1 /* 2 * Copyright (c) 2005 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Jeffrey Hsu. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * 35 * Copyright (c) 1982, 1986, 1989, 1991, 1993 36 * The Regents of the University of California. All rights reserved. 37 * (c) UNIX System Laboratories, Inc. 38 * All or some portions of this file are derived from material licensed 39 * to the University of California by American Telephone and Telegraph 40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 41 * the permission of UNIX System Laboratories, Inc. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed by the University of 54 * California, Berkeley and its contributors. 55 * 4. Neither the name of the University nor the names of its contributors 56 * may be used to endorse or promote products derived from this software 57 * without specific prior written permission. 58 * 59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 69 * SUCH DAMAGE. 70 * 71 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 72 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $ 73 * $DragonFly: src/sys/kern/kern_descrip.c,v 1.45 2005/06/22 19:58:44 dillon Exp $ 74 */ 75 76 #include "opt_compat.h" 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/malloc.h> 80 #include <sys/sysproto.h> 81 #include <sys/conf.h> 82 #include <sys/filedesc.h> 83 #include <sys/kernel.h> 84 #include <sys/sysctl.h> 85 #include <sys/vnode.h> 86 #include <sys/proc.h> 87 #include <sys/nlookup.h> 88 #include <sys/file.h> 89 #include <sys/stat.h> 90 #include <sys/filio.h> 91 #include <sys/fcntl.h> 92 #include <sys/unistd.h> 93 #include <sys/resourcevar.h> 94 #include <sys/event.h> 95 #include <sys/kern_syscall.h> 96 #include <sys/kcore.h> 97 #include <sys/kinfo.h> 98 99 #include <vm/vm.h> 100 #include <vm/vm_extern.h> 101 102 #include <sys/thread2.h> 103 #include <sys/file2.h> 104 105 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 106 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 107 "file desc to leader structures"); 108 MALLOC_DEFINE(M_FILE, "file", "Open file structure"); 109 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 110 111 static d_open_t fdopen; 112 #define NUMFDESC 64 113 114 #define CDEV_MAJOR 22 115 static struct cdevsw fildesc_cdevsw = { 116 /* name */ "FD", 117 /* maj */ CDEV_MAJOR, 118 /* flags */ 0, 119 /* port */ NULL, 120 /* clone */ NULL, 121 122 /* open */ fdopen, 123 /* close */ noclose, 124 /* read */ noread, 125 /* write */ nowrite, 126 /* ioctl */ noioctl, 127 /* poll */ nopoll, 128 /* mmap */ nommap, 129 /* strategy */ nostrategy, 130 /* dump */ nodump, 131 /* psize */ nopsize 132 }; 133 134 static int badfo_readwrite (struct file *fp, struct uio *uio, 135 struct ucred *cred, int flags, struct thread *td); 136 static int badfo_ioctl (struct file *fp, u_long com, caddr_t data, 137 struct thread *td); 138 static int badfo_poll (struct file *fp, int events, 139 struct ucred *cred, struct thread *td); 140 static int badfo_kqfilter (struct file *fp, struct knote *kn); 141 static int badfo_stat (struct file *fp, struct stat *sb, struct thread *td); 142 static int badfo_close (struct file *fp, struct thread *td); 143 144 /* 145 * Descriptor management. 146 */ 147 struct filelist filehead; /* head of list of open files */ 148 int nfiles; /* actual number of open files */ 149 extern int cmask; 150 151 /* 152 * System calls on descriptors. 153 */ 154 /* ARGSUSED */ 155 int 156 getdtablesize(struct getdtablesize_args *uap) 157 { 158 struct proc *p = curproc; 159 160 uap->sysmsg_result = 161 min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 162 return (0); 163 } 164 165 /* 166 * Duplicate a file descriptor to a particular value. 167 * 168 * note: keep in mind that a potential race condition exists when closing 169 * descriptors from a shared descriptor table (via rfork). 170 */ 171 /* ARGSUSED */ 172 int 173 dup2(struct dup2_args *uap) 174 { 175 int error; 176 177 error = kern_dup(DUP_FIXED, uap->from, uap->to, uap->sysmsg_fds); 178 179 return (error); 180 } 181 182 /* 183 * Duplicate a file descriptor. 184 */ 185 /* ARGSUSED */ 186 int 187 dup(struct dup_args *uap) 188 { 189 int error; 190 191 error = kern_dup(DUP_VARIABLE, uap->fd, 0, uap->sysmsg_fds); 192 193 return (error); 194 } 195 196 int 197 kern_fcntl(int fd, int cmd, union fcntl_dat *dat) 198 { 199 struct thread *td = curthread; 200 struct proc *p = td->td_proc; 201 struct filedesc *fdp = p->p_fd; 202 struct file *fp; 203 char *pop; 204 struct vnode *vp; 205 u_int newmin; 206 int tmp, error, flg = F_POSIX; 207 208 KKASSERT(p); 209 210 if ((unsigned)fd >= fdp->fd_nfiles || 211 (fp = fdp->fd_files[fd].fp) == NULL) 212 return (EBADF); 213 pop = &fdp->fd_files[fd].fileflags; 214 215 switch (cmd) { 216 case F_DUPFD: 217 newmin = dat->fc_fd; 218 if (newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 219 newmin > maxfilesperproc) 220 return (EINVAL); 221 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd); 222 return (error); 223 224 case F_GETFD: 225 dat->fc_cloexec = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 226 return (0); 227 228 case F_SETFD: 229 *pop = (*pop &~ UF_EXCLOSE) | 230 (dat->fc_cloexec & FD_CLOEXEC ? UF_EXCLOSE : 0); 231 return (0); 232 233 case F_GETFL: 234 dat->fc_flags = OFLAGS(fp->f_flag); 235 return (0); 236 237 case F_SETFL: 238 fhold(fp); 239 fp->f_flag &= ~FCNTLFLAGS; 240 fp->f_flag |= FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 241 tmp = fp->f_flag & FNONBLOCK; 242 error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td); 243 if (error) { 244 fdrop(fp, td); 245 return (error); 246 } 247 tmp = fp->f_flag & FASYNC; 248 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, td); 249 if (!error) { 250 fdrop(fp, td); 251 return (0); 252 } 253 fp->f_flag &= ~FNONBLOCK; 254 tmp = 0; 255 fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td); 256 fdrop(fp, td); 257 return (error); 258 259 case F_GETOWN: 260 fhold(fp); 261 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, td); 262 fdrop(fp, td); 263 return(error); 264 265 case F_SETOWN: 266 fhold(fp); 267 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, td); 268 fdrop(fp, td); 269 return(error); 270 271 case F_SETLKW: 272 flg |= F_WAIT; 273 /* Fall into F_SETLK */ 274 275 case F_SETLK: 276 if (fp->f_type != DTYPE_VNODE) 277 return (EBADF); 278 vp = (struct vnode *)fp->f_data; 279 280 /* 281 * copyin/lockop may block 282 */ 283 fhold(fp); 284 if (dat->fc_flock.l_whence == SEEK_CUR) 285 dat->fc_flock.l_start += fp->f_offset; 286 287 switch (dat->fc_flock.l_type) { 288 case F_RDLCK: 289 if ((fp->f_flag & FREAD) == 0) { 290 error = EBADF; 291 break; 292 } 293 p->p_leader->p_flag |= P_ADVLOCK; 294 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 295 &dat->fc_flock, flg); 296 break; 297 case F_WRLCK: 298 if ((fp->f_flag & FWRITE) == 0) { 299 error = EBADF; 300 break; 301 } 302 p->p_leader->p_flag |= P_ADVLOCK; 303 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 304 &dat->fc_flock, flg); 305 break; 306 case F_UNLCK: 307 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 308 &dat->fc_flock, F_POSIX); 309 break; 310 default: 311 error = EINVAL; 312 break; 313 } 314 /* Check for race with close */ 315 if ((unsigned) fd >= fdp->fd_nfiles || 316 fp != fdp->fd_files[fd].fp) { 317 dat->fc_flock.l_whence = SEEK_SET; 318 dat->fc_flock.l_start = 0; 319 dat->fc_flock.l_len = 0; 320 dat->fc_flock.l_type = F_UNLCK; 321 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 322 F_UNLCK, &dat->fc_flock, F_POSIX); 323 } 324 fdrop(fp, td); 325 return(error); 326 327 case F_GETLK: 328 if (fp->f_type != DTYPE_VNODE) 329 return (EBADF); 330 vp = (struct vnode *)fp->f_data; 331 /* 332 * copyin/lockop may block 333 */ 334 fhold(fp); 335 if (dat->fc_flock.l_type != F_RDLCK && 336 dat->fc_flock.l_type != F_WRLCK && 337 dat->fc_flock.l_type != F_UNLCK) { 338 fdrop(fp, td); 339 return (EINVAL); 340 } 341 if (dat->fc_flock.l_whence == SEEK_CUR) 342 dat->fc_flock.l_start += fp->f_offset; 343 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 344 &dat->fc_flock, F_POSIX); 345 fdrop(fp, td); 346 return(error); 347 default: 348 return (EINVAL); 349 } 350 /* NOTREACHED */ 351 } 352 353 /* 354 * The file control system call. 355 */ 356 int 357 fcntl(struct fcntl_args *uap) 358 { 359 union fcntl_dat dat; 360 int error; 361 362 switch (uap->cmd) { 363 case F_DUPFD: 364 dat.fc_fd = uap->arg; 365 break; 366 case F_SETFD: 367 dat.fc_cloexec = uap->arg; 368 break; 369 case F_SETFL: 370 dat.fc_flags = uap->arg; 371 break; 372 case F_SETOWN: 373 dat.fc_owner = uap->arg; 374 break; 375 case F_SETLKW: 376 case F_SETLK: 377 case F_GETLK: 378 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 379 sizeof(struct flock)); 380 if (error) 381 return (error); 382 break; 383 } 384 385 error = kern_fcntl(uap->fd, uap->cmd, &dat); 386 387 if (error == 0) { 388 switch (uap->cmd) { 389 case F_DUPFD: 390 uap->sysmsg_result = dat.fc_fd; 391 break; 392 case F_GETFD: 393 uap->sysmsg_result = dat.fc_cloexec; 394 break; 395 case F_GETFL: 396 uap->sysmsg_result = dat.fc_flags; 397 break; 398 case F_GETOWN: 399 uap->sysmsg_result = dat.fc_owner; 400 case F_GETLK: 401 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 402 sizeof(struct flock)); 403 break; 404 } 405 } 406 407 return (error); 408 } 409 410 /* 411 * Common code for dup, dup2, and fcntl(F_DUPFD). 412 * 413 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells 414 * kern_dup() to destructively dup over an existing file descriptor if new 415 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest 416 * unused file descriptor that is greater than or equal to new. 417 */ 418 int 419 kern_dup(enum dup_type type, int old, int new, int *res) 420 { 421 struct thread *td = curthread; 422 struct proc *p = td->td_proc; 423 struct filedesc *fdp = p->p_fd; 424 struct file *fp; 425 struct file *delfp; 426 int holdleaders; 427 boolean_t fdalloced = FALSE; 428 int error, newfd; 429 430 /* 431 * Verify that we have a valid descriptor to dup from and 432 * possibly to dup to. 433 */ 434 if (old < 0 || new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 435 new >= maxfilesperproc) 436 return (EBADF); 437 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) 438 return (EBADF); 439 if (type == DUP_FIXED && old == new) { 440 *res = new; 441 return (0); 442 } 443 fp = fdp->fd_files[old].fp; 444 fhold(fp); 445 446 /* 447 * Expand the table for the new descriptor if needed. This may 448 * block and drop and reacquire the fidedesc lock. 449 */ 450 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) { 451 error = fdalloc(p, new, &newfd); 452 if (error) { 453 fdrop(fp, td); 454 return (error); 455 } 456 fdalloced = TRUE; 457 } 458 if (type == DUP_VARIABLE) 459 new = newfd; 460 461 /* 462 * If the old file changed out from under us then treat it as a 463 * bad file descriptor. Userland should do its own locking to 464 * avoid this case. 465 */ 466 if (fdp->fd_files[old].fp != fp) { 467 if (fdp->fd_files[new].fp == NULL) { 468 if (fdalloced) 469 fdreserve(fdp, newfd, -1); 470 if (new < fdp->fd_freefile) 471 fdp->fd_freefile = new; 472 while (fdp->fd_lastfile > 0 && 473 fdp->fd_files[fdp->fd_lastfile].fp == NULL) 474 fdp->fd_lastfile--; 475 } 476 fdrop(fp, td); 477 return (EBADF); 478 } 479 KASSERT(old != new, ("new fd is same as old")); 480 481 /* 482 * Save info on the descriptor being overwritten. We have 483 * to do the unmap now, but we cannot close it without 484 * introducing an ownership race for the slot. 485 */ 486 delfp = fdp->fd_files[new].fp; 487 if (delfp != NULL && p->p_fdtol != NULL) { 488 /* 489 * Ask fdfree() to sleep to ensure that all relevant 490 * process leaders can be traversed in closef(). 491 */ 492 fdp->fd_holdleaderscount++; 493 holdleaders = 1; 494 } else 495 holdleaders = 0; 496 KASSERT(delfp == NULL || type == DUP_FIXED, 497 ("dup() picked an open file")); 498 #if 0 499 if (delfp && (fdp->fd_files[new].fileflags & UF_MAPPED)) 500 (void) munmapfd(p, new); 501 #endif 502 503 /* 504 * Duplicate the source descriptor, update lastfile 505 */ 506 if (new > fdp->fd_lastfile) 507 fdp->fd_lastfile = new; 508 if (!fdalloced && fdp->fd_files[new].fp == NULL) 509 fdreserve(fdp, new, 1); 510 fdp->fd_files[new].fp = fp; 511 fdp->fd_files[new].fileflags = 512 fdp->fd_files[old].fileflags & ~UF_EXCLOSE; 513 *res = new; 514 515 /* 516 * If we dup'd over a valid file, we now own the reference to it 517 * and must dispose of it using closef() semantics (as if a 518 * close() were performed on it). 