1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1992, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software donated to Berkeley by 8 * Jan-Simon Pendry. 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 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)fdesc_vnops.c 8.9 (Berkeley) 1/21/94 35 * 36 * $FreeBSD$ 37 */ 38 39 /* 40 * /dev/fd Filesystem 41 */ 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/capsicum.h> 46 #include <sys/conf.h> 47 #include <sys/dirent.h> 48 #include <sys/filedesc.h> 49 #include <sys/kernel.h> /* boottime */ 50 #include <sys/lock.h> 51 #include <sys/mutex.h> 52 #include <sys/malloc.h> 53 #include <sys/file.h> /* Must come after sys/malloc.h */ 54 #include <sys/mount.h> 55 #include <sys/namei.h> 56 #include <sys/proc.h> 57 #include <sys/stat.h> 58 #include <sys/syscallsubr.h> 59 #include <sys/unistd.h> 60 #include <sys/vnode.h> 61 62 #include <fs/fdescfs/fdesc.h> 63 64 #define NFDCACHE 4 65 #define FD_NHASH(ix) \ 66 (&fdhashtbl[(ix) & fdhash]) 67 static LIST_HEAD(fdhashhead, fdescnode) *fdhashtbl; 68 static u_long fdhash; 69 70 struct mtx fdesc_hashmtx; 71 72 static vop_getattr_t fdesc_getattr; 73 static vop_lookup_t fdesc_lookup; 74 static vop_open_t fdesc_open; 75 static vop_pathconf_t fdesc_pathconf; 76 static vop_readdir_t fdesc_readdir; 77 static vop_readlink_t fdesc_readlink; 78 static vop_reclaim_t fdesc_reclaim; 79 static vop_setattr_t fdesc_setattr; 80 81 static struct vop_vector fdesc_vnodeops = { 82 .vop_default = &default_vnodeops, 83 84 .vop_access = VOP_NULL, 85 .vop_getattr = fdesc_getattr, 86 .vop_lookup = fdesc_lookup, 87 .vop_open = fdesc_open, 88 .vop_pathconf = fdesc_pathconf, 89 .vop_readdir = fdesc_readdir, 90 .vop_readlink = fdesc_readlink, 91 .vop_reclaim = fdesc_reclaim, 92 .vop_setattr = fdesc_setattr, 93 }; 94 VFS_VOP_VECTOR_REGISTER(fdesc_vnodeops); 95 96 static void fdesc_remove_entry(struct fdescnode *); 97 98 /* 99 * Initialise cache headers 100 */ 101 int 102 fdesc_init(struct vfsconf *vfsp) 103 { 104 105 mtx_init(&fdesc_hashmtx, "fdescfs_hash", NULL, MTX_DEF); 106 fdhashtbl = hashinit(NFDCACHE, M_CACHE, &fdhash); 107 return (0); 108 } 109 110 /* 111 * Uninit ready for unload. 112 */ 113 int 114 fdesc_uninit(struct vfsconf *vfsp) 115 { 116 117 hashdestroy(fdhashtbl, M_CACHE, fdhash); 118 mtx_destroy(&fdesc_hashmtx); 119 return (0); 120 } 121 122 /* 123 * Remove an entry from the hash if it exists. 124 */ 125 static void 126 fdesc_remove_entry(struct fdescnode *fd) 127 { 128 struct fdhashhead *fc; 129 struct fdescnode *fd2; 130 131 fc = FD_NHASH(fd->fd_ix); 132 mtx_lock(&fdesc_hashmtx); 133 LIST_FOREACH(fd2, fc, fd_hash) { 134 if (fd == fd2) { 135 LIST_REMOVE(fd, fd_hash); 136 break; 137 } 138 } 139 mtx_unlock(&fdesc_hashmtx); 140 } 141 142 int 143 fdesc_allocvp(fdntype ftype, unsigned fd_fd, int ix, struct mount *mp, 144 struct vnode **vpp) 145 { 146 struct fdescmount *fmp; 147 struct fdhashhead *fc; 148 struct fdescnode *fd, *fd2; 149 struct vnode *vp, *vp2; 150 enum vgetstate vgs; 151 int error; 152 153 fc = FD_NHASH(ix); 154 loop: 155 mtx_lock(&fdesc_hashmtx); 156 /* 157 * If a forced unmount is progressing, we need to drop it. The flags are 158 * protected by the hashmtx. 