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. Neither the name of the University nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 * 67 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 68 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $ 69 */ 70 71 #include "opt_compat.h" 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/malloc.h> 75 #include <sys/sysproto.h> 76 #include <sys/conf.h> 77 #include <sys/device.h> 78 #include <sys/file.h> 79 #include <sys/filedesc.h> 80 #include <sys/kernel.h> 81 #include <sys/sysctl.h> 82 #include <sys/vnode.h> 83 #include <sys/proc.h> 84 #include <sys/nlookup.h> 85 #include <sys/stat.h> 86 #include <sys/filio.h> 87 #include <sys/fcntl.h> 88 #include <sys/unistd.h> 89 #include <sys/resourcevar.h> 90 #include <sys/event.h> 91 #include <sys/kern_syscall.h> 92 #include <sys/kcore.h> 93 #include <sys/kinfo.h> 94 #include <sys/un.h> 95 96 #include <vm/vm.h> 97 #include <vm/vm_extern.h> 98 99 #include <sys/thread2.h> 100 #include <sys/file2.h> 101 #include <sys/spinlock2.h> 102 103 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd); 104 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr); 105 static struct file *funsetfd_locked (struct filedesc *fdp, int fd); 106 static void ffree(struct file *fp); 107 108 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 109 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 110 "file desc to leader structures"); 111 MALLOC_DEFINE(M_FILE, "file", "Open file structure"); 112 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 113 114 static struct krate krate_uidinfo = { .freq = 1 }; 115 116 static d_open_t fdopen; 117 #define NUMFDESC 64 118 119 #define CDEV_MAJOR 22 120 static struct dev_ops fildesc_ops = { 121 { "FD", 0, 0 }, 122 .d_open = fdopen, 123 }; 124 125 /* 126 * Descriptor management. 127 */ 128 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead); 129 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin); 130 static int nfiles; /* actual number of open files */ 131 extern int cmask; 132 133 /* 134 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared. 135 * 136 * MPSAFE - must be called with fdp->fd_spin exclusively held 137 */ 138 static __inline 139 void 140 fdfixup_locked(struct filedesc *fdp, int fd) 141 { 142 if (fd < fdp->fd_freefile) { 143 fdp->fd_freefile = fd; 144 } 145 while (fdp->fd_lastfile >= 0 && 146 fdp->fd_files[fdp->fd_lastfile].fp == NULL && 147 fdp->fd_files[fdp->fd_lastfile].reserved == 0 148 ) { 149 --fdp->fd_lastfile; 150 } 151 } 152 153 /* 154 * System calls on descriptors. 155 * 156 * MPSAFE 157 */ 158 int 159 sys_getdtablesize(struct getdtablesize_args *uap) 160 { 161 struct proc *p = curproc; 162 struct plimit *limit = p->p_limit; 163 int dtsize; 164 165 spin_lock(&limit->p_spin); 166 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 167 dtsize = INT_MAX; 168 else 169 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur; 170 spin_unlock(&limit->p_spin); 171 172 if (dtsize > maxfilesperproc) 173 dtsize = maxfilesperproc; 174 if (dtsize < minfilesperproc) 175 dtsize = minfilesperproc; 176 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser) 177 dtsize = maxfilesperuser; 178 uap->sysmsg_result = dtsize; 179 return (0); 180 } 181 182 /* 183 * Duplicate a file descriptor to a particular value. 184 * 185 * note: keep in mind that a potential race condition exists when closing 186 * descriptors from a shared descriptor table (via rfork). 187 * 188 * MPSAFE 189 */ 190 int 191 sys_dup2(struct dup2_args *uap) 192 { 193 int error; 194 int fd = 0; 195 196 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd); 197 uap->sysmsg_fds[0] = fd; 198 199 return (error); 200 } 201 202 /* 203 * Duplicate a file descriptor. 204 * 205 * MPSAFE 206 */ 207 int 208 sys_dup(struct dup_args *uap) 209 { 210 int error; 211 int fd = 0; 212 213 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd); 214 uap->sysmsg_fds[0] = fd; 215 216 return (error); 217 } 218 219 /* 220 * MPALMOSTSAFE - acquires mplock for fp operations 221 */ 222 int 223 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred) 224 { 225 struct thread *td = curthread; 226 struct proc *p = td->td_proc; 227 struct file *fp; 228 struct vnode *vp; 229 u_int newmin; 230 u_int oflags; 231 u_int nflags; 232 int tmp, error, flg = F_POSIX; 233 234 KKASSERT(p); 235 236 /* 237 * Operations on file descriptors that do not require a file pointer. 238 */ 239 switch (cmd) { 240 case F_GETFD: 241 error = fgetfdflags(p->p_fd, fd, &tmp); 242 if (error == 0) 243 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0; 244 return (error); 245 246 case F_SETFD: 247 if (dat->fc_cloexec & FD_CLOEXEC) 248 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE); 249 else 250 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE); 251 return (error); 252 case F_DUPFD: 253 newmin = dat->fc_fd; 254 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd); 255 return (error); 256 default: 257 break; 258 } 259 260 /* 261 * Operations on file pointers 262 */ 263 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 264 return (EBADF); 265 266 switch (cmd) { 267 case F_GETFL: 268 dat->fc_flags = OFLAGS(fp->f_flag); 269 error = 0; 270 break; 271 272 case F_SETFL: 273 oflags = fp->f_flag; 274 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 275 nflags |= oflags & ~FCNTLFLAGS; 276 277 error = 0; 278 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY)) 279 error = EINVAL; 280 if (error == 0 && ((nflags ^ oflags) & FASYNC)) { 281 tmp = nflags & FASYNC; 282 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, 283 cred, NULL); 284 } 285 if (error == 0) 286 fp->f_flag = nflags; 287 break; 288 289 case F_GETOWN: 290 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, 291 cred, NULL); 292 break; 293 294 case F_SETOWN: 295 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, 296 cred, NULL); 297 break; 298 299 case F_SETLKW: 300 flg |= F_WAIT; 301 /* Fall into F_SETLK */ 302 303 case F_SETLK: 304 if (fp->f_type != DTYPE_VNODE) { 305 error = EBADF; 306 break; 307 } 308 vp = (struct vnode *)fp->f_data; 309 310 /* 311 * copyin/lockop may block 312 */ 313 if (dat->fc_flock.l_whence == SEEK_CUR) 314 dat->fc_flock.l_start += fp->f_offset; 315 316 switch (dat->fc_flock.l_type) { 317 case F_RDLCK: 318 if ((fp->f_flag & FREAD) == 0) { 319 error = EBADF; 320 break; 321 } 322 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) { 323 lwkt_gettoken(&p->p_leader->p_token); 324 p->p_leader->p_flags |= P_ADVLOCK; 325 lwkt_reltoken(&p->p_leader->p_token); 326 } 327 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 328 &dat->fc_flock, flg); 329 break; 330 case F_WRLCK: 331 if ((fp->f_flag & FWRITE) == 0) { 332 error = EBADF; 333 break; 334 } 335 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) { 336 lwkt_gettoken(&p->p_leader->p_token); 337 p->p_leader->p_flags |= P_ADVLOCK; 338 lwkt_reltoken(&p->p_leader->p_token); 339 } 340 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 341 &dat->fc_flock, flg); 342 break; 343 case F_UNLCK: 344 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 345 &dat->fc_flock, F_POSIX); 346 break; 347 default: 348 error = EINVAL; 349 break; 350 } 351 352 /* 353 * It is possible to race a close() on the descriptor while 354 * we were blocked getting the lock. If this occurs the 355 * close might not have caught the lock. 356 */ 357 if (checkfdclosed(p->p_fd, fd, fp)) { 358 dat->fc_flock.l_whence = SEEK_SET; 359 dat->fc_flock.l_start = 0; 360 dat->fc_flock.l_len = 0; 361 dat->fc_flock.l_type = F_UNLCK; 362 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 363 F_UNLCK, &dat->fc_flock, F_POSIX); 364 } 365 break; 366 367 case F_GETLK: 368 if (fp->f_type != DTYPE_VNODE) { 369 error = EBADF; 370 break; 371 } 372 vp = (struct vnode *)fp->f_data; 373 /* 374 * copyin/lockop may block 375 */ 376 if (dat->fc_flock.l_type != F_RDLCK && 377 dat->fc_flock.l_type != F_WRLCK && 378 dat->fc_flock.l_type != F_UNLCK) { 379 error = EINVAL; 380 break; 381 } 382 if (dat->fc_flock.l_whence == SEEK_CUR) 383 dat->fc_flock.l_start += fp->f_offset; 384 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 385 &dat->fc_flock, F_POSIX); 386 break; 387 default: 388 error = EINVAL; 389 break; 390 } 391 392 fdrop(fp); 393 return (error); 394 } 395 396 /* 397 * The file control system call. 398 * 399 * MPSAFE 400 */ 401 int 402 sys_fcntl(struct fcntl_args *uap) 403 { 404 union fcntl_dat dat; 405 int error; 406 407 switch (uap->cmd) { 408 case F_DUPFD: 409 dat.fc_fd = uap->arg; 410 break; 411 case F_SETFD: 412 dat.fc_cloexec = uap->arg; 413 break; 414 case F_SETFL: 415 dat.fc_flags = uap->arg; 416 break; 417 case F_SETOWN: 418 dat.fc_owner = uap->arg; 419 break; 420 case F_SETLKW: 421 case F_SETLK: 422 case F_GETLK: 423 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 424 sizeof(struct flock)); 425 if (error) 426 return (error); 427 break; 428 } 429 430 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred); 431 432 if (error == 0) { 433 switch (uap->cmd) { 434 case F_DUPFD: 435 uap->sysmsg_result = dat.fc_fd; 436 break; 437 case F_GETFD: 438 uap->sysmsg_result = dat.fc_cloexec; 439 break; 440 case F_GETFL: 441 uap->sysmsg_result = dat.fc_flags; 442 break; 443 case F_GETOWN: 444 uap->sysmsg_result = dat.fc_owner; 445 case F_GETLK: 446 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 447 sizeof(struct flock)); 448 break; 449 } 450 } 451 452 return (error); 453 } 454 455 /* 456 * Common code for dup, dup2, and fcntl(F_DUPFD). 