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, "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 * 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 int 157 sys_getdtablesize(struct getdtablesize_args *uap) 158 { 159 struct proc *p = curproc; 160 struct plimit *limit = p->p_limit; 161 int dtsize; 162 163 spin_lock(&limit->p_spin); 164 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 165 dtsize = INT_MAX; 166 else 167 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur; 168 spin_unlock(&limit->p_spin); 169 170 if (dtsize > maxfilesperproc) 171 dtsize = maxfilesperproc; 172 if (dtsize < minfilesperproc) 173 dtsize = minfilesperproc; 174 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser) 175 dtsize = maxfilesperuser; 176 uap->sysmsg_result = dtsize; 177 return (0); 178 } 179 180 /* 181 * Duplicate a file descriptor to a particular value. 182 * 183 * note: keep in mind that a potential race condition exists when closing 184 * descriptors from a shared descriptor table (via rfork). 185 */ 186 int 187 sys_dup2(struct dup2_args *uap) 188 { 189 int error; 190 int fd = 0; 191 192 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd); 193 uap->sysmsg_fds[0] = fd; 194 195 return (error); 196 } 197 198 /* 199 * Duplicate a file descriptor. 200 */ 201 int 202 sys_dup(struct dup_args *uap) 203 { 204 int error; 205 int fd = 0; 206 207 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd); 208 uap->sysmsg_fds[0] = fd; 209 210 return (error); 211 } 212 213 /* 214 * MPALMOSTSAFE - acquires mplock for fp operations 215 */ 216 int 217 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred) 218 { 219 struct thread *td = curthread; 220 struct proc *p = td->td_proc; 221 struct file *fp; 222 struct vnode *vp; 223 u_int newmin; 224 u_int oflags; 225 u_int nflags; 226 int tmp, error, flg = F_POSIX; 227 228 KKASSERT(p); 229 230 /* 231 * Operations on file descriptors that do not require a file pointer. 232 */ 233 switch (cmd) { 234 case F_GETFD: 235 error = fgetfdflags(p->p_fd, fd, &tmp); 236 if (error == 0) 237 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0; 238 return (error); 239 240 case F_SETFD: 241 if (dat->fc_cloexec & FD_CLOEXEC) 242 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE); 243 else 244 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE); 245 return (error); 246 case F_DUPFD: 247 newmin = dat->fc_fd; 248 error = kern_dup(DUP_VARIABLE | DUP_FCNTL, fd, newmin, 249 &dat->fc_fd); 250 return (error); 251 case F_DUPFD_CLOEXEC: 252 newmin = dat->fc_fd; 253 error = kern_dup(DUP_VARIABLE | DUP_CLOEXEC | DUP_FCNTL, 254 fd, newmin, &dat->fc_fd); 255 return (error); 256 case F_DUP2FD: 257 newmin = dat->fc_fd; 258 error = kern_dup(DUP_FIXED, fd, newmin, &dat->fc_fd); 259 return (error); 260 case F_DUP2FD_CLOEXEC: 261 newmin = dat->fc_fd; 262 error = kern_dup(DUP_FIXED | DUP_CLOEXEC, fd, newmin, 263 &dat->fc_fd); 264 return (error); 265 default: 266 break; 267 } 268 269 /* 270 * Operations on file pointers 271 */ 272 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 273 return (EBADF); 274 275 switch (cmd) { 276 case F_GETFL: 277 dat->fc_flags = OFLAGS(fp->f_flag); 278 error = 0; 279 break; 280 281 case F_SETFL: 282 oflags = fp->f_flag; 283 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 284 nflags |= oflags & ~FCNTLFLAGS; 285 286 error = 0; 287 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY)) 288 error = EINVAL; 289 if (error == 0 && ((nflags ^ oflags) & FASYNC)) { 290 tmp = nflags & FASYNC; 291 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, 292 cred, NULL); 293 } 294 295 /* 296 * If no error, must be atomically set. 297 */ 298 while (error == 0) { 299 oflags = fp->f_flag; 300 cpu_ccfence(); 301 nflags = (oflags & ~FCNTLFLAGS) | (nflags & FCNTLFLAGS); 302 if (atomic_cmpset_int(&fp->f_flag, oflags, nflags)) 303 break; 304 cpu_pause(); 305 } 306 break; 307 308 case F_GETOWN: 309 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, 310 cred, NULL); 311 break; 312 313 case F_SETOWN: 314 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, 315 cred, NULL); 316 break; 317 318 case F_SETLKW: 319 flg |= F_WAIT; 320 /* Fall into F_SETLK */ 321 322 case F_SETLK: 323 if (fp->f_type != DTYPE_VNODE) { 324 error = EBADF; 325 break; 326 } 327 vp = (struct vnode *)fp->f_data; 328 329 /* 330 * copyin/lockop may block 331 */ 332 if (dat->fc_flock.l_whence == SEEK_CUR) 333 dat->fc_flock.l_start += fp->f_offset; 334 335 switch (dat->fc_flock.l_type) { 336 case F_RDLCK: 337 if ((fp->f_flag & FREAD) == 0) { 338 error = EBADF; 339 break; 340 } 341 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) { 342 lwkt_gettoken(&p->p_leader->p_token); 343 p->p_leader->p_flags |= P_ADVLOCK; 344 lwkt_reltoken(&p->p_leader->p_token); 345 } 346 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 347 &dat->fc_flock, flg); 348 break; 349 case F_WRLCK: 350 if ((fp->f_flag & FWRITE) == 0) { 351 error = EBADF; 352 break; 353 } 354 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) { 355 lwkt_gettoken(&p->p_leader->p_token); 356 p->p_leader->p_flags |= P_ADVLOCK; 357 lwkt_reltoken(&p->p_leader->p_token); 358 } 359 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 360 &dat->fc_flock, flg); 361 break; 362 case F_UNLCK: 363 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 364 &dat->fc_flock, F_POSIX); 365 break; 366 default: 367 error = EINVAL; 368 break; 369 } 370 371 /* 372 * It is possible to race a close() on the descriptor while 373 * we were blocked getting the lock. If this occurs the 374 * close might not have caught the lock. 375 */ 376 if (checkfdclosed(p->p_fd, fd, fp)) { 377 dat->fc_flock.l_whence = SEEK_SET; 378 dat->fc_flock.l_start = 0; 379 dat->fc_flock.l_len = 0; 380 dat->fc_flock.l_type = F_UNLCK; 381 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 382 F_UNLCK, &dat->fc_flock, F_POSIX); 383 } 384 break; 385 386 case F_GETLK: 387 if (fp->f_type != DTYPE_VNODE) { 388 error = EBADF; 389 break; 390 } 391 vp = (struct vnode *)fp->f_data; 392 /* 393 * copyin/lockop may block 394 */ 395 if (dat->fc_flock.l_type != F_RDLCK && 396 dat->fc_flock.l_type != F_WRLCK && 397 dat->fc_flock.l_type != F_UNLCK) { 398 error = EINVAL; 399 break; 400 } 401 if (dat->fc_flock.l_whence == SEEK_CUR) 402 dat->fc_flock.l_start += fp->f_offset; 403 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 404 &dat->fc_flock, F_POSIX); 405 break; 406 default: 407 error = EINVAL; 408 break; 409 } 410 411 fdrop(fp); 412 return (error); 413 } 414 415 /* 416 * The file control system call. 417 */ 418 int 419 sys_fcntl(struct fcntl_args *uap) 420 { 421 union fcntl_dat dat; 422 int error; 423 424 switch (uap->cmd) { 425 case F_DUPFD: 426 case F_DUP2FD: 427 case F_DUPFD_CLOEXEC: 428 case F_DUP2FD_CLOEXEC: 429 dat.fc_fd = uap->arg; 430 break; 431 case F_SETFD: 432 dat.fc_cloexec = uap->arg; 433 break; 434 case F_SETFL: 435 dat.fc_flags = uap->arg; 436 break; 437 case F_SETOWN: 438 dat.fc_owner = uap->arg; 439 break; 440 case F_SETLKW: 441 case F_SETLK: 442 case F_GETLK: 443 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 444 sizeof(struct flock)); 445 if (error) 446 return (error); 447 break; 448 } 449 450 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred); 451 452 if (error == 0) { 453 switch (uap->cmd) { 454 case F_DUPFD: 455 case F_DUP2FD: 456 case F_DUPFD_CLOEXEC: 457 case F_DUP2FD_CLOEXEC: 458 uap->sysmsg_result = dat.fc_fd; 459 break; 460 case F_GETFD: 461 uap->sysmsg_result = dat.fc_cloexec; 462 break; 463 case F_GETFL: 464 uap->sysmsg_result = dat.fc_flags; 465 break; 466 case F_GETOWN: 467 uap->sysmsg_result = dat.fc_owner; 468 break; 469 case F_GETLK: 470 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 471 sizeof(struct flock)); 472 break; 473 } 474 } 475 476 return (error); 477 } 478 479 /* 480 * Common code for dup, dup2, and fcntl(F_DUPFD). 481 * 482 * There are four type flags: DUP_FCNTL, DUP_FIXED, DUP_VARIABLE, and 483 * DUP_CLOEXEC. 484 * 485 * DUP_FCNTL is for handling EINVAL vs. EBADF differences between 486 * fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC and dup2() (per POSIX). 487 * The next two flags are mutually exclusive, and the fourth is optional. 488 * DUP_FIXED tells kern_dup() to destructively dup over an existing file 489 * descriptor if "new" is already open. DUP_VARIABLE tells kern_dup() 490 * to find the lowest unused file descriptor that is greater than or 491 * equal to "new". DUP_CLOEXEC, which works with either of the first 492 * two flags, sets the close-on-exec flag on the "new" file descriptor. 493 */ 494 int 495 kern_dup(int flags, int old, int new, int *res) 496 { 497 struct thread *td = curthread; 498 struct proc *p = td->td_proc; 499 struct filedesc *fdp = p->p_fd; 500 struct file *fp; 501 struct file *delfp; 502 int oldflags; 503 int holdleaders; 504 int dtsize; 505 int error, newfd; 506 507 /* 508 * Verify that we have a valid descriptor to dup from and 509 * possibly to dup to. When the new descriptor is out of 510 * bounds, fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC must 511 * return EINVAL, while dup2() returns EBADF in 512 * this case. 513 * 514 * NOTE: maxfilesperuser is not applicable to dup() 515 */ 516 retry: 517 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 518 dtsize = INT_MAX; 519 else 520 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 521 if (dtsize > maxfilesperproc) 522 dtsize = maxfilesperproc; 523 if (dtsize < minfilesperproc) 524 dtsize = minfilesperproc; 525 526 if (new < 0 || new > dtsize) 527 return (flags & DUP_FCNTL ? EINVAL : EBADF); 528 529 spin_lock(&fdp->fd_spin); 530 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) { 531 spin_unlock(&fdp->fd_spin); 532 return (EBADF); 533 } 534 if ((flags & DUP_FIXED) && old == new) { 535 *res = new; 536 if (flags & DUP_CLOEXEC) 537 fdp->fd_files[new].fileflags |= UF_EXCLOSE; 538 spin_unlock(&fdp->fd_spin); 539 return (0); 540 } 541 fp = fdp->fd_files[old].fp; 542 oldflags = fdp->fd_files[old].fileflags; 543 fhold(fp); 544 545 /* 546 * Allocate a new descriptor if DUP_VARIABLE, or expand the table 547 * if the requested descriptor is beyond the current table size. 548 * 549 * This can block. Retry if the source descriptor no longer matches 550 * or if our expectation in the expansion case races. 