1 /* 2 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 /* 35 * nlookup() is the 'new' namei interface. Rather then return directory and 36 * leaf vnodes (in various lock states) the new interface instead deals in 37 * namecache records. Namecache records may represent both a positive or 38 * a negative hit. The namespace is locked via the namecache record instead 39 * of via the vnode, and only the leaf namecache record (representing the 40 * filename) needs to be locked. 41 * 42 * This greatly improves filesystem parallelism and is a huge simplification 43 * of the API verses the old vnode locking / namei scheme. 44 * 45 * Filesystems must actively control the caching aspects of the namecache, 46 * and since namecache pointers are used as handles they are non-optional 47 * even for filesystems which do not generally wish to cache things. It is 48 * intended that a separate cache coherency API will be constructed to handle 49 * these issues. 50 */ 51 52 #include "opt_ktrace.h" 53 54 #include <sys/param.h> 55 #include <sys/systm.h> 56 #include <sys/kernel.h> 57 #include <sys/vnode.h> 58 #include <sys/mount.h> 59 #include <sys/filedesc.h> 60 #include <sys/proc.h> 61 #include <sys/namei.h> 62 #include <sys/nlookup.h> 63 #include <sys/malloc.h> 64 #include <sys/stat.h> 65 #include <sys/objcache.h> 66 #include <sys/file.h> 67 68 #ifdef KTRACE 69 #include <sys/ktrace.h> 70 #endif 71 72 static int naccess(struct nchandle *nch, int vmode, struct ucred *cred, 73 int *stickyp); 74 75 /* 76 * Initialize a nlookup() structure, early error return for copyin faults 77 * or a degenerate empty string (which is not allowed). 78 * 79 * The first process proc0's credentials are used if the calling thread 80 * is not associated with a process context. 81 * 82 * MPSAFE 83 */ 84 int 85 nlookup_init(struct nlookupdata *nd, 86 const char *path, enum uio_seg seg, int flags) 87 { 88 size_t pathlen; 89 struct proc *p; 90 thread_t td; 91 int error; 92 93 td = curthread; 94 p = td->td_proc; 95 96 /* 97 * note: the pathlen set by copy*str() includes the terminating \0. 98 */ 99 bzero(nd, sizeof(struct nlookupdata)); 100 nd->nl_path = objcache_get(namei_oc, M_WAITOK); 101 nd->nl_flags |= NLC_HASBUF; 102 if (seg == UIO_SYSSPACE) 103 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen); 104 else 105 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen); 106 107 /* 108 * Don't allow empty pathnames. 109 * POSIX.1 requirement: "" is not a vaild file name. 110 */ 111 if (error == 0 && pathlen <= 1) 112 error = ENOENT; 113 114 if (error == 0) { 115 if (p && p->p_fd) { 116 cache_copy(&p->p_fd->fd_ncdir, &nd->nl_nch); 117 cache_copy(&p->p_fd->fd_nrdir, &nd->nl_rootnch); 118 if (p->p_fd->fd_njdir.ncp) 119 cache_copy(&p->p_fd->fd_njdir, &nd->nl_jailnch); 120 nd->nl_cred = crhold(p->p_ucred); 121 } else { 122 cache_copy(&rootnch, &nd->nl_nch); 123 cache_copy(&nd->nl_nch, &nd->nl_rootnch); 124 cache_copy(&nd->nl_nch, &nd->nl_jailnch); 125 nd->nl_cred = crhold(proc0.p_ucred); 126 } 127 nd->nl_td = td; 128 nd->nl_flags |= flags; 129 } else { 130 nlookup_done(nd); 131 } 132 return(error); 133 } 134 135 136 /* 137 * nlookup_init() for "at" family of syscalls. 138 * 139 * Works similarly to nlookup_init() but if path is relative and fd is not 140 * AT_FDCWD, path is interpreted relative to the directory pointed to by fd. 141 * In this case, the file entry pointed to by fd is ref'ed and returned in 142 * *fpp. 143 * 144 * If the call succeeds, nlookup_done_at() must be called to clean-up the nd 145 * and release the ref to the file entry. 146 */ 147 int 148 nlookup_init_at(struct nlookupdata *nd, struct file **fpp, int fd, 149 const char *path, enum uio_seg seg, int flags) 150 { 151 struct thread *td = curthread; 152 struct proc *p = td->td_proc; 153 struct file* fp; 154 struct vnode *vp; 155 int error; 156 157 *fpp = NULL; 158 159 if ((error = nlookup_init(nd, path, seg, flags)) != 0) { 160 return (error); 161 } 162 163 if (nd->nl_path[0] != '/' && fd != AT_FDCWD) { 164 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 165 goto done; 166 vp = (struct vnode*)fp->f_data; 167 if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL) { 168 fdrop(fp); 169 fp = NULL; 170 error = ENOTDIR; 171 goto done; 172 } 173 cache_drop(&nd->nl_nch); 174 cache_copy(&fp->f_nchandle, &nd->nl_nch); 175 *fpp = fp; 176 } 177 178 179 done: 180 if (error) 181 nlookup_done(nd); 182 return (error); 183 184 } 185 186 /* 187 * This works similarly to nlookup_init() but does not assume a process 188 * context. rootnch is always chosen for the root directory and the cred 189 * and starting directory are supplied in arguments. 