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 1.51 2005/12/01 18:30:08 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 static int badfo_shutdown (struct file *fp, int how, struct thread *td); 144 145 /* 146 * Descriptor management. 147 */ 148 struct filelist filehead; /* head of list of open files */ 149 int nfiles; /* actual number of open files */ 150 extern int cmask; 151 152 /* 153 * System calls on descriptors. 154 */ 155 /* ARGSUSED */ 156 int 157 getdtablesize(struct getdtablesize_args *uap) 158 { 159 struct proc *p = curproc; 160 161 uap->sysmsg_result = 162 min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 163 return (0); 164 } 165 166 /* 167 * Duplicate a file descriptor to a particular value. 168 * 169 * note: keep in mind that a potential race condition exists when closing 170 * descriptors from a shared descriptor table (via rfork). 171 */ 172 /* ARGSUSED */ 173 int 174 dup2(struct dup2_args *uap) 175 { 176 int error; 177 178 error = kern_dup(DUP_FIXED, uap->from, uap->to, uap->sysmsg_fds); 179 180 return (error); 181 } 182 183 /* 184 * Duplicate a file descriptor. 185 */ 186 /* ARGSUSED */ 187 int 188 dup(struct dup_args *uap) 189 { 190 int error; 191 192 error = kern_dup(DUP_VARIABLE, uap->fd, 0, uap->sysmsg_fds); 193 194 return (error); 195 } 196 197 int 198 kern_fcntl(int fd, int cmd, union fcntl_dat *dat) 199 { 200 struct thread *td = curthread; 201 struct proc *p = td->td_proc; 202 struct filedesc *fdp = p->p_fd; 203 struct file *fp; 204 char *pop; 205 struct vnode *vp; 206 u_int newmin; 207 int tmp, error, flg = F_POSIX; 208 209 KKASSERT(p); 210 211 if ((unsigned)fd >= fdp->fd_nfiles || 212 (fp = fdp->fd_files[fd].fp) == NULL) 213 return (EBADF); 214 pop = &fdp->fd_files[fd].fileflags; 215 216 switch (cmd) { 217 case F_DUPFD: 218 newmin = dat->fc_fd; 219 if (newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 220 newmin > maxfilesperproc) 221 return (EINVAL); 222 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd); 223 return (error); 224 225 case F_GETFD: 226 dat->fc_cloexec = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 227 return (0); 228 229 case F_SETFD: 230 *pop = (*pop &~ UF_EXCLOSE) | 231 (dat->fc_cloexec & FD_CLOEXEC ? UF_EXCLOSE : 0); 232 return (0); 233 234 case F_GETFL: 235 dat->fc_flags = OFLAGS(fp->f_flag); 236 return (0); 237 238 case F_SETFL: 239 fhold(fp); 240 fp->f_flag &= ~FCNTLFLAGS; 241 fp->f_flag |= FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 242 tmp = fp->f_flag & FNONBLOCK; 243 error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td); 244 if (error) { 245 fdrop(fp, td); 246 return (error); 247 } 248 tmp = fp->f_flag & FASYNC; 249 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, td); 250 if (!error) { 251 fdrop(fp, td); 252 return (0); 253 } 254 fp->f_flag &= ~FNONBLOCK; 255 tmp = 0; 256 fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td); 257 fdrop(fp, td); 258 return (error); 259 260 case F_GETOWN: 261 fhold(fp); 262 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, td); 263 fdrop(fp, td); 264 return(error); 265 266 case F_SETOWN: 267 fhold(fp); 268 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, td); 269 fdrop(fp, td); 270 return(error); 271 272 case F_SETLKW: 273 flg |= F_WAIT; 274 /* Fall into F_SETLK */ 275 276 case F_SETLK: 277 if (fp->f_type != DTYPE_VNODE) 278 return (EBADF); 279 vp = (struct vnode *)fp->f_data; 280 281 /* 282 * copyin/lockop may block 283 */ 284 fhold(fp); 285 if (dat->fc_flock.l_whence == SEEK_CUR) 286 dat->fc_flock.l_start += fp->f_offset; 287 288 switch (dat->fc_flock.l_type) { 289 case F_RDLCK: 290 if ((fp->f_flag & FREAD) == 0) { 291 error = EBADF; 292 break; 293 } 294 p->p_leader->p_flag |= P_ADVLOCK; 295 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 296 &dat->fc_flock, flg); 297 break; 298 case F_WRLCK: 299 if ((fp->f_flag & FWRITE) == 0) { 300 error = EBADF; 301 break; 302 } 303 p->p_leader->p_flag |= P_ADVLOCK; 304 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 305 &dat->fc_flock, flg); 306 break; 307 case F_UNLCK: 308 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 309 &dat->fc_flock, F_POSIX); 310 break; 311 default: 312 error = EINVAL; 313 break; 314 } 315 /* Check for race with close */ 316 if ((unsigned) fd >= fdp->fd_nfiles || 317 fp != fdp->fd_files[fd].fp) { 318 dat->fc_flock.l_whence = SEEK_SET; 319 dat->fc_flock.l_start = 0; 320 dat->fc_flock.l_len = 0; 321 dat->fc_flock.l_type = F_UNLCK; 322 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 323 F_UNLCK, &dat->fc_flock, F_POSIX); 324 } 325 fdrop(fp, td); 326 return(error); 327 328 case F_GETLK: 329 if (fp->f_type != DTYPE_VNODE) 330 return (EBADF); 331 vp = (struct vnode *)fp->f_data; 332 /* 333 * copyin/lockop may block 334 */ 335 fhold(fp); 336 if (dat->fc_flock.l_type != F_RDLCK && 337 dat->fc_flock.l_type != F_WRLCK && 338 dat->fc_flock.l_type != F_UNLCK) { 339 fdrop(fp, td); 340 return (EINVAL); 341 } 342 if (dat->fc_flock.l_whence == SEEK_CUR) 343 dat->fc_flock.l_start += fp->f_offset; 344 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 345 &dat->fc_flock, F_POSIX); 346 fdrop(fp, td); 347 return(error); 348 default: 349 return (EINVAL); 350 } 351 /* NOTREACHED */ 352 } 353 354 /* 355 * The file control system call. 356 */ 357 int 358 fcntl(struct fcntl_args *uap) 359 { 360 union fcntl_dat dat; 361 int error; 362 363 switch (uap->cmd) { 364 case F_DUPFD: 365 dat.fc_fd = uap->arg; 366 break; 367 case F_SETFD: 368 dat.fc_cloexec = uap->arg; 369 break; 370 case F_SETFL: 371 dat.fc_flags = uap->arg; 372 break; 373 case F_SETOWN: 374 dat.fc_owner = uap->arg; 375 break; 376 case F_SETLKW: 377 case F_SETLK: 378 case F_GETLK: 379 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 380 sizeof(struct flock)); 381 if (error) 382 return (error); 383 break; 384 } 385 386 error = kern_fcntl(uap->fd, uap->cmd, &dat); 387 388 if (error == 0) { 389 switch (uap->cmd) { 390 case F_DUPFD: 391 uap->sysmsg_result = dat.fc_fd; 392 break; 393 case F_GETFD: 394 uap->sysmsg_result = dat.fc_cloexec; 395 break; 396 case F_GETFL: 397 uap->sysmsg_result = dat.fc_flags; 398 break; 399 case F_GETOWN: 400 uap->sysmsg_result = dat.fc_owner; 401 case F_GETLK: 402 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 403 sizeof(struct flock)); 404 break; 405 } 406 } 407 408 return (error); 409 } 410 411 /* 412 * Common code for dup, dup2, and fcntl(F_DUPFD). 