551 * 552 * If we are not expanding or allocating a new decriptor, then reset 553 * the target descriptor to a reserved state so we have a uniform 554 * setup for the next code block. 555 */ 556 if ((flags & DUP_VARIABLE) || new >= fdp->fd_nfiles) { 557 spin_unlock(&fdp->fd_spin); 558 error = fdalloc(p, new, &newfd); 559 spin_lock(&fdp->fd_spin); 560 if (error) { 561 spin_unlock(&fdp->fd_spin); 562 fdrop(fp); 563 return (error); 564 } 565 /* 566 * Check for ripout 567 */ 568 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) { 569 fsetfd_locked(fdp, NULL, newfd); 570 spin_unlock(&fdp->fd_spin); 571 fdrop(fp); 572 goto retry; 573 } 574 /* 575 * Check for expansion race 576 */ 577 if ((flags & DUP_VARIABLE) == 0 && new != newfd) { 578 fsetfd_locked(fdp, NULL, newfd); 579 spin_unlock(&fdp->fd_spin); 580 fdrop(fp); 581 goto retry; 582 } 583 /* 584 * Check for ripout, newfd reused old (this case probably 585 * can't occur). 586 */ 587 if (old == newfd) { 588 fsetfd_locked(fdp, NULL, newfd); 589 spin_unlock(&fdp->fd_spin); 590 fdrop(fp); 591 goto retry; 592 } 593 new = newfd; 594 delfp = NULL; 595 } else { 596 if (fdp->fd_files[new].reserved) { 597 spin_unlock(&fdp->fd_spin); 598 fdrop(fp); 599 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new); 600 tsleep(fdp, 0, "fdres", hz); 601 goto retry; 602 } 603 604 /* 605 * If the target descriptor was never allocated we have 606 * to allocate it. If it was we have to clean out the 607 * old descriptor. delfp inherits the ref from the 608 * descriptor table. 609 */ 610 delfp = fdp->fd_files[new].fp; 611 fdp->fd_files[new].fp = NULL; 612 fdp->fd_files[new].reserved = 1; 613 if (delfp == NULL) { 614 fdreserve_locked(fdp, new, 1); 615 if (new > fdp->fd_lastfile) 616 fdp->fd_lastfile = new; 617 } 618 619 } 620 621 /* 622 * NOTE: still holding an exclusive spinlock 623 */ 624 625 /* 626 * If a descriptor is being overwritten we may hve to tell 627 * fdfree() to sleep to ensure that all relevant process 628 * leaders can be traversed in closef(). 629 */ 630 if (delfp != NULL && p->p_fdtol != NULL) { 631 fdp->fd_holdleaderscount++; 632 holdleaders = 1; 633 } else { 634 holdleaders = 0; 635 } 636 KASSERT(delfp == NULL || (flags & DUP_FIXED), 637 ("dup() picked an open file")); 638 639 /* 640 * Duplicate the source descriptor, update lastfile. If the new 641 * descriptor was not allocated and we aren't replacing an existing 642 * descriptor we have to mark the descriptor as being in use. 643 * 644 * The fd_files[] array inherits fp's hold reference. 645 */ 646 fsetfd_locked(fdp, fp, new); 647 if ((flags & DUP_CLOEXEC) != 0) 648 fdp->fd_files[new].fileflags = oldflags | UF_EXCLOSE; 649 else 650 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE; 651 spin_unlock(&fdp->fd_spin); 652 fdrop(fp); 653 *res = new; 654 655 /* 656 * If we dup'd over a valid file, we now own the reference to it 657 * and must dispose of it using closef() semantics (as if a 658 * close() were performed on it). 659 */ 660 if (delfp) { 661 if (SLIST_FIRST(&delfp->f_klist)) 662 knote_fdclose(delfp, fdp, new); 663 closef(delfp, p); 664 if (holdleaders) { 665 spin_lock(&fdp->fd_spin); 666 fdp->fd_holdleaderscount--; 667 if (fdp->fd_holdleaderscount == 0 && 668 fdp->fd_holdleaderswakeup != 0) { 669 fdp->fd_holdleaderswakeup = 0; 670 spin_unlock(&fdp->fd_spin); 671 wakeup(&fdp->fd_holdleaderscount); 672 } else { 673 spin_unlock(&fdp->fd_spin); 674 } 675 } 676 } 677 return (0); 678 } 679 680 /* 681 * If sigio is on the list associated with a process or process group, 682 * disable signalling from the device, remove sigio from the list and 683 * free sigio. 684 */ 685 void 686 funsetown(struct sigio **sigiop) 687 { 688 struct pgrp *pgrp; 689 struct proc *p; 690 struct sigio *sigio; 691 692 if ((sigio = *sigiop) != NULL) { 693 lwkt_gettoken(&sigio_token); /* protect sigio */ 694 KKASSERT(sigiop == sigio->sio_myref); 695 sigio = *sigiop; 696 *sigiop = NULL; 697 lwkt_reltoken(&sigio_token); 698 } 699 if (sigio == NULL) 700 return; 701 702 if (sigio->sio_pgid < 0) { 703 pgrp = sigio->sio_pgrp; 704 sigio->sio_pgrp = NULL; 705 lwkt_gettoken(&pgrp->pg_token); 706 SLIST_REMOVE(&pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio); 707 lwkt_reltoken(&pgrp->pg_token); 708 pgrel(pgrp); 709 } else /* if ((*sigiop)->sio_pgid > 0) */ { 710 p = sigio->sio_proc; 711 sigio->sio_proc = NULL; 712 PHOLD(p); 713 lwkt_gettoken(&p->p_token); 714 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); 715 lwkt_reltoken(&p->p_token); 716 PRELE(p); 717 } 718 crfree(sigio->sio_ucred); 719 sigio->sio_ucred = NULL; 720 kfree(sigio, M_SIGIO); 721 } 722 723 /* 724 * Free a list of sigio structures. Caller is responsible for ensuring 725 * that the list is MPSAFE. 726 */ 727 void 728 funsetownlst(struct sigiolst *sigiolst) 729 { 730 struct sigio *sigio; 731 732 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) 733 funsetown(sigio->sio_myref); 734 } 735 736 /* 737 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 738 * 739 * After permission checking, add a sigio structure to the sigio list for 740 * the process or process group. 