190 */ 191 int 192 nlookup_init_raw(struct nlookupdata *nd, 193 const char *path, enum uio_seg seg, int flags, 194 struct ucred *cred, struct nchandle *ncstart) 195 { 196 size_t pathlen; 197 thread_t td; 198 int error; 199 200 td = curthread; 201 202 bzero(nd, sizeof(struct nlookupdata)); 203 nd->nl_path = objcache_get(namei_oc, M_WAITOK); 204 nd->nl_flags |= NLC_HASBUF; 205 if (seg == UIO_SYSSPACE) 206 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen); 207 else 208 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen); 209 210 /* 211 * Don't allow empty pathnames. 212 * POSIX.1 requirement: "" is not a vaild file name. 213 */ 214 if (error == 0 && pathlen <= 1) 215 error = ENOENT; 216 217 if (error == 0) { 218 cache_copy(ncstart, &nd->nl_nch); 219 cache_copy(&rootnch, &nd->nl_rootnch); 220 cache_copy(&rootnch, &nd->nl_jailnch); 221 nd->nl_cred = crhold(cred); 222 nd->nl_td = td; 223 nd->nl_flags |= flags; 224 } else { 225 nlookup_done(nd); 226 } 227 return(error); 228 } 229 230 /* 231 * This works similarly to nlookup_init_raw() but does not rely 232 * on rootnch being initialized yet. 233 */ 234 int 235 nlookup_init_root(struct nlookupdata *nd, 236 const char *path, enum uio_seg seg, int flags, 237 struct ucred *cred, struct nchandle *ncstart, 238 struct nchandle *ncroot) 239 { 240 size_t pathlen; 241 thread_t td; 242 int error; 243 244 td = curthread; 245 246 bzero(nd, sizeof(struct nlookupdata)); 247 nd->nl_path = objcache_get(namei_oc, M_WAITOK); 248 nd->nl_flags |= NLC_HASBUF; 249 if (seg == UIO_SYSSPACE) 250 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen); 251 else 252 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen); 253 254 /* 255 * Don't allow empty pathnames. 256 * POSIX.1 requirement: "" is not a vaild file name. 257 */ 258 if (error == 0 && pathlen <= 1) 259 error = ENOENT; 260 261 if (error == 0) { 262 cache_copy(ncstart, &nd->nl_nch); 263 cache_copy(ncroot, &nd->nl_rootnch); 264 cache_copy(ncroot, &nd->nl_jailnch); 265 nd->nl_cred = crhold(cred); 266 nd->nl_td = td; 267 nd->nl_flags |= flags; 268 } else { 269 nlookup_done(nd); 270 } 271 return(error); 272 } 273 274 /* 275 * Set a different credential; this credential will be used by future 276 * operations performed on nd.nl_open_vp and nlookupdata structure. 277 */ 278 void 279 nlookup_set_cred(struct nlookupdata *nd, struct ucred *cred) 280 { 281 KKASSERT(nd->nl_cred != NULL); 282 283 if (nd->nl_cred != cred) { 284 cred = crhold(cred); 285 crfree(nd->nl_cred); 286 nd->nl_cred = cred; 287 } 288 } 289 290 /* 291 * Cleanup a nlookupdata structure after we are through with it. This may 292 * be called on any nlookupdata structure initialized with nlookup_init(). 293 * Calling nlookup_done() is mandatory in all cases except where nlookup_init() 294 * returns an error, even if as a consumer you believe you have taken all 295 * dynamic elements out of the nlookupdata structure. 296 */ 297 void 298 nlookup_done(struct nlookupdata *nd) 299 { 300 if (nd->nl_nch.ncp) { 301 if (nd->nl_flags & NLC_NCPISLOCKED) { 302 nd->nl_flags &= ~NLC_NCPISLOCKED; 303 cache_unlock(&nd->nl_nch); 304 } 305 cache_drop(&nd->nl_nch); /* NULL's out the nch */ 306 } 307 if (nd->nl_rootnch.ncp) 308 cache_drop(&nd->nl_rootnch); 309 if (nd->nl_jailnch.ncp) 310 cache_drop(&nd->nl_jailnch); 311 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) { 312 objcache_put(namei_oc, nd->nl_path); 313 nd->nl_path = NULL; 314 } 315 if (nd->nl_cred) { 316 crfree(nd->nl_cred); 317 nd->nl_cred = NULL; 318 } 319 if (nd->nl_open_vp) { 320 if (nd->nl_flags & NLC_LOCKVP) { 321 vn_unlock(nd->nl_open_vp); 322 nd->nl_flags &= ~NLC_LOCKVP; 323 } 324 vn_close(nd->nl_open_vp, nd->nl_vp_fmode); 325 nd->nl_open_vp = NULL; 326 } 327 if (nd->nl_dvp) { 328 vrele(nd->nl_dvp); 329 nd->nl_dvp = NULL; 330 } 331 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */ 332 } 333 334 /* 335 * Works similarly to nlookup_done() when nd initialized with 336 * nlookup_init_at(). 