457 * 458 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells 459 * kern_dup() to destructively dup over an existing file descriptor if new 460 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest 461 * unused file descriptor that is greater than or equal to new. 462 * 463 * MPSAFE 464 */ 465 int 466 kern_dup(enum dup_type type, int old, int new, int *res) 467 { 468 struct thread *td = curthread; 469 struct proc *p = td->td_proc; 470 struct filedesc *fdp = p->p_fd; 471 struct file *fp; 472 struct file *delfp; 473 int oldflags; 474 int holdleaders; 475 int dtsize; 476 int error, newfd; 477 478 /* 479 * Verify that we have a valid descriptor to dup from and 480 * possibly to dup to. 481 * 482 * NOTE: maxfilesperuser is not applicable to dup() 483 */ 484 retry: 485 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 486 dtsize = INT_MAX; 487 else 488 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 489 if (dtsize > maxfilesperproc) 490 dtsize = maxfilesperproc; 491 if (dtsize < minfilesperproc) 492 dtsize = minfilesperproc; 493 494 if (new < 0 || new > dtsize) 495 return (EINVAL); 496 497 spin_lock(&fdp->fd_spin); 498 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) { 499 spin_unlock(&fdp->fd_spin); 500 return (EBADF); 501 } 502 if (type == DUP_FIXED && old == new) { 503 *res = new; 504 spin_unlock(&fdp->fd_spin); 505 return (0); 506 } 507 fp = fdp->fd_files[old].fp; 508 oldflags = fdp->fd_files[old].fileflags; 509 fhold(fp); /* MPSAFE - can be called with a spinlock held */ 510 511 /* 512 * Allocate a new descriptor if DUP_VARIABLE, or expand the table 513 * if the requested descriptor is beyond the current table size. 514 * 515 * This can block. Retry if the source descriptor no longer matches 516 * or if our expectation in the expansion case races. 517 * 518 * If we are not expanding or allocating a new decriptor, then reset 519 * the target descriptor to a reserved state so we have a uniform 520 * setup for the next code block. 521 */ 522 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) { 523 spin_unlock(&fdp->fd_spin); 524 error = fdalloc(p, new, &newfd); 525 spin_lock(&fdp->fd_spin); 526 if (error) { 527 spin_unlock(&fdp->fd_spin); 528 fdrop(fp); 529 return (error); 530 } 531 /* 532 * Check for ripout 533 */ 534 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) { 535 fsetfd_locked(fdp, NULL, newfd); 536 spin_unlock(&fdp->fd_spin); 537 fdrop(fp); 538 goto retry; 539 } 540 /* 541 * Check for expansion race 542 */ 543 if (type != DUP_VARIABLE && new != newfd) { 544 fsetfd_locked(fdp, NULL, newfd); 545 spin_unlock(&fdp->fd_spin); 546 fdrop(fp); 547 goto retry; 548 } 549 /* 550 * Check for ripout, newfd reused old (this case probably 551 * can't occur). 552 */ 553 if (old == newfd) { 554 fsetfd_locked(fdp, NULL, newfd); 555 spin_unlock(&fdp->fd_spin); 556 fdrop(fp); 557 goto retry; 558 } 559 new = newfd; 560 delfp = NULL; 561 } else { 562 if (fdp->fd_files[new].reserved) { 563 spin_unlock(&fdp->fd_spin); 564 fdrop(fp); 565 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new); 566 tsleep(fdp, 0, "fdres", hz); 567 goto retry; 568 } 569 570 /* 571 * If the target descriptor was never allocated we have 572 * to allocate it. If it was we have to clean out the 573 * old descriptor. delfp inherits the ref from the 574 * descriptor table. 575 */ 576 delfp = fdp->fd_files[new].fp; 577 fdp->fd_files[new].fp = NULL; 578 fdp->fd_files[new].reserved = 1; 579 if (delfp == NULL) { 580 fdreserve_locked(fdp, new, 1); 581 if (new > fdp->fd_lastfile) 582 fdp->fd_lastfile = new; 583 } 584 585 } 586 587 /* 588 * NOTE: still holding an exclusive spinlock 589 */ 590 591 /* 592 * If a descriptor is being overwritten we may hve to tell 593 * fdfree() to sleep to ensure that all relevant process 594 * leaders can be traversed in closef(). 595 */ 596 if (delfp != NULL && p->p_fdtol != NULL) { 597 fdp->fd_holdleaderscount++; 598 holdleaders = 1; 599 } else { 600 holdleaders = 0; 601 } 602 KASSERT(delfp == NULL || type == DUP_FIXED, 603 ("dup() picked an open file")); 604 605 /* 606 * Duplicate the source descriptor, update lastfile. If the new 607 * descriptor was not allocated and we aren't replacing an existing 608 * descriptor we have to mark the descriptor as being in use. 609 * 610 * The fd_files[] array inherits fp's hold reference. 611 */ 612 fsetfd_locked(fdp, fp, new); 613 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE; 614 spin_unlock(&fdp->fd_spin); 615 fdrop(fp); 616 *res = new; 617 618 /* 619 * If we dup'd over a valid file, we now own the reference to it 620 * and must dispose of it using closef() semantics (as if a 621 * close() were performed on it). 622 */ 623 if (delfp) { 624 if (SLIST_FIRST(&delfp->f_klist)) 625 knote_fdclose(delfp, fdp, new); 626 closef(delfp, p); 627 if (holdleaders) { 628 spin_lock(&fdp->fd_spin); 629 fdp->fd_holdleaderscount--; 630 if (fdp->fd_holdleaderscount == 0 && 631 fdp->fd_holdleaderswakeup != 0) { 632 fdp->fd_holdleaderswakeup = 0; 633 spin_unlock(&fdp->fd_spin); 634 wakeup(&fdp->fd_holdleaderscount); 635 } else { 636 spin_unlock(&fdp->fd_spin); 637 } 638 } 639 } 640 return (0); 641 } 642 643 /* 644 * If sigio is on the list associated with a process or process group, 645 * disable signalling from the device, remove sigio from the list and 646 * free sigio. 647 * 648 * MPSAFE 649 */ 650 void 651 funsetown(struct sigio **sigiop) 652 { 653 struct pgrp *pgrp; 654 struct proc *p; 655 struct sigio *sigio; 656 657 if ((sigio = *sigiop) != NULL) { 658 lwkt_gettoken(&proc_token); /* protect sigio */ 659 KKASSERT(sigiop == sigio->sio_myref); 660 sigio = *sigiop; 661 *sigiop = NULL; 662 lwkt_reltoken(&proc_token); 663 } 664 if (sigio == NULL) 665 return; 666 667 if (sigio->sio_pgid < 0) { 668 pgrp = sigio->sio_pgrp; 669 sigio->sio_pgrp = NULL; 670 lwkt_gettoken(&pgrp->pg_token); 671 SLIST_REMOVE(&pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio); 672 lwkt_reltoken(&pgrp->pg_token); 673 pgrel(pgrp); 674 } else /* if ((*sigiop)->sio_pgid > 0) */ { 675 p = sigio->sio_proc; 676 sigio->sio_proc = NULL; 677 PHOLD(p); 678 lwkt_gettoken(&p->p_token); 679 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); 680 lwkt_reltoken(&p->p_token); 681 PRELE(p); 682 } 683 crfree(sigio->sio_ucred); 684 sigio->sio_ucred = NULL; 685 kfree(sigio, M_SIGIO); 686 } 687 688 /* 689 * Free a list of sigio structures. Caller is responsible for ensuring 690 * that the list is MPSAFE. 691 * 692 * MPSAFE 693 */ 694 void 695 funsetownlst(struct sigiolst *sigiolst) 696 { 697 struct sigio *sigio; 698 699 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) 700 funsetown(sigio->sio_myref); 701 } 702 703 /* 704 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 705 * 706 * After permission checking, add a sigio structure to the sigio list for 707 * the process or process group. 708 * 709 * MPSAFE 710 */ 711 int 712 fsetown(pid_t pgid, struct sigio **sigiop) 713 { 714 struct proc *proc = NULL; 715 struct pgrp *pgrp = NULL; 716 struct sigio *sigio; 717 int error; 718 719 if (pgid == 0) { 720 funsetown(sigiop); 721 return (0); 722 } 723 724 if (pgid > 0) { 725 proc = pfind(pgid); 726 if (proc == NULL) { 727 error = ESRCH; 728 goto done; 729 } 730 731 /* 732 * Policy - Don't allow a process to FSETOWN a process 733 * in another session. 734 * 735 * Remove this test to allow maximum flexibility or 736 * restrict FSETOWN to the current process or process 737 * group for maximum safety. 738 */ 739 if (proc->p_session != curproc->p_session) { 740 error = EPERM; 741 goto done; 742 } 743 } else /* if (pgid < 0) */ { 744 pgrp = pgfind(-pgid); 745 if (pgrp == NULL) { 746 error = ESRCH; 747 goto done; 748 } 749 750 /* 751 * Policy - Don't allow a process to FSETOWN a process 752 * in another session. 753 * 754 * Remove this test to allow maximum flexibility or 755 * restrict FSETOWN to the current process or process 756 * group for maximum safety. 757 */ 758 if (pgrp->pg_session != curproc->p_session) { 759 error = EPERM; 760 goto done; 761 } 762 } 763 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK | M_ZERO); 764 if (pgid > 0) { 765 KKASSERT(pgrp == NULL); 766 lwkt_gettoken(&proc->p_token); 767 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 768 sigio->sio_proc = proc; 769 lwkt_reltoken(&proc->p_token); 770 } else { 771 KKASSERT(proc == NULL); 772 lwkt_gettoken(&pgrp->pg_token); 773 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 774 sigio->sio_pgrp = pgrp; 775 lwkt_reltoken(&pgrp->pg_token); 776 pgrp = NULL; 777 } 778 sigio->sio_pgid = pgid; 779 sigio->sio_ucred = crhold(curthread->td_ucred); 780 /* It would be convenient if p_ruid was in ucred. */ 781 sigio->sio_ruid = sigio->sio_ucred->cr_ruid; 782 sigio->sio_myref = sigiop; 783 784 lwkt_gettoken(&proc_token); 785 while (*sigiop) 786 funsetown(sigiop); 787 *sigiop = sigio; 788 lwkt_reltoken(&proc_token); 789 error = 0; 790 done: 791 if (pgrp) 792 pgrel(pgrp); 793 if (proc) 794 PRELE(proc); 795 return (error); 796 } 797 798 /* 799 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 800 * 801 * MPSAFE 802 */ 803 pid_t 804 fgetown(struct sigio **sigiop) 805 { 806 struct sigio *sigio; 807 pid_t own; 808 809 lwkt_gettoken(&proc_token); 810 sigio = *sigiop; 811 own = (sigio != NULL ? sigio->sio_pgid : 0); 812 lwkt_reltoken(&proc_token); 813 814 return (own); 815 } 816 817 /* 818 * Close many file descriptors. 