1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 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. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 39 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $ 40 * $DragonFly: src/sys/kern/kern_descrip.c,v 1.34 2004/11/24 22:51:01 joerg Exp $ 41 */ 42 43 #include "opt_compat.h" 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/malloc.h> 47 #include <sys/sysproto.h> 48 #include <sys/conf.h> 49 #include <sys/filedesc.h> 50 #include <sys/kernel.h> 51 #include <sys/sysctl.h> 52 #include <sys/vnode.h> 53 #include <sys/proc.h> 54 #include <sys/nlookup.h> 55 #include <sys/file.h> 56 #include <sys/stat.h> 57 #include <sys/filio.h> 58 #include <sys/fcntl.h> 59 #include <sys/unistd.h> 60 #include <sys/resourcevar.h> 61 #include <sys/event.h> 62 #include <sys/kern_syscall.h> 63 #include <sys/kcore.h> 64 #include <sys/kinfo.h> 65 66 #include <vm/vm.h> 67 #include <vm/vm_extern.h> 68 69 #include <sys/file2.h> 70 71 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 72 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 73 "file desc to leader structures"); 74 MALLOC_DEFINE(M_FILE, "file", "Open file structure"); 75 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 76 77 static d_open_t fdopen; 78 #define NUMFDESC 64 79 80 #define CDEV_MAJOR 22 81 static struct cdevsw fildesc_cdevsw = { 82 /* name */ "FD", 83 /* maj */ CDEV_MAJOR, 84 /* flags */ 0, 85 /* port */ NULL, 86 /* clone */ NULL, 87 88 /* open */ fdopen, 89 /* close */ noclose, 90 /* read */ noread, 91 /* write */ nowrite, 92 /* ioctl */ noioctl, 93 /* poll */ nopoll, 94 /* mmap */ nommap, 95 /* strategy */ nostrategy, 96 /* dump */ nodump, 97 /* psize */ nopsize 98 }; 99 100 static int badfo_readwrite (struct file *fp, struct uio *uio, 101 struct ucred *cred, int flags, struct thread *td); 102 static int badfo_ioctl (struct file *fp, u_long com, caddr_t data, 103 struct thread *td); 104 static int badfo_poll (struct file *fp, int events, 105 struct ucred *cred, struct thread *td); 106 static int badfo_kqfilter (struct file *fp, struct knote *kn); 107 static int badfo_stat (struct file *fp, struct stat *sb, struct thread *td); 108 static int badfo_close (struct file *fp, struct thread *td); 109 110 /* 111 * Descriptor management. 112 */ 113 struct filelist filehead; /* head of list of open files */ 114 int nfiles; /* actual number of open files */ 115 extern int cmask; 116 117 /* 118 * System calls on descriptors. 119 */ 120 /* ARGSUSED */ 121 int 122 getdtablesize(struct getdtablesize_args *uap) 123 { 124 struct proc *p = curproc; 125 126 uap->sysmsg_result = 127 min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 128 return (0); 129 } 130 131 /* 132 * Duplicate a file descriptor to a particular value. 133 * 134 * note: keep in mind that a potential race condition exists when closing 135 * descriptors from a shared descriptor table (via rfork). 136 */ 137 /* ARGSUSED */ 138 int 139 dup2(struct dup2_args *uap) 140 { 141 int error; 142 143 error = kern_dup(DUP_FIXED, uap->from, uap->to, uap->sysmsg_fds); 144 145 return (error); 146 } 147 148 /* 149 * Duplicate a file descriptor. 150 */ 151 /* ARGSUSED */ 152 int 153 dup(struct dup_args *uap) 154 { 155 int error; 156 157 error = kern_dup(DUP_VARIABLE, uap->fd, 0, uap->sysmsg_fds); 158 159 return (error); 160 } 161 162 int 163 kern_fcntl(int fd, int cmd, union fcntl_dat *dat) 164 { 165 struct thread *td = curthread; 166 struct proc *p = td->td_proc; 167 struct filedesc *fdp = p->p_fd; 168 struct file *fp; 169 char *pop; 170 struct vnode *vp; 171 u_int newmin; 172 int tmp, error, flg = F_POSIX; 173 174 KKASSERT(p); 175 176 if ((unsigned)fd >= fdp->fd_nfiles || 177 (fp = fdp->fd_ofiles[fd]) == NULL) 178 return (EBADF); 179 pop = &fdp->fd_ofileflags[fd]; 180 181 switch (cmd) { 182 case F_DUPFD: 183 newmin = dat->fc_fd; 184 if (newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 185 newmin > maxfilesperproc) 186 return (EINVAL); 187 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd); 188 return (error); 189 190 case F_GETFD: 191 dat->fc_cloexec = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 192 return (0); 193 194 case F_SETFD: 195 *pop = (*pop &~ UF_EXCLOSE) | 196 (dat->fc_cloexec & FD_CLOEXEC ? UF_EXCLOSE : 0); 197 return (0); 198 199 case F_GETFL: 200 dat->fc_flags = OFLAGS(fp->f_flag); 201 return (0); 202 203 case F_SETFL: 204 fhold(fp); 205 fp->f_flag &= ~FCNTLFLAGS; 206 fp->f_flag |= FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 207 tmp = fp->f_flag & FNONBLOCK; 208 error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td); 209 if (error) { 210 fdrop(fp, td); 211 return (error); 212 } 213 tmp = fp->f_flag & FASYNC; 214 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, td); 215 if (!error) { 216 fdrop(fp, td); 217 return (0); 218 } 219 fp->f_flag &= ~FNONBLOCK; 220 tmp = 0; 221 fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td); 222 fdrop(fp, td); 223 return (error); 224 225 case F_GETOWN: 226 fhold(fp); 227 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, td); 228 fdrop(fp, td); 229 return(error); 230 231 case F_SETOWN: 232 fhold(fp); 233 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, td); 234 fdrop(fp, td); 235 return(error); 236 237 case F_SETLKW: 238 flg |= F_WAIT; 239 /* Fall into F_SETLK */ 240 241 case F_SETLK: 242 if (fp->f_type != DTYPE_VNODE) 243 return (EBADF); 244 vp = (struct vnode *)fp->f_data; 245 246 /* 247 * copyin/lockop may block 248 */ 249 fhold(fp); 250 if (dat->fc_flock.l_whence == SEEK_CUR) 251 dat->fc_flock.l_start += fp->f_offset; 252 253 switch (dat->fc_flock.l_type) { 254 case F_RDLCK: 255 if ((fp->f_flag & FREAD) == 0) { 256 error = EBADF; 257 break; 258 } 259 p->p_leader->p_flag |= P_ADVLOCK; 260 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 261 &dat->fc_flock, flg); 