1 /* $NetBSD: vfs_getcwd.c,v 1.46 2010/07/21 09:01:36 hannken Exp $ */ 2 3 /*- 4 * Copyright (c) 1999 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Bill Sommerfeld. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __KERNEL_RCSID(0, "$NetBSD: vfs_getcwd.c,v 1.46 2010/07/21 09:01:36 hannken Exp $"); 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/namei.h> 38 #include <sys/filedesc.h> 39 #include <sys/kernel.h> 40 #include <sys/file.h> 41 #include <sys/stat.h> 42 #include <sys/vnode.h> 43 #include <sys/mount.h> 44 #include <sys/proc.h> 45 #include <sys/uio.h> 46 #include <sys/kmem.h> 47 #include <sys/dirent.h> 48 #include <sys/kauth.h> 49 50 #include <ufs/ufs/dir.h> /* XXX only for DIRBLKSIZ */ 51 52 #include <sys/syscallargs.h> 53 54 /* 55 * Vnode variable naming conventions in this file: 56 * 57 * rvp: the current root we're aiming towards. 58 * lvp, *lvpp: the "lower" vnode 59 * uvp, *uvpp: the "upper" vnode. 60 * 61 * Since all the vnodes we're dealing with are directories, and the 62 * lookups are going *up* in the filesystem rather than *down*, the 63 * usual "pvp" (parent) or "dvp" (directory) naming conventions are 64 * too confusing. 65 */ 66 67 /* 68 * XXX Will infinite loop in certain cases if a directory read reliably 69 * returns EINVAL on last block. 70 * XXX is EINVAL the right thing to return if a directory is malformed? 71 */ 72 73 /* 74 * XXX Untested vs. mount -o union; probably does the wrong thing. 75 */ 76 77 /* 78 * Find parent vnode of *lvpp, return in *uvpp 79 * 80 * If we care about the name, scan it looking for name of directory 81 * entry pointing at lvp. 82 * 83 * Place the name in the buffer which starts at bufp, immediately 84 * before *bpp, and move bpp backwards to point at the start of it. 85 * 86 * On entry, *lvpp is a locked vnode reference; on exit, it is vput and NULL'ed 87 * On exit, *uvpp is either NULL or is a locked vnode reference. 88 */ 89 static int 90 getcwd_scandir(struct vnode **lvpp, struct vnode **uvpp, char **bpp, 91 char *bufp, struct lwp *l) 92 { 93 int error = 0; 94 int eofflag; 95 off_t off; 96 int tries; 97 struct uio uio; 98 struct iovec iov; 99 char *dirbuf = NULL; 100 int dirbuflen; 101 ino_t fileno; 102 struct vattr va; 103 struct vnode *uvp = NULL; 104 struct vnode *lvp = *lvpp; 105 kauth_cred_t cred = l->l_cred; 106 struct componentname cn; 107 int len, reclen; 108 tries = 0; 109 110 /* 111 * If we want the filename, get some info we need while the 112 * current directory is still locked. 113 */ 114 if (bufp != NULL) { 115 error = VOP_GETATTR(lvp, &va, cred); 116 if (error) { 117 vput(lvp); 118 *lvpp = NULL; 119 *uvpp = NULL; 120 return error; 121 } 122 } 123 124 /* 125 * Ok, we have to do it the hard way.. 126 * Next, get parent vnode using lookup of .. 127 */ 128 cn.cn_nameiop = LOOKUP; 129 cn.cn_flags = ISLASTCN | ISDOTDOT | RDONLY; 130 cn.cn_cred = cred; 131 cn.cn_pnbuf = NULL; 132 cn.cn_nameptr = ".."; 133 cn.cn_namelen = 2; 134 cn.cn_hash = 0; 135 cn.cn_consume = 0; 136 137 /* 138 * At this point, lvp is locked. 