819 * 820 * MPSAFE 821 */ 822 int 823 sys_closefrom(struct closefrom_args *uap) 824 { 825 return(kern_closefrom(uap->fd)); 826 } 827 828 /* 829 * Close all file descriptors greater then or equal to fd 830 * 831 * MPSAFE 832 */ 833 int 834 kern_closefrom(int fd) 835 { 836 struct thread *td = curthread; 837 struct proc *p = td->td_proc; 838 struct filedesc *fdp; 839 840 KKASSERT(p); 841 fdp = p->p_fd; 842 843 if (fd < 0) 844 return (EINVAL); 845 846 /* 847 * NOTE: This function will skip unassociated descriptors and 848 * reserved descriptors that have not yet been assigned. 849 * fd_lastfile can change as a side effect of kern_close(). 850 */ 851 spin_lock(&fdp->fd_spin); 852 while (fd <= fdp->fd_lastfile) { 853 if (fdp->fd_files[fd].fp != NULL) { 854 spin_unlock(&fdp->fd_spin); 855 /* ok if this races another close */ 856 if (kern_close(fd) == EINTR) 857 return (EINTR); 858 spin_lock(&fdp->fd_spin); 859 } 860 ++fd; 861 } 862 spin_unlock(&fdp->fd_spin); 863 return (0); 864 } 865 866 /* 867 * Close a file descriptor. 868 * 869 * MPSAFE 870 */ 871 int 872 sys_close(struct close_args *uap) 873 { 874 return(kern_close(uap->fd)); 875 } 876 877 /* 878 * MPSAFE 879 */ 880 int 881 kern_close(int fd) 882 { 883 struct thread *td = curthread; 884 struct proc *p = td->td_proc; 885 struct filedesc *fdp; 886 struct file *fp; 887 int error; 888 int holdleaders; 889 890 KKASSERT(p); 891 fdp = p->p_fd; 892 893 spin_lock(&fdp->fd_spin); 894 if ((fp = funsetfd_locked(fdp, fd)) == NULL) { 895 spin_unlock(&fdp->fd_spin); 896 return (EBADF); 897 } 898 holdleaders = 0; 899 if (p->p_fdtol != NULL) { 900 /* 901 * Ask fdfree() to sleep to ensure that all relevant 902 * process leaders can be traversed in closef(). 903 */ 904 fdp->fd_holdleaderscount++; 905 holdleaders = 1; 906 } 907 908 /* 909 * we now hold the fp reference that used to be owned by the descriptor 910 * array. 911 */ 912 spin_unlock(&fdp->fd_spin); 913 if (SLIST_FIRST(&fp->f_klist)) 914 knote_fdclose(fp, fdp, fd); 915 error = closef(fp, p); 916 if (holdleaders) { 917 spin_lock(&fdp->fd_spin); 918 fdp->fd_holdleaderscount--; 919 if (fdp->fd_holdleaderscount == 0 && 920 fdp->fd_holdleaderswakeup != 0) { 921 fdp->fd_holdleaderswakeup = 0; 922 spin_unlock(&fdp->fd_spin); 923 wakeup(&fdp->fd_holdleaderscount); 924 } else { 925 spin_unlock(&fdp->fd_spin); 926 } 927 } 928 return (error); 929 } 930 931 /* 932 * shutdown_args(int fd, int how) 933 */ 934 int 935 kern_shutdown(int fd, int how) 936 { 937 struct thread *td = curthread; 938 struct proc *p = td->td_proc; 939 struct file *fp; 940 int error; 941 942 KKASSERT(p); 943 944 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 945 return (EBADF); 946 error = fo_shutdown(fp, how); 947 fdrop(fp); 948 949 return (error); 950 } 951 952 /* 953 * MPALMOSTSAFE 954 */ 955 int 956 sys_shutdown(struct shutdown_args *uap) 957 { 958 int error; 959 960 error = kern_shutdown(uap->s, uap->how); 961 962 return (error); 963 } 964 965 /* 966 * MPSAFE 967 */ 968 int 969 kern_fstat(int fd, struct stat *ub) 970 { 971 struct thread *td = curthread; 972 struct proc *p = td->td_proc; 973 struct file *fp; 974 int error; 975 976 KKASSERT(p); 977 978 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 979 return (EBADF); 980 error = fo_stat(fp, ub, td->td_ucred); 981 fdrop(fp); 982 983 return (error); 984 } 985 986 /* 987 * Return status information about a file descriptor. 988 * 989 * MPSAFE 990 */ 991 int 992 sys_fstat(struct fstat_args *uap) 993 { 994 struct stat st; 995 int error; 996 997 error = kern_fstat(uap->fd, &st); 998 999 if (error == 0) 1000 error = copyout(&st, uap->sb, sizeof(st)); 1001 return (error); 1002 } 1003 1004 /* 1005 * Return pathconf information about a file descriptor. 1006 * 1007 * MPALMOSTSAFE 1008 */ 1009 int 1010 sys_fpathconf(struct fpathconf_args *uap) 1011 { 1012 struct thread *td = curthread; 1013 struct proc *p = td->td_proc; 1014 struct file *fp; 1015 struct vnode *vp; 1016 int error = 0; 1017 1018 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 1019 return (EBADF); 1020 1021 switch (fp->f_type) { 1022 case DTYPE_PIPE: 1023 case DTYPE_SOCKET: 1024 if (uap->name != _PC_PIPE_BUF) { 1025 error = EINVAL; 1026 } else { 1027 uap->sysmsg_result = PIPE_BUF; 1028 error = 0; 1029 } 1030 break; 1031 case DTYPE_FIFO: 1032 case DTYPE_VNODE: 1033 vp = (struct vnode *)fp->f_data; 1034 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg); 1035 break; 1036 default: 1037 error = EOPNOTSUPP; 1038 break; 1039 } 1040 fdrop(fp); 1041 return(error); 1042 } 1043 1044 static int fdexpand; 1045 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, 1046 "Number of times a file table has been expanded"); 1047 1048 /* 1049 * Grow the file table so it can hold through descriptor (want). 1050 * 1051 * The fdp's spinlock must be held exclusively on entry and may be held 1052 * exclusively on return. The spinlock may be cycled by the routine. 1053 * 1054 * MPSAFE 1055 */ 1056 static void 1057 fdgrow_locked(struct filedesc *fdp, int want) 1058 { 1059 struct fdnode *newfiles; 1060 struct fdnode *oldfiles; 1061 int nf, extra; 1062 1063 nf = fdp->fd_nfiles; 1064 do { 1065 /* nf has to be of the form 2^n - 1 */ 1066 nf = 2 * nf + 1; 1067 } while (nf <= want); 1068 1069 spin_unlock(&fdp->fd_spin); 1070 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK); 1071 spin_lock(&fdp->fd_spin); 1072 1073 /* 1074 * We could have raced another extend while we were not holding 1075 * the spinlock. 1076 */ 1077 if (fdp->fd_nfiles >= nf) { 1078 spin_unlock(&fdp->fd_spin); 1079 kfree(newfiles, M_FILEDESC); 1080 spin_lock(&fdp->fd_spin); 1081 return; 1082 } 1083 /* 1084 * Copy the existing ofile and ofileflags arrays 1085 * and zero the new portion of each array. 1086 */ 1087 extra = nf - fdp->fd_nfiles; 1088 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode)); 1089 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode)); 1090 1091 oldfiles = fdp->fd_files; 1092 fdp->fd_files = newfiles; 1093 fdp->fd_nfiles = nf; 1094 1095 if (oldfiles != fdp->fd_builtin_files) { 1096 spin_unlock(&fdp->fd_spin); 1097 kfree(oldfiles, M_FILEDESC); 1098 spin_lock(&fdp->fd_spin); 1099 } 1100 fdexpand++; 1101 } 1102 1103 /* 1104 * Number of nodes in right subtree, including the root. 1105 */ 1106 static __inline int 1107 right_subtree_size(int n) 1108 { 1109 return (n ^ (n | (n + 1))); 1110 } 1111 1112 /* 1113 * Bigger ancestor. 1114 */ 1115 static __inline int 1116 right_ancestor(int n) 1117 { 1118 return (n | (n + 1)); 1119 } 1120 1121 /* 1122 * Smaller ancestor. 1123 */ 1124 static __inline int 1125 left_ancestor(int n) 1126 { 1127 return ((n & (n + 1)) - 1); 1128 } 1129 1130 /* 1131 * Traverse the in-place binary tree buttom-up adjusting the allocation 1132 * count so scans can determine where free descriptors are located. 1133 * 1134 * MPSAFE - caller must be holding an exclusive spinlock on fdp 1135 */ 1136 static 1137 void 1138 fdreserve_locked(struct filedesc *fdp, int fd, int incr) 1139 { 1140 while (fd >= 0) { 1141 fdp->fd_files[fd].allocated += incr; 1142 KKASSERT(fdp->fd_files[fd].allocated >= 0); 1143 fd = left_ancestor(fd); 1144 } 1145 } 1146 1147 /* 1148 * Reserve a file descriptor for the process. If no error occurs, the 1149 * caller MUST at some point call fsetfd() or assign a file pointer 1150 * or dispose of the reservation. 1151 * 1152 * MPSAFE 1153 */ 1154 int 1155 fdalloc(struct proc *p, int want, int *result) 1156 { 1157 struct filedesc *fdp = p->p_fd; 1158 struct uidinfo *uip; 1159 int fd, rsize, rsum, node, lim; 1160 1161 /* 1162 * Check dtable size limit 1163 */ 1164 spin_lock(&p->p_limit->p_spin); 1165 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 1166 lim = INT_MAX; 1167 else 1168 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 1169 spin_unlock(&p->p_limit->p_spin); 1170 1171 if (lim > maxfilesperproc) 1172 lim = maxfilesperproc; 1173 if (lim < minfilesperproc) 1174 lim = minfilesperproc; 1175 if (want >= lim) 1176 return (EMFILE); 1177 1178 /* 1179 * Check that the user has not run out of descriptors (non-root only). 1180 * As a safety measure the dtable is allowed to have at least 1181 * minfilesperproc open fds regardless of the maxfilesperuser limit. 1182 */ 1183 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) { 1184 uip = p->p_ucred->cr_uidinfo; 1185 if (uip->ui_openfiles > maxfilesperuser) { 1186 krateprintf(&krate_uidinfo, 1187 "Warning: user %d pid %d (%s) ran out of " 1188 "file descriptors (%d/%d)\n", 1189 p->p_ucred->cr_uid, (int)p->p_pid, 1190 p->p_comm, 1191 uip->ui_openfiles, maxfilesperuser); 1192 return(ENFILE); 1193 } 1194 } 1195 1196 /* 1197 * Grow the dtable if necessary 1198 */ 1199 spin_lock(&fdp->fd_spin); 1200 if (want >= fdp->fd_nfiles) 1201 fdgrow_locked(fdp, want); 1202 1203 /* 1204 * Search for a free descriptor starting at the higher 1205 * of want or fd_freefile. If that fails, consider 1206 * expanding the ofile array. 