159 */ 160 fmp = mp->mnt_data; 161 if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) { 162 mtx_unlock(&fdesc_hashmtx); 163 return (-1); 164 } 165 166 LIST_FOREACH(fd, fc, fd_hash) { 167 if (fd->fd_ix == ix && fd->fd_vnode->v_mount == mp) { 168 /* Get reference to vnode in case it's being free'd */ 169 vp = fd->fd_vnode; 170 vgs = vget_prep(vp); 171 mtx_unlock(&fdesc_hashmtx); 172 if (vget_finish(vp, LK_EXCLUSIVE, vgs) != 0) 173 goto loop; 174 *vpp = vp; 175 return (0); 176 } 177 } 178 mtx_unlock(&fdesc_hashmtx); 179 180 fd = malloc(sizeof(struct fdescnode), M_TEMP, M_WAITOK); 181 182 error = getnewvnode("fdescfs", mp, &fdesc_vnodeops, &vp); 183 if (error) { 184 free(fd, M_TEMP); 185 return (error); 186 } 187 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 188 vp->v_data = fd; 189 fd->fd_vnode = vp; 190 fd->fd_type = ftype; 191 fd->fd_fd = fd_fd; 192 fd->fd_ix = ix; 193 if (ftype == Fdesc) { 194 if ((fmp->flags & FMNT_RDLNKF) != 0) 195 vp->v_type = VLNK; 196 else if ((fmp->flags & FMNT_LINRDLNKF) != 0) 197 vp->v_vflag |= VV_READLINK; 198 } 199 error = insmntque1(vp, mp); 200 if (error != 0) { 201 vgone(vp); 202 vput(vp); 203 *vpp = NULLVP; 204 return (error); 205 } 206 207 /* Make sure that someone didn't beat us when inserting the vnode. */ 208 mtx_lock(&fdesc_hashmtx); 209 /* 210 * If a forced unmount is progressing, we need to drop it. The flags are 211 * protected by the hashmtx. 212 */ 213 fmp = mp->mnt_data; 214 if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) { 215 mtx_unlock(&fdesc_hashmtx); 216 vgone(vp); 217 vput(vp); 218 *vpp = NULLVP; 219 return (-1); 220 } 221 222 LIST_FOREACH(fd2, fc, fd_hash) { 223 if (fd2->fd_ix == ix && fd2->fd_vnode->v_mount == mp) { 224 /* Get reference to vnode in case it's being free'd */ 225 vp2 = fd2->fd_vnode; 226 vgs = vget_prep(vp2); 227 mtx_unlock(&fdesc_hashmtx); 228 error = vget_finish(vp2, LK_EXCLUSIVE, vgs); 229 /* Someone beat us, dec use count and wait for reclaim */ 230 vgone(vp); 231 vput(vp); 232 /* If we didn't get it, return no vnode. */ 233 if (error) 234 vp2 = NULLVP; 235 *vpp = vp2; 236 return (error); 237 } 238 } 239 240 /* If we came here, we can insert it safely. */ 241 LIST_INSERT_HEAD(fc, fd, fd_hash); 242 mtx_unlock(&fdesc_hashmtx); 243 vn_set_state(vp, VSTATE_CONSTRUCTED); 244 *vpp = vp; 245 return (0); 246 } 247 248 struct fdesc_get_ino_args { 249 fdntype ftype; 250 unsigned fd_fd; 251 int ix; 252 struct file *fp; 253 struct thread *td; 254 bool fdropped; 255 }; 256 257 static int 258 fdesc_get_ino_alloc(struct mount *mp, void *arg, int lkflags, 259 struct vnode **rvp) 260 { 261 struct fdesc_get_ino_args *a; 262 struct fdescmount *fdm; 263 struct vnode *vp; 264 int error; 265 266 a = arg; 267 fdm = VFSTOFDESC(mp); 268 if ((fdm->flags & FMNT_NODUP) != 0 && a->fp->f_type == DTYPE_VNODE) { 269 vp = a->fp->f_vnode; 270 vget(vp, lkflags | LK_RETRY); 271 *rvp = vp; 272 error = 0; 273 } else { 274 error = fdesc_allocvp(a->ftype, a->fd_fd, a->ix, mp, rvp); 275 } 276 fdrop(a->fp, a->td); 277 a->fdropped = true; 278 return (error); 279 } 280 281 /* 282 * vp is the current namei directory 283 * ndp is the name to locate in that directory... 