| 1 | /* |
| 2 | * Copyright (c) 2000-2015 Apple Inc. All rights reserved. |
| 3 | * |
| 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
| 5 | * |
| 6 | * This file contains Original Code and/or Modifications of Original Code |
| 7 | * as defined in and that are subject to the Apple Public Source License |
| 8 | * Version 2.0 (the 'License'). You may not use this file except in |
| 9 | * compliance with the License. The rights granted to you under the License |
| 10 | * may not be used to create, or enable the creation or redistribution of, |
| 11 | * unlawful or unlicensed copies of an Apple operating system, or to |
| 12 | * circumvent, violate, or enable the circumvention or violation of, any |
| 13 | * terms of an Apple operating system software license agreement. |
| 14 | * |
| 15 | * Please obtain a copy of the License at |
| 16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. |
| 17 | * |
| 18 | * The Original Code and all software distributed under the License are |
| 19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
| 20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
| 21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
| 22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
| 23 | * Please see the License for the specific language governing rights and |
| 24 | * limitations under the License. |
| 25 | * |
| 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
| 27 | */ |
| 28 | /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */ |
| 29 | /* |
| 30 | * Copyright (c) 1982, 1986, 1989, 1993 |
| 31 | * The Regents of the University of California. All rights reserved. |
| 32 | * (c) UNIX System Laboratories, Inc. |
| 33 | * All or some portions of this file are derived from material licensed |
| 34 | * to the University of California by American Telephone and Telegraph |
| 35 | * Co. or Unix System Laboratories, Inc. and are reproduced herein with |
| 36 | * the permission of UNIX System Laboratories, Inc. |
| 37 | * |
| 38 | * Redistribution and use in source and binary forms, with or without |
| 39 | * modification, are permitted provided that the following conditions |
| 40 | * are met: |
| 41 | * 1. Redistributions of source code must retain the above copyright |
| 42 | * notice, this list of conditions and the following disclaimer. |
| 43 | * 2. Redistributions in binary form must reproduce the above copyright |
| 44 | * notice, this list of conditions and the following disclaimer in the |
| 45 | * documentation and/or other materials provided with the distribution. |
| 46 | * 3. All advertising materials mentioning features or use of this software |
| 47 | * must display the following acknowledgement: |
| 48 | * This product includes software developed by the University of |
| 49 | * California, Berkeley and its contributors. |
| 50 | * 4. Neither the name of the University nor the names of its contributors |
| 51 | * may be used to endorse or promote products derived from this software |
| 52 | * without specific prior written permission. |
| 53 | * |
| 54 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 55 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 56 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 57 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 58 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 59 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 60 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 61 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 62 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 63 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 64 | * SUCH DAMAGE. |
| 65 | * |
| 66 | * @(#)vfs_lookup.c 8.10 (Berkeley) 5/27/95 |
| 67 | */ |
| 68 | /* |
| 69 | * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce |
| 70 | * support for mandatory and extensible security protections. This notice |
| 71 | * is included in support of clause 2.2 (b) of the Apple Public License, |
| 72 | * Version 2.0. |
| 73 | */ |
| 74 | |
| 75 | #include <sys/param.h> |
| 76 | #include <sys/systm.h> |
| 77 | #include <sys/syslimits.h> |
| 78 | #include <sys/time.h> |
| 79 | #include <sys/namei.h> |
| 80 | #include <sys/vm.h> |
| 81 | #include <sys/vnode_internal.h> |
| 82 | #include <sys/mount_internal.h> |
| 83 | #include <sys/errno.h> |
| 84 | #include <sys/malloc.h> |
| 85 | #include <sys/filedesc.h> |
| 86 | #include <sys/proc_internal.h> |
| 87 | #include <sys/kdebug.h> |
| 88 | #include <sys/unistd.h> /* For _PC_NAME_MAX */ |
| 89 | #include <sys/uio_internal.h> |
| 90 | #include <sys/kauth.h> |
| 91 | #include <kern/kalloc.h> |
| 92 | #include <security/audit/audit.h> |
| 93 | #include <sys/dtrace.h> /* to get the prototype for strstr() in sys/dtrace_glue.h */ |
| 94 | #if CONFIG_MACF |
| 95 | #include <security/mac_framework.h> |
| 96 | #endif |
| 97 | |
| 98 | #if NAMEDRSRCFORK |
| 99 | #include <sys/xattr.h> |
| 100 | #endif |
| 101 | /* |
| 102 | * The minimum volfs-style pathname is 9. |
| 103 | * Example: "/.vol/1/2" |
| 104 | */ |
| 105 | #define VOLFS_MIN_PATH_LEN 9 |
| 106 | |
| 107 | |
| 108 | #if CONFIG_VOLFS |
| 109 | static int vfs_getrealpath(const char * path, char * realpath, size_t bufsize, vfs_context_t ctx); |
| 110 | #define MAX_VOLFS_RESTARTS 5 |
| 111 | #endif |
| 112 | |
| 113 | static int lookup_traverse_mountpoints(struct nameidata *ndp, struct componentname *cnp, vnode_t dp, int vbusyflags, vfs_context_t ctx); |
| 114 | static int lookup_handle_symlink(struct nameidata *ndp, vnode_t *new_dp, vfs_context_t ctx); |
| 115 | static int lookup_authorize_search(vnode_t dp, struct componentname *cnp, int dp_authorized_in_cache, vfs_context_t ctx); |
| 116 | static void lookup_consider_update_cache(vnode_t dvp, vnode_t vp, struct componentname *cnp, int nc_generation); |
| 117 | static int lookup_handle_found_vnode(struct nameidata *ndp, struct componentname *cnp, int rdonly, |
| 118 | int vbusyflags, int *keep_going, int nc_generation, |
| 119 | int wantparent, int atroot, vfs_context_t ctx); |
| 120 | static int lookup_handle_emptyname(struct nameidata *ndp, struct componentname *cnp, int wantparent); |
| 121 | |
| 122 | #if NAMEDRSRCFORK |
| 123 | static int lookup_handle_rsrc_fork(vnode_t dp, struct nameidata *ndp, struct componentname *cnp, int wantparent, vfs_context_t ctx); |
| 124 | #endif |
| 125 | |
| 126 | |
| 127 | |
| 128 | /* |
| 129 | * Convert a pathname into a pointer to a locked inode. |
| 130 | * |
| 131 | * The FOLLOW flag is set when symbolic links are to be followed |
| 132 | * when they occur at the end of the name translation process. |
| 133 | * Symbolic links are always followed for all other pathname |
| 134 | * components other than the last. |
| 135 | * |
| 136 | * The segflg defines whether the name is to be copied from user |
| 137 | * space or kernel space. |
| 138 | * |
| 139 | * Overall outline of namei: |
| 140 | * |
| 141 | * copy in name |
| 142 | * get starting directory |
| 143 | * while (!done && !error) { |
| 144 | * call lookup to search path. |
| 145 | * if symbolic link, massage name in buffer and continue |
| 146 | * } |
| 147 | * |
| 148 | * Returns: 0 Success |
| 149 | * ENOENT No such file or directory |
| 150 | * ELOOP Too many levels of symbolic links |
| 151 | * ENAMETOOLONG Filename too long |
| 152 | * copyinstr:EFAULT Bad address |
| 153 | * copyinstr:ENAMETOOLONG Filename too long |
| 154 | * lookup:EBADF Bad file descriptor |
| 155 | * lookup:EROFS |
| 156 | * lookup:EACCES |
| 157 | * lookup:EPERM |
| 158 | * lookup:ERECYCLE vnode was recycled from underneath us in lookup. |
| 159 | * This means we should re-drive lookup from this point. |
| 160 | * lookup: ??? |
| 161 | * VNOP_READLINK:??? |
| 162 | */ |
| 163 | int |
| 164 | namei(struct nameidata *ndp) |
| 165 | { |
| 166 | struct filedesc *fdp; /* pointer to file descriptor state */ |
| 167 | struct vnode *dp; /* the directory we are searching */ |
| 168 | struct vnode *usedvp = ndp->ni_dvp; /* store pointer to vp in case we must loop due to |
| 169 | heavy vnode pressure */ |
| 170 | u_long cnpflags = ndp->ni_cnd.cn_flags; /* store in case we have to restore after loop */ |
| 171 | int error; |
| 172 | struct componentname *cnp = &ndp->ni_cnd; |
| 173 | vfs_context_t ctx = cnp->cn_context; |
| 174 | proc_t p = vfs_context_proc(ctx); |
| 175 | #if CONFIG_AUDIT |
| 176 | /* XXX ut should be from context */ |
| 177 | uthread_t ut = (struct uthread *)get_bsdthread_info(current_thread()); |
| 178 | #endif |
| 179 | |
| 180 | #if CONFIG_VOLFS |
| 181 | int volfs_restarts = 0; |
| 182 | #endif |
| 183 | size_t bytes_copied = 0; |
| 184 | |
| 185 | fdp = p->p_fd; |
| 186 | |
| 187 | #if DIAGNOSTIC |
| 188 | if (!vfs_context_ucred(ctx) || !p) |
| 189 | panic ("namei: bad cred/proc" ); |
| 190 | if (cnp->cn_nameiop & (~OPMASK)) |
| 191 | panic ("namei: nameiop contaminated with flags" ); |