413 * 414 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells 415 * kern_dup() to destructively dup over an existing file descriptor if new 416 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest 417 * unused file descriptor that is greater than or equal to new. 418 */ 419 int 420 kern_dup(enum dup_type type, int old, int new, int *res) 421 { 422 struct thread *td = curthread; 423 struct proc *p = td->td_proc; 424 struct filedesc *fdp = p->p_fd; 425 struct file *fp; 426 struct file *delfp; 427 int holdleaders; 428 boolean_t fdalloced = FALSE; 429 int error, newfd; 430 431 /* 432 * Verify that we have a valid descriptor to dup from and 433 * possibly to dup to. 434 */ 435 if (old < 0 || new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 436 new >= maxfilesperproc) 437 return (EBADF); 438 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) 439 return (EBADF); 440 if (type == DUP_FIXED && old == new) { 441 *res = new; 442 return (0); 443 } 444 fp = fdp->fd_files[old].fp; 445 fhold(fp); 446 447 /* 448 * Expand the table for the new descriptor if needed. This may 449 * block and drop and reacquire the fidedesc lock. 450 */ 451 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) { 452 error = fdalloc(p, new, &newfd); 453 if (error) { 454 fdrop(fp, td); 455 return (error); 456 } 457 fdalloced = TRUE; 458 } 459 if (type == DUP_VARIABLE) 460 new = newfd; 461 462 /* 463 * If the old file changed out from under us then treat it as a 464 * bad file descriptor. Userland should do its own locking to 465 * avoid this case. 466 */ 467 if (fdp->fd_files[old].fp != fp) { 468 if (fdp->fd_files[new].fp == NULL) { 469 if (fdalloced) 470 fdreserve(fdp, newfd, -1); 471 if (new < fdp->fd_freefile) 472 fdp->fd_freefile = new; 473 while (fdp->fd_lastfile > 0 && 474 fdp->fd_files[fdp->fd_lastfile].fp == NULL) 475 fdp->fd_lastfile--; 476 } 477 fdrop(fp, td); 478 return (EBADF); 479 } 480 KASSERT(old != new, ("new fd is same as old")); 481 482 /* 483 * Save info on the descriptor being overwritten. We have 484 * to do the unmap now, but we cannot close it without 485 * introducing an ownership race for the slot. 486 */ 487 delfp = fdp->fd_files[new].fp; 488 if (delfp != NULL && p->p_fdtol != NULL) { 489 /* 490 * Ask fdfree() to sleep to ensure that all relevant 491 * process leaders can be traversed in closef(). 492 */ 493 fdp->fd_holdleaderscount++; 494 holdleaders = 1; 495 } else 496 holdleaders = 0; 497 KASSERT(delfp == NULL || type == DUP_FIXED, 498 ("dup() picked an open file")); 499 #if 0 500 if (delfp && (fdp->fd_files[new].fileflags & UF_MAPPED)) 501 (void) munmapfd(p, new); 502 #endif 503 504 /* 505 * Duplicate the source descriptor, update lastfile 506 */ 507 if (new > fdp->fd_lastfile) 508 fdp->fd_lastfile = new; 509 if (!fdalloced && fdp->fd_files[new].fp == NULL) 510 fdreserve(fdp, new, 1); 511 fdp->fd_files[new].fp = fp; 512 fdp->fd_files[new].fileflags = 513 fdp->fd_files[old].fileflags & ~UF_EXCLOSE; 514 *res = new; 515 516 /* 517 * If we dup'd over a valid file, we now own the reference to it 518 * and must dispose of it using closef() semantics (as if a 519 * close() were performed on it). 520 */ 521 if (delfp) { 522 (void) closef(delfp, td); 523 if (holdleaders) { 524 fdp->fd_holdleaderscount--; 525 if (fdp->fd_holdleaderscount == 0 && 526 fdp->fd_holdleaderswakeup != 0) { 527 fdp->fd_holdleaderswakeup = 0; 528 wakeup(&fdp->fd_holdleaderscount); 529 } 530 } 531 } 532 return (0); 533 } 534 535 /* 536 * If sigio is on the list associated with a process or process group, 537 * disable signalling from the device, remove sigio from the list and 538 * free sigio. 539 */ 540 void 541 funsetown(struct sigio *sigio) 542 { 543 if (sigio == NULL) 544 return; 545 crit_enter(); 546 *(sigio->sio_myref) = NULL; 547 crit_exit(); 548 if (sigio->sio_pgid < 0) { 549 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 550 sigio, sio_pgsigio); 551 } else /* if ((*sigiop)->sio_pgid > 0) */ { 552 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 553 sigio, sio_pgsigio); 554 } 555 crfree(sigio->sio_ucred); 556 free(sigio, M_SIGIO); 557 } 558 559 /* Free a list of sigio structures. */ 560 void 561 funsetownlst(struct sigiolst *sigiolst) 562 { 563 struct sigio *sigio; 564 565 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) 566 funsetown(sigio); 567 } 568 569 /* 570 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 571 * 572 * After permission checking, add a sigio structure to the sigio list for 573 * the process or process group. 574 */ 575 int 576 fsetown(pid_t pgid, struct sigio **sigiop) 577 { 578 struct proc *proc; 579 struct pgrp *pgrp; 580 struct sigio *sigio; 581 582 if (pgid == 0) { 583 funsetown(*sigiop); 584 return (0); 585 } 586 if (pgid > 0) { 587 proc = pfind(pgid); 588 if (proc == NULL) 589 return (ESRCH); 590 591 /* 592 * Policy - Don't allow a process to FSETOWN a process 593 * in another session. 594 * 595 * Remove this test to allow maximum flexibility or 596 * restrict FSETOWN to the current process or process 597 * group for maximum safety. 598 */ 599 if (proc->p_session != curproc->p_session) 600 return (EPERM); 601 602 pgrp = NULL; 603 } else /* if (pgid < 0) */ { 604 pgrp = pgfind(-pgid); 605 if (pgrp == NULL) 606 return (ESRCH); 607 608 /* 609 * Policy - Don't allow a process to FSETOWN a process 610 * in another session. 611 * 612 * Remove this test to allow maximum flexibility or 613 * restrict FSETOWN to the current process or process 614 * group for maximum safety. 