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 struct vnode *hvp; /* hold to prevent recyclement */ 416 int wasdotordotdot; 417 char *ptr; 418 char *xptr; 419 int error; 420 int len; 421 int dflags; 422 int hit = 1; 423 424 #ifdef KTRACE 425 if (KTRPOINT(nd->nl_td, KTR_NAMEI)) 426 ktrnamei(nd->nl_td->td_lwp, nd->nl_path); 427 #endif 428 bzero(&nlc, sizeof(nlc)); 429 430 /* 431 * Setup for the loop. The current working namecache element is 432 * always at least referenced. We lock it as required, but always 433 * return a locked, resolved namecache entry. 434 */ 435 nd->nl_loopcnt = 0; 436 if (nd->nl_dvp) { 437 vrele(nd->nl_dvp); 438 nd->nl_dvp = NULL; 439 } 440 ptr = nd->nl_path; 441 442 /* 443 * Loop on the path components. At the top of the loop nd->nl_nch 444 * is ref'd and unlocked and represents our current position. 445 */ 446 for (;;) { 447 /* 448 * Make sure nl_nch is locked so we can access the vnode, resolution 449 * state, etc. 450 */ 451 if ((nd->nl_flags & NLC_NCPISLOCKED) == 0) { 452 nd->nl_flags |= NLC_NCPISLOCKED; 453 cache_lock(&nd->nl_nch); 454 } 455 456 /* 457 * Check if the root directory should replace the current 458 * directory. This is done at the start of a translation 459 * or after a symbolic link has been found. In other cases 460 * ptr will never be pointing at a '/'. 461 */ 462 if (*ptr == '/') { 463 do { 464 ++ptr; 465 } while (*ptr == '/'); 466 cache_get(&nd->nl_rootnch, &nch); 467 cache_put(&nd->nl_nch); 468 nd->nl_nch = nch; /* remains locked */ 469 470 /* 471 * Fast-track termination. There is no parent directory of 472 * the root in the same mount from the point of view of 473 * the caller so return EACCES if NLC_REFDVP is specified, 474 * and EEXIST if NLC_CREATE is also specified. 475 * e.g. 'rmdir /' or 'mkdir /' are not allowed. 476 */ 477 if (*ptr == 0) { 478 if (nd->nl_flags & NLC_REFDVP) 479 error = (nd->nl_flags & NLC_CREATE) ? EEXIST : EACCES; 480 else 481 error = 0; 482 break; 483 } 484 continue; 485 } 486 487 /* 488 * Check directory search permissions. 489 */ 490 dflags = 0; 491 error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags); 492 if (error) 493 break; 494 495 /* 496 * Extract the path component. Path components are limited to 497 * 255 characters. 498 */ 499 nlc.nlc_nameptr = ptr; 500 while (*ptr && *ptr != '/') 501 ++ptr; 502 nlc.nlc_namelen = ptr - nlc.nlc_nameptr; 503 if (nlc.nlc_namelen >= 256) { 504 error = ENAMETOOLONG; 505 break; 506 } 507 508 /* 509 * Lookup the path component in the cache, creating an unresolved 510 * entry if necessary. We have to handle "." and ".." as special 511 * cases. 512 * 513 * When handling ".." we have to detect a traversal back through a 514 * mount point. If we are at the root, ".." just returns the root. 515 * 516 * When handling "." or ".." we also have to recalculate dflags 517 * since our dflags will be for some sub-directory instead of the 518 * parent dir. 519 * 520 * This subsection returns a locked, refd 'nch' unless it errors out. 521 * The namecache topology is not allowed to be disconnected, so 522 * encountering a NULL parent will generate EINVAL. This typically 523 * occurs when a directory is removed out from under a process. 