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 1.81.2.17 2003/06/06 20:21:32 tegge Exp $ 40 * $DragonFly: src/sys/kern/kern_descrip.c,v 1.19 2004/03/01 06:33:17 dillon 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/namei.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 64 #include <vm/vm.h> 65 #include <vm/vm_extern.h> 66 67 #include <sys/file2.h> 68 69 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 70 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 71 "file desc to leader structures"); 72 MALLOC_DEFINE(M_FILE, "file", "Open file structure"); 73 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 74 75 static d_open_t fdopen; 76 #define NUMFDESC 64 77 78 #define CDEV_MAJOR 22 79 static struct cdevsw fildesc_cdevsw = { 80 /* name */ "FD", 81 /* maj */ CDEV_MAJOR, 82 /* flags */ 0, 83 /* port */ NULL, 84 /* autoq */ 0, 85 86 /* open */ fdopen, 87 /* close */ noclose, 88 /* read */ noread, 89 /* write */ nowrite, 90 /* ioctl */ noioctl, 91 /* poll */ nopoll, 92 /* mmap */ nommap, 93 /* strategy */ nostrategy, 94 /* dump */ nodump, 95 /* psize */ nopsize 96 }; 97 98 static int badfo_readwrite (struct file *fp, struct uio *uio, 99 struct ucred *cred, int flags, struct thread *td); 100 static int badfo_ioctl (struct file *fp, u_long com, caddr_t data, 101 struct thread *td); 102 static int badfo_poll (struct file *fp, int events, 103 struct ucred *cred, struct thread *td); 104 static int badfo_kqfilter (struct file *fp, struct knote *kn); 105 static int badfo_stat (struct file *fp, struct stat *sb, struct thread *td); 106 static int badfo_close (struct file *fp, struct thread *td); 107 108 /* 109 * Descriptor management. 110 */ 111 struct filelist filehead; /* head of list of open files */ 112 int nfiles; /* actual number of open files */ 113 extern int cmask; 114 115 /* 116 * System calls on descriptors. 117 */ 118 /* ARGSUSED */ 119 int 120 getdtablesize(struct getdtablesize_args *uap) 121 { 122 struct proc *p = curproc; 123 124 uap->sysmsg_result = 125 min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 126 return (0); 127 } 128 129 /* 130 * Duplicate a file descriptor to a particular value. 131 * 132 * note: keep in mind that a potential race condition exists when closing 133 * descriptors from a shared descriptor table (via rfork). 134 */ 135 /* ARGSUSED */ 136 int 137 dup2(struct dup2_args *uap) 138 { 139 int error; 140 141 error = kern_dup(DUP_FIXED, uap->from, uap->to, uap->sysmsg_fds); 142 143 return (error); 144 } 145 146 /* 147 * Duplicate a file descriptor. 148 */ 149 /* ARGSUSED */ 150 int 151 dup(struct dup_args *uap) 152 { 153 int error; 154 155 error = kern_dup(DUP_VARIABLE, uap->fd, 0, uap->sysmsg_fds); 156 157 return (error); 158 } 159 160 int 161 kern_fcntl(int fd, int cmd, union fcntl_dat *dat) 162 { 163 struct thread *td = curthread; 164 struct proc *p = td->td_proc; 165 struct filedesc *fdp = p->p_fd; 166 struct file *fp; 167 char *pop; 168 struct vnode *vp; 169 u_int newmin; 170 int tmp, error, flg = F_POSIX; 171 172 KKASSERT(p); 173 174 if ((unsigned)fd >= fdp->fd_nfiles || 175 (fp = fdp->fd_ofiles[fd]) == NULL) 176 return (EBADF); 177 pop = &fdp->fd_ofileflags[fd]; 178 179 switch (cmd) { 180 case F_DUPFD: 181 newmin = dat->fc_fd; 182 if (newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 183 newmin > maxfilesperproc) 184 return (EINVAL); 185 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd); 186 return (error); 187 188 case F_GETFD: 189 dat->fc_cloexec = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 190 return (0); 191 192 case F_SETFD: 193 *pop = (*pop &~ UF_EXCLOSE) | 194 (dat->fc_cloexec & FD_CLOEXEC ? UF_EXCLOSE : 0); 195 return (0); 196 197 case F_GETFL: 198 dat->fc_flags = OFLAGS(fp->f_flag); 199 return (0); 200 201 case F_SETFL: 202 fhold(fp); 203 fp->f_flag &= ~FCNTLFLAGS; 204 fp->f_flag |= FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 205 tmp = fp->f_flag & FNONBLOCK; 206 error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td); 207 if (error) { 208 fdrop(fp, td); 209 return (error); 210 } 211 tmp = fp->f_flag & FASYNC; 212 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, td); 213 if (!error) { 214 fdrop(fp, td); 215 return (0); 216 } 217 fp->f_flag &= ~FNONBLOCK; 218 tmp = 0; 219 fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td); 220 fdrop(fp, td); 221 return (error); 222 223 case F_GETOWN: 224 fhold(fp); 225 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, td); 226 fdrop(fp, td); 227 return(error); 228 229 case F_SETOWN: 230 fhold(fp); 231 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, td); 232 fdrop(fp, td); 233 return(error); 234 235 case F_SETLKW: 236 flg |= F_WAIT; 237 /* Fall into F_SETLK */ 238 239 case F_SETLK: 240 if (fp->f_type != DTYPE_VNODE) 241 return (EBADF); 242 vp = (struct vnode *)fp->f_data; 243 244 /* 245 * copyin/lockop may block 246 */ 247 fhold(fp); 248 if (dat->fc_flock.l_whence == SEEK_CUR) 249 dat->fc_flock.l_start += fp->f_offset; 250 251 switch (dat->fc_flock.l_type) { 252 case F_RDLCK: 253 if ((fp->f_flag & FREAD) == 0) { 254 error = EBADF; 255 break; 256 } 257 p->p_leader->p_flag |= P_ADVLOCK; 258 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 259 &dat->fc_flock, flg); 260 break; 261 case F_WRLCK: 262 if ((fp->f_flag & FWRITE) == 0) { 263 error = EBADF; 264 break; 265 } 266 p->p_leader->p_flag |= P_ADVLOCK; 267 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 268 &dat->fc_flock, flg); 269 break; 270 case F_UNLCK: 271 