615 */ 616 if (pgrp->pg_session != curproc->p_session) 617 return (EPERM); 618 619 proc = NULL; 620 } 621 funsetown(*sigiop); 622 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 623 if (pgid > 0) { 624 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 625 sigio->sio_proc = proc; 626 } else { 627 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 628 sigio->sio_pgrp = pgrp; 629 } 630 sigio->sio_pgid = pgid; 631 sigio->sio_ucred = crhold(curproc->p_ucred); 632 /* It would be convenient if p_ruid was in ucred. */ 633 sigio->sio_ruid = curproc->p_ucred->cr_ruid; 634 sigio->sio_myref = sigiop; 635 crit_enter(); 636 *sigiop = sigio; 637 crit_exit(); 638 return (0); 639 } 640 641 /* 642 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 643 */ 644 pid_t 645 fgetown(struct sigio *sigio) 646 { 647 return (sigio != NULL ? sigio->sio_pgid : 0); 648 } 649 650 /* 651 * Close many file descriptors. 652 */ 653 /* ARGSUSED */ 654 655 int 656 closefrom(struct closefrom_args *uap) 657 { 658 return(kern_closefrom(uap->fd)); 659 } 660 661 int 662 kern_closefrom(int fd) 663 { 664 struct thread *td = curthread; 665 struct proc *p = td->td_proc; 666 struct filedesc *fdp; 667 668 KKASSERT(p); 669 fdp = p->p_fd; 670 671 if (fd < 0 || fd > fdp->fd_lastfile) 672 return (0); 673 674 do { 675 if (kern_close(fdp->fd_lastfile) == EINTR) 676 return (EINTR); 677 } while (fdp->fd_lastfile > fd); 678 679 return (0); 680 } 681 682 /* 683 * Close a file descriptor. 684 */ 685 /* ARGSUSED */ 686 687 int 688 close(struct close_args *uap) 689 { 690 return(kern_close(uap->fd)); 691 } 692 693 int 694 kern_close(int fd) 695 { 696 struct thread *td = curthread; 697 struct proc *p = td->td_proc; 698 struct filedesc *fdp; 699 struct file *fp; 700 int error; 701 int holdleaders; 702 703 KKASSERT(p); 704 fdp = p->p_fd; 705 706 if ((unsigned)fd >= fdp->fd_nfiles || 707 (fp = fdp->fd_files[fd].fp) == NULL) 708 return (EBADF); 709 #if 0 710 if (fdp->fd_files[fd].fileflags & UF_MAPPED) 711 (void) munmapfd(p, fd); 712 #endif 713 funsetfd(fdp, fd); 714 holdleaders = 0; 715 if (p->p_fdtol != NULL) { 716 /* 717 * Ask fdfree() to sleep to ensure that all relevant 718 * process leaders can be traversed in closef(). 719 */ 720 fdp->fd_holdleaderscount++; 721 holdleaders = 1; 722 } 723 724 /* 725 * we now hold the fp reference that used to be owned by the descriptor 726 * array. 727 */ 728 while (fdp->fd_lastfile > 0 && fdp->fd_files[fdp->fd_lastfile].fp == NULL) 729 fdp->fd_lastfile--; 730 if (fd < fdp->fd_knlistsize) 731 knote_fdclose(p, fd); 732 error = closef(fp, td); 733 if (holdleaders) { 734 fdp->fd_holdleaderscount--; 735 if (fdp->fd_holdleaderscount == 0 && 736 fdp->fd_holdleaderswakeup != 0) { 737 fdp->fd_holdleaderswakeup = 0; 738 wakeup(&fdp->fd_holdleaderscount); 739 } 740 } 741 return (error); 742 } 743 744 /* 745 * shutdown_args(int fd, int how) 746 */ 747 int 748 kern_shutdown(int fd, int how) 749 { 750 struct thread *td = curthread; 751 struct proc *p = td->td_proc; 752 struct filedesc *fdp; 753 struct file *fp; 754 int error; 755 756 KKASSERT(p); 757 758 fdp = p->p_fd; 759 if ((unsigned)fd >= fdp->fd_nfiles || 760 (fp = fdp->fd_files[fd].fp) == NULL) 761 return (EBADF); 762 fhold(fp); 763 error = fo_shutdown(fp, how, td); 764 fdrop(fp, td); 765 766 return (error); 767 } 768 769 int 770 shutdown(struct shutdown_args *uap) 771 { 772 int error; 773 774 error = kern_shutdown(uap->s, uap->how); 775 776 return (error); 777 } 778 779 int 780 kern_fstat(int fd, struct stat *ub) 781 { 782 struct thread *td = curthread; 783 struct proc *p = td->td_proc; 784 struct filedesc *fdp; 785 struct file *fp; 786 int error; 787 788 KKASSERT(p); 789 790 fdp = p->p_fd; 791 if ((unsigned)fd >= fdp->fd_nfiles || 792 (fp = fdp->fd_files[fd].fp) == NULL) 793 return (EBADF); 794 fhold(fp); 795 error = fo_stat(fp, ub, td); 796 fdrop(fp, td); 797 798 return (error); 799 } 800 801 /* 802 * Return status information about a file descriptor. 803 */ 804 int 805 fstat(struct fstat_args *uap) 806 { 807 struct stat st; 808 int error; 809 810 error = kern_fstat(uap->fd, &st); 811 812 if (error == 0) 813 error = copyout(&st, uap->sb, sizeof(st)); 814 return (error); 815 } 816 817 /* 818 * Return pathconf information about a file descriptor. 819 */ 820 /* ARGSUSED */ 821 int 822 fpathconf(struct fpathconf_args *uap) 823 { 824 struct thread *td = curthread; 825 struct proc *p = td->td_proc; 826 struct filedesc *fdp; 827 struct file *fp; 828 struct vnode *vp; 829 int error = 0; 830 831 KKASSERT(p); 832 fdp = p->p_fd; 833 if ((unsigned)uap->fd >= fdp->fd_nfiles || 834 (fp = fdp->fd_files[uap->fd].fp) == NULL) 835 return (EBADF); 836 837 fhold(fp); 838 839 switch (fp->f_type) { 840 case DTYPE_PIPE: 841 case DTYPE_SOCKET: 842 if (uap->name != _PC_PIPE_BUF) { 843 error = EINVAL; 844 } else { 845 uap->sysmsg_result = PIPE_BUF; 846 error = 0; 847 } 848 break; 849 case DTYPE_FIFO: 850 case DTYPE_VNODE: 851 vp = (struct vnode *)fp->f_data; 852 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds); 853 break; 854 default: 855 error = EOPNOTSUPP; 856 break; 857 } 858 fdrop(fp, td); 859 return(error); 860 } 861 862 static int fdexpand; 863 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, ""); 864 865 static void 866 fdgrow(struct filedesc *fdp, int want) 867 { 868 struct fdnode *newfiles; 869 struct fdnode *oldfiles; 870 int nf, extra; 871 872 nf = fdp->fd_nfiles; 873 do { 874 /* nf has to be of the form 2^n - 1 */ 875 nf = 2 * nf + 1; 876 } while (nf <= want); 877 878 newfiles = malloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK); 879 880 /* 881 * deal with file-table extend race that might have occured 882 * when malloc was blocked. 883 */ 884 if (fdp->fd_nfiles >= nf) { 885 free(newfiles, M_FILEDESC); 886 return; 887 } 888 /* 889 * Copy the existing ofile and ofileflags arrays 890 * and zero the new portion of each array. 891 */ 892 extra = nf - fdp->fd_nfiles; 893 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode)); 894 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode)); 895 896 oldfiles = fdp->fd_files; 897 fdp->fd_files = newfiles; 898 fdp->fd_nfiles = nf; 899 900 if (oldfiles != fdp->fd_builtin_files) 901 free(oldfiles, M_FILEDESC); 902 fdexpand++; 903 } 904 905 /* 906 * Number of nodes in right subtree, including the root. 907 */ 908 static __inline int 909 right_subtree_size(int n) 910 { 911 return (n ^ (n | (n + 1))); 912 } 913 914 /* 915 * Bigger ancestor. 