The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_lookup.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1982, 1986, 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  * (c) UNIX System Laboratories, Inc.
    5  * All or some portions of this file are derived from material licensed
    6  * to the University of California by American Telephone and Telegraph
    7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
    8  * the permission of UNIX System Laboratories, Inc.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 4. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)vfs_lookup.c        8.4 (Berkeley) 2/16/94
   35  */
   36 
   37 #include <sys/cdefs.h>
   38 __FBSDID("$FreeBSD$");
   39 
   40 #include "opt_capsicum.h"
   41 #include "opt_kdtrace.h"
   42 #include "opt_ktrace.h"
   43 
   44 #include <sys/param.h>
   45 #include <sys/systm.h>
   46 #include <sys/kernel.h>
   47 #include <sys/capability.h>
   48 #include <sys/fcntl.h>
   49 #include <sys/jail.h>
   50 #include <sys/lock.h>
   51 #include <sys/mutex.h>
   52 #include <sys/namei.h>
   53 #include <sys/vnode.h>
   54 #include <sys/mount.h>
   55 #include <sys/filedesc.h>
   56 #include <sys/proc.h>
   57 #include <sys/sdt.h>
   58 #include <sys/syscallsubr.h>
   59 #include <sys/sysctl.h>
   60 #ifdef KTRACE
   61 #include <sys/ktrace.h>
   62 #endif
   63 
   64 #include <security/audit/audit.h>
   65 #include <security/mac/mac_framework.h>
   66 
   67 #include <vm/uma.h>
   68 
   69 #define NAMEI_DIAGNOSTIC 1
   70 #undef NAMEI_DIAGNOSTIC
   71 
   72 SDT_PROVIDER_DECLARE(vfs);
   73 SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *",
   74     "unsigned long");
   75 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *");
   76 
   77 /*
   78  * Allocation zone for namei
   79  */
   80 uma_zone_t namei_zone;
   81 /*
   82  * Placeholder vnode for mp traversal
   83  */
   84 static struct vnode *vp_crossmp;
   85 
   86 static void
   87 nameiinit(void *dummy __unused)
   88 {
   89 
   90         namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
   91             UMA_ALIGN_PTR, 0);
   92         getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp);
   93         vn_lock(vp_crossmp, LK_EXCLUSIVE);
   94         VN_LOCK_ASHARE(vp_crossmp);
   95         VOP_UNLOCK(vp_crossmp, 0);
   96 }
   97 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
   98 
   99 static int lookup_shared = 1;
  100 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
  101     "Enables/Disables shared locks for path name translation");
  102 TUNABLE_INT("vfs.lookup_shared", &lookup_shared);
  103 
  104 /*
  105  * Convert a pathname into a pointer to a locked vnode.
  106  *
  107  * The FOLLOW flag is set when symbolic links are to be followed
  108  * when they occur at the end of the name translation process.
  109  * Symbolic links are always followed for all other pathname
  110  * components other than the last.
  111  *
  112  * The segflg defines whether the name is to be copied from user
  113  * space or kernel space.
  114  *
  115  * Overall outline of namei:
  116  *
  117  *      copy in name
  118  *      get starting directory
  119  *      while (!done && !error) {
  120  *              call lookup to search path.
  121  *              if symbolic link, massage name in buffer and continue
  122  *      }
  123  */
  124 static void
  125 namei_cleanup_cnp(struct componentname *cnp)
  126 {
  127         uma_zfree(namei_zone, cnp->cn_pnbuf);
  128 #ifdef DIAGNOSTIC
  129         cnp->cn_pnbuf = NULL;
  130         cnp->cn_nameptr = NULL;
  131 #endif
  132 }
  133 
  134 int
  135 namei(struct nameidata *ndp)
  136 {
  137         struct filedesc *fdp;   /* pointer to file descriptor state */
  138         char *cp;               /* pointer into pathname argument */
  139         struct vnode *dp;       /* the directory we are searching */
  140         struct iovec aiov;              /* uio for reading symbolic links */
  141         struct uio auio;
  142         int error, linklen;
  143         struct componentname *cnp = &ndp->ni_cnd;
  144         struct thread *td = cnp->cn_thread;
  145         struct proc *p = td->td_proc;
  146         int vfslocked;
  147 
  148         KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
  149             ("NOT MPSAFE and Giant not held"));
  150         ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
  151         KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
  152         KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
  153             ("namei: nameiop contaminated with flags"));
  154         KASSERT((cnp->cn_flags & OPMASK) == 0,
  155             ("namei: flags contaminated with nameiops"));
  156         if (!lookup_shared)
  157                 cnp->cn_flags &= ~LOCKSHARED;
  158         fdp = p->p_fd;
  159 
  160         /* We will set this ourselves if we need it. */
  161         cnp->cn_flags &= ~TRAILINGSLASH;
  162 
  163         /*
  164          * Get a buffer for the name to be translated, and copy the
  165          * name into the buffer.
  166          */
  167         if ((cnp->cn_flags & HASBUF) == 0)
  168                 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
  169         if (ndp->ni_segflg == UIO_SYSSPACE)
  170                 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
  171                             MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
  172         else
  173                 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
  174                             MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
  175 
  176         /*
  177          * Don't allow empty pathnames.
  178          */
  179         if (!error && *cnp->cn_pnbuf == '\0')
  180                 error = ENOENT;
  181 
  182 #ifdef CAPABILITY_MODE
  183         /*
  184          * In capability mode, lookups must be "strictly relative" (i.e.
  185          * not an absolute path, and not containing '..' components) to
  186          * a real file descriptor, not the pseudo-descriptor AT_FDCWD.