524 */ 525 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') { 526 cache_get(&nd->nl_nch, &nch); 527 wasdotordotdot = 1; 528 } else if (nlc.nlc_namelen == 2 && 529 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') { 530 if (nd->nl_nch.mount == nd->nl_rootnch.mount && 531 nd->nl_nch.ncp == nd->nl_rootnch.ncp 532 ) { 533 /* 534 * ".." at the root returns the root 535 */ 536 cache_get(&nd->nl_nch, &nch); 537 } else { 538 /* 539 * Locate the parent ncp. If we are at the root of a 540 * filesystem mount we have to skip to the mounted-on 541 * point in the underlying filesystem. 542 * 543 * Expect the parent to always be good since the 544 * mountpoint doesn't go away. XXX hack. cache_get() 545 * requires the ncp to already have a ref as a safety. 546 */ 547 nctmp = nd->nl_nch; 548 while (nctmp.ncp == nctmp.mount->mnt_ncmountpt.ncp) 549 nctmp = nctmp.mount->mnt_ncmounton; 550 nctmp.ncp = nctmp.ncp->nc_parent; 551 KKASSERT(nctmp.ncp != NULL); 552 cache_hold(&nctmp); 553 cache_get(&nctmp, &nch); 554 cache_drop(&nctmp); /* NOTE: zero's nctmp */ 555 } 556 wasdotordotdot = 2; 557 } else { 558 /* 559 * Must unlock nl_nch when traversing down the path. However, 560 * the child ncp has not yet been found/created and the parent's 561 * child list might be empty. Thus releasing the lock can 562 * allow a race whereby the parent ncp's vnode is recycled. 563 * This case can occur especially when maxvnodes is set very low. 564 * 565 * We need the parent's ncp to remain resolved for all normal 566 * filesystem activities, so we vhold() the vp during the lookup 567 * to prevent recyclement due to vnlru / maxvnodes. 568 * 569 * If we race an unlink or rename the ncp might be marked 570 * DESTROYED after resolution, requiring a retry. 571 */ 572 if ((hvp = nd->nl_nch.ncp->nc_vp) != NULL) 573 vhold(hvp); 574 cache_unlock(&nd->nl_nch); 575 nd->nl_flags &= ~NLC_NCPISLOCKED; 576 nch = cache_nlookup(&nd->nl_nch, &nlc); 577 if (nch.ncp->nc_flag & NCF_UNRESOLVED) 578 hit = 0; 579 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN || 580 (nch.ncp->nc_flag & NCF_DESTROYED)) { 581 kprintf("[diagnostic] nlookup: relookup %*.*s\n", 582 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name); 583 cache_put(&nch); 584 nch = cache_nlookup(&nd->nl_nch, &nlc); 585 } 586 if (hvp) 587 vdrop(hvp); 588 wasdotordotdot = 0; 589 } 590 591 /* 592 * If the last component was "." or ".." our dflags no longer 593 * represents the parent directory and we have to explicitly 594 * look it up. 595 * 596 * Expect the parent to be good since nch is locked. 597 */ 598 if (wasdotordotdot && error == 0) { 599 dflags = 0; 600 if ((par.ncp = nch.ncp->nc_parent) != NULL) { 601 par.mount = nch.mount; 602 cache_hold(&par); 603 cache_lock(&par); 604 error = naccess(&par, 0, nd->nl_cred, &dflags); 605 cache_put(&par); 606 } 607 } 608 if (nd->nl_flags & NLC_NCPISLOCKED) { 609 cache_unlock(&nd->nl_nch); 610 nd->nl_flags &= ~NLC_NCPISLOCKED; 611 } 612 613 /* 614 * [end of subsection] 615 * 616 * nch is locked and referenced. 617 * nd->nl_nch is unlocked and referenced. 618 * 619 * nl_nch must be unlocked or we could chain lock to the root 620 * if a resolve gets stuck (e.g. in NFS). 621 */ 622 623 /* 624 * Resolve the namespace if necessary. The ncp returned by 625 * cache_nlookup() is referenced and locked. 