519 */ 520 if (delfp) { 521 (void) closef(delfp, td); 522 if (holdleaders) { 523 fdp->fd_holdleaderscount--; 524 if (fdp->fd_holdleaderscount == 0 && 525 fdp->fd_holdleaderswakeup != 0) { 526 fdp->fd_holdleaderswakeup = 0; 527 wakeup(&fdp->fd_holdleaderscount); 528 } 529 } 530 } 531 return (0); 532 } 533 534 /* 535 * If sigio is on the list associated with a process or process group, 536 * disable signalling from the device, remove sigio from the list and 537 * free sigio. 538 */ 539 void 540 funsetown(struct sigio *sigio) 541 { 542 if (sigio == NULL) 543 return; 544 crit_enter(); 545 *(sigio->sio_myref) = NULL; 546 crit_exit(); 547 if (sigio->sio_pgid < 0) { 548 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 549 sigio, sio_pgsigio); 550 } else /* if ((*sigiop)->sio_pgid > 0) */ { 551 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 552 sigio, sio_pgsigio); 553 } 554 crfree(sigio->sio_ucred); 555 free(sigio, M_SIGIO); 556 } 557 558 /* Free a list of sigio structures. */ 559 void 560 funsetownlst(struct sigiolst *sigiolst) 561 { 562 struct sigio *sigio; 563 564 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) 565 funsetown(sigio); 566 } 567 568 /* 569 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 570 * 571 * After permission checking, add a sigio structure to the sigio list for 572 * the process or process group. 573 */ 574 int 575 fsetown(pid_t pgid, struct sigio **sigiop) 576 { 577 struct proc *proc; 578 struct pgrp *pgrp; 579 struct sigio *sigio; 580 581 if (pgid == 0) { 582 funsetown(*sigiop); 583 return (0); 584 } 585 if (pgid > 0) { 586 proc = pfind(pgid); 587 if (proc == NULL) 588 return (ESRCH); 589 590 /* 591 * Policy - Don't allow a process to FSETOWN a process 592 * in another session. 593 * 594 * Remove this test to allow maximum flexibility or 595 * restrict FSETOWN to the current process or process 596 * group for maximum safety. 597 */ 598 if (proc->p_session != curproc->p_session) 599 return (EPERM); 600 601 pgrp = NULL; 602 } else /* if (pgid < 0) */ { 603 pgrp = pgfind(-pgid); 604 if (pgrp == NULL) 605 return (ESRCH); 606 607 /* 608 * Policy - Don't allow a process to FSETOWN a process 609 * in another session. 610 * 611 * Remove this test to allow maximum flexibility or 612 * restrict FSETOWN to the current process or process 613 * group for maximum safety. 614 */ 615 if (pgrp->pg_session != curproc->p_session) 616 return (EPERM); 617 618 proc = NULL; 619 } 620 funsetown(*sigiop); 621 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 622 if (pgid > 0) { 623 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 624 sigio->sio_proc = proc; 625 } else { 626 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 627 sigio->sio_pgrp = pgrp; 628 } 629 sigio->sio_pgid = pgid; 630 sigio->sio_ucred = crhold(curproc->p_ucred); 631 /* It would be convenient if p_ruid was in ucred. */ 632 sigio->sio_ruid = curproc->p_ucred->cr_ruid; 633 sigio->sio_myref = sigiop; 634 crit_enter(); 635 *sigiop = sigio; 636 crit_exit(); 637 return (0); 638 } 639 640 /* 641 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 642 */ 643 pid_t 644 fgetown(struct sigio *sigio) 645 { 646 return (sigio != NULL ? sigio->sio_pgid : 0); 647 } 648 649 /* 650 * Close many file descriptors. 651 */ 652 /* ARGSUSED */ 653 654 int 655 closefrom(struct closefrom_args *uap) 656 { 657 return(kern_closefrom(uap->fd)); 658 } 659 660 int 661 kern_closefrom(int fd) 662 { 663 struct thread *td = curthread; 664 struct proc *p = td->td_proc; 665 struct filedesc *fdp; 666 667 KKASSERT(p); 668 fdp = p->p_fd; 669 670 if (fd < 0 || fd > fdp->fd_lastfile) 671 return (0); 672 673 do { 674 if (kern_close(fdp->fd_lastfile) == EINTR) 675 return (EINTR); 676 } while (fdp->fd_lastfile > fd); 677 678 return (0); 679 } 680 681 /* 682 * Close a file descriptor. 683 */ 684 /* ARGSUSED */ 685 686 int 687 close(struct close_args *uap) 688 { 689 return(kern_close(uap->fd)); 690 } 691 692 int 693 kern_close(int fd) 694 { 695 struct thread *td = curthread; 696 struct proc *p = td->td_proc; 697 struct filedesc *fdp; 698 struct file *fp; 699 int error; 700 int holdleaders; 701 702 KKASSERT(p); 703 fdp = p->p_fd; 704 705 if ((unsigned)fd >= fdp->fd_nfiles || 706 (fp = fdp->fd_files[fd].fp) == NULL) 707 return (EBADF); 708 #if 0 709 if (fdp->fd_files[fd].fileflags & UF_MAPPED) 710 (void) munmapfd(p, fd); 711 #endif 712 funsetfd(fdp, fd); 713 holdleaders = 0; 714 if (p->p_fdtol != NULL) { 715 /* 716 * Ask fdfree() to sleep to ensure that all relevant 717 * process leaders can be traversed in closef(). 718 */ 719 fdp->fd_holdleaderscount++; 720 holdleaders = 1; 721 } 722 723 /* 724 * we now hold the fp reference that used to be owned by the descriptor 725 * array. 726 */ 727 while (fdp->fd_lastfile > 0 && fdp->fd_files[fdp->fd_lastfile].fp == NULL) 728 fdp->fd_lastfile--; 729 if (fd < fdp->fd_knlistsize) 730 knote_fdclose(p, fd); 731 error = closef(fp, td); 732 if (holdleaders) { 733 fdp->fd_holdleaderscount--; 734 if (fdp->fd_holdleaderscount == 0 && 735 fdp->fd_holdleaderswakeup != 0) { 736 fdp->fd_holdleaderswakeup = 0; 737 wakeup(&fdp->fd_holdleaderscount); 738 } 739 } 740 return (error); 741 } 742 743 int 744 kern_fstat(int fd, struct stat *ub) 745 { 746 struct thread *td = curthread; 747 struct proc *p = td->td_proc; 748 struct filedesc *fdp; 749 struct file *fp; 750 int error; 751 752 KKASSERT(p); 753 754 fdp = p->p_fd; 755 if ((unsigned)fd >= fdp->fd_nfiles || 756 (fp = fdp->fd_files[fd].fp) == NULL) 757 return (EBADF); 758 fhold(fp); 759 error = fo_stat(fp, ub, td); 760 fdrop(fp, td); 761 762 return (error); 763 } 764 765 /* 766 * Return status information about a file descriptor. 767 */ 768 int 769 fstat(struct fstat_args *uap) 770 { 771 struct stat st; 772 int error; 773 774 error = kern_fstat(uap->fd, &st); 775 776 if (error == 0) 777 error = copyout(&st, uap->sb, sizeof(st)); 778 return (error); 779 } 780 781 /* 782 * XXX: This is for source compatibility with NetBSD. Probably doesn't 783 * belong here. 784 */ 785 int 786 nfstat(struct nfstat_args *uap) 787 { 788 struct stat st; 789 struct nstat nst; 790 int error; 791 792 error = kern_fstat(uap->fd, &st); 793 794 if (error == 0) { 795 cvtnstat(&st, &nst); 796 error = copyout(&nst, uap->sb, sizeof (nst)); 797 } 798 return (error); 799 } 800 801 /* 802 * Return pathconf information about a file descriptor. 