  187          */
  188         if (error == 0 && IN_CAPABILITY_MODE(td)) {
  189                 ndp->ni_strictrelative = 1;
  190                 if (ndp->ni_dirfd == AT_FDCWD)
  191                         error = ECAPMODE;
  192         }
  193 #endif
  194         if (error) {
  195                 namei_cleanup_cnp(cnp);
  196                 ndp->ni_vp = NULL;
  197                 return (error);
  198         }
  199         ndp->ni_loopcnt = 0;
  200 #ifdef KTRACE
  201         if (KTRPOINT(td, KTR_NAMEI)) {
  202                 KASSERT(cnp->cn_thread == curthread,
  203                     ("namei not using curthread"));
  204                 ktrnamei(cnp->cn_pnbuf);
  205         }
  206 #endif
  207         /*
  208          * Get starting point for the translation.
  209          */
  210         FILEDESC_SLOCK(fdp);
  211         ndp->ni_rootdir = fdp->fd_rdir;
  212         ndp->ni_topdir = fdp->fd_jdir;
  213 
  214         /*
  215          * If we are auditing the kernel pathname, save the user pathname.
  216          */
  217         if (cnp->cn_flags & AUDITVNODE1)
  218                 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
  219         if (cnp->cn_flags & AUDITVNODE2)
  220                 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
  221 
  222         dp = NULL;
  223         if (cnp->cn_pnbuf[0] != '/') {
  224                 if (ndp->ni_startdir != NULL) {
  225                         dp = ndp->ni_startdir;
  226                         error = 0;
  227                 } else if (ndp->ni_dirfd != AT_FDCWD) {
  228                         if (cnp->cn_flags & AUDITVNODE1)
  229                                 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
  230                         if (cnp->cn_flags & AUDITVNODE2)
  231                                 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
  232                         error = fgetvp_rights(td, ndp->ni_dirfd,
  233                             ndp->ni_rightsneeded | CAP_LOOKUP,
  234                             &(ndp->ni_baserights), &dp);
  235 #ifdef CAPABILITIES
  236                         /*
  237                          * Lookups relative to a capability must also be
  238                          * strictly relative.
  239                          *
  240                          * Note that a capability with rights CAP_MASK_VALID
  241                          * is treated exactly like a regular file descriptor.
  242                          */
  243                         if (ndp->ni_baserights != CAP_MASK_VALID)
  244                                 ndp->ni_strictrelative = 1;
  245 #endif
  246                 }
  247                 if (error != 0 || dp != NULL) {
  248                         FILEDESC_SUNLOCK(fdp);
  249                         if (error == 0 && dp->v_type != VDIR) {
  250                                 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
  251                                 vrele(dp);
  252                                 VFS_UNLOCK_GIANT(vfslocked);
  253                                 error = ENOTDIR;
  254                         }
  255                 }
  256                 if (error) {
  257                         namei_cleanup_cnp(cnp);
  258                         return (error);
  259                 }
  260         }
  261         if (dp == NULL) {
  262                 dp = fdp->fd_cdir;
  263                 VREF(dp);
  264                 FILEDESC_SUNLOCK(fdp);
  265                 if (ndp->ni_startdir != NULL) {
  266                         vfslocked = VFS_LOCK_GIANT(ndp->ni_startdir->v_mount);
  267                         vrele(ndp->ni_startdir);
  268                         VFS_UNLOCK_GIANT(vfslocked);
  269                 }
  270         }
  271         SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
  272             cnp->cn_flags);
  273         vfslocked = VFS_LOCK_GIANT(dp->v_mount);
  274         for (;;) {
  275                 /*
  276                  * Check if root directory should replace current directory.
  277                  * Done at start of translation and after symbolic link.
  278                  */
  279                 cnp->cn_nameptr = cnp->cn_pnbuf;
  280                 if (*(cnp->cn_nameptr) == '/') {
  281                         vrele(dp);
  282                         VFS_UNLOCK_GIANT(vfslocked);
  283                         if (ndp->ni_strictrelative != 0) {
  284                                 namei_cleanup_cnp(cnp);
  285                                 return (ENOTCAPABLE);
  286                         }
  287                         while (*(cnp->cn_nameptr) == '/') {
  288                                 cnp->cn_nameptr++;
  289                                 ndp->ni_pathlen--;
  290                         }
  291                         dp = ndp->ni_rootdir;
  292                         vfslocked = VFS_LOCK_GIANT(dp->v_mount);
  293                         VREF(dp);
  294                 }
  295                 if (vfslocked)
  296                         ndp->ni_cnd.cn_flags |= GIANTHELD;
  297                 ndp->ni_startdir = dp;
  298                 error = lookup(ndp);
  299                 if (error) {
  300                         namei_cleanup_cnp(cnp);
  301                         SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
  302                         return (error);
  303                 }
  304                 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
  305                 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
  306                 /*
  307                  * If not a symbolic link, we're done.