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 272 &dat->fc_flock, F_POSIX); 273 break; 274 default: 275 error = EINVAL; 276 break; 277 } 278 /* Check for race with close */ 279 if ((unsigned) fd >= fdp->fd_nfiles || 280 fp != fdp->fd_ofiles[fd]) { 281 dat->fc_flock.l_whence = SEEK_SET; 282 dat->fc_flock.l_start = 0; 283 dat->fc_flock.l_len = 0; 284 dat->fc_flock.l_type = F_UNLCK; 285 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 286 F_UNLCK, &dat->fc_flock, F_POSIX); 287 } 288 fdrop(fp, td); 289 return(error); 290 291 case F_GETLK: 292 if (fp->f_type != DTYPE_VNODE) 293 return (EBADF); 294 vp = (struct vnode *)fp->f_data; 295 /* 296 * copyin/lockop may block 297 */ 298 fhold(fp); 299 if (dat->fc_flock.l_type != F_RDLCK && 300 dat->fc_flock.l_type != F_WRLCK && 301 dat->fc_flock.l_type != F_UNLCK) { 302 fdrop(fp, td); 303 return (EINVAL); 304 } 305 if (dat->fc_flock.l_whence == SEEK_CUR) 306 dat->fc_flock.l_start += fp->f_offset; 307 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 308 &dat->fc_flock, F_POSIX); 309 fdrop(fp, td); 310 return(error); 311 default: 312 return (EINVAL); 313 } 314 /* NOTREACHED */ 315 } 316 317 /* 318 * The file control system call. 319 */ 320 int 321 fcntl(struct fcntl_args *uap) 322 { 323 union fcntl_dat dat; 324 int error; 325 326 switch (uap->cmd) { 327 case F_DUPFD: 328 dat.fc_fd = uap->arg; 329 break; 330 case F_SETFD: 331 dat.fc_cloexec = uap->arg; 332 break; 333 case F_SETFL: 334 dat.fc_flags = uap->arg; 335 break; 336 case F_SETOWN: 337 dat.fc_owner = uap->arg; 338 break; 339 case F_SETLKW: 340 case F_SETLK: 341 case F_GETLK: 342 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 343 sizeof(struct flock)); 344 if (error) 345 return (error); 346 break; 347 } 348 349 error = kern_fcntl(uap->fd, uap->cmd, &dat); 350 351 if (error == 0) { 352 switch (uap->cmd) { 353 case F_DUPFD: 354 uap->sysmsg_result = dat.fc_fd; 355 break; 356 case F_GETFD: 357 uap->sysmsg_result = dat.fc_cloexec; 358 break; 359 case F_GETFL: 360 uap->sysmsg_result = dat.fc_flags; 361 break; 362 case F_GETOWN: 363 uap->sysmsg_result = dat.fc_owner; 364 case F_GETLK: 365 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 366 sizeof(struct flock)); 367 break; 368 } 369 } 370 371 return (error); 372 } 373 374 /* 375 * Common code for dup, dup2, and fcntl(F_DUPFD). 376 * 377 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells 378 * kern_dup() to destructively dup over an existing file descriptor if new 379 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest 380 * unused file descriptor that is greater than or equal to new. 381 */ 382 int 383 kern_dup(enum dup_type type, int old, int new, int *res) 384 { 385 struct thread *td = curthread; 386 struct proc *p = td->td_proc; 387 struct filedesc *fdp = p->p_fd; 388 struct file *fp; 389 struct file *delfp; 390 int holdleaders; 391 int error, newfd; 392 393 /* 394 * Verify that we have a valid descriptor to dup from and 395 * possibly to dup to. 396 */ 397 if (old < 0 || new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 398 new >= maxfilesperproc) 399 return (EBADF); 400 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) 401 return (EBADF); 402 if (type == DUP_FIXED && old == new) { 403 *res = new; 404 return (0); 405 } 406 fp = fdp->fd_ofiles[old]; 407 fhold(fp); 408 409 /* 410 * Expand the table for the new descriptor if needed. This may 411 * block and drop and reacquire the fidedesc lock. 412 */ 413 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) { 414 error = fdalloc(p, new, &newfd); 415 if (error) { 416 fdrop(fp, td); 417 return (error); 418 } 419 } 420 if (type == DUP_VARIABLE) 421 new = newfd; 422 423 /* 424 * If the old file changed out from under us then treat it as a 425 * bad file descriptor. Userland should do its own locking to 426 * avoid this case. 427 */ 428 if (fdp->fd_ofiles[old] != fp) { 429 if (fdp->fd_ofiles[new] == NULL) { 430 if (new < fdp->fd_freefile) 431 fdp->fd_freefile = new; 432 while (fdp->fd_lastfile > 0 && 433 fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 434 fdp->fd_lastfile--; 435 } 436 fdrop(fp, td); 437 return (EBADF); 438 } 439 KASSERT(old != new, ("new fd is same as old")); 440 441 /* 442 * Save info on the descriptor being overwritten. We have 443 * to do the unmap now, but we cannot close it without 444 * introducing an ownership race for the slot. 445 */ 446 delfp = fdp->fd_ofiles[new]; 447 if (delfp != NULL && p->p_fdtol != NULL) { 448 /* 449 * Ask fdfree() to sleep to ensure that all relevant 450 * process leaders can be traversed in closef(). 451 */ 452 fdp->fd_holdleaderscount++; 453 holdleaders = 1; 454 } else 455 holdleaders = 0; 456 KASSERT(delfp == NULL || type == DUP_FIXED, 457 ("dup() picked an open file")); 458 #if 0 459 if (delfp && (fdp->fd_ofileflags[new] & UF_MAPPED)) 460 (void) munmapfd(p, new); 461 #endif 462 463 /* 464 * Duplicate the source descriptor, update lastfile 465 */ 466 fdp->fd_ofiles[new] = fp; 467 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 468 if (new > fdp->fd_lastfile) 469 fdp->fd_lastfile = new; 470 *res = new; 471 472 /* 473 * If we dup'd over a valid file, we now own the reference to it 474 * and must dispose of it using closef() semantics (as if a 475 * close() were performed on it). 476 */ 477 if (delfp) { 478 (void) closef(delfp, td); 479 if (holdleaders) { 480 fdp->fd_holdleaderscount--; 481 if (fdp->fd_holdleaderscount == 0 && 482 fdp->fd_holdleaderswakeup != 0) { 483 fdp->fd_holdleaderswakeup = 0; 484 wakeup(&fdp->fd_holdleaderscount); 485 } 486 } 487 } 488 return (0); 489 } 490 491 /* 492 * If sigio is on the list associated with a process or process group, 493 * disable signalling from the device, remove sigio from the list and 494 * free sigio. 