626 * 627 * XXX neither '.' nor '..' should return EAGAIN since they were 628 * previously resolved and thus cannot be newly created ncp's. 629 */ 630 if (nch.ncp->nc_flag & NCF_UNRESOLVED) { 631 hit = 0; 632 error = cache_resolve(&nch, nd->nl_cred); 633 KKASSERT(error != EAGAIN); 634 } else { 635 error = nch.ncp->nc_error; 636 } 637 638 /* 639 * Early completion. ENOENT is not an error if this is the last 640 * component and NLC_CREATE or NLC_RENAME (rename target) was 641 * requested. Note that ncp->nc_error is left as ENOENT in that 642 * case, which we check later on. 643 * 644 * Also handle invalid '.' or '..' components terminating a path 645 * for a create/rename/delete. The standard requires this and pax 646 * pretty stupidly depends on it. 647 */ 648 for (xptr = ptr; *xptr == '/'; ++xptr) 649 ; 650 if (*xptr == 0) { 651 if (error == ENOENT && 652 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST)) 653 ) { 654 if (nd->nl_flags & NLC_NFS_RDONLY) { 655 error = EROFS; 656 } else { 657 error = naccess(&nch, nd->nl_flags | dflags, 658 nd->nl_cred, NULL); 659 } 660 } 661 if (error == 0 && wasdotordotdot && 662 (nd->nl_flags & (NLC_CREATE | NLC_DELETE | 663 NLC_RENAME_SRC | NLC_RENAME_DST))) { 664 /* 665 * POSIX junk 666 */ 667 if (nd->nl_flags & NLC_CREATE) 668 error = EEXIST; 669 else if (nd->nl_flags & NLC_DELETE) 670 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY; 671 else 672 error = EINVAL; 673 } 674 } 675 676 /* 677 * Early completion on error. 678 */ 679 if (error) { 680 cache_put(&nch); 681 break; 682 } 683 684 /* 685 * If the element is a symlink and it is either not the last 686 * element or it is the last element and we are allowed to 687 * follow symlinks, resolve the symlink. 688 */ 689 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) && 690 (*ptr || (nd->nl_flags & NLC_FOLLOW)) 691 ) { 692 if (nd->nl_loopcnt++ >= MAXSYMLINKS) { 693 error = ELOOP; 694 cache_put(&nch); 695 break; 696 } 697 error = nreadsymlink(nd, &nch, &nlc); 698 cache_put(&nch); 699 if (error) 700 break; 701 702 /* 703 * Concatenate trailing path elements onto the returned symlink. 704 * Note that if the path component (ptr) is not exhausted, it 705 * will being with a '/', so we do not have to add another one. 706 * 707 * The symlink may not be empty. 708 */ 709 len = strlen(ptr); 710 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) { 711 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT; 712 objcache_put(namei_oc, nlc.nlc_nameptr); 713 break; 714 } 715 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1); 716 if (nd->nl_flags & NLC_HASBUF) 717 objcache_put(namei_oc, nd->nl_path); 718 nd->nl_path = nlc.nlc_nameptr; 719 nd->nl_flags |= NLC_HASBUF; 720 ptr = nd->nl_path; 721 722 /* 723 * Go back up to the top to resolve any initial '/'s in the 724 * symlink. 725 */ 726 continue; 727 } 728 729 /* 730 * If the element is a directory and we are crossing a mount point, 731 * Locate the mount. 732 */ 733 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && 734 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 && 735 (mp = cache_findmount(&nch)) != NULL 736 ) { 737 struct vnode *tdp; 738 739 cache_put(&nch); 740 cache_get(&mp->mnt_ncmountpt, &nch); 741 742 if (nch.ncp->nc_flag & NCF_UNRESOLVED) { 743 while (vfs_busy(mp, 0)) 744 ; 745 error = VFS_ROOT(mp, &tdp); 746 vfs_unbusy(mp); 747 if (error) { 748 cache_dropmount(mp); 749 break; 750 } 751 cache_setvp(&nch, tdp); 752 vput(tdp); 753 } 754 cache_dropmount(mp); 755 } 756 if (error) { 757 cache_put(&nch); 758 break; 759 } 760 761 /* 762 * Skip any slashes to get to the next element. If there 763 * are any slashes at all the current element must be a 764 * directory or, in the create case, intended to become a directory. 765 * If it isn't we break without incrementing ptr and fall through 766 * to the failure case below. 