139 * On successful return, *uvpp will be locked 140 */ 141 error = VOP_LOOKUP(lvp, uvpp, &cn); 142 vput(lvp); 143 if (error) { 144 *lvpp = NULL; 145 *uvpp = NULL; 146 return error; 147 } 148 uvp = *uvpp; 149 150 /* If we don't care about the pathname, we're done */ 151 if (bufp == NULL) { 152 *lvpp = NULL; 153 return 0; 154 } 155 156 fileno = va.va_fileid; 157 158 dirbuflen = DIRBLKSIZ; 159 if (dirbuflen < va.va_blocksize) 160 dirbuflen = va.va_blocksize; 161 dirbuf = kmem_alloc(dirbuflen, KM_SLEEP); 162 163 #if 0 164 unionread: 165 #endif 166 off = 0; 167 do { 168 /* call VOP_READDIR of parent */ 169 iov.iov_base = dirbuf; 170 iov.iov_len = dirbuflen; 171 172 uio.uio_iov = &iov; 173 uio.uio_iovcnt = 1; 174 uio.uio_offset = off; 175 uio.uio_resid = dirbuflen; 176 uio.uio_rw = UIO_READ; 177 UIO_SETUP_SYSSPACE(&uio); 178 179 eofflag = 0; 180 181 error = VOP_READDIR(uvp, &uio, cred, &eofflag, 0, 0); 182 183 off = uio.uio_offset; 184 185 /* 186 * Try again if NFS tosses its cookies. 187 * XXX this can still loop forever if the directory is busted 188 * such that the second or subsequent page of it always 189 * returns EINVAL 190 */ 191 if ((error == EINVAL) && (tries < 3)) { 192 off = 0; 193 tries++; 194 continue; /* once more, with feeling */ 195 } 196 197 if (!error) { 198 char *cpos; 199 struct dirent *dp; 200 201 cpos = dirbuf; 202 tries = 0; 203 204 /* scan directory page looking for matching vnode */ 205 for (len = (dirbuflen - uio.uio_resid); len > 0; 206 len -= reclen) { 207 dp = (struct dirent *) cpos; 208 reclen = dp->d_reclen; 209 210 /* check for malformed directory.. */ 211 if (reclen < _DIRENT_MINSIZE(dp)) { 212 error = EINVAL; 213 goto out; 214 } 215 /* 216 * XXX should perhaps do VOP_LOOKUP to 217 * check that we got back to the right place, 218 * but getting the locking games for that 219 * right would be heinous. 220 */ 221 if ((dp->d_type != DT_WHT) && 222 (dp->d_fileno == fileno)) { 223 char *bp = *bpp; 224 225 bp -= dp->d_namlen; 226 if (bp <= bufp) { 227 error = ERANGE; 228 goto out; 229 } 230 memcpy(bp, dp->d_name, dp->d_namlen); 231 error = 0; 232 *bpp = bp; 233 goto out; 234 } 235 cpos += reclen; 236 } 237 } else 238 goto out; 239 } while (!eofflag); 240 #if 0 241 /* 242 * Deal with mount -o union, which unions only the 243 * root directory of the mount. 244 */ 245 if ((uvp->v_vflag & VV_ROOT) && 246 (uvp->v_mount->mnt_flag & MNT_UNION)) { 247 struct vnode *tvp = uvp; 248 249 uvp = uvp->v_mount->mnt_vnodecovered; 250 vput(tvp); 251 vref(uvp); 252 *uvpp = uvp; 253 vn_lock(uvp, LK_EXCLUSIVE | LK_RETRY); 254 goto unionread; 255 } 256 #endif 257 error = ENOENT; 258 259 out: 260 *lvpp = NULL; 261 kmem_free(dirbuf, dirbuflen); 262 return error; 263 } 264 265 /* 266 * Look in the vnode-to-name reverse cache to see if 267 * we can find things the easy way. 268 * 269 * XXX vget failure path is untested. 270 * 271 * On entry, *lvpp is a locked vnode reference. 272 * On exit, one of the following is the case: 273 * 0) Both *lvpp and *uvpp are NULL and failure is returned. 274 * 1) *uvpp is NULL, *lvpp remains locked and -1 is returned (cache miss) 275 * 2) *uvpp is a locked vnode reference, *lvpp is vput and NULL'ed 276 * and 0 is returned (cache hit) 277 */ 278 279 static int 280 getcwd_getcache(struct vnode **lvpp, struct vnode **uvpp, char **bpp, 281 char *bufp) 282 { 283 struct vnode *lvp, *uvp = NULL; 284 int error; 285 286 lvp = *lvpp; 287 288 /* 289 * This returns 0 on a cache hit, -1 on a clean cache miss, 290 * or an errno on other failure. 