495 */ 496 void 497 funsetown(sigio) 498 struct sigio *sigio; 499 { 500 int s; 501 502 if (sigio == NULL) 503 return; 504 s = splhigh(); 505 *(sigio->sio_myref) = NULL; 506 splx(s); 507 if (sigio->sio_pgid < 0) { 508 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 509 sigio, sio_pgsigio); 510 } else /* if ((*sigiop)->sio_pgid > 0) */ { 511 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 512 sigio, sio_pgsigio); 513 } 514 crfree(sigio->sio_ucred); 515 FREE(sigio, M_SIGIO); 516 } 517 518 /* Free a list of sigio structures. */ 519 void 520 funsetownlst(sigiolst) 521 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(pgid, sigiop) 537 pid_t pgid; 538 struct sigio **sigiop; 539 { 540 struct proc *proc; 541 struct pgrp *pgrp; 542 struct sigio *sigio; 543 int s; 544 545 if (pgid == 0) { 546 funsetown(*sigiop); 547 return (0); 548 } 549 if (pgid > 0) { 550 proc = pfind(pgid); 551 if (proc == NULL) 552 return (ESRCH); 553 554 /* 555 * Policy - Don't allow a process to FSETOWN a process 556 * in another session. 557 * 558 * Remove this test to allow maximum flexibility or 559 * restrict FSETOWN to the current process or process 560 * group for maximum safety. 561 */ 562 if (proc->p_session != curproc->p_session) 563 return (EPERM); 564 565 pgrp = NULL; 566 } else /* if (pgid < 0) */ { 567 pgrp = pgfind(-pgid); 568 if (pgrp == NULL) 569 return (ESRCH); 570 571 /* 572 * Policy - Don't allow a process to FSETOWN a process 573 * in another session. 574 * 575 * Remove this test to allow maximum flexibility or 576 * restrict FSETOWN to the current process or process 577 * group for maximum safety. 578 */ 579 if (pgrp->pg_session != curproc->p_session) 580 return (EPERM); 581 582 proc = NULL; 583 } 584 funsetown(*sigiop); 585 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK); 586 if (pgid > 0) { 587 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 588 sigio->sio_proc = proc; 589 } else { 590 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 591 sigio->sio_pgrp = pgrp; 592 } 593 sigio->sio_pgid = pgid; 594 sigio->sio_ucred = crhold(curproc->p_ucred); 595 /* It would be convenient if p_ruid was in ucred. */ 596 sigio->sio_ruid = curproc->p_ucred->cr_ruid; 597 sigio->sio_myref = sigiop; 598 s = splhigh(); 599 *sigiop = sigio; 600 splx(s); 601 return (0); 602 } 603 604 /* 605 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 606 */ 607 pid_t 608 fgetown(sigio) 609 struct sigio *sigio; 610 { 611 return (sigio != NULL ? sigio->sio_pgid : 0); 612 } 613 614 /* 615 * Close a file descriptor. 616 */ 617 /* ARGSUSED */ 618 int 619 close(struct close_args *uap) 620 { 621 struct thread *td = curthread; 622 struct proc *p = td->td_proc; 623 struct filedesc *fdp; 624 struct file *fp; 625 int fd = uap->fd; 626 int error; 627 int holdleaders; 628 629 KKASSERT(p); 630 fdp = p->p_fd; 631 632 if ((unsigned)fd >= fdp->fd_nfiles || 633 (fp = fdp->fd_ofiles[fd]) == NULL) 634 return (EBADF); 635 #if 0 636 if (fdp->fd_ofileflags[fd] & UF_MAPPED) 637 (void) munmapfd(p, fd); 638 #endif 639 fdp->fd_ofiles[fd] = NULL; 640 fdp->fd_ofileflags[fd] = 0; 641 holdleaders = 0; 642 if (p->p_fdtol != NULL) { 643 /* 644 * Ask fdfree() to sleep to ensure that all relevant 645 * process leaders can be traversed in closef(). 646 */ 647 fdp->fd_holdleaderscount++; 648 holdleaders = 1; 649 } 650 651 /* 652 * we now hold the fp reference that used to be owned by the descriptor 653 * array. 654 */ 655 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 656 fdp->fd_lastfile--; 657 if (fd < fdp->fd_freefile) 658 fdp->fd_freefile = fd; 659 if (fd < fdp->fd_knlistsize) 660 knote_fdclose(p, fd); 661 error = closef(fp, td); 662 if (holdleaders) { 663 fdp->fd_holdleaderscount--; 664 if (fdp->fd_holdleaderscount == 0 && 665 fdp->fd_holdleaderswakeup != 0) { 666 fdp->fd_holdleaderswakeup = 0; 667 wakeup(&fdp->fd_holdleaderscount); 668 } 669 } 670 return (error); 671 } 672 673 int 674 kern_fstat(int fd, struct stat *ub) 675 { 676 struct thread *td = curthread; 677 struct proc *p = td->td_proc; 678 struct filedesc *fdp; 679 struct file *fp; 680 int error; 681 682 KKASSERT(p); 683 684 fdp = p->p_fd; 685 if ((unsigned)fd >= fdp->fd_nfiles || 686 (fp = fdp->fd_ofiles[fd]) == NULL) 687 return (EBADF); 688 fhold(fp); 689 error = fo_stat(fp, ub, td); 690 fdrop(fp, td); 691 692 return (error); 693 } 694 695 /* 696 * Return status information about a file descriptor. 697 */ 698 int 699 fstat(struct fstat_args *uap) 700 { 701 struct stat st; 702 int error; 703 704 error = kern_fstat(uap->fd, &st); 705 706 if (error == 0) 707 error = copyout(&st, uap->sb, sizeof(st)); 708 return (error); 709 } 710 711 /* 712 * XXX: This is for source compatibility with NetBSD. Probably doesn't 713 * belong here. 714 */ 715 int 716 nfstat(struct nfstat_args *uap) 717 { 718 struct stat st; 719 struct nstat nst; 720 int error; 721 722 error = kern_fstat(uap->fd, &st); 723 724 if (error == 0) { 725 cvtnstat(&st, &nst); 726 error = copyout(&nst, uap->sb, sizeof (nst)); 727 } 728 return (error); 729 } 730 731 /* 732 * Return pathconf information about a file descriptor. 