916 */ 917 static __inline int 918 right_ancestor(int n) 919 { 920 return (n | (n + 1)); 921 } 922 923 /* 924 * Smaller ancestor. 925 */ 926 static __inline int 927 left_ancestor(int n) 928 { 929 return ((n & (n + 1)) - 1); 930 } 931 932 void 933 fdreserve(struct filedesc *fdp, int fd, int incr) 934 { 935 while (fd >= 0) { 936 fdp->fd_files[fd].allocated += incr; 937 KKASSERT(fdp->fd_files[fd].allocated >= 0); 938 fd = left_ancestor(fd); 939 } 940 } 941 942 /* 943 * Allocate a file descriptor for the process. 944 */ 945 int 946 fdalloc(struct proc *p, int want, int *result) 947 { 948 struct filedesc *fdp = p->p_fd; 949 int fd, rsize, rsum, node, lim; 950 951 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 952 if (want >= lim) 953 return (EMFILE); 954 if (want >= fdp->fd_nfiles) 955 fdgrow(fdp, want); 956 957 /* 958 * Search for a free descriptor starting at the higher 959 * of want or fd_freefile. If that fails, consider 960 * expanding the ofile array. 961 */ 962 retry: 963 /* move up the tree looking for a subtree with a free node */ 964 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim); 965 fd = right_ancestor(fd)) { 966 if (fdp->fd_files[fd].allocated == 0) 967 goto found; 968 969 rsize = right_subtree_size(fd); 970 if (fdp->fd_files[fd].allocated == rsize) 971 continue; /* right subtree full */ 972 973 /* 974 * Free fd is in the right subtree of the tree rooted at fd. 975 * Call that subtree R. Look for the smallest (leftmost) 976 * subtree of R with an unallocated fd: continue moving 977 * down the left branch until encountering a full left 978 * subtree, then move to the right. 979 */ 980 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) { 981 node = fd + rsize; 982 rsum += fdp->fd_files[node].allocated; 983 if (fdp->fd_files[fd].allocated == rsum + rsize) { 984 fd = node; /* move to the right */ 985 if (fdp->fd_files[node].allocated == 0) 986 goto found; 987 rsum = 0; 988 } 989 } 990 goto found; 991 } 992 993 /* 994 * No space in current array. Expand? 995 */ 996 if (fdp->fd_nfiles >= lim) 997 return (EMFILE); 998 fdgrow(fdp, want); 999 goto retry; 1000 1001 found: 1002 KKASSERT(fd < fdp->fd_nfiles); 1003 fdp->fd_files[fd].fileflags = 0; 1004 if (fd > fdp->fd_lastfile) 1005 fdp->fd_lastfile = fd; 1006 if (want <= fdp->fd_freefile) 1007 fdp->fd_freefile = fd; 1008 *result = fd; 1009 KKASSERT(fdp->fd_files[fd].fp == NULL); 1010 fdreserve(fdp, fd, 1); 1011 return (0); 1012 } 1013 1014 /* 1015 * Check to see whether n user file descriptors 1016 * are available to the process p. 1017 */ 1018 int 1019 fdavail(struct proc *p, int n) 1020 { 1021 struct filedesc *fdp = p->p_fd; 1022 struct fdnode *fdnode; 1023 int i, lim, last; 1024 1025 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 1026 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1027 return (1); 1028 1029 last = min(fdp->fd_nfiles, lim); 1030 fdnode = &fdp->fd_files[fdp->fd_freefile]; 1031 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) { 1032 if (fdnode->fp == NULL && --n <= 0) 1033 return (1); 1034 } 1035 return (0); 1036 } 1037 1038 /* 1039 * falloc: 1040 * Create a new open file structure and allocate a file decriptor 1041 * for the process that refers to it. If p is NULL, no descriptor 1042 * is allocated and the file pointer is returned unassociated with 1043 * any process. resultfd is only used if p is not NULL and may 1044 * separately be NULL indicating that you don't need the returned fd. 1045 * 1046 * A held file pointer is returned. If a descriptor has been allocated 1047 * an additional hold on the fp will be made due to the fd_files[] 1048 * reference. 1049 */ 1050 int 1051 falloc(struct proc *p, struct file **resultfp, int *resultfd) 1052 { 1053 static struct timeval lastfail; 1054 static int curfail; 1055 struct file *fp; 1056 int error; 1057 1058 fp = NULL; 1059 1060 /* 1061 * Handle filetable full issues and root overfill. 1062 */ 1063 if (nfiles >= maxfiles - maxfilesrootres && 1064 ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) { 1065 if (ppsratecheck(&lastfail, &curfail, 1)) { 1066 printf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n", 1067 (p ? p->p_ucred->cr_ruid : -1)); 1068 } 1069 error = ENFILE; 1070 goto done; 1071 } 1072 1073 /* 1074 * Allocate a new file descriptor. 1075 */ 1076 nfiles++; 1077 fp = malloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO); 1078 fp->f_count = 1; 1079 fp->f_ops = &badfileops; 1080 fp->f_seqcount = 1; 1081 if (p) 1082 fp->f_cred = crhold(p->p_ucred); 1083 else 1084 fp->f_cred = crhold(proc0.p_ucred); 1085 LIST_INSERT_HEAD(&filehead, fp, f_list); 1086 if (resultfd) { 1087 if ((error = fsetfd(p, fp, resultfd)) != 0) { 1088 fdrop(fp, p->p_thread); 1089 fp = NULL; 1090 } 1091 } else { 1092 error = 0; 1093 } 1094 done: 1095 *resultfp = fp; 1096 return (error); 1097 } 1098 1099 /* 1100 * Associate a file pointer with a file descriptor. On success the fp 1101 * will have an additional ref representing the fd_files[] association. 1102 */ 1103 int 1104 fsetfd(struct proc *p, struct file *fp, int *resultfd) 1105 { 1106 int fd, error; 1107 1108 fd = -1; 1109 if ((error = fdalloc(p, 0, &fd)) == 0) { 1110 fhold(fp); 1111 p->p_fd->fd_files[fd].fp = fp; 1112 } 1113 *resultfd = fd; 1114 return (error); 1115 } 1116 1117 void 1118 funsetfd(struct filedesc *fdp, int fd) 1119 { 1120 fdp->fd_files[fd].fp = NULL; 1121 fdp->fd_files[fd].fileflags = 0; 1122 fdreserve(fdp, fd, -1); 1123 if (fd < fdp->fd_freefile) 1124 fdp->fd_freefile = fd; 1125 } 1126 1127 void 1128 fsetcred(struct file *fp, struct ucred *cr) 1129 { 1130 crhold(cr); 1131 crfree(fp->f_cred); 1132 fp->f_cred = cr; 1133 } 1134 1135 /* 1136 * Free a file descriptor. 1137 */ 1138 void 1139 ffree(struct file *fp) 1140 { 1141 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 1142 LIST_REMOVE(fp, f_list); 1143 crfree(fp->f_cred); 1144 if (fp->f_ncp) { 1145 cache_drop(fp->f_ncp); 1146 fp->f_ncp = NULL; 1147 } 1148 nfiles--; 1149 free(fp, M_FILE); 1150 } 1151 1152 /* 1153 * Build a new filedesc structure. 