337 */ 338 void 339 nlookup_done_at(struct nlookupdata *nd, struct file *fp) 340 { 341 nlookup_done(nd); 342 if (fp != NULL) 343 fdrop(fp); 344 } 345 346 void 347 nlookup_zero(struct nlookupdata *nd) 348 { 349 bzero(nd, sizeof(struct nlookupdata)); 350 } 351 352 /* 353 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL 354 * if an error occured. 355 * 356 * Note that the returned ncp is not checked for permissions, though VEXEC 357 * is checked on the directory path leading up to the result. The caller 358 * must call naccess() to check the permissions of the returned leaf. 359 */ 360 struct nchandle 361 nlookup_simple(const char *str, enum uio_seg seg, 362 int niflags, int *error) 363 { 364 struct nlookupdata nd; 365 struct nchandle nch; 366 367 *error = nlookup_init(&nd, str, seg, niflags); 368 if (*error == 0) { 369 if ((*error = nlookup(&nd)) == 0) { 370 nch = nd.nl_nch; /* keep hold ref from structure */ 371 cache_zero(&nd.nl_nch); /* and NULL out */ 372 } else { 373 cache_zero(&nch); 374 } 375 nlookup_done(&nd); 376 } else { 377 cache_zero(&nch); 378 } 379 return(nch); 380 } 381 382 /* 383 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d 384 * on return, even if an error occurs. If no error occurs the returned 385 * nl_nch is always referenced and locked, otherwise it may or may not be. 386 * 387 * Intermediate directory elements, including the current directory, require 388 * execute (search) permission. nlookup does not examine the access 389 * permissions on the returned element. 390 * 391 * If NLC_CREATE is set the last directory must allow node creation, 392 * and an error code of 0 will be returned for a non-existant 393 * target (not ENOENT). 394 * 395 * If NLC_RENAME_DST is set the last directory mut allow node deletion, 396 * plus the sticky check is made, and an error code of 0 will be returned 397 * for a non-existant target (not ENOENT). 398 * 399 * If NLC_DELETE is set the last directory mut allow node deletion, 400 * plus the sticky check is made. 401 * 402 * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode 403 * of the returned entry. The vnode will be referenced, but not locked, 404 * and will be released by nlookup_done() along with everything else. 405 */ 406 int 407 nlookup(struct nlookupdata *nd) 408 { 409 globaldata_t gd = mycpu; 410 struct nlcomponent nlc; 411 struct nchandle nch; 412 struct nchandle par; 413 struct nchandle nctmp; 414 struct mount *mp; 415 int wasdotordotdot; 416 char *ptr; 417 char *xptr; 418 int error; 419 int len; 420 int dflags; 421 int hit = 1; 422 423 #ifdef KTRACE 424 if (KTRPOINT(nd->nl_td, KTR_NAMEI)) 425 ktrnamei(nd->nl_td->td_lwp, nd->nl_path); 426 #endif 427 bzero(&nlc, sizeof(nlc)); 428 429 /* 430 * Setup for the loop. The current working namecache element is 431 * always at least referenced. We lock it as required, but always 432 * return a locked, resolved namecache entry. 433 */ 434 nd->nl_loopcnt = 0; 435 if (nd->nl_dvp) { 436 vrele(nd->nl_dvp); 437 nd->nl_dvp = NULL; 438 } 439 ptr = nd->nl_path; 440 441 /* 442 * Loop on the path components. At the top of the loop nd->nl_nch 443 * is ref'd and unlocked and represents our current position. 444 */ 445 for (;;) { 446 /* 447 * Make sure nl_nch is locked so we can access the vnode, resolution 448 * state, etc. 449 */ 450 if ((nd->nl_flags & NLC_NCPISLOCKED) == 0) { 451 nd->nl_flags |= NLC_NCPISLOCKED; 452 cache_lock(&nd->nl_nch); 453 } 454 455 /* 456 * Check if the root directory should replace the current 457 * directory. This is done at the start of a translation 458 * or after a symbolic link has been found. In other cases 459 * ptr will never be pointing at a '/'. 460 */ 461 if (*ptr == '/') { 462 do { 463 ++ptr; 464 } while (*ptr == '/'); 465 cache_get(&nd->nl_rootnch, &nch); 466 cache_put(&nd->nl_nch); 467 nd->nl_nch = nch; /* remains locked */ 468 469 /* 470 * Fast-track termination. There is no parent directory of 471 * the root in the same mount from the point of view of 472 * the caller so return EACCES if NLC_REFDVP is specified, 473 * and EEXIST if NLC_CREATE is also specified. 474 * e.g. 'rmdir /' or 'mkdir /' are not allowed. 475 */ 476 if (*ptr == 0) { 477 if (nd->nl_flags & NLC_REFDVP) 478 error = (nd->nl_flags & NLC_CREATE) ? EEXIST : EACCES; 479 else 480 error = 0; 481 break; 482 } 483 continue; 484 } 485 486 /* 487 * Check directory search permissions. 