733 */ 734 /* ARGSUSED */ 735 int 736 fpathconf(struct fpathconf_args *uap) 737 { 738 struct thread *td = curthread; 739 struct proc *p = td->td_proc; 740 struct filedesc *fdp; 741 struct file *fp; 742 struct vnode *vp; 743 int error = 0; 744 745 KKASSERT(p); 746 fdp = p->p_fd; 747 if ((unsigned)uap->fd >= fdp->fd_nfiles || 748 (fp = fdp->fd_ofiles[uap->fd]) == NULL) 749 return (EBADF); 750 751 fhold(fp); 752 753 switch (fp->f_type) { 754 case DTYPE_PIPE: 755 case DTYPE_SOCKET: 756 if (uap->name != _PC_PIPE_BUF) { 757 error = EINVAL; 758 } else { 759 uap->sysmsg_result = PIPE_BUF; 760 error = 0; 761 } 762 break; 763 case DTYPE_FIFO: 764 case DTYPE_VNODE: 765 vp = (struct vnode *)fp->f_data; 766 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds); 767 break; 768 default: 769 error = EOPNOTSUPP; 770 break; 771 } 772 fdrop(fp, td); 773 return(error); 774 } 775 776 /* 777 * Allocate a file descriptor for the process. 778 */ 779 static int fdexpand; 780 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, ""); 781 782 int 783 fdalloc(struct proc *p, int want, int *result) 784 { 785 struct filedesc *fdp = p->p_fd; 786 int i; 787 int lim, last, nfiles; 788 struct file **newofile; 789 char *newofileflags; 790 791 /* 792 * Search for a free descriptor starting at the higher 793 * of want or fd_freefile. If that fails, consider 794 * expanding the ofile array. 795 */ 796 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 797 for (;;) { 798 last = min(fdp->fd_nfiles, lim); 799 if ((i = want) < fdp->fd_freefile) 800 i = fdp->fd_freefile; 801 for (; i < last; i++) { 802 if (fdp->fd_ofiles[i] == NULL) { 803 fdp->fd_ofileflags[i] = 0; 804 if (i > fdp->fd_lastfile) 805 fdp->fd_lastfile = i; 806 if (want <= fdp->fd_freefile) 807 fdp->fd_freefile = i; 808 *result = i; 809 return (0); 810 } 811 } 812 813 /* 814 * No space in current array. Expand? 815 */ 816 if (fdp->fd_nfiles >= lim) 817 return (EMFILE); 818 if (fdp->fd_nfiles < NDEXTENT) 819 nfiles = NDEXTENT; 820 else 821 nfiles = 2 * fdp->fd_nfiles; 822 MALLOC(newofile, struct file **, nfiles * OFILESIZE, 823 M_FILEDESC, M_WAITOK); 824 825 /* 826 * deal with file-table extend race that might have occured 827 * when malloc was blocked. 828 */ 829 if (fdp->fd_nfiles >= nfiles) { 830 FREE(newofile, M_FILEDESC); 831 continue; 832 } 833 newofileflags = (char *) &newofile[nfiles]; 834 /* 835 * Copy the existing ofile and ofileflags arrays 836 * and zero the new portion of each array. 837 */ 838 bcopy(fdp->fd_ofiles, newofile, 839 (i = sizeof(struct file *) * fdp->fd_nfiles)); 840 bzero((char *)newofile + i, nfiles * sizeof(struct file *) - i); 841 bcopy(fdp->fd_ofileflags, newofileflags, 842 (i = sizeof(char) * fdp->fd_nfiles)); 843 bzero(newofileflags + i, nfiles * sizeof(char) - i); 844 if (fdp->fd_nfiles > NDFILE) 845 FREE(fdp->fd_ofiles, M_FILEDESC); 846 fdp->fd_ofiles = newofile; 847 fdp->fd_ofileflags = newofileflags; 848 fdp->fd_nfiles = nfiles; 849 fdexpand++; 850 } 851 return (0); 852 } 853 854 /* 855 * Check to see whether n user file descriptors 856 * are available to the process p. 857 */ 858 int 859 fdavail(p, n) 860 struct proc *p; 861 int n; 862 { 863 struct filedesc *fdp = p->p_fd; 864 struct file **fpp; 865 int i, lim, last; 866 867 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 868 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 869 return (1); 870 871 last = min(fdp->fd_nfiles, lim); 872 fpp = &fdp->fd_ofiles[fdp->fd_freefile]; 873 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) { 874 if (*fpp == NULL && --n <= 0) 875 return (1); 876 } 877 return (0); 878 } 879 880 /* 881 * falloc: 882 * Create a new open file structure and allocate a file decriptor 883 * for the process that refers to it. If p is NULL, no descriptor 884 * is allocated and the file pointer is returned unassociated with 885 * any process. resultfd is only used if p is not NULL and may 886 * separately be NULL indicating that you don't need the returned fd. 887 */ 888 int 889 falloc(struct proc *p, struct file **resultfp, int *resultfd) 890 { 891 struct file *fp, *fq; 892 int error, i; 893 894 if (nfiles >= maxfiles) { 895 tablefull("file"); 896 return (ENFILE); 897 } 898 /* 899 * Allocate a new file descriptor. 900 * If the process has file descriptor zero open, add to the list 901 * of open files at that point, otherwise put it at the front of 902 * the list of open files. 903 */ 904 nfiles++; 905 MALLOC(fp, struct file *, sizeof(struct file), M_FILE, M_WAITOK); 906 bzero(fp, sizeof(struct file)); 907 908 /* 909 * wait until after malloc (which may have blocked) returns before 910 * allocating the slot, else a race might have shrunk it if we had 911 * allocated it before the malloc. 912 */ 913 i = -1; 914 if (p && (error = fdalloc(p, 0, &i))) { 915 nfiles--; 916 FREE(fp, M_FILE); 917 return (error); 918 } 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 if ((fq = p->p_fd->fd_ofiles[0]) != NULL) { 925 LIST_INSERT_AFTER(fq, fp, f_list); 926 } else { 927 LIST_INSERT_HEAD(&filehead, fp, f_list); 928 } 929 p->p_fd->fd_ofiles[i] = fp; 930 } else { 931 fp->f_cred = crhold(proc0.p_ucred); 932 LIST_INSERT_HEAD(&filehead, fp, f_list); 933 } 934 if (resultfp) 935 *resultfp = fp; 936 if (resultfd) 937 *resultfd = i; 938 return (0); 939 } 940 941 void 942 fsetcred(struct file *fp, struct ucred *cr) 943 { 944 crhold(cr); 945 crfree(fp->f_cred); 946 fp->f_cred = cr; 947 } 948 949 /* 950 * Free a file descriptor. 951 */ 952 void 953 ffree(fp) 954 struct file *fp; 955 { 956 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 957 LIST_REMOVE(fp, f_list); 958 crfree(fp->f_cred); 959 nfiles--; 960 FREE(fp, M_FILE); 961 } 962 963 /* 964 * Build a new filedesc structure. 