803 */ 804 /* ARGSUSED */ 805 int 806 fpathconf(struct fpathconf_args *uap) 807 { 808 struct thread *td = curthread; 809 struct proc *p = td->td_proc; 810 struct filedesc *fdp; 811 struct file *fp; 812 struct vnode *vp; 813 int error = 0; 814 815 KKASSERT(p); 816 fdp = p->p_fd; 817 if ((unsigned)uap->fd >= fdp->fd_nfiles || 818 (fp = fdp->fd_files[uap->fd].fp) == NULL) 819 return (EBADF); 820 821 fhold(fp); 822 823 switch (fp->f_type) { 824 case DTYPE_PIPE: 825 case DTYPE_SOCKET: 826 if (uap->name != _PC_PIPE_BUF) { 827 error = EINVAL; 828 } else { 829 uap->sysmsg_result = PIPE_BUF; 830 error = 0; 831 } 832 break; 833 case DTYPE_FIFO: 834 case DTYPE_VNODE: 835 vp = (struct vnode *)fp->f_data; 836 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds); 837 break; 838 default: 839 error = EOPNOTSUPP; 840 break; 841 } 842 fdrop(fp, td); 843 return(error); 844 } 845 846 static int fdexpand; 847 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, ""); 848 849 static void 850 fdgrow(struct filedesc *fdp, int want) 851 { 852 struct fdnode *newfiles; 853 struct fdnode *oldfiles; 854 int nf, extra; 855 856 nf = fdp->fd_nfiles; 857 do { 858 /* nf has to be of the form 2^n - 1 */ 859 nf = 2 * nf + 1; 860 } while (nf <= want); 861 862 newfiles = malloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK); 863 864 /* 865 * deal with file-table extend race that might have occured 866 * when malloc was blocked. 867 */ 868 if (fdp->fd_nfiles >= nf) { 869 free(newfiles, M_FILEDESC); 870 return; 871 } 872 /* 873 * Copy the existing ofile and ofileflags arrays 874 * and zero the new portion of each array. 875 */ 876 extra = nf - fdp->fd_nfiles; 877 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode)); 878 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode)); 879 880 oldfiles = fdp->fd_files; 881 fdp->fd_files = newfiles; 882 fdp->fd_nfiles = nf; 883 884 if (oldfiles != fdp->fd_builtin_files) 885 free(oldfiles, M_FILEDESC); 886 fdexpand++; 887 } 888 889 /* 890 * Number of nodes in right subtree, including the root. 891 */ 892 static __inline int 893 right_subtree_size(int n) 894 { 895 return (n ^ (n | (n + 1))); 896 } 897 898 /* 899 * Bigger ancestor. 900 */ 901 static __inline int 902 right_ancestor(int n) 903 { 904 return (n | (n + 1)); 905 } 906 907 /* 908 * Smaller ancestor. 909 */ 910 static __inline int 911 left_ancestor(int n) 912 { 913 return ((n & (n + 1)) - 1); 914 } 915 916 void 917 fdreserve(struct filedesc *fdp, int fd, int incr) 918 { 919 while (fd >= 0) { 920 fdp->fd_files[fd].allocated += incr; 921 KKASSERT(fdp->fd_files[fd].allocated >= 0); 922 fd = left_ancestor(fd); 923 } 924 } 925 926 /* 927 * Allocate a file descriptor for the process. 928 */ 929 int 930 fdalloc(struct proc *p, int want, int *result) 931 { 932 struct filedesc *fdp = p->p_fd; 933 int fd, rsize, rsum, node, lim; 934 935 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 936 if (want >= lim) 937 return (EMFILE); 938 if (want >= fdp->fd_nfiles) 939 fdgrow(fdp, want); 940 941 /* 942 * Search for a free descriptor starting at the higher 943 * of want or fd_freefile. If that fails, consider 944 * expanding the ofile array. 945 */ 946 retry: 947 /* move up the tree looking for a subtree with a free node */ 948 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim); 949 fd = right_ancestor(fd)) { 950 if (fdp->fd_files[fd].allocated == 0) 951 goto found; 952 953 rsize = right_subtree_size(fd); 954 if (fdp->fd_files[fd].allocated == rsize) 955 continue; /* right subtree full */ 956 957 /* 958 * Free fd is in the right subtree of the tree rooted at fd. 959 * Call that subtree R. Look for the smallest (leftmost) 960 * subtree of R with an unallocated fd: continue moving 961 * down the left branch until encountering a full left 962 * subtree, then move to the right. 963 */ 964 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) { 965 node = fd + rsize; 966 rsum += fdp->fd_files[node].allocated; 967 if (fdp->fd_files[fd].allocated == rsum + rsize) { 968 fd = node; /* move to the right */ 969 if (fdp->fd_files[node].allocated == 0) 970 goto found; 971 rsum = 0; 972 } 973 } 974 goto found; 975 } 976 977 /* 978 * No space in current array. Expand? 979 */ 980 if (fdp->fd_nfiles >= lim) 981 return (EMFILE); 982 fdgrow(fdp, want); 983 goto retry; 984 985 found: 986 KKASSERT(fd < fdp->fd_nfiles); 987 fdp->fd_files[fd].fileflags = 0; 988 if (fd > fdp->fd_lastfile) 989 fdp->fd_lastfile = fd; 990 if (want <= fdp->fd_freefile) 991 fdp->fd_freefile = fd; 992 *result = fd; 993 KKASSERT(fdp->fd_files[fd].fp == NULL); 994 fdreserve(fdp, fd, 1); 995 return (0); 996 } 997 998 /* 999 * Check to see whether n user file descriptors 1000 * are available to the process p. 1001 */ 1002 int 1003 fdavail(struct proc *p, int n) 1004 { 1005 struct filedesc *fdp = p->p_fd; 1006 struct fdnode *fdnode; 1007 int i, lim, last; 1008 1009 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 1010 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1011 return (1); 1012 1013 last = min(fdp->fd_nfiles, lim); 1014 fdnode = &fdp->fd_files[fdp->fd_freefile]; 1015 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) { 1016 if (fdnode->fp == NULL && --n <= 0) 1017 return (1); 1018 } 1019 return (0); 1020 } 1021 1022 /* 1023 * falloc: 1024 * Create a new open file structure and allocate a file decriptor 1025 * for the process that refers to it. If p is NULL, no descriptor 1026 * is allocated and the file pointer is returned unassociated with 1027 * any process. resultfd is only used if p is not NULL and may 1028 * separately be NULL indicating that you don't need the returned fd. 1029 * 1030 * A held file pointer is returned. If a descriptor has been allocated 1031 * an additional hold on the fp will be made due to the fd_files[] 1032 * reference. 1033 */ 1034 int 1035 falloc(struct proc *p, struct file **resultfp, int *resultfd) 1036 { 1037 static struct timeval lastfail; 1038 static int curfail; 1039 struct file *fp; 1040 int error; 1041 1042 fp = NULL; 1043 1044 /* 1045 * Handle filetable full issues and root overfill. 1046 */ 1047 if (nfiles >= maxfiles - maxfilesrootres && 1048 ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) { 1049 if (ppsratecheck(&lastfail, &curfail, 1)) { 1050 printf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n", 1051 (p ? p->p_ucred->cr_ruid : -1)); 1052 } 1053 error = ENFILE; 1054 goto done; 1055 } 1056 1057 /* 1058 * Allocate a new file descriptor. 