| 192 | if (cnp->cn_flags & OPMASK) |
| 193 | panic ("namei: flags contaminated with nameiops" ); |
| 194 | #endif |
| 195 | |
| 196 | /* |
| 197 | * A compound VNOP found something that needs further processing: |
| 198 | * either a trigger vnode, a covered directory, or a symlink. |
| 199 | */ |
| 200 | if (ndp->ni_flag & NAMEI_CONTLOOKUP) { |
| 201 | int rdonly, vbusyflags, keep_going, wantparent; |
| 202 | |
| 203 | rdonly = cnp->cn_flags & RDONLY; |
| 204 | vbusyflags = ((cnp->cn_flags & CN_NBMOUNTLOOK) != 0) ? LK_NOWAIT : 0; |
| 205 | keep_going = 0; |
| 206 | wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT); |
| 207 | |
| 208 | ndp->ni_flag &= ~(NAMEI_CONTLOOKUP); |
| 209 | |
| 210 | error = lookup_handle_found_vnode(ndp, &ndp->ni_cnd, rdonly, vbusyflags, |
| 211 | &keep_going, ndp->ni_ncgeneration, wantparent, 0, ctx); |
| 212 | if (error) |
| 213 | goto out_drop; |
| 214 | if (keep_going) { |
| 215 | if ((cnp->cn_flags & ISSYMLINK) == 0) { |
| 216 | panic("We need to keep going on a continued lookup, but for vp type %d (tag %d)\n" , ndp->ni_vp->v_type, ndp->ni_vp->v_tag); |
| 217 | } |
| 218 | goto continue_symlink; |
| 219 | } |
| 220 | |
| 221 | return 0; |
| 222 | |
| 223 | } |
| 224 | |
| 225 | vnode_recycled: |
| 226 | |
| 227 | /* |
| 228 | * Get a buffer for the name to be translated, and copy the |
| 229 | * name into the buffer. |
| 230 | */ |
| 231 | if ((cnp->cn_flags & HASBUF) == 0) { |
| 232 | cnp->cn_pnbuf = ndp->ni_pathbuf; |
| 233 | cnp->cn_pnlen = PATHBUFLEN; |
| 234 | } |
| 235 | #if LP64_DEBUG |
| 236 | if ((UIO_SEG_IS_USER_SPACE(ndp->ni_segflg) == 0) |
| 237 | && (ndp->ni_segflg != UIO_SYSSPACE) |
| 238 | && (ndp->ni_segflg != UIO_SYSSPACE32)) { |
| 239 | panic("%s :%d - invalid ni_segflg\n" , __FILE__, __LINE__); |
| 240 | } |
| 241 | #endif /* LP64_DEBUG */ |
| 242 | |
| 243 | retry_copy: |
| 244 | if (UIO_SEG_IS_USER_SPACE(ndp->ni_segflg)) { |
| 245 | error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, |
| 246 | cnp->cn_pnlen, &bytes_copied); |
| 247 | } else { |
| 248 | error = copystr(CAST_DOWN(void *, ndp->ni_dirp), cnp->cn_pnbuf, |
| 249 | cnp->cn_pnlen, &bytes_copied); |
| 250 | } |
| 251 | if (error == ENAMETOOLONG && !(cnp->cn_flags & HASBUF)) { |
| 252 | MALLOC_ZONE(cnp->cn_pnbuf, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK); |
| 253 | if (cnp->cn_pnbuf == NULL) { |
| 254 | error = ENOMEM; |
| 255 | goto error_out; |
| 256 | } |
| 257 | |
| 258 | cnp->cn_flags |= HASBUF; |
| 259 | cnp->cn_pnlen = MAXPATHLEN; |
| 260 | bytes_copied = 0; |
| 261 | |
| 262 | goto retry_copy; |
| 263 | } |
| 264 | if (error) |
| 265 | goto error_out; |
| 266 | ndp->ni_pathlen = bytes_copied; |
| 267 | bytes_copied = 0; |
| 268 | |
| 269 | /* |
| 270 | * Since the name cache may contain positive entries of |
| 271 | * the incorrect case, force lookup() to bypass the cache |
| 272 | * and call directly into the filesystem for each path |
| 273 | * component. Note: the FS may still consult the cache, |
| 274 | * but can apply rules to validate the results. |
| 275 | */ |
| 276 | if (proc_is_forcing_hfs_case_sensitivity(p)) |
| 277 | cnp->cn_flags |= CN_SKIPNAMECACHE; |
| 278 | |
| 279 | #if CONFIG_VOLFS |
| 280 | /* |
| 281 | * Check for legacy volfs style pathnames. |
| 282 | * |
| 283 | * For compatibility reasons we currently allow these paths, |
| 284 | * but future versions of the OS may not support them. |
| 285 | */ |
| 286 | if (ndp->ni_pathlen >= VOLFS_MIN_PATH_LEN && |
| 287 | cnp->cn_pnbuf[0] == '/' && |
| 288 | cnp->cn_pnbuf[1] == '.' && |
| 289 | cnp->cn_pnbuf[2] == 'v' && |
| 290 | cnp->cn_pnbuf[3] == 'o' && |
| 291 | cnp->cn_pnbuf[4] == 'l' && |
| 292 | cnp->cn_pnbuf[5] == '/' ) { |
| 293 | char * realpath; |
| 294 | int realpath_err; |
| 295 | /* Attempt to resolve a legacy volfs style pathname. */ |
| 296 | MALLOC_ZONE(realpath, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK); |
| 297 | if (realpath) { |
| 298 | /* |
| 299 | * We only error out on the ENAMETOOLONG cases where we know that |
| 300 | * vfs_getrealpath translation succeeded but the path could not fit into |
| 301 | * MAXPATHLEN characters. In other failure cases, we may be dealing with a path |
| 302 | * that legitimately looks like /.vol/1234/567 and is not meant to be translated |
| 303 | */ |
| 304 | if ((realpath_err= vfs_getrealpath(&cnp->cn_pnbuf[6], realpath, MAXPATHLEN, ctx))) { |
| 305 | FREE_ZONE(realpath, MAXPATHLEN, M_NAMEI); |
| 306 | if (realpath_err == ENOSPC || realpath_err == ENAMETOOLONG){ |
| 307 | error = ENAMETOOLONG; |
| 308 | goto error_out; |
| 309 | } |
| 310 | } else { |
| 311 | if (cnp->cn_flags & HASBUF) { |
| 312 | FREE_ZONE(cnp->cn_pnbuf, cnp->cn_pnlen, M_NAMEI); |
| 313 | } |
| 314 | cnp->cn_pnbuf = realpath; |
| 315 | cnp->cn_pnlen = MAXPATHLEN; |
| 316 | ndp->ni_pathlen = strlen(realpath) + 1; |
| 317 | cnp->cn_flags |= HASBUF | CN_VOLFSPATH; |
| 318 | } |
| 319 | } |
| 320 | } |
| 321 | #endif /* CONFIG_VOLFS */ |
| 322 | |
| 323 | #if CONFIG_AUDIT |
| 324 | /* If we are auditing the kernel pathname, save the user pathname */ |
| 325 | if (cnp->cn_flags & AUDITVNPATH1) |
| 326 | AUDIT_ARG(upath, ut->uu_cdir, cnp->cn_pnbuf, ARG_UPATH1); |
| 327 | if (cnp->cn_flags & AUDITVNPATH2) |
| 328 | AUDIT_ARG(upath, ut->uu_cdir, cnp->cn_pnbuf, ARG_UPATH2); |
| 329 | #endif /* CONFIG_AUDIT */ |
| 330 | |
| 331 | /* |
| 332 | * Do not allow empty pathnames |
| 333 | */ |
| 334 | if (*cnp->cn_pnbuf == '\0') { |
| 335 | error = ENOENT; |
| 336 | goto error_out; |
| 337 | } |
| 338 | ndp->ni_loopcnt = 0; |
| 339 | |
| 340 | /* |
| 341 | * determine the starting point for the translation. |
| 342 | */ |
| 343 | if ((ndp->ni_rootdir = fdp->fd_rdir) == NULLVP) { |
| 344 | if ( !(fdp->fd_flags & FD_CHROOT)) |
| 345 | ndp->ni_rootdir = rootvnode; |
| 346 | } |
| 347 | cnp->cn_nameptr = cnp->cn_pnbuf; |
| 348 | |
| 349 | ndp->ni_usedvp = NULLVP; |
| 350 | |
| 351 | if (*(cnp->cn_nameptr) == '/') { |
| 352 | while (*(cnp->cn_nameptr) == '/') { |
| 353 | cnp->cn_nameptr++; |
| 354 | ndp->ni_pathlen--; |
| 355 | } |
| 356 | dp = ndp->ni_rootdir; |
| 357 | } else if (cnp->cn_flags & USEDVP) { |
| 358 | dp = ndp->ni_dvp; |
| 359 | ndp->ni_usedvp = dp; |
| 360 | } else |
| 361 | dp = vfs_context_cwd(ctx); |
| 362 | |
| 363 | if (dp == NULLVP || (dp->v_lflag & VL_DEAD)) { |
| 364 | error = ENOENT; |
| 365 | goto error_out; |
| 366 | } |
| 367 | ndp->ni_dvp = NULLVP; |
| 368 | ndp->ni_vp = NULLVP; |
| 369 | |
| 370 | for (;;) { |
| 371 | #if CONFIG_MACF |
| 372 | /* |
| 373 | * Give MACF policies a chance to reject the lookup |
| 374 | * before performing any filesystem operations. |
| 375 | * This hook is called before resolving the path and |
| 376 | * again each time a symlink is encountered. |
| 377 | * NB: policies receive path information as supplied |
| 378 | * by the caller and thus cannot be trusted. |
| 379 | */ |
| 380 | error = mac_vnode_check_lookup_preflight(ctx, dp, cnp->cn_nameptr, cnp->cn_namelen); |
| 381 | if (error) { |
| 382 | goto error_out; |
| 383 | } |
| 384 | #endif |
| 385 | |
| 386 | ndp->ni_startdir = dp; |
| 387 | |
| 388 | if ( (error = lookup(ndp)) ) { |
| 389 | goto error_out; |
| 390 | } |
| 391 | |
| 392 | /* |
| 393 | * Check for symbolic link |
| 394 | */ |
| 395 | if ((cnp->cn_flags & ISSYMLINK) == 0) { |
| 396 | return (0); |
| 397 | } |
| 398 | |
| 399 | continue_symlink: |
| 400 | /* Gives us a new path to process, and a starting dir */ |
| 401 | error = lookup_handle_symlink(ndp, &dp, ctx); |
| 402 | if (error != 0) { |
| 403 | break; |
| 404 | } |
| 405 | } |
| 406 | /* |
| 407 | * only come here if we fail to handle a SYMLINK... |
| 408 | * if either ni_dvp or ni_vp is non-NULL, then |
| 409 | * we need to drop the iocount that was picked |
| 410 | * up in the lookup routine |
| 411 | */ |
| 412 | out_drop: |
| 413 | if (ndp->ni_dvp) |
| 414 | vnode_put(ndp->ni_dvp); |
| 415 | if (ndp->ni_vp) |
| 416 | vnode_put(ndp->ni_vp); |
| 417 | error_out: |
| 418 | if ( (cnp->cn_flags & HASBUF) ) { |
| 419 | cnp->cn_flags &= ~HASBUF; |
| 420 | FREE_ZONE(cnp->cn_pnbuf, cnp->cn_pnlen, M_NAMEI); |
| 421 | } |
| 422 | cnp->cn_pnbuf = NULL; |
| 423 | ndp->ni_vp = NULLVP; |
| 424 | ndp->ni_dvp = NULLVP; |
| 425 | |
| 426 | #if CONFIG_VOLFS |
| 427 | /* |
| 428 | * Deal with volfs fallout. |
| 429 | * |
| 430 | * At this point, if we were originally given a volfs path that |
| 431 | * looks like /.vol/123/456, then we would have had to convert it into |
| 432 | * a full path. Assuming that part worked properly, we will now attempt |
| 433 | * to conduct a lookup of the item in the namespace. Under normal |
| 434 | * circumstances, if a user looked up /tmp/foo and it was not there, it |
| 435 | * would be permissible to return ENOENT. |
| 436 | * |
| 437 | * However, we may not want to do that here. Specifically, the volfs path |