1207 * 1208 * NOTE! the 'allocated' field is a cumulative recursive allocation 1209 * count. If we happen to see a value of 0 then we can shortcut 1210 * our search. Otherwise we run through through the tree going 1211 * down branches we know have free descriptor(s) until we hit a 1212 * leaf node. The leaf node will be free but will not necessarily 1213 * have an allocated field of 0. 1214 */ 1215 retry: 1216 /* move up the tree looking for a subtree with a free node */ 1217 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim); 1218 fd = right_ancestor(fd)) { 1219 if (fdp->fd_files[fd].allocated == 0) 1220 goto found; 1221 1222 rsize = right_subtree_size(fd); 1223 if (fdp->fd_files[fd].allocated == rsize) 1224 continue; /* right subtree full */ 1225 1226 /* 1227 * Free fd is in the right subtree of the tree rooted at fd. 1228 * Call that subtree R. Look for the smallest (leftmost) 1229 * subtree of R with an unallocated fd: continue moving 1230 * down the left branch until encountering a full left 1231 * subtree, then move to the right. 1232 */ 1233 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) { 1234 node = fd + rsize; 1235 rsum += fdp->fd_files[node].allocated; 1236 if (fdp->fd_files[fd].allocated == rsum + rsize) { 1237 fd = node; /* move to the right */ 1238 if (fdp->fd_files[node].allocated == 0) 1239 goto found; 1240 rsum = 0; 1241 } 1242 } 1243 goto found; 1244 } 1245 1246 /* 1247 * No space in current array. Expand? 1248 */ 1249 if (fdp->fd_nfiles >= lim) { 1250 spin_unlock(&fdp->fd_spin); 1251 return (EMFILE); 1252 } 1253 fdgrow_locked(fdp, want); 1254 goto retry; 1255 1256 found: 1257 KKASSERT(fd < fdp->fd_nfiles); 1258 if (fd > fdp->fd_lastfile) 1259 fdp->fd_lastfile = fd; 1260 if (want <= fdp->fd_freefile) 1261 fdp->fd_freefile = fd; 1262 *result = fd; 1263 KKASSERT(fdp->fd_files[fd].fp == NULL); 1264 KKASSERT(fdp->fd_files[fd].reserved == 0); 1265 fdp->fd_files[fd].fileflags = 0; 1266 fdp->fd_files[fd].reserved = 1; 1267 fdreserve_locked(fdp, fd, 1); 1268 spin_unlock(&fdp->fd_spin); 1269 return (0); 1270 } 1271 1272 /* 1273 * Check to see whether n user file descriptors 1274 * are available to the process p. 1275 * 1276 * MPSAFE 1277 */ 1278 int 1279 fdavail(struct proc *p, int n) 1280 { 1281 struct filedesc *fdp = p->p_fd; 1282 struct fdnode *fdnode; 1283 int i, lim, last; 1284 1285 spin_lock(&p->p_limit->p_spin); 1286 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 1287 lim = INT_MAX; 1288 else 1289 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 1290 spin_unlock(&p->p_limit->p_spin); 1291 1292 if (lim > maxfilesperproc) 1293 lim = maxfilesperproc; 1294 if (lim < minfilesperproc) 1295 lim = minfilesperproc; 1296 1297 spin_lock(&fdp->fd_spin); 1298 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) { 1299 spin_unlock(&fdp->fd_spin); 1300 return (1); 1301 } 1302 last = min(fdp->fd_nfiles, lim); 1303 fdnode = &fdp->fd_files[fdp->fd_freefile]; 1304 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) { 1305 if (fdnode->fp == NULL && --n <= 0) { 1306 spin_unlock(&fdp->fd_spin); 1307 return (1); 1308 } 1309 } 1310 spin_unlock(&fdp->fd_spin); 1311 return (0); 1312 } 1313 1314 /* 1315 * Revoke open descriptors referencing (f_data, f_type) 1316 * 1317 * Any revoke executed within a prison is only able to 1318 * revoke descriptors for processes within that prison. 1319 * 1320 * Returns 0 on success or an error code. 1321 */ 1322 struct fdrevoke_info { 1323 void *data; 1324 short type; 1325 short unused; 1326 int count; 1327 int intransit; 1328 struct ucred *cred; 1329 struct file *nfp; 1330 }; 1331 1332 static int fdrevoke_check_callback(struct file *fp, void *vinfo); 1333 static int fdrevoke_proc_callback(struct proc *p, void *vinfo); 1334 1335 int 1336 fdrevoke(void *f_data, short f_type, struct ucred *cred) 1337 { 1338 struct fdrevoke_info info; 1339 int error; 1340 1341 bzero(&info, sizeof(info)); 1342 info.data = f_data; 1343 info.type = f_type; 1344 info.cred = cred; 1345 error = falloc(NULL, &info.nfp, NULL); 1346 if (error) 1347 return (error); 1348 1349 /* 1350 * Scan the file pointer table once. dups do not dup file pointers, 1351 * only descriptors, so there is no leak. Set FREVOKED on the fps 1352 * being revoked. 1353 */ 1354 allfiles_scan_exclusive(fdrevoke_check_callback, &info); 1355 1356 /* 1357 * If any fps were marked track down the related descriptors 1358 * and close them. Any dup()s at this point will notice 1359 * the FREVOKED already set in the fp and do the right thing. 1360 * 1361 * Any fps with non-zero msgcounts (aka sent over a unix-domain 1362 * socket) bumped the intransit counter and will require a 1363 * scan. Races against fps leaving the socket are closed by 1364 * the socket code checking for FREVOKED. 1365 */ 1366 if (info.count) 1367 allproc_scan(fdrevoke_proc_callback, &info); 1368 if (info.intransit) 1369 unp_revoke_gc(info.nfp); 1370 fdrop(info.nfp); 1371 return(0); 1372 } 1373 1374 /* 1375 * Locate matching file pointers directly. 1376 * 1377 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls! 1378 */ 1379 static int 1380 fdrevoke_check_callback(struct file *fp, void *vinfo) 1381 { 1382 struct fdrevoke_info *info = vinfo; 1383 1384 /* 1385 * File pointers already flagged for revokation are skipped. 1386 */ 1387 if (fp->f_flag & FREVOKED) 1388 return(0); 1389 1390 /* 1391 * If revoking from a prison file pointers created outside of 1392 * that prison, or file pointers without creds, cannot be revoked. 1393 */ 1394 if (info->cred->cr_prison && 1395 (fp->f_cred == NULL || 1396 info->cred->cr_prison != fp->f_cred->cr_prison)) { 1397 return(0); 1398 } 1399 1400 /* 1401 * If the file pointer matches then mark it for revocation. The 1402 * flag is currently only used by unp_revoke_gc(). 1403 * 1404 * info->count is a heuristic and can race in a SMP environment. 1405 */ 1406 if (info->data == fp->f_data && info->type == fp->f_type) { 1407 atomic_set_int(&fp->f_flag, FREVOKED); 1408 info->count += fp->f_count; 1409 if (fp->f_msgcount) 1410 ++info->intransit; 1411 } 1412 return(0); 1413 } 1414 1415 /* 1416 * Locate matching file pointers via process descriptor tables. 1417 */ 1418 static int 1419 fdrevoke_proc_callback(struct proc *p, void *vinfo) 1420 { 1421 struct fdrevoke_info *info = vinfo; 1422 struct filedesc *fdp; 1423 struct file *fp; 1424 int n; 1425 1426 if (p->p_stat == SIDL || p->p_stat == SZOMB) 1427 return(0); 1428 if (info->cred->cr_prison && 1429 info->cred->cr_prison != p->p_ucred->cr_prison) { 1430 return(0); 1431 } 1432 1433 /* 1434 * If the controlling terminal of the process matches the 1435 * vnode being revoked we clear the controlling terminal. 1436 * 1437 * The normal spec_close() may not catch this because it 1438 * uses curproc instead of p. 1439 */ 1440 if (p->p_session && info->type == DTYPE_VNODE && 1441 info->data == p->p_session->s_ttyvp) { 1442 p->p_session->s_ttyvp = NULL; 1443 vrele(info->data); 1444 } 1445 1446 /* 1447 * Softref the fdp to prevent it from being destroyed 1448 */ 1449 spin_lock(&p->p_spin); 1450 if ((fdp = p->p_fd) == NULL) { 1451 spin_unlock(&p->p_spin); 1452 return(0); 1453 } 1454 atomic_add_int(&fdp->fd_softrefs, 1); 1455 spin_unlock(&p->p_spin); 1456 1457 /* 1458 * Locate and close any matching file descriptors. 1459 */ 1460 spin_lock(&fdp->fd_spin); 1461 for (n = 0; n < fdp->fd_nfiles; ++n) { 1462 if ((fp = fdp->fd_files[n].fp) == NULL) 1463 continue; 1464 if (fp->f_flag & FREVOKED) { 1465 fhold(info->nfp); 1466 fdp->fd_files[n].fp = info->nfp; 1467 spin_unlock(&fdp->fd_spin); 1468 knote_fdclose(fp, fdp, n); /* XXX */ 1469 closef(fp, p); 1470 spin_lock(&fdp->fd_spin); 1471 --info->count; 1472 } 1473 } 1474 spin_unlock(&fdp->fd_spin); 1475 atomic_subtract_int(&fdp->fd_softrefs, 1); 1476 return(0); 1477 } 1478 1479 /* 1480 * falloc: 1481 * Create a new open file structure and reserve a file decriptor 1482 * for the process that refers to it. 1483 * 1484 * Root creds are checked using lp, or assumed if lp is NULL. If 1485 * resultfd is non-NULL then lp must also be non-NULL. No file 1486 * descriptor is reserved (and no process context is needed) if 1487 * resultfd is NULL. 1488 * 1489 * A file pointer with a refcount of 1 is returned. Note that the 1490 * file pointer is NOT associated with the descriptor. If falloc 1491 * returns success, fsetfd() MUST be called to either associate the 1492 * file pointer or clear the reservation. 1493 * 1494 * MPSAFE 1495 */ 1496 int 1497 falloc(struct lwp *lp, struct file **resultfp, int *resultfd) 1498 { 1499 static struct timeval lastfail; 1500 static int curfail; 1501 struct file *fp; 1502 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred; 1503 int error; 1504 1505 fp = NULL; 1506 1507 /* 1508 * Handle filetable full issues and root overfill. 