1059 */ 1060 nfiles++; 1061 fp = malloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO); 1062 fp->f_count = 1; 1063 fp->f_ops = &badfileops; 1064 fp->f_seqcount = 1; 1065 if (p) 1066 fp->f_cred = crhold(p->p_ucred); 1067 else 1068 fp->f_cred = crhold(proc0.p_ucred); 1069 LIST_INSERT_HEAD(&filehead, fp, f_list); 1070 if (resultfd) { 1071 if ((error = fsetfd(p, fp, resultfd)) != 0) { 1072 fdrop(fp, p->p_thread); 1073 fp = NULL; 1074 } 1075 } else { 1076 error = 0; 1077 } 1078 done: 1079 *resultfp = fp; 1080 return (error); 1081 } 1082 1083 /* 1084 * Associate a file pointer with a file descriptor. On success the fp 1085 * will have an additional ref representing the fd_files[] association. 1086 */ 1087 int 1088 fsetfd(struct proc *p, struct file *fp, int *resultfd) 1089 { 1090 int fd, error; 1091 1092 fd = -1; 1093 if ((error = fdalloc(p, 0, &fd)) == 0) { 1094 fhold(fp); 1095 p->p_fd->fd_files[fd].fp = fp; 1096 } 1097 *resultfd = fd; 1098 return (0); 1099 } 1100 1101 void 1102 funsetfd(struct filedesc *fdp, int fd) 1103 { 1104 fdp->fd_files[fd].fp = NULL; 1105 fdp->fd_files[fd].fileflags = 0; 1106 fdreserve(fdp, fd, -1); 1107 if (fd < fdp->fd_freefile) 1108 fdp->fd_freefile = fd; 1109 } 1110 1111 void 1112 fsetcred(struct file *fp, struct ucred *cr) 1113 { 1114 crhold(cr); 1115 crfree(fp->f_cred); 1116 fp->f_cred = cr; 1117 } 1118 1119 /* 1120 * Free a file descriptor. 1121 */ 1122 void 1123 ffree(struct file *fp) 1124 { 1125 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 1126 LIST_REMOVE(fp, f_list); 1127 crfree(fp->f_cred); 1128 if (fp->f_ncp) { 1129 cache_drop(fp->f_ncp); 1130 fp->f_ncp = NULL; 1131 } 1132 nfiles--; 1133 free(fp, M_FILE); 1134 } 1135 1136 /* 1137 * Build a new filedesc structure. 1138 */ 1139 struct filedesc * 1140 fdinit(struct proc *p) 1141 { 1142 struct filedesc *newfdp; 1143 struct filedesc *fdp = p->p_fd; 1144 1145 newfdp = malloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO); 1146 if (fdp->fd_cdir) { 1147 newfdp->fd_cdir = fdp->fd_cdir; 1148 vref(newfdp->fd_cdir); 1149 newfdp->fd_ncdir = cache_hold(fdp->fd_ncdir); 1150 } 1151 1152 /* 1153 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of 1154 * proc0, but should unconditionally exist in other processes. 1155 */ 1156 if (fdp->fd_rdir) { 1157 newfdp->fd_rdir = fdp->fd_rdir; 1158 vref(newfdp->fd_rdir); 1159 newfdp->fd_nrdir = cache_hold(fdp->fd_nrdir); 1160 } 1161 if (fdp->fd_jdir) { 1162 newfdp->fd_jdir = fdp->fd_jdir; 1163 vref(newfdp->fd_jdir); 1164 newfdp->fd_njdir = cache_hold(fdp->fd_njdir); 1165 } 1166 1167 /* Create the file descriptor table. */ 1168 newfdp->fd_refcnt = 1; 1169 newfdp->fd_cmask = cmask; 1170 newfdp->fd_files = newfdp->fd_builtin_files; 1171 newfdp->fd_nfiles = NDFILE; 1172 newfdp->fd_knlistsize = -1; 1173 1174 return (newfdp); 1175 } 1176 1177 /* 1178 * Share a filedesc structure. 1179 */ 1180 struct filedesc * 1181 fdshare(struct proc *p) 1182 { 1183 p->p_fd->fd_refcnt++; 1184 return (p->p_fd); 1185 } 1186 1187 /* 1188 * Copy a filedesc structure. 1189 */ 1190 struct filedesc * 1191 fdcopy(struct proc *p) 1192 { 1193 struct filedesc *newfdp, *fdp = p->p_fd; 1194 struct fdnode *fdnode; 1195 int i; 1196 1197 /* Certain daemons might not have file descriptors. */ 1198 if (fdp == NULL) 1199 return (NULL); 1200 1201 newfdp = malloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK); 1202 *newfdp = *fdp; 1203 if (newfdp->fd_cdir) { 1204 vref(newfdp->fd_cdir); 1205 newfdp->fd_ncdir = cache_hold(newfdp->fd_ncdir); 1206 } 1207 /* 1208 * We must check for fd_rdir here, at least for now because 1209 * the init process is created before we have access to the 1210 * rootvode to take a reference to it. 1211 */ 1212 if (newfdp->fd_rdir) { 1213 vref(newfdp->fd_rdir); 1214 newfdp->fd_nrdir = cache_hold(newfdp->fd_nrdir); 1215 } 1216 if (newfdp->fd_jdir) { 1217 vref(newfdp->fd_jdir); 1218 newfdp->fd_njdir = cache_hold(newfdp->fd_njdir); 1219 } 1220 newfdp->fd_refcnt = 1; 1221 1222 /* 1223 * If the number of open files fits in the internal arrays 1224 * of the open file structure, use them, otherwise allocate 1225 * additional memory for the number of descriptors currently 1226 * in use. 1227 */ 1228 if (newfdp->fd_lastfile < NDFILE) { 1229 newfdp->fd_files = newfdp->fd_builtin_files; 1230 i = NDFILE; 1231 } else { 1232 /* 1233 * Compute the smallest file table size 1234 * for the file descriptors currently in use, 1235 * allowing the table to shrink. 1236 */ 1237 i = newfdp->fd_nfiles; 1238 while ((i-1)/2 > newfdp->fd_lastfile && (i-1)/2 > NDFILE) 1239 i = (i-1)/2; 1240 newfdp->fd_files = malloc(i * sizeof(struct fdnode), 1241 M_FILEDESC, M_WAITOK); 1242 } 1243 newfdp->fd_nfiles = i; 1244 1245 if (fdp->fd_files != fdp->fd_builtin_files || 1246 newfdp->fd_files != newfdp->fd_builtin_files 1247 ) { 1248 bcopy(fdp->fd_files, newfdp->fd_files, 1249 i * sizeof(struct fdnode)); 1250 } 1251 1252 /* 1253 * kq descriptors cannot be copied. 1254 */ 1255 if (newfdp->fd_knlistsize != -1) { 1256 fdnode = &newfdp->fd_files[newfdp->fd_lastfile]; 1257 for (i = newfdp->fd_lastfile; i >= 0; i--, fdnode--) { 1258 if (fdnode->fp != NULL && fdnode->fp->f_type == DTYPE_KQUEUE) 1259 funsetfd(newfdp, i); /* nulls out *fpp */ 1260 if (fdnode->fp == NULL && i == newfdp->fd_lastfile && i > 0) 1261 newfdp->fd_lastfile--; 1262 } 1263 newfdp->fd_knlist = NULL; 1264 newfdp->fd_knlistsize = -1; 1265 newfdp->fd_knhash = NULL; 1266 newfdp->fd_knhashmask = 0; 1267 } 1268 1269 fdnode = newfdp->fd_files; 1270 for (i = newfdp->fd_lastfile; i-- >= 0; fdnode++) { 1271 if (fdnode->fp != NULL) 1272 fhold(fdnode->fp); 1273 } 1274 return (newfdp); 1275 } 1276 1277 /* 1278 * Release a filedesc structure. 