| 438 | * uniquely identifies a certain item in the namespace regardless of where it |
| 439 | * lives. If the item has moved in between the time we constructed the |
| 440 | * path and now, when we're trying to do a lookup/authorization on the full |
| 441 | * path, we may have gotten an ENOENT. |
| 442 | * |
| 443 | * At this point we can no longer tell if the path no longer exists |
| 444 | * or if the item in question no longer exists. It could have been renamed |
| 445 | * away, in which case the /.vol identifier is still valid. |
| 446 | * |
| 447 | * Do this dance a maximum of MAX_VOLFS_RESTARTS times. |
| 448 | */ |
| 449 | if ((error == ENOENT) && (ndp->ni_cnd.cn_flags & CN_VOLFSPATH)) { |
| 450 | if (volfs_restarts < MAX_VOLFS_RESTARTS) { |
| 451 | volfs_restarts++; |
| 452 | goto vnode_recycled; |
| 453 | } |
| 454 | } |
| 455 | #endif |
| 456 | |
| 457 | if (error == ERECYCLE){ |
| 458 | /* vnode was recycled underneath us. re-drive lookup to start at |
| 459 | the beginning again, since recycling invalidated last lookup*/ |
| 460 | ndp->ni_cnd.cn_flags = cnpflags; |
| 461 | ndp->ni_dvp = usedvp; |
| 462 | goto vnode_recycled; |
| 463 | } |
| 464 | |
| 465 | |
| 466 | return (error); |
| 467 | } |
| 468 | |
| 469 | int |
| 470 | namei_compound_available(vnode_t dp, struct nameidata *ndp) |
| 471 | { |
| 472 | if ((ndp->ni_flag & NAMEI_COMPOUNDOPEN) != 0) { |
| 473 | return vnode_compound_open_available(dp); |
| 474 | } |
| 475 | |
| 476 | return 0; |
| 477 | } |
| 478 | |
| 479 | static int |
| 480 | lookup_authorize_search(vnode_t dp, struct componentname *cnp, int dp_authorized_in_cache, vfs_context_t ctx) |
| 481 | { |
| 482 | #if !CONFIG_MACF |
| 483 | #pragma unused(cnp) |
| 484 | #endif |
| 485 | |
| 486 | int error; |
| 487 | |
| 488 | if (!dp_authorized_in_cache) { |
| 489 | error = vnode_authorize(dp, NULL, KAUTH_VNODE_SEARCH, ctx); |
| 490 | if (error) |
| 491 | return error; |
| 492 | } |
| 493 | #if CONFIG_MACF |
| 494 | error = mac_vnode_check_lookup(ctx, dp, cnp); |
| 495 | if (error) |
| 496 | return error; |
| 497 | #endif /* CONFIG_MACF */ |
| 498 | |
| 499 | return 0; |
| 500 | } |
| 501 | |
| 502 | static void |
| 503 | lookup_consider_update_cache(vnode_t dvp, vnode_t vp, struct componentname *cnp, int nc_generation) |
| 504 | { |
| 505 | int isdot_or_dotdot; |
| 506 | isdot_or_dotdot = (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') || (cnp->cn_flags & ISDOTDOT); |
| 507 | |
| 508 | if (vp->v_name == NULL || vp->v_parent == NULLVP) { |
| 509 | int update_flags = 0; |
| 510 | |
| 511 | if (isdot_or_dotdot == 0) { |
| 512 | if (vp->v_name == NULL) |
| 513 | update_flags |= VNODE_UPDATE_NAME; |
| 514 | if (dvp != NULLVP && vp->v_parent == NULLVP) |
| 515 | update_flags |= VNODE_UPDATE_PARENT; |
| 516 | |
| 517 | if (update_flags) |
| 518 | vnode_update_identity(vp, dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_hash, update_flags); |
| 519 | } |
| 520 | } |
| 521 | if ( (cnp->cn_flags & MAKEENTRY) && (vp->v_flag & VNCACHEABLE) && LIST_FIRST(&vp->v_nclinks) == NULL) { |
| 522 | /* |
| 523 | * missing from name cache, but should |
| 524 | * be in it... this can happen if volfs |
| 525 | * causes the vnode to be created or the |
| 526 | * name cache entry got recycled but the |
| 527 | * vnode didn't... |
| 528 | * check to make sure that ni_dvp is valid |
| 529 | * cache_lookup_path may return a NULL |
| 530 | * do a quick check to see if the generation of the |
| 531 | * directory matches our snapshot... this will get |
| 532 | * rechecked behind the name cache lock, but if it |
| 533 | * already fails to match, no need to go any further |
| 534 | */ |
| 535 | if (dvp != NULLVP && (nc_generation == dvp->v_nc_generation) && (!isdot_or_dotdot)) |
| 536 | cache_enter_with_gen(dvp, vp, cnp, nc_generation); |
| 537 | } |
| 538 | |
| 539 | } |
| 540 | |
| 541 | #if NAMEDRSRCFORK |
| 542 | /* |
| 543 | * Can change ni_dvp and ni_vp. On success, returns with iocounts on stream vnode (always) and |
| 544 | * data fork if requested. On failure, returns with iocount data fork (always) and its parent directory |
| 545 | * (if one was provided). |
| 546 | */ |
| 547 | static int |
| 548 | lookup_handle_rsrc_fork(vnode_t dp, struct nameidata *ndp, struct componentname *cnp, int wantparent, vfs_context_t ctx) |
| 549 | { |
| 550 | vnode_t svp = NULLVP; |
| 551 | enum nsoperation nsop; |
| 552 | int nsflags; |
| 553 | int error; |
| 554 | |
| 555 | if (dp->v_type != VREG) { |
| 556 | error = ENOENT; |
| 557 | goto out; |
| 558 | } |
| 559 | switch (cnp->cn_nameiop) { |
| 560 | case DELETE: |
| 561 | if (cnp->cn_flags & CN_ALLOWRSRCFORK) { |
| 562 | nsop = NS_DELETE; |
| 563 | } else { |
| 564 | error = EPERM; |
| 565 | goto out; |
| 566 | } |
| 567 | break; |
| 568 | case CREATE: |
| 569 | if (cnp->cn_flags & CN_ALLOWRSRCFORK) { |
| 570 | nsop = NS_CREATE; |
| 571 | } else { |
| 572 | error = EPERM; |
| 573 | goto out; |
| 574 | } |
| 575 | break; |
| 576 | case LOOKUP: |
| 577 | /* Make sure our lookup of "/..namedfork/rsrc" is allowed. */ |
| 578 | if (cnp->cn_flags & CN_ALLOWRSRCFORK) { |
| 579 | nsop = NS_OPEN; |
| 580 | } else { |
| 581 | error = EPERM; |
| 582 | goto out; |
| 583 | } |
| 584 | break; |
| 585 | default: |
| 586 | error = EPERM; |
| 587 | goto out; |
| 588 | } |
| 589 | |
| 590 | nsflags = 0; |
| 591 | if (cnp->cn_flags & CN_RAW_ENCRYPTED) |
| 592 | nsflags |= NS_GETRAWENCRYPTED; |
| 593 | |
| 594 | /* Ask the file system for the resource fork. */ |
| 595 | error = vnode_getnamedstream(dp, &svp, XATTR_RESOURCEFORK_NAME, nsop, nsflags, ctx); |
| 596 | |
| 597 | /* During a create, it OK for stream vnode to be missing. */ |
| 598 | if (error == ENOATTR || error == ENOENT) { |
| 599 | error = (nsop == NS_CREATE) ? 0 : ENOENT; |
| 600 | } |
| 601 | if (error) { |
| 602 | goto out; |
| 603 | } |
| 604 | /* The "parent" of the stream is the file. */ |
| 605 | if (wantparent) { |
| 606 | if (ndp->ni_dvp) { |
| 607 | vnode_put(ndp->ni_dvp); |
| 608 | } |
| 609 | ndp->ni_dvp = dp; |
| 610 | } else { |
| 611 | vnode_put(dp); |
| 612 | } |
| 613 | ndp->ni_vp = svp; /* on create this may be null */ |
| 614 | |
| 615 | /* Restore the truncated pathname buffer (for audits). */ |
| 616 | if (ndp->ni_pathlen == 1 && ndp->ni_next[0] == '\0') { |
| 617 | ndp->ni_next[0] = '/'; |
| 618 | } |
| 619 | cnp->cn_flags &= ~MAKEENTRY; |
| 620 | |
| 621 | return 0; |
| 622 | out: |
| 623 | return error; |
| 624 | } |
| 625 | #endif /* NAMEDRSRCFORK */ |
| 626 | |
| 627 | /* |
| 628 | * iocounts in: |
| 629 | * --One on ni_vp. One on ni_dvp if there is more path, or we didn't come through the |
| 630 | * cache, or we came through the cache and the caller doesn't want the parent. |
| 631 | * |
| 632 | * iocounts out: |
| 633 | * --Leaves us in the correct state for the next step, whatever that might be. |
| 634 | * --If we find a symlink, returns with iocounts on both ni_vp and ni_dvp. |
| 635 | * --If we are to look up another component, then we have an iocount on ni_vp and |
| 636 | * nothing else. |
| 637 | * --If we are done, returns an iocount on ni_vp, and possibly on ni_dvp depending on nameidata flags. |
| 638 | * --In the event of an error, may return with ni_dvp NULL'ed out (in which case, iocount |
| 639 | * was dropped). |
| 640 | */ |
| 641 | static int |
| 642 | lookup_handle_found_vnode(struct nameidata *ndp, struct componentname *cnp, int rdonly, |
| 643 | int vbusyflags, int *keep_going, int nc_generation, |
| 644 | int wantparent, int atroot, vfs_context_t ctx) |
| 645 | { |
| 646 | vnode_t dp; |
| 647 | int error; |
| 648 | char *cp; |
| 649 | |
| 650 | dp = ndp->ni_vp; |
| 651 | *keep_going = 0; |
| 652 | |
| 653 | if (ndp->ni_vp == NULLVP) { |
| 654 | panic("NULL ni_vp in %s\n" , __FUNCTION__); |
| 655 | } |
| 656 | |
| 657 | if (atroot) { |
| 658 | goto nextname; |
| 659 | } |
| 660 | |
| 661 | /* |
| 662 | * Take into account any additional components consumed by |
| 663 | * the underlying filesystem. |
| 664 | */ |
| 665 | if (cnp->cn_consume > 0) { |
| 666 | cnp->cn_nameptr += cnp->cn_consume; |
| 667 | ndp->ni_next += cnp->cn_consume; |
| 668 | ndp->ni_pathlen -= cnp->cn_consume; |
| 669 | cnp->cn_consume = 0; |
| 670 | } else { |
| 671 | lookup_consider_update_cache(ndp->ni_dvp, dp, cnp, nc_generation); |
| 672 | } |
| 673 | |
| 674 | /* |
| 675 | * Check to see if the vnode has been mounted on... |
| 676 | * if so find the root of the mounted file system. |
| 677 | * Updates ndp->ni_vp. |
| 678 | */ |
| 679 | error = lookup_traverse_mountpoints(ndp, cnp, dp, vbusyflags, ctx); |
| 680 | dp = ndp->ni_vp; |
| 681 | if (error) { |
| 682 | goto out; |
| 683 | } |
| 684 | |
| 685 | #if CONFIG_MACF |
| 686 | if (vfs_flags(vnode_mount(dp)) & MNT_MULTILABEL) { |
| 687 | error = vnode_label(vnode_mount(dp), NULL, dp, NULL, 0, ctx); |
| 688 | if (error) |