1509 */ 1510 if (nfiles >= maxfiles - maxfilesrootres && 1511 (cred->cr_ruid != 0 || nfiles >= maxfiles)) { 1512 if (ppsratecheck(&lastfail, &curfail, 1)) { 1513 kprintf("kern.maxfiles limit exceeded by uid %d, " 1514 "please see tuning(7).\n", 1515 cred->cr_ruid); 1516 } 1517 error = ENFILE; 1518 goto done; 1519 } 1520 1521 /* 1522 * Allocate a new file descriptor. 1523 */ 1524 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO); 1525 spin_init(&fp->f_spin); 1526 SLIST_INIT(&fp->f_klist); 1527 fp->f_count = 1; 1528 fp->f_ops = &badfileops; 1529 fp->f_seqcount = 1; 1530 fsetcred(fp, cred); 1531 spin_lock(&filehead_spin); 1532 nfiles++; 1533 LIST_INSERT_HEAD(&filehead, fp, f_list); 1534 spin_unlock(&filehead_spin); 1535 if (resultfd) { 1536 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) { 1537 fdrop(fp); 1538 fp = NULL; 1539 } 1540 } else { 1541 error = 0; 1542 } 1543 done: 1544 *resultfp = fp; 1545 return (error); 1546 } 1547 1548 /* 1549 * Check for races against a file descriptor by determining that the 1550 * file pointer is still associated with the specified file descriptor, 1551 * and a close is not currently in progress. 1552 * 1553 * MPSAFE 1554 */ 1555 int 1556 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp) 1557 { 1558 int error; 1559 1560 spin_lock_shared(&fdp->fd_spin); 1561 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp) 1562 error = EBADF; 1563 else 1564 error = 0; 1565 spin_unlock_shared(&fdp->fd_spin); 1566 return (error); 1567 } 1568 1569 /* 1570 * Associate a file pointer with a previously reserved file descriptor. 1571 * This function always succeeds. 1572 * 1573 * If fp is NULL, the file descriptor is returned to the pool. 1574 */ 1575 1576 /* 1577 * MPSAFE (exclusive spinlock must be held on call) 1578 */ 1579 static void 1580 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd) 1581 { 1582 KKASSERT((unsigned)fd < fdp->fd_nfiles); 1583 KKASSERT(fdp->fd_files[fd].reserved != 0); 1584 if (fp) { 1585 fhold(fp); 1586 fdp->fd_files[fd].fp = fp; 1587 fdp->fd_files[fd].reserved = 0; 1588 } else { 1589 fdp->fd_files[fd].reserved = 0; 1590 fdreserve_locked(fdp, fd, -1); 1591 fdfixup_locked(fdp, fd); 1592 } 1593 } 1594 1595 /* 1596 * MPSAFE 1597 */ 1598 void 1599 fsetfd(struct filedesc *fdp, struct file *fp, int fd) 1600 { 1601 spin_lock(&fdp->fd_spin); 1602 fsetfd_locked(fdp, fp, fd); 1603 spin_unlock(&fdp->fd_spin); 1604 } 1605 1606 /* 1607 * MPSAFE (exclusive spinlock must be held on call) 1608 */ 1609 static 1610 struct file * 1611 funsetfd_locked(struct filedesc *fdp, int fd) 1612 { 1613 struct file *fp; 1614 1615 if ((unsigned)fd >= fdp->fd_nfiles) 1616 return (NULL); 1617 if ((fp = fdp->fd_files[fd].fp) == NULL) 1618 return (NULL); 1619 fdp->fd_files[fd].fp = NULL; 1620 fdp->fd_files[fd].fileflags = 0; 1621 1622 fdreserve_locked(fdp, fd, -1); 1623 fdfixup_locked(fdp, fd); 1624 return(fp); 1625 } 1626 1627 /* 1628 * MPSAFE 1629 */ 1630 int 1631 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp) 1632 { 1633 int error; 1634 1635 spin_lock(&fdp->fd_spin); 1636 if (((u_int)fd) >= fdp->fd_nfiles) { 1637 error = EBADF; 1638 } else if (fdp->fd_files[fd].fp == NULL) { 1639 error = EBADF; 1640 } else { 1641 *flagsp = fdp->fd_files[fd].fileflags; 1642 error = 0; 1643 } 1644 spin_unlock(&fdp->fd_spin); 1645 return (error); 1646 } 1647 1648 /* 1649 * MPSAFE 1650 */ 1651 int 1652 fsetfdflags(struct filedesc *fdp, int fd, int add_flags) 1653 { 1654 int error; 1655 1656 spin_lock(&fdp->fd_spin); 1657 if (((u_int)fd) >= fdp->fd_nfiles) { 1658 error = EBADF; 1659 } else if (fdp->fd_files[fd].fp == NULL) { 1660 error = EBADF; 1661 } else { 1662 fdp->fd_files[fd].fileflags |= add_flags; 1663 error = 0; 1664 } 1665 spin_unlock(&fdp->fd_spin); 1666 return (error); 1667 } 1668 1669 /* 1670 * MPSAFE 1671 */ 1672 int 1673 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags) 1674 { 1675 int error; 1676 1677 spin_lock(&fdp->fd_spin); 1678 if (((u_int)fd) >= fdp->fd_nfiles) { 1679 error = EBADF; 1680 } else if (fdp->fd_files[fd].fp == NULL) { 1681 error = EBADF; 1682 } else { 1683 fdp->fd_files[fd].fileflags &= ~rem_flags; 1684 error = 0; 1685 } 1686 spin_unlock(&fdp->fd_spin); 1687 return (error); 1688 } 1689 1690 /* 1691 * Set/Change/Clear the creds for a fp and synchronize the uidinfo. 1692 */ 1693 void 1694 fsetcred(struct file *fp, struct ucred *ncr) 1695 { 1696 struct ucred *ocr; 1697 struct uidinfo *uip; 1698 1699 ocr = fp->f_cred; 1700 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) { 1701 if (ocr) { 1702 uip = ocr->cr_uidinfo; 1703 atomic_add_int(&uip->ui_openfiles, -1); 1704 } 1705 if (ncr) { 1706 uip = ncr->cr_uidinfo; 1707 atomic_add_int(&uip->ui_openfiles, 1); 1708 } 1709 } 1710 if (ncr) 1711 crhold(ncr); 1712 fp->f_cred = ncr; 1713 if (ocr) 1714 crfree(ocr); 1715 } 1716 1717 /* 1718 * Free a file descriptor. 1719 */ 1720 static 1721 void 1722 ffree(struct file *fp) 1723 { 1724 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 1725 spin_lock(&filehead_spin); 1726 LIST_REMOVE(fp, f_list); 1727 nfiles--; 1728 spin_unlock(&filehead_spin); 1729 fsetcred(fp, NULL); 1730 if (fp->f_nchandle.ncp) 1731 cache_drop(&fp->f_nchandle); 1732 kfree(fp, M_FILE); 1733 } 1734 1735 /* 1736 * called from init_main, initialize filedesc0 for proc0. 1737 */ 1738 void 1739 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask) 1740 { 1741 p0->p_fd = fdp0; 1742 p0->p_fdtol = NULL; 1743 fdp0->fd_refcnt = 1; 1744 fdp0->fd_cmask = cmask; 1745 fdp0->fd_files = fdp0->fd_builtin_files; 1746 fdp0->fd_nfiles = NDFILE; 1747 fdp0->fd_lastfile = -1; 1748 spin_init(&fdp0->fd_spin); 1749 } 1750 1751 /* 1752 * Build a new filedesc structure. 1753 * 1754 * NOT MPSAFE (vref) 1755 */ 1756 struct filedesc * 1757 fdinit(struct proc *p) 1758 { 1759 struct filedesc *newfdp; 1760 struct filedesc *fdp = p->p_fd; 1761 1762 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO); 1763 spin_lock(&fdp->fd_spin); 1764 if (fdp->fd_cdir) { 1765 newfdp->fd_cdir = fdp->fd_cdir; 1766 vref(newfdp->fd_cdir); 1767 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1768 } 1769 1770 /* 1771 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of 1772 * proc0, but should unconditionally exist in other processes. 1773 */ 1774 if (fdp->fd_rdir) { 1775 newfdp->fd_rdir = fdp->fd_rdir; 1776 vref(newfdp->fd_rdir); 1777 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1778 } 1779 if (fdp->fd_jdir) { 1780 newfdp->fd_jdir = fdp->fd_jdir; 1781 vref(newfdp->fd_jdir); 1782 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1783 } 1784 spin_unlock(&fdp->fd_spin); 1785 1786 /* Create the file descriptor table. */ 1787 newfdp->fd_refcnt = 1; 1788 newfdp->fd_cmask = cmask; 1789 newfdp->fd_files = newfdp->fd_builtin_files; 1790 newfdp->fd_nfiles = NDFILE; 1791 newfdp->fd_lastfile = -1; 1792 spin_init(&newfdp->fd_spin); 1793 1794 return (newfdp); 1795 } 1796 1797 /* 1798 * Share a filedesc structure. 1799 * 1800 * MPSAFE 1801 */ 1802 struct filedesc * 1803 fdshare(struct proc *p) 1804 { 1805 struct filedesc *fdp; 1806 1807 fdp = p->p_fd; 1808 spin_lock(&fdp->fd_spin); 1809 fdp->fd_refcnt++; 1810 spin_unlock(&fdp->fd_spin); 1811 return (fdp); 1812 } 1813 1814 /* 1815 * Copy a filedesc structure. 1816 * 1817 * MPSAFE 1818 */ 1819 int 1820 fdcopy(struct proc *p, struct filedesc **fpp) 1821 { 1822 struct filedesc *fdp = p->p_fd; 1823 struct filedesc *newfdp; 1824 struct fdnode *fdnode; 1825 int i; 1826 int ni; 1827 1828 /* 1829 * Certain daemons might not have file descriptors. 1830 */ 1831 if (fdp == NULL) 1832 return (0); 1833 1834 /* 1835 * Allocate the new filedesc and fd_files[] array. This can race 1836 * with operations by other threads on the fdp so we have to be 1837 * careful. 1838 */ 1839 newfdp = kmalloc(sizeof(struct filedesc), 1840 M_FILEDESC, M_WAITOK | M_ZERO | M_NULLOK); 1841 if (newfdp == NULL) { 1842 *fpp = NULL; 1843 return (-1); 1844 } 1845 again: 1846 spin_lock(&fdp->fd_spin); 1847 if (fdp->fd_lastfile < NDFILE) { 1848 newfdp->fd_files = newfdp->fd_builtin_files; 1849 i = NDFILE; 1850 } else { 1851 /* 1852 * We have to allocate (N^2-1) entries for our in-place 1853 * binary tree. Allow the table to shrink. 1854 */ 1855 i = fdp->fd_nfiles; 1856 ni = (i - 1) / 2; 1857 while (ni > fdp->fd_lastfile && ni > NDFILE) { 1858 i = ni; 1859 ni = (i - 1) / 2; 1860 } 1861 spin_unlock(&fdp->fd_spin); 1862 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode), 1863 M_FILEDESC, M_WAITOK | M_ZERO); 1864 1865 /* 1866 * Check for race, retry 1867 */ 1868 spin_lock(&fdp->fd_spin); 1869 if (i <= fdp->fd_lastfile) { 1870 spin_unlock(&fdp->fd_spin); 1871 kfree(newfdp->fd_files, M_FILEDESC); 1872 goto again; 1873 } 1874 } 1875 1876 /* 1877 * Dup the remaining fields. vref() and cache_hold() can be 1878 * safely called while holding the read spinlock on fdp. 1879 * 1880 * The read spinlock on fdp is still being held. 1881 * 1882 * NOTE: vref and cache_hold calls for the case where the vnode 1883 * or cache entry already has at least one ref may be called 1884 * while holding spin locks. 