262 break; 263 case F_WRLCK: 264 if ((fp->f_flag & FWRITE) == 0) { 265 error = EBADF; 266 break; 267 } 268 p->p_leader->p_flag |= P_ADVLOCK; 269 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 270 &dat->fc_flock, flg); 271 break; 272 case F_UNLCK: 273 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 274 &dat->fc_flock, F_POSIX); 275 break; 276 default: 277 error = EINVAL; 278 break; 279 } 280 /* Check for race with close */ 281 if ((unsigned) fd >= fdp->fd_nfiles || 282 fp != fdp->fd_ofiles[fd]) { 283 dat->fc_flock.l_whence = SEEK_SET; 284 dat->fc_flock.l_start = 0; 285 dat->fc_flock.l_len = 0; 286 dat->fc_flock.l_type = F_UNLCK; 287 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 288 F_UNLCK, &dat->fc_flock, F_POSIX); 289 } 290 fdrop(fp, td); 291 return(error); 292 293 case F_GETLK: 294 if (fp->f_type != DTYPE_VNODE) 295 return (EBADF); 296 vp = (struct vnode *)fp->f_data; 297 /* 298 * copyin/lockop may block 299 */ 300 fhold(fp); 301 if (dat->fc_flock.l_type != F_RDLCK && 302 dat->fc_flock.l_type != F_WRLCK && 303 dat->fc_flock.l_type != F_UNLCK) { 304 fdrop(fp, td); 305 return (EINVAL); 306 } 307 if (dat->fc_flock.l_whence == SEEK_CUR) 308 dat->fc_flock.l_start += fp->f_offset; 309 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 310 &dat->fc_flock, F_POSIX); 311 fdrop(fp, td); 312 return(error); 313 default: 314 return (EINVAL); 315 } 316 /* NOTREACHED */ 317 } 318 319 /* 320 * The file control system call. 321 */ 322 int 323 fcntl(struct fcntl_args *uap) 324 { 325 union fcntl_dat dat; 326 int error; 327 328 switch (uap->cmd) { 329 case F_DUPFD: 330 dat.fc_fd = uap->arg; 331 break; 332 case F_SETFD: 333 dat.fc_cloexec = uap->arg; 334 break; 335 case F_SETFL: 336 dat.fc_flags = uap->arg; 337 break; 338 case F_SETOWN: 339 dat.fc_owner = uap->arg; 340 break; 341 case F_SETLKW: 342 case F_SETLK: 343 case F_GETLK: 344 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 345 sizeof(struct flock)); 346 if (error) 347 return (error); 348 break; 349 } 350 351 error = kern_fcntl(uap->fd, uap->cmd, &dat); 352 353 if (error == 0) { 354 switch (uap->cmd) { 355 case F_DUPFD: 356 uap->sysmsg_result = dat.fc_fd; 357 break; 358 case F_GETFD: 359 uap->sysmsg_result = dat.fc_cloexec; 360 break; 361 case F_GETFL: 362 uap->sysmsg_result = dat.fc_flags; 363 break; 364 case F_GETOWN: 365 uap->sysmsg_result = dat.fc_owner; 366 case F_GETLK: 367 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 368 sizeof(struct flock)); 369 break; 370 } 371 } 372 373 return (error); 374 } 375 376 /* 377 * Common code for dup, dup2, and fcntl(F_DUPFD). 378 * 379 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells 380 * kern_dup() to destructively dup over an existing file descriptor if new 381 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest 382 * unused file descriptor that is greater than or equal to new. 383 */ 384 int 385 kern_dup(enum dup_type type, int old, int new, int *res) 386 { 387 struct thread *td = curthread; 388 struct proc *p = td->td_proc; 389 struct filedesc *fdp = p->p_fd; 390 struct file *fp; 391 struct file *delfp; 392 int holdleaders; 393 int error, newfd; 394 395 /* 396 * Verify that we have a valid descriptor to dup from and 397 * possibly to dup to. 398 */ 399 if (old < 0 || new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 400 new >= maxfilesperproc) 401 return (EBADF); 402 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) 403 return (EBADF); 404 if (type == DUP_FIXED && old == new) { 405 *res = new; 406 return (0); 407 } 408 fp = fdp->fd_ofiles[old]; 409 fhold(fp); 410 411 /* 412 * Expand the table for the new descriptor if needed. This may 413 * block and drop and reacquire the fidedesc lock. 414 */ 415 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) { 416 error = fdalloc(p, new, &newfd); 417 if (error) { 418 fdrop(fp, td); 419 return (error); 420 } 421 } 422 if (type == DUP_VARIABLE) 423 new = newfd; 424 425 /* 426 * If the old file changed out from under us then treat it as a 427 * bad file descriptor. Userland should do its own locking to 428 * avoid this case. 429 */ 430 if (fdp->fd_ofiles[old] != fp) { 431 if (fdp->fd_ofiles[new] == NULL) { 432 if (new < fdp->fd_freefile) 433 fdp->fd_freefile = new; 434 while (fdp->fd_lastfile > 0 && 435 fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 436 fdp->fd_lastfile--; 437 } 438 fdrop(fp, td); 439 return (EBADF); 440 } 441 KASSERT(old != new, ("new fd is same as old")); 442 443 /* 444 * Save info on the descriptor being overwritten. We have 445 * to do the unmap now, but we cannot close it without 446 * introducing an ownership race for the slot. 447 */ 448 delfp = fdp->fd_ofiles[new]; 449 if (delfp != NULL && p->p_fdtol != NULL) { 450 /* 451 * Ask fdfree() to sleep to ensure that all relevant 452 * process leaders can be traversed in closef(). 453 */ 454 fdp->fd_holdleaderscount++; 455 holdleaders = 1; 456 } else 457 holdleaders = 0; 458 KASSERT(delfp == NULL || type == DUP_FIXED, 459 ("dup() picked an open file")); 460 #if 0 461 if (delfp && (fdp->fd_ofileflags[new] & UF_MAPPED)) 462 (void) munmapfd(p, new); 463 #endif 464 465 /* 466 * Duplicate the source descriptor, update lastfile 467 */ 468 fdp->fd_ofiles[new] = fp; 469 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 470 if (new > fdp->fd_lastfile) 471 fdp->fd_lastfile = new; 472 *res = new; 473 474 /* 475 * If we dup'd over a valid file, we now own the reference to it 476 * and must dispose of it using closef() semantics (as if a 477 * close() were performed on it). 478 */ 479 if (delfp) { 480 (void) closef(delfp, td); 481 if (holdleaders) { 482 fdp->fd_holdleaderscount--; 483 if (fdp->fd_holdleaderscount == 0 && 484 fdp->fd_holdleaderswakeup != 0) { 485 fdp->fd_holdleaderswakeup = 0; 486 wakeup(&fdp->fd_holdleaderscount); 487 } 488 } 489 } 490 return (0); 491 } 492 493 /* 494 * If sigio is on the list associated with a process or process group, 495 * disable signalling from the device, remove sigio from the list and 496 * free sigio. 