1154 */ 1155 struct filedesc * 1156 fdinit(struct proc *p) 1157 { 1158 struct filedesc *newfdp; 1159 struct filedesc *fdp = p->p_fd; 1160 1161 newfdp = malloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO); 1162 if (fdp->fd_cdir) { 1163 newfdp->fd_cdir = fdp->fd_cdir; 1164 vref(newfdp->fd_cdir); 1165 newfdp->fd_ncdir = cache_hold(fdp->fd_ncdir); 1166 } 1167 1168 /* 1169 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of 1170 * proc0, but should unconditionally exist in other processes. 1171 */ 1172 if (fdp->fd_rdir) { 1173 newfdp->fd_rdir = fdp->fd_rdir; 1174 vref(newfdp->fd_rdir); 1175 newfdp->fd_nrdir = cache_hold(fdp->fd_nrdir); 1176 } 1177 if (fdp->fd_jdir) { 1178 newfdp->fd_jdir = fdp->fd_jdir; 1179 vref(newfdp->fd_jdir); 1180 newfdp->fd_njdir = cache_hold(fdp->fd_njdir); 1181 } 1182 1183 /* Create the file descriptor table. */ 1184 newfdp->fd_refcnt = 1; 1185 newfdp->fd_cmask = cmask; 1186 newfdp->fd_files = newfdp->fd_builtin_files; 1187 newfdp->fd_nfiles = NDFILE; 1188 newfdp->fd_knlistsize = -1; 1189 1190 return (newfdp); 1191 } 1192 1193 /* 1194 * Share a filedesc structure. 1195 */ 1196 struct filedesc * 1197 fdshare(struct proc *p) 1198 { 1199 p->p_fd->fd_refcnt++; 1200 return (p->p_fd); 1201 } 1202 1203 /* 1204 * Copy a filedesc structure. 1205 */ 1206 struct filedesc * 1207 fdcopy(struct proc *p) 1208 { 1209 struct filedesc *newfdp, *fdp = p->p_fd; 1210 struct fdnode *fdnode; 1211 int i; 1212 1213 /* Certain daemons might not have file descriptors. */ 1214 if (fdp == NULL) 1215 return (NULL); 1216 1217 newfdp = malloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK); 1218 *newfdp = *fdp; 1219 if (newfdp->fd_cdir) { 1220 vref(newfdp->fd_cdir); 1221 newfdp->fd_ncdir = cache_hold(newfdp->fd_ncdir); 1222 } 1223 /* 1224 * We must check for fd_rdir here, at least for now because 1225 * the init process is created before we have access to the 1226 * rootvode to take a reference to it. 1227 */ 1228 if (newfdp->fd_rdir) { 1229 vref(newfdp->fd_rdir); 1230 newfdp->fd_nrdir = cache_hold(newfdp->fd_nrdir); 1231 } 1232 if (newfdp->fd_jdir) { 1233 vref(newfdp->fd_jdir); 1234 newfdp->fd_njdir = cache_hold(newfdp->fd_njdir); 1235 } 1236 newfdp->fd_refcnt = 1; 1237 1238 /* 1239 * If the number of open files fits in the internal arrays 1240 * of the open file structure, use them, otherwise allocate 1241 * additional memory for the number of descriptors currently 1242 * in use. 1243 */ 1244 if (newfdp->fd_lastfile < NDFILE) { 1245 newfdp->fd_files = newfdp->fd_builtin_files; 1246 i = NDFILE; 1247 } else { 1248 /* 1249 * Compute the smallest file table size 1250 * for the file descriptors currently in use, 1251 * allowing the table to shrink. 1252 */ 1253 i = newfdp->fd_nfiles; 1254 while ((i-1)/2 > newfdp->fd_lastfile && (i-1)/2 > NDFILE) 1255 i = (i-1)/2; 1256 newfdp->fd_files = malloc(i * sizeof(struct fdnode), 1257 M_FILEDESC, M_WAITOK); 1258 } 1259 newfdp->fd_nfiles = i; 1260 1261 if (fdp->fd_files != fdp->fd_builtin_files || 1262 newfdp->fd_files != newfdp->fd_builtin_files 1263 ) { 1264 bcopy(fdp->fd_files, newfdp->fd_files, 1265 i * sizeof(struct fdnode)); 1266 } 1267 1268 /* 1269 * kq descriptors cannot be copied. 1270 */ 1271 if (newfdp->fd_knlistsize != -1) { 1272 fdnode = &newfdp->fd_files[newfdp->fd_lastfile]; 1273 for (i = newfdp->fd_lastfile; i >= 0; i--, fdnode--) { 1274 if (fdnode->fp != NULL && fdnode->fp->f_type == DTYPE_KQUEUE) 1275 funsetfd(newfdp, i); /* nulls out *fpp */ 1276 if (fdnode->fp == NULL && i == newfdp->fd_lastfile && i > 0) 1277 newfdp->fd_lastfile--; 1278 } 1279 newfdp->fd_knlist = NULL; 1280 newfdp->fd_knlistsize = -1; 1281 newfdp->fd_knhash = NULL; 1282 newfdp->fd_knhashmask = 0; 1283 } 1284 1285 fdnode = newfdp->fd_files; 1286 for (i = newfdp->fd_lastfile; i-- >= 0; fdnode++) { 1287 if (fdnode->fp != NULL) 1288 fhold(fdnode->fp); 1289 } 1290 return (newfdp); 1291 } 1292 1293 /* 1294 * Release a filedesc structure. 1295 */ 1296 void 1297 fdfree(struct proc *p) 1298 { 1299 struct thread *td = p->p_thread; 1300 struct filedesc *fdp = p->p_fd; 1301 struct fdnode *fdnode; 1302 int i; 1303 struct filedesc_to_leader *fdtol; 1304 struct file *fp; 1305 struct vnode *vp; 1306 struct flock lf; 1307 1308 /* Certain daemons might not have file descriptors. */ 1309 if (fdp == NULL) 1310 return; 1311 1312 /* Check for special need to clear POSIX style locks */ 1313 fdtol = p->p_fdtol; 1314 if (fdtol != NULL) { 1315 KASSERT(fdtol->fdl_refcount > 0, 1316 ("filedesc_to_refcount botch: fdl_refcount=%d", 1317 fdtol->fdl_refcount)); 1318 if (fdtol->fdl_refcount == 1 && 1319 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1320 i = 0; 1321 fdnode = fdp->fd_files; 1322 for (i = 0; i <= fdp->fd_lastfile; i++, fdnode++) { 1323 if (fdnode->fp == NULL || 1324 fdnode->fp->f_type != DTYPE_VNODE) 1325 continue; 1326 fp = fdnode->fp; 1327 fhold(fp); 1328 lf.l_whence = SEEK_SET; 1329 lf.l_start = 0; 1330 lf.l_len = 0; 1331 lf.l_type = F_UNLCK; 1332 vp = (struct vnode *)fp->f_data; 1333 (void) VOP_ADVLOCK(vp, 1334 (caddr_t)p->p_leader, 1335 F_UNLCK, 1336 &lf, 1337 F_POSIX); 1338 fdrop(fp, p->p_thread); 1339 /* reload due to possible reallocation */ 1340 fdnode = &fdp->fd_files[i]; 1341 } 1342 } 1343 retry: 1344 if (fdtol->fdl_refcount == 1) { 1345 if (fdp->fd_holdleaderscount > 0 && 1346 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1347 /* 1348 * close() or do_dup() has cleared a reference 1349 * in a shared file descriptor table. 1350 */ 1351 fdp->fd_holdleaderswakeup = 1; 1352 tsleep(&fdp->fd_holdleaderscount, 1353 0, "fdlhold", 0); 1354 goto retry; 1355 } 1356 if (fdtol->fdl_holdcount > 0) { 1357 /* 1358 * Ensure that fdtol->fdl_leader 1359 * remains valid in closef(). 