767 */ 768 while (*ptr == '/') { 769 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 && 770 !(nd->nl_flags & NLC_WILLBEDIR) 771 ) { 772 break; 773 } 774 ++ptr; 775 } 776 777 /* 778 * Continuation case: additional elements and the current 779 * element is a directory. 780 */ 781 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) { 782 cache_drop(&nd->nl_nch); 783 cache_unlock(&nch); 784 KKASSERT((nd->nl_flags & NLC_NCPISLOCKED) == 0); 785 nd->nl_nch = nch; 786 continue; 787 } 788 789 /* 790 * Failure case: additional elements and the current element 791 * is not a directory 792 */ 793 if (*ptr) { 794 cache_put(&nch); 795 error = ENOTDIR; 796 break; 797 } 798 799 /* 800 * Successful lookup of last element. 801 * 802 * Check permissions if the target exists. If the target does not 803 * exist directory permissions were already tested in the early 804 * completion code above. 805 * 806 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY 807 * if the file is marked append-only, and NLC_STICKY if the directory 808 * containing the file is sticky. 809 */ 810 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) { 811 error = naccess(&nch, nd->nl_flags | dflags, 812 nd->nl_cred, NULL); 813 if (error) { 814 cache_put(&nch); 815 break; 816 } 817 } 818 819 /* 820 * Termination: no more elements. 821 * 822 * If NLC_REFDVP is set acquire a referenced parent dvp. 823 */ 824 if (nd->nl_flags & NLC_REFDVP) { 825 cache_lock(&nd->nl_nch); 826 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp); 827 cache_unlock(&nd->nl_nch); 828 if (error) { 829 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp); 830 cache_put(&nch); 831 break; 832 } 833 } 834 cache_drop(&nd->nl_nch); 835 nd->nl_nch = nch; 836 nd->nl_flags |= NLC_NCPISLOCKED; 837 error = 0; 838 break; 839 } 840 841 if (hit) 842 ++gd->gd_nchstats->ncs_longhits; 843 else 844 ++gd->gd_nchstats->ncs_longmiss; 845 846 /* 847 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit 848 * (ncp->nc_error will be ENOENT), but we will still return an error 849 * code of 0. 850 */ 851 return(error); 852 } 853 854 /* 855 * Resolve a mount point's glue ncp. This ncp connects creates the illusion 856 * of continuity in the namecache tree by connecting the ncp related to the 857 * vnode under the mount to the ncp related to the mount's root vnode. 858 * 859 * If no error occured a locked, ref'd ncp is stored in *ncpp. 860 */ 861 int 862 nlookup_mp(struct mount *mp, struct nchandle *nch) 863 { 864 struct vnode *vp; 865 int error; 866 867 error = 0; 868 cache_get(&mp->mnt_ncmountpt, nch); 869 if (nch->ncp->nc_flag & NCF_UNRESOLVED) { 870 while (vfs_busy(mp, 0)) 871 ; 872 error = VFS_ROOT(mp, &vp); 873 vfs_unbusy(mp); 874 if (error) { 875 cache_put(nch); 876 } else { 877 cache_setvp(nch, vp); 878 vput(vp); 879 } 880 } 881 return(error); 882 } 883 884 /* 885 * Read the contents of a symlink, allocate a path buffer out of the 886 * namei_oc and initialize the supplied nlcomponent with the result. 