284 */ 285 static int 286 fdesc_lookup(struct vop_lookup_args *ap) 287 { 288 struct vnode **vpp = ap->a_vpp; 289 struct vnode *dvp = ap->a_dvp; 290 struct componentname *cnp = ap->a_cnp; 291 char *pname = cnp->cn_nameptr; 292 struct thread *td = curthread; 293 struct file *fp; 294 struct fdesc_get_ino_args arg; 295 int nlen = cnp->cn_namelen; 296 u_int fd, fd1; 297 int error; 298 struct vnode *fvp; 299 300 if ((cnp->cn_flags & ISLASTCN) && 301 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 302 error = EROFS; 303 goto bad; 304 } 305 306 if (cnp->cn_namelen == 1 && *pname == '.') { 307 *vpp = dvp; 308 VREF(dvp); 309 return (0); 310 } 311 312 if (VTOFDESC(dvp)->fd_type != Froot) { 313 error = ENOTDIR; 314 goto bad; 315 } 316 317 fd = 0; 318 /* the only time a leading 0 is acceptable is if it's "0" */ 319 if (*pname == '0' && nlen != 1) { 320 error = ENOENT; 321 goto bad; 322 } 323 while (nlen--) { 324 if (*pname < '0' || *pname > '9') { 325 error = ENOENT; 326 goto bad; 327 } 328 fd1 = 10 * fd + *pname++ - '0'; 329 if (fd1 < fd) { 330 error = ENOENT; 331 goto bad; 332 } 333 fd = fd1; 334 } 335 336 /* 337 * No rights to check since 'fp' isn't actually used. 338 */ 339 if ((error = fget(td, fd, &cap_no_rights, &fp)) != 0) 340 goto bad; 341 342 /* 343 * Make sure we do not deadlock looking up the dvp itself. 344 * 345 * Unlock our root node (dvp) when doing this, since we might 346 * deadlock since the vnode might be locked by another thread 347 * and the root vnode lock will be obtained afterwards (in case 348 * we're looking up the fd of the root vnode), which will be the 349 * opposite lock order. 350 */ 351 arg.ftype = Fdesc; 352 arg.fd_fd = fd; 353 arg.ix = FD_DESC + fd; 354 arg.fp = fp; 355 arg.td = td; 356 arg.fdropped = false; 357 error = vn_vget_ino_gen(dvp, fdesc_get_ino_alloc, &arg, 358 LK_EXCLUSIVE, &fvp); 359 360 if (!arg.fdropped) { 361 /* 362 * In case we're holding the last reference to the file, the dvp 363 * will be re-acquired. 364 */ 365 VOP_UNLOCK(dvp); 366 fdrop(fp, td); 367 368 vn_lock(dvp, LK_RETRY | LK_EXCLUSIVE); 369 fvp = dvp; 370 if (error == 0 && VN_IS_DOOMED(dvp)) 371 error = ENOENT; 372 } 373 374 if (error) 375 goto bad; 376 *vpp = fvp; 377 return (0); 378 379 bad: 380 *vpp = NULL; 381 return (error); 382 } 383 384 static int 385 fdesc_open(struct vop_open_args *ap) 386 { 387 struct vnode *vp = ap->a_vp; 388 389 if (VTOFDESC(vp)->fd_type == Froot) 390 return (0); 391 392 /* 393 * XXX Kludge: set td->td_proc->p_dupfd to contain the value of the file 394 * descriptor being sought for duplication. The error return ensures 395 * that the vnode for this device will be released by vn_open. Open 396 * will detect this special error and take the actions in dupfdopen. 397 * Other callers of vn_open or VOP_OPEN will simply report the 398 * error. 