pp->pn_buf,
 531                                     rpnp->pn_path + rpnp->pn_pathlen,
 532                                     rpnp->pn_bufsize - rpnp->pn_pathlen, &len);
 533                         } else {
 534                                 error = copystr(component,
 535                                     rpnp->pn_path + rpnp->pn_pathlen,
 536                                     rpnp->pn_bufsize - rpnp->pn_pathlen, &len);
 537                         }
 538                         if (error)      /* copystr() returns ENAMETOOLONG */
 539                                 goto bad;
 540                         rpnp->pn_pathlen += (len - 1);
 541                         ASSERT(rpnp->pn_bufsize > rpnp->pn_pathlen);
 542                 }
 543         }
 544 
 545         /*
 546          * If no more components, return last directory (if wanted) and
 547          * last component (if wanted).
 548          */
 549         if (pn_pathleft(pnp) == 0) {
 550                 /*
 551                  * If there was a trailing slash in the pathname,
 552                  * make sure the last component is a directory.
 553                  */
 554                 if (must_be_directory && cvp->v_type != VDIR) {
 555                         error = ENOTDIR;
 556                         goto bad;
 557                 }
 558                 if (dirvpp != NULL) {
 559                         /*
 560                          * Check that we have the real parent and not
 561                          * an alias of the last component.
 562                          */
 563                         if (vn_compare(vp, cvp)) {
 564                                 if (auditing)
 565                                         (void) audit_savepath(pnp, cvp, vp,
 566                                             EINVAL, cr);
 567                                 pn_setlast(pnp);
 568                                 VN_RELE(vp);
 569                                 VN_RELE(cvp);
 570                                 if (rootvp != rootdir)
 571                                         VN_RELE(rootvp);
 572                                 if (pp)
 573                                         pn_free(pp);
 574                                 return (EINVAL);
 575                         }
 576                         *dirvpp = vp;
 577                 } else
 578                         VN_RELE(vp);
 579                 if (auditing)
 580                         (void) audit_savepath(pnp, cvp, vp, 0, cr);
 581                 if (pnp->pn_path == pnp->pn_buf)
 582                         (void) pn_set(pnp, ".");
 583                 else
 584                         pn_setlast(pnp);
 585                 if (rpnp) {
 586                         if (VN_CMP(cvp, rootvp))
 587                                 (void) pn_set(rpnp, "/");
 588                         else if (rpnp->pn_pathlen == 0)
 589                                 (void) pn_set(rpnp, ".");
 590                 }
 591 
 592                 if (compvpp != NULL)
 593                         *compvpp = cvp;
 594                 else
 595                         VN_RELE(cvp);
 596                 if (rootvp != rootdir)
 597                         VN_RELE(rootvp);
 598                 if (pp)
 599                         pn_free(pp);
 600                 return (0);
 601         }
 602 
 603         /*
 604          * Skip over slashes from end of last component.
 605          */
 606         while (pnp->pn_path[0] == '/') {
 607                 pnp->pn_path++;
 608                 pnp->pn_pathlen--;
 609         }
 610 
 611         /*
 612          * Searched through another level of directory:
 613          * release previous directory handle and save new (result
 614          * of lookup) as current directory.
 615          */
 616         VN_RELE(vp);
 617         vp = cvp;
 618         cvp = NULL;
 619         goto next;
 620 
 621 bad:
 622         if (auditing)   /* reached end of path */
 623                 (void) audit_savepath(pnp, cvp, vp, error, cr);
 624 bad_noaudit:
 625         /*
 626          * Error.  Release vnodes and return.
 627          */
 628         if (cvp)
 629                 VN_RELE(cvp);
 630         /*
 631          * If the error was ESTALE and the current directory to look in
 632          * was the root for this lookup, the root for a mounted file
 633          * system, or the starting directory for lookups, then
 634          * return ENOENT instead of ESTALE.  In this case, no recovery
 635          * is possible by the higher level.  If ESTALE was returned for
 636          * some intermediate directory along the path, then recovery
 637          * is potentially possible and retrying from the higher level
 638          * will either correct the situation by purging stale cache
 639          * entries or eventually get back to the point where no recovery
 640          * is possible.
 641          */
 642         if (error == ESTALE &&
 643             (VN_CMP(vp, rootvp) || (vp->v_flag & VROOT) || vp == startvp))
 644                 error = ENOENT;
 645         VN_RELE(vp);
 646         if (rootvp != rootdir)
 647                 VN_RELE(rootvp);
 648         if (pp)
 649                 pn_free(pp);
 650         return (error);
 651 }
 652 
 653 /*
 654  * Traverse a mount point.  Routine accepts a vnode pointer as a reference
 655  * parameter and performs the indirection, releasing the original vnode.