887 * 888 * If an error occurs no buffer will be allocated or returned in the nlc. 889 */ 890 int 891 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch, 892 struct nlcomponent *nlc) 893 { 894 struct vnode *vp; 895 struct iovec aiov; 896 struct uio auio; 897 int linklen; 898 int error; 899 char *cp; 900 901 nlc->nlc_nameptr = NULL; 902 nlc->nlc_namelen = 0; 903 if (nch->ncp->nc_vp == NULL) 904 return(ENOENT); 905 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0) 906 return(error); 907 cp = objcache_get(namei_oc, M_WAITOK); 908 aiov.iov_base = cp; 909 aiov.iov_len = MAXPATHLEN; 910 auio.uio_iov = &aiov; 911 auio.uio_iovcnt = 1; 912 auio.uio_offset = 0; 913 auio.uio_rw = UIO_READ; 914 auio.uio_segflg = UIO_SYSSPACE; 915 auio.uio_td = nd->nl_td; 916 auio.uio_resid = MAXPATHLEN - 1; 917 error = VOP_READLINK(vp, &auio, nd->nl_cred); 918 if (error) 919 goto fail; 920 linklen = MAXPATHLEN - 1 - auio.uio_resid; 921 if (varsym_enable) { 922 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1); 923 if (linklen < 0) { 924 error = ENAMETOOLONG; 925 goto fail; 926 } 927 } 928 cp[linklen] = 0; 929 nlc->nlc_nameptr = cp; 930 nlc->nlc_namelen = linklen; 931 vput(vp); 932 return(0); 933 fail: 934 objcache_put(namei_oc, cp); 935 vput(vp); 936 return(error); 937 } 938 939 /* 940 * Check access [XXX cache vattr!] [XXX quota] 941 * 942 * Generally check the NLC_* access bits. All specified bits must pass 943 * for this function to return 0. 944 * 945 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST 946 * access, otherwise it must exist. No error is returned in this case. 947 * 948 * The file must not exist if NLC_EXCL is specified. 949 * 950 * Directory permissions in general are tested for NLC_CREATE if the file 951 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST 952 * whether the file exists or not. 953 * 954 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST, 955 * the latter is only tested if the target exists. 956 * 957 * The passed ncp must be referenced and locked. 958 */ 959 static int 960 naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp) 961 { 962 struct vnode *vp; 963 struct vattr va; 964 struct namecache *ncp; 965 int error; 966 int cflags; 967 968 ASSERT_NCH_LOCKED(nch); 969 ncp = nch->ncp; 970 if (ncp->nc_flag & NCF_UNRESOLVED) { 971 cache_resolve(nch, cred); 972 ncp = nch->ncp; 973 } 974 error = ncp->nc_error; 975 976 /* 977 * Directory permissions checks. Silently ignore ENOENT if these 978 * tests pass. It isn't an error. 979 * 980 * We can safely resolve ncp->nc_parent because ncp is currently 981 * locked. 982 */ 983 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) { 984 if (((nflags & NLC_CREATE) && ncp->nc_vp == NULL) || 985 ((nflags & NLC_DELETE) && ncp->nc_vp != NULL) || 986 ((nflags & NLC_RENAME_SRC) && ncp->nc_vp != NULL) || 987 (nflags & NLC_RENAME_DST) 988 ) { 989 struct nchandle par; 990 991 if ((par.ncp = ncp->nc_parent) == NULL) { 992 if (error != EAGAIN) 993 error = EINVAL; 994 } else if (error == 0 || error == ENOENT) { 995 par.mount = nch->mount; 996 cache_hold(&par); 997 cache_lock(&par); 998 error = naccess(&par, NLC_WRITE, cred, NULL); 999 cache_put(&par); 1000 } 1001 } 1002 } 1003 1004 /* 1005 * NLC_EXCL check. Target file must not exist. 1006 */ 1007 if (error == 0 && (nflags & NLC_EXCL) && ncp->nc_vp != NULL) 1008 error = EEXIST; 1009 1010 /* 1011 * Get the vnode attributes so we can do the rest of our checks. 1012 * 1013 * NOTE: We only call naccess_va() if the target exists. 1014 */ 1015 if (error == 0) { 1016 error = cache_vget(nch, cred, LK_SHARED, &vp); 1017 if (error == ENOENT) { 1018 /* 1019 * Silently zero-out ENOENT if creating or renaming 1020 * (rename target). It isn't an error. 1021 */ 1022 if (nflags & (NLC_CREATE | NLC_RENAME_DST)) 1023 error = 0; 1024 } else if (error == 0) { 1025 /* 1026 * Get the vnode attributes and check for illegal O_TRUNC 1027 * requests and read-only mounts. 