741 */ 742 int 743 fsetown(pid_t pgid, struct sigio **sigiop) 744 { 745 struct proc *proc = NULL; 746 struct pgrp *pgrp = NULL; 747 struct sigio *sigio; 748 int error; 749 750 if (pgid == 0) { 751 funsetown(sigiop); 752 return (0); 753 } 754 755 if (pgid > 0) { 756 proc = pfind(pgid); 757 if (proc == NULL) { 758 error = ESRCH; 759 goto done; 760 } 761 762 /* 763 * Policy - Don't allow a process to FSETOWN a process 764 * in another session. 765 * 766 * Remove this test to allow maximum flexibility or 767 * restrict FSETOWN to the current process or process 768 * group for maximum safety. 769 */ 770 if (proc->p_session != curproc->p_session) { 771 error = EPERM; 772 goto done; 773 } 774 } else /* if (pgid < 0) */ { 775 pgrp = pgfind(-pgid); 776 if (pgrp == NULL) { 777 error = ESRCH; 778 goto done; 779 } 780 781 /* 782 * Policy - Don't allow a process to FSETOWN a process 783 * in another session. 784 * 785 * Remove this test to allow maximum flexibility or 786 * restrict FSETOWN to the current process or process 787 * group for maximum safety. 788 */ 789 if (pgrp->pg_session != curproc->p_session) { 790 error = EPERM; 791 goto done; 792 } 793 } 794 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK | M_ZERO); 795 if (pgid > 0) { 796 KKASSERT(pgrp == NULL); 797 lwkt_gettoken(&proc->p_token); 798 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 799 sigio->sio_proc = proc; 800 lwkt_reltoken(&proc->p_token); 801 } else { 802 KKASSERT(proc == NULL); 803 lwkt_gettoken(&pgrp->pg_token); 804 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 805 sigio->sio_pgrp = pgrp; 806 lwkt_reltoken(&pgrp->pg_token); 807 pgrp = NULL; 808 } 809 sigio->sio_pgid = pgid; 810 sigio->sio_ucred = crhold(curthread->td_ucred); 811 /* It would be convenient if p_ruid was in ucred. */ 812 sigio->sio_ruid = sigio->sio_ucred->cr_ruid; 813 sigio->sio_myref = sigiop; 814 815 lwkt_gettoken(&sigio_token); 816 while (*sigiop) 817 funsetown(sigiop); 818 *sigiop = sigio; 819 lwkt_reltoken(&sigio_token); 820 error = 0; 821 done: 822 if (pgrp) 823 pgrel(pgrp); 824 if (proc) 825 PRELE(proc); 826 return (error); 827 } 828 829 /* 830 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 831 */ 832 pid_t 833 fgetown(struct sigio **sigiop) 834 { 835 struct sigio *sigio; 836 pid_t own; 837 838 lwkt_gettoken_shared(&sigio_token); 839 sigio = *sigiop; 840 own = (sigio != NULL ? sigio->sio_pgid : 0); 841 lwkt_reltoken(&sigio_token); 842 843 return (own); 844 } 845 846 /* 847 * Close many file descriptors. 848 */ 849 int 850 sys_closefrom(struct closefrom_args *uap) 851 { 852 return(kern_closefrom(uap->fd)); 853 } 854 855 /* 856 * Close all file descriptors greater then or equal to fd 857 */ 858 int 859 kern_closefrom(int fd) 860 { 861 struct thread *td = curthread; 862 struct proc *p = td->td_proc; 863 struct filedesc *fdp; 864 865 KKASSERT(p); 866 fdp = p->p_fd; 867 868 if (fd < 0) 869 return (EINVAL); 870 871 /* 872 * NOTE: This function will skip unassociated descriptors and 873 * reserved descriptors that have not yet been assigned. 874 * fd_lastfile can change as a side effect of kern_close(). 875 */ 876 spin_lock(&fdp->fd_spin); 877 while (fd <= fdp->fd_lastfile) { 878 if (fdp->fd_files[fd].fp != NULL) { 879 spin_unlock(&fdp->fd_spin); 880 /* ok if this races another close */ 881 if (kern_close(fd) == EINTR) 882 return (EINTR); 883 spin_lock(&fdp->fd_spin); 884 } 885 ++fd; 886 } 887 spin_unlock(&fdp->fd_spin); 888 return (0); 889 } 890 891 /* 892 * Close a file descriptor. 893 */ 894 int 895 sys_close(struct close_args *uap) 896 { 897 return(kern_close(uap->fd)); 898 } 899 900 /* 901 * close() helper 902 */ 903 int 904 kern_close(int fd) 905 { 906 struct thread *td = curthread; 907 struct proc *p = td->td_proc; 908 struct filedesc *fdp; 909 struct file *fp; 910 int error; 911 int holdleaders; 912 913 KKASSERT(p); 914 fdp = p->p_fd; 915 916 spin_lock(&fdp->fd_spin); 917 if ((fp = funsetfd_locked(fdp, fd)) == NULL) { 918 spin_unlock(&fdp->fd_spin); 919 return (EBADF); 920 } 921 holdleaders = 0; 922 if (p->p_fdtol != NULL) { 923 /* 924 * Ask fdfree() to sleep to ensure that all relevant 925 * process leaders can be traversed in closef(). 926 */ 927 fdp->fd_holdleaderscount++; 928 holdleaders = 1; 929 } 930 931 /* 932 * we now hold the fp reference that used to be owned by the descriptor 933 * array. 