1360 */ 1361 fdtol->fdl_wakeup = 1; 1362 tsleep(fdtol, 0, "fdlhold", 0); 1363 goto retry; 1364 } 1365 } 1366 fdtol->fdl_refcount--; 1367 if (fdtol->fdl_refcount == 0 && 1368 fdtol->fdl_holdcount == 0) { 1369 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1370 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1371 } else 1372 fdtol = NULL; 1373 p->p_fdtol = NULL; 1374 if (fdtol != NULL) 1375 free(fdtol, M_FILEDESC_TO_LEADER); 1376 } 1377 if (--fdp->fd_refcnt > 0) 1378 return; 1379 /* 1380 * we are the last reference to the structure, we can 1381 * safely assume it will not change out from under us. 1382 */ 1383 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1384 if (fdp->fd_files[i].fp) 1385 closef(fdp->fd_files[i].fp, td); 1386 } 1387 if (fdp->fd_files != fdp->fd_builtin_files) 1388 free(fdp->fd_files, M_FILEDESC); 1389 if (fdp->fd_cdir) { 1390 cache_drop(fdp->fd_ncdir); 1391 vrele(fdp->fd_cdir); 1392 } 1393 if (fdp->fd_rdir) { 1394 cache_drop(fdp->fd_nrdir); 1395 vrele(fdp->fd_rdir); 1396 } 1397 if (fdp->fd_jdir) { 1398 cache_drop(fdp->fd_njdir); 1399 vrele(fdp->fd_jdir); 1400 } 1401 if (fdp->fd_knlist) 1402 free(fdp->fd_knlist, M_KQUEUE); 1403 if (fdp->fd_knhash) 1404 free(fdp->fd_knhash, M_KQUEUE); 1405 free(fdp, M_FILEDESC); 1406 } 1407 1408 /* 1409 * For setugid programs, we don't want to people to use that setugidness 1410 * to generate error messages which write to a file which otherwise would 1411 * otherwise be off-limits to the process. 1412 * 1413 * This is a gross hack to plug the hole. A better solution would involve 1414 * a special vop or other form of generalized access control mechanism. We 1415 * go ahead and just reject all procfs file systems accesses as dangerous. 1416 * 1417 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1418 * sufficient. We also don't for check setugidness since we know we are. 1419 */ 1420 static int 1421 is_unsafe(struct file *fp) 1422 { 1423 if (fp->f_type == DTYPE_VNODE && 1424 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 1425 return (1); 1426 return (0); 1427 } 1428 1429 /* 1430 * Make this setguid thing safe, if at all possible. 1431 */ 1432 void 1433 setugidsafety(struct proc *p) 1434 { 1435 struct thread *td = p->p_thread; 1436 struct filedesc *fdp = p->p_fd; 1437 int i; 1438 1439 /* Certain daemons might not have file descriptors. */ 1440 if (fdp == NULL) 1441 return; 1442 1443 /* 1444 * note: fdp->fd_files may be reallocated out from under us while 1445 * we are blocked in a close. Be careful! 1446 */ 1447 for (i = 0; i <= fdp->fd_lastfile; i++) { 1448 if (i > 2) 1449 break; 1450 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) { 1451 struct file *fp; 1452 1453 #if 0 1454 if ((fdp->fd_files[i].fileflags & UF_MAPPED) != 0) 1455 (void) munmapfd(p, i); 1456 #endif 1457 if (i < fdp->fd_knlistsize) 1458 knote_fdclose(p, i); 1459 /* 1460 * NULL-out descriptor prior to close to avoid 1461 * a race while close blocks. 1462 */ 1463 fp = fdp->fd_files[i].fp; 1464 funsetfd(fdp, i); 1465 closef(fp, td); 1466 } 1467 } 1468 while (fdp->fd_lastfile > 0 && fdp->fd_files[fdp->fd_lastfile].fp == NULL) 1469 fdp->fd_lastfile--; 1470 } 1471 1472 /* 1473 * Close any files on exec? 1474 */ 1475 void 1476 fdcloseexec(struct proc *p) 1477 { 1478 struct thread *td = p->p_thread; 1479 struct filedesc *fdp = p->p_fd; 1480 int i; 1481 1482 /* Certain daemons might not have file descriptors. */ 1483 if (fdp == NULL) 1484 return; 1485 1486 /* 1487 * We cannot cache fd_files since operations may block and rip 1488 * them out from under us. 1489 */ 1490 for (i = 0; i <= fdp->fd_lastfile; i++) { 1491 if (fdp->fd_files[i].fp != NULL && 1492 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) { 1493 struct file *fp; 1494 1495 #if 0 1496 if (fdp->fd_files[i].fileflags & UF_MAPPED) 1497 (void) munmapfd(p, i); 1498 #endif 1499 if (i < fdp->fd_knlistsize) 1500 knote_fdclose(p, i); 1501 /* 1502 * NULL-out descriptor prior to close to avoid 1503 * a race while close blocks. 1504 */ 1505 fp = fdp->fd_files[i].fp; 1506 funsetfd(fdp, i); 1507 closef(fp, td); 1508 } 1509 } 1510 while (fdp->fd_lastfile > 0 && fdp->fd_files[fdp->fd_lastfile].fp == NULL) 1511 fdp->fd_lastfile--; 1512 } 1513 1514 /* 1515 * It is unsafe for set[ug]id processes to be started with file 1516 * descriptors 0..2 closed, as these descriptors are given implicit 1517 * significance in the Standard C library. fdcheckstd() will create a 1518 * descriptor referencing /dev/null for each of stdin, stdout, and 1519 * stderr that is not already open. 1520 */ 1521 int 1522 fdcheckstd(struct proc *p) 1523 { 1524 struct thread *td = p->p_thread; 1525 struct nlookupdata nd; 1526 struct filedesc *fdp; 1527 struct file *fp; 1528 register_t retval; 1529 int fd, i, error, flags, devnull; 1530 1531 fdp = p->p_fd; 1532 if (fdp == NULL) 1533 return (0); 1534 devnull = -1; 1535 error = 0; 1536 for (i = 0; i < 3; i++) { 1537 if (fdp->fd_files[i].fp != NULL) 1538 continue; 1539 if (devnull < 0) { 1540 if ((error = falloc(p, &fp, NULL)) != 0) 1541 break; 1542 1543 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, 1544 NLC_FOLLOW|NLC_LOCKVP); 1545 flags = FREAD | FWRITE; 1546 if (error == 0) 1547 error = vn_open(&nd, fp, flags, 0); 1548 if (error == 0) 1549 error = fsetfd(p, fp, &fd); 1550 fdrop(fp, td); 1551 nlookup_done(&nd); 1552 if (error) 1553 break; 1554 KKASSERT(i == fd); 1555 devnull = fd; 1556 } else { 1557 error = kern_dup(DUP_FIXED, devnull, i, &retval); 1558 if (error != 0) 1559 break; 1560 } 1561 } 1562 return (error); 1563 } 1564 1565 /* 1566 * Internal form of close. 1567 * Decrement reference count on file structure. 1568 * Note: td and/or p may be NULL when closing a file 1569 * that was being passed in a message. 1570 */ 1571 int 1572 closef(struct file *fp, struct thread *td) 1573 { 1574 struct vnode *vp; 1575 struct flock lf; 1576 struct filedesc_to_leader *fdtol; 1577 struct proc *p; 1578 1579 if (fp == NULL) 1580 return (0); 1581 if (td == NULL) { 1582 td = curthread; 1583 p = NULL; /* allow no proc association */ 1584 } else { 1585 p = td->td_proc; /* can also be NULL */ 1586 } 1587 /* 1588 * POSIX record locking dictates that any close releases ALL 1589 * locks owned by this process. This is handled by setting 1590 * a flag in the unlock to free ONLY locks obeying POSIX 1591 * semantics, and not to free BSD-style file locks. 