  308                  */
  309                 if ((cnp->cn_flags & ISSYMLINK) == 0) {
  310                         if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
  311                                 namei_cleanup_cnp(cnp);
  312                         } else
  313                                 cnp->cn_flags |= HASBUF;
  314 
  315                         if ((cnp->cn_flags & MPSAFE) == 0) {
  316                                 VFS_UNLOCK_GIANT(vfslocked);
  317                         } else if (vfslocked)
  318                                 ndp->ni_cnd.cn_flags |= GIANTHELD;
  319                         SDT_PROBE2(vfs, namei, lookup, return, 0, ndp->ni_vp);
  320                         return (0);
  321                 }
  322                 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
  323                         error = ELOOP;
  324                         break;
  325                 }
  326 #ifdef MAC
  327                 if ((cnp->cn_flags & NOMACCHECK) == 0) {
  328                         error = mac_vnode_check_readlink(td->td_ucred,
  329                             ndp->ni_vp);
  330                         if (error)
  331                                 break;
  332                 }
  333 #endif
  334                 if (ndp->ni_pathlen > 1)
  335                         cp = uma_zalloc(namei_zone, M_WAITOK);
  336                 else
  337                         cp = cnp->cn_pnbuf;
  338                 aiov.iov_base = cp;
  339                 aiov.iov_len = MAXPATHLEN;
  340                 auio.uio_iov = &aiov;
  341                 auio.uio_iovcnt = 1;
  342                 auio.uio_offset = 0;
  343                 auio.uio_rw = UIO_READ;
  344                 auio.uio_segflg = UIO_SYSSPACE;
  345                 auio.uio_td = td;
  346                 auio.uio_resid = MAXPATHLEN;
  347                 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
  348                 if (error) {
  349                         if (ndp->ni_pathlen > 1)
  350                                 uma_zfree(namei_zone, cp);
  351                         break;
  352                 }
  353                 linklen = MAXPATHLEN - auio.uio_resid;
  354                 if (linklen == 0) {
  355                         if (ndp->ni_pathlen > 1)
  356                                 uma_zfree(namei_zone, cp);
  357                         error = ENOENT;
  358                         break;
  359                 }
  360                 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
  361                         if (ndp->ni_pathlen > 1)
  362                                 uma_zfree(namei_zone, cp);
  363                         error = ENAMETOOLONG;
  364                         break;
  365                 }
  366                 if (ndp->ni_pathlen > 1) {
  367                         bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
  368                         uma_zfree(namei_zone, cnp->cn_pnbuf);
  369                         cnp->cn_pnbuf = cp;
  370                 } else
  371                         cnp->cn_pnbuf[linklen] = '\0';
  372                 ndp->ni_pathlen += linklen;
  373                 vput(ndp->ni_vp);
  374                 dp = ndp->ni_dvp;
  375         }
  376         namei_cleanup_cnp(cnp);
  377         vput(ndp->ni_vp);
  378         ndp->ni_vp = NULL;
  379         vrele(ndp->ni_dvp);
  380         VFS_UNLOCK_GIANT(vfslocked);
  381         SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
  382         return (error);
  383 }
  384 
  385 static int
  386 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
  387 {
  388 
  389         if (mp == NULL || ((lkflags & LK_SHARED) &&
  390             (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
  391             ((cnflags & ISDOTDOT) &&
  392             (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
  393                 lkflags &= ~LK_SHARED;
  394                 lkflags |= LK_EXCLUSIVE;
  395         }
  396         return (lkflags);
  397 }
  398 
  399 static __inline int
  400 needs_exclusive_leaf(struct mount *mp, int flags)
  401 {
  402 
  403         /*
  404          * Intermediate nodes can use shared locks, we only need to
  405          * force an exclusive lock for leaf nodes.
  406          */
  407         if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
  408                 return (0);
  409 
  410         /* Always use exclusive locks if LOCKSHARED isn't set. */
  411         if (!(flags & LOCKSHARED))
  412                 return (1);
  413 
  414         /*
  415          * For lookups during open(), if the mount point supports
  416          * extended shared operations, then use a shared lock for the
  417          * leaf node, otherwise use an exclusive lock.
  418          */
  419         if (flags & ISOPEN) {
  420                 if (mp != NULL &&
  421                     (mp->mnt_kern_flag & MNTK_EXTENDED_SHARED))
  422                         return (0);
  423                 else
  424                         return (1);
  425         }
  426 
  427         /*
  428          * Lookup requests outside of open() that specify LOCKSHARED
  429          * only need a shared lock on the leaf vnode.
  430          */
  431         return (0);
  432 }
  433 
  434 /*
  435  * Search a pathname.
  436  * This is a very central and rather complicated routine.
  437  *
  438  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
  439  * The starting directory is taken from ni_startdir. The pathname is
  440  * descended until done, or a symbolic link is encountered. The variable
  441  * ni_more is clear if the path is completed; it is set to one if a
  442  * symbolic link needing interpretation is encountered.
  443  *
  444  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
  445  * whether the name is to be looked up, created, renamed, or deleted.
  446  * When CREATE, RENAME, or DELETE is specified, information usable in
  447  * creating, renaming, or deleting a directory entry may be calculated.
  448  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
  449  * locked. If flag has WANTPARENT or'ed into it, the parent directory is
  450  * returned unlocked. Otherwise the parent directory is not returned. If
  451  * the target of the pathname exists and LOCKLEAF is or'ed into the flag
  452  * the target is returned locked, otherwise it is returned unlocked.
  453  * When creating or renaming and LOCKPARENT is specified, the target may not
  454  * be ".".  When deleting and LOCKPARENT is specified, the target may be ".".
  455  *
  456  * Overall outline of lookup:
  457  *
  458  * dirloop:
  459  *      identify next component of name at ndp->ni_ptr
  460  *      handle degenerate case where name is null string
  461  *      if .. and crossing mount points and on mounted filesys, find parent
  462  *      call VOP_LOOKUP routine for next component name
  463  *          directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
  464  *          component vnode returned in ni_vp (if it exists), locked.
  465  *      if result vnode is mounted on and crossing mount points,
  466  *          find mounted on vnode
  467  *      if more components of name, do next level at dirloop
  468  *      return the answer in ni_vp, locked if LOCKLEAF set
  469  *          if LOCKPARENT set, return locked parent in ni_dvp
  470  *          if WANTPARENT set, return unlocked parent in ni_dvp
  471  */
  472 int
  473 lookup(struct nameidata *ndp)
  474 {
  475         char *cp;               /* pointer into pathname argument */
  476         struct vnode *dp = 0;   /* the directory we are searching */
  477         struct vnode *tdp;              /* saved dp */
  478         struct mount *mp;               /* mount table entry */
  479         struct prison *pr;
  480         int docache;                    /* == 0 do not cache last component */
  481         int wantparent;                 /* 1 => wantparent or lockparent flag */
  482         int rdonly;                     /* lookup read-only flag bit */
  483         int error = 0;
  484         int dpunlocked = 0;             /* dp has already been unlocked */
  485         struct componentname *cnp = &ndp->ni_cnd;
  486         int vfslocked;                  /* VFS Giant state for child */
  487         int dvfslocked;                 /* VFS Giant state for parent */
  488         int tvfslocked;
  489         int lkflags_save;
  490         int ni_dvp_unlocked;
  491         
  492         /*
  493          * Setup: break out flag bits into variables.