965 */ 966 struct filedesc * 967 fdinit(p) 968 struct proc *p; 969 { 970 struct filedesc0 *newfdp; 971 struct filedesc *fdp = p->p_fd; 972 973 MALLOC(newfdp, struct filedesc0 *, sizeof(struct filedesc0), 974 M_FILEDESC, M_WAITOK); 975 bzero(newfdp, sizeof(struct filedesc0)); 976 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 977 if (newfdp->fd_fd.fd_cdir) 978 VREF(newfdp->fd_fd.fd_cdir); 979 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 980 if (newfdp->fd_fd.fd_rdir) 981 VREF(newfdp->fd_fd.fd_rdir); 982 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 983 if (newfdp->fd_fd.fd_jdir) 984 VREF(newfdp->fd_fd.fd_jdir); 985 986 /* Create the file descriptor table. */ 987 newfdp->fd_fd.fd_refcnt = 1; 988 newfdp->fd_fd.fd_cmask = cmask; 989 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 990 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 991 newfdp->fd_fd.fd_nfiles = NDFILE; 992 newfdp->fd_fd.fd_knlistsize = -1; 993 994 return (&newfdp->fd_fd); 995 } 996 997 /* 998 * Share a filedesc structure. 999 */ 1000 struct filedesc * 1001 fdshare(p) 1002 struct proc *p; 1003 { 1004 p->p_fd->fd_refcnt++; 1005 return (p->p_fd); 1006 } 1007 1008 /* 1009 * Copy a filedesc structure. 1010 */ 1011 struct filedesc * 1012 fdcopy(p) 1013 struct proc *p; 1014 { 1015 struct filedesc *newfdp, *fdp = p->p_fd; 1016 struct file **fpp; 1017 int i; 1018 1019 /* Certain daemons might not have file descriptors. */ 1020 if (fdp == NULL) 1021 return (NULL); 1022 1023 MALLOC(newfdp, struct filedesc *, sizeof(struct filedesc0), 1024 M_FILEDESC, M_WAITOK); 1025 bcopy(fdp, newfdp, sizeof(struct filedesc)); 1026 if (newfdp->fd_cdir) 1027 VREF(newfdp->fd_cdir); 1028 if (newfdp->fd_rdir) 1029 VREF(newfdp->fd_rdir); 1030 if (newfdp->fd_jdir) 1031 VREF(newfdp->fd_jdir); 1032 newfdp->fd_refcnt = 1; 1033 1034 /* 1035 * If the number of open files fits in the internal arrays 1036 * of the open file structure, use them, otherwise allocate 1037 * additional memory for the number of descriptors currently 1038 * in use. 1039 */ 1040 if (newfdp->fd_lastfile < NDFILE) { 1041 newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles; 1042 newfdp->fd_ofileflags = 1043 ((struct filedesc0 *) newfdp)->fd_dfileflags; 1044 i = NDFILE; 1045 } else { 1046 /* 1047 * Compute the smallest multiple of NDEXTENT needed 1048 * for the file descriptors currently in use, 1049 * allowing the table to shrink. 1050 */ 1051 i = newfdp->fd_nfiles; 1052 while (i > 2 * NDEXTENT && i > newfdp->fd_lastfile * 2) 1053 i /= 2; 1054 MALLOC(newfdp->fd_ofiles, struct file **, i * OFILESIZE, 1055 M_FILEDESC, M_WAITOK); 1056 newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i]; 1057 } 1058 newfdp->fd_nfiles = i; 1059 bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **)); 1060 bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char)); 1061 1062 /* 1063 * kq descriptors cannot be copied. 1064 */ 1065 if (newfdp->fd_knlistsize != -1) { 1066 fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile]; 1067 for (i = newfdp->fd_lastfile; i >= 0; i--, fpp--) { 1068 if (*fpp != NULL && (*fpp)->f_type == DTYPE_KQUEUE) { 1069 *fpp = NULL; 1070 if (i < newfdp->fd_freefile) 1071 newfdp->fd_freefile = i; 1072 } 1073 if (*fpp == NULL && i == newfdp->fd_lastfile && i > 0) 1074 newfdp->fd_lastfile--; 1075 } 1076 newfdp->fd_knlist = NULL; 1077 newfdp->fd_knlistsize = -1; 1078 newfdp->fd_knhash = NULL; 1079 newfdp->fd_knhashmask = 0; 1080 } 1081 1082 fpp = newfdp->fd_ofiles; 1083 for (i = newfdp->fd_lastfile; i-- >= 0; fpp++) { 1084 if (*fpp != NULL) 1085 fhold(*fpp); 1086 } 1087 return (newfdp); 1088 } 1089 1090 /* 1091 * Release a filedesc structure. 1092 */ 1093 void 1094 fdfree(struct proc *p) 1095 { 1096 struct thread *td = p->p_thread; 1097 struct filedesc *fdp = p->p_fd; 1098 struct file **fpp; 1099 int i; 1100 struct filedesc_to_leader *fdtol; 1101 struct file *fp; 1102 struct vnode *vp; 1103 struct flock lf; 1104 1105 /* Certain daemons might not have file descriptors. */ 1106 if (fdp == NULL) 1107 return; 1108 1109 /* Check for special need to clear POSIX style locks */ 1110 fdtol = p->p_fdtol; 1111 if (fdtol != NULL) { 1112 KASSERT(fdtol->fdl_refcount > 0, 1113 ("filedesc_to_refcount botch: fdl_refcount=%d", 1114 fdtol->fdl_refcount)); 1115 if (fdtol->fdl_refcount == 1 && 1116 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1117 i = 0; 1118 fpp = fdp->fd_ofiles; 1119 for (i = 0, fpp = fdp->fd_ofiles; 1120 i <= fdp->fd_lastfile; 1121 i++, fpp++) { 1122 if (*fpp == NULL || 1123 (*fpp)->f_type != DTYPE_VNODE) 1124 continue; 1125 fp = *fpp; 1126 fhold(fp); 1127 lf.l_whence = SEEK_SET; 1128 lf.l_start = 0; 1129 lf.l_len = 0; 1130 lf.l_type = F_UNLCK; 1131 vp = (struct vnode *)fp->f_data; 1132 (void) VOP_ADVLOCK(vp, 1133 (caddr_t)p->p_leader, 1134 F_UNLCK, 1135 &lf, 1136 F_POSIX); 1137 fdrop(fp, p->p_thread); 1138 fpp = fdp->fd_ofiles + i; 1139 } 1140 } 1141 retry: 1142 if (fdtol->fdl_refcount == 1) { 1143 if (fdp->fd_holdleaderscount > 0 && 1144 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1145 /* 1146 * close() or do_dup() has cleared a reference 1147 * in a shared file descriptor table. 1148 */ 1149 fdp->fd_holdleaderswakeup = 1; 1150 tsleep(&fdp->fd_holdleaderscount, 1151 0, "fdlhold", 0); 1152 goto retry; 1153 } 1154 if (fdtol->fdl_holdcount > 0) { 1155 /* 1156 * Ensure that fdtol->fdl_leader 1157 * remains valid in closef(). 