934 */ 935 spin_unlock(&fdp->fd_spin); 936 if (SLIST_FIRST(&fp->f_klist)) 937 knote_fdclose(fp, fdp, fd); 938 error = closef(fp, p); 939 if (holdleaders) { 940 spin_lock(&fdp->fd_spin); 941 fdp->fd_holdleaderscount--; 942 if (fdp->fd_holdleaderscount == 0 && 943 fdp->fd_holdleaderswakeup != 0) { 944 fdp->fd_holdleaderswakeup = 0; 945 spin_unlock(&fdp->fd_spin); 946 wakeup(&fdp->fd_holdleaderscount); 947 } else { 948 spin_unlock(&fdp->fd_spin); 949 } 950 } 951 return (error); 952 } 953 954 /* 955 * shutdown_args(int fd, int how) 956 */ 957 int 958 kern_shutdown(int fd, int how) 959 { 960 struct thread *td = curthread; 961 struct proc *p = td->td_proc; 962 struct file *fp; 963 int error; 964 965 KKASSERT(p); 966 967 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 968 return (EBADF); 969 error = fo_shutdown(fp, how); 970 fdrop(fp); 971 972 return (error); 973 } 974 975 /* 976 * MPALMOSTSAFE 977 */ 978 int 979 sys_shutdown(struct shutdown_args *uap) 980 { 981 int error; 982 983 error = kern_shutdown(uap->s, uap->how); 984 985 return (error); 986 } 987 988 /* 989 * fstat() helper 990 */ 991 int 992 kern_fstat(int fd, struct stat *ub) 993 { 994 struct thread *td = curthread; 995 struct proc *p = td->td_proc; 996 struct file *fp; 997 int error; 998 999 KKASSERT(p); 1000 1001 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 1002 return (EBADF); 1003 error = fo_stat(fp, ub, td->td_ucred); 1004 fdrop(fp); 1005 1006 return (error); 1007 } 1008 1009 /* 1010 * Return status information about a file descriptor. 1011 */ 1012 int 1013 sys_fstat(struct fstat_args *uap) 1014 { 1015 struct stat st; 1016 int error; 1017 1018 error = kern_fstat(uap->fd, &st); 1019 1020 if (error == 0) 1021 error = copyout(&st, uap->sb, sizeof(st)); 1022 return (error); 1023 } 1024 1025 /* 1026 * Return pathconf information about a file descriptor. 1027 * 1028 * MPALMOSTSAFE 1029 */ 1030 int 1031 sys_fpathconf(struct fpathconf_args *uap) 1032 { 1033 struct thread *td = curthread; 1034 struct proc *p = td->td_proc; 1035 struct file *fp; 1036 struct vnode *vp; 1037 int error = 0; 1038 1039 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 1040 return (EBADF); 1041 1042 switch (fp->f_type) { 1043 case DTYPE_PIPE: 1044 case DTYPE_SOCKET: 1045 if (uap->name != _PC_PIPE_BUF) { 1046 error = EINVAL; 1047 } else { 1048 uap->sysmsg_result = PIPE_BUF; 1049 error = 0; 1050 } 1051 break; 1052 case DTYPE_FIFO: 1053 case DTYPE_VNODE: 1054 vp = (struct vnode *)fp->f_data; 1055 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg); 1056 break; 1057 default: 1058 error = EOPNOTSUPP; 1059 break; 1060 } 1061 fdrop(fp); 1062 return(error); 1063 } 1064 1065 static int fdexpand; 1066 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, 1067 "Number of times a file table has been expanded"); 1068 1069 /* 1070 * Grow the file table so it can hold through descriptor (want). 1071 * 1072 * The fdp's spinlock must be held exclusively on entry and may be held 1073 * exclusively on return. The spinlock may be cycled by the routine. 1074 */ 1075 static void 1076 fdgrow_locked(struct filedesc *fdp, int want) 1077 { 1078 struct fdnode *newfiles; 1079 struct fdnode *oldfiles; 1080 int nf, extra; 1081 1082 nf = fdp->fd_nfiles; 1083 do { 1084 /* nf has to be of the form 2^n - 1 */ 1085 nf = 2 * nf + 1; 1086 } while (nf <= want); 1087 1088 spin_unlock(&fdp->fd_spin); 1089 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK); 1090 spin_lock(&fdp->fd_spin); 1091 1092 /* 1093 * We could have raced another extend while we were not holding 1094 * the spinlock. 1095 */ 1096 if (fdp->fd_nfiles >= nf) { 1097 spin_unlock(&fdp->fd_spin); 1098 kfree(newfiles, M_FILEDESC); 1099 spin_lock(&fdp->fd_spin); 1100 return; 1101 } 1102 /* 1103 * Copy the existing ofile and ofileflags arrays 1104 * and zero the new portion of each array. 1105 */ 1106 extra = nf - fdp->fd_nfiles; 1107 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode)); 1108 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode)); 1109 1110 oldfiles = fdp->fd_files; 1111 fdp->fd_files = newfiles; 1112 fdp->fd_nfiles = nf; 1113 1114 if (oldfiles != fdp->fd_builtin_files) { 1115 spin_unlock(&fdp->fd_spin); 1116 kfree(oldfiles, M_FILEDESC); 1117 spin_lock(&fdp->fd_spin); 1118 } 1119 fdexpand++; 1120 } 1121 1122 /* 1123 * Number of nodes in right subtree, including the root. 1124 */ 1125 static __inline int 1126 right_subtree_size(int n) 1127 { 1128 return (n ^ (n | (n + 1))); 1129 } 1130 1131 /* 1132 * Bigger ancestor. 1133 */ 1134 static __inline int 1135 right_ancestor(int n) 1136 { 1137 return (n | (n + 1)); 1138 } 1139 1140 /* 1141 * Smaller ancestor. 1142 */ 1143 static __inline int 1144 left_ancestor(int n) 1145 { 1146 return ((n & (n + 1)) - 1); 1147 } 1148 1149 /* 1150 * Traverse the in-place binary tree buttom-up adjusting the allocation 1151 * count so scans can determine where free descriptors are located. 1152 * 1153 * caller must be holding an exclusive spinlock on fdp 1154 */ 1155 static 1156 void 1157 fdreserve_locked(struct filedesc *fdp, int fd, int incr) 1158 { 1159 while (fd >= 0) { 1160 fdp->fd_files[fd].allocated += incr; 1161 KKASSERT(fdp->fd_files[fd].allocated >= 0); 1162 fd = left_ancestor(fd); 1163 } 1164 } 1165 1166 /* 1167 * Reserve a file descriptor for the process. If no error occurs, the 1168 * caller MUST at some point call fsetfd() or assign a file pointer 1169 * or dispose of the reservation. 