497 */ 498 void 499 funsetown(struct sigio *sigio) 500 { 501 int s; 502 503 if (sigio == NULL) 504 return; 505 s = splhigh(); 506 *(sigio->sio_myref) = NULL; 507 splx(s); 508 if (sigio->sio_pgid < 0) { 509 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 510 sigio, sio_pgsigio); 511 } else /* if ((*sigiop)->sio_pgid > 0) */ { 512 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 513 sigio, sio_pgsigio); 514 } 515 crfree(sigio->sio_ucred); 516 free(sigio, M_SIGIO); 517 } 518 519 /* Free a list of sigio structures. */ 520 void 521 funsetownlst(struct sigiolst *sigiolst) 522 { 523 struct sigio *sigio; 524 525 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) 526 funsetown(sigio); 527 } 528 529 /* 530 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 531 * 532 * After permission checking, add a sigio structure to the sigio list for 533 * the process or process group. 534 */ 535 int 536 fsetown(pid_t pgid, struct sigio **sigiop) 537 { 538 struct proc *proc; 539 struct pgrp *pgrp; 540 struct sigio *sigio; 541 int s; 542 543 if (pgid == 0) { 544 funsetown(*sigiop); 545 return (0); 546 } 547 if (pgid > 0) { 548 proc = pfind(pgid); 549 if (proc == NULL) 550 return (ESRCH); 551 552 /* 553 * Policy - Don't allow a process to FSETOWN a process 554 * in another session. 555 * 556 * Remove this test to allow maximum flexibility or 557 * restrict FSETOWN to the current process or process 558 * group for maximum safety. 559 */ 560 if (proc->p_session != curproc->p_session) 561 return (EPERM); 562 563 pgrp = NULL; 564 } else /* if (pgid < 0) */ { 565 pgrp = pgfind(-pgid); 566 if (pgrp == NULL) 567 return (ESRCH); 568 569 /* 570 * Policy - Don't allow a process to FSETOWN a process 571 * in another session. 572 * 573 * Remove this test to allow maximum flexibility or 574 * restrict FSETOWN to the current process or process 575 * group for maximum safety. 576 */ 577 if (pgrp->pg_session != curproc->p_session) 578 return (EPERM); 579 580 proc = NULL; 581 } 582 funsetown(*sigiop); 583 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 584 if (pgid > 0) { 585 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 586 sigio->sio_proc = proc; 587 } else { 588 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 589 sigio->sio_pgrp = pgrp; 590 } 591 sigio->sio_pgid = pgid; 592 sigio->sio_ucred = crhold(curproc->p_ucred); 593 /* It would be convenient if p_ruid was in ucred. */ 594 sigio->sio_ruid = curproc->p_ucred->cr_ruid; 595 sigio->sio_myref = sigiop; 596 s = splhigh(); 597 *sigiop = sigio; 598 splx(s); 599 return (0); 600 } 601 602 /* 603 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 604 */ 605 pid_t 606 fgetown(struct sigio *sigio) 607 { 608 return (sigio != NULL ? sigio->sio_pgid : 0); 609 } 610 611 /* 612 * Close a file descriptor. 613 */ 614 /* ARGSUSED */ 615 int 616 close(struct close_args *uap) 617 { 618 return(kern_close(uap->fd)); 619 } 620 621 int 622 kern_close(int fd) 623 { 624 struct thread *td = curthread; 625 struct proc *p = td->td_proc; 626 struct filedesc *fdp; 627 struct file *fp; 628 int error; 629 int holdleaders; 630 631 KKASSERT(p); 632 fdp = p->p_fd; 633 634 if ((unsigned)fd >= fdp->fd_nfiles || 635 (fp = fdp->fd_ofiles[fd]) == NULL) 636 return (EBADF); 637 #if 0 638 if (fdp->fd_ofileflags[fd] & UF_MAPPED) 639 (void) munmapfd(p, fd); 640 #endif 641 fdp->fd_ofiles[fd] = NULL; 642 fdp->fd_ofileflags[fd] = 0; 643 holdleaders = 0; 644 if (p->p_fdtol != NULL) { 645 /* 646 * Ask fdfree() to sleep to ensure that all relevant 647 * process leaders can be traversed in closef(). 648 */ 649 fdp->fd_holdleaderscount++; 650 holdleaders = 1; 651 } 652 653 /* 654 * we now hold the fp reference that used to be owned by the descriptor 655 * array. 656 */ 657 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 658 fdp->fd_lastfile--; 659 if (fd < fdp->fd_freefile) 660 fdp->fd_freefile = fd; 661 if (fd < fdp->fd_knlistsize) 662 knote_fdclose(p, fd); 663 error = closef(fp, td); 664 if (holdleaders) { 665 fdp->fd_holdleaderscount--; 666 if (fdp->fd_holdleaderscount == 0 && 667 fdp->fd_holdleaderswakeup != 0) { 668 fdp->fd_holdleaderswakeup = 0; 669 wakeup(&fdp->fd_holdleaderscount); 670 } 671 } 672 return (error); 673 } 674 675 int 676 kern_fstat(int fd, struct stat *ub) 677 { 678 struct thread *td = curthread; 679 struct proc *p = td->td_proc; 680 struct filedesc *fdp; 681 struct file *fp; 682 int error; 683 684 KKASSERT(p); 685 686 fdp = p->p_fd; 687 if ((unsigned)fd >= fdp->fd_nfiles || 688 (fp = fdp->fd_ofiles[fd]) == NULL) 689 return (EBADF); 690 fhold(fp); 691 error = fo_stat(fp, ub, td); 692 fdrop(fp, td); 693 694 return (error); 695 } 696 697 /* 698 * Return status information about a file descriptor. 699 */ 700 int 701 fstat(struct fstat_args *uap) 702 { 703 struct stat st; 704 int error; 705 706 error = kern_fstat(uap->fd, &st); 707 708 if (error == 0) 709 error = copyout(&st, uap->sb, sizeof(st)); 710 return (error); 711 } 712 713 /* 714 * XXX: This is for source compatibility with NetBSD. Probably doesn't 715 * belong here. 716 */ 717 int 718 nfstat(struct nfstat_args *uap) 719 { 720 struct stat st; 721 struct nstat nst; 722 int error; 723 724 error = kern_fstat(uap->fd, &st); 725 726 if (error == 0) { 727 cvtnstat(&st, &nst); 728 error = copyout(&nst, uap->sb, sizeof (nst)); 729 } 730 return (error); 731 } 732 733 /* 734 * Return pathconf information about a file descriptor. 