1279 */ 1280 void 1281 fdfree(struct proc *p) 1282 { 1283 struct thread *td = p->p_thread; 1284 struct filedesc *fdp = p->p_fd; 1285 struct fdnode *fdnode; 1286 int i; 1287 struct filedesc_to_leader *fdtol; 1288 struct file *fp; 1289 struct vnode *vp; 1290 struct flock lf; 1291 1292 /* Certain daemons might not have file descriptors. */ 1293 if (fdp == NULL) 1294 return; 1295 1296 /* Check for special need to clear POSIX style locks */ 1297 fdtol = p->p_fdtol; 1298 if (fdtol != NULL) { 1299 KASSERT(fdtol->fdl_refcount > 0, 1300 ("filedesc_to_refcount botch: fdl_refcount=%d", 1301 fdtol->fdl_refcount)); 1302 if (fdtol->fdl_refcount == 1 && 1303 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1304 i = 0; 1305 fdnode = fdp->fd_files; 1306 for (i = 0; i <= fdp->fd_lastfile; i++, fdnode++) { 1307 if (fdnode->fp == NULL || 1308 fdnode->fp->f_type != DTYPE_VNODE) 1309 continue; 1310 fp = fdnode->fp; 1311 fhold(fp); 1312 lf.l_whence = SEEK_SET; 1313 lf.l_start = 0; 1314 lf.l_len = 0; 1315 lf.l_type = F_UNLCK; 1316 vp = (struct vnode *)fp->f_data; 1317 (void) VOP_ADVLOCK(vp, 1318 (caddr_t)p->p_leader, 1319 F_UNLCK, 1320 &lf, 1321 F_POSIX); 1322 fdrop(fp, p->p_thread); 1323 /* reload due to possible reallocation */ 1324 fdnode = &fdp->fd_files[i]; 1325 } 1326 } 1327 retry: 1328 if (fdtol->fdl_refcount == 1) { 1329 if (fdp->fd_holdleaderscount > 0 && 1330 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1331 /* 1332 * close() or do_dup() has cleared a reference 1333 * in a shared file descriptor table. 1334 */ 1335 fdp->fd_holdleaderswakeup = 1; 1336 tsleep(&fdp->fd_holdleaderscount, 1337 0, "fdlhold", 0); 1338 goto retry; 1339 } 1340 if (fdtol->fdl_holdcount > 0) { 1341 /* 1342 * Ensure that fdtol->fdl_leader 1343 * remains valid in closef(). 1344 */ 1345 fdtol->fdl_wakeup = 1; 1346 tsleep(fdtol, 0, "fdlhold", 0); 1347 goto retry; 1348 } 1349 } 1350 fdtol->fdl_refcount--; 1351 if (fdtol->fdl_refcount == 0 && 1352 fdtol->fdl_holdcount == 0) { 1353 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1354 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1355 } else 1356 fdtol = NULL; 1357 p->p_fdtol = NULL; 1358 if (fdtol != NULL) 1359 free(fdtol, M_FILEDESC_TO_LEADER); 1360 } 1361 if (--fdp->fd_refcnt > 0) 1362 return; 1363 /* 1364 * we are the last reference to the structure, we can 1365 * safely assume it will not change out from under us. 1366 */ 1367 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1368 if (fdp->fd_files[i].fp) 1369 closef(fdp->fd_files[i].fp, td); 1370 } 1371 if (fdp->fd_files != fdp->fd_builtin_files) 1372 free(fdp->fd_files, M_FILEDESC); 1373 if (fdp->fd_cdir) { 1374 cache_drop(fdp->fd_ncdir); 1375 vrele(fdp->fd_cdir); 1376 } 1377 if (fdp->fd_rdir) { 1378 cache_drop(fdp->fd_nrdir); 1379 vrele(fdp->fd_rdir); 1380 } 1381 if (fdp->fd_jdir) { 1382 cache_drop(fdp->fd_njdir); 1383 vrele(fdp->fd_jdir); 1384 } 1385 if (fdp->fd_knlist) 1386 free(fdp->fd_knlist, M_KQUEUE); 1387 if (fdp->fd_knhash) 1388 free(fdp->fd_knhash, M_KQUEUE); 1389 free(fdp, M_FILEDESC); 1390 } 1391 1392 /* 1393 * For setugid programs, we don't want to people to use that setugidness 1394 * to generate error messages which write to a file which otherwise would 1395 * otherwise be off-limits to the process. 1396 * 1397 * This is a gross hack to plug the hole. A better solution would involve 1398 * a special vop or other form of generalized access control mechanism. We 1399 * go ahead and just reject all procfs file systems accesses as dangerous. 1400 * 1401 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1402 * sufficient. We also don't for check setugidness since we know we are. 1403 */ 1404 static int 1405 is_unsafe(struct file *fp) 1406 { 1407 if (fp->f_type == DTYPE_VNODE && 1408 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 1409 return (1); 1410 return (0); 1411 } 1412 1413 /* 1414 * Make this setguid thing safe, if at all possible. 1415 */ 1416 void 1417 setugidsafety(struct proc *p) 1418 { 1419 struct thread *td = p->p_thread; 1420 struct filedesc *fdp = p->p_fd; 1421 int i; 1422 1423 /* Certain daemons might not have file descriptors. */ 1424 if (fdp == NULL) 1425 return; 1426 1427 /* 1428 * note: fdp->fd_files may be reallocated out from under us while 1429 * we are blocked in a close. Be careful! 1430 */ 1431 for (i = 0; i <= fdp->fd_lastfile; i++) { 1432 if (i > 2) 1433 break; 1434 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) { 1435 struct file *fp; 1436 1437 #if 0 1438 if ((fdp->fd_files[i].fileflags & UF_MAPPED) != 0) 1439 (void) munmapfd(p, i); 1440 #endif 1441 if (i < fdp->fd_knlistsize) 1442 knote_fdclose(p, i); 1443 /* 1444 * NULL-out descriptor prior to close to avoid 1445 * a race while close blocks. 1446 */ 1447 fp = fdp->fd_files[i].fp; 1448 funsetfd(fdp, i); 1449 closef(fp, td); 1450 } 1451 } 1452 while (fdp->fd_lastfile > 0 && fdp->fd_files[fdp->fd_lastfile].fp == NULL) 1453 fdp->fd_lastfile--; 1454 } 1455 1456 /* 1457 * Close any files on exec? 1458 */ 1459 void 1460 fdcloseexec(struct proc *p) 1461 { 1462 struct thread *td = p->p_thread; 1463 struct filedesc *fdp = p->p_fd; 1464 int i; 1465 1466 /* Certain daemons might not have file descriptors. */ 1467 if (fdp == NULL) 1468 return; 1469 1470 /* 1471 * We cannot cache fd_files since operations may block and rip 1472 * them out from under us. 1473 */ 1474 for (i = 0; i <= fdp->fd_lastfile; i++) { 1475 if (fdp->fd_files[i].fp != NULL && 1476 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) { 1477 struct file *fp; 1478 1479 #if 0 1480 if (fdp->fd_files[i].fileflags & UF_MAPPED) 1481 (void) munmapfd(p, i); 1482 #endif 1483 if (i < fdp->fd_knlistsize) 1484 knote_fdclose(p, i); 1485 /* 1486 * NULL-out descriptor prior to close to avoid 1487 * a race while close blocks. 1488 */ 1489 fp = fdp->fd_files[i].fp; 1490 funsetfd(fdp, i); 1491 closef(fp, td); 1492 } 1493 } 1494 while (fdp->fd_lastfile > 0 && fdp->fd_files[fdp->fd_lastfile].fp == NULL) 1495 fdp->fd_lastfile--; 1496 } 1497 1498 /* 1499 * It is unsafe for set[ug]id processes to be started with file 1500 * descriptors 0..2 closed, as these descriptors are given implicit 1501 * significance in the Standard C library. fdcheckstd() will create a 1502 * descriptor referencing /dev/null for each of stdin, stdout, and 1503 * stderr that is not already open. 1504 */ 1505 int 1506 fdcheckstd(struct proc *p) 1507 { 1508 struct thread *td = p->p_thread; 1509 struct nlookupdata nd; 1510 struct filedesc *fdp; 1511 struct file *fp; 1512 register_t retval; 1513 int fd, i, error, flags, devnull; 1514 1515 fdp = p->p_fd; 1516 if (fdp == NULL) 1517 return (0); 1518 devnull = -1; 1519 error = 0; 1520 for (i = 0; i < 3; i++) { 1521 if (fdp->fd_files[i].fp != NULL) 1522 continue; 1523 if (devnull < 0) { 1524 if ((error = falloc(p, &fp, NULL)) != 0) 1525 break; 1526 1527 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, 1528 NLC_FOLLOW|NLC_LOCKVP); 1529 flags = FREAD | FWRITE; 1530 if (error == 0) 1531 error = vn_open(&nd, fp, flags, 0); 1532 if (error == 0) 1533 error = fsetfd(p, fp, &fd); 1534 fdrop(fp, td); 1535 nlookup_done(&nd); 1536 if (error) 1537 break; 1538 KKASSERT(i == fd); 1539 devnull = fd; 1540 } else { 1541 error = kern_dup(DUP_FIXED, devnull, i, &retval); 1542 if (error != 0) 1543 break; 1544 } 1545 } 1546 return (error); 1547 } 1548 1549 /* 1550 * Internal form of close. 