291 */ 292 error = cache_revlookup(lvp, uvpp, bpp, bufp); 293 if (error) { 294 if (error != -1) { 295 vput(lvp); 296 *lvpp = NULL; 297 *uvpp = NULL; 298 } 299 return error; 300 } 301 uvp = *uvpp; 302 303 /* 304 * Since we're going up, we have to release the current lock 305 * before we take the parent lock. 306 */ 307 308 VOP_UNLOCK(lvp); 309 vn_lock(uvp, LK_EXCLUSIVE | LK_RETRY); 310 vrele(lvp); 311 *lvpp = NULL; 312 313 return error; 314 } 315 316 /* 317 * common routine shared by sys___getcwd() and vn_isunder() 318 */ 319 320 int 321 getcwd_common(struct vnode *lvp, struct vnode *rvp, char **bpp, char *bufp, 322 int limit, int flags, struct lwp *l) 323 { 324 struct cwdinfo *cwdi = l->l_proc->p_cwdi; 325 kauth_cred_t cred = l->l_cred; 326 struct vnode *uvp = NULL; 327 char *bp = NULL; 328 int error; 329 int perms = VEXEC; 330 331 error = 0; 332 if (rvp == NULL) { 333 rvp = cwdi->cwdi_rdir; 334 if (rvp == NULL) 335 rvp = rootvnode; 336 } 337 338 vref(rvp); 339 vref(lvp); 340 341 /* 342 * Error handling invariant: 343 * Before a `goto out': 344 * lvp is either NULL, or locked and held. 345 * uvp is either NULL, or locked and held. 346 */ 347 348 vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY); 349 if (bufp) 350 bp = *bpp; 351 352 /* 353 * this loop will terminate when one of the following happens: 354 * - we hit the root 355 * - getdirentries or lookup fails 356 * - we run out of space in the buffer. 357 */ 358 if (lvp == rvp) { 359 if (bp) 360 *(--bp) = '/'; 361 goto out; 362 } 363 do { 364 /* 365 * access check here is optional, depending on 366 * whether or not caller cares. 367 */ 368 if (flags & GETCWD_CHECK_ACCESS) { 369 error = VOP_ACCESS(lvp, perms, cred); 370 if (error) 371 goto out; 372 perms = VEXEC|VREAD; 373 } 374 375 /* 376 * step up if we're a covered vnode.. 377 */ 378 while (lvp->v_vflag & VV_ROOT) { 379 struct vnode *tvp; 380 381 if (lvp == rvp) 382 goto out; 383 384 tvp = lvp; 385 lvp = lvp->v_mount->mnt_vnodecovered; 386 vput(tvp); 387 /* 388 * hodie natus est radici frater 389 */ 390 if (lvp == NULL) { 391 error = ENOENT; 392 goto out; 393 } 394 vref(lvp); 395 error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY); 396 if (error != 0) { 397 vrele(lvp); 398 lvp = NULL; 399 goto out; 400 } 401 } 402 /* 403 * Look in the name cache; if that fails, look in the 404 * directory.. 405 */ 406 error = getcwd_getcache(&lvp, &uvp, &bp, bufp); 407 if (error == -1) { 408 if (lvp->v_type != VDIR) { 409 error = ENOTDIR; 410 goto out; 411 } 412 error = getcwd_scandir(&lvp, &uvp, &bp, bufp, l); 413 } 414 if (error) 415 goto out; 416 #if DIAGNOSTIC 417 if (lvp != NULL) 418 panic("getcwd: oops, forgot to null lvp"); 419 if (bufp && (bp <= bufp)) { 420 panic("getcwd: oops, went back too far"); 421 } 422 #endif 423 if (bp) 424 *(--bp) = '/'; 425 lvp = uvp; 426 uvp = NULL; 427 limit--; 428 } while ((lvp != rvp) && (limit > 0)); 429 430 out: 431 if (bpp) 432 *bpp = bp; 433 if (uvp) 434 vput(uvp); 435 if (lvp) 436 vput(lvp); 437 vrele(rvp); 438 return error; 439 } 440 441 /* 442 * Check if one directory can be found inside another in the directory 443 * hierarchy. 444 * 445 * Intended to be used in chroot, chdir, fchdir, etc., to ensure that 446 * chroot() actually means something. 447 */ 448 int 449 vn_isunder(struct vnode *lvp, struct vnode *rvp, struct lwp *l) 450 { 451 int error; 452 453 error = getcwd_common(lvp, rvp, NULL, NULL, MAXPATHLEN / 2, 0, l); 454 455 if (!error) 456 return 1; 457 else 458 return 0; 459 } 460 461 /* 462 * Returns true if proc p1's root directory equal to or under p2's 463 * root directory. 