| 689 | goto out; |
| 690 | } |
| 691 | #endif |
| 692 | |
| 693 | /* |
| 694 | * Check for symbolic link |
| 695 | */ |
| 696 | if ((dp->v_type == VLNK) && |
| 697 | ((cnp->cn_flags & FOLLOW) || (ndp->ni_flag & NAMEI_TRAILINGSLASH) || *ndp->ni_next == '/')) { |
| 698 | cnp->cn_flags |= ISSYMLINK; |
| 699 | *keep_going = 1; |
| 700 | return (0); |
| 701 | } |
| 702 | |
| 703 | /* |
| 704 | * Check for bogus trailing slashes. |
| 705 | */ |
| 706 | if ((ndp->ni_flag & NAMEI_TRAILINGSLASH)) { |
| 707 | if (dp->v_type != VDIR) { |
| 708 | error = ENOTDIR; |
| 709 | goto out; |
| 710 | } |
| 711 | ndp->ni_flag &= ~(NAMEI_TRAILINGSLASH); |
| 712 | } |
| 713 | |
| 714 | #if NAMEDSTREAMS |
| 715 | /* |
| 716 | * Deny namei/lookup requests to resolve paths that point to shadow files. |
| 717 | * Access to shadow files must be conducted by explicit calls to VNOP_LOOKUP |
| 718 | * directly, and not use lookup/namei |
| 719 | */ |
| 720 | if (vnode_isshadow (dp)) { |
| 721 | error = ENOENT; |
| 722 | goto out; |
| 723 | } |
| 724 | #endif |
| 725 | |
| 726 | nextname: |
| 727 | /* |
| 728 | * Not a symbolic link. If more pathname, |
| 729 | * continue at next component, else return. |
| 730 | * |
| 731 | * Definitely have a dvp if there's another slash |
| 732 | */ |
| 733 | if (*ndp->ni_next == '/') { |
| 734 | cnp->cn_nameptr = ndp->ni_next + 1; |
| 735 | ndp->ni_pathlen--; |
| 736 | while (*cnp->cn_nameptr == '/') { |
| 737 | cnp->cn_nameptr++; |
| 738 | ndp->ni_pathlen--; |
| 739 | } |
| 740 | |
| 741 | cp = cnp->cn_nameptr; |
| 742 | vnode_put(ndp->ni_dvp); |
| 743 | ndp->ni_dvp = NULLVP; |
| 744 | |
| 745 | if (*cp == '\0') { |
| 746 | goto emptyname; |
| 747 | } |
| 748 | |
| 749 | *keep_going = 1; |
| 750 | return 0; |
| 751 | } |
| 752 | |
| 753 | /* |
| 754 | * Disallow directory write attempts on read-only file systems. |
| 755 | */ |
| 756 | if (rdonly && |
| 757 | (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { |
| 758 | error = EROFS; |
| 759 | goto out; |
| 760 | } |
| 761 | |
| 762 | /* If SAVESTART is set, we should have a dvp */ |
| 763 | if (cnp->cn_flags & SAVESTART) { |
| 764 | /* |
| 765 | * note that we already hold a reference |
| 766 | * on both dp and ni_dvp, but for some reason |
| 767 | * can't get another one... in this case we |
| 768 | * need to do vnode_put on dp in 'bad2' |
| 769 | */ |
| 770 | if ( (vnode_get(ndp->ni_dvp)) ) { |
| 771 | error = ENOENT; |
| 772 | goto out; |
| 773 | } |
| 774 | ndp->ni_startdir = ndp->ni_dvp; |
| 775 | } |
| 776 | if (!wantparent && ndp->ni_dvp) { |
| 777 | vnode_put(ndp->ni_dvp); |
| 778 | ndp->ni_dvp = NULLVP; |
| 779 | } |
| 780 | |
| 781 | if (cnp->cn_flags & AUDITVNPATH1) |
| 782 | AUDIT_ARG(vnpath, dp, ARG_VNODE1); |
| 783 | else if (cnp->cn_flags & AUDITVNPATH2) |
| 784 | AUDIT_ARG(vnpath, dp, ARG_VNODE2); |
| 785 | |
| 786 | #if NAMEDRSRCFORK |
| 787 | /* |
| 788 | * Caller wants the resource fork. |
| 789 | */ |
| 790 | if ((cnp->cn_flags & CN_WANTSRSRCFORK) && (dp != NULLVP)) { |
| 791 | error = lookup_handle_rsrc_fork(dp, ndp, cnp, wantparent, ctx); |
| 792 | if (error != 0) |
| 793 | goto out; |
| 794 | |
| 795 | dp = ndp->ni_vp; |
| 796 | } |
| 797 | #endif |
| 798 | if (kdebug_enable) |
| 799 | kdebug_lookup(ndp->ni_vp, cnp); |
| 800 | |
| 801 | return 0; |
| 802 | |
| 803 | emptyname: |
| 804 | error = lookup_handle_emptyname(ndp, cnp, wantparent); |
| 805 | if (error != 0) |
| 806 | goto out; |
| 807 | |
| 808 | return 0; |
| 809 | out: |
| 810 | return error; |
| 811 | |
| 812 | } |
| 813 | |
| 814 | /* |
| 815 | * Comes in iocount on ni_vp. May overwrite ni_dvp, but doesn't interpret incoming value. |
| 816 | */ |
| 817 | static int |
| 818 | lookup_handle_emptyname(struct nameidata *ndp, struct componentname *cnp, int wantparent) |
| 819 | { |
| 820 | vnode_t dp; |
| 821 | int error = 0; |
| 822 | |
| 823 | dp = ndp->ni_vp; |
| 824 | cnp->cn_namelen = 0; |
| 825 | /* |
| 826 | * A degenerate name (e.g. / or "") which is a way of |
| 827 | * talking about a directory, e.g. like "/." or ".". |
| 828 | */ |
| 829 | if (dp->v_type != VDIR) { |
| 830 | error = ENOTDIR; |
| 831 | goto out; |
| 832 | } |
| 833 | if (cnp->cn_nameiop != LOOKUP) { |
| 834 | error = EISDIR; |
| 835 | goto out; |
| 836 | } |
| 837 | if (wantparent) { |
| 838 | /* |
| 839 | * note that we already hold a reference |
| 840 | * on dp, but for some reason can't |
| 841 | * get another one... in this case we |
| 842 | * need to do vnode_put on dp in 'bad' |
| 843 | */ |
| 844 | if ( (vnode_get(dp)) ) { |
| 845 | error = ENOENT; |
| 846 | goto out; |
| 847 | } |
| 848 | ndp->ni_dvp = dp; |
| 849 | } |
| 850 | cnp->cn_flags &= ~ISDOTDOT; |
| 851 | cnp->cn_flags |= ISLASTCN; |
| 852 | ndp->ni_next = cnp->cn_nameptr; |
| 853 | ndp->ni_vp = dp; |
| 854 | |
| 855 | if (cnp->cn_flags & AUDITVNPATH1) |
| 856 | AUDIT_ARG(vnpath, dp, ARG_VNODE1); |
| 857 | else if (cnp->cn_flags & AUDITVNPATH2) |
| 858 | AUDIT_ARG(vnpath, dp, ARG_VNODE2); |
| 859 | if (cnp->cn_flags & SAVESTART) |
| 860 | panic("lookup: SAVESTART" ); |
| 861 | |
| 862 | return 0; |
| 863 | out: |
| 864 | return error; |
| 865 | } |
| 866 | /* |
| 867 | * Search a pathname. |
| 868 | * This is a very central and rather complicated routine. |
| 869 | * |
| 870 | * The pathname is pointed to by ni_ptr and is of length ni_pathlen. |
| 871 | * The starting directory is taken from ni_startdir. The pathname is |
| 872 | * descended until done, or a symbolic link is encountered. The variable |
| 873 | * ni_more is clear if the path is completed; it is set to one if a |
| 874 | * symbolic link needing interpretation is encountered. |
| 875 | * |
| 876 | * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on |
| 877 | * whether the name is to be looked up, created, renamed, or deleted. |
| 878 | * When CREATE, RENAME, or DELETE is specified, information usable in |
| 879 | * creating, renaming, or deleting a directory entry may be calculated. |
| 880 | * If flag has LOCKPARENT or'ed into it, the parent directory is returned |
| 881 | * locked. If flag has WANTPARENT or'ed into it, the parent directory is |
| 882 | * returned unlocked. Otherwise the parent directory is not returned. If |
| 883 | * the target of the pathname exists and LOCKLEAF is or'ed into the flag |
| 884 | * the target is returned locked, otherwise it is returned unlocked. |
| 885 | * When creating or renaming and LOCKPARENT is specified, the target may not |
| 886 | * be ".". When deleting and LOCKPARENT is specified, the target may be ".". |
| 887 | * |
| 888 | * Overall outline of lookup: |
| 889 | * |
| 890 | * dirloop: |
| 891 | * identify next component of name at ndp->ni_ptr |
| 892 | * handle degenerate case where name is null string |
| 893 | * if .. and crossing mount points and on mounted filesys, find parent |
| 894 | * call VNOP_LOOKUP routine for next component name |
| 895 | * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set |
| 896 | * component vnode returned in ni_vp (if it exists), locked. |
| 897 | * if result vnode is mounted on and crossing mount points, |
| 898 | * find mounted on vnode |
| 899 | * if more components of name, do next level at dirloop |
| 900 | * return the answer in ni_vp, locked if LOCKLEAF set |
| 901 | * if LOCKPARENT set, return locked parent in ni_dvp |
| 902 | * if WANTPARENT set, return unlocked parent in ni_dvp |
| 903 | * |
| 904 | * Returns: 0 Success |
| 905 | * ENOENT No such file or directory |
| 906 | * EBADF Bad file descriptor |
| 907 | * ENOTDIR Not a directory |
| 908 | * EROFS Read-only file system [CREATE] |
| 909 | * EISDIR Is a directory [CREATE] |
| 910 | * cache_lookup_path:ERECYCLE (vnode was recycled from underneath us, redrive lookup again) |
| 911 | * vnode_authorize:EROFS |
| 912 | * vnode_authorize:EACCES |
| 913 | * vnode_authorize:EPERM |
| 914 | * vnode_authorize:??? |
| 915 | * VNOP_LOOKUP:ENOENT No such file or directory |
| 916 | * VNOP_LOOKUP:EJUSTRETURN Restart system call (INTERNAL) |
| 917 | * VNOP_LOOKUP:??? |
| 918 | * VFS_ROOT:ENOTSUP |
| 919 | * VFS_ROOT:ENOENT |
| 920 | * VFS_ROOT:??? |
| 921 | */ |
| 922 | int |
| 923 | lookup(struct nameidata *ndp) |
| 924 | { |
| 925 | char *cp; /* pointer into pathname argument */ |
| 926 | vnode_t tdp; /* saved dp */ |
| 927 | vnode_t dp; /* the directory we are searching */ |
| 928 | int docache = 1; /* == 0 do not cache last component */ |
| 929 | int wantparent; /* 1 => wantparent or lockparent flag */ |
| 930 | int rdonly; /* lookup read-only flag bit */ |
| 931 | int dp_authorized = 0; |
| 932 | int error = 0; |
| 933 | struct componentname *cnp = &ndp->ni_cnd; |
| 934 | vfs_context_t ctx = cnp->cn_context; |
| 935 | int vbusyflags = 0; |
| 936 | int nc_generation = 0; |
| 937 | vnode_t last_dp = NULLVP; |
| 938 | int keep_going; |
| 939 | int atroot; |
| 940 | |
| 941 | /* |