488 */ 489 dflags = 0; 490 error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags); 491 if (error) 492 break; 493 494 /* 495 * Extract the path component. Path components are limited to 496 * 255 characters. 497 */ 498 nlc.nlc_nameptr = ptr; 499 while (*ptr && *ptr != '/') 500 ++ptr; 501 nlc.nlc_namelen = ptr - nlc.nlc_nameptr; 502 if (nlc.nlc_namelen >= 256) { 503 error = ENAMETOOLONG; 504 break; 505 } 506 507 /* 508 * Lookup the path component in the cache, creating an unresolved 509 * entry if necessary. We have to handle "." and ".." as special 510 * cases. 511 * 512 * When handling ".." we have to detect a traversal back through a 513 * mount point. If we are at the root, ".." just returns the root. 514 * 515 * When handling "." or ".." we also have to recalculate dflags 516 * since our dflags will be for some sub-directory instead of the 517 * parent dir. 518 * 519 * This subsection returns a locked, refd 'nch' unless it errors out. 520 * The namecache topology is not allowed to be disconnected, so 521 * encountering a NULL parent will generate EINVAL. This typically 522 * occurs when a directory is removed out from under a process. 523 */ 524 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') { 525 cache_get(&nd->nl_nch, &nch); 526 wasdotordotdot = 1; 527 } else if (nlc.nlc_namelen == 2 && 528 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') { 529 if (nd->nl_nch.mount == nd->nl_rootnch.mount && 530 nd->nl_nch.ncp == nd->nl_rootnch.ncp 531 ) { 532 /* 533 * ".." at the root returns the root 534 */ 535 cache_get(&nd->nl_nch, &nch); 536 } else { 537 /* 538 * Locate the parent ncp. If we are at the root of a 539 * filesystem mount we have to skip to the mounted-on 540 * point in the underlying filesystem. 541 * 542 * Expect the parent to always be good since the 543 * mountpoint doesn't go away. XXX hack. cache_get() 544 * requires the ncp to already have a ref as a safety. 545 */ 546 nctmp = nd->nl_nch; 547 while (nctmp.ncp == nctmp.mount->mnt_ncmountpt.ncp) 548 nctmp = nctmp.mount->mnt_ncmounton; 549 nctmp.ncp = nctmp.ncp->nc_parent; 550 KKASSERT(nctmp.ncp != NULL); 551 cache_hold(&nctmp); 552 cache_get(&nctmp, &nch); 553 cache_drop(&nctmp); /* NOTE: zero's nctmp */ 554 } 555 wasdotordotdot = 2; 556 } else { 557 /* 558 * Must unlock nl_nch when traversing down the path. 559 */ 560 cache_unlock(&nd->nl_nch); 561 nd->nl_flags &= ~NLC_NCPISLOCKED; 562 nch = cache_nlookup(&nd->nl_nch, &nlc); 563 if (nch.ncp->nc_flag & NCF_UNRESOLVED) 564 hit = 0; 565 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) { 566 kprintf("[diagnostic] nlookup: relookup %*.*s\n", 567 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name); 568 cache_put(&nch); 569 nch = cache_nlookup(&nd->nl_nch, &nlc); 570 } 571 wasdotordotdot = 0; 572 } 573 574 /* 575 * If the last component was "." or ".." our dflags no longer 576 * represents the parent directory and we have to explicitly 577 * look it up. 578 * 579 * Expect the parent to be good since nch is locked. 580 */ 581 if (wasdotordotdot && error == 0) { 582 dflags = 0; 583 if ((par.ncp = nch.ncp->nc_parent) != NULL) { 584 par.mount = nch.mount; 585 cache_hold(&par); 586 cache_lock(&par); 587 error = naccess(&par, 0, nd->nl_cred, &dflags); 588 cache_put(&par); 589 } 590 } 591 if (nd->nl_flags & NLC_NCPISLOCKED) { 592 cache_unlock(&nd->nl_nch); 593 nd->nl_flags &= ~NLC_NCPISLOCKED; 594 } 595 596 /* 597 * [end of subsection] 598 * 599 * nch is locked and referenced. 600 * nd->nl_nch is unlocked and referenced. 601 * 602 * nl_nch must be unlocked or we could chain lock to the root 603 * if a resolve gets stuck (e.g. in NFS). 604 */ 605 606 /* 607 * Resolve the namespace if necessary. The ncp returned by 608 * cache_nlookup() is referenced and locked. 