1028 * 1029 * NOTE: You can still open devices on read-only mounts for 1030 * writing. 1031 * 1032 * NOTE: creates/deletes/renames are handled by the NLC_WRITE 1033 * check on the parent directory above. 1034 * 1035 * XXX cache the va in the namecache or in the vnode 1036 */ 1037 error = VOP_GETATTR(vp, &va); 1038 if (error == 0 && (nflags & NLC_TRUNCATE)) { 1039 switch(va.va_type) { 1040 case VREG: 1041 case VDATABASE: 1042 case VCHR: 1043 case VBLK: 1044 case VFIFO: 1045 break; 1046 case VDIR: 1047 error = EISDIR; 1048 break; 1049 default: 1050 error = EINVAL; 1051 break; 1052 } 1053 } 1054 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount && 1055 (vp->v_mount->mnt_flag & MNT_RDONLY) 1056 ) { 1057 switch(va.va_type) { 1058 case VDIR: 1059 case VLNK: 1060 case VREG: 1061 case VDATABASE: 1062 error = EROFS; 1063 break; 1064 default: 1065 break; 1066 } 1067 } 1068 vput(vp); 1069 1070 /* 1071 * Check permissions based on file attributes. The passed 1072 * flags (*nflagsp) are modified with feedback based on 1073 * special attributes and requirements. 1074 */ 1075 if (error == 0) { 1076 /* 1077 * Adjust the returned (*nflagsp) if non-NULL. 1078 */ 1079 if (nflagsp) { 1080 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid) 1081 *nflagsp |= NLC_STICKY; 1082 if (va.va_flags & APPEND) 1083 *nflagsp |= NLC_APPENDONLY; 1084 if (va.va_flags & IMMUTABLE) 1085 *nflagsp |= NLC_IMMUTABLE; 1086 } 1087 1088 /* 1089 * Track swapcache management flags in the namecache. 1090 * 1091 * Calculate the flags based on the current vattr info 1092 * and recalculate the inherited flags from the parent 1093 * (the original cache linkage may have occurred without 1094 * getattrs and thus have stale flags). 1095 */ 1096 cflags = 0; 1097 if (va.va_flags & SF_NOCACHE) 1098 cflags |= NCF_SF_NOCACHE; 1099 if (va.va_flags & UF_CACHE) 1100 cflags |= NCF_UF_CACHE; 1101 if (ncp->nc_parent) { 1102 if (ncp->nc_parent->nc_flag & 1103 (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) { 1104 cflags |= NCF_SF_PNOCACHE; 1105 } 1106 if (ncp->nc_parent->nc_flag & 1107 (NCF_UF_CACHE | NCF_UF_PCACHE)) { 1108 cflags |= NCF_UF_PCACHE; 1109 } 1110 } 1111 ncp->nc_flag &= ~(NCF_SF_NOCACHE | NCF_UF_CACHE | 1112 NCF_SF_PNOCACHE | NCF_UF_PCACHE); 1113 ncp->nc_flag |= cflags; 1114 1115 /* 1116 * Process general access. 1117 */ 1118 error = naccess_va(&va, nflags, cred); 1119 } 1120 } 1121 } 1122 return(error); 1123 } 1124 1125 /* 1126 * Check the requested access against the given vattr using cred. 1127 */ 1128 int 1129 naccess_va(struct vattr *va, int nflags, struct ucred *cred) 1130 { 1131 int i; 1132 int vmode; 1133 1134 /* 1135 * Test the immutable bit. Creations, deletions, renames (source 1136 * or destination) are not allowed. chown/chmod/other is also not 1137 * allowed but is handled by SETATTR. Hardlinks to the immutable 1138 * file are allowed. 1139 * 1140 * If the directory is set to immutable then creations, deletions, 1141 * renames (source or dest) and hardlinks to files within the directory 1142 * are not allowed, and regular files opened through the directory may 1143 * not be written to or truncated (unless a special device). 1144 * 1145 * NOTE! New hardlinks to immutable files work but new hardlinks to 1146 * files, immutable or not, sitting inside an immutable directory are 1147 * not allowed. As always if the file is hardlinked via some other 1148 * path additional hardlinks may be possible even if the file is marked 1149 * immutable. The sysop needs to create a closure by checking the hard 1150 * link count. Once closure is achieved you are good, and security 1151 * scripts should check link counts anyway. 