1592 * If the descriptor was in a message, POSIX-style locks 1593 * aren't passed with the descriptor. 1594 */ 1595 if (p != NULL && 1596 fp->f_type == DTYPE_VNODE) { 1597 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) { 1598 lf.l_whence = SEEK_SET; 1599 lf.l_start = 0; 1600 lf.l_len = 0; 1601 lf.l_type = F_UNLCK; 1602 vp = (struct vnode *)fp->f_data; 1603 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 1604 &lf, F_POSIX); 1605 } 1606 fdtol = p->p_fdtol; 1607 if (fdtol != NULL) { 1608 /* 1609 * Handle special case where file descriptor table 1610 * is shared between multiple process leaders. 1611 */ 1612 for (fdtol = fdtol->fdl_next; 1613 fdtol != p->p_fdtol; 1614 fdtol = fdtol->fdl_next) { 1615 if ((fdtol->fdl_leader->p_flag & 1616 P_ADVLOCK) == 0) 1617 continue; 1618 fdtol->fdl_holdcount++; 1619 lf.l_whence = SEEK_SET; 1620 lf.l_start = 0; 1621 lf.l_len = 0; 1622 lf.l_type = F_UNLCK; 1623 vp = (struct vnode *)fp->f_data; 1624 (void) VOP_ADVLOCK(vp, 1625 (caddr_t)p->p_leader, 1626 F_UNLCK, &lf, F_POSIX); 1627 fdtol->fdl_holdcount--; 1628 if (fdtol->fdl_holdcount == 0 && 1629 fdtol->fdl_wakeup != 0) { 1630 fdtol->fdl_wakeup = 0; 1631 wakeup(fdtol); 1632 } 1633 } 1634 } 1635 } 1636 return (fdrop(fp, td)); 1637 } 1638 1639 int 1640 fdrop(struct file *fp, struct thread *td) 1641 { 1642 struct flock lf; 1643 struct vnode *vp; 1644 int error; 1645 1646 if (--fp->f_count > 0) 1647 return (0); 1648 if (fp->f_count < 0) 1649 panic("fdrop: count < 0"); 1650 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) { 1651 lf.l_whence = SEEK_SET; 1652 lf.l_start = 0; 1653 lf.l_len = 0; 1654 lf.l_type = F_UNLCK; 1655 vp = (struct vnode *)fp->f_data; 1656 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 1657 } 1658 if (fp->f_ops != &badfileops) 1659 error = fo_close(fp, td); 1660 else 1661 error = 0; 1662 ffree(fp); 1663 return (error); 1664 } 1665 1666 /* 1667 * Apply an advisory lock on a file descriptor. 1668 * 1669 * Just attempt to get a record lock of the requested type on 1670 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 1671 */ 1672 /* ARGSUSED */ 1673 int 1674 flock(struct flock_args *uap) 1675 { 1676 struct proc *p = curproc; 1677 struct filedesc *fdp = p->p_fd; 1678 struct file *fp; 1679 struct vnode *vp; 1680 struct flock lf; 1681 1682 if ((unsigned)uap->fd >= fdp->fd_nfiles || 1683 (fp = fdp->fd_files[uap->fd].fp) == NULL) 1684 return (EBADF); 1685 if (fp->f_type != DTYPE_VNODE) 1686 return (EOPNOTSUPP); 1687 vp = (struct vnode *)fp->f_data; 1688 lf.l_whence = SEEK_SET; 1689 lf.l_start = 0; 1690 lf.l_len = 0; 1691 if (uap->how & LOCK_UN) { 1692 lf.l_type = F_UNLCK; 1693 fp->f_flag &= ~FHASLOCK; 1694 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK)); 1695 } 1696 if (uap->how & LOCK_EX) 1697 lf.l_type = F_WRLCK; 1698 else if (uap->how & LOCK_SH) 1699 lf.l_type = F_RDLCK; 1700 else 1701 return (EBADF); 1702 fp->f_flag |= FHASLOCK; 1703 if (uap->how & LOCK_NB) 1704 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK)); 1705 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT)); 1706 } 1707 1708 /* 1709 * File Descriptor pseudo-device driver (/dev/fd/). 1710 * 1711 * Opening minor device N dup()s the file (if any) connected to file 1712 * descriptor N belonging to the calling process. Note that this driver 1713 * consists of only the ``open()'' routine, because all subsequent 1714 * references to this file will be direct to the other driver. 1715 */ 1716 /* ARGSUSED */ 1717 static int 1718 fdopen(dev_t dev, int mode, int type, struct thread *td) 1719 { 1720 KKASSERT(td->td_lwp != NULL); 1721 1722 /* 1723 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the 1724 * the file descriptor being sought for duplication. The error 1725 * return ensures that the vnode for this device will be released 1726 * by vn_open. Open will detect this special error and take the 1727 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 1728 * will simply report the error. 1729 */ 1730 td->td_lwp->lwp_dupfd = minor(dev); 1731 return (ENODEV); 1732 } 1733 1734 /* 1735 * Duplicate the specified descriptor to a free descriptor. 1736 */ 1737 int 1738 dupfdopen(struct filedesc *fdp, int indx, int dfd, int mode, int error) 1739 { 1740 struct file *wfp; 1741 struct file *fp; 1742 1743 /* 1744 * If the to-be-dup'd fd number is greater than the allowed number 1745 * of file descriptors, or the fd to be dup'd has already been 1746 * closed, then reject. 1747 */ 1748 if ((u_int)dfd >= fdp->fd_nfiles || 1749 (wfp = fdp->fd_files[dfd].fp) == NULL) { 1750 return (EBADF); 1751 } 1752 1753 /* 1754 * There are two cases of interest here. 1755 * 1756 * For ENODEV simply dup (dfd) to file descriptor 1757 * (indx) and return. 1758 * 1759 * For ENXIO steal away the file structure from (dfd) and 1760 * store it in (indx). (dfd) is effectively closed by 1761 * this operation. 1762 * 1763 * Any other error code is just returned. 1764 */ 1765 switch (error) { 1766 case ENODEV: 1767 /* 1768 * Check that the mode the file is being opened for is a 1769 * subset of the mode of the existing descriptor. 1770 */ 1771 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) 1772 return (EACCES); 1773 fp = fdp->fd_files[indx].fp; 1774 #if 0 1775 if (fp && fdp->fd_files[indx].fileflags & UF_MAPPED) 1776 (void) munmapfd(p, indx); 1777 #endif 1778 fdp->fd_files[indx].fp = wfp; 1779 fdp->fd_files[indx].fileflags = fdp->fd_files[dfd].fileflags; 1780 fhold(wfp); 1781 if (indx > fdp->fd_lastfile) 1782 fdp->fd_lastfile = indx; 1783 /* 1784 * we now own the reference to fp that the ofiles[] array 1785 * used to own. Release it. 1786 */ 1787 if (fp) 1788 fdrop(fp, curthread); 1789 return (0); 1790 1791 case ENXIO: 1792 /* 1793 * Steal away the file pointer from dfd, and stuff it into indx. 1794 */ 1795 fp = fdp->fd_files[indx].fp; 1796 #if 0 1797 if (fp && fdp->fd_files[indx].fileflags & UF_MAPPED) 1798 (void) munmapfd(p, indx); 1799 #endif 1800 fdp->fd_files[indx].fp = fdp->fd_files[dfd].fp; 1801 fdp->fd_files[indx].fileflags = fdp->fd_files[dfd].fileflags; 1802 funsetfd(fdp, dfd); 1803 1804 /* 1805 * we now own the reference to fp that the files[] array 1806 * used to own. Release it. 1807 */ 1808 if (fp) 1809 fdrop(fp, curthread); 1810 /* 1811 * Complete the clean up of the filedesc structure by 1812 * recomputing the various hints. 