609 * 610 * XXX neither '.' nor '..' should return EAGAIN since they were 611 * previously resolved and thus cannot be newly created ncp's. 612 */ 613 if (nch.ncp->nc_flag & NCF_UNRESOLVED) { 614 hit = 0; 615 error = cache_resolve(&nch, nd->nl_cred); 616 KKASSERT(error != EAGAIN); 617 } else { 618 error = nch.ncp->nc_error; 619 } 620 621 /* 622 * Early completion. ENOENT is not an error if this is the last 623 * component and NLC_CREATE or NLC_RENAME (rename target) was 624 * requested. Note that ncp->nc_error is left as ENOENT in that 625 * case, which we check later on. 626 * 627 * Also handle invalid '.' or '..' components terminating a path 628 * for a create/rename/delete. The standard requires this and pax 629 * pretty stupidly depends on it. 630 */ 631 for (xptr = ptr; *xptr == '/'; ++xptr) 632 ; 633 if (*xptr == 0) { 634 if (error == ENOENT && 635 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST)) 636 ) { 637 if (nd->nl_flags & NLC_NFS_RDONLY) { 638 error = EROFS; 639 } else { 640 error = naccess(&nch, nd->nl_flags | dflags, 641 nd->nl_cred, NULL); 642 } 643 } 644 if (error == 0 && wasdotordotdot && 645 (nd->nl_flags & (NLC_CREATE | NLC_DELETE | 646 NLC_RENAME_SRC | NLC_RENAME_DST))) { 647 /* 648 * POSIX junk 649 */ 650 if (nd->nl_flags & NLC_CREATE) 651 error = EEXIST; 652 else if (nd->nl_flags & NLC_DELETE) 653 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY; 654 else 655 error = EINVAL; 656 } 657 } 658 659 /* 660 * Early completion on error. 661 */ 662 if (error) { 663 cache_put(&nch); 664 break; 665 } 666 667 /* 668 * If the element is a symlink and it is either not the last 669 * element or it is the last element and we are allowed to 670 * follow symlinks, resolve the symlink. 671 */ 672 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) && 673 (*ptr || (nd->nl_flags & NLC_FOLLOW)) 674 ) { 675 if (nd->nl_loopcnt++ >= MAXSYMLINKS) { 676 error = ELOOP; 677 cache_put(&nch); 678 break; 679 } 680 error = nreadsymlink(nd, &nch, &nlc); 681 cache_put(&nch); 682 if (error) 683 break; 684 685 /* 686 * Concatenate trailing path elements onto the returned symlink. 687 * Note that if the path component (ptr) is not exhausted, it 688 * will being with a '/', so we do not have to add another one. 689 * 690 * The symlink may not be empty. 691 */ 692 len = strlen(ptr); 693 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) { 694 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT; 695 objcache_put(namei_oc, nlc.nlc_nameptr); 696 break; 697 } 698 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1); 699 if (nd->nl_flags & NLC_HASBUF) 700 objcache_put(namei_oc, nd->nl_path); 701 nd->nl_path = nlc.nlc_nameptr; 702 nd->nl_flags |= NLC_HASBUF; 703 ptr = nd->nl_path; 704 705 /* 706 * Go back up to the top to resolve any initial '/'s in the 707 * symlink. 708 */ 709 continue; 710 } 711 712 /* 713 * If the element is a directory and we are crossing a mount point, 714 * Locate the mount. 715 */ 716 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && 717 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 && 718 (mp = cache_findmount(&nch)) != NULL 719 ) { 720 struct vnode *tdp; 721 722 cache_put(&nch); 723 cache_get(&mp->mnt_ncmountpt, &nch); 724 725 if (nch.ncp->nc_flag & NCF_UNRESOLVED) { 726 while (vfs_busy(mp, 0)) 727 ; 728 error = VFS_ROOT(mp, &tdp); 729 vfs_unbusy(mp); 730 if (error) 731 break; 732 cache_setvp(&nch, tdp); 733 vput(tdp); 734 } 735 } 736 if (error) { 737 cache_put(&nch); 738 break; 739 } 740 741 /* 742 * Skip any slashes to get to the next element. If there 743 * are any slashes at all the current element must be a 744 * directory or, in the create case, intended to become a directory. 745 * If it isn't we break without incrementing ptr and fall through 746 * to the failure case below. 747 */ 748 while (*ptr == '/') { 749 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 && 750 !(nd->nl_flags & NLC_WILLBEDIR) 751 ) { 752 break; 753 } 754 ++ptr; 755 } 756 757 /* 758 * Continuation case: additional elements and the current 759 * element is a directory. 760 */ 761 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) { 762 cache_drop(&nd->nl_nch); 763 cache_unlock(&nch); 764 KKASSERT((nd->nl_flags & NLC_NCPISLOCKED) == 0); 765 nd->nl_nch = nch; 766 continue; 767 } 768 769 /* 770 * Failure case: additional elements and the current element 771 * is not a directory 772 */ 773 if (*ptr) { 774 cache_put(&nch); 775 error = ENOTDIR; 776 break; 777 } 778 779 /* 780 * Successful lookup of last element. 781 * 782 * Check permissions if the target exists. If the target does not 783 * exist directory permissions were already tested in the early 784 * completion code above. 