464 * 465 * Intended to be used from ptrace/procfs sorts of things. 466 */ 467 468 int 469 proc_isunder(struct proc *p1, struct lwp *l2) 470 { 471 struct vnode *r1 = p1->p_cwdi->cwdi_rdir; 472 struct vnode *r2 = l2->l_proc->p_cwdi->cwdi_rdir; 473 474 if (r1 == NULL) 475 return (r2 == NULL); 476 else if (r2 == NULL) 477 return 1; 478 else 479 return vn_isunder(r1, r2, l2); 480 } 481 482 /* 483 * Find pathname of process's current directory. 484 * 485 * Use vfs vnode-to-name reverse cache; if that fails, fall back 486 * to reading directory contents. 487 */ 488 489 int 490 sys___getcwd(struct lwp *l, const struct sys___getcwd_args *uap, register_t *retval) 491 { 492 /* { 493 syscallarg(char *) bufp; 494 syscallarg(size_t) length; 495 } */ 496 497 int error; 498 char *path; 499 char *bp, *bend; 500 int len = SCARG(uap, length); 501 int lenused; 502 struct cwdinfo *cwdi; 503 504 if (len > MAXPATHLEN * 4) 505 len = MAXPATHLEN * 4; 506 else if (len < 2) 507 return ERANGE; 508 509 path = kmem_alloc(len, KM_SLEEP); 510 if (!path) 511 return ENOMEM; 512 513 bp = &path[len]; 514 bend = bp; 515 *(--bp) = '\0'; 516 517 /* 518 * 5th argument here is "max number of vnodes to traverse". 519 * Since each entry takes up at least 2 bytes in the output buffer, 520 * limit it to N/2 vnodes for an N byte buffer. 521 */ 522 cwdi = l->l_proc->p_cwdi; 523 rw_enter(&cwdi->cwdi_lock, RW_READER); 524 error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, 525 len/2, GETCWD_CHECK_ACCESS, l); 526 rw_exit(&cwdi->cwdi_lock); 527 528 if (error) 529 goto out; 530 lenused = bend - bp; 531 *retval = lenused; 532 /* put the result into user buffer */ 533 error = copyout(bp, SCARG(uap, bufp), lenused); 534 535 out: 536 kmem_free(path, len); 537 return error; 538 } 539 540 /* 541 * Try to find a pathname for a vnode. Since there is no mapping 542 * vnode -> parent directory, this needs the NAMECACHE_ENTER_REVERSE 543 * option to work (to make cache_revlookup succeed). Caller holds a 544 * reference to the vnode. 545 */ 546 int 547 vnode_to_path(char *path, size_t len, struct vnode *vp, struct lwp *curl, 548 struct proc *p) 549 { 550 struct proc *curp = curl->l_proc; 551 int error, lenused, elen; 552 char *bp, *bend; 553 struct vnode *dvp; 554 555 KASSERT(vp->v_usecount > 0); 556 557 bp = bend = &path[len]; 558 *(--bp) = '\0'; 559 560 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 561 if (error != 0) 562 return error; 563 error = cache_revlookup(vp, &dvp, &bp, path); 564 VOP_UNLOCK(vp); 565 if (error != 0) 566 return (error == -1 ? ENOENT : error); 567 568 *(--bp) = '/'; 569 error = getcwd_common(dvp, NULL, &bp, path, len / 2, 570 GETCWD_CHECK_ACCESS, curl); 571 vrele(dvp); 572 573 /* 574 * Strip off emulation path for emulated processes looking at 575 * the maps file of a process of the same emulation. (Won't 576 * work if /emul/xxx is a symlink..) 577 */ 578 if (curp->p_emul == p->p_emul && curp->p_emul->e_path != NULL) { 579 elen = strlen(curp->p_emul->e_path); 580 if (!strncmp(bp, curp->p_emul->e_path, elen)) 581 bp = &bp[elen]; 582 } 583 584 lenused = bend - bp; 585 586 memcpy(path, bp, lenused); 587 path[lenused] = 0; 588 589 return 0; 590 } 591