735 */ 736 /* ARGSUSED */ 737 int 738 fpathconf(struct fpathconf_args *uap) 739 { 740 struct thread *td = curthread; 741 struct proc *p = td->td_proc; 742 struct filedesc *fdp; 743 struct file *fp; 744 struct vnode *vp; 745 int error = 0; 746 747 KKASSERT(p); 748 fdp = p->p_fd; 749 if ((unsigned)uap->fd >= fdp->fd_nfiles || 750 (fp = fdp->fd_ofiles[uap->fd]) == NULL) 751 return (EBADF); 752 753 fhold(fp); 754 755 switch (fp->f_type) { 756 case DTYPE_PIPE: 757 case DTYPE_SOCKET: 758 if (uap->name != _PC_PIPE_BUF) { 759 error = EINVAL; 760 } else { 761 uap->sysmsg_result = PIPE_BUF; 762 error = 0; 763 } 764 break; 765 case DTYPE_FIFO: 766 case DTYPE_VNODE: 767 vp = (struct vnode *)fp->f_data; 768 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds); 769 break; 770 default: 771 error = EOPNOTSUPP; 772 break; 773 } 774 fdrop(fp, td); 775 return(error); 776 } 777 778 /* 779 * Allocate a file descriptor for the process. 780 */ 781 static int fdexpand; 782 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, ""); 783 784 int 785 fdalloc(struct proc *p, int want, int *result) 786 { 787 struct filedesc *fdp = p->p_fd; 788 int i; 789 int lim, last, nfiles; 790 struct file **newofile; 791 char *newofileflags; 792 793 /* 794 * Search for a free descriptor starting at the higher 795 * of want or fd_freefile. If that fails, consider 796 * expanding the ofile array. 797 */ 798 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 799 for (;;) { 800 last = min(fdp->fd_nfiles, lim); 801 if ((i = want) < fdp->fd_freefile) 802 i = fdp->fd_freefile; 803 for (; i < last; i++) { 804 if (fdp->fd_ofiles[i] == NULL) { 805 fdp->fd_ofileflags[i] = 0; 806 if (i > fdp->fd_lastfile) 807 fdp->fd_lastfile = i; 808 if (want <= fdp->fd_freefile) 809 fdp->fd_freefile = i; 810 *result = i; 811 return (0); 812 } 813 } 814 815 /* 816 * No space in current array. Expand? 817 */ 818 if (fdp->fd_nfiles >= lim) 819 return (EMFILE); 820 if (fdp->fd_nfiles < NDEXTENT) 821 nfiles = NDEXTENT; 822 else 823 nfiles = 2 * fdp->fd_nfiles; 824 newofile = malloc(nfiles * OFILESIZE, M_FILEDESC, M_WAITOK); 825 826 /* 827 * deal with file-table extend race that might have occured 828 * when malloc was blocked. 829 */ 830 if (fdp->fd_nfiles >= nfiles) { 831 free(newofile, M_FILEDESC); 832 continue; 833 } 834 newofileflags = (char *) &newofile[nfiles]; 835 /* 836 * Copy the existing ofile and ofileflags arrays 837 * and zero the new portion of each array. 838 */ 839 bcopy(fdp->fd_ofiles, newofile, 840 (i = sizeof(struct file *) * fdp->fd_nfiles)); 841 bzero((char *)newofile + i, nfiles * sizeof(struct file *) - i); 842 bcopy(fdp->fd_ofileflags, newofileflags, 843 (i = sizeof(char) * fdp->fd_nfiles)); 844 bzero(newofileflags + i, nfiles * sizeof(char) - i); 845 if (fdp->fd_nfiles > NDFILE) 846 free(fdp->fd_ofiles, M_FILEDESC); 847 fdp->fd_ofiles = newofile; 848 fdp->fd_ofileflags = newofileflags; 849 fdp->fd_nfiles = nfiles; 850 fdexpand++; 851 } 852 return (0); 853 } 854 855 /* 856 * Check to see whether n user file descriptors 857 * are available to the process p. 858 */ 859 int 860 fdavail(struct proc *p, int n) 861 { 862 struct filedesc *fdp = p->p_fd; 863 struct file **fpp; 864 int i, lim, last; 865 866 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 867 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 868 return (1); 869 870 last = min(fdp->fd_nfiles, lim); 871 fpp = &fdp->fd_ofiles[fdp->fd_freefile]; 872 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) { 873 if (*fpp == NULL && --n <= 0) 874 return (1); 875 } 876 return (0); 877 } 878 879 /* 880 * falloc: 881 * Create a new open file structure and allocate a file decriptor 882 * for the process that refers to it. If p is NULL, no descriptor 883 * is allocated and the file pointer is returned unassociated with 884 * any process. resultfd is only used if p is not NULL and may 885 * separately be NULL indicating that you don't need the returned fd. 886 * 887 * A held file pointer is returned. If a descriptor has been allocated 888 * an additional hold on the fp will be made due to the fd_ofiles[] 889 * reference. 890 */ 891 int 892 falloc(struct proc *p, struct file **resultfp, int *resultfd) 893 { 894 static struct timeval lastfail; 895 static int curfail; 896 struct file *fp; 897 int error; 898 899 fp = NULL; 900 901 /* 902 * Handle filetable full issues and root overfill. 903 */ 904 if (nfiles >= maxfiles - maxfilesrootres && 905 ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) { 906 if (ppsratecheck(&lastfail, &curfail, 1)) { 907 printf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n", 908 (p ? p->p_ucred->cr_ruid : -1)); 909 } 910 error = ENFILE; 911 goto done; 912 } 913 914 /* 915 * Allocate a new file descriptor. 916 */ 917 nfiles++; 918 fp = malloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO); 919 fp->f_count = 1; 920 fp->f_ops = &badfileops; 921 fp->f_seqcount = 1; 922 if (p) 923 fp->f_cred = crhold(p->p_ucred); 924 else 925 fp->f_cred = crhold(proc0.p_ucred); 926 LIST_INSERT_HEAD(&filehead, fp, f_list); 927 if (resultfd) { 928 if ((error = fsetfd(p, fp, resultfd)) != 0) { 929 fdrop(fp, p->p_thread); 930 fp = NULL; 931 } 932 } else { 933 error = 0; 934 } 935 done: 936 *resultfp = fp; 937 return (0); 938 } 939 940 /* 941 * Associate a file pointer with a file descriptor. On success the fp 942 * will have an additional ref representing the fd_ofiles[] association. 943 */ 944 int 945 fsetfd(struct proc *p, struct file *fp, int *resultfd) 946 { 947 int i; 948 int error; 949 950 KKASSERT(p); 951 952 i = -1; 953 if ((error = fdalloc(p, 0, &i)) == 0) { 954 fhold(fp); 955 p->p_fd->fd_ofiles[i] = fp; 956 } 957 *resultfd = i; 958 return (0); 959 } 960 961 void 962 fsetcred(struct file *fp, struct ucred *cr) 963 { 964 crhold(cr); 965 crfree(fp->f_cred); 966 fp->f_cred = cr; 967 } 968 969 /* 970 * Free a file descriptor. 971 */ 972 void 973 ffree(struct file *fp) 974 { 975 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 976 LIST_REMOVE(fp, f_list); 977 crfree(fp->f_cred); 978 if (fp->f_ncp) { 979 cache_drop(fp->f_ncp); 980 fp->f_ncp = NULL; 981 } 982 nfiles--; 983 free(fp, M_FILE); 984 } 985 986 /* 987 * Build a new filedesc structure. 