785 * 786 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY 787 * if the file is marked append-only, and NLC_STICKY if the directory 788 * containing the file is sticky. 789 */ 790 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) { 791 error = naccess(&nch, nd->nl_flags | dflags, 792 nd->nl_cred, NULL); 793 if (error) { 794 cache_put(&nch); 795 break; 796 } 797 } 798 799 /* 800 * Termination: no more elements. 801 * 802 * If NLC_REFDVP is set acquire a referenced parent dvp. 803 */ 804 if (nd->nl_flags & NLC_REFDVP) { 805 cache_lock(&nd->nl_nch); 806 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp); 807 cache_unlock(&nd->nl_nch); 808 if (error) { 809 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp); 810 cache_put(&nch); 811 break; 812 } 813 } 814 cache_drop(&nd->nl_nch); 815 nd->nl_nch = nch; 816 nd->nl_flags |= NLC_NCPISLOCKED; 817 error = 0; 818 break; 819 } 820 821 if (hit) 822 ++gd->gd_nchstats->ncs_longhits; 823 else 824 ++gd->gd_nchstats->ncs_longmiss; 825 826 /* 827 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit 828 * (ncp->nc_error will be ENOENT), but we will still return an error 829 * code of 0. 830 */ 831 return(error); 832 } 833 834 /* 835 * Resolve a mount point's glue ncp. This ncp connects creates the illusion 836 * of continuity in the namecache tree by connecting the ncp related to the 837 * vnode under the mount to the ncp related to the mount's root vnode. 838 * 839 * If no error occured a locked, ref'd ncp is stored in *ncpp. 840 */ 841 int 842 nlookup_mp(struct mount *mp, struct nchandle *nch) 843 { 844 struct vnode *vp; 845 int error; 846 847 error = 0; 848 cache_get(&mp->mnt_ncmountpt, nch); 849 if (nch->ncp->nc_flag & NCF_UNRESOLVED) { 850 while (vfs_busy(mp, 0)) 851 ; 852 error = VFS_ROOT(mp, &vp); 853 vfs_unbusy(mp); 854 if (error) { 855 cache_put(nch); 856 } else { 857 cache_setvp(nch, vp); 858 vput(vp); 859 } 860 } 861 return(error); 862 } 863 864 /* 865 * Read the contents of a symlink, allocate a path buffer out of the 866 * namei_oc and initialize the supplied nlcomponent with the result. 867 * 868 * If an error occurs no buffer will be allocated or returned in the nlc. 869 */ 870 int 871 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch, 872 struct nlcomponent *nlc) 873 { 874 struct vnode *vp; 875 struct iovec aiov; 876 struct uio auio; 877 int linklen; 878 int error; 879 char *cp; 880 881 nlc->nlc_nameptr = NULL; 882 nlc->nlc_namelen = 0; 883 if (nch->ncp->nc_vp == NULL) 884 return(ENOENT); 885 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0) 886 return(error); 887 cp = objcache_get(namei_oc, M_WAITOK); 888 aiov.iov_base = cp; 889 aiov.iov_len = MAXPATHLEN; 890 auio.uio_iov = &aiov; 891 auio.uio_iovcnt = 1; 892 auio.uio_offset = 0; 893 auio.uio_rw = UIO_READ; 894 auio.uio_segflg = UIO_SYSSPACE; 895 auio.uio_td = nd->nl_td; 896 auio.uio_resid = MAXPATHLEN - 1; 897 error = VOP_READLINK(vp, &auio, nd->nl_cred); 898 if (error) 899 goto fail; 900 linklen = MAXPATHLEN - 1 - auio.uio_resid; 901 if (varsym_enable) { 902 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1); 903 if (linklen < 0) { 904 error = ENAMETOOLONG; 905 goto fail; 906 } 907 } 908 cp[linklen] = 0; 909 nlc->nlc_nameptr = cp; 910 nlc->nlc_namelen = linklen; 911 vput(vp); 912 return(0); 913 fail: 914 objcache_put(namei_oc, cp); 915 vput(vp); 916 return(error); 917 } 918 919 /* 920 * Check access [XXX cache vattr!] [XXX quota] 921 * 922 * Generally check the NLC_* access bits. All specified bits must pass 923 * for this function to return 0. 924 * 925 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST 926 * access, otherwise it must exist. No error is returned in this case. 927 * 928 * The file must not exist if NLC_EXCL is specified. 929 * 930 * Directory permissions in general are tested for NLC_CREATE if the file 931 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST 932 * whether the file exists or not. 933 * 934 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST, 935 * the latter is only tested if the target exists. 936 * 937 * The passed ncp must be referenced and locked. 