1551 * Decrement reference count on file structure. 1552 * Note: td and/or p may be NULL when closing a file 1553 * that was being passed in a message. 1554 */ 1555 int 1556 closef(struct file *fp, struct thread *td) 1557 { 1558 struct vnode *vp; 1559 struct flock lf; 1560 struct filedesc_to_leader *fdtol; 1561 struct proc *p; 1562 1563 if (fp == NULL) 1564 return (0); 1565 if (td == NULL) { 1566 td = curthread; 1567 p = NULL; /* allow no proc association */ 1568 } else { 1569 p = td->td_proc; /* can also be NULL */ 1570 } 1571 /* 1572 * POSIX record locking dictates that any close releases ALL 1573 * locks owned by this process. This is handled by setting 1574 * a flag in the unlock to free ONLY locks obeying POSIX 1575 * semantics, and not to free BSD-style file locks. 1576 * If the descriptor was in a message, POSIX-style locks 1577 * aren't passed with the descriptor. 1578 */ 1579 if (p != NULL && 1580 fp->f_type == DTYPE_VNODE) { 1581 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) { 1582 lf.l_whence = SEEK_SET; 1583 lf.l_start = 0; 1584 lf.l_len = 0; 1585 lf.l_type = F_UNLCK; 1586 vp = (struct vnode *)fp->f_data; 1587 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 1588 &lf, F_POSIX); 1589 } 1590 fdtol = p->p_fdtol; 1591 if (fdtol != NULL) { 1592 /* 1593 * Handle special case where file descriptor table 1594 * is shared between multiple process leaders. 1595 */ 1596 for (fdtol = fdtol->fdl_next; 1597 fdtol != p->p_fdtol; 1598 fdtol = fdtol->fdl_next) { 1599 if ((fdtol->fdl_leader->p_flag & 1600 P_ADVLOCK) == 0) 1601 continue; 1602 fdtol->fdl_holdcount++; 1603 lf.l_whence = SEEK_SET; 1604 lf.l_start = 0; 1605 lf.l_len = 0; 1606 lf.l_type = F_UNLCK; 1607 vp = (struct vnode *)fp->f_data; 1608 (void) VOP_ADVLOCK(vp, 1609 (caddr_t)p->p_leader, 1610 F_UNLCK, &lf, F_POSIX); 1611 fdtol->fdl_holdcount--; 1612 if (fdtol->fdl_holdcount == 0 && 1613 fdtol->fdl_wakeup != 0) { 1614 fdtol->fdl_wakeup = 0; 1615 wakeup(fdtol); 1616 } 1617 } 1618 } 1619 } 1620 return (fdrop(fp, td)); 1621 } 1622 1623 int 1624 fdrop(struct file *fp, struct thread *td) 1625 { 1626 struct flock lf; 1627 struct vnode *vp; 1628 int error; 1629 1630 if (--fp->f_count > 0) 1631 return (0); 1632 if (fp->f_count < 0) 1633 panic("fdrop: count < 0"); 1634 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) { 1635 lf.l_whence = SEEK_SET; 1636 lf.l_start = 0; 1637 lf.l_len = 0; 1638 lf.l_type = F_UNLCK; 1639 vp = (struct vnode *)fp->f_data; 1640 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 1641 } 1642 if (fp->f_ops != &badfileops) 1643 error = fo_close(fp, td); 1644 else 1645 error = 0; 1646 ffree(fp); 1647 return (error); 1648 } 1649 1650 /* 1651 * Apply an advisory lock on a file descriptor. 1652 * 1653 * Just attempt to get a record lock of the requested type on 1654 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 1655 */ 1656 /* ARGSUSED */ 1657 int 1658 flock(struct flock_args *uap) 1659 { 1660 struct proc *p = curproc; 1661 struct filedesc *fdp = p->p_fd; 1662 struct file *fp; 1663 struct vnode *vp; 1664 struct flock lf; 1665 1666 if ((unsigned)uap->fd >= fdp->fd_nfiles || 1667 (fp = fdp->fd_files[uap->fd].fp) == NULL) 1668 return (EBADF); 1669 if (fp->f_type != DTYPE_VNODE) 1670 return (EOPNOTSUPP); 1671 vp = (struct vnode *)fp->f_data; 1672 lf.l_whence = SEEK_SET; 1673 lf.l_start = 0; 1674 lf.l_len = 0; 1675 if (uap->how & LOCK_UN) { 1676 lf.l_type = F_UNLCK; 1677 fp->f_flag &= ~FHASLOCK; 1678 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK)); 1679 } 1680 if (uap->how & LOCK_EX) 1681 lf.l_type = F_WRLCK; 1682 else if (uap->how & LOCK_SH) 1683 lf.l_type = F_RDLCK; 1684 else 1685 return (EBADF); 1686 fp->f_flag |= FHASLOCK; 1687 if (uap->how & LOCK_NB) 1688 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK)); 1689 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT)); 1690 } 1691 1692 /* 1693 * File Descriptor pseudo-device driver (/dev/fd/). 1694 * 1695 * Opening minor device N dup()s the file (if any) connected to file 1696 * descriptor N belonging to the calling process. Note that this driver 1697 * consists of only the ``open()'' routine, because all subsequent 1698 * references to this file will be direct to the other driver. 1699 */ 1700 /* ARGSUSED */ 1701 static int 1702 fdopen(dev_t dev, int mode, int type, struct thread *td) 1703 { 1704 KKASSERT(td->td_proc != NULL); 1705 1706 /* 1707 * XXX Kludge: set curproc->p_dupfd to contain the value of the 1708 * the file descriptor being sought for duplication. The error 1709 * return ensures that the vnode for this device will be released 1710 * by vn_open. Open will detect this special error and take the 1711 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 1712 * will simply report the error. 1713 */ 1714 td->td_proc->p_dupfd = minor(dev); 1715 return (ENODEV); 1716 } 1717 1718 /* 1719 * Duplicate the specified descriptor to a free descriptor. 1720 */ 1721 int 1722 dupfdopen(struct filedesc *fdp, int indx, int dfd, int mode, int error) 1723 { 1724 struct file *wfp; 1725 struct file *fp; 1726 1727 /* 1728 * If the to-be-dup'd fd number is greater than the allowed number 1729 * of file descriptors, or the fd to be dup'd has already been 1730 * closed, then reject. 1731 */ 1732 if ((u_int)dfd >= fdp->fd_nfiles || 1733 (wfp = fdp->fd_files[dfd].fp) == NULL) { 1734 return (EBADF); 1735 } 1736 1737 /* 1738 * There are two cases of interest here. 1739 * 1740 * For ENODEV simply dup (dfd) to file descriptor 1741 * (indx) and return. 1742 * 1743 * For ENXIO steal away the file structure from (dfd) and 1744 * store it in (indx). (dfd) is effectively closed by 1745 * this operation. 1746 * 1747 * Any other error code is just returned. 