1885 */ 1886 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) { 1887 vref(newfdp->fd_cdir); 1888 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1889 } 1890 /* 1891 * We must check for fd_rdir here, at least for now because 1892 * the init process is created before we have access to the 1893 * rootvode to take a reference to it. 1894 */ 1895 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) { 1896 vref(newfdp->fd_rdir); 1897 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1898 } 1899 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) { 1900 vref(newfdp->fd_jdir); 1901 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1902 } 1903 newfdp->fd_refcnt = 1; 1904 newfdp->fd_nfiles = i; 1905 newfdp->fd_lastfile = fdp->fd_lastfile; 1906 newfdp->fd_freefile = fdp->fd_freefile; 1907 newfdp->fd_cmask = fdp->fd_cmask; 1908 spin_init(&newfdp->fd_spin); 1909 1910 /* 1911 * Copy the descriptor table through (i). This also copies the 1912 * allocation state. Then go through and ref the file pointers 1913 * and clean up any KQ descriptors. 1914 * 1915 * kq descriptors cannot be copied. Since we haven't ref'd the 1916 * copied files yet we can ignore the return value from funsetfd(). 1917 * 1918 * The read spinlock on fdp is still being held. 1919 */ 1920 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode)); 1921 for (i = 0 ; i < newfdp->fd_nfiles; ++i) { 1922 fdnode = &newfdp->fd_files[i]; 1923 if (fdnode->reserved) { 1924 fdreserve_locked(newfdp, i, -1); 1925 fdnode->reserved = 0; 1926 fdfixup_locked(newfdp, i); 1927 } else if (fdnode->fp) { 1928 if (fdnode->fp->f_type == DTYPE_KQUEUE) { 1929 (void)funsetfd_locked(newfdp, i); 1930 } else { 1931 fhold(fdnode->fp); 1932 } 1933 } 1934 } 1935 spin_unlock(&fdp->fd_spin); 1936 *fpp = newfdp; 1937 return (0); 1938 } 1939 1940 /* 1941 * Release a filedesc structure. 1942 * 1943 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE) 1944 */ 1945 void 1946 fdfree(struct proc *p, struct filedesc *repl) 1947 { 1948 struct filedesc *fdp; 1949 struct fdnode *fdnode; 1950 int i; 1951 struct filedesc_to_leader *fdtol; 1952 struct file *fp; 1953 struct vnode *vp; 1954 struct flock lf; 1955 1956 /* 1957 * Certain daemons might not have file descriptors. 1958 */ 1959 fdp = p->p_fd; 1960 if (fdp == NULL) { 1961 p->p_fd = repl; 1962 return; 1963 } 1964 1965 /* 1966 * Severe messing around to follow. 1967 */ 1968 spin_lock(&fdp->fd_spin); 1969 1970 /* Check for special need to clear POSIX style locks */ 1971 fdtol = p->p_fdtol; 1972 if (fdtol != NULL) { 1973 KASSERT(fdtol->fdl_refcount > 0, 1974 ("filedesc_to_refcount botch: fdl_refcount=%d", 1975 fdtol->fdl_refcount)); 1976 if (fdtol->fdl_refcount == 1 && 1977 (p->p_leader->p_flags & P_ADVLOCK) != 0) { 1978 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1979 fdnode = &fdp->fd_files[i]; 1980 if (fdnode->fp == NULL || 1981 fdnode->fp->f_type != DTYPE_VNODE) { 1982 continue; 1983 } 1984 fp = fdnode->fp; 1985 fhold(fp); 1986 spin_unlock(&fdp->fd_spin); 1987 1988 lf.l_whence = SEEK_SET; 1989 lf.l_start = 0; 1990 lf.l_len = 0; 1991 lf.l_type = F_UNLCK; 1992 vp = (struct vnode *)fp->f_data; 1993 (void) VOP_ADVLOCK(vp, 1994 (caddr_t)p->p_leader, 1995 F_UNLCK, 1996 &lf, 1997 F_POSIX); 1998 fdrop(fp); 1999 spin_lock(&fdp->fd_spin); 2000 } 2001 } 2002 retry: 2003 if (fdtol->fdl_refcount == 1) { 2004 if (fdp->fd_holdleaderscount > 0 && 2005 (p->p_leader->p_flags & P_ADVLOCK) != 0) { 2006 /* 2007 * close() or do_dup() has cleared a reference 2008 * in a shared file descriptor table. 2009 */ 2010 fdp->fd_holdleaderswakeup = 1; 2011 ssleep(&fdp->fd_holdleaderscount, 2012 &fdp->fd_spin, 0, "fdlhold", 0); 2013 goto retry; 2014 } 2015 if (fdtol->fdl_holdcount > 0) { 2016 /* 2017 * Ensure that fdtol->fdl_leader 2018 * remains valid in closef(). 2019 */ 2020 fdtol->fdl_wakeup = 1; 2021 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0); 2022 goto retry; 2023 } 2024 } 2025 fdtol->fdl_refcount--; 2026 if (fdtol->fdl_refcount == 0 && 2027 fdtol->fdl_holdcount == 0) { 2028 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2029 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2030 } else { 2031 fdtol = NULL; 2032 } 2033 p->p_fdtol = NULL; 2034 if (fdtol != NULL) { 2035 spin_unlock(&fdp->fd_spin); 2036 kfree(fdtol, M_FILEDESC_TO_LEADER); 2037 spin_lock(&fdp->fd_spin); 2038 } 2039 } 2040 if (--fdp->fd_refcnt > 0) { 2041 spin_unlock(&fdp->fd_spin); 2042 spin_lock(&p->p_spin); 2043 p->p_fd = repl; 2044 spin_unlock(&p->p_spin); 2045 return; 2046 } 2047 2048 /* 2049 * Even though we are the last reference to the structure allproc 2050 * scans may still reference the structure. Maintain proper 2051 * locks until we can replace p->p_fd. 2052 * 2053 * Also note that kqueue's closef still needs to reference the 2054 * fdp via p->p_fd, so we have to close the descriptors before 2055 * we replace p->p_fd. 2056 */ 2057 for (i = 0; i <= fdp->fd_lastfile; ++i) { 2058 if (fdp->fd_files[i].fp) { 2059 fp = funsetfd_locked(fdp, i); 2060 if (fp) { 2061 spin_unlock(&fdp->fd_spin); 2062 if (SLIST_FIRST(&fp->f_klist)) 2063 knote_fdclose(fp, fdp, i); 2064 closef(fp, p); 2065 spin_lock(&fdp->fd_spin); 2066 } 2067 } 2068 } 2069 spin_unlock(&fdp->fd_spin); 2070 2071 /* 2072 * Interlock against an allproc scan operations (typically frevoke). 2073 */ 2074 spin_lock(&p->p_spin); 2075 p->p_fd = repl; 2076 spin_unlock(&p->p_spin); 2077 2078 /* 2079 * Wait for any softrefs to go away. This race rarely occurs so 2080 * we can use a non-critical-path style poll/sleep loop. The 2081 * race only occurs against allproc scans. 2082 * 2083 * No new softrefs can occur with the fdp disconnected from the 2084 * process. 2085 */ 2086 if (fdp->fd_softrefs) { 2087 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid); 2088 while (fdp->fd_softrefs) 2089 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1); 2090 } 2091 2092 if (fdp->fd_files != fdp->fd_builtin_files) 2093 kfree(fdp->fd_files, M_FILEDESC); 2094 if (fdp->fd_cdir) { 2095 cache_drop(&fdp->fd_ncdir); 2096 vrele(fdp->fd_cdir); 2097 } 2098 if (fdp->fd_rdir) { 2099 cache_drop(&fdp->fd_nrdir); 2100 vrele(fdp->fd_rdir); 2101 } 2102 if (fdp->fd_jdir) { 2103 cache_drop(&fdp->fd_njdir); 2104 vrele(fdp->fd_jdir); 2105 } 2106 kfree(fdp, M_FILEDESC); 2107 } 2108 2109 /* 2110 * Retrieve and reference the file pointer associated with a descriptor. 2111 * 2112 * MPSAFE 2113 */ 2114 struct file * 2115 holdfp(struct filedesc *fdp, int fd, int flag) 2116 { 2117 struct file* fp; 2118 2119 spin_lock_shared(&fdp->fd_spin); 2120 if (((u_int)fd) >= fdp->fd_nfiles) { 2121 fp = NULL; 2122 goto done; 2123 } 2124 if ((fp = fdp->fd_files[fd].fp) == NULL) 2125 goto done; 2126 if ((fp->f_flag & flag) == 0 && flag != -1) { 2127 fp = NULL; 2128 goto done; 2129 } 2130 fhold(fp); 2131 done: 2132 spin_unlock_shared(&fdp->fd_spin); 2133 return (fp); 2134 } 2135 2136 /* 2137 * holdsock() - load the struct file pointer associated 2138 * with a socket into *fpp. If an error occurs, non-zero 2139 * will be returned and *fpp will be set to NULL. 2140 * 2141 * MPSAFE 2142 */ 2143 int 2144 holdsock(struct filedesc *fdp, int fd, struct file **fpp) 2145 { 2146 struct file *fp; 2147 int error; 2148 2149 spin_lock_shared(&fdp->fd_spin); 2150 if ((unsigned)fd >= fdp->fd_nfiles) { 2151 error = EBADF; 2152 fp = NULL; 2153 goto done; 2154 } 2155 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2156 error = EBADF; 2157 goto done; 2158 } 2159 if (fp->f_type != DTYPE_SOCKET) { 2160 error = ENOTSOCK; 2161 goto done; 2162 } 2163 fhold(fp); 2164 error = 0; 2165 done: 2166 spin_unlock_shared(&fdp->fd_spin); 2167 *fpp = fp; 2168 return (error); 2169 } 2170 2171 /* 2172 * Convert a user file descriptor to a held file pointer. 2173 * 2174 * MPSAFE 2175 */ 2176 int 2177 holdvnode(struct filedesc *fdp, int fd, struct file **fpp) 2178 { 2179 struct file *fp; 2180 int error; 2181 2182 spin_lock_shared(&fdp->fd_spin); 2183 if ((unsigned)fd >= fdp->fd_nfiles) { 2184 error = EBADF; 2185 fp = NULL; 2186 goto done; 2187 } 2188 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2189 error = EBADF; 2190 goto done; 2191 } 2192 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 2193 fp = NULL; 2194 error = EINVAL; 2195 goto done; 2196 } 2197 fhold(fp); 2198 error = 0; 2199 done: 2200 spin_unlock_shared(&fdp->fd_spin); 2201 *fpp = fp; 2202 return (error); 2203 } 2204 2205 /* 2206 * For setugid programs, we don't want to people to use that setugidness 2207 * to generate error messages which write to a file which otherwise would 2208 * otherwise be off-limits to the process. 2209 * 2210 * This is a gross hack to plug the hole. A better solution would involve 2211 * a special vop or other form of generalized access control mechanism. We 2212 * go ahead and just reject all procfs file systems accesses as dangerous. 2213 * 2214 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2215 * sufficient. We also don't for check setugidness since we know we are. 2216 */ 2217 static int 2218 is_unsafe(struct file *fp) 2219 { 2220 if (fp->f_type == DTYPE_VNODE && 2221 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 2222 return (1); 2223 return (0); 2224 } 2225 2226 /* 2227 * Make this setguid thing safe, if at all possible. 