1170 */ 1171 int 1172 fdalloc(struct proc *p, int want, int *result) 1173 { 1174 struct filedesc *fdp = p->p_fd; 1175 struct uidinfo *uip; 1176 int fd, rsize, rsum, node, lim; 1177 1178 /* 1179 * Check dtable size limit 1180 */ 1181 spin_lock(&p->p_limit->p_spin); 1182 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 1183 lim = INT_MAX; 1184 else 1185 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 1186 spin_unlock(&p->p_limit->p_spin); 1187 1188 if (lim > maxfilesperproc) 1189 lim = maxfilesperproc; 1190 if (lim < minfilesperproc) 1191 lim = minfilesperproc; 1192 if (want >= lim) 1193 return (EMFILE); 1194 1195 /* 1196 * Check that the user has not run out of descriptors (non-root only). 1197 * As a safety measure the dtable is allowed to have at least 1198 * minfilesperproc open fds regardless of the maxfilesperuser limit. 1199 */ 1200 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) { 1201 uip = p->p_ucred->cr_uidinfo; 1202 if (uip->ui_openfiles > maxfilesperuser) { 1203 krateprintf(&krate_uidinfo, 1204 "Warning: user %d pid %d (%s) ran out of " 1205 "file descriptors (%d/%d)\n", 1206 p->p_ucred->cr_uid, (int)p->p_pid, 1207 p->p_comm, 1208 uip->ui_openfiles, maxfilesperuser); 1209 return(ENFILE); 1210 } 1211 } 1212 1213 /* 1214 * Grow the dtable if necessary 1215 */ 1216 spin_lock(&fdp->fd_spin); 1217 if (want >= fdp->fd_nfiles) 1218 fdgrow_locked(fdp, want); 1219 1220 /* 1221 * Search for a free descriptor starting at the higher 1222 * of want or fd_freefile. If that fails, consider 1223 * expanding the ofile array. 1224 * 1225 * NOTE! the 'allocated' field is a cumulative recursive allocation 1226 * count. If we happen to see a value of 0 then we can shortcut 1227 * our search. Otherwise we run through through the tree going 1228 * down branches we know have free descriptor(s) until we hit a 1229 * leaf node. The leaf node will be free but will not necessarily 1230 * have an allocated field of 0. 1231 */ 1232 retry: 1233 /* move up the tree looking for a subtree with a free node */ 1234 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim); 1235 fd = right_ancestor(fd)) { 1236 if (fdp->fd_files[fd].allocated == 0) 1237 goto found; 1238 1239 rsize = right_subtree_size(fd); 1240 if (fdp->fd_files[fd].allocated == rsize) 1241 continue; /* right subtree full */ 1242 1243 /* 1244 * Free fd is in the right subtree of the tree rooted at fd. 1245 * Call that subtree R. Look for the smallest (leftmost) 1246 * subtree of R with an unallocated fd: continue moving 1247 * down the left branch until encountering a full left 1248 * subtree, then move to the right. 1249 */ 1250 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) { 1251 node = fd + rsize; 1252 rsum += fdp->fd_files[node].allocated; 1253 if (fdp->fd_files[fd].allocated == rsum + rsize) { 1254 fd = node; /* move to the right */ 1255 if (fdp->fd_files[node].allocated == 0) 1256 goto found; 1257 rsum = 0; 1258 } 1259 } 1260 goto found; 1261 } 1262 1263 /* 1264 * No space in current array. Expand? 1265 */ 1266 if (fdp->fd_nfiles >= lim) { 1267 spin_unlock(&fdp->fd_spin); 1268 return (EMFILE); 1269 } 1270 fdgrow_locked(fdp, want); 1271 goto retry; 1272 1273 found: 1274 KKASSERT(fd < fdp->fd_nfiles); 1275 if (fd > fdp->fd_lastfile) 1276 fdp->fd_lastfile = fd; 1277 if (want <= fdp->fd_freefile) 1278 fdp->fd_freefile = fd; 1279 *result = fd; 1280 KKASSERT(fdp->fd_files[fd].fp == NULL); 1281 KKASSERT(fdp->fd_files[fd].reserved == 0); 1282 fdp->fd_files[fd].fileflags = 0; 1283 fdp->fd_files[fd].reserved = 1; 1284 fdreserve_locked(fdp, fd, 1); 1285 spin_unlock(&fdp->fd_spin); 1286 return (0); 1287 } 1288 1289 /* 1290 * Check to see whether n user file descriptors 1291 * are available to the process p. 1292 */ 1293 int 1294 fdavail(struct proc *p, int n) 1295 { 1296 struct filedesc *fdp = p->p_fd; 1297 struct fdnode *fdnode; 1298 int i, lim, last; 1299 1300 spin_lock(&p->p_limit->p_spin); 1301 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 1302 lim = INT_MAX; 1303 else 1304 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 1305 spin_unlock(&p->p_limit->p_spin); 1306 1307 if (lim > maxfilesperproc) 1308 lim = maxfilesperproc; 1309 if (lim < minfilesperproc) 1310 lim = minfilesperproc; 1311 1312 spin_lock(&fdp->fd_spin); 1313 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) { 1314 spin_unlock(&fdp->fd_spin); 1315 return (1); 1316 } 1317 last = min(fdp->fd_nfiles, lim); 1318 fdnode = &fdp->fd_files[fdp->fd_freefile]; 1319 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) { 1320 if (fdnode->fp == NULL && --n <= 0) { 1321 spin_unlock(&fdp->fd_spin); 1322 return (1); 1323 } 1324 } 1325 spin_unlock(&fdp->fd_spin); 1326 return (0); 1327 } 1328 1329 /* 1330 * Revoke open descriptors referencing (f_data, f_type) 1331 * 1332 * Any revoke executed within a prison is only able to 1333 * revoke descriptors for processes within that prison. 