1748 */ 1749 switch (error) { 1750 case ENODEV: 1751 /* 1752 * Check that the mode the file is being opened for is a 1753 * subset of the mode of the existing descriptor. 1754 */ 1755 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) 1756 return (EACCES); 1757 fp = fdp->fd_files[indx].fp; 1758 #if 0 1759 if (fp && fdp->fd_files[indx].fileflags & UF_MAPPED) 1760 (void) munmapfd(p, indx); 1761 #endif 1762 fdp->fd_files[indx].fp = wfp; 1763 fdp->fd_files[indx].fileflags = fdp->fd_files[dfd].fileflags; 1764 fhold(wfp); 1765 if (indx > fdp->fd_lastfile) 1766 fdp->fd_lastfile = indx; 1767 /* 1768 * we now own the reference to fp that the ofiles[] array 1769 * used to own. Release it. 1770 */ 1771 if (fp) 1772 fdrop(fp, curthread); 1773 return (0); 1774 1775 case ENXIO: 1776 /* 1777 * Steal away the file pointer from dfd, and stuff it into indx. 1778 */ 1779 fp = fdp->fd_files[indx].fp; 1780 #if 0 1781 if (fp && fdp->fd_files[indx].fileflags & UF_MAPPED) 1782 (void) munmapfd(p, indx); 1783 #endif 1784 fdp->fd_files[indx].fp = fdp->fd_files[dfd].fp; 1785 fdp->fd_files[indx].fileflags = fdp->fd_files[dfd].fileflags; 1786 funsetfd(fdp, dfd); 1787 1788 /* 1789 * we now own the reference to fp that the files[] array 1790 * used to own. Release it. 1791 */ 1792 if (fp) 1793 fdrop(fp, curthread); 1794 /* 1795 * Complete the clean up of the filedesc structure by 1796 * recomputing the various hints. 1797 */ 1798 if (indx > fdp->fd_lastfile) { 1799 fdp->fd_lastfile = indx; 1800 } else { 1801 while (fdp->fd_lastfile > 0 && 1802 fdp->fd_files[fdp->fd_lastfile].fp == NULL) { 1803 fdp->fd_lastfile--; 1804 } 1805 } 1806 return (0); 1807 1808 default: 1809 return (error); 1810 } 1811 /* NOTREACHED */ 1812 } 1813 1814 1815 struct filedesc_to_leader * 1816 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 1817 struct proc *leader) 1818 { 1819 struct filedesc_to_leader *fdtol; 1820 1821 fdtol = malloc(sizeof(struct filedesc_to_leader), 1822 M_FILEDESC_TO_LEADER, M_WAITOK); 1823 fdtol->fdl_refcount = 1; 1824 fdtol->fdl_holdcount = 0; 1825 fdtol->fdl_wakeup = 0; 1826 fdtol->fdl_leader = leader; 1827 if (old != NULL) { 1828 fdtol->fdl_next = old->fdl_next; 1829 fdtol->fdl_prev = old; 1830 old->fdl_next = fdtol; 1831 fdtol->fdl_next->fdl_prev = fdtol; 1832 } else { 1833 fdtol->fdl_next = fdtol; 1834 fdtol->fdl_prev = fdtol; 1835 } 1836 return fdtol; 1837 } 1838 1839 /* 1840 * Get file structures. 1841 */ 1842 static int 1843 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 1844 { 1845 struct kinfo_file kf; 1846 struct filedesc *fdp; 1847 struct file *fp; 1848 struct proc *p; 1849 int count; 1850 int error; 1851 int n; 1852 1853 /* 1854 * Note: because the number of file descriptors is calculated 1855 * in different ways for sizing vs returning the data, 1856 * there is information leakage from the first loop. However, 1857 * it is of a similar order of magnitude to the leakage from 1858 * global system statistics such as kern.openfiles. 1859 * 1860 * When just doing a count, note that we cannot just count 1861 * the elements and add f_count via the filehead list because 1862 * threaded processes share their descriptor table and f_count might 1863 * still be '1' in that case. 1864 */ 1865 count = 0; 1866 error = 0; 1867 LIST_FOREACH(p, &allproc, p_list) { 1868 if (p->p_stat == SIDL) 1869 continue; 1870 if (!PRISON_CHECK(req->td->td_proc->p_ucred, p->p_ucred) != 0) 1871 continue; 1872 if ((fdp = p->p_fd) == NULL) 1873 continue; 1874 for (n = 0; n < fdp->fd_nfiles; ++n) { 1875 if ((fp = fdp->fd_files[n].fp) == NULL) 1876 continue; 1877 if (req->oldptr == NULL) { 1878 ++count; 1879 } else { 1880 kcore_make_file(&kf, fp, p->p_pid, 1881 p->p_ucred->cr_uid, n); 1882 error = SYSCTL_OUT(req, &kf, sizeof(kf)); 1883 if (error) 1884 break; 1885 } 1886 } 1887 if (error) 1888 break; 1889 } 1890 1891 /* 1892 * When just calculating the size, overestimate a bit to try to 1893 * prevent system activity from causing the buffer-fill call 1894 * to fail later on. 1895 */ 1896 if (req->oldptr == NULL) { 1897 count = (count + 16) + (count / 10); 1898 error = SYSCTL_OUT(req, NULL, count * sizeof(kf)); 1899 } 1900 return (error); 1901 } 1902 1903 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 1904 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 1905 1906 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 1907 &maxfilesperproc, 0, "Maximum files allowed open per process"); 1908 1909 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 1910 &maxfiles, 0, "Maximum number of files"); 1911 1912 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, 1913 &maxfilesrootres, 0, "Descriptors reserved for root use"); 1914 1915 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 1916 &nfiles, 0, "System-wide number of open files"); 1917 1918 static void 1919 fildesc_drvinit(void *unused) 1920 { 1921 int fd; 1922 1923 cdevsw_add(&fildesc_cdevsw, 0, 0); 1924 for (fd = 0; fd < NUMFDESC; fd++) { 1925 make_dev(&fildesc_cdevsw, fd, 1926 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 1927 } 1928 make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 1929 make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 1930 make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 1931 } 1932 1933 struct fileops badfileops = { 1934 NULL, /* port */ 1935 NULL, /* clone */ 1936 badfo_readwrite, 1937 badfo_readwrite, 1938 badfo_ioctl, 1939 badfo_poll, 1940 badfo_kqfilter, 1941 badfo_stat, 1942 badfo_close 1943 }; 1944 1945 static int 1946 badfo_readwrite( 1947 struct file *fp, 1948 struct uio *uio, 1949 struct ucred *cred, 1950 int flags, 1951 struct thread *td 1952 ) { 1953 return (EBADF); 1954 } 1955 1956 static int 1957 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td) 1958 { 1959 return (EBADF); 1960 } 1961 1962 static int 1963 badfo_poll(struct file *fp, int events, struct ucred *cred, struct thread *td) 1964 { 1965 return (0); 1966 } 1967 1968 static int 1969 badfo_kqfilter(struct file *fp, struct knote *kn) 1970 { 1971 return (0); 1972 } 1973 1974 static int 1975 badfo_stat(struct file *fp, struct stat *sb, struct thread *td) 1976 { 1977 return (EBADF); 1978 } 1979 1980 static int 1981 badfo_close(struct file *fp, struct thread *td) 1982 { 1983 return (EBADF); 1984 } 1985 1986 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, 1987 fildesc_drvinit,NULL) 1988