 656  */
 657 int
 658 traverse(vnode_t **cvpp)
 659 {
 660         int error = 0;
 661         vnode_t *cvp;
 662         vnode_t *tvp;
 663         vfs_t *vfsp;
 664 
 665         cvp = *cvpp;
 666 
 667         /*
 668          * If this vnode is mounted on, then we transparently indirect
 669          * to the vnode which is the root of the mounted file system.
 670          * Before we do this we must check that an unmount is not in
 671          * progress on this vnode.
 672          */
 673 
 674         for (;;) {
 675                 /*
 676                  * Try to read lock the vnode.  If this fails because
 677                  * the vnode is already write locked, then check to
 678                  * see whether it is the current thread which locked
 679                  * the vnode.  If it is not, then read lock the vnode
 680                  * by waiting to acquire the lock.
 681                  *
 682                  * The code path in domount() is an example of support
 683                  * which needs to look up two pathnames and locks one
 684                  * of them in between the two lookups.
 685                  */
 686                 error = vn_vfsrlock(cvp);
 687                 if (error) {
 688                         if (!vn_vfswlock_held(cvp))
 689                                 error = vn_vfsrlock_wait(cvp);
 690                         if (error != 0) {
 691                                 /*
 692                                  * lookuppn() expects a held vnode to be
 693                                  * returned because it promptly calls
 694                                  * VN_RELE after the error return
 695                                  */
 696                                 *cvpp = cvp;
 697                                 return (error);
 698                         }
 699                 }
 700 
 701                 /*
 702                  * Reached the end of the mount chain?
 703                  */
 704                 vfsp = vn_mountedvfs(cvp);
 705                 if (vfsp == NULL) {
 706                         vn_vfsunlock(cvp);
 707                         break;
 708                 }
 709 
 710                 /*
 711                  * The read lock must be held across the call to VFS_ROOT() to
 712                  * prevent a concurrent unmount from destroying the vfs.
 713                  */
 714                 error = VFS_ROOT(vfsp, &tvp);
 715                 vn_vfsunlock(cvp);
 716 
 717                 if (error)
 718                         break;
 719 
 720                 VN_RELE(cvp);
 721 
 722                 cvp = tvp;
 723         }
 724 
 725         *cvpp = cvp;
 726         return (error);
 727 }
 728 
 729 /*
 730  * Return the lowermost vnode if this is a mountpoint.
 731  */
 732 static vnode_t *
 733 vn_under(vnode_t *vp)
 734 {
 735         vnode_t *uvp;
 736         vfs_t *vfsp;
 737 
 738         while (vp->v_flag & VROOT) {
 739 
 740                 vfsp = vp->v_vfsp;
 741                 vfs_rlock_wait(vfsp);
 742                 if ((uvp = vfsp->vfs_vnodecovered) == NULL ||
 743                     (vfsp->vfs_flag & VFS_UNMOUNTED)) {
 744                         vfs_unlock(vfsp);
 745                         break;
 746                 }
 747                 VN_HOLD(uvp);
 748                 vfs_unlock(vfsp);
 749                 VN_RELE(vp);
 750                 vp = uvp;
 751         }
 752 
 753         return (vp);
 754 }
 755 
 756 static int
 757 vnode_match(vnode_t *v1, vnode_t *v2, cred_t *cr)
 758 {
 759         vattr_t v1attr, v2attr;
 760 
 761         /*
 762          * If we have a device file, check to see if is a cloned open of the
 763          * same device.  For self-cloning devices, the major numbers will match.
 764          * For devices cloned through the 'clone' driver, the minor number of
 765          * the source device will be the same as the major number of the cloned
 766          * device.
 767          */
 768         if ((v1->v_type == VCHR || v1->v_type == VBLK) &&
 769             v1->v_type == v2->v_type) {
 770                 if ((spec_is_selfclone(v1) || spec_is_selfclone(v2)) &&
 771                     getmajor(v1->v_rdev) == getmajor(v2->v_rdev))
 772                         return (1);
 773 
 774                 if (spec_is_clone(v1) &&
 775                     getmajor(v1->v_rdev) == getminor(v2->v_rdev))
 776                         return (1);
 777 
 778                 if (spec_is_clone(v2) &&
 779                     getmajor(v2->v_rdev) == getminor(v1->v_rdev))
 780                         return (1);
 781         }
 782 
 783         v1attr.va_mask = v2attr.va_mask = AT_TYPE;
 784 
 785         /*
 786          * This check for symbolic links handles the pseudo-symlinks in procfs.
 787          * These particular links have v_type of VDIR, but the attributes have a
 788          * type of VLNK.  We need to avoid these links because otherwise if we
 789          * are currently in '/proc/self/fd', then '/proc/self/cwd' will compare
 790          * as the same vnode.