  494          */
  495         dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
  496         vfslocked = 0;
  497         ni_dvp_unlocked = 0;
  498         ndp->ni_cnd.cn_flags &= ~GIANTHELD;
  499         wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
  500         KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
  501             ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
  502         docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
  503         if (cnp->cn_nameiop == DELETE ||
  504             (wantparent && cnp->cn_nameiop != CREATE &&
  505              cnp->cn_nameiop != LOOKUP))
  506                 docache = 0;
  507         rdonly = cnp->cn_flags & RDONLY;
  508         cnp->cn_flags &= ~ISSYMLINK;
  509         ndp->ni_dvp = NULL;
  510         /*
  511          * We use shared locks until we hit the parent of the last cn then
  512          * we adjust based on the requesting flags.
  513          */
  514         if (lookup_shared)
  515                 cnp->cn_lkflags = LK_SHARED;
  516         else
  517                 cnp->cn_lkflags = LK_EXCLUSIVE;
  518         dp = ndp->ni_startdir;
  519         ndp->ni_startdir = NULLVP;
  520         vn_lock(dp,
  521             compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
  522             cnp->cn_flags));
  523 
  524 dirloop:
  525         /*
  526          * Search a new directory.
  527          *
  528          * The last component of the filename is left accessible via
  529          * cnp->cn_nameptr for callers that need the name. Callers needing
  530          * the name set the SAVENAME flag. When done, they assume
  531          * responsibility for freeing the pathname buffer.
  532          */
  533         cnp->cn_consume = 0;
  534         for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
  535                 continue;
  536         cnp->cn_namelen = cp - cnp->cn_nameptr;
  537         if (cnp->cn_namelen > NAME_MAX) {
  538                 error = ENAMETOOLONG;
  539                 goto bad;
  540         }
  541 #ifdef NAMEI_DIAGNOSTIC
  542         { char c = *cp;
  543         *cp = '\0';
  544         printf("{%s}: ", cnp->cn_nameptr);
  545         *cp = c; }
  546 #endif
  547         ndp->ni_pathlen -= cnp->cn_namelen;
  548         ndp->ni_next = cp;
  549 
  550         /*
  551          * Replace multiple slashes by a single slash and trailing slashes
  552          * by a null.  This must be done before VOP_LOOKUP() because some
  553          * fs's don't know about trailing slashes.  Remember if there were
  554          * trailing slashes to handle symlinks, existing non-directories
  555          * and non-existing files that won't be directories specially later.
  556          */
  557         while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
  558                 cp++;
  559                 ndp->ni_pathlen--;
  560                 if (*cp == '\0') {
  561                         *ndp->ni_next = '\0';
  562                         cnp->cn_flags |= TRAILINGSLASH;
  563                 }
  564         }
  565         ndp->ni_next = cp;
  566 
  567         cnp->cn_flags |= MAKEENTRY;
  568         if (*cp == '\0' && docache == 0)
  569                 cnp->cn_flags &= ~MAKEENTRY;
  570         if (cnp->cn_namelen == 2 &&
  571             cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
  572                 cnp->cn_flags |= ISDOTDOT;
  573         else
  574                 cnp->cn_flags &= ~ISDOTDOT;
  575         if (*ndp->ni_next == 0)
  576                 cnp->cn_flags |= ISLASTCN;
  577         else
  578                 cnp->cn_flags &= ~ISLASTCN;
  579 
  580         if ((cnp->cn_flags & ISLASTCN) != 0 &&
  581             cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
  582             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
  583                 error = EINVAL;
  584                 goto bad;
  585         }
  586 
  587         /*
  588          * Check for degenerate name (e.g. / or "")
  589          * which is a way of talking about a directory,
  590          * e.g. like "/." or ".".
  591          */
  592         if (cnp->cn_nameptr[0] == '\0') {
  593                 if (dp->v_type != VDIR) {
  594                         error = ENOTDIR;
  595                         goto bad;
  596                 }
  597                 if (cnp->cn_nameiop != LOOKUP) {
  598                         error = EISDIR;
  599                         goto bad;
  600                 }
  601                 if (wantparent) {
  602                         ndp->ni_dvp = dp;
  603                         VREF(dp);
  604                 }
  605                 ndp->ni_vp = dp;
  606 
  607                 if (cnp->cn_flags & AUDITVNODE1)
  608                         AUDIT_ARG_VNODE1(dp);
  609                 else if (cnp->cn_flags & AUDITVNODE2)
  610                         AUDIT_ARG_VNODE2(dp);
  611 
  612                 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
  613                         VOP_UNLOCK(dp, 0);
  614                 /* XXX This should probably move to the top of function. */
  615                 if (cnp->cn_flags & SAVESTART)
  616                         panic("lookup: SAVESTART");
  617                 goto success;
  618         }
  619 
  620         /*
  621          * Handle "..": five special cases.
  622          * 0. If doing a capability lookup, return ENOTCAPABLE (this is a
  623          *    fairly conservative design choice, but it's the only one that we
  624          *    are satisfied guarantees the property we're looking for).
  625          * 1. Return an error if this is the last component of
  626          *    the name and the operation is DELETE or RENAME.
  627          * 2. If at root directory (e.g. after chroot)
  628          *    or at absolute root directory
  629          *    then ignore it so can't get out.