| 942 | * Setup: break out flag bits into variables. |
| 943 | */ |
| 944 | if (cnp->cn_flags & NOCACHE) { |
| 945 | docache = 0; |
| 946 | } |
| 947 | wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT); |
| 948 | rdonly = cnp->cn_flags & RDONLY; |
| 949 | cnp->cn_flags &= ~ISSYMLINK; |
| 950 | cnp->cn_consume = 0; |
| 951 | |
| 952 | dp = ndp->ni_startdir; |
| 953 | ndp->ni_startdir = NULLVP; |
| 954 | |
| 955 | if ((cnp->cn_flags & CN_NBMOUNTLOOK) != 0) |
| 956 | vbusyflags = LK_NOWAIT; |
| 957 | cp = cnp->cn_nameptr; |
| 958 | |
| 959 | if (*cp == '\0') { |
| 960 | if ( (vnode_getwithref(dp)) ) { |
| 961 | dp = NULLVP; |
| 962 | error = ENOENT; |
| 963 | goto bad; |
| 964 | } |
| 965 | ndp->ni_vp = dp; |
| 966 | error = lookup_handle_emptyname(ndp, cnp, wantparent); |
| 967 | if (error) { |
| 968 | goto bad; |
| 969 | } |
| 970 | |
| 971 | return 0; |
| 972 | } |
| 973 | dirloop: |
| 974 | atroot = 0; |
| 975 | ndp->ni_vp = NULLVP; |
| 976 | |
| 977 | if ( (error = cache_lookup_path(ndp, cnp, dp, ctx, &dp_authorized, last_dp)) ) { |
| 978 | dp = NULLVP; |
| 979 | goto bad; |
| 980 | } |
| 981 | if ((cnp->cn_flags & ISLASTCN)) { |
| 982 | if (docache) |
| 983 | cnp->cn_flags |= MAKEENTRY; |
| 984 | } else |
| 985 | cnp->cn_flags |= MAKEENTRY; |
| 986 | |
| 987 | dp = ndp->ni_dvp; |
| 988 | |
| 989 | if (ndp->ni_vp != NULLVP) { |
| 990 | /* |
| 991 | * cache_lookup_path returned a non-NULL ni_vp then, |
| 992 | * we're guaranteed that the dp is a VDIR, it's |
| 993 | * been authorized, and vp is not ".." |
| 994 | * |
| 995 | * make sure we don't try to enter the name back into |
| 996 | * the cache if this vp is purged before we get to that |
| 997 | * check since we won't have serialized behind whatever |
| 998 | * activity is occurring in the FS that caused the purge |
| 999 | */ |
| 1000 | if (dp != NULLVP) |
| 1001 | nc_generation = dp->v_nc_generation - 1; |
| 1002 | |
| 1003 | goto returned_from_lookup_path; |
| 1004 | } |
| 1005 | |
| 1006 | /* |
| 1007 | * Handle "..": two special cases. |
| 1008 | * 1. If at root directory (e.g. after chroot) |
| 1009 | * or at absolute root directory |
| 1010 | * then ignore it so can't get out. |
| 1011 | * 2. If this vnode is the root of a mounted |
| 1012 | * filesystem, then replace it with the |
| 1013 | * vnode which was mounted on so we take the |
| 1014 | * .. in the other file system. |
| 1015 | */ |
| 1016 | if ( (cnp->cn_flags & ISDOTDOT) ) { |
| 1017 | /* |
| 1018 | * if this is a chroot'ed process, check if the current |
| 1019 | * directory is still a subdirectory of the process's |
| 1020 | * root directory. |
| 1021 | */ |
| 1022 | if (ndp->ni_rootdir && (ndp->ni_rootdir != rootvnode) && |
| 1023 | dp != ndp->ni_rootdir) { |
| 1024 | int sdir_error; |
| 1025 | int is_subdir = FALSE; |
| 1026 | |
| 1027 | sdir_error = vnode_issubdir(dp, ndp->ni_rootdir, |
| 1028 | &is_subdir, vfs_context_kernel()); |
| 1029 | |
| 1030 | /* |
| 1031 | * If we couldn't determine if dp is a subdirectory of |
| 1032 | * ndp->ni_rootdir (sdir_error != 0), we let the request |
| 1033 | * proceed. |
| 1034 | */ |
| 1035 | if (!sdir_error && !is_subdir) { |
| 1036 | vnode_put(dp); |
| 1037 | dp = ndp->ni_rootdir; |
| 1038 | /* |
| 1039 | * There's a ref on the process's root directory |
| 1040 | * but we can't use vnode_getwithref here as |
| 1041 | * there is nothing preventing that ref being |
| 1042 | * released by another thread. |
| 1043 | */ |
| 1044 | if (vnode_get(dp)) { |
| 1045 | error = ENOENT; |
| 1046 | goto bad; |
| 1047 | } |
| 1048 | } |
| 1049 | } |
| 1050 | |
| 1051 | for (;;) { |
| 1052 | if (dp == ndp->ni_rootdir || dp == rootvnode) { |
| 1053 | ndp->ni_dvp = dp; |
| 1054 | ndp->ni_vp = dp; |
| 1055 | /* |
| 1056 | * we're pinned at the root |
| 1057 | * we've already got one reference on 'dp' |
| 1058 | * courtesy of cache_lookup_path... take |
| 1059 | * another one for the ".." |
| 1060 | * if we fail to get the new reference, we'll |
| 1061 | * drop our original down in 'bad' |
| 1062 | */ |
| 1063 | if ( (vnode_get(dp)) ) { |
| 1064 | error = ENOENT; |
| 1065 | goto bad; |
| 1066 | } |
| 1067 | atroot = 1; |
| 1068 | goto returned_from_lookup_path; |
| 1069 | } |
| 1070 | if ((dp->v_flag & VROOT) == 0 || |
| 1071 | (cnp->cn_flags & NOCROSSMOUNT)) |
| 1072 | break; |
| 1073 | if (dp->v_mount == NULL) { /* forced umount */ |
| 1074 | error = EBADF; |
| 1075 | goto bad; |
| 1076 | } |
| 1077 | tdp = dp; |
| 1078 | dp = tdp->v_mount->mnt_vnodecovered; |
| 1079 | |
| 1080 | vnode_put(tdp); |
| 1081 | |
| 1082 | if ( (vnode_getwithref(dp)) ) { |
| 1083 | dp = NULLVP; |
| 1084 | error = ENOENT; |
| 1085 | goto bad; |
| 1086 | } |
| 1087 | ndp->ni_dvp = dp; |
| 1088 | dp_authorized = 0; |
| 1089 | } |
| 1090 | } |
| 1091 | |
| 1092 | /* |
| 1093 | * We now have a segment name to search for, and a directory to search. |
| 1094 | */ |
| 1095 | unionlookup: |
| 1096 | ndp->ni_vp = NULLVP; |
| 1097 | |
| 1098 | if (dp->v_type != VDIR) { |
| 1099 | error = ENOTDIR; |
| 1100 | goto lookup_error; |
| 1101 | } |
| 1102 | if ( (cnp->cn_flags & DONOTAUTH) != DONOTAUTH ) { |
| 1103 | error = lookup_authorize_search(dp, cnp, dp_authorized, ctx); |
| 1104 | if (error) { |
| 1105 | goto lookup_error; |
| 1106 | } |
| 1107 | } |
| 1108 | |
| 1109 | /* |
| 1110 | * Now that we've authorized a lookup, can bail out if the filesystem |
| 1111 | * will be doing a batched operation. Return an iocount on dvp. |
| 1112 | */ |
| 1113 | #if NAMEDRSRCFORK |
| 1114 | if ((cnp->cn_flags & ISLASTCN) && namei_compound_available(dp, ndp) && !(cnp->cn_flags & CN_WANTSRSRCFORK)) { |
| 1115 | #else |
| 1116 | if ((cnp->cn_flags & ISLASTCN) && namei_compound_available(dp, ndp)) { |
| 1117 | #endif /* NAMEDRSRCFORK */ |
| 1118 | ndp->ni_flag |= NAMEI_UNFINISHED; |
| 1119 | ndp->ni_ncgeneration = dp->v_nc_generation; |
| 1120 | return 0; |
| 1121 | } |
| 1122 | |
| 1123 | nc_generation = dp->v_nc_generation; |
| 1124 | |
| 1125 | /* |
| 1126 | * Note: |
| 1127 | * Filesystems that support hardlinks may want to call vnode_update_identity |
| 1128 | * if the lookup operation below will modify the in-core vnode to belong to a new point |
| 1129 | * in the namespace. VFS cannot infer whether or not the look up operation makes the vnode |
| 1130 | * name change or change parents. Without this, the lookup may make update |
| 1131 | * filesystem-specific in-core metadata but fail to update the v_parent or v_name |
| 1132 | * fields in the vnode. If VFS were to do this, it would be necessary to call |
| 1133 | * vnode_update_identity on every lookup operation -- expensive! |
| 1134 | * |
| 1135 | * However, even with this in place, multiple lookups may occur in between this lookup |
| 1136 | * and the subsequent vnop, so, at best, we could only guarantee that you would get a |
| 1137 | * valid path back, and not necessarily the one that you wanted. |
| 1138 | * |
| 1139 | * Example: |
| 1140 | * /tmp/a == /foo/b |
| 1141 | * |
| 1142 | * If you are now looking up /foo/b and the vnode for this link represents /tmp/a, |
| 1143 | * vnode_update_identity will fix the parentage so that you can get /foo/b back |
| 1144 | * through the v_parent chain (preventing you from getting /tmp/b back). It would |
| 1145 | * not fix whether or not you should or should not get /tmp/a vs. /foo/b. |
| 1146 | */ |
| 1147 | |
| 1148 | error = VNOP_LOOKUP(dp, &ndp->ni_vp, cnp, ctx); |
| 1149 | |
| 1150 | if ( error ) { |
| 1151 | lookup_error: |
| 1152 | if ((error == ENOENT) && |
| 1153 | (dp->v_mount != NULL) && |
| 1154 | (dp->v_mount->mnt_flag & MNT_UNION)) { |
| 1155 | tdp = dp; |
| 1156 | error = lookup_traverse_union(tdp, &dp, ctx); |
| 1157 | vnode_put(tdp); |
| 1158 | if (error) { |
| 1159 | dp = NULLVP; |
| 1160 | goto bad; |
| 1161 | } |
| 1162 | |
| 1163 | ndp->ni_dvp = dp; |
| 1164 | dp_authorized = 0; |
| 1165 | goto unionlookup; |
| 1166 | } |
| 1167 | |
| 1168 | if (error != EJUSTRETURN) |
| 1169 | goto bad; |
| 1170 | |
| 1171 | if (ndp->ni_vp != NULLVP) |
| 1172 | panic("leaf should be empty" ); |
| 1173 | |
| 1174 | #if NAMEDRSRCFORK |
| 1175 | /* |
| 1176 | * At this point, error should be EJUSTRETURN. |
| 1177 | * |
| 1178 | * If CN_WANTSRSRCFORK is set, that implies that the |
| 1179 | * underlying filesystem could not find the "parent" of the |
| 1180 | * resource fork (the data fork), and we are doing a lookup |
| 1181 | * for a CREATE event. |
| 1182 | * |
| 1183 | * However, this should be converted to an error, as the |
| 1184 | * failure to find this parent should disallow further |
| 1185 | * progress to try and acquire a resource fork vnode. |
| 1186 | */ |
| 1187 | if (cnp->cn_flags & CN_WANTSRSRCFORK) { |
| 1188 | error = ENOENT; |
| 1189 | goto bad; |
| 1190 | } |
| 1191 | #endif |
| 1192 | |
| 1193 | error = lookup_validate_creation_path(ndp); |
| 1194 | if (error) |
| 1195 | goto bad; |