1158 */ 1159 fdtol->fdl_wakeup = 1; 1160 tsleep(fdtol, 0, "fdlhold", 0); 1161 goto retry; 1162 } 1163 } 1164 fdtol->fdl_refcount--; 1165 if (fdtol->fdl_refcount == 0 && 1166 fdtol->fdl_holdcount == 0) { 1167 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1168 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1169 } else 1170 fdtol = NULL; 1171 p->p_fdtol = NULL; 1172 if (fdtol != NULL) 1173 FREE(fdtol, M_FILEDESC_TO_LEADER); 1174 } 1175 if (--fdp->fd_refcnt > 0) 1176 return; 1177 /* 1178 * we are the last reference to the structure, we can 1179 * safely assume it will not change out from under us. 1180 */ 1181 fpp = fdp->fd_ofiles; 1182 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) { 1183 if (*fpp) 1184 (void) closef(*fpp, td); 1185 } 1186 if (fdp->fd_nfiles > NDFILE) 1187 FREE(fdp->fd_ofiles, M_FILEDESC); 1188 if (fdp->fd_cdir) 1189 vrele(fdp->fd_cdir); 1190 if (fdp->fd_rdir) 1191 vrele(fdp->fd_rdir); 1192 if (fdp->fd_jdir) 1193 vrele(fdp->fd_jdir); 1194 if (fdp->fd_knlist) 1195 FREE(fdp->fd_knlist, M_KQUEUE); 1196 if (fdp->fd_knhash) 1197 FREE(fdp->fd_knhash, M_KQUEUE); 1198 FREE(fdp, M_FILEDESC); 1199 } 1200 1201 /* 1202 * For setugid programs, we don't want to people to use that setugidness 1203 * to generate error messages which write to a file which otherwise would 1204 * otherwise be off-limits to the process. 1205 * 1206 * This is a gross hack to plug the hole. A better solution would involve 1207 * a special vop or other form of generalized access control mechanism. We 1208 * go ahead and just reject all procfs file systems accesses as dangerous. 1209 * 1210 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1211 * sufficient. We also don't for check setugidness since we know we are. 1212 */ 1213 static int 1214 is_unsafe(struct file *fp) 1215 { 1216 if (fp->f_type == DTYPE_VNODE && 1217 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 1218 return (1); 1219 return (0); 1220 } 1221 1222 /* 1223 * Make this setguid thing safe, if at all possible. 1224 */ 1225 void 1226 setugidsafety(struct proc *p) 1227 { 1228 struct thread *td = p->p_thread; 1229 struct filedesc *fdp = p->p_fd; 1230 int i; 1231 1232 /* Certain daemons might not have file descriptors. */ 1233 if (fdp == NULL) 1234 return; 1235 1236 /* 1237 * note: fdp->fd_ofiles may be reallocated out from under us while 1238 * we are blocked in a close. Be careful! 1239 */ 1240 for (i = 0; i <= fdp->fd_lastfile; i++) { 1241 if (i > 2) 1242 break; 1243 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 1244 struct file *fp; 1245 1246 #if 0 1247 if ((fdp->fd_ofileflags[i] & UF_MAPPED) != 0) 1248 (void) munmapfd(p, i); 1249 #endif 1250 if (i < fdp->fd_knlistsize) 1251 knote_fdclose(p, i); 1252 /* 1253 * NULL-out descriptor prior to close to avoid 1254 * a race while close blocks. 1255 */ 1256 fp = fdp->fd_ofiles[i]; 1257 fdp->fd_ofiles[i] = NULL; 1258 fdp->fd_ofileflags[i] = 0; 1259 if (i < fdp->fd_freefile) 1260 fdp->fd_freefile = i; 1261 (void) closef(fp, td); 1262 } 1263 } 1264 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 1265 fdp->fd_lastfile--; 1266 } 1267 1268 /* 1269 * Close any files on exec? 1270 */ 1271 void 1272 fdcloseexec(struct proc *p) 1273 { 1274 struct thread *td = p->p_thread; 1275 struct filedesc *fdp = p->p_fd; 1276 int i; 1277 1278 /* Certain daemons might not have file descriptors. */ 1279 if (fdp == NULL) 1280 return; 1281 1282 /* 1283 * We cannot cache fd_ofiles or fd_ofileflags since operations 1284 * may block and rip them out from under us. 1285 */ 1286 for (i = 0; i <= fdp->fd_lastfile; i++) { 1287 if (fdp->fd_ofiles[i] != NULL && 1288 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) { 1289 struct file *fp; 1290 1291 #if 0 1292 if (fdp->fd_ofileflags[i] & UF_MAPPED) 1293 (void) munmapfd(p, i); 1294 #endif 1295 if (i < fdp->fd_knlistsize) 1296 knote_fdclose(p, i); 1297 /* 1298 * NULL-out descriptor prior to close to avoid 1299 * a race while close blocks. 1300 */ 1301 fp = fdp->fd_ofiles[i]; 1302 fdp->fd_ofiles[i] = NULL; 1303 fdp->fd_ofileflags[i] = 0; 1304 if (i < fdp->fd_freefile) 1305 fdp->fd_freefile = i; 1306 (void) closef(fp, td); 1307 } 1308 } 1309 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 1310 fdp->fd_lastfile--; 1311 } 1312 1313 /* 1314 * It is unsafe for set[ug]id processes to be started with file 1315 * descriptors 0..2 closed, as these descriptors are given implicit 1316 * significance in the Standard C library. fdcheckstd() will create a 1317 * descriptor referencing /dev/null for each of stdin, stdout, and 1318 * stderr that is not already open. 1319 */ 1320 int 1321 fdcheckstd(struct proc *p) 1322 { 1323 struct thread *td = p->p_thread; 1324 struct nameidata nd; 1325 struct filedesc *fdp; 1326 struct file *fp; 1327 register_t retval; 1328 int fd, i, error, flags, devnull; 1329 1330 fdp = p->p_fd; 1331 if (fdp == NULL) 1332 return (0); 1333 devnull = -1; 1334 error = 0; 1335 for (i = 0; i < 3; i++) { 1336 if (fdp->fd_ofiles[i] != NULL) 1337 continue; 1338 if (devnull < 0) { 1339 error = falloc(p, &fp, &fd); 1340 if (error != 0) 1341 break; 1342 NDINIT(&nd, NAMEI_LOOKUP, CNP_FOLLOW, UIO_SYSSPACE, 1343 "/dev/null", td); 1344 flags = FREAD | FWRITE; 1345 error = vn_open(&nd, flags, 0); 1346 if (error != 0) { 1347 fdp->fd_ofiles[i] = NULL; 1348 fdrop(fp, td); 1349 break; 1350 } 1351 NDFREE(&nd, NDF_ONLY_PNBUF); 1352 fp->f_data = (caddr_t)nd.ni_vp; 1353 fp->f_flag = flags; 1354 fp->f_ops = &vnops; 1355 fp->f_type = DTYPE_VNODE; 1356 VOP_UNLOCK(nd.ni_vp, NULL, 0, td); 1357 devnull = fd; 1358 } else { 1359 error = kern_dup(DUP_FIXED, devnull, i, &retval); 1360 if (error != 0) 1361 break; 1362 } 1363 } 1364 return (error); 1365 } 1366 1367 /* 1368 * Internal form of close. 