 791          */
 792         if (VOP_GETATTR(v1, &v1attr, 0, cr, NULL) != 0 ||
 793             VOP_GETATTR(v2, &v2attr, 0, cr, NULL) != 0 ||
 794             v1attr.va_type == VLNK || v2attr.va_type == VLNK)
 795                 return (0);
 796 
 797         v1attr.va_mask = v2attr.va_mask = AT_TYPE | AT_FSID | AT_NODEID;
 798 
 799         if (VOP_GETATTR(v1, &v1attr, ATTR_REAL, cr, NULL) != 0 ||
 800             VOP_GETATTR(v2, &v2attr, ATTR_REAL, cr, NULL) != 0)
 801                 return (0);
 802 
 803         return (v1attr.va_fsid == v2attr.va_fsid &&
 804             v1attr.va_nodeid == v2attr.va_nodeid);
 805 }
 806 
 807 
 808 /*
 809  * Find the entry in the directory corresponding to the target vnode.
 810  */
 811 int
 812 dirfindvp(vnode_t *vrootp, vnode_t *dvp, vnode_t *tvp, cred_t *cr, char *dbuf,
 813     size_t dlen, dirent64_t **rdp)
 814 {
 815         size_t dbuflen;
 816         struct iovec iov;
 817         struct uio uio;
 818         int error;
 819         int eof;
 820         vnode_t *cmpvp;
 821         struct dirent64 *dp;
 822         pathname_t pnp;
 823 
 824         ASSERT(dvp->v_type == VDIR);
 825 
 826         /*
 827          * This is necessary because of the strange semantics of VOP_LOOKUP().
 828          */
 829         bzero(&pnp, sizeof (pnp));
 830 
 831         eof = 0;
 832 
 833         uio.uio_iov = &iov;
 834         uio.uio_iovcnt = 1;
 835         uio.uio_segflg = UIO_SYSSPACE;
 836         uio.uio_fmode = 0;
 837         uio.uio_extflg = UIO_COPY_CACHED;
 838         uio.uio_loffset = 0;
 839 
 840         if ((error = VOP_ACCESS(dvp, VREAD, 0, cr, NULL)) != 0)
 841                 return (error);
 842 
 843         while (!eof) {
 844                 uio.uio_resid = dlen;
 845                 iov.iov_base = dbuf;
 846                 iov.iov_len = dlen;
 847 
 848                 (void) VOP_RWLOCK(dvp, V_WRITELOCK_FALSE, NULL);
 849                 error = VOP_READDIR(dvp, &uio, cr, &eof, NULL, 0);
 850                 VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
 851 
 852                 dbuflen = dlen - uio.uio_resid;
 853 
 854                 if (error || dbuflen == 0)
 855                         break;
 856 
 857                 dp = (dirent64_t *)dbuf;
 858                 while ((intptr_t)dp < (intptr_t)dbuf + dbuflen) {
 859                         /*
 860                          * Ignore '.' and '..' entries
 861                          */
 862                         if (strcmp(dp->d_name, ".") == 0 ||
 863                             strcmp(dp->d_name, "..") == 0) {
 864                                 dp = (dirent64_t *)((intptr_t)dp +
 865                                     dp->d_reclen);
 866                                 continue;
 867                         }
 868 
 869                         error = VOP_LOOKUP(dvp, dp->d_name, &cmpvp, &pnp, 0,
 870                             vrootp, cr, NULL, NULL, NULL);
 871 
 872                         /*
 873                          * We only want to bail out if there was an error other
 874                          * than ENOENT.  Otherwise, it could be that someone
 875                          * just removed an entry since the readdir() call, and
 876                          * the entry we want is further on in the directory.
 877                          */
 878                         if (error == 0) {
 879                                 if (vnode_match(tvp, cmpvp, cr)) {
 880                                         VN_RELE(cmpvp);
 881                                         *rdp = dp;
 882                                         return (0);
 883                                 }
 884 
 885                                 VN_RELE(cmpvp);
 886                         } else if (error != ENOENT) {
 887                                 return (error);
 888                         }
 889 
 890                         dp = (dirent64_t *)((intptr_t)dp + dp->d_reclen);
 891                 }
 892         }
 893 
 894         /*
 895          * Something strange has happened, this directory does not contain the
 896          * specified vnode.  This should never happen in the normal case, since
 897          * we ensured that dvp is the parent of vp.  This is possible in some
 898          * rare conditions (races and the special .zfs directory).
 899          */
 900         if (error == 0) {
 901                 error = VOP_LOOKUP(dvp, ".zfs", &cmpvp, &pnp, 0, vrootp, cr,
 902                     NULL, NULL, NULL);
 903                 if (error == 0) {
 904                         if (vnode_match(tvp, cmpvp, cr)) {
 905                                 (void) strcpy(dp->d_name, ".zfs");
 906                                 dp->d_reclen = strlen(".zfs");
 907                                 dp->d_off = 2;
 908                                 dp->d_ino = 1;
 909                                 *rdp = dp;
 910                         } else {
 911                                 error = ENOENT;
 912                         }
 913                         VN_RELE(cmpvp);
 914                 }
 915         }
 916 
 917         return (error);
 918 }
 919 
 920 /*
 921  * Given a global path (from rootdir), and a vnode that is the current root,
 922  * return the portion of the path that is beneath the current root or NULL on
 923  * failure.  The path MUST be a resolved path (no '..' entries or symlinks),
 924  * otherwise this function will fail.