  630          * 3. If this vnode is the root of a mounted
  631          *    filesystem, then replace it with the
  632          *    vnode which was mounted on so we take the
  633          *    .. in the other filesystem.
  634          * 4. If the vnode is the top directory of
  635          *    the jail or chroot, don't let them out.
  636          */
  637         if (cnp->cn_flags & ISDOTDOT) {
  638                 if (ndp->ni_strictrelative != 0) {
  639                         error = ENOTCAPABLE;
  640                         goto bad;
  641                 }
  642                 if ((cnp->cn_flags & ISLASTCN) != 0 &&
  643                     (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
  644                         error = EINVAL;
  645                         goto bad;
  646                 }
  647                 for (;;) {
  648                         for (pr = cnp->cn_cred->cr_prison; pr != NULL;
  649                              pr = pr->pr_parent)
  650                                 if (dp == pr->pr_root)
  651                                         break;
  652                         if (dp == ndp->ni_rootdir || 
  653                             dp == ndp->ni_topdir || 
  654                             dp == rootvnode ||
  655                             pr != NULL ||
  656                             ((dp->v_vflag & VV_ROOT) != 0 &&
  657                              (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
  658                                 ndp->ni_dvp = dp;
  659                                 ndp->ni_vp = dp;
  660                                 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
  661                                 VREF(dp);
  662                                 goto nextname;
  663                         }
  664                         if ((dp->v_vflag & VV_ROOT) == 0)
  665                                 break;
  666                         if (dp->v_iflag & VI_DOOMED) {  /* forced unmount */
  667                                 error = ENOENT;
  668                                 goto bad;
  669                         }
  670                         tdp = dp;
  671                         dp = dp->v_mount->mnt_vnodecovered;
  672                         tvfslocked = dvfslocked;
  673                         dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
  674                         VREF(dp);
  675                         vput(tdp);
  676                         VFS_UNLOCK_GIANT(tvfslocked);
  677                         vn_lock(dp,
  678                             compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
  679                             LK_RETRY, ISDOTDOT));
  680                 }
  681         }
  682 
  683         /*
  684          * We now have a segment name to search for, and a directory to search.
  685          */
  686 unionlookup:
  687 #ifdef MAC
  688         if ((cnp->cn_flags & NOMACCHECK) == 0) {
  689                 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
  690                     cnp);
  691                 if (error)
  692                         goto bad;
  693         }
  694 #endif
  695         ndp->ni_dvp = dp;
  696         ndp->ni_vp = NULL;
  697         ASSERT_VOP_LOCKED(dp, "lookup");
  698         VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
  699         /*
  700          * If we have a shared lock we may need to upgrade the lock for the
  701          * last operation.
  702          */
  703         if (dp != vp_crossmp &&
  704             VOP_ISLOCKED(dp) == LK_SHARED &&
  705             (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
  706                 vn_lock(dp, LK_UPGRADE|LK_RETRY);
  707         if ((dp->v_iflag & VI_DOOMED) != 0) {
  708                 error = ENOENT;
  709                 goto bad;
  710         }
  711         /*
  712          * If we're looking up the last component and we need an exclusive
  713          * lock, adjust our lkflags.
  714          */
  715         if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
  716                 cnp->cn_lkflags = LK_EXCLUSIVE;
  717 #ifdef NAMEI_DIAGNOSTIC
  718         vprint("lookup in", dp);
  719 #endif
  720         lkflags_save = cnp->cn_lkflags;
  721         cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
  722             cnp->cn_flags);
  723         if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
  724                 cnp->cn_lkflags = lkflags_save;
  725                 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
  726 #ifdef NAMEI_DIAGNOSTIC
  727                 printf("not found\n");
  728 #endif
  729                 if ((error == ENOENT) &&
  730                     (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
  731                     (dp->v_mount->mnt_flag & MNT_UNION)) {
  732                         tdp = dp;
  733                         dp = dp->v_mount->mnt_vnodecovered;
  734                         tvfslocked = dvfslocked;
  735                         dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
  736                         VREF(dp);
  737                         vput(tdp);
  738                         VFS_UNLOCK_GIANT(tvfslocked);
  739                         vn_lock(dp,
  740                             compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
  741                             LK_RETRY, cnp->cn_flags));
  742                         goto unionlookup;
  743                 }
  744 
  745                 if (error != EJUSTRETURN)
  746                         goto bad;
  747                 /*
  748                  * At this point, we know we're at the end of the
  749                  * pathname.  If creating / renaming, we can consider
  750                  * allowing the file or directory to be created / renamed,
  751                  * provided we're not on a read-only filesystem.
  752                  */
  753                 if (rdonly) {
  754                         error = EROFS;
  755                         goto bad;
  756                 }
  757                 /* trailing slash only allowed for directories */
  758                 if ((cnp->cn_flags & TRAILINGSLASH) &&
  759                     !(cnp->cn_flags & WILLBEDIR)) {
  760                         error = ENOENT;
  761                         goto bad;
  762                 }
  763                 if ((cnp->cn_flags & LOCKPARENT) == 0)
  764                         VOP_UNLOCK(dp, 0);
  765                 /*
  766                  * We return with ni_vp NULL to indicate that the entry
  767                  * doesn't currently exist, leaving a pointer to the
  768                  * (possibly locked) directory vnode in ndp->ni_dvp.
  769                  */
  770                 if (cnp->cn_flags & SAVESTART) {
  771                         ndp->ni_startdir = ndp->ni_dvp;
  772                         VREF(ndp->ni_startdir);
  773                 }
  774                 goto success;
  775         } else
  776                 cnp->cn_lkflags = lkflags_save;
  777 #ifdef NAMEI_DIAGNOSTIC
  778         printf("found\n");
  779 #endif
  780         /*
  781          * Take into account any additional components consumed by
  782          * the underlying filesystem.