2228 * 2229 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2230 */ 2231 void 2232 setugidsafety(struct proc *p) 2233 { 2234 struct filedesc *fdp = p->p_fd; 2235 int i; 2236 2237 /* Certain daemons might not have file descriptors. */ 2238 if (fdp == NULL) 2239 return; 2240 2241 /* 2242 * note: fdp->fd_files may be reallocated out from under us while 2243 * we are blocked in a close. Be careful! 2244 */ 2245 for (i = 0; i <= fdp->fd_lastfile; i++) { 2246 if (i > 2) 2247 break; 2248 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) { 2249 struct file *fp; 2250 2251 /* 2252 * NULL-out descriptor prior to close to avoid 2253 * a race while close blocks. 2254 */ 2255 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2256 knote_fdclose(fp, fdp, i); 2257 closef(fp, p); 2258 } 2259 } 2260 } 2261 } 2262 2263 /* 2264 * Close any files on exec? 2265 * 2266 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2267 */ 2268 void 2269 fdcloseexec(struct proc *p) 2270 { 2271 struct filedesc *fdp = p->p_fd; 2272 int i; 2273 2274 /* Certain daemons might not have file descriptors. */ 2275 if (fdp == NULL) 2276 return; 2277 2278 /* 2279 * We cannot cache fd_files since operations may block and rip 2280 * them out from under us. 2281 */ 2282 for (i = 0; i <= fdp->fd_lastfile; i++) { 2283 if (fdp->fd_files[i].fp != NULL && 2284 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) { 2285 struct file *fp; 2286 2287 /* 2288 * NULL-out descriptor prior to close to avoid 2289 * a race while close blocks. 2290 */ 2291 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2292 knote_fdclose(fp, fdp, i); 2293 closef(fp, p); 2294 } 2295 } 2296 } 2297 } 2298 2299 /* 2300 * It is unsafe for set[ug]id processes to be started with file 2301 * descriptors 0..2 closed, as these descriptors are given implicit 2302 * significance in the Standard C library. fdcheckstd() will create a 2303 * descriptor referencing /dev/null for each of stdin, stdout, and 2304 * stderr that is not already open. 2305 * 2306 * NOT MPSAFE - calls falloc, vn_open, etc 2307 */ 2308 int 2309 fdcheckstd(struct lwp *lp) 2310 { 2311 struct nlookupdata nd; 2312 struct filedesc *fdp; 2313 struct file *fp; 2314 int retval; 2315 int i, error, flags, devnull; 2316 2317 fdp = lp->lwp_proc->p_fd; 2318 if (fdp == NULL) 2319 return (0); 2320 devnull = -1; 2321 error = 0; 2322 for (i = 0; i < 3; i++) { 2323 if (fdp->fd_files[i].fp != NULL) 2324 continue; 2325 if (devnull < 0) { 2326 if ((error = falloc(lp, &fp, &devnull)) != 0) 2327 break; 2328 2329 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, 2330 NLC_FOLLOW|NLC_LOCKVP); 2331 flags = FREAD | FWRITE; 2332 if (error == 0) 2333 error = vn_open(&nd, fp, flags, 0); 2334 if (error == 0) 2335 fsetfd(fdp, fp, devnull); 2336 else 2337 fsetfd(fdp, NULL, devnull); 2338 fdrop(fp); 2339 nlookup_done(&nd); 2340 if (error) 2341 break; 2342 KKASSERT(i == devnull); 2343 } else { 2344 error = kern_dup(DUP_FIXED, devnull, i, &retval); 2345 if (error != 0) 2346 break; 2347 } 2348 } 2349 return (error); 2350 } 2351 2352 /* 2353 * Internal form of close. 2354 * Decrement reference count on file structure. 2355 * Note: td and/or p may be NULL when closing a file 2356 * that was being passed in a message. 2357 * 2358 * MPALMOSTSAFE - acquires mplock for VOP operations 2359 */ 2360 int 2361 closef(struct file *fp, struct proc *p) 2362 { 2363 struct vnode *vp; 2364 struct flock lf; 2365 struct filedesc_to_leader *fdtol; 2366 2367 if (fp == NULL) 2368 return (0); 2369 2370 /* 2371 * POSIX record locking dictates that any close releases ALL 2372 * locks owned by this process. This is handled by setting 2373 * a flag in the unlock to free ONLY locks obeying POSIX 2374 * semantics, and not to free BSD-style file locks. 2375 * If the descriptor was in a message, POSIX-style locks 2376 * aren't passed with the descriptor. 2377 */ 2378 if (p != NULL && fp->f_type == DTYPE_VNODE && 2379 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2380 ) { 2381 if ((p->p_leader->p_flags & P_ADVLOCK) != 0) { 2382 lf.l_whence = SEEK_SET; 2383 lf.l_start = 0; 2384 lf.l_len = 0; 2385 lf.l_type = F_UNLCK; 2386 vp = (struct vnode *)fp->f_data; 2387 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 2388 &lf, F_POSIX); 2389 } 2390 fdtol = p->p_fdtol; 2391 if (fdtol != NULL) { 2392 lwkt_gettoken(&p->p_token); 2393 /* 2394 * Handle special case where file descriptor table 2395 * is shared between multiple process leaders. 2396 */ 2397 for (fdtol = fdtol->fdl_next; 2398 fdtol != p->p_fdtol; 2399 fdtol = fdtol->fdl_next) { 2400 if ((fdtol->fdl_leader->p_flags & 2401 P_ADVLOCK) == 0) 2402 continue; 2403 fdtol->fdl_holdcount++; 2404 lf.l_whence = SEEK_SET; 2405 lf.l_start = 0; 2406 lf.l_len = 0; 2407 lf.l_type = F_UNLCK; 2408 vp = (struct vnode *)fp->f_data; 2409 (void) VOP_ADVLOCK(vp, 2410 (caddr_t)fdtol->fdl_leader, 2411 F_UNLCK, &lf, F_POSIX); 2412 fdtol->fdl_holdcount--; 2413 if (fdtol->fdl_holdcount == 0 && 2414 fdtol->fdl_wakeup != 0) { 2415 fdtol->fdl_wakeup = 0; 2416 wakeup(fdtol); 2417 } 2418 } 2419 lwkt_reltoken(&p->p_token); 2420 } 2421 } 2422 return (fdrop(fp)); 2423 } 2424 2425 /* 2426 * MPSAFE 2427 * 2428 * fhold() can only be called if f_count is already at least 1 (i.e. the 2429 * caller of fhold() already has a reference to the file pointer in some 2430 * manner or other). 2431 * 2432 * f_count is not spin-locked. Instead, atomic ops are used for 2433 * incrementing, decrementing, and handling the 1->0 transition. 2434 */ 2435 void 2436 fhold(struct file *fp) 2437 { 2438 atomic_add_int(&fp->f_count, 1); 2439 } 2440 2441 /* 2442 * fdrop() - drop a reference to a descriptor 2443 * 2444 * MPALMOSTSAFE - acquires mplock for final close sequence 2445 */ 2446 int 2447 fdrop(struct file *fp) 2448 { 2449 struct flock lf; 2450 struct vnode *vp; 2451 int error; 2452 2453 /* 2454 * A combined fetch and subtract is needed to properly detect 2455 * 1->0 transitions, otherwise two cpus dropping from a ref 2456 * count of 2 might both try to run the 1->0 code. 2457 */ 2458 if (atomic_fetchadd_int(&fp->f_count, -1) > 1) 2459 return (0); 2460 2461 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL); 2462 2463 /* 2464 * The last reference has gone away, we own the fp structure free 2465 * and clear. 2466 */ 2467 if (fp->f_count < 0) 2468 panic("fdrop: count < 0"); 2469 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE && 2470 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2471 ) { 2472 lf.l_whence = SEEK_SET; 2473 lf.l_start = 0; 2474 lf.l_len = 0; 2475 lf.l_type = F_UNLCK; 2476 vp = (struct vnode *)fp->f_data; 2477 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2478 } 2479 if (fp->f_ops != &badfileops) 2480 error = fo_close(fp); 2481 else 2482 error = 0; 2483 ffree(fp); 2484 return (error); 2485 } 2486 2487 /* 2488 * Apply an advisory lock on a file descriptor. 2489 * 2490 * Just attempt to get a record lock of the requested type on 2491 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2492 * 2493 * MPALMOSTSAFE 2494 */ 2495 int 2496 sys_flock(struct flock_args *uap) 2497 { 2498 struct proc *p = curproc; 2499 struct file *fp; 2500 struct vnode *vp; 2501 struct flock lf; 2502 int error; 2503 2504 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 2505 return (EBADF); 2506 if (fp->f_type != DTYPE_VNODE) { 2507 error = EOPNOTSUPP; 2508 goto done; 2509 } 2510 vp = (struct vnode *)fp->f_data; 2511 lf.l_whence = SEEK_SET; 2512 lf.l_start = 0; 2513 lf.l_len = 0; 2514 if (uap->how & LOCK_UN) { 2515 lf.l_type = F_UNLCK; 2516 fp->f_flag &= ~FHASLOCK; 2517 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2518 goto done; 2519 } 2520 if (uap->how & LOCK_EX) 2521 lf.l_type = F_WRLCK; 2522 else if (uap->how & LOCK_SH) 2523 lf.l_type = F_RDLCK; 2524 else { 2525 error = EBADF; 2526 goto done; 2527 } 2528 fp->f_flag |= FHASLOCK; 2529 if (uap->how & LOCK_NB) 2530 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0); 2531 else 2532 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT); 2533 done: 2534 fdrop(fp); 2535 return (error); 2536 } 2537 2538 /* 2539 * File Descriptor pseudo-device driver (/dev/fd/). 2540 * 2541 * Opening minor device N dup()s the file (if any) connected to file 2542 * descriptor N belonging to the calling process. Note that this driver 2543 * consists of only the ``open()'' routine, because all subsequent 2544 * references to this file will be direct to the other driver. 2545 */ 2546 static int 2547 fdopen(struct dev_open_args *ap) 2548 { 2549 thread_t td = curthread; 2550 2551 KKASSERT(td->td_lwp != NULL); 2552 2553 /* 2554 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the 2555 * the file descriptor being sought for duplication. The error 2556 * return ensures that the vnode for this device will be released 2557 * by vn_open. Open will detect this special error and take the 2558 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2559 * will simply report the error. 