 925  */
 926 static char *
 927 localpath(char *path, struct vnode *vrootp, cred_t *cr)
 928 {
 929         vnode_t *vp;
 930         vnode_t *cvp;
 931         char component[MAXNAMELEN];
 932         char *ret = NULL;
 933         pathname_t pn;
 934 
 935         /*
 936          * We use vn_compare() instead of VN_CMP() in order to detect lofs
 937          * mounts and stacked vnodes.
 938          */
 939         if (vn_compare(vrootp, rootdir))
 940                 return (path);
 941 
 942         if (pn_get(path, UIO_SYSSPACE, &pn) != 0)
 943                 return (NULL);
 944 
 945         vp = rootdir;
 946         VN_HOLD(vp);
 947 
 948         if (vn_ismntpt(vp) && traverse(&vp) != 0) {
 949                 VN_RELE(vp);
 950                 pn_free(&pn);
 951                 return (NULL);
 952         }
 953 
 954         while (pn_pathleft(&pn)) {
 955                 pn_skipslash(&pn);
 956 
 957                 if (pn_getcomponent(&pn, component) != 0)
 958                         break;
 959 
 960                 if (VOP_LOOKUP(vp, component, &cvp, &pn, 0, rootdir, cr,
 961                     NULL, NULL, NULL) != 0)
 962                         break;
 963                 VN_RELE(vp);
 964                 vp = cvp;
 965 
 966                 if (vn_ismntpt(vp) && traverse(&vp) != 0)
 967                         break;
 968 
 969                 if (vn_compare(vp, vrootp)) {
 970                         ret = path + (pn.pn_path - pn.pn_buf);
 971                         break;
 972                 }
 973         }
 974 
 975         VN_RELE(vp);
 976         pn_free(&pn);
 977 
 978         return (ret);
 979 }
 980 
 981 /*
 982  * Given a directory, return the full, resolved path.  This looks up "..",
 983  * searches for the given vnode in the parent, appends the component, etc.  It
 984  * is used to implement vnodetopath() and getcwd() when the cached path fails.
 985  */
 986 static int
 987 dirtopath(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen, int flags,
 988     cred_t *cr)
 989 {
 990         pathname_t pn, rpn, emptypn;
 991         vnode_t *cmpvp, *pvp = NULL;
 992         vnode_t *startvp = vp;
 993         int err = 0, vprivs;
 994         size_t complen;
 995         char *dbuf;
 996         dirent64_t *dp;
 997         char            *bufloc;
 998         size_t          dlen = DIRENT64_RECLEN(MAXPATHLEN);
 999         refstr_t        *mntpt;
1000 
1001         /* Operation only allowed on directories */
1002         ASSERT(vp->v_type == VDIR);
1003 
1004         /* We must have at least enough space for "/" */
1005         if (buflen < 2)
1006                 return (ENAMETOOLONG);
1007 
1008         /* Start at end of string with terminating null */
1009         bufloc = &buf[buflen - 1];
1010         *bufloc = '\0';
1011 
1012         pn_alloc(&pn);
1013         pn_alloc(&rpn);
1014         dbuf = kmem_alloc(dlen, KM_SLEEP);
1015         bzero(&emptypn, sizeof (emptypn));
1016 
1017         /*
1018          * Begin with an additional reference on vp.  This will be decremented
1019          * during the loop.
1020          */
1021         VN_HOLD(vp);
1022 
1023         for (;;) {
1024                 /*
1025                  * Return if we've reached the root.  If the buffer is empty,
1026                  * return '/'.  We explicitly don't use vn_compare(), since it
1027                  * compares the real vnodes.  A lofs mount of '/' would produce
1028                  * incorrect results otherwise.
1029                  */
1030                 if (VN_CMP(vrootp, vp)) {
1031                         if (*bufloc == '\0')
1032                                 *--bufloc = '/';
1033                         break;
1034                 }
1035 
1036                 /*
1037                  * If we've reached the VFS root, something has gone wrong.  We
1038                  * should have reached the root in the above check.  The only
1039                  * explantation is that 'vp' is not contained withing the given
1040                  * root, in which case we return EPERM.
1041                  */
1042                 if (VN_CMP(rootdir, vp)) {
1043                         err = EPERM;
1044                         goto out;
1045                 }
1046 
1047                 /*
1048                  * Shortcut: see if this vnode is a mountpoint.  If so,
1049                  * grab the path information from the vfs_t.