1334 * 1335 * Returns 0 on success or an error code. 1336 */ 1337 struct fdrevoke_info { 1338 void *data; 1339 short type; 1340 short unused; 1341 int count; 1342 int intransit; 1343 struct ucred *cred; 1344 struct file *nfp; 1345 }; 1346 1347 static int fdrevoke_check_callback(struct file *fp, void *vinfo); 1348 static int fdrevoke_proc_callback(struct proc *p, void *vinfo); 1349 1350 int 1351 fdrevoke(void *f_data, short f_type, struct ucred *cred) 1352 { 1353 struct fdrevoke_info info; 1354 int error; 1355 1356 bzero(&info, sizeof(info)); 1357 info.data = f_data; 1358 info.type = f_type; 1359 info.cred = cred; 1360 error = falloc(NULL, &info.nfp, NULL); 1361 if (error) 1362 return (error); 1363 1364 /* 1365 * Scan the file pointer table once. dups do not dup file pointers, 1366 * only descriptors, so there is no leak. Set FREVOKED on the fps 1367 * being revoked. 1368 */ 1369 allfiles_scan_exclusive(fdrevoke_check_callback, &info); 1370 1371 /* 1372 * If any fps were marked track down the related descriptors 1373 * and close them. Any dup()s at this point will notice 1374 * the FREVOKED already set in the fp and do the right thing. 1375 * 1376 * Any fps with non-zero msgcounts (aka sent over a unix-domain 1377 * socket) bumped the intransit counter and will require a 1378 * scan. Races against fps leaving the socket are closed by 1379 * the socket code checking for FREVOKED. 1380 */ 1381 if (info.count) 1382 allproc_scan(fdrevoke_proc_callback, &info); 1383 if (info.intransit) 1384 unp_revoke_gc(info.nfp); 1385 fdrop(info.nfp); 1386 return(0); 1387 } 1388 1389 /* 1390 * Locate matching file pointers directly. 1391 * 1392 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls! 1393 */ 1394 static int 1395 fdrevoke_check_callback(struct file *fp, void *vinfo) 1396 { 1397 struct fdrevoke_info *info = vinfo; 1398 1399 /* 1400 * File pointers already flagged for revokation are skipped. 1401 */ 1402 if (fp->f_flag & FREVOKED) 1403 return(0); 1404 1405 /* 1406 * If revoking from a prison file pointers created outside of 1407 * that prison, or file pointers without creds, cannot be revoked. 1408 */ 1409 if (info->cred->cr_prison && 1410 (fp->f_cred == NULL || 1411 info->cred->cr_prison != fp->f_cred->cr_prison)) { 1412 return(0); 1413 } 1414 1415 /* 1416 * If the file pointer matches then mark it for revocation. The 1417 * flag is currently only used by unp_revoke_gc(). 1418 * 1419 * info->count is a heuristic and can race in a SMP environment. 1420 */ 1421 if (info->data == fp->f_data && info->type == fp->f_type) { 1422 atomic_set_int(&fp->f_flag, FREVOKED); 1423 info->count += fp->f_count; 1424 if (fp->f_msgcount) 1425 ++info->intransit; 1426 } 1427 return(0); 1428 } 1429 1430 /* 1431 * Locate matching file pointers via process descriptor tables. 1432 */ 1433 static int 1434 fdrevoke_proc_callback(struct proc *p, void *vinfo) 1435 { 1436 struct fdrevoke_info *info = vinfo; 1437 struct filedesc *fdp; 1438 struct file *fp; 1439 int n; 1440 1441 if (p->p_stat == SIDL || p->p_stat == SZOMB) 1442 return(0); 1443 if (info->cred->cr_prison && 1444 info->cred->cr_prison != p->p_ucred->cr_prison) { 1445 return(0); 1446 } 1447 1448 /* 1449 * If the controlling terminal of the process matches the 1450 * vnode being revoked we clear the controlling terminal. 1451 * 1452 * The normal spec_close() may not catch this because it 1453 * uses curproc instead of p. 1454 */ 1455 if (p->p_session && info->type == DTYPE_VNODE && 1456 info->data == p->p_session->s_ttyvp) { 1457 p->p_session->s_ttyvp = NULL; 1458 vrele(info->data); 1459 } 1460 1461 /* 1462 * Softref the fdp to prevent it from being destroyed 1463 */ 1464 spin_lock(&p->p_spin); 1465 if ((fdp = p->p_fd) == NULL) { 1466 spin_unlock(&p->p_spin); 1467 return(0); 1468 } 1469 atomic_add_int(&fdp->fd_softrefs, 1); 1470 spin_unlock(&p->p_spin); 1471 1472 /* 1473 * Locate and close any matching file descriptors. 1474 */ 1475 spin_lock(&fdp->fd_spin); 1476 for (n = 0; n < fdp->fd_nfiles; ++n) { 1477 if ((fp = fdp->fd_files[n].fp) == NULL) 1478 continue; 1479 if (fp->f_flag & FREVOKED) { 1480 fhold(info->nfp); 1481 fdp->fd_files[n].fp = info->nfp; 1482 spin_unlock(&fdp->fd_spin); 1483 knote_fdclose(fp, fdp, n); /* XXX */ 1484 closef(fp, p); 1485 spin_lock(&fdp->fd_spin); 1486 --info->count; 1487 } 1488 } 1489 spin_unlock(&fdp->fd_spin); 1490 atomic_subtract_int(&fdp->fd_softrefs, 1); 1491 return(0); 1492 } 1493 1494 /* 1495 * falloc: 1496 * Create a new open file structure and reserve a file decriptor 1497 * for the process that refers to it. 1498 * 1499 * Root creds are checked using lp, or assumed if lp is NULL. If 1500 * resultfd is non-NULL then lp must also be non-NULL. No file 1501 * descriptor is reserved (and no process context is needed) if 1502 * resultfd is NULL. 1503 * 1504 * A file pointer with a refcount of 1 is returned. Note that the 1505 * file pointer is NOT associated with the descriptor. If falloc 1506 * returns success, fsetfd() MUST be called to either associate the 1507 * file pointer or clear the reservation. 1508 */ 1509 int 1510 falloc(struct lwp *lp, struct file **resultfp, int *resultfd) 1511 { 1512 static struct timeval lastfail; 1513 static int curfail; 1514 struct file *fp; 1515 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred; 1516 int error; 1517 1518 fp = NULL; 1519 1520 /* 1521 * Handle filetable full issues and root overfill. 