| 1196 | /* |
| 1197 | * We return with ni_vp NULL to indicate that the entry |
| 1198 | * doesn't currently exist, leaving a pointer to the |
| 1199 | * referenced directory vnode in ndp->ni_dvp. |
| 1200 | */ |
| 1201 | if (cnp->cn_flags & SAVESTART) { |
| 1202 | if ( (vnode_get(ndp->ni_dvp)) ) { |
| 1203 | error = ENOENT; |
| 1204 | goto bad; |
| 1205 | } |
| 1206 | ndp->ni_startdir = ndp->ni_dvp; |
| 1207 | } |
| 1208 | if (!wantparent) |
| 1209 | vnode_put(ndp->ni_dvp); |
| 1210 | |
| 1211 | if (kdebug_enable) |
| 1212 | kdebug_lookup(ndp->ni_dvp, cnp); |
| 1213 | return (0); |
| 1214 | } |
| 1215 | returned_from_lookup_path: |
| 1216 | /* We'll always have an iocount on ni_vp when this finishes. */ |
| 1217 | error = lookup_handle_found_vnode(ndp, cnp, rdonly, vbusyflags, &keep_going, nc_generation, wantparent, atroot, ctx); |
| 1218 | if (error != 0) { |
| 1219 | goto bad2; |
| 1220 | } |
| 1221 | |
| 1222 | if (keep_going) { |
| 1223 | dp = ndp->ni_vp; |
| 1224 | |
| 1225 | /* namei() will handle symlinks */ |
| 1226 | if ((dp->v_type == VLNK) && |
| 1227 | ((cnp->cn_flags & FOLLOW) || (ndp->ni_flag & NAMEI_TRAILINGSLASH) || *ndp->ni_next == '/')) { |
| 1228 | return 0; |
| 1229 | } |
| 1230 | |
| 1231 | /* |
| 1232 | * Otherwise, there's more path to process. |
| 1233 | * cache_lookup_path is now responsible for dropping io ref on dp |
| 1234 | * when it is called again in the dirloop. This ensures we hold |
| 1235 | * a ref on dp until we complete the next round of lookup. |
| 1236 | */ |
| 1237 | last_dp = dp; |
| 1238 | |
| 1239 | goto dirloop; |
| 1240 | } |
| 1241 | |
| 1242 | return (0); |
| 1243 | bad2: |
| 1244 | if (ndp->ni_dvp) |
| 1245 | vnode_put(ndp->ni_dvp); |
| 1246 | |
| 1247 | vnode_put(ndp->ni_vp); |
| 1248 | ndp->ni_vp = NULLVP; |
| 1249 | |
| 1250 | if (kdebug_enable) |
| 1251 | kdebug_lookup(dp, cnp); |
| 1252 | return (error); |
| 1253 | |
| 1254 | bad: |
| 1255 | if (dp) |
| 1256 | vnode_put(dp); |
| 1257 | ndp->ni_vp = NULLVP; |
| 1258 | |
| 1259 | if (kdebug_enable) |
| 1260 | kdebug_lookup(dp, cnp); |
| 1261 | return (error); |
| 1262 | } |
| 1263 | |
| 1264 | /* |
| 1265 | * Given a vnode in a union mount, traverse to the equivalent |
| 1266 | * vnode in the underlying mount. |
| 1267 | */ |
| 1268 | int |
| 1269 | lookup_traverse_union(vnode_t dvp, vnode_t *new_dvp, vfs_context_t ctx) |
| 1270 | { |
| 1271 | char *path = NULL, *pp; |
| 1272 | const char *name, *np; |
| 1273 | int len; |
| 1274 | int error = 0; |
| 1275 | struct nameidata nd; |
| 1276 | vnode_t vp = dvp; |
| 1277 | |
| 1278 | *new_dvp = NULL; |
| 1279 | |
| 1280 | if (vp && vp->v_flag & VROOT) { |
| 1281 | *new_dvp = vp->v_mount->mnt_vnodecovered; |
| 1282 | if (vnode_getwithref(*new_dvp)) |
| 1283 | return ENOENT; |
| 1284 | return 0; |
| 1285 | } |
| 1286 | |
| 1287 | path = (char *) kalloc(MAXPATHLEN); |
| 1288 | if (path == NULL) { |
| 1289 | error = ENOMEM; |
| 1290 | goto done; |
| 1291 | } |
| 1292 | |
| 1293 | /* |
| 1294 | * Walk back up to the mountpoint following the |
| 1295 | * v_parent chain and build a slash-separated path. |
| 1296 | * Then lookup that path starting with the covered vnode. |
| 1297 | */ |
| 1298 | pp = path + (MAXPATHLEN - 1); |
| 1299 | *pp = '\0'; |
| 1300 | |
| 1301 | while (1) { |
| 1302 | name = vnode_getname(vp); |
| 1303 | if (name == NULL) { |
| 1304 | printf("lookup_traverse_union: null parent name: .%s\n" , pp); |
| 1305 | error = ENOENT; |
| 1306 | goto done; |
| 1307 | } |
| 1308 | len = strlen(name); |
| 1309 | if ((len + 1) > (pp - path)) { // Enough space for this name ? |
| 1310 | error = ENAMETOOLONG; |
| 1311 | vnode_putname(name); |
| 1312 | goto done; |
| 1313 | } |
| 1314 | for (np = name + len; len > 0; len--) // Copy name backwards |
| 1315 | *--pp = *--np; |
| 1316 | vnode_putname(name); |
| 1317 | vp = vp->v_parent; |
| 1318 | if (vp == NULLVP || vp->v_flag & VROOT) |
| 1319 | break; |
| 1320 | *--pp = '/'; |
| 1321 | } |
| 1322 | |
| 1323 | /* Evaluate the path in the underlying mount */ |
| 1324 | NDINIT(&nd, LOOKUP, OP_LOOKUP, USEDVP, UIO_SYSSPACE, CAST_USER_ADDR_T(pp), ctx); |
| 1325 | nd.ni_dvp = dvp->v_mount->mnt_vnodecovered; |
| 1326 | error = namei(&nd); |
| 1327 | if (error == 0) |
| 1328 | *new_dvp = nd.ni_vp; |
| 1329 | nameidone(&nd); |
| 1330 | done: |
| 1331 | if (path) |
| 1332 | kfree(path, MAXPATHLEN); |
| 1333 | return error; |
| 1334 | } |
| 1335 | |
| 1336 | int |
| 1337 | lookup_validate_creation_path(struct nameidata *ndp) |
| 1338 | { |
| 1339 | struct componentname *cnp = &ndp->ni_cnd; |
| 1340 | |
| 1341 | /* |
| 1342 | * If creating and at end of pathname, then can consider |
| 1343 | * allowing file to be created. |
| 1344 | */ |
| 1345 | if (cnp->cn_flags & RDONLY) { |
| 1346 | return EROFS; |
| 1347 | } |
| 1348 | if ((cnp->cn_flags & ISLASTCN) && (ndp->ni_flag & NAMEI_TRAILINGSLASH) && !(cnp->cn_flags & WILLBEDIR)) { |
| 1349 | return ENOENT; |
| 1350 | } |
| 1351 | |
| 1352 | return 0; |
| 1353 | } |
| 1354 | |
| 1355 | /* |
| 1356 | * Modifies only ni_vp. Always returns with ni_vp still valid (iocount held). |
| 1357 | */ |
| 1358 | static int |
| 1359 | lookup_traverse_mountpoints(struct nameidata *ndp, struct componentname *cnp, vnode_t dp, |
| 1360 | int vbusyflags, vfs_context_t ctx) |
| 1361 | { |
| 1362 | mount_t mp; |
| 1363 | vnode_t tdp; |
| 1364 | int error = 0; |
| 1365 | uint32_t depth = 0; |
| 1366 | vnode_t mounted_on_dp; |
| 1367 | int current_mount_generation = 0; |
| 1368 | #if CONFIG_TRIGGERS |
| 1369 | vnode_t triggered_dp = NULLVP; |
| 1370 | int retry_cnt = 0; |
| 1371 | #define MAX_TRIGGER_RETRIES 1 |
| 1372 | #endif |
| 1373 | |
| 1374 | if (dp->v_type != VDIR || cnp->cn_flags & NOCROSSMOUNT) |
| 1375 | return 0; |
| 1376 | |
| 1377 | mounted_on_dp = dp; |
| 1378 | #if CONFIG_TRIGGERS |
| 1379 | restart: |
| 1380 | #endif |
| 1381 | current_mount_generation = mount_generation; |
| 1382 | |
| 1383 | while (dp->v_mountedhere) { |
| 1384 | vnode_lock_spin(dp); |
| 1385 | if ((mp = dp->v_mountedhere)) { |
| 1386 | mp->mnt_crossref++; |
| 1387 | vnode_unlock(dp); |
| 1388 | } else { |
| 1389 | vnode_unlock(dp); |
| 1390 | break; |
| 1391 | } |
| 1392 | |
| 1393 | if (ISSET(mp->mnt_lflag, MNT_LFORCE)) { |
| 1394 | mount_dropcrossref(mp, dp, 0); |
| 1395 | break; // don't traverse into a forced unmount |
| 1396 | } |
| 1397 | |
| 1398 | |
| 1399 | if (vfs_busy(mp, vbusyflags)) { |
| 1400 | mount_dropcrossref(mp, dp, 0); |
| 1401 | if (vbusyflags == LK_NOWAIT) { |
| 1402 | error = ENOENT; |
| 1403 | goto out; |
| 1404 | } |
| 1405 | |
| 1406 | continue; |
| 1407 | } |
| 1408 | |
| 1409 | error = VFS_ROOT(mp, &tdp, ctx); |
| 1410 | |
| 1411 | mount_dropcrossref(mp, dp, 0); |
| 1412 | vfs_unbusy(mp); |
| 1413 | |
| 1414 | if (error) { |
| 1415 | goto out; |
| 1416 | } |
| 1417 | |
| 1418 | vnode_put(dp); |
| 1419 | ndp->ni_vp = dp = tdp; |
| 1420 | if (dp->v_type != VDIR) { |
| 1421 | #if DEVELOPMENT || DEBUG |
| 1422 | panic("%s : Root of filesystem not a directory\n" , |
| 1423 | __FUNCTION__); |
| 1424 | #else |
| 1425 | break; |
| 1426 | #endif |
| 1427 | } |
| 1428 | depth++; |
| 1429 | } |
| 1430 | |
| 1431 | #if CONFIG_TRIGGERS |
| 1432 | /* |
| 1433 | * The triggered_dp check here is required but is susceptible to a |
| 1434 | * (unlikely) race in which trigger mount is done from here and is |
| 1435 | * unmounted before we get past vfs_busy above. We retry to deal with |
| 1436 | * that case but it has the side effect of unwanted retries for |
| 1437 | * "special" processes which don't want to trigger mounts. |
| 1438 | */ |
| 1439 | if (dp->v_resolve && retry_cnt < MAX_TRIGGER_RETRIES) { |
| 1440 | error = vnode_trigger_resolve(dp, ndp, ctx); |
| 1441 | if (error) |
| 1442 | goto out; |
| 1443 | if (dp == triggered_dp) |
| 1444 | retry_cnt += 1; |
| 1445 | else |
| 1446 | retry_cnt = 0; |
| 1447 | triggered_dp = dp; |
| 1448 | goto restart; |
| 1449 | } |
| 1450 | #endif /* CONFIG_TRIGGERS */ |
| 1451 | |
| 1452 | if (depth) { |
| 1453 | mp = mounted_on_dp->v_mountedhere; |
| 1454 | |
| 1455 | if (mp) { |
| 1456 | mount_lock_spin(mp); |
| 1457 | mp->mnt_realrootvp_vid = dp->v_id; |
| 1458 | mp->mnt_realrootvp = dp; |
| 1459 | mp->mnt_generation = current_mount_generation; |
| 1460 | mount_unlock(mp); |
| 1461 | } |
| 1462 | } |
| 1463 | |
| 1464 | return 0; |
| 1465 | |
| 1466 | out: |
| 1467 | return error; |
| 1468 | } |
| 1469 | |
| 1470 | /* |
| 1471 | * Takes ni_vp and ni_dvp non-NULL. Returns with *new_dp set to the location |
| 1472 | * at which to start a lookup with a resolved path, and all other iocounts dropped. |
| 1473 | */ |
| 1474 | static int |
| 1475 | lookup_handle_symlink(struct nameidata *ndp, vnode_t *new_dp, vfs_context_t ctx) |
| 1476 | { |
| 1477 | int error; |
| 1478 | char *cp; /* pointer into pathname argument */ |
| 1479 | uio_t auio; |
| 1480 | union { |