1050                  */
1051                 if (vp->v_flag & VROOT) {
1052 
1053                         mntpt = vfs_getmntpoint(vp->v_vfsp);
1054                         if ((err = pn_set(&pn, (char *)refstr_value(mntpt)))
1055                             == 0) {
1056                                 refstr_rele(mntpt);
1057                                 rpn.pn_path = rpn.pn_buf;
1058 
1059                                 /*
1060                                  * Ensure the mountpoint still exists.
1061                                  */
1062                                 VN_HOLD(vrootp);
1063                                 if (vrootp != rootdir)
1064                                         VN_HOLD(vrootp);
1065                                 if (lookuppnvp(&pn, &rpn, flags, NULL,
1066                                     &cmpvp, vrootp, vrootp, cr) == 0) {
1067 
1068                                         if (VN_CMP(vp, cmpvp)) {
1069                                                 VN_RELE(cmpvp);
1070 
1071                                                 complen = strlen(rpn.pn_path);
1072                                                 bufloc -= complen;
1073                                                 if (bufloc < buf) {
1074                                                         err = ERANGE;
1075                                                         goto out;
1076                                                 }
1077                                                 bcopy(rpn.pn_path, bufloc,
1078                                                     complen);
1079                                                 break;
1080                                         } else {
1081                                                 VN_RELE(cmpvp);
1082                                         }
1083                                 }
1084                         } else {
1085                                 refstr_rele(mntpt);
1086                         }
1087                 }
1088 
1089                 /*
1090                  * Shortcut: see if this vnode has correct v_path. If so,
1091                  * we have the work done.
1092                  */
1093                 mutex_enter(&vp->v_lock);
1094                 if (vp->v_path != NULL) {
1095 
1096                         if ((err = pn_set(&pn, vp->v_path)) == 0) {
1097                                 mutex_exit(&vp->v_lock);
1098                                 rpn.pn_path = rpn.pn_buf;
1099 
1100                                 /*
1101                                  * Ensure the v_path pointing to correct vnode
1102                                  */
1103                                 VN_HOLD(vrootp);
1104                                 if (vrootp != rootdir)
1105                                         VN_HOLD(vrootp);
1106                                 if (lookuppnvp(&pn, &rpn, flags, NULL,
1107                                     &cmpvp, vrootp, vrootp, cr) == 0) {
1108 
1109                                         if (VN_CMP(vp, cmpvp)) {
1110                                                 VN_RELE(cmpvp);
1111 
1112                                                 complen = strlen(rpn.pn_path);
1113                                                 bufloc -= complen;
1114                                                 if (bufloc < buf) {
1115                                                         err = ERANGE;
1116                                                         goto out;
1117                                                 }
1118                                                 bcopy(rpn.pn_path, bufloc,
1119                                                     complen);
1120                                                 break;
1121                                         } else {
1122                                                 VN_RELE(cmpvp);
1123                                         }
1124                                 }
1125                         } else {
1126                                 mutex_exit(&vp->v_lock);
1127                         }
1128                 } else {
1129                         mutex_exit(&vp->v_lock);
1130                 }
1131 
1132                 /*
1133                  * Shortcuts failed, search for this vnode in its parent.  If
1134                  * this is a mountpoint, then get the vnode underneath.
1135                  */
1136                 if (vp->v_flag & VROOT)
1137                         vp = vn_under(vp);
1138                 if ((err = VOP_LOOKUP(vp, "..", &pvp, &emptypn, 0, vrootp, cr,
1139                     NULL, NULL, NULL)) != 0)
1140                         goto out;
1141 
1142                 /*
1143                  * With extended attributes, it's possible for a directory to
1144                  * have a parent that is a regular file.  Check for that here.
1145                  */
1146                 if (pvp->v_type != VDIR) {
1147                         err = ENOTDIR;
1148                         goto out;
1149                 }
1150 
1151                 /*
1152                  * If this is true, something strange has happened.  This is
1153                  * only true if we are the root of a filesystem, which should
1154                  * have been caught by the check above.
1155                  */
1156                 if (VN_CMP(pvp, vp)) {
1157                         err = ENOENT;
1158                         goto out;
1159                 }
1160 
1161                 /*
1162                  * Check if we have read and search privilege so, that
1163                  * we can lookup the path in the directory
1164                  */
1165                 vprivs = (flags & LOOKUP_CHECKREAD) ? VREAD | VEXEC : VEXEC;
1166                 if ((err = VOP_ACCESS(pvp, vprivs, 0, cr, NULL)) != 0) {
1167                         goto out;
1168                 }
1169 
1170                 /*
1171                  * Try to obtain the path component from dnlc cache
1172                  * before searching through the directory.
1173                  */
1174                 if ((cmpvp = dnlc_reverse_lookup(vp, dbuf, dlen)) != NULL) {
1175                         /*
1176                          * If we got parent vnode as a result,
1177                          * then the answered path is correct.