399 */ 400 ap->a_td->td_dupfd = VTOFDESC(vp)->fd_fd; /* XXX */ 401 return (ENODEV); 402 } 403 404 static int 405 fdesc_pathconf(struct vop_pathconf_args *ap) 406 { 407 struct vnode *vp = ap->a_vp; 408 int error; 409 410 switch (ap->a_name) { 411 case _PC_NAME_MAX: 412 *ap->a_retval = NAME_MAX; 413 return (0); 414 case _PC_LINK_MAX: 415 if (VTOFDESC(vp)->fd_type == Froot) 416 *ap->a_retval = 2; 417 else 418 *ap->a_retval = 1; 419 return (0); 420 default: 421 if (VTOFDESC(vp)->fd_type == Froot) 422 return (vop_stdpathconf(ap)); 423 vref(vp); 424 VOP_UNLOCK(vp); 425 error = kern_fpathconf(curthread, VTOFDESC(vp)->fd_fd, 426 ap->a_name, ap->a_retval); 427 vn_lock(vp, LK_SHARED | LK_RETRY); 428 vunref(vp); 429 return (error); 430 } 431 } 432 433 static int 434 fdesc_getattr(struct vop_getattr_args *ap) 435 { 436 struct vnode *vp = ap->a_vp; 437 struct vattr *vap = ap->a_vap; 438 struct timeval boottime; 439 440 getboottime(&boottime); 441 vap->va_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH; 442 vap->va_fileid = VTOFDESC(vp)->fd_ix; 443 vap->va_uid = 0; 444 vap->va_gid = 0; 445 vap->va_blocksize = DEV_BSIZE; 446 vap->va_atime.tv_sec = boottime.tv_sec; 447 vap->va_atime.tv_nsec = 0; 448 vap->va_mtime = vap->va_atime; 449 vap->va_ctime = vap->va_mtime; 450 vap->va_gen = 0; 451 vap->va_flags = 0; 452 vap->va_bytes = 0; 453 vap->va_filerev = 0; 454 455 switch (VTOFDESC(vp)->fd_type) { 456 case Froot: 457 vap->va_type = VDIR; 458 vap->va_nlink = 2; 459 vap->va_size = DEV_BSIZE; 460 vap->va_rdev = NODEV; 461 break; 462 463 case Fdesc: 464 vap->va_type = (VFSTOFDESC(vp->v_mount)->flags & 465 (FMNT_RDLNKF | FMNT_LINRDLNKF)) == 0 ? VCHR : VLNK; 466 vap->va_nlink = 1; 467 vap->va_size = 0; 468 vap->va_rdev = makedev(0, vap->va_fileid); 469 break; 470 471 default: 472 panic("fdesc_getattr"); 473 break; 474 } 475 476 vp->v_type = vap->va_type; 477 return (0); 478 } 479 480 static int 481 fdesc_setattr(struct vop_setattr_args *ap) 482 { 483 struct vattr *vap = ap->a_vap; 484 struct vnode *vp; 485 struct mount *mp; 486 struct file *fp; 487 struct thread *td = curthread; 488 cap_rights_t rights; 489 unsigned fd; 490 int error; 491 492 /* 493 * Can't mess with the root vnode 494 */ 495 if (VTOFDESC(ap->a_vp)->fd_type == Froot) 496 return (EACCES); 497 498 fd = VTOFDESC(ap->a_vp)->fd_fd; 499 500 /* 501 * Allow setattr where there is an underlying vnode. 502 * For O_PATH descriptors, disallow truncate. 503 */ 504 if (vap->va_size != VNOVAL) { 505 error = getvnode(td, fd, 506 cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); 507 } else { 508 error = getvnode_path(td, fd, 509 cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); 510 } 511 if (error) { 512 /* 513 * getvnode() returns EINVAL if the file descriptor is not 514 * backed by a vnode. Silently drop all changes except 515 * chflags(2) in this case. 516 */ 517 if (error == EINVAL) { 518 if (vap->va_flags != VNOVAL) 519 error = EOPNOTSUPP; 520 else 521 error = 0; 522 } 523 return (error); 524 } 525 vp = fp->f_vnode; 526 if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) == 0) { 527 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 528 error = VOP_SETATTR(vp, ap->a_vap, ap->a_cred); 529 VOP_UNLOCK(vp); 530 vn_finished_write(mp); 531 } 532 fdrop(fp, td); 533 return (error); 534 } 535 536 #define UIO_MX _GENERIC_DIRLEN(10) /* number of symbols in INT_MAX printout */ 537 538 static int 539 fdesc_readdir(struct