1369 * Decrement reference count on file structure. 1370 * Note: td and/or p may be NULL when closing a file 1371 * that was being passed in a message. 1372 */ 1373 int 1374 closef(struct file *fp, struct thread *td) 1375 { 1376 struct vnode *vp; 1377 struct flock lf; 1378 struct filedesc_to_leader *fdtol; 1379 struct proc *p; 1380 1381 if (fp == NULL) 1382 return (0); 1383 if (td == NULL) { 1384 td = curthread; 1385 p = NULL; /* allow no proc association */ 1386 } else { 1387 p = td->td_proc; /* can also be NULL */ 1388 } 1389 /* 1390 * POSIX record locking dictates that any close releases ALL 1391 * locks owned by this process. This is handled by setting 1392 * a flag in the unlock to free ONLY locks obeying POSIX 1393 * semantics, and not to free BSD-style file locks. 1394 * If the descriptor was in a message, POSIX-style locks 1395 * aren't passed with the descriptor. 1396 */ 1397 if (p != NULL && 1398 fp->f_type == DTYPE_VNODE) { 1399 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) { 1400 lf.l_whence = SEEK_SET; 1401 lf.l_start = 0; 1402 lf.l_len = 0; 1403 lf.l_type = F_UNLCK; 1404 vp = (struct vnode *)fp->f_data; 1405 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 1406 &lf, F_POSIX); 1407 } 1408 fdtol = p->p_fdtol; 1409 if (fdtol != NULL) { 1410 /* 1411 * Handle special case where file descriptor table 1412 * is shared between multiple process leaders. 1413 */ 1414 for (fdtol = fdtol->fdl_next; 1415 fdtol != p->p_fdtol; 1416 fdtol = fdtol->fdl_next) { 1417 if ((fdtol->fdl_leader->p_flag & 1418 P_ADVLOCK) == 0) 1419 continue; 1420 fdtol->fdl_holdcount++; 1421 lf.l_whence = SEEK_SET; 1422 lf.l_start = 0; 1423 lf.l_len = 0; 1424 lf.l_type = F_UNLCK; 1425 vp = (struct vnode *)fp->f_data; 1426 (void) VOP_ADVLOCK(vp, 1427 (caddr_t)p->p_leader, 1428 F_UNLCK, &lf, F_POSIX); 1429 fdtol->fdl_holdcount--; 1430 if (fdtol->fdl_holdcount == 0 && 1431 fdtol->fdl_wakeup != 0) { 1432 fdtol->fdl_wakeup = 0; 1433 wakeup(fdtol); 1434 } 1435 } 1436 } 1437 } 1438 return (fdrop(fp, td)); 1439 } 1440 1441 int 1442 fdrop(struct file *fp, struct thread *td) 1443 { 1444 struct flock lf; 1445 struct vnode *vp; 1446 int error; 1447 1448 if (--fp->f_count > 0) 1449 return (0); 1450 if (fp->f_count < 0) 1451 panic("fdrop: count < 0"); 1452 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) { 1453 lf.l_whence = SEEK_SET; 1454 lf.l_start = 0; 1455 lf.l_len = 0; 1456 lf.l_type = F_UNLCK; 1457 vp = (struct vnode *)fp->f_data; 1458 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 1459 } 1460 if (fp->f_ops != &badfileops) 1461 error = fo_close(fp, td); 1462 else 1463 error = 0; 1464 ffree(fp); 1465 return (error); 1466 } 1467 1468 /* 1469 * Apply an advisory lock on a file descriptor. 1470 * 1471 * Just attempt to get a record lock of the requested type on 1472 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 1473 */ 1474 /* ARGSUSED */ 1475 int 1476 flock(struct flock_args *uap) 1477 { 1478 struct proc *p = curproc; 1479 struct filedesc *fdp = p->p_fd; 1480 struct file *fp; 1481 struct vnode *vp; 1482 struct flock lf; 1483 1484 if ((unsigned)uap->fd >= fdp->fd_nfiles || 1485 (fp = fdp->fd_ofiles[uap->fd]) == NULL) 1486 return (EBADF); 1487 if (fp->f_type != DTYPE_VNODE) 1488 return (EOPNOTSUPP); 1489 vp = (struct vnode *)fp->f_data; 1490 lf.l_whence = SEEK_SET; 1491 lf.l_start = 0; 1492 lf.l_len = 0; 1493 if (uap->how & LOCK_UN) { 1494 lf.l_type = F_UNLCK; 1495 fp->f_flag &= ~FHASLOCK; 1496 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK)); 1497 } 1498 if (uap->how & LOCK_EX) 1499 lf.l_type = F_WRLCK; 1500 else if (uap->how & LOCK_SH) 1501 lf.l_type = F_RDLCK; 1502 else 1503 return (EBADF); 1504 fp->f_flag |= FHASLOCK; 1505 if (uap->how & LOCK_NB) 1506 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK)); 1507 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT)); 1508 } 1509 1510 /* 1511 * File Descriptor pseudo-device driver (/dev/fd/). 1512 * 1513 * Opening minor device N dup()s the file (if any) connected to file 1514 * descriptor N belonging to the calling process. Note that this driver 1515 * consists of only the ``open()'' routine, because all subsequent 1516 * references to this file will be direct to the other driver. 1517 */ 1518 /* ARGSUSED */ 1519 static int 1520 fdopen(dev_t dev, int mode, int type, struct thread *td) 1521 { 1522 KKASSERT(td->td_proc != NULL); 1523 1524 /* 1525 * XXX Kludge: set curproc->p_dupfd to contain the value of the 1526 * the file descriptor being sought for duplication. The error 1527 * return ensures that the vnode for this device will be released 1528 * by vn_open. Open will detect this special error and take the 1529 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 1530 * will simply report the error. 1531 */ 1532 td->td_proc->p_dupfd = minor(dev); 1533 return (ENODEV); 1534 } 1535 1536 /* 1537 * Duplicate the specified descriptor to a free descriptor. 1538 */ 1539 int 1540 dupfdopen(struct filedesc *fdp, int indx, int dfd, int mode, int error) 1541 { 1542 struct file *wfp; 1543 struct file *fp; 1544 1545 /* 1546 * If the to-be-dup'd fd number is greater than the allowed number 1547 * of file descriptors, or the fd to be dup'd has already been 1548 * closed, then reject. 1549 */ 1550 if ((u_int)dfd >= fdp->fd_nfiles || 1551 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 1552 return (EBADF); 1553 } 1554 1555 /* 1556 * There are two cases of interest here. 1557 * 1558 * For ENODEV simply dup (dfd) to file descriptor 1559 * (indx) and return. 