| 1481 | union { |
| 1482 | struct user_iovec s_uiovec; |
| 1483 | struct kern_iovec s_kiovec; |
| 1484 | } u_iovec; |
| 1485 | struct uio s_uio; |
| 1486 | char uio_buf[ UIO_SIZEOF(1) ]; |
| 1487 | } u_uio_buf; /* union only for aligning uio_buf correctly */ |
| 1488 | int need_newpathbuf; |
| 1489 | u_int linklen; |
| 1490 | struct componentname *cnp = &ndp->ni_cnd; |
| 1491 | vnode_t dp; |
| 1492 | char *tmppn; |
| 1493 | |
| 1494 | if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { |
| 1495 | return ELOOP; |
| 1496 | } |
| 1497 | #if CONFIG_MACF |
| 1498 | if ((error = mac_vnode_check_readlink(ctx, ndp->ni_vp)) != 0) |
| 1499 | return error; |
| 1500 | #endif /* MAC */ |
| 1501 | if (ndp->ni_pathlen > 1 || !(cnp->cn_flags & HASBUF)) |
| 1502 | need_newpathbuf = 1; |
| 1503 | else |
| 1504 | need_newpathbuf = 0; |
| 1505 | |
| 1506 | if (need_newpathbuf) { |
| 1507 | MALLOC_ZONE(cp, char *, MAXPATHLEN, M_NAMEI, M_WAITOK); |
| 1508 | if (cp == NULL) { |
| 1509 | return ENOMEM; |
| 1510 | } |
| 1511 | } else { |
| 1512 | cp = cnp->cn_pnbuf; |
| 1513 | } |
| 1514 | auio = uio_createwithbuffer(1, 0, UIO_SYSSPACE, UIO_READ, |
| 1515 | &u_uio_buf.uio_buf[0], sizeof(u_uio_buf.uio_buf)); |
| 1516 | |
| 1517 | uio_addiov(auio, CAST_USER_ADDR_T(cp), MAXPATHLEN); |
| 1518 | |
| 1519 | error = VNOP_READLINK(ndp->ni_vp, auio, ctx); |
| 1520 | if (error) { |
| 1521 | if (need_newpathbuf) |
| 1522 | FREE_ZONE(cp, MAXPATHLEN, M_NAMEI); |
| 1523 | return error; |
| 1524 | } |
| 1525 | |
| 1526 | /* |
| 1527 | * Safe to set unsigned with a [larger] signed type here |
| 1528 | * because 0 <= uio_resid <= MAXPATHLEN and MAXPATHLEN |
| 1529 | * is only 1024. |
| 1530 | */ |
| 1531 | linklen = MAXPATHLEN - (u_int)uio_resid(auio); |
| 1532 | if (linklen + ndp->ni_pathlen > MAXPATHLEN) { |
| 1533 | if (need_newpathbuf) |
| 1534 | FREE_ZONE(cp, MAXPATHLEN, M_NAMEI); |
| 1535 | |
| 1536 | return ENAMETOOLONG; |
| 1537 | } |
| 1538 | if (need_newpathbuf) { |
| 1539 | long len = cnp->cn_pnlen; |
| 1540 | |
| 1541 | tmppn = cnp->cn_pnbuf; |
| 1542 | bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen); |
| 1543 | cnp->cn_pnbuf = cp; |
| 1544 | cnp->cn_pnlen = MAXPATHLEN; |
| 1545 | |
| 1546 | if ( (cnp->cn_flags & HASBUF) ) |
| 1547 | FREE_ZONE(tmppn, len, M_NAMEI); |
| 1548 | else |
| 1549 | cnp->cn_flags |= HASBUF; |
| 1550 | } else |
| 1551 | cnp->cn_pnbuf[linklen] = '\0'; |
| 1552 | |
| 1553 | ndp->ni_pathlen += linklen; |
| 1554 | cnp->cn_nameptr = cnp->cn_pnbuf; |
| 1555 | |
| 1556 | /* |
| 1557 | * starting point for 'relative' |
| 1558 | * symbolic link path |
| 1559 | */ |
| 1560 | dp = ndp->ni_dvp; |
| 1561 | |
| 1562 | /* |
| 1563 | * get rid of references returned via 'lookup' |
| 1564 | */ |
| 1565 | vnode_put(ndp->ni_vp); |
| 1566 | vnode_put(ndp->ni_dvp); /* ALWAYS have a dvp for a symlink */ |
| 1567 | |
| 1568 | ndp->ni_vp = NULLVP; |
| 1569 | ndp->ni_dvp = NULLVP; |
| 1570 | |
| 1571 | /* |
| 1572 | * Check if symbolic link restarts us at the root |
| 1573 | */ |
| 1574 | if (*(cnp->cn_nameptr) == '/') { |
| 1575 | while (*(cnp->cn_nameptr) == '/') { |
| 1576 | cnp->cn_nameptr++; |
| 1577 | ndp->ni_pathlen--; |
| 1578 | } |
| 1579 | if ((dp = ndp->ni_rootdir) == NULLVP) { |
| 1580 | return ENOENT; |
| 1581 | } |
| 1582 | } |
| 1583 | |
| 1584 | *new_dp = dp; |
| 1585 | |
| 1586 | return 0; |
| 1587 | } |
| 1588 | |
| 1589 | /* |
| 1590 | * relookup - lookup a path name component |
| 1591 | * Used by lookup to re-aquire things. |
| 1592 | */ |
| 1593 | int |
| 1594 | relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) |
| 1595 | { |
| 1596 | struct vnode *dp = NULL; /* the directory we are searching */ |
| 1597 | int wantparent; /* 1 => wantparent or lockparent flag */ |
| 1598 | int rdonly; /* lookup read-only flag bit */ |
| 1599 | int error = 0; |
| 1600 | #ifdef NAMEI_DIAGNOSTIC |
| 1601 | int i, newhash; /* DEBUG: check name hash */ |
| 1602 | char *cp; /* DEBUG: check name ptr/len */ |
| 1603 | #endif |
| 1604 | vfs_context_t ctx = cnp->cn_context;; |
| 1605 | |
| 1606 | /* |
| 1607 | * Setup: break out flag bits into variables. |
| 1608 | */ |
| 1609 | wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT); |
| 1610 | rdonly = cnp->cn_flags & RDONLY; |
| 1611 | cnp->cn_flags &= ~ISSYMLINK; |
| 1612 | |
| 1613 | if (cnp->cn_flags & NOCACHE) |
| 1614 | cnp->cn_flags &= ~MAKEENTRY; |
| 1615 | else |
| 1616 | cnp->cn_flags |= MAKEENTRY; |
| 1617 | |
| 1618 | dp = dvp; |
| 1619 | |
| 1620 | /* |
| 1621 | * Check for degenerate name (e.g. / or "") |
| 1622 | * which is a way of talking about a directory, |
| 1623 | * e.g. like "/." or ".". |
| 1624 | */ |
| 1625 | if (cnp->cn_nameptr[0] == '\0') { |
| 1626 | if (cnp->cn_nameiop != LOOKUP || wantparent) { |
| 1627 | error = EISDIR; |
| 1628 | goto bad; |
| 1629 | } |
| 1630 | if (dp->v_type != VDIR) { |
| 1631 | error = ENOTDIR; |
| 1632 | goto bad; |
| 1633 | } |
| 1634 | if ( (vnode_get(dp)) ) { |
| 1635 | error = ENOENT; |
| 1636 | goto bad; |
| 1637 | } |
| 1638 | *vpp = dp; |
| 1639 | |
| 1640 | if (cnp->cn_flags & SAVESTART) |
| 1641 | panic("lookup: SAVESTART" ); |
| 1642 | return (0); |
| 1643 | } |
| 1644 | /* |
| 1645 | * We now have a segment name to search for, and a directory to search. |
| 1646 | */ |
| 1647 | if ( (error = VNOP_LOOKUP(dp, vpp, cnp, ctx)) ) { |
| 1648 | if (error != EJUSTRETURN) |
| 1649 | goto bad; |
| 1650 | #if DIAGNOSTIC |
| 1651 | if (*vpp != NULL) |
| 1652 | panic("leaf should be empty" ); |
| 1653 | #endif |
| 1654 | /* |
| 1655 | * If creating and at end of pathname, then can consider |
| 1656 | * allowing file to be created. |
| 1657 | */ |
| 1658 | if (rdonly) { |
| 1659 | error = EROFS; |
| 1660 | goto bad; |
| 1661 | } |
| 1662 | /* |
| 1663 | * We return with ni_vp NULL to indicate that the entry |
| 1664 | * doesn't currently exist, leaving a pointer to the |
| 1665 | * (possibly locked) directory inode in ndp->ni_dvp. |
| 1666 | */ |
| 1667 | return (0); |
| 1668 | } |
| 1669 | dp = *vpp; |
| 1670 | |
| 1671 | #if DIAGNOSTIC |
| 1672 | /* |
| 1673 | * Check for symbolic link |
| 1674 | */ |
| 1675 | if (dp->v_type == VLNK && (cnp->cn_flags & FOLLOW)) |
| 1676 | panic ("relookup: symlink found.\n" ); |
| 1677 | #endif |
| 1678 | |
| 1679 | /* |
| 1680 | * Disallow directory write attempts on read-only file systems. |
| 1681 | */ |
| 1682 | if (rdonly && |
| 1683 | (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { |
| 1684 | error = EROFS; |
| 1685 | goto bad2; |
| 1686 | } |
| 1687 | /* ASSERT(dvp == ndp->ni_startdir) */ |
| 1688 | |
| 1689 | return (0); |
| 1690 | |
| 1691 | bad2: |
| 1692 | vnode_put(dp); |
| 1693 | bad: |
| 1694 | *vpp = NULL; |
| 1695 | |
| 1696 | return (error); |
| 1697 | } |
| 1698 | |
| 1699 | /* |
| 1700 | * Free pathname buffer |
| 1701 | */ |
| 1702 | void |
| 1703 | nameidone(struct nameidata *ndp) |
| 1704 | { |
| 1705 | if (ndp->ni_cnd.cn_flags & HASBUF) { |
| 1706 | char *tmp = ndp->ni_cnd.cn_pnbuf; |
| 1707 | |
| 1708 | ndp->ni_cnd.cn_pnbuf = NULL; |
| 1709 | ndp->ni_cnd.cn_flags &= ~HASBUF; |
| 1710 | FREE_ZONE(tmp, ndp->ni_cnd.cn_pnlen, M_NAMEI); |
| 1711 | } |
| 1712 | } |
| 1713 | |
| 1714 | |
| 1715 | /* |
| 1716 | * Log (part of) a pathname using the KERNEL_DEBUG_CONSTANT mechanism, as used |
| 1717 | * by fs_usage. The path up to and including the current component name are |
| 1718 | * logged. Up to NUMPARMS*4 bytes of pathname will be logged. If the path |
| 1719 | * to be logged is longer than that, then the last NUMPARMS*4 bytes are logged. |
| 1720 | * That is, the truncation removes the leading portion of the path. |
| 1721 | * |
| 1722 | * The logging is done via multiple KERNEL_DEBUG_CONSTANT calls. The first one |
| 1723 | * is marked with DBG_FUNC_START. The last one is marked with DBG_FUNC_END |
| 1724 | * (in addition to DBG_FUNC_START if it is also the first). There may be |
| 1725 | * intermediate ones with neither DBG_FUNC_START nor DBG_FUNC_END. |
| 1726 | * |
| 1727 | * The first KERNEL_DEBUG_CONSTANT passes the vnode pointer and 12 bytes of |
| 1728 | * pathname. The remaining KERNEL_DEBUG_CONSTANT calls add 16 bytes of pathname |
| 1729 | * each. The minimum number of KERNEL_DEBUG_CONSTANT calls required to pass |
| 1730 | * the path are used. Any excess padding in the final KERNEL_DEBUG_CONSTANT |
| 1731 | * (because not all of the 12 or 16 bytes are needed for the remainder of the |
| 1732 | * path) is set to zero bytes, or '>' if there is more path beyond the |
| 1733 | * current component name (usually because an intermediate component was not |
| 1734 | * found). |
| 1735 | * |
| 1736 | * NOTE: If the path length is greater than NUMPARMS*4, or is not of the form |
| 1737 | * 12+N*16, there will be no padding. |
| 1738 | * |
| 1739 | * TODO: If there is more path beyond the current component name, should we |