vop_readdir_args *ap) 540 { 541 struct fdescmount *fmp; 542 struct uio *uio = ap->a_uio; 543 struct filedesc *fdp; 544 struct dirent d; 545 struct dirent *dp = &d; 546 int error, i, off, fcnt; 547 548 if (VTOFDESC(ap->a_vp)->fd_type != Froot) 549 panic("fdesc_readdir: not dir"); 550 551 fmp = VFSTOFDESC(ap->a_vp->v_mount); 552 if (ap->a_ncookies != NULL) 553 *ap->a_ncookies = 0; 554 555 off = (int)uio->uio_offset; 556 if (off != uio->uio_offset || off < 0 || (u_int)off % UIO_MX != 0 || 557 uio->uio_resid < UIO_MX) 558 return (EINVAL); 559 i = (u_int)off / UIO_MX; 560 fdp = uio->uio_td->td_proc->p_fd; 561 error = 0; 562 563 fcnt = i - 2; /* The first two nodes are `.' and `..' */ 564 565 FILEDESC_SLOCK(fdp); 566 while (i < fdp->fd_nfiles + 2 && uio->uio_resid >= UIO_MX) { 567 bzero((caddr_t)dp, UIO_MX); 568 switch (i) { 569 case 0: /* `.' */ 570 case 1: /* `..' */ 571 dp->d_fileno = i + FD_ROOT; 572 dp->d_namlen = i + 1; 573 dp->d_reclen = UIO_MX; 574 bcopy("..", dp->d_name, dp->d_namlen); 575 dp->d_type = DT_DIR; 576 dirent_terminate(dp); 577 break; 578 default: 579 if (fdp->fd_ofiles[fcnt].fde_file == NULL) 580 break; 581 dp->d_namlen = sprintf(dp->d_name, "%d", fcnt); 582 dp->d_reclen = UIO_MX; 583 dp->d_type = (fmp->flags & (FMNT_RDLNKF | 584 FMNT_LINRDLNKF)) == 0 ? DT_CHR : DT_LNK; 585 dp->d_fileno = i + FD_DESC; 586 dirent_terminate(dp); 587 break; 588 } 589 /* NOTE: d_off is the offset of the *next* entry. */ 590 dp->d_off = UIO_MX * (i + 1); 591 if (dp->d_namlen != 0) { 592 /* 593 * And ship to userland 594 */ 595 FILEDESC_SUNLOCK(fdp); 596 error = uiomove(dp, UIO_MX, uio); 597 if (error) 598 goto done; 599 FILEDESC_SLOCK(fdp); 600 } 601 i++; 602 fcnt++; 603 } 604 FILEDESC_SUNLOCK(fdp); 605 606 done: 607 uio->uio_offset = i * UIO_MX; 608 return (error); 609 } 610 611 static int 612 fdesc_reclaim(struct vop_reclaim_args *ap) 613 { 614 struct vnode *vp; 615 struct fdescnode *fd; 616 617 vp = ap->a_vp; 618 fd = VTOFDESC(vp); 619 fdesc_remove_entry(fd); 620 free(vp->v_data, M_TEMP); 621 vp->v_data = NULL; 622 return (0); 623 } 624 625 static int 626 fdesc_readlink(struct vop_readlink_args *va) 627 { 628 struct vnode *vp, *vn; 629 struct thread *td; 630 struct uio *uio; 631 struct file *fp; 632 char *freepath, *fullpath; 633 size_t pathlen; 634 int lockflags, fd_fd; 635 int error; 636 637 freepath = NULL; 638 vn = va->a_vp; 639 if (VTOFDESC(vn)->fd_type != Fdesc) 640 panic("fdesc_readlink: not fdescfs link"); 641 fd_fd = ((struct fdescnode *)vn->v_data)->fd_fd; 642 lockflags = VOP_ISLOCKED(vn); 643 VOP_UNLOCK(vn); 644 645 td = curthread; 646 error = fget_cap(td, fd_fd, &cap_no_rights, &fp, NULL); 647 if (error != 0) 648 goto out; 649 650 switch (fp->f_type) { 651 case DTYPE_VNODE: 652 vp = fp->f_vnode; 653 error = vn_fullpath(vp, &fullpath, &freepath); 654 break; 655 default: 656 fullpath = "anon_inode:[unknown]"; 657 break; 658 } 659 if (error == 0) { 660 uio = va->a_uio; 661 pathlen = strlen(fullpath); 662 error = uiomove(fullpath, pathlen, uio); 663 } 664 if (freepath != NULL) 665 free(freepath, M_TEMP); 666 fdrop(fp, td); 667 668 out: 669 vn_lock(vn, lockflags | LK_RETRY); 670 return (error); 671 } 672