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