  783          */
  784         if (cnp->cn_consume > 0) {
  785                 cnp->cn_nameptr += cnp->cn_consume;
  786                 ndp->ni_next += cnp->cn_consume;
  787                 ndp->ni_pathlen -= cnp->cn_consume;
  788                 cnp->cn_consume = 0;
  789         }
  790 
  791         dp = ndp->ni_vp;
  792         vfslocked = VFS_LOCK_GIANT(dp->v_mount);
  793 
  794         /*
  795          * Check to see if the vnode has been mounted on;
  796          * if so find the root of the mounted filesystem.
  797          */
  798         while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
  799                (cnp->cn_flags & NOCROSSMOUNT) == 0) {
  800                 if (vfs_busy(mp, 0))
  801                         continue;
  802                 vput(dp);
  803                 VFS_UNLOCK_GIANT(vfslocked);
  804                 vfslocked = VFS_LOCK_GIANT(mp);
  805                 if (dp != ndp->ni_dvp)
  806                         vput(ndp->ni_dvp);
  807                 else
  808                         vrele(ndp->ni_dvp);
  809                 VFS_UNLOCK_GIANT(dvfslocked);
  810                 dvfslocked = 0;
  811                 vref(vp_crossmp);
  812                 ndp->ni_dvp = vp_crossmp;
  813                 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
  814                     cnp->cn_flags), &tdp);
  815                 vfs_unbusy(mp);
  816                 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
  817                         panic("vp_crossmp exclusively locked or reclaimed");
  818                 if (error) {
  819                         dpunlocked = 1;
  820                         goto bad2;
  821                 }
  822                 ndp->ni_vp = dp = tdp;
  823         }
  824 
  825         /*
  826          * Check for symbolic link
  827          */
  828         if ((dp->v_type == VLNK) &&
  829             ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
  830              *ndp->ni_next == '/')) {
  831                 cnp->cn_flags |= ISSYMLINK;
  832                 if (dp->v_iflag & VI_DOOMED) {
  833                         /*
  834                          * We can't know whether the directory was mounted with
  835                          * NOSYMFOLLOW, so we can't follow safely.
  836                          */
  837                         error = ENOENT;
  838                         goto bad2;
  839                 }
  840                 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
  841                         error = EACCES;
  842                         goto bad2;
  843                 }
  844                 /*
  845                  * Symlink code always expects an unlocked dvp.
  846                  */
  847                 if (ndp->ni_dvp != ndp->ni_vp) {
  848                         VOP_UNLOCK(ndp->ni_dvp, 0);
  849                         ni_dvp_unlocked = 1;
  850                 }
  851                 goto success;
  852         }
  853 
  854 nextname:
  855         /*
  856          * Not a symbolic link that we will follow.  Continue with the
  857          * next component if there is any; otherwise, we're done.
  858          */
  859         KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
  860             ("lookup: invalid path state."));
  861         if (*ndp->ni_next == '/') {
  862                 cnp->cn_nameptr = ndp->ni_next;
  863                 while (*cnp->cn_nameptr == '/') {
  864                         cnp->cn_nameptr++;
  865                         ndp->ni_pathlen--;
  866                 }
  867                 if (ndp->ni_dvp != dp)
  868                         vput(ndp->ni_dvp);
  869                 else
  870                         vrele(ndp->ni_dvp);
  871                 VFS_UNLOCK_GIANT(dvfslocked);
  872                 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
  873                 vfslocked = 0;
  874                 goto dirloop;
  875         }
  876         /*
  877          * If we're processing a path with a trailing slash,
  878          * check that the end result is a directory.
  879          */
  880         if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
  881                 error = ENOTDIR;
  882                 goto bad2;
  883         }
  884         /*
  885          * Disallow directory write attempts on read-only filesystems.
  886          */
  887         if (rdonly &&
  888             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
  889                 error = EROFS;
  890                 goto bad2;
  891         }
  892         if (cnp->cn_flags & SAVESTART) {
  893                 ndp->ni_startdir = ndp->ni_dvp;
  894                 VREF(ndp->ni_startdir);
  895         }
  896         if (!wantparent) {
  897                 ni_dvp_unlocked = 2;
  898                 if (ndp->ni_dvp != dp)
  899                         vput(ndp->ni_dvp);
  900                 else
  901                         vrele(ndp->ni_dvp);
  902                 VFS_UNLOCK_GIANT(dvfslocked);
  903                 dvfslocked = 0;
  904         } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
  905                 VOP_UNLOCK(ndp->ni_dvp, 0);
  906                 ni_dvp_unlocked = 1;
  907         }
  908 
  909         if (cnp->cn_flags & AUDITVNODE1)
  910                 AUDIT_ARG_VNODE1(dp);
  911         else if (cnp->cn_flags & AUDITVNODE2)
  912                 AUDIT_ARG_VNODE2(dp);
  913 
  914         if ((cnp->cn_flags & LOCKLEAF) == 0)
  915                 VOP_UNLOCK(dp, 0);
  916 success:
  917         /*
  918          * Because of lookup_shared we may have the vnode shared locked, but
  919          * the caller may want it to be exclusively locked.
  920          */
  921         if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
  922             VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
  923                 vn_lock(dp, LK_UPGRADE | LK_RETRY);
  924                 if (dp->v_iflag & VI_DOOMED) {
  925                         error = ENOENT;
  926                         goto bad2;
  927                 }
  928         }
  929         if (vfslocked && dvfslocked)
  930                 VFS_UNLOCK_GIANT(dvfslocked);   /* Only need one */
  931         if (vfslocked || dvfslocked)
  932                 ndp->ni_cnd.cn_flags |= GIANTHELD;
  933         return (0);
  934 
  935 bad2:
  936         if (ni_dvp_unlocked != 2) {
  937                 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
  938                         vput(ndp->ni_dvp);
  939                 else
  940                         vrele(ndp->ni_dvp);
  941         }
  942 bad:
  943         if (!dpunlocked)
  944                 vput(dp);
  945         VFS_UNLOCK_GIANT(vfslocked);
  946         VFS_UNLOCK_GIANT(dvfslocked);
  947         ndp->ni_cnd.cn_flags &= ~GIANTHELD;
  948         ndp->ni_vp = NULL;
  949         return (error);
  950 }
  951 
  952 /*
  953  * relookup - lookup a path name component
  954  *    Used by lookup to re-acquire things.