1813 */ 1814 if (indx > fdp->fd_lastfile) { 1815 fdp->fd_lastfile = indx; 1816 } else { 1817 while (fdp->fd_lastfile > 0 && 1818 fdp->fd_files[fdp->fd_lastfile].fp == NULL) { 1819 fdp->fd_lastfile--; 1820 } 1821 } 1822 return (0); 1823 1824 default: 1825 return (error); 1826 } 1827 /* NOTREACHED */ 1828 } 1829 1830 1831 struct filedesc_to_leader * 1832 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 1833 struct proc *leader) 1834 { 1835 struct filedesc_to_leader *fdtol; 1836 1837 fdtol = malloc(sizeof(struct filedesc_to_leader), 1838 M_FILEDESC_TO_LEADER, M_WAITOK); 1839 fdtol->fdl_refcount = 1; 1840 fdtol->fdl_holdcount = 0; 1841 fdtol->fdl_wakeup = 0; 1842 fdtol->fdl_leader = leader; 1843 if (old != NULL) { 1844 fdtol->fdl_next = old->fdl_next; 1845 fdtol->fdl_prev = old; 1846 old->fdl_next = fdtol; 1847 fdtol->fdl_next->fdl_prev = fdtol; 1848 } else { 1849 fdtol->fdl_next = fdtol; 1850 fdtol->fdl_prev = fdtol; 1851 } 1852 return fdtol; 1853 } 1854 1855 /* 1856 * Get file structures. 1857 */ 1858 static int 1859 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 1860 { 1861 struct kinfo_file kf; 1862 struct filedesc *fdp; 1863 struct file *fp; 1864 struct proc *p; 1865 uid_t uid; 1866 int count; 1867 int error; 1868 int n; 1869 1870 /* 1871 * Note: because the number of file descriptors is calculated 1872 * in different ways for sizing vs returning the data, 1873 * there is information leakage from the first loop. However, 1874 * it is of a similar order of magnitude to the leakage from 1875 * global system statistics such as kern.openfiles. 1876 * 1877 * When just doing a count, note that we cannot just count 1878 * the elements and add f_count via the filehead list because 1879 * threaded processes share their descriptor table and f_count might 1880 * still be '1' in that case. 1881 * 1882 * Since the SYSCTL op can block, we must hold the process to 1883 * prevent it being ripped out from under us either in the 1884 * file descriptor loop or in the greater LIST_FOREACH. The 1885 * process may be in varying states of disrepair. If the process 1886 * is in SZOMB we may have caught it just as it is being removed 1887 * from the allproc list, we must skip it in that case to maintain 1888 * an unbroken chain through the allproc list. 1889 */ 1890 count = 0; 1891 error = 0; 1892 LIST_FOREACH(p, &allproc, p_list) { 1893 if (p->p_stat == SIDL || (p->p_flag & P_ZOMBIE)) 1894 continue; 1895 if (!PRISON_CHECK(req->td->td_proc->p_ucred, p->p_ucred) != 0) 1896 continue; 1897 if ((fdp = p->p_fd) == NULL) 1898 continue; 1899 PHOLD(p); 1900 for (n = 0; n < fdp->fd_nfiles; ++n) { 1901 if ((fp = fdp->fd_files[n].fp) == NULL) 1902 continue; 1903 if (req->oldptr == NULL) { 1904 ++count; 1905 } else { 1906 uid = p->p_ucred ? p->p_ucred->cr_uid : -1; 1907 kcore_make_file(&kf, fp, p->p_pid, uid, n); 1908 error = SYSCTL_OUT(req, &kf, sizeof(kf)); 1909 if (error) 1910 break; 1911 } 1912 } 1913 PRELE(p); 1914 if (error) 1915 break; 1916 } 1917 1918 /* 1919 * When just calculating the size, overestimate a bit to try to 1920 * prevent system activity from causing the buffer-fill call 1921 * to fail later on. 1922 */ 1923 if (req->oldptr == NULL) { 1924 count = (count + 16) + (count / 10); 1925 error = SYSCTL_OUT(req, NULL, count * sizeof(kf)); 1926 } 1927 return (error); 1928 } 1929 1930 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 1931 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 1932 1933 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 1934 &maxfilesperproc, 0, "Maximum files allowed open per process"); 1935 1936 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 1937 &maxfiles, 0, "Maximum number of files"); 1938 1939 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, 1940 &maxfilesrootres, 0, "Descriptors reserved for root use"); 1941 1942 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 1943 &nfiles, 0, "System-wide number of open files"); 1944 1945 static void 1946 fildesc_drvinit(void *unused) 1947 { 1948 int fd; 1949 1950 cdevsw_add(&fildesc_cdevsw, 0, 0); 1951 for (fd = 0; fd < NUMFDESC; fd++) { 1952 make_dev(&fildesc_cdevsw, fd, 1953 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 1954 } 1955 make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 1956 make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 1957 make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 1958 } 1959 1960 struct fileops badfileops = { 1961 NULL, /* port */ 1962 NULL, /* clone */ 1963 badfo_readwrite, 1964 badfo_readwrite, 1965 badfo_ioctl, 1966 badfo_poll, 1967 badfo_kqfilter, 1968 badfo_stat, 1969 badfo_close, 1970 badfo_shutdown 1971 }; 1972 1973 static int 1974 badfo_readwrite( 1975 struct file *fp, 1976 struct uio *uio, 1977 struct ucred *cred, 1978 int flags, 1979 struct thread *td 1980 ) { 1981 return (EBADF); 1982 } 1983 1984 static int 1985 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td) 1986 { 1987 return (EBADF); 1988 } 1989 1990 static int 1991 badfo_poll(struct file *fp, int events, struct ucred *cred, struct thread *td) 1992 { 1993 return (0); 1994 } 1995 1996 static int 1997 badfo_kqfilter(struct file *fp, struct knote *kn) 1998 { 1999 return (0); 2000 } 2001 2002 static int 2003 badfo_stat(struct file *fp, struct stat *sb, struct thread *td) 2004 { 2005 return (EBADF); 2006 } 2007 2008 static int 2009 badfo_close(struct file *fp, struct thread *td) 2010 { 2011 return (EBADF); 2012 } 2013 2014 static int 2015 badfo_shutdown(struct file *fp, int how, struct thread *td) 2016 { 2017 return (EBADF); 2018 } 2019 2020 int 2021 nofo_shutdown(struct file *fp, int how, struct thread *td) 2022 { 2023 return (EOPNOTSUPP); 2024 } 2025 2026 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, 2027 fildesc_drvinit,NULL) 2028