938 */ 939 static int 940 naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp) 941 { 942 struct vnode *vp; 943 struct vattr va; 944 struct namecache *ncp; 945 int error; 946 int cflags; 947 948 ASSERT_NCH_LOCKED(nch); 949 ncp = nch->ncp; 950 if (ncp->nc_flag & NCF_UNRESOLVED) { 951 cache_resolve(nch, cred); 952 ncp = nch->ncp; 953 } 954 error = ncp->nc_error; 955 956 /* 957 * Directory permissions checks. Silently ignore ENOENT if these 958 * tests pass. It isn't an error. 959 * 960 * We can safely resolve ncp->nc_parent because ncp is currently 961 * locked. 962 */ 963 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) { 964 if (((nflags & NLC_CREATE) && ncp->nc_vp == NULL) || 965 ((nflags & NLC_DELETE) && ncp->nc_vp != NULL) || 966 ((nflags & NLC_RENAME_SRC) && ncp->nc_vp != NULL) || 967 (nflags & NLC_RENAME_DST) 968 ) { 969 struct nchandle par; 970 971 if ((par.ncp = ncp->nc_parent) == NULL) { 972 if (error != EAGAIN) 973 error = EINVAL; 974 } else if (error == 0 || error == ENOENT) { 975 par.mount = nch->mount; 976 cache_hold(&par); 977 cache_lock(&par); 978 error = naccess(&par, NLC_WRITE, cred, NULL); 979 cache_put(&par); 980 } 981 } 982 } 983 984 /* 985 * NLC_EXCL check. Target file must not exist. 986 */ 987 if (error == 0 && (nflags & NLC_EXCL) && ncp->nc_vp != NULL) 988 error = EEXIST; 989 990 /* 991 * Get the vnode attributes so we can do the rest of our checks. 992 * 993 * NOTE: We only call naccess_va() if the target exists. 994 */ 995 if (error == 0) { 996 error = cache_vget(nch, cred, LK_SHARED, &vp); 997 if (error == ENOENT) { 998 /* 999 * Silently zero-out ENOENT if creating or renaming 1000 * (rename target). It isn't an error. 1001 */ 1002 if (nflags & (NLC_CREATE | NLC_RENAME_DST)) 1003 error = 0; 1004 } else if (error == 0) { 1005 /* 1006 * Get the vnode attributes and check for illegal O_TRUNC 1007 * requests and read-only mounts. 1008 * 1009 * NOTE: You can still open devices on read-only mounts for 1010 * writing. 1011 * 1012 * NOTE: creates/deletes/renames are handled by the NLC_WRITE 1013 * check on the parent directory above. 1014 * 1015 * XXX cache the va in the namecache or in the vnode 1016 */ 1017 error = VOP_GETATTR(vp, &va); 1018 if (error == 0 && (nflags & NLC_TRUNCATE)) { 1019 switch(va.va_type) { 1020 case VREG: 1021 case VDATABASE: 1022 case VCHR: 1023 case VBLK: 1024 case VFIFO: 1025 break; 1026 case VDIR: 1027 error = EISDIR; 1028 break; 1029 default: 1030 error = EINVAL; 1031 break; 1032 } 1033 } 1034 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount && 1035 (vp->v_mount->mnt_flag & MNT_RDONLY) 1036 ) { 1037 switch(va.va_type) { 1038 case VDIR: 1039 case VLNK: 1040 case VREG: 1041 case VDATABASE: 1042 error = EROFS; 1043 break; 1044 default: 1045 break; 1046 } 1047 } 1048 vput(vp); 1049 1050 /* 1051 * Check permissions based on file attributes. The passed 1052 * flags (*nflagsp) are modified with feedback based on 1053 * special attributes and requirements. 1054 */ 1055 if (error == 0) { 1056 /* 1057 * Adjust the returned (*nflagsp) if non-NULL. 1058 */ 1059 if (nflagsp) { 1060 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid) 1061 *nflagsp |= NLC_STICKY; 1062 if (va.va_flags & APPEND) 1063 *nflagsp |= NLC_APPENDONLY; 1064 if (va.va_flags & IMMUTABLE) 1065 *nflagsp |= NLC_IMMUTABLE; 1066 } 1067 1068 /* 1069 * Track swapcache management flags in the namecache. 1070 * 1071 * Calculate the flags based on the current vattr info 1072 * and recalculate the inherited flags from the parent 1073 * (the original cache linkage may have occurred without 1074 * getattrs and thus have stale flags). 1075 */ 1076 cflags = 0; 1077 if (va.va_flags & SF_NOCACHE) 1078 cflags |= NCF_SF_NOCACHE; 1079 if (va.va_flags & UF_CACHE) 1080 cflags |= NCF_UF_CACHE; 1081 if (ncp->nc_parent) { 1082 if (ncp->nc_parent->nc_flag & 1083 (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) { 1084 cflags |= NCF_SF_PNOCACHE; 1085 } 1086 if (ncp->nc_parent->nc_flag & 1087 (NCF_UF_CACHE | NCF_UF_PCACHE)) { 1088 cflags |= NCF_UF_PCACHE; 1089 } 1090 } 1091 ncp->nc_flag &= ~(NCF_SF_NOCACHE | NCF_UF_CACHE | 1092 NCF_SF_PNOCACHE | NCF_UF_PCACHE); 1093 ncp->nc_flag |= cflags; 1094 1095 /* 1096 * Process general access. 1097 */ 1098 error = naccess_va(&va, nflags, cred); 1099 } 1100 } 1101 } 1102 return(error); 1103 } 1104 1105 /* 1106 * Check the requested access against the given vattr using cred. 1107 */ 1108 int 1109 naccess_va(struct vattr *va, int nflags, struct ucred *cred) 1110 { 1111 int i; 1112 int vmode; 1113 1114 /* 1115 * Test the immutable bit. Creations, deletions, renames (source 1116 * or destination) are not allowed. chown/chmod/other is also not 1117 * allowed but is handled by SETATTR. Hardlinks to the immutable 1118 * file are allowed. 