1178                          */
1179                         if (VN_CMP(cmpvp, pvp)) {
1180                                 VN_RELE(cmpvp);
1181                                 complen = strlen(dbuf);
1182                                 bufloc -= complen;
1183                                 if (bufloc <= buf) {
1184                                         err = ENAMETOOLONG;
1185                                         goto out;
1186                                 }
1187                                 bcopy(dbuf, bufloc, complen);
1188 
1189                                 /* Prepend a slash to the current path */
1190                                 *--bufloc = '/';
1191 
1192                                 /* And continue with the next component */
1193                                 VN_RELE(vp);
1194                                 vp = pvp;
1195                                 pvp = NULL;
1196                                 continue;
1197                         } else {
1198                                 VN_RELE(cmpvp);
1199                         }
1200                 }
1201 
1202                 /*
1203                  * Search the parent directory for the entry corresponding to
1204                  * this vnode.
1205                  */
1206                 if ((err = dirfindvp(vrootp, pvp, vp, cr, dbuf, dlen, &dp))
1207                     != 0)
1208                         goto out;
1209                 complen = strlen(dp->d_name);
1210                 bufloc -= complen;
1211                 if (bufloc <= buf) {
1212                         err = ENAMETOOLONG;
1213                         goto out;
1214                 }
1215                 bcopy(dp->d_name, bufloc, complen);
1216 
1217                 /* Prepend a slash to the current path.  */
1218                 *--bufloc = '/';
1219 
1220                 /* And continue with the next component */
1221                 VN_RELE(vp);
1222                 vp = pvp;
1223                 pvp = NULL;
1224         }
1225 
1226         /*
1227          * Place the path at the beginning of the buffer.
1228          */
1229         if (bufloc != buf)
1230                 ovbcopy(bufloc, buf, buflen - (bufloc - buf));
1231 
1232 out:
1233         /*
1234          * If the error was ESTALE and the current directory to look in
1235          * was the root for this lookup, the root for a mounted file
1236          * system, or the starting directory for lookups, then
1237          * return ENOENT instead of ESTALE.  In this case, no recovery
1238          * is possible by the higher level.  If ESTALE was returned for
1239          * some intermediate directory along the path, then recovery
1240          * is potentially possible and retrying from the higher level
1241          * will either correct the situation by purging stale cache
1242          * entries or eventually get back to the point where no recovery
1243          * is possible.
1244          */
1245         if (err == ESTALE &&
1246             (VN_CMP(vp, vrootp) || (vp->v_flag & VROOT) || vp == startvp))
1247                 err = ENOENT;
1248 
1249         kmem_free(dbuf, dlen);
1250         VN_RELE(vp);
1251         if (pvp)
1252                 VN_RELE(pvp);
1253         pn_free(&pn);
1254         pn_free(&rpn);
1255 
1256         return (err);
1257 }
1258 
1259 /*
1260  * The additional flag, LOOKUP_CHECKREAD, is used to enforce artificial
1261  * constraints in order to be standards compliant.  For example, if we have
1262  * the cached path of '/foo/bar', and '/foo' has permissions 100 (execute
1263  * only), then we can legitimately look up the path to the current working
1264  * directory without needing read permission.  Existing standards tests,
1265  * however, assume that we are determining the path by repeatedly looking up
1266  * "..".  We need to keep this behavior in order to maintain backwards
1267  * compatibility.
1268  */
1269 static int
1270 vnodetopath_common(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen,
1271     cred_t *cr, int flags)
1272 {
1273         pathname_t pn, rpn;
1274         int ret, len;
1275         vnode_t *compvp, *pvp, *realvp;
1276         proc_t *p = curproc;
1277         char path[MAXNAMELEN];
1278         int doclose = 0;
1279 
1280         /*
1281          * If vrootp is NULL, get the root for curproc.  Callers with any other
1282          * requirements should pass in a different vrootp.
1283          */
1284         if (vrootp == NULL) {
1285                 mutex_enter(&p->p_lock);
1286                 if ((vrootp = PTOU(p)->u_rdir) == NULL)
1287                         vrootp = rootdir;
1288                 VN_HOLD(vrootp);
1289                 mutex_exit(&p->p_lock);
1290         } else {
1291                 VN_HOLD(vrootp);
1292         }
1293 
1294         /*
1295          * This is to get around an annoying artifact of the /proc filesystem,
1296          * which is the behavior of {cwd/root}.  Trying to resolve this path
1297          * will result in /proc/pid/cwd instead of whatever the real working
1298          * directory is.  We can't rely on VOP_REALVP(), since that will break
1299          * lofs.  The only difference between procfs and lofs is that opening
1300          * the file will return the underling vnode in the case of procfs.
1301          */
1302         if (vp->v_type == VDIR && VOP_REALVP(vp, &realvp, NULL) == 0 &&
1303             realvp != vp) {
1304                 VN_HOLD(vp);
1305                 if (VOP_OPEN(&vp, FREAD, cr, NULL) == 0)
1306                         doclose = 1;
1307                 else
1308                         VN_RELE(vp);
1309         }
1310 
1311         pn_alloc(&pn);
1312 
1313         /*
1314          * Check to see if we have a cached path in the vnode.