988 */ 989 struct filedesc * 990 fdinit(struct proc *p) 991 { 992 struct filedesc0 *newfdp; 993 struct filedesc *fdp = p->p_fd; 994 995 newfdp = malloc(sizeof(struct filedesc0), M_FILEDESC, M_WAITOK|M_ZERO); 996 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 997 if (newfdp->fd_fd.fd_cdir) { 998 vref(newfdp->fd_fd.fd_cdir); 999 newfdp->fd_fd.fd_ncdir = cache_hold(fdp->fd_ncdir); 1000 } 1001 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1002 newfdp->fd_fd.fd_nrdir = cache_hold(fdp->fd_nrdir); 1003 vref(newfdp->fd_fd.fd_rdir); 1004 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1005 if (newfdp->fd_fd.fd_jdir) { 1006 vref(newfdp->fd_fd.fd_jdir); 1007 newfdp->fd_fd.fd_njdir = cache_hold(fdp->fd_njdir); 1008 } 1009 1010 1011 /* Create the file descriptor table. */ 1012 newfdp->fd_fd.fd_refcnt = 1; 1013 newfdp->fd_fd.fd_cmask = cmask; 1014 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1015 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1016 newfdp->fd_fd.fd_nfiles = NDFILE; 1017 newfdp->fd_fd.fd_knlistsize = -1; 1018 1019 return (&newfdp->fd_fd); 1020 } 1021 1022 /* 1023 * Share a filedesc structure. 1024 */ 1025 struct filedesc * 1026 fdshare(struct proc *p) 1027 { 1028 p->p_fd->fd_refcnt++; 1029 return (p->p_fd); 1030 } 1031 1032 /* 1033 * Copy a filedesc structure. 1034 */ 1035 struct filedesc * 1036 fdcopy(struct proc *p) 1037 { 1038 struct filedesc *newfdp, *fdp = p->p_fd; 1039 struct file **fpp; 1040 int i; 1041 1042 /* Certain daemons might not have file descriptors. */ 1043 if (fdp == NULL) 1044 return (NULL); 1045 1046 newfdp = malloc(sizeof(struct filedesc0), M_FILEDESC, M_WAITOK); 1047 bcopy(fdp, newfdp, sizeof(struct filedesc)); 1048 if (newfdp->fd_cdir) { 1049 vref(newfdp->fd_cdir); 1050 newfdp->fd_ncdir = cache_hold(newfdp->fd_ncdir); 1051 } 1052 /* 1053 * We must check for fd_rdir here, at least for now because 1054 * the init process is created before we have access to the 1055 * rootvode to take a reference to it. 1056 */ 1057 if (newfdp->fd_rdir) { 1058 vref(newfdp->fd_rdir); 1059 newfdp->fd_nrdir = cache_hold(newfdp->fd_nrdir); 1060 } 1061 if (newfdp->fd_jdir) { 1062 vref(newfdp->fd_jdir); 1063 newfdp->fd_njdir = cache_hold(newfdp->fd_njdir); 1064 } 1065 newfdp->fd_refcnt = 1; 1066 1067 /* 1068 * If the number of open files fits in the internal arrays 1069 * of the open file structure, use them, otherwise allocate 1070 * additional memory for the number of descriptors currently 1071 * in use. 1072 */ 1073 if (newfdp->fd_lastfile < NDFILE) { 1074 newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles; 1075 newfdp->fd_ofileflags = 1076 ((struct filedesc0 *) newfdp)->fd_dfileflags; 1077 i = NDFILE; 1078 } else { 1079 /* 1080 * Compute the smallest multiple of NDEXTENT needed 1081 * for the file descriptors currently in use, 1082 * allowing the table to shrink. 1083 */ 1084 i = newfdp->fd_nfiles; 1085 while (i > 2 * NDEXTENT && i > newfdp->fd_lastfile * 2) 1086 i /= 2; 1087 newfdp->fd_ofiles = malloc(i * OFILESIZE, M_FILEDESC, M_WAITOK); 1088 newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i]; 1089 } 1090 newfdp->fd_nfiles = i; 1091 bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **)); 1092 bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char)); 1093 1094 /* 1095 * kq descriptors cannot be copied. 1096 */ 1097 if (newfdp->fd_knlistsize != -1) { 1098 fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile]; 1099 for (i = newfdp->fd_lastfile; i >= 0; i--, fpp--) { 1100 if (*fpp != NULL && (*fpp)->f_type == DTYPE_KQUEUE) { 1101 *fpp = NULL; 1102 if (i < newfdp->fd_freefile) 1103 newfdp->fd_freefile = i; 1104 } 1105 if (*fpp == NULL && i == newfdp->fd_lastfile && i > 0) 1106 newfdp->fd_lastfile--; 1107 } 1108 newfdp->fd_knlist = NULL; 1109 newfdp->fd_knlistsize = -1; 1110 newfdp->fd_knhash = NULL; 1111 newfdp->fd_knhashmask = 0; 1112 } 1113 1114 fpp = newfdp->fd_ofiles; 1115 for (i = newfdp->fd_lastfile; i-- >= 0; fpp++) { 1116 if (*fpp != NULL) 1117 fhold(*fpp); 1118 } 1119 return (newfdp); 1120 } 1121 1122 /* 1123 * Release a filedesc structure. 1124 */ 1125 void 1126 fdfree(struct proc *p) 1127 { 1128 struct thread *td = p->p_thread; 1129 struct filedesc *fdp = p->p_fd; 1130 struct file **fpp; 1131 int i; 1132 struct filedesc_to_leader *fdtol; 1133 struct file *fp; 1134 struct vnode *vp; 1135 struct flock lf; 1136 1137 /* Certain daemons might not have file descriptors. */ 1138 if (fdp == NULL) 1139 return; 1140 1141 /* Check for special need to clear POSIX style locks */ 1142 fdtol = p->p_fdtol; 1143 if (fdtol != NULL) { 1144 KASSERT(fdtol->fdl_refcount > 0, 1145 ("filedesc_to_refcount botch: fdl_refcount=%d", 1146 fdtol->fdl_refcount)); 1147 if (fdtol->fdl_refcount == 1 && 1148 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1149 i = 0; 1150 fpp = fdp->fd_ofiles; 1151 for (i = 0, fpp = fdp->fd_ofiles; 1152 i <= fdp->fd_lastfile; 1153 i++, fpp++) { 1154 if (*fpp == NULL || 1155 (*fpp)->f_type != DTYPE_VNODE) 1156 continue; 1157 fp = *fpp; 1158 fhold(fp); 1159 lf.l_whence = SEEK_SET; 1160 lf.l_start = 0; 1161 lf.l_len = 0; 1162 lf.l_type = F_UNLCK; 1163 vp = (struct vnode *)fp->f_data; 1164 (void) VOP_ADVLOCK(vp, 1165 (caddr_t)p->p_leader, 1166 F_UNLCK, 1167 &lf, 1168 F_POSIX); 1169 fdrop(fp, p->p_thread); 1170 fpp = fdp->fd_ofiles + i; 1171 } 1172 } 1173 retry: 1174 if (fdtol->fdl_refcount == 1) { 1175 if (fdp->fd_holdleaderscount > 0 && 1176 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1177 /* 1178 * close() or do_dup() has cleared a reference 1179 * in a shared file descriptor table. 1180 */ 1181 fdp->fd_holdleaderswakeup = 1; 1182 tsleep(&fdp->fd_holdleaderscount, 1183 0, "fdlhold", 0); 1184 goto retry; 1185 } 1186 if (fdtol->fdl_holdcount > 0) { 1187 /* 1188 * Ensure that fdtol->fdl_leader 1189 * remains valid in closef(). 