  955  */
  956 int
  957 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
  958 {
  959         struct vnode *dp = 0;           /* the directory we are searching */
  960         int wantparent;                 /* 1 => wantparent or lockparent flag */
  961         int rdonly;                     /* lookup read-only flag bit */
  962         int error = 0;
  963 
  964         KASSERT(cnp->cn_flags & ISLASTCN,
  965             ("relookup: Not given last component."));
  966         /*
  967          * Setup: break out flag bits into variables.
  968          */
  969         wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
  970         KASSERT(wantparent, ("relookup: parent not wanted."));
  971         rdonly = cnp->cn_flags & RDONLY;
  972         cnp->cn_flags &= ~ISSYMLINK;
  973         dp = dvp;
  974         cnp->cn_lkflags = LK_EXCLUSIVE;
  975         vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
  976 
  977         /*
  978          * Search a new directory.
  979          *
  980          * The last component of the filename is left accessible via
  981          * cnp->cn_nameptr for callers that need the name. Callers needing
  982          * the name set the SAVENAME flag. When done, they assume
  983          * responsibility for freeing the pathname buffer.
  984          */
  985 #ifdef NAMEI_DIAGNOSTIC
  986         printf("{%s}: ", cnp->cn_nameptr);
  987 #endif
  988 
  989         /*
  990          * Check for "" which represents the root directory after slash
  991          * removal.
  992          */
  993         if (cnp->cn_nameptr[0] == '\0') {
  994                 /*
  995                  * Support only LOOKUP for "/" because lookup()
  996                  * can't succeed for CREATE, DELETE and RENAME.
  997                  */
  998                 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
  999                 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
 1000 
 1001                 if (!(cnp->cn_flags & LOCKLEAF))
 1002                         VOP_UNLOCK(dp, 0);
 1003                 *vpp = dp;
 1004                 /* XXX This should probably move to the top of function. */
 1005                 if (cnp->cn_flags & SAVESTART)
 1006                         panic("lookup: SAVESTART");
 1007                 return (0);
 1008         }
 1009 
 1010         if (cnp->cn_flags & ISDOTDOT)
 1011                 panic ("relookup: lookup on dot-dot");
 1012 
 1013         /*
 1014          * We now have a segment name to search for, and a directory to search.
 1015          */
 1016 #ifdef NAMEI_DIAGNOSTIC
 1017         vprint("search in:", dp);
 1018 #endif
 1019         if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
 1020                 KASSERT(*vpp == NULL, ("leaf should be empty"));
 1021                 if (error != EJUSTRETURN)
 1022                         goto bad;
 1023                 /*
 1024                  * If creating and at end of pathname, then can consider
 1025                  * allowing file to be created.
 1026                  */
 1027                 if (rdonly) {
 1028                         error = EROFS;
 1029                         goto bad;
 1030                 }
 1031                 /* ASSERT(dvp == ndp->ni_startdir) */
 1032                 if (cnp->cn_flags & SAVESTART)
 1033                         VREF(dvp);
 1034                 if ((cnp->cn_flags & LOCKPARENT) == 0)
 1035                         VOP_UNLOCK(dp, 0);
 1036                 /*
 1037                  * We return with ni_vp NULL to indicate that the entry
 1038                  * doesn't currently exist, leaving a pointer to the
 1039                  * (possibly locked) directory vnode in ndp->ni_dvp.
 1040                  */
 1041                 return (0);
 1042         }
 1043 
 1044         dp = *vpp;
 1045 
 1046         /*
 1047          * Disallow directory write attempts on read-only filesystems.
 1048          */
 1049         if (rdonly &&
 1050             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
 1051                 if (dvp == dp)
 1052                         vrele(dvp);
 1053                 else
 1054                         vput(dvp);
 1055                 error = EROFS;
 1056                 goto bad;
 1057         }
 1058         /*
 1059          * Set the parent lock/ref state to the requested state.
 1060          */
 1061         if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
 1062                 if (wantparent)
 1063                         VOP_UNLOCK(dvp, 0);
 1064                 else
 1065                         vput(dvp);
 1066         } else if (!wantparent)
 1067                 vrele(dvp);
 1068         /*
 1069          * Check for symbolic link
 1070          */
 1071         KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
 1072             ("relookup: symlink found.\n"));
 1073 
 1074         /* ASSERT(dvp == ndp->ni_startdir) */
 1075         if (cnp->cn_flags & SAVESTART)
 1076                 VREF(dvp);
 1077         
 1078         if ((cnp->cn_flags & LOCKLEAF) == 0)
 1079                 VOP_UNLOCK(dp, 0);
 1080         return (0);
 1081 bad:
 1082         vput(dp);
 1083         *vpp = NULL;
 1084         return (error);
 1085 }
 1086 
 1087 /*
 1088  * Free data allocated by namei(); see namei(9) for details.