2560 */ 2561 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev); 2562 return (ENODEV); 2563 } 2564 2565 /* 2566 * The caller has reserved the file descriptor dfd for us. On success we 2567 * must fsetfd() it. On failure the caller will clean it up. 2568 * 2569 * MPSAFE 2570 */ 2571 int 2572 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error) 2573 { 2574 struct file *wfp; 2575 struct file *xfp; 2576 int werror; 2577 2578 if ((wfp = holdfp(fdp, sfd, -1)) == NULL) 2579 return (EBADF); 2580 2581 /* 2582 * Close a revoke/dup race. Duping a descriptor marked as revoked 2583 * will dup a dummy descriptor instead of the real one. 2584 */ 2585 if (wfp->f_flag & FREVOKED) { 2586 kprintf("Warning: attempt to dup() a revoked descriptor\n"); 2587 fdrop(wfp); 2588 wfp = NULL; 2589 werror = falloc(NULL, &wfp, NULL); 2590 if (werror) 2591 return (werror); 2592 } 2593 2594 /* 2595 * There are two cases of interest here. 2596 * 2597 * For ENODEV simply dup sfd to file descriptor dfd and return. 2598 * 2599 * For ENXIO steal away the file structure from sfd and store it 2600 * dfd. sfd is effectively closed by this operation. 2601 * 2602 * Any other error code is just returned. 2603 */ 2604 switch (error) { 2605 case ENODEV: 2606 /* 2607 * Check that the mode the file is being opened for is a 2608 * subset of the mode of the existing descriptor. 2609 */ 2610 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2611 error = EACCES; 2612 break; 2613 } 2614 spin_lock(&fdp->fd_spin); 2615 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2616 fsetfd_locked(fdp, wfp, dfd); 2617 spin_unlock(&fdp->fd_spin); 2618 error = 0; 2619 break; 2620 case ENXIO: 2621 /* 2622 * Steal away the file pointer from dfd, and stuff it into indx. 2623 */ 2624 spin_lock(&fdp->fd_spin); 2625 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2626 fsetfd(fdp, wfp, dfd); 2627 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) { 2628 spin_unlock(&fdp->fd_spin); 2629 fdrop(xfp); 2630 } else { 2631 spin_unlock(&fdp->fd_spin); 2632 } 2633 error = 0; 2634 break; 2635 default: 2636 break; 2637 } 2638 fdrop(wfp); 2639 return (error); 2640 } 2641 2642 /* 2643 * NOT MPSAFE - I think these refer to a common file descriptor table 2644 * and we need to spinlock that to link fdtol in. 2645 */ 2646 struct filedesc_to_leader * 2647 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 2648 struct proc *leader) 2649 { 2650 struct filedesc_to_leader *fdtol; 2651 2652 fdtol = kmalloc(sizeof(struct filedesc_to_leader), 2653 M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO); 2654 fdtol->fdl_refcount = 1; 2655 fdtol->fdl_holdcount = 0; 2656 fdtol->fdl_wakeup = 0; 2657 fdtol->fdl_leader = leader; 2658 if (old != NULL) { 2659 fdtol->fdl_next = old->fdl_next; 2660 fdtol->fdl_prev = old; 2661 old->fdl_next = fdtol; 2662 fdtol->fdl_next->fdl_prev = fdtol; 2663 } else { 2664 fdtol->fdl_next = fdtol; 2665 fdtol->fdl_prev = fdtol; 2666 } 2667 return fdtol; 2668 } 2669 2670 /* 2671 * Scan all file pointers in the system. The callback is made with 2672 * the master list spinlock held exclusively. 2673 * 2674 * MPSAFE 2675 */ 2676 void 2677 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data) 2678 { 2679 struct file *fp; 2680 int res; 2681 2682 spin_lock(&filehead_spin); 2683 LIST_FOREACH(fp, &filehead, f_list) { 2684 res = callback(fp, data); 2685 if (res < 0) 2686 break; 2687 } 2688 spin_unlock(&filehead_spin); 2689 } 2690 2691 /* 2692 * Get file structures. 2693 * 2694 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe) 2695 */ 2696 2697 struct sysctl_kern_file_info { 2698 int count; 2699 int error; 2700 struct sysctl_req *req; 2701 }; 2702 2703 static int sysctl_kern_file_callback(struct proc *p, void *data); 2704 2705 static int 2706 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2707 { 2708 struct sysctl_kern_file_info info; 2709 2710 /* 2711 * Note: because the number of file descriptors is calculated 2712 * in different ways for sizing vs returning the data, 2713 * there is information leakage from the first loop. However, 2714 * it is of a similar order of magnitude to the leakage from 2715 * global system statistics such as kern.openfiles. 2716 * 2717 * When just doing a count, note that we cannot just count 2718 * the elements and add f_count via the filehead list because 2719 * threaded processes share their descriptor table and f_count might 2720 * still be '1' in that case. 2721 * 2722 * Since the SYSCTL op can block, we must hold the process to 2723 * prevent it being ripped out from under us either in the 2724 * file descriptor loop or in the greater LIST_FOREACH. The 2725 * process may be in varying states of disrepair. If the process 2726 * is in SZOMB we may have caught it just as it is being removed 2727 * from the allproc list, we must skip it in that case to maintain 2728 * an unbroken chain through the allproc list. 2729 */ 2730 info.count = 0; 2731 info.error = 0; 2732 info.req = req; 2733 allproc_scan(sysctl_kern_file_callback, &info); 2734 2735 /* 2736 * When just calculating the size, overestimate a bit to try to 2737 * prevent system activity from causing the buffer-fill call 2738 * to fail later on. 2739 */ 2740 if (req->oldptr == NULL) { 2741 info.count = (info.count + 16) + (info.count / 10); 2742 info.error = SYSCTL_OUT(req, NULL, 2743 info.count * sizeof(struct kinfo_file)); 2744 } 2745 return (info.error); 2746 } 2747 2748 static int 2749 sysctl_kern_file_callback(struct proc *p, void *data) 2750 { 2751 struct sysctl_kern_file_info *info = data; 2752 struct kinfo_file kf; 2753 struct filedesc *fdp; 2754 struct file *fp; 2755 uid_t uid; 2756 int n; 2757 2758 if (p->p_stat == SIDL || p->p_stat == SZOMB) 2759 return(0); 2760 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0) 2761 return(0); 2762 2763 /* 2764 * Softref the fdp to prevent it from being destroyed 2765 */ 2766 spin_lock(&p->p_spin); 2767 if ((fdp = p->p_fd) == NULL) { 2768 spin_unlock(&p->p_spin); 2769 return(0); 2770 } 2771 atomic_add_int(&fdp->fd_softrefs, 1); 2772 spin_unlock(&p->p_spin); 2773 2774 /* 2775 * The fdp's own spinlock prevents the contents from being 2776 * modified. 2777 */ 2778 spin_lock_shared(&fdp->fd_spin); 2779 for (n = 0; n < fdp->fd_nfiles; ++n) { 2780 if ((fp = fdp->fd_files[n].fp) == NULL) 2781 continue; 2782 if (info->req->oldptr == NULL) { 2783 ++info->count; 2784 } else { 2785 uid = p->p_ucred ? p->p_ucred->cr_uid : -1; 2786 kcore_make_file(&kf, fp, p->p_pid, uid, n); 2787 spin_unlock_shared(&fdp->fd_spin); 2788 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf)); 2789 spin_lock_shared(&fdp->fd_spin); 2790 if (info->error) 2791 break; 2792 } 2793 } 2794 spin_unlock_shared(&fdp->fd_spin); 2795 atomic_subtract_int(&fdp->fd_softrefs, 1); 2796 if (info->error) 2797 return(-1); 2798 return(0); 2799 } 2800 2801 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2802 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 2803 2804 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW, 2805 &minfilesperproc, 0, "Minimum files allowed open per process"); 2806 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2807 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2808 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW, 2809 &maxfilesperuser, 0, "Maximum files allowed open per user"); 2810 2811 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2812 &maxfiles, 0, "Maximum number of files"); 2813 2814 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, 2815 &maxfilesrootres, 0, "Descriptors reserved for root use"); 2816 2817 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2818 &nfiles, 0, "System-wide number of open files"); 2819 2820 static void 2821 fildesc_drvinit(void *unused) 2822 { 2823 int fd; 2824 2825 for (fd = 0; fd < NUMFDESC; fd++) { 2826 make_dev(&fildesc_ops, fd, 2827 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 2828 } 2829 2830 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 2831 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 2832 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 2833 } 2834 2835 /* 2836 * MPSAFE 2837 */ 2838 struct fileops badfileops = { 2839 .fo_read = badfo_readwrite, 2840 .fo_write = badfo_readwrite, 2841 .fo_ioctl = badfo_ioctl, 2842 .fo_kqfilter = badfo_kqfilter, 2843 .fo_stat = badfo_stat, 2844 .fo_close = badfo_close, 2845 .fo_shutdown = badfo_shutdown 2846 }; 2847 2848 int 2849 badfo_readwrite( 2850 struct file *fp, 2851 struct uio *uio, 2852 struct ucred *cred, 2853 int flags 2854 ) { 2855 return (EBADF); 2856 } 2857 2858 int 2859 badfo_ioctl(struct file *fp, u_long com, caddr_t data, 2860 struct ucred *cred, struct sysmsg *msgv) 2861 { 2862 return (EBADF); 2863 } 2864 2865 /* 2866 * Must return an error to prevent registration, typically 2867 * due to a revoked descriptor (file_filtops assigned). 2868 */ 2869 int 2870 badfo_kqfilter(struct file *fp, struct knote *kn) 2871 { 2872 return (EOPNOTSUPP); 2873 } 2874 2875 /* 2876 * MPSAFE 2877 */ 2878 int 2879 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred) 2880 { 2881 return (EBADF); 2882 } 2883 2884 /* 2885 * MPSAFE 2886 */ 2887 int 2888 badfo_close(struct file *fp) 2889 { 2890 return (EBADF); 2891 } 2892 2893 /* 2894 * MPSAFE 2895 */ 2896 int 2897 badfo_shutdown(struct file *fp, int how) 2898 { 2899 return (EBADF); 2900 } 2901 2902 /* 2903 * MPSAFE 2904 */ 2905 int 2906 nofo_shutdown(struct file *fp, int how) 2907 { 2908 return (EOPNOTSUPP); 2909 } 2910 2911 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, 2912 fildesc_drvinit,NULL) 2913