1119 * 1120 * If the directory is set to immutable then creations, deletions, 1121 * renames (source or dest) and hardlinks to files within the directory 1122 * are not allowed, and regular files opened through the directory may 1123 * not be written to or truncated (unless a special device). 1124 * 1125 * NOTE! New hardlinks to immutable files work but new hardlinks to 1126 * files, immutable or not, sitting inside an immutable directory are 1127 * not allowed. As always if the file is hardlinked via some other 1128 * path additional hardlinks may be possible even if the file is marked 1129 * immutable. The sysop needs to create a closure by checking the hard 1130 * link count. Once closure is achieved you are good, and security 1131 * scripts should check link counts anyway. 1132 * 1133 * Writes and truncations are only allowed on special devices. 1134 */ 1135 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) { 1136 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK)) 1137 return (EPERM); 1138 if (nflags & (NLC_CREATE | NLC_DELETE | 1139 NLC_RENAME_SRC | NLC_RENAME_DST)) { 1140 return (EPERM); 1141 } 1142 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) { 1143 switch(va->va_type) { 1144 case VDIR: 1145 return (EISDIR); 1146 case VLNK: 1147 case VREG: 1148 case VDATABASE: 1149 return (EPERM); 1150 default: 1151 break; 1152 } 1153 } 1154 } 1155 1156 /* 1157 * Test the no-unlink and append-only bits for opens, rename targets, 1158 * and deletions. These bits are not tested for creations or 1159 * rename sources. 1160 * 1161 * Unlike FreeBSD we allow a file with APPEND set to be renamed. 1162 * If you do not wish this you must also set NOUNLINK. 1163 * 1164 * If the governing directory is marked APPEND-only it implies 1165 * NOUNLINK for all entries in the directory. 1166 */ 1167 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) && 1168 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) 1169 ) { 1170 return (EPERM); 1171 } 1172 1173 /* 1174 * A file marked append-only may not be deleted but can be renamed. 1175 */ 1176 if ((va->va_flags & APPEND) && 1177 (nflags & (NLC_DELETE | NLC_RENAME_DST)) 1178 ) { 1179 return (EPERM); 1180 } 1181 1182 /* 1183 * A file marked append-only which is opened for writing must also 1184 * be opened O_APPEND. 1185 */ 1186 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) { 1187 if (nflags & NLC_TRUNCATE) 1188 return (EPERM); 1189 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) { 1190 if ((nflags & NLC_APPEND) == 0) 1191 return (EPERM); 1192 } 1193 } 1194 1195 /* 1196 * root gets universal access 1197 */ 1198 if (cred->cr_uid == 0) 1199 return(0); 1200 1201 /* 1202 * Check owner perms. 1203 * 1204 * If NLC_OWN is set the owner of the file is allowed no matter when 1205 * the owner-mode bits say (utimes). 1206 */ 1207 vmode = 0; 1208 if (nflags & NLC_READ) 1209 vmode |= S_IRUSR; 1210 if (nflags & NLC_WRITE) 1211 vmode |= S_IWUSR; 1212 if (nflags & NLC_EXEC) 1213 vmode |= S_IXUSR; 1214 1215 if (cred->cr_uid == va->va_uid) { 1216 if ((nflags & NLC_OWN) == 0) { 1217 if ((vmode & va->va_mode) != vmode) 1218 return(EACCES); 1219 } 1220 return(0); 1221 } 1222 1223 /* 1224 * If NLC_STICKY is set only the owner may delete or rename a file. 1225 * This bit is typically set on /tmp. 1226 * 1227 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in 1228 * the specific delete or rename case. For deletions and renames we 1229 * usually just care about directory permissions, not file permissions. 1230 */ 1231 if ((nflags & NLC_STICKY) && 1232 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) { 1233 return(EACCES); 1234 } 1235 1236 /* 1237 * Check group perms 1238 */ 1239 vmode >>= 3; 1240 for (i = 0; i < cred->cr_ngroups; ++i) { 1241 if (va->va_gid == cred->cr_groups[i]) { 1242 if ((vmode & va->va_mode) != vmode) 1243 return(EACCES); 1244 return(0); 1245 } 1246 } 1247 1248 /* 1249 * Check world perms 1250 */ 1251 vmode >>= 3; 1252 if ((vmode & va->va_mode) != vmode) 1253 return(EACCES); 1254 return(0); 1255 } 1256 1257