 1089  */
 1090 void
 1091 NDFREE(struct nameidata *ndp, const u_int flags)
 1092 {
 1093         int unlock_dvp;
 1094         int unlock_vp;
 1095 
 1096         unlock_dvp = 0;
 1097         unlock_vp = 0;
 1098 
 1099         if (!(flags & NDF_NO_FREE_PNBUF) &&
 1100             (ndp->ni_cnd.cn_flags & HASBUF)) {
 1101                 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
 1102                 ndp->ni_cnd.cn_flags &= ~HASBUF;
 1103         }
 1104         if (!(flags & NDF_NO_VP_UNLOCK) &&
 1105             (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
 1106                 unlock_vp = 1;
 1107         if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
 1108                 if (unlock_vp) {
 1109                         vput(ndp->ni_vp);
 1110                         unlock_vp = 0;
 1111                 } else
 1112                         vrele(ndp->ni_vp);
 1113                 ndp->ni_vp = NULL;
 1114         }
 1115         if (unlock_vp)
 1116                 VOP_UNLOCK(ndp->ni_vp, 0);
 1117         if (!(flags & NDF_NO_DVP_UNLOCK) &&
 1118             (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
 1119             ndp->ni_dvp != ndp->ni_vp)
 1120                 unlock_dvp = 1;
 1121         if (!(flags & NDF_NO_DVP_RELE) &&
 1122             (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
 1123                 if (unlock_dvp) {
 1124                         vput(ndp->ni_dvp);
 1125                         unlock_dvp = 0;
 1126                 } else
 1127                         vrele(ndp->ni_dvp);
 1128                 ndp->ni_dvp = NULL;
 1129         }
 1130         if (unlock_dvp)
 1131                 VOP_UNLOCK(ndp->ni_dvp, 0);
 1132         if (!(flags & NDF_NO_STARTDIR_RELE) &&
 1133             (ndp->ni_cnd.cn_flags & SAVESTART)) {
 1134                 vrele(ndp->ni_startdir);
 1135                 ndp->ni_startdir = NULL;
 1136         }
 1137 }
 1138 
 1139 /*
 1140  * Determine if there is a suitable alternate filename under the specified
 1141  * prefix for the specified path.  If the create flag is set, then the
 1142  * alternate prefix will be used so long as the parent directory exists.
 1143  * This is used by the various compatiblity ABIs so that Linux binaries prefer
 1144  * files under /compat/linux for example.  The chosen path (whether under
 1145  * the prefix or under /) is returned in a kernel malloc'd buffer pointed
 1146  * to by pathbuf.  The caller is responsible for free'ing the buffer from
 1147  * the M_TEMP bucket if one is returned.
 1148  */
 1149 int
 1150 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
 1151     enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
 1152 {
 1153         struct nameidata nd, ndroot;
 1154         char *ptr, *buf, *cp;
 1155         size_t len, sz;
 1156         int error;
 1157 
 1158         buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 1159         *pathbuf = buf;
 1160 
 1161         /* Copy the prefix into the new pathname as a starting point. */
 1162         len = strlcpy(buf, prefix, MAXPATHLEN);
 1163         if (len >= MAXPATHLEN) {
 1164                 *pathbuf = NULL;
 1165                 free(buf, M_TEMP);
 1166                 return (EINVAL);
 1167         }
 1168         sz = MAXPATHLEN - len;
 1169         ptr = buf + len;
 1170 
 1171         /* Append the filename to the prefix. */
 1172         if (pathseg == UIO_SYSSPACE)
 1173                 error = copystr(path, ptr, sz, &len);
 1174         else
 1175                 error = copyinstr(path, ptr, sz, &len);
 1176 
 1177         if (error) {
 1178                 *pathbuf = NULL;
 1179                 free(buf, M_TEMP);
 1180                 return (error);
 1181         }
 1182 
 1183         /* Only use a prefix with absolute pathnames. */
 1184         if (*ptr != '/') {
 1185                 error = EINVAL;
 1186                 goto keeporig;
 1187         }
 1188 
 1189         if (dirfd != AT_FDCWD) {
 1190                 /*
 1191                  * We want the original because the "prefix" is
 1192                  * included in the already opened dirfd.
 1193                  */
 1194                 bcopy(ptr, buf, len);
 1195                 return (0);
 1196         }
 1197 
 1198         /*
 1199          * We know that there is a / somewhere in this pathname.
 1200          * Search backwards for it, to find the file's parent dir
 1201          * to see if it exists in the alternate tree. If it does,
 1202          * and we want to create a file (cflag is set). We don't
 1203          * need to worry about the root comparison in this case.
 1204          */
 1205 
 1206         if (create) {
 1207                 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
 1208                 *cp = '\0';
 1209 
 1210                 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
 1211                 error = namei(&nd);
 1212                 *cp = '/';
 1213                 if (error != 0)
 1214                         goto keeporig;
 1215         } else {
 1216                 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
 1217 
 1218                 error = namei(&nd);
 1219                 if (error != 0)
 1220                         goto keeporig;
 1221 
 1222                 /*
 1223                  * We now compare the vnode of the prefix to the one
 1224                  * vnode asked. If they resolve to be the same, then we
 1225                  * ignore the match so that the real root gets used.
 1226                  * This avoids the problem of traversing "../.." to find the
 1227                  * root directory and never finding it, because "/" resolves
 1228                  * to the emulation root directory. This is expensive :-(
 1229                  */
 1230                 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
 1231                     td);
 1232 
 1233                 /* We shouldn't ever get an error from this namei(). */
 1234                 error = namei(&ndroot);
 1235                 if (error == 0) {
 1236                         if (nd.ni_vp == ndroot.ni_vp)
 1237                                 error = ENOENT;
 1238 
 1239                         NDFREE(&ndroot, NDF_ONLY_PNBUF);
 1240                         vrele(ndroot.ni_vp);
 1241                         VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
 1242                 }
 1243         }
 1244 
 1245         NDFREE(&nd, NDF_ONLY_PNBUF);
 1246         vrele(nd.ni_vp);
 1247         VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
 1248 
 1249 keeporig:
 1250         /* If there was an error, use the original path name. */
 1251         if (error)
 1252                 bcopy(ptr, buf, len);
 1253         return (error);
 1254 }

Cache object: 2c88062aa6aab8100198f24458a521ee


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.