1315          */
1316         mutex_enter(&vp->v_lock);
1317         if (vp->v_path != NULL) {
1318                 (void) pn_set(&pn, vp->v_path);
1319                 mutex_exit(&vp->v_lock);
1320 
1321                 pn_alloc(&rpn);
1322 
1323                 /* We should only cache absolute paths */
1324                 ASSERT(pn.pn_buf[0] == '/');
1325 
1326                 /*
1327                  * If we are in a zone or a chroot environment, then we have to
1328                  * take additional steps, since the path to the root might not
1329                  * be readable with the current credentials, even though the
1330                  * process can legitmately access the file.  In this case, we
1331                  * do the following:
1332                  *
1333                  * lookuppnvp() with all privileges to get the resolved path.
1334                  * call localpath() to get the local portion of the path, and
1335                  * continue as normal.
1336                  *
1337                  * If the the conversion to a local path fails, then we continue
1338                  * as normal.  This is a heuristic to make process object file
1339                  * paths available from within a zone.  Because lofs doesn't
1340                  * support page operations, the vnode stored in the seg_t is
1341                  * actually the underlying real vnode, not the lofs node itself.
1342                  * Most of the time, the lofs path is the same as the underlying
1343                  * vnode (for example, /usr/lib/libc.so.1).
1344                  */
1345                 if (vrootp != rootdir) {
1346                         char *local = NULL;
1347                         VN_HOLD(rootdir);
1348                         if (lookuppnvp(&pn, &rpn, FOLLOW,
1349                             NULL, &compvp, rootdir, rootdir, kcred) == 0) {
1350                                 local = localpath(rpn.pn_path, vrootp,
1351                                     kcred);
1352                                 VN_RELE(compvp);
1353                         }
1354 
1355                         /*
1356                          * The original pn was changed through lookuppnvp().
1357                          * Set it to local for next validation attempt.
1358                          */
1359                         if (local) {
1360                                 (void) pn_set(&pn, local);
1361                         } else {
1362                                 goto notcached;
1363                         }
1364                 }
1365 
1366                 /*
1367                  * We should have a local path at this point, so start the
1368                  * search from the root of the current process.
1369                  */
1370                 VN_HOLD(vrootp);
1371                 if (vrootp != rootdir)
1372                         VN_HOLD(vrootp);
1373                 ret = lookuppnvp(&pn, &rpn, FOLLOW | flags, NULL,
1374                     &compvp, vrootp, vrootp, cr);
1375                 if (ret == 0) {
1376                         /*
1377                          * Check to see if the returned vnode is the same as
1378                          * the one we expect.  If not, give up.
1379                          */
1380                         if (!vn_compare(vp, compvp) &&
1381                             !vnode_match(vp, compvp, cr)) {
1382                                 VN_RELE(compvp);
1383                                 goto notcached;
1384                         }
1385 
1386                         VN_RELE(compvp);
1387 
1388                         /*
1389                          * Return the result.
1390                          */
1391                         if (buflen <= rpn.pn_pathlen)
1392                                 goto notcached;
1393 
1394                         bcopy(rpn.pn_path, buf, rpn.pn_pathlen + 1);
1395                         pn_free(&pn);
1396                         pn_free(&rpn);
1397                         VN_RELE(vrootp);
1398                         if (doclose) {
1399                                 (void) VOP_CLOSE(vp, FREAD, 1, 0, cr, NULL);
1400                                 VN_RELE(vp);
1401                         }
1402                         return (0);
1403                 }
1404 
1405 notcached:
1406                 pn_free(&rpn);
1407         } else {
1408                 mutex_exit(&vp->v_lock);
1409         }
1410 
1411         pn_free(&pn);
1412 
1413         if (PROC_IS_BRANDED(curproc)) {
1414                 /*
1415                  * If v_path doesn't work out and we're in a branded zone,
1416                  * we're not going to bother doing more work here:  because
1417                  * directories from the global can be lofs mounted into odd
1418                  * locations (e.g., /native in an lx zone), it is likely that
1419                  * the DNLC reverse lookup will yield nothing.  Indeed, the
1420                  * only certainty is that the DNLC reverse lookup will be
1421                  * exceedingly painful; we save ourselves the substantial
1422                  * grief of scanning the entire DNLC and kick out with ENOENT
1423                  * in this case.
1424                  */
1425                 ret = ENOENT;
1426         } else if (vp->v_type != VDIR) {
1427                 /*
1428                  * If we don't have a directory, try to find it in the dnlc via
1429                  * reverse lookup.  Once this is found, we can use the regular
1430                  * directory search to find the full path.