1152 * 1153 * Writes and truncations are only allowed on special devices. 1154 */ 1155 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) { 1156 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK)) 1157 return (EPERM); 1158 if (nflags & (NLC_CREATE | NLC_DELETE | 1159 NLC_RENAME_SRC | NLC_RENAME_DST)) { 1160 return (EPERM); 1161 } 1162 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) { 1163 switch(va->va_type) { 1164 case VDIR: 1165 return (EISDIR); 1166 case VLNK: 1167 case VREG: 1168 case VDATABASE: 1169 return (EPERM); 1170 default: 1171 break; 1172 } 1173 } 1174 } 1175 1176 /* 1177 * Test the no-unlink and append-only bits for opens, rename targets, 1178 * and deletions. These bits are not tested for creations or 1179 * rename sources. 1180 * 1181 * Unlike FreeBSD we allow a file with APPEND set to be renamed. 1182 * If you do not wish this you must also set NOUNLINK. 1183 * 1184 * If the governing directory is marked APPEND-only it implies 1185 * NOUNLINK for all entries in the directory. 1186 */ 1187 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) && 1188 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) 1189 ) { 1190 return (EPERM); 1191 } 1192 1193 /* 1194 * A file marked append-only may not be deleted but can be renamed. 1195 */ 1196 if ((va->va_flags & APPEND) && 1197 (nflags & (NLC_DELETE | NLC_RENAME_DST)) 1198 ) { 1199 return (EPERM); 1200 } 1201 1202 /* 1203 * A file marked append-only which is opened for writing must also 1204 * be opened O_APPEND. 1205 */ 1206 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) { 1207 if (nflags & NLC_TRUNCATE) 1208 return (EPERM); 1209 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) { 1210 if ((nflags & NLC_APPEND) == 0) 1211 return (EPERM); 1212 } 1213 } 1214 1215 /* 1216 * root gets universal access 1217 */ 1218 if (cred->cr_uid == 0) 1219 return(0); 1220 1221 /* 1222 * Check owner perms. 1223 * 1224 * If NLC_OWN is set the owner of the file is allowed no matter when 1225 * the owner-mode bits say (utimes). 1226 */ 1227 vmode = 0; 1228 if (nflags & NLC_READ) 1229 vmode |= S_IRUSR; 1230 if (nflags & NLC_WRITE) 1231 vmode |= S_IWUSR; 1232 if (nflags & NLC_EXEC) 1233 vmode |= S_IXUSR; 1234 1235 if (cred->cr_uid == va->va_uid) { 1236 if ((nflags & NLC_OWN) == 0) { 1237 if ((vmode & va->va_mode) != vmode) 1238 return(EACCES); 1239 } 1240 return(0); 1241 } 1242 1243 /* 1244 * If NLC_STICKY is set only the owner may delete or rename a file. 1245 * This bit is typically set on /tmp. 1246 * 1247 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in 1248 * the specific delete or rename case. For deletions and renames we 1249 * usually just care about directory permissions, not file permissions. 1250 */ 1251 if ((nflags & NLC_STICKY) && 1252 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) { 1253 return(EACCES); 1254 } 1255 1256 /* 1257 * Check group perms 1258 */ 1259 vmode >>= 3; 1260 for (i = 0; i < cred->cr_ngroups; ++i) { 1261 if (va->va_gid == cred->cr_groups[i]) { 1262 if ((vmode & va->va_mode) != vmode) 1263 return(EACCES); 1264 return(0); 1265 } 1266 } 1267 1268 /* 1269 * Check world perms 1270 */ 1271 vmode >>= 3; 1272 if ((vmode & va->va_mode) != vmode) 1273 return(EACCES); 1274 return(0); 1275 } 1276 1277