1190 */ 1191 fdtol->fdl_wakeup = 1; 1192 tsleep(fdtol, 0, "fdlhold", 0); 1193 goto retry; 1194 } 1195 } 1196 fdtol->fdl_refcount--; 1197 if (fdtol->fdl_refcount == 0 && 1198 fdtol->fdl_holdcount == 0) { 1199 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1200 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1201 } else 1202 fdtol = NULL; 1203 p->p_fdtol = NULL; 1204 if (fdtol != NULL) 1205 free(fdtol, M_FILEDESC_TO_LEADER); 1206 } 1207 if (--fdp->fd_refcnt > 0) 1208 return; 1209 /* 1210 * we are the last reference to the structure, we can 1211 * safely assume it will not change out from under us. 1212 */ 1213 fpp = fdp->fd_ofiles; 1214 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) { 1215 if (*fpp) 1216 (void) closef(*fpp, td); 1217 } 1218 if (fdp->fd_nfiles > NDFILE) 1219 free(fdp->fd_ofiles, M_FILEDESC); 1220 if (fdp->fd_cdir) { 1221 cache_drop(fdp->fd_ncdir); 1222 vrele(fdp->fd_cdir); 1223 } 1224 cache_drop(fdp->fd_nrdir); 1225 vrele(fdp->fd_rdir); 1226 if (fdp->fd_jdir) { 1227 cache_drop(fdp->fd_njdir); 1228 vrele(fdp->fd_jdir); 1229 } 1230 if (fdp->fd_knlist) 1231 free(fdp->fd_knlist, M_KQUEUE); 1232 if (fdp->fd_knhash) 1233 free(fdp->fd_knhash, M_KQUEUE); 1234 free(fdp, M_FILEDESC); 1235 } 1236 1237 /* 1238 * For setugid programs, we don't want to people to use that setugidness 1239 * to generate error messages which write to a file which otherwise would 1240 * otherwise be off-limits to the process. 1241 * 1242 * This is a gross hack to plug the hole. A better solution would involve 1243 * a special vop or other form of generalized access control mechanism. We 1244 * go ahead and just reject all procfs file systems accesses as dangerous. 1245 * 1246 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1247 * sufficient. We also don't for check setugidness since we know we are. 1248 */ 1249 static int 1250 is_unsafe(struct file *fp) 1251 { 1252 if (fp->f_type == DTYPE_VNODE && 1253 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 1254 return (1); 1255 return (0); 1256 } 1257 1258 /* 1259 * Make this setguid thing safe, if at all possible. 1260 */ 1261 void 1262 setugidsafety(struct proc *p) 1263 { 1264 struct thread *td = p->p_thread; 1265 struct filedesc *fdp = p->p_fd; 1266 int i; 1267 1268 /* Certain daemons might not have file descriptors. */ 1269 if (fdp == NULL) 1270 return; 1271 1272 /* 1273 * note: fdp->fd_ofiles may be reallocated out from under us while 1274 * we are blocked in a close. Be careful! 1275 */ 1276 for (i = 0; i <= fdp->fd_lastfile; i++) { 1277 if (i > 2) 1278 break; 1279 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 1280 struct file *fp; 1281 1282 #if 0 1283 if ((fdp->fd_ofileflags[i] & UF_MAPPED) != 0) 1284 (void) munmapfd(p, i); 1285 #endif 1286 if (i < fdp->fd_knlistsize) 1287 knote_fdclose(p, i); 1288 /* 1289 * NULL-out descriptor prior to close to avoid 1290 * a race while close blocks. 1291 */ 1292 fp = fdp->fd_ofiles[i]; 1293 fdp->fd_ofiles[i] = NULL; 1294 fdp->fd_ofileflags[i] = 0; 1295 if (i < fdp->fd_freefile) 1296 fdp->fd_freefile = i; 1297 (void) closef(fp, td); 1298 } 1299 } 1300 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 1301 fdp->fd_lastfile--; 1302 } 1303 1304 /* 1305 * Close any files on exec? 1306 */ 1307 void 1308 fdcloseexec(struct proc *p) 1309 { 1310 struct thread *td = p->p_thread; 1311 struct filedesc *fdp = p->p_fd; 1312 int i; 1313 1314 /* Certain daemons might not have file descriptors. */ 1315 if (fdp == NULL) 1316 return; 1317 1318 /* 1319 * We cannot cache fd_ofiles or fd_ofileflags since operations 1320 * may block and rip them out from under us. 1321 */ 1322 for (i = 0; i <= fdp->fd_lastfile; i++) { 1323 if (fdp->fd_ofiles[i] != NULL && 1324 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) { 1325 struct file *fp; 1326 1327 #if 0 1328 if (fdp->fd_ofileflags[i] & UF_MAPPED) 1329 (void) munmapfd(p, i); 1330 #endif 1331 if (i < fdp->fd_knlistsize) 1332 knote_fdclose(p, i); 1333 /* 1334 * NULL-out descriptor prior to close to avoid 1335 * a race while close blocks. 1336 */ 1337 fp = fdp->fd_ofiles[i]; 1338 fdp->fd_ofiles[i] = NULL; 1339 fdp->fd_ofileflags[i] = 0; 1340 if (i < fdp->fd_freefile) 1341 fdp->fd_freefile = i; 1342 (void) closef(fp, td); 1343 } 1344 } 1345 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 1346 fdp->fd_lastfile--; 1347 } 1348 1349 /* 1350 * It is unsafe for set[ug]id processes to be started with file 1351 * descriptors 0..2 closed, as these descriptors are given implicit 1352 * significance in the Standard C library. fdcheckstd() will create a 1353 * descriptor referencing /dev/null for each of stdin, stdout, and 1354 * stderr that is not already open. 1355 */ 1356 int 1357 fdcheckstd(struct proc *p) 1358 { 1359 struct thread *td = p->p_thread; 1360 struct nlookupdata nd; 1361 struct filedesc *fdp; 1362 struct file *fp; 1363 register_t retval; 1364 int fd, i, error, flags, devnull; 1365 1366 fdp = p->p_fd; 1367 if (fdp == NULL) 1368 return (0); 1369 devnull = -1; 1370 error = 0; 1371 for (i = 0; i < 3; i++) { 1372 if (fdp->fd_ofiles[i] != NULL) 1373 continue; 1374 if (devnull < 0) { 1375 if ((error = falloc(p, &fp, NULL)) != 0) 1376 break; 1377 1378 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, 1379 NLC_FOLLOW|NLC_LOCKVP); 1380 flags = FREAD | FWRITE; 1381 if (error == 0) 1382 error = vn_open(&nd, fp, flags, 0); 1383 if (error == 0) 1384 error = fsetfd(p, fp, &fd); 1385 fdrop(fp, td); 1386 nlookup_done(&nd); 1387 if (error) 1388 break; 1389 KKASSERT(i == fd); 1390 devnull = fd; 1391 } else { 1392 error = kern_dup(DUP_FIXED, devnull, i, &retval); 1393 if (error != 0) 1394 break; 1395 } 1396 } 1397 return (error); 1398 } 1399 1400 /* 1401 * Internal form of close. 1402 * Decrement reference count on file structure. 