1560 * 1561 * For ENXIO steal away the file structure from (dfd) and 1562 * store it in (indx). (dfd) is effectively closed by 1563 * this operation. 1564 * 1565 * Any other error code is just returned. 1566 */ 1567 switch (error) { 1568 case ENODEV: 1569 /* 1570 * Check that the mode the file is being opened for is a 1571 * subset of the mode of the existing descriptor. 1572 */ 1573 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) 1574 return (EACCES); 1575 fp = fdp->fd_ofiles[indx]; 1576 #if 0 1577 if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED) 1578 (void) munmapfd(p, indx); 1579 #endif 1580 fdp->fd_ofiles[indx] = wfp; 1581 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 1582 fhold(wfp); 1583 if (indx > fdp->fd_lastfile) 1584 fdp->fd_lastfile = indx; 1585 /* 1586 * we now own the reference to fp that the ofiles[] array 1587 * used to own. Release it. 1588 */ 1589 if (fp) 1590 fdrop(fp, curthread); 1591 return (0); 1592 1593 case ENXIO: 1594 /* 1595 * Steal away the file pointer from dfd, and stuff it into indx. 1596 */ 1597 fp = fdp->fd_ofiles[indx]; 1598 #if 0 1599 if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED) 1600 (void) munmapfd(p, indx); 1601 #endif 1602 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 1603 fdp->fd_ofiles[dfd] = NULL; 1604 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 1605 fdp->fd_ofileflags[dfd] = 0; 1606 1607 /* 1608 * we now own the reference to fp that the ofiles[] array 1609 * used to own. Release it. 1610 */ 1611 if (fp) 1612 fdrop(fp, curthread); 1613 /* 1614 * Complete the clean up of the filedesc structure by 1615 * recomputing the various hints. 1616 */ 1617 if (indx > fdp->fd_lastfile) { 1618 fdp->fd_lastfile = indx; 1619 } else { 1620 while (fdp->fd_lastfile > 0 && 1621 fdp->fd_ofiles[fdp->fd_lastfile] == NULL) { 1622 fdp->fd_lastfile--; 1623 } 1624 if (dfd < fdp->fd_freefile) 1625 fdp->fd_freefile = dfd; 1626 } 1627 return (0); 1628 1629 default: 1630 return (error); 1631 } 1632 /* NOTREACHED */ 1633 } 1634 1635 1636 struct filedesc_to_leader * 1637 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 1638 struct proc *leader) 1639 { 1640 struct filedesc_to_leader *fdtol; 1641 1642 MALLOC(fdtol, struct filedesc_to_leader *, 1643 sizeof(struct filedesc_to_leader), 1644 M_FILEDESC_TO_LEADER, 1645 M_WAITOK); 1646 fdtol->fdl_refcount = 1; 1647 fdtol->fdl_holdcount = 0; 1648 fdtol->fdl_wakeup = 0; 1649 fdtol->fdl_leader = leader; 1650 if (old != NULL) { 1651 fdtol->fdl_next = old->fdl_next; 1652 fdtol->fdl_prev = old; 1653 old->fdl_next = fdtol; 1654 fdtol->fdl_next->fdl_prev = fdtol; 1655 } else { 1656 fdtol->fdl_next = fdtol; 1657 fdtol->fdl_prev = fdtol; 1658 } 1659 return fdtol; 1660 } 1661 1662 /* 1663 * Get file structures. 1664 */ 1665 static int 1666 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 1667 { 1668 int error; 1669 struct file *fp; 1670 1671 if (!req->oldptr) { 1672 /* 1673 * overestimate by 10 files 1674 */ 1675 return (SYSCTL_OUT(req, 0, sizeof(filehead) + 1676 (nfiles + 10) * sizeof(struct file))); 1677 } 1678 1679 error = SYSCTL_OUT(req, (caddr_t)&filehead, sizeof(filehead)); 1680 if (error) 1681 return (error); 1682 1683 /* 1684 * followed by an array of file structures 1685 */ 1686 LIST_FOREACH(fp, &filehead, f_list) { 1687 error = SYSCTL_OUT(req, (caddr_t)fp, sizeof (struct file)); 1688 if (error) 1689 return (error); 1690 } 1691 return (0); 1692 } 1693 1694 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 1695 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 1696 1697 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 1698 &maxfilesperproc, 0, "Maximum files allowed open per process"); 1699 1700 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 1701 &maxfiles, 0, "Maximum number of files"); 1702 1703 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 1704 &nfiles, 0, "System-wide number of open files"); 1705 1706 static void 1707 fildesc_drvinit(void *unused) 1708 { 1709 int fd; 1710 1711 for (fd = 0; fd < NUMFDESC; fd++) 1712 make_dev(&fildesc_cdevsw, fd, 1713 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 1714 make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 1715 make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 1716 make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 1717 } 1718 1719 struct fileops badfileops = { 1720 NULL, /* port */ 1721 0, /* autoq */ 1722 badfo_readwrite, 1723 badfo_readwrite, 1724 badfo_ioctl, 1725 badfo_poll, 1726 badfo_kqfilter, 1727 badfo_stat, 1728 badfo_close 1729 }; 1730 1731 static int 1732 badfo_readwrite( 1733 struct file *fp, 1734 struct uio *uio, 1735 struct ucred *cred, 1736 int flags, 1737 struct thread *td 1738 ) { 1739 return (EBADF); 1740 } 1741 1742 static int 1743 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td) 1744 { 1745 return (EBADF); 1746 } 1747 1748 static int 1749 badfo_poll(struct file *fp, int events, struct ucred *cred, struct thread *td) 1750 { 1751 return (0); 1752 } 1753 1754 static int 1755 badfo_kqfilter(struct file *fp, struct knote *kn) 1756 { 1757 return (0); 1758 } 1759 1760 static int 1761 badfo_stat(struct file *fp, struct stat *sb, struct thread *td) 1762 { 1763 return (EBADF); 1764 } 1765 1766 static int 1767 badfo_close(struct file *fp, struct thread *td) 1768 { 1769 return (EBADF); 1770 } 1771 1772 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, 1773 fildesc_drvinit,NULL) 1774