| 1740 | * force some padding? For example, a lookup for /foo_bar_baz/spam that |
| 1741 | * fails because /foo_bar_baz is not found will only log "/foo_bar_baz", with |
| 1742 | * no '>' padding. But /foo_bar/spam would log "/foo_bar>>>>". |
| 1743 | */ |
| 1744 | #if (KDEBUG_LEVEL >= KDEBUG_LEVEL_IST) |
| 1745 | |
| 1746 | void |
| 1747 | kdebug_vfs_lookup(long *dbg_parms, int dbg_namelen, void *dp, uint32_t flags) |
| 1748 | { |
| 1749 | int code; |
| 1750 | unsigned int i; |
| 1751 | bool lookup = flags & KDBG_VFS_LOOKUP_FLAG_LOOKUP; |
| 1752 | bool noprocfilt = flags & KDBG_VFS_LOOKUP_FLAG_NOPROCFILT; |
| 1753 | |
| 1754 | /* |
| 1755 | * In the event that we collect multiple, consecutive pathname |
| 1756 | * entries, we must mark the start of the path's string and the end. |
| 1757 | */ |
| 1758 | if (lookup) { |
| 1759 | code = VFS_LOOKUP | DBG_FUNC_START; |
| 1760 | } else { |
| 1761 | code = VFS_LOOKUP_DONE | DBG_FUNC_START; |
| 1762 | } |
| 1763 | |
| 1764 | if (dbg_namelen <= (int)(3 * sizeof(long))) |
| 1765 | code |= DBG_FUNC_END; |
| 1766 | |
| 1767 | if (noprocfilt) { |
| 1768 | KDBG_RELEASE_NOPROCFILT(code, kdebug_vnode(dp), dbg_parms[0], |
| 1769 | dbg_parms[1], dbg_parms[2]); |
| 1770 | } else { |
| 1771 | KDBG_RELEASE(code, kdebug_vnode(dp), dbg_parms[0], dbg_parms[1], |
| 1772 | dbg_parms[2]); |
| 1773 | } |
| 1774 | |
| 1775 | code &= ~DBG_FUNC_START; |
| 1776 | |
| 1777 | for (i=3, dbg_namelen -= (3 * sizeof(long)); dbg_namelen > 0; i+=4, dbg_namelen -= (4 * sizeof(long))) { |
| 1778 | if (dbg_namelen <= (int)(4 * sizeof(long))) |
| 1779 | code |= DBG_FUNC_END; |
| 1780 | |
| 1781 | if (noprocfilt) { |
| 1782 | KDBG_RELEASE_NOPROCFILT(code, dbg_parms[i], dbg_parms[i + 1], |
| 1783 | dbg_parms[i + 2], dbg_parms[i + 3]); |
| 1784 | } else { |
| 1785 | KDBG_RELEASE(code, dbg_parms[i], dbg_parms[i + 1], dbg_parms[i + 2], |
| 1786 | dbg_parms[i + 3]); |
| 1787 | } |
| 1788 | } |
| 1789 | } |
| 1790 | |
| 1791 | void |
| 1792 | kdebug_lookup_gen_events(long *dbg_parms, int dbg_namelen, void *dp, |
| 1793 | boolean_t lookup) |
| 1794 | { |
| 1795 | kdebug_vfs_lookup(dbg_parms, dbg_namelen, dp, |
| 1796 | lookup ? KDBG_VFS_LOOKUP_FLAG_LOOKUP : 0); |
| 1797 | } |
| 1798 | |
| 1799 | void |
| 1800 | kdebug_lookup(vnode_t dp, struct componentname *cnp) |
| 1801 | { |
| 1802 | int dbg_namelen; |
| 1803 | char *dbg_nameptr; |
| 1804 | long dbg_parms[NUMPARMS]; |
| 1805 | |
| 1806 | /* Collect the pathname for tracing */ |
| 1807 | dbg_namelen = (cnp->cn_nameptr - cnp->cn_pnbuf) + cnp->cn_namelen; |
| 1808 | dbg_nameptr = cnp->cn_nameptr + cnp->cn_namelen; |
| 1809 | |
| 1810 | if (dbg_namelen > (int)sizeof(dbg_parms)) |
| 1811 | dbg_namelen = sizeof(dbg_parms); |
| 1812 | dbg_nameptr -= dbg_namelen; |
| 1813 | |
| 1814 | /* Copy the (possibly truncated) path itself */ |
| 1815 | memcpy(dbg_parms, dbg_nameptr, dbg_namelen); |
| 1816 | |
| 1817 | /* Pad with '\0' or '>' */ |
| 1818 | if (dbg_namelen < (int)sizeof(dbg_parms)) { |
| 1819 | memset((char *)dbg_parms + dbg_namelen, |
| 1820 | *(cnp->cn_nameptr + cnp->cn_namelen) ? '>' : 0, |
| 1821 | sizeof(dbg_parms) - dbg_namelen); |
| 1822 | } |
| 1823 | kdebug_vfs_lookup(dbg_parms, dbg_namelen, (void *)dp, |
| 1824 | KDBG_VFS_LOOKUP_FLAG_LOOKUP); |
| 1825 | } |
| 1826 | |
| 1827 | #else /* (KDEBUG_LEVEL >= KDEBUG_LEVEL_IST) */ |
| 1828 | |
| 1829 | void |
| 1830 | kdebug_vfs_lookup(long *dbg_parms __unused, int dbg_namelen __unused, |
| 1831 | void *dp __unused, __unused uint32_t flags) |
| 1832 | { |
| 1833 | } |
| 1834 | |
| 1835 | static void |
| 1836 | kdebug_lookup(struct vnode *dp __unused, struct componentname *cnp __unused) |
| 1837 | { |
| 1838 | } |
| 1839 | #endif /* (KDEBUG_LEVEL >= KDEBUG_LEVEL_IST) */ |
| 1840 | |
| 1841 | int |
| 1842 | vfs_getbyid(fsid_t *fsid, ino64_t ino, vnode_t *vpp, vfs_context_t ctx) |
| 1843 | { |
| 1844 | mount_t mp; |
| 1845 | int error; |
| 1846 | |
| 1847 | mp = mount_lookupby_volfsid(fsid->val[0], 1); |
| 1848 | if (mp == NULL) { |
| 1849 | return EINVAL; |
| 1850 | } |
| 1851 | |
| 1852 | /* Get the target vnode. */ |
| 1853 | if (ino == 2) { |
| 1854 | error = VFS_ROOT(mp, vpp, ctx); |
| 1855 | } else { |
| 1856 | error = VFS_VGET(mp, ino, vpp, ctx); |
| 1857 | } |
| 1858 | |
| 1859 | vfs_unbusy(mp); |
| 1860 | return error; |
| 1861 | } |
| 1862 | /* |
| 1863 | * Obtain the real path from a legacy volfs style path. |
| 1864 | * |
| 1865 | * Valid formats of input path: |
| 1866 | * |
| 1867 | * "555/@" |
| 1868 | * "555/2" |
| 1869 | * "555/123456" |
| 1870 | * "555/123456/foobar" |
| 1871 | * |
| 1872 | * Where: |
| 1873 | * 555 represents the volfs file system id |
| 1874 | * '@' and '2' are aliases to the root of a file system |
| 1875 | * 123456 represents a file id |
| 1876 | * "foobar" represents a file name |
| 1877 | */ |
| 1878 | #if CONFIG_VOLFS |
| 1879 | static int |
| 1880 | vfs_getrealpath(const char * path, char * realpath, size_t bufsize, vfs_context_t ctx) |
| 1881 | { |
| 1882 | vnode_t vp; |
| 1883 | struct mount *mp = NULL; |
| 1884 | char *str; |
| 1885 | char ch; |
| 1886 | uint32_t id; |
| 1887 | ino64_t ino; |
| 1888 | int error; |
| 1889 | int length; |
| 1890 | |
| 1891 | /* Get file system id and move str to next component. */ |
| 1892 | id = strtoul(path, &str, 10); |
| 1893 | if (id == 0 || str[0] != '/') { |
| 1894 | return (EINVAL); |
| 1895 | } |
| 1896 | while (*str == '/') { |
| 1897 | str++; |
| 1898 | } |
| 1899 | ch = *str; |
| 1900 | |
| 1901 | mp = mount_lookupby_volfsid(id, 1); |
| 1902 | if (mp == NULL) { |
| 1903 | return (EINVAL); /* unexpected failure */ |
| 1904 | } |
| 1905 | /* Check for an alias to a file system root. */ |
| 1906 | if (ch == '@' && str[1] == '\0') { |
| 1907 | ino = 2; |
| 1908 | str++; |
| 1909 | } else { |
| 1910 | /* Get file id and move str to next component. */ |
| 1911 | ino = strtouq(str, &str, 10); |
| 1912 | } |
| 1913 | |
| 1914 | /* Get the target vnode. */ |
| 1915 | if (ino == 2) { |
| 1916 | error = VFS_ROOT(mp, &vp, ctx); |
| 1917 | } else { |
| 1918 | error = VFS_VGET(mp, ino, &vp, ctx); |
| 1919 | } |
| 1920 | vfs_unbusy(mp); |
| 1921 | if (error) { |
| 1922 | goto out; |
| 1923 | } |
| 1924 | realpath[0] = '\0'; |
| 1925 | |
| 1926 | /* Get the absolute path to this vnode. */ |
| 1927 | error = build_path(vp, realpath, bufsize, &length, 0, ctx); |
| 1928 | vnode_put(vp); |
| 1929 | |
| 1930 | if (error == 0 && *str != '\0') { |
| 1931 | int attempt = strlcat(realpath, str, MAXPATHLEN); |
| 1932 | if (attempt > MAXPATHLEN){ |
| 1933 | error = ENAMETOOLONG; |
| 1934 | } |
| 1935 | } |
| 1936 | out: |
| 1937 | return (error); |
| 1938 | } |
| 1939 | #endif |
| 1940 | |
| 1941 | void |
| 1942 | lookup_compound_vnop_post_hook(int error, vnode_t dvp, vnode_t vp, struct nameidata *ndp, int did_create) |
| 1943 | { |
| 1944 | if (error == 0 && vp == NULLVP) { |
| 1945 | panic("NULL vp with error == 0.\n" ); |
| 1946 | } |
| 1947 | |
| 1948 | /* |
| 1949 | * We don't want to do any of this if we didn't use the compound vnop |
| 1950 | * to perform the lookup... i.e. if we're allowing and using the legacy pattern, |
| 1951 | * where we did a full lookup. |
| 1952 | */ |
| 1953 | if ((ndp->ni_flag & NAMEI_COMPOUND_OP_MASK) == 0) { |
| 1954 | return; |
| 1955 | } |
| 1956 | |
| 1957 | /* |
| 1958 | * If we're going to continue the lookup, we'll handle |
| 1959 | * all lookup-related updates at that time. |
| 1960 | */ |
| 1961 | if (error == EKEEPLOOKING) { |
| 1962 | return; |
| 1963 | } |
| 1964 | |
| 1965 | /* |
| 1966 | * Only audit or update cache for *found* vnodes. For creation |
| 1967 | * neither would happen in the non-compound-vnop case. |
| 1968 | */ |
| 1969 | if ((vp != NULLVP) && !did_create) { |
| 1970 | /* |
| 1971 | * If MAKEENTRY isn't set, and we've done a successful compound VNOP, |
| 1972 | * then we certainly don't want to update cache or identity. |
| 1973 | */ |
| 1974 | if ((error != 0) || (ndp->ni_cnd.cn_flags & MAKEENTRY)) { |
| 1975 | lookup_consider_update_cache(dvp, vp, &ndp->ni_cnd, ndp->ni_ncgeneration); |
| 1976 | } |
| 1977 | if (ndp->ni_cnd.cn_flags & AUDITVNPATH1) |
| 1978 | AUDIT_ARG(vnpath, vp, ARG_VNODE1); |
| 1979 | else if (ndp->ni_cnd.cn_flags & AUDITVNPATH2) |
| 1980 | AUDIT_ARG(vnpath, vp, ARG_VNODE2); |
| 1981 | } |
| 1982 | |
| 1983 | /* |
| 1984 | * If you created (whether you opened or not), cut a lookup tracepoint |
| 1985 | * for the parent dir (as would happen without a compound vnop). Note: we may need |
| 1986 | * a vnode despite failure in this case! |
| 1987 | * |
| 1988 | * If you did not create: |
| 1989 | * Found child (succeeded or not): cut a tracepoint for the child. |
| 1990 | * Did not find child: cut a tracepoint with the parent. |
| 1991 | */ |
| 1992 | if (kdebug_enable) { |
| 1993 | kdebug_lookup(vp ? vp : dvp, &ndp->ni_cnd); |
| 1994 | } |
| 1995 | } |
| 1996 | |