1403 * Note: td and/or p may be NULL when closing a file 1404 * that was being passed in a message. 1405 */ 1406 int 1407 closef(struct file *fp, struct thread *td) 1408 { 1409 struct vnode *vp; 1410 struct flock lf; 1411 struct filedesc_to_leader *fdtol; 1412 struct proc *p; 1413 1414 if (fp == NULL) 1415 return (0); 1416 if (td == NULL) { 1417 td = curthread; 1418 p = NULL; /* allow no proc association */ 1419 } else { 1420 p = td->td_proc; /* can also be NULL */ 1421 } 1422 /* 1423 * POSIX record locking dictates that any close releases ALL 1424 * locks owned by this process. This is handled by setting 1425 * a flag in the unlock to free ONLY locks obeying POSIX 1426 * semantics, and not to free BSD-style file locks. 1427 * If the descriptor was in a message, POSIX-style locks 1428 * aren't passed with the descriptor. 1429 */ 1430 if (p != NULL && 1431 fp->f_type == DTYPE_VNODE) { 1432 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) { 1433 lf.l_whence = SEEK_SET; 1434 lf.l_start = 0; 1435 lf.l_len = 0; 1436 lf.l_type = F_UNLCK; 1437 vp = (struct vnode *)fp->f_data; 1438 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 1439 &lf, F_POSIX); 1440 } 1441 fdtol = p->p_fdtol; 1442 if (fdtol != NULL) { 1443 /* 1444 * Handle special case where file descriptor table 1445 * is shared between multiple process leaders. 1446 */ 1447 for (fdtol = fdtol->fdl_next; 1448 fdtol != p->p_fdtol; 1449 fdtol = fdtol->fdl_next) { 1450 if ((fdtol->fdl_leader->p_flag & 1451 P_ADVLOCK) == 0) 1452 continue; 1453 fdtol->fdl_holdcount++; 1454 lf.l_whence = SEEK_SET; 1455 lf.l_start = 0; 1456 lf.l_len = 0; 1457 lf.l_type = F_UNLCK; 1458 vp = (struct vnode *)fp->f_data; 1459 (void) VOP_ADVLOCK(vp, 1460 (caddr_t)p->p_leader, 1461 F_UNLCK, &lf, F_POSIX); 1462 fdtol->fdl_holdcount--; 1463 if (fdtol->fdl_holdcount == 0 && 1464 fdtol->fdl_wakeup != 0) { 1465 fdtol->fdl_wakeup = 0; 1466 wakeup(fdtol); 1467 } 1468 } 1469 } 1470 } 1471 return (fdrop(fp, td)); 1472 } 1473 1474 int 1475 fdrop(struct file *fp, struct thread *td) 1476 { 1477 struct flock lf; 1478 struct vnode *vp; 1479 int error; 1480 1481 if (--fp->f_count > 0) 1482 return (0); 1483 if (fp->f_count < 0) 1484 panic("fdrop: count < 0"); 1485 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) { 1486 lf.l_whence = SEEK_SET; 1487 lf.l_start = 0; 1488 lf.l_len = 0; 1489 lf.l_type = F_UNLCK; 1490 vp = (struct vnode *)fp->f_data; 1491 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 1492 } 1493 if (fp->f_ops != &badfileops) 1494 error = fo_close(fp, td); 1495 else 1496 error = 0; 1497 ffree(fp); 1498 return (error); 1499 } 1500 1501 /* 1502 * Apply an advisory lock on a file descriptor. 1503 * 1504 * Just attempt to get a record lock of the requested type on 1505 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 1506 */ 1507 /* ARGSUSED */ 1508 int 1509 flock(struct flock_args *uap) 1510 { 1511 struct proc *p = curproc; 1512 struct filedesc *fdp = p->p_fd; 1513 struct file *fp; 1514 struct vnode *vp; 1515 struct flock lf; 1516 1517 if ((unsigned)uap->fd >= fdp->fd_nfiles || 1518 (fp = fdp->fd_ofiles[uap->fd]) == NULL) 1519 return (EBADF); 1520 if (fp->f_type != DTYPE_VNODE) 1521 return (EOPNOTSUPP); 1522 vp = (struct vnode *)fp->f_data; 1523 lf.l_whence = SEEK_SET; 1524 lf.l_start = 0; 1525 lf.l_len = 0; 1526 if (uap->how & LOCK_UN) { 1527 lf.l_type = F_UNLCK; 1528 fp->f_flag &= ~FHASLOCK; 1529 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK)); 1530 } 1531 if (uap->how & LOCK_EX) 1532 lf.l_type = F_WRLCK; 1533 else if (uap->how & LOCK_SH) 1534 lf.l_type = F_RDLCK; 1535 else 1536 return (EBADF); 1537 fp->f_flag |= FHASLOCK; 1538 if (uap->how & LOCK_NB) 1539 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK)); 1540 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT)); 1541 } 1542 1543 /* 1544 * File Descriptor pseudo-device driver (/dev/fd/). 1545 * 1546 * Opening minor device N dup()s the file (if any) connected to file 1547 * descriptor N belonging to the calling process. Note that this driver 1548 * consists of only the ``open()'' routine, because all subsequent 1549 * references to this file will be direct to the other driver. 1550 */ 1551 /* ARGSUSED */ 1552 static int 1553 fdopen(dev_t dev, int mode, int type, struct thread *td) 1554 { 1555 KKASSERT(td->td_proc != NULL); 1556 1557 /* 1558 * XXX Kludge: set curproc->p_dupfd to contain the value of the 1559 * the file descriptor being sought for duplication. The error 1560 * return ensures that the vnode for this device will be released 1561 * by vn_open. Open will detect this special error and take the 1562 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 1563 * will simply report the error. 1564 */ 1565 td->td_proc->p_dupfd = minor(dev); 1566 return (ENODEV); 1567 } 1568 1569 /* 1570 * Duplicate the specified descriptor to a free descriptor. 1571 */ 1572 int 1573 dupfdopen(struct filedesc *fdp, int indx, int dfd, int mode, int error) 1574 { 1575 struct file *wfp; 1576 struct file *fp; 1577 1578 /* 1579 * If the to-be-dup'd fd number is greater than the allowed number 1580 * of file descriptors, or the fd to be dup'd has already been 1581 * closed, then reject. 1582 */ 1583 if ((u_int)dfd >= fdp->fd_nfiles || 1584 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 1585 return (EBADF); 1586 } 1587 1588 /* 1589 * There are two cases of interest here. 1590 * 1591 * For ENODEV simply dup (dfd) to file descriptor 1592 * (indx) and return. 1593 * 1594 * For ENXIO steal away the file structure from (dfd) and 1595 * store it in (indx). (dfd) is effectively closed by 1596 * this operation. 1597 * 1598 * Any other error code is just returned. 1599 */ 1600 switch (error) { 1601 case ENODEV: 1602 /* 1603 * Check that the mode the file is being opened for is a 1604 * subset of the mode of the existing descriptor. 