1431                  */
1432                 if ((pvp = dnlc_reverse_lookup(vp, path, MAXNAMELEN)) != NULL) {
1433                         /*
1434                          * Check if we have read privilege so, that
1435                          * we can lookup the path in the directory
1436                          */
1437                         ret = 0;
1438                         if ((flags & LOOKUP_CHECKREAD)) {
1439                                 ret = VOP_ACCESS(pvp, VREAD, 0, cr, NULL);
1440                         }
1441                         if (ret == 0) {
1442                                 ret = dirtopath(vrootp, pvp, buf, buflen,
1443                                     flags, cr);
1444                         }
1445                         if (ret == 0) {
1446                                 len = strlen(buf);
1447                                 if (len + strlen(path) + 1 >= buflen) {
1448                                         ret = ENAMETOOLONG;
1449                                 } else {
1450                                         if (buf[len - 1] != '/')
1451                                                 buf[len++] = '/';
1452                                         bcopy(path, buf + len,
1453                                             strlen(path) + 1);
1454                                 }
1455                         }
1456 
1457                         VN_RELE(pvp);
1458                 } else
1459                         ret = ENOENT;
1460         } else
1461                 ret = dirtopath(vrootp, vp, buf, buflen, flags, cr);
1462 
1463         VN_RELE(vrootp);
1464         if (doclose) {
1465                 (void) VOP_CLOSE(vp, FREAD, 1, 0, cr, NULL);
1466                 VN_RELE(vp);
1467         }
1468 
1469         return (ret);
1470 }
1471 
1472 int
1473 vnodetopath(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen, cred_t *cr)
1474 {
1475         return (vnodetopath_common(vrootp, vp, buf, buflen, cr, 0));
1476 }
1477 
1478 int
1479 dogetcwd(char *buf, size_t buflen)
1480 {
1481         int ret;
1482         vnode_t *vp;
1483         vnode_t *compvp;
1484         refstr_t *cwd, *oldcwd;
1485         const char *value;
1486         pathname_t rpnp, pnp;
1487         proc_t *p = curproc;
1488 
1489         /*
1490          * Check to see if there is a cached version of the cwd.  If so, lookup
1491          * the cached value and make sure it is the same vnode.
1492          */
1493         mutex_enter(&p->p_lock);
1494         if ((cwd = PTOU(p)->u_cwd) != NULL)
1495                 refstr_hold(cwd);
1496         vp = PTOU(p)->u_cdir;
1497         VN_HOLD(vp);
1498         mutex_exit(&p->p_lock);
1499 
1500         /*
1501          * Make sure we have permission to access the current directory.
1502          */
1503         if ((ret = VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) != 0) {
1504                 if (cwd != NULL)
1505                         refstr_rele(cwd);
1506                 VN_RELE(vp);
1507                 return (ret);
1508         }
1509 
1510         if (cwd) {
1511                 value = refstr_value(cwd);
1512                 if ((ret = pn_get((char *)value, UIO_SYSSPACE, &pnp)) != 0) {
1513                         refstr_rele(cwd);
1514                         VN_RELE(vp);
1515                         return (ret);
1516                 }
1517 
1518                 pn_alloc(&rpnp);
1519 
1520                 if (lookuppn(&pnp, &rpnp, NO_FOLLOW, NULL, &compvp) == 0) {
1521 
1522                         if (VN_CMP(vp, compvp) &&
1523                             strcmp(value, rpnp.pn_path) == 0) {
1524                                 VN_RELE(compvp);
1525                                 VN_RELE(vp);
1526                                 pn_free(&pnp);
1527                                 pn_free(&rpnp);
1528                                 if (strlen(value) + 1 > buflen) {
1529                                         refstr_rele(cwd);
1530                                         return (ENAMETOOLONG);
1531                                 }
1532                                 bcopy(value, buf, strlen(value) + 1);
1533                                 refstr_rele(cwd);
1534                                 return (0);
1535                         }
1536 
1537                         VN_RELE(compvp);
1538                 }
1539 
1540                 pn_free(&rpnp);
1541                 pn_free(&pnp);
1542 
1543                 refstr_rele(cwd);
1544         }
1545 
1546         ret = vnodetopath_common(NULL, vp, buf, buflen, CRED(),
1547             LOOKUP_CHECKREAD);
1548 
1549         VN_RELE(vp);
1550 
1551         /*
1552          * Store the new cwd and replace the existing cached copy.
1553          */
1554         if (ret == 0)
1555                 cwd = refstr_alloc(buf);
1556         else
1557                 cwd = NULL;
1558 
1559         mutex_enter(&p->p_lock);
1560         oldcwd = PTOU(p)->u_cwd;
1561         PTOU(p)->u_cwd = cwd;
1562         mutex_exit(&p->p_lock);
1563 
1564         if (oldcwd)
1565                 refstr_rele(oldcwd);
1566 
1567         return (ret);
1568 }