1605 */ 1606 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) 1607 return (EACCES); 1608 fp = fdp->fd_ofiles[indx]; 1609 #if 0 1610 if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED) 1611 (void) munmapfd(p, indx); 1612 #endif 1613 fdp->fd_ofiles[indx] = wfp; 1614 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 1615 fhold(wfp); 1616 if (indx > fdp->fd_lastfile) 1617 fdp->fd_lastfile = indx; 1618 /* 1619 * we now own the reference to fp that the ofiles[] array 1620 * used to own. Release it. 1621 */ 1622 if (fp) 1623 fdrop(fp, curthread); 1624 return (0); 1625 1626 case ENXIO: 1627 /* 1628 * Steal away the file pointer from dfd, and stuff it into indx. 1629 */ 1630 fp = fdp->fd_ofiles[indx]; 1631 #if 0 1632 if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED) 1633 (void) munmapfd(p, indx); 1634 #endif 1635 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 1636 fdp->fd_ofiles[dfd] = NULL; 1637 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 1638 fdp->fd_ofileflags[dfd] = 0; 1639 1640 /* 1641 * we now own the reference to fp that the ofiles[] array 1642 * used to own. Release it. 1643 */ 1644 if (fp) 1645 fdrop(fp, curthread); 1646 /* 1647 * Complete the clean up of the filedesc structure by 1648 * recomputing the various hints. 1649 */ 1650 if (indx > fdp->fd_lastfile) { 1651 fdp->fd_lastfile = indx; 1652 } else { 1653 while (fdp->fd_lastfile > 0 && 1654 fdp->fd_ofiles[fdp->fd_lastfile] == NULL) { 1655 fdp->fd_lastfile--; 1656 } 1657 if (dfd < fdp->fd_freefile) 1658 fdp->fd_freefile = dfd; 1659 } 1660 return (0); 1661 1662 default: 1663 return (error); 1664 } 1665 /* NOTREACHED */ 1666 } 1667 1668 1669 struct filedesc_to_leader * 1670 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 1671 struct proc *leader) 1672 { 1673 struct filedesc_to_leader *fdtol; 1674 1675 fdtol = malloc(sizeof(struct filedesc_to_leader), 1676 M_FILEDESC_TO_LEADER, M_WAITOK); 1677 fdtol->fdl_refcount = 1; 1678 fdtol->fdl_holdcount = 0; 1679 fdtol->fdl_wakeup = 0; 1680 fdtol->fdl_leader = leader; 1681 if (old != NULL) { 1682 fdtol->fdl_next = old->fdl_next; 1683 fdtol->fdl_prev = old; 1684 old->fdl_next = fdtol; 1685 fdtol->fdl_next->fdl_prev = fdtol; 1686 } else { 1687 fdtol->fdl_next = fdtol; 1688 fdtol->fdl_prev = fdtol; 1689 } 1690 return fdtol; 1691 } 1692 1693 /* 1694 * Get file structures. 1695 */ 1696 static int 1697 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 1698 { 1699 struct kinfo_file kf; 1700 struct filedesc *fdp; 1701 struct file *fp; 1702 struct proc *p; 1703 int error, n; 1704 1705 /* 1706 * Note: because the number of file descriptors is calculated 1707 * in different ways for sizing vs returning the data, 1708 * there is information leakage from the first loop. However, 1709 * it is of a similar order of magnitude to the leakage from 1710 * global system statistics such as kern.openfiles. 1711 */ 1712 if (req->oldptr == NULL) { 1713 n = 16; /* A slight overestimate. */ 1714 LIST_FOREACH(fp, &filehead, f_list) 1715 n += fp->f_count; 1716 return (SYSCTL_OUT(req, 0, n * sizeof(kf))); 1717 } 1718 error = 0; 1719 LIST_FOREACH(p, &allproc, p_list) { 1720 if (p->p_stat == SIDL) 1721 continue; 1722 if (!PRISON_CHECK(req->td->td_proc->p_ucred, p->p_ucred) != 0) { 1723 continue; 1724 } 1725 if ((fdp = p->p_fd) == NULL) { 1726 continue; 1727 } 1728 for (n = 0; n < fdp->fd_nfiles; ++n) { 1729 if ((fp = fdp->fd_ofiles[n]) == NULL) 1730 continue; 1731 kcore_make_file(&kf, fp, p->p_pid, 1732 p->p_ucred->cr_uid, n); 1733 error = SYSCTL_OUT(req, &kf, sizeof(kf)); 1734 if (error) 1735 break; 1736 } 1737 if (error) 1738 break; 1739 } 1740 return (error); 1741 } 1742 1743 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 1744 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 1745 1746 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 1747 &maxfilesperproc, 0, "Maximum files allowed open per process"); 1748 1749 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 1750 &maxfiles, 0, "Maximum number of files"); 1751 1752 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, 1753 &maxfilesrootres, 0, "Descriptors reserved for root use"); 1754 1755 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 1756 &nfiles, 0, "System-wide number of open files"); 1757 1758 static void 1759 fildesc_drvinit(void *unused) 1760 { 1761 int fd; 1762 1763 cdevsw_add(&fildesc_cdevsw, 0, 0); 1764 for (fd = 0; fd < NUMFDESC; fd++) { 1765 make_dev(&fildesc_cdevsw, fd, 1766 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 1767 } 1768 make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 1769 make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 1770 make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 1771 } 1772 1773 struct fileops badfileops = { 1774 NULL, /* port */ 1775 NULL, /* clone */ 1776 badfo_readwrite, 1777 badfo_readwrite, 1778 badfo_ioctl, 1779 badfo_poll, 1780 badfo_kqfilter, 1781 badfo_stat, 1782 badfo_close 1783 }; 1784 1785 static int 1786 badfo_readwrite( 1787 struct file *fp, 1788 struct uio *uio, 1789 struct ucred *cred, 1790 int flags, 1791 struct thread *td 1792 ) { 1793 return (EBADF); 1794 } 1795 1796 static int 1797 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td) 1798 { 1799 return (EBADF); 1800 } 1801 1802 static int 1803 badfo_poll(struct file *fp, int events, struct ucred *cred, struct thread *td) 1804 { 1805 return (0); 1806 } 1807 1808 static int 1809 badfo_kqfilter(struct file *fp, struct knote *kn) 1810 { 1811 return (0); 1812 } 1813 1814 static int 1815 badfo_stat(struct file *fp, struct stat *sb, struct thread *td) 1816 { 1817 return (EBADF); 1818 } 1819 1820 static int 1821 badfo_close(struct file *fp, struct thread *td) 1822 { 1823 return (EBADF); 1824 } 1825 1826 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, 1827 fildesc_drvinit,NULL) 1828