1522 */ 1523 if (nfiles >= maxfiles - maxfilesrootres && 1524 (cred->cr_ruid != 0 || nfiles >= maxfiles)) { 1525 if (ppsratecheck(&lastfail, &curfail, 1)) { 1526 kprintf("kern.maxfiles limit exceeded by uid %d, " 1527 "please see tuning(7).\n", 1528 cred->cr_ruid); 1529 } 1530 error = ENFILE; 1531 goto done; 1532 } 1533 1534 /* 1535 * Allocate a new file descriptor. 1536 */ 1537 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO); 1538 spin_init(&fp->f_spin, "falloc"); 1539 SLIST_INIT(&fp->f_klist); 1540 fp->f_count = 1; 1541 fp->f_ops = &badfileops; 1542 fp->f_seqcount = 1; 1543 fsetcred(fp, cred); 1544 spin_lock(&filehead_spin); 1545 nfiles++; 1546 LIST_INSERT_HEAD(&filehead, fp, f_list); 1547 spin_unlock(&filehead_spin); 1548 if (resultfd) { 1549 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) { 1550 fdrop(fp); 1551 fp = NULL; 1552 } 1553 } else { 1554 error = 0; 1555 } 1556 done: 1557 *resultfp = fp; 1558 return (error); 1559 } 1560 1561 /* 1562 * Check for races against a file descriptor by determining that the 1563 * file pointer is still associated with the specified file descriptor, 1564 * and a close is not currently in progress. 1565 */ 1566 int 1567 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp) 1568 { 1569 int error; 1570 1571 spin_lock_shared(&fdp->fd_spin); 1572 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp) 1573 error = EBADF; 1574 else 1575 error = 0; 1576 spin_unlock_shared(&fdp->fd_spin); 1577 return (error); 1578 } 1579 1580 /* 1581 * Associate a file pointer with a previously reserved file descriptor. 1582 * This function always succeeds. 1583 * 1584 * If fp is NULL, the file descriptor is returned to the pool. 1585 */ 1586 1587 /* 1588 * (exclusive spinlock must be held on call) 1589 */ 1590 static void 1591 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd) 1592 { 1593 KKASSERT((unsigned)fd < fdp->fd_nfiles); 1594 KKASSERT(fdp->fd_files[fd].reserved != 0); 1595 if (fp) { 1596 fhold(fp); 1597 fdp->fd_files[fd].fp = fp; 1598 fdp->fd_files[fd].reserved = 0; 1599 } else { 1600 fdp->fd_files[fd].reserved = 0; 1601 fdreserve_locked(fdp, fd, -1); 1602 fdfixup_locked(fdp, fd); 1603 } 1604 } 1605 1606 void 1607 fsetfd(struct filedesc *fdp, struct file *fp, int fd) 1608 { 1609 spin_lock(&fdp->fd_spin); 1610 fsetfd_locked(fdp, fp, fd); 1611 spin_unlock(&fdp->fd_spin); 1612 } 1613 1614 /* 1615 * (exclusive spinlock must be held on call) 1616 */ 1617 static 1618 struct file * 1619 funsetfd_locked(struct filedesc *fdp, int fd) 1620 { 1621 struct file *fp; 1622 1623 if ((unsigned)fd >= fdp->fd_nfiles) 1624 return (NULL); 1625 if ((fp = fdp->fd_files[fd].fp) == NULL) 1626 return (NULL); 1627 fdp->fd_files[fd].fp = NULL; 1628 fdp->fd_files[fd].fileflags = 0; 1629 1630 fdreserve_locked(fdp, fd, -1); 1631 fdfixup_locked(fdp, fd); 1632 return(fp); 1633 } 1634 1635 /* 1636 * WARNING: May not be called before initial fsetfd(). 1637 */ 1638 int 1639 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp) 1640 { 1641 int error; 1642 1643 spin_lock(&fdp->fd_spin); 1644 if (((u_int)fd) >= fdp->fd_nfiles) { 1645 error = EBADF; 1646 } else if (fdp->fd_files[fd].fp == NULL) { 1647 error = EBADF; 1648 } else { 1649 *flagsp = fdp->fd_files[fd].fileflags; 1650 error = 0; 1651 } 1652 spin_unlock(&fdp->fd_spin); 1653 return (error); 1654 } 1655 1656 /* 1657 * WARNING: May not be called before initial fsetfd(). 1658 */ 1659 int 1660 fsetfdflags(struct filedesc *fdp, int fd, int add_flags) 1661 { 1662 int error; 1663 1664 spin_lock(&fdp->fd_spin); 1665 if (((u_int)fd) >= fdp->fd_nfiles) { 1666 error = EBADF; 1667 } else if (fdp->fd_files[fd].fp == NULL) { 1668 error = EBADF; 1669 } else { 1670 fdp->fd_files[fd].fileflags |= add_flags; 1671 error = 0; 1672 } 1673 spin_unlock(&fdp->fd_spin); 1674 return (error); 1675 } 1676 1677 /* 1678 * WARNING: May not be called before initial fsetfd(). 1679 */ 1680 int 1681 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags) 1682 { 1683 int error; 1684 1685 spin_lock(&fdp->fd_spin); 1686 if (((u_int)fd) >= fdp->fd_nfiles) { 1687 error = EBADF; 1688 } else if (fdp->fd_files[fd].fp == NULL) { 1689 error = EBADF; 1690 } else { 1691 fdp->fd_files[fd].fileflags &= ~rem_flags; 1692 error = 0; 1693 } 1694 spin_unlock(&fdp->fd_spin); 1695 return (error); 1696 } 1697 1698 /* 1699 * Set/Change/Clear the creds for a fp and synchronize the uidinfo. 1700 */ 1701 void 1702 fsetcred(struct file *fp, struct ucred *ncr) 1703 { 1704 struct ucred *ocr; 1705 struct uidinfo *uip; 1706 1707 ocr = fp->f_cred; 1708 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) { 1709 if (ocr) { 1710 uip = ocr->cr_uidinfo; 1711 atomic_add_int(&uip->ui_openfiles, -1); 1712 } 1713 if (ncr) { 1714 uip = ncr->cr_uidinfo; 1715 atomic_add_int(&uip->ui_openfiles, 1); 1716 } 1717 } 1718 if (ncr) 1719 crhold(ncr); 1720 fp->f_cred = ncr; 1721 if (ocr) 1722 crfree(ocr); 1723 } 1724 1725 /* 1726 * Free a file descriptor. 1727 */ 1728 static 1729 void 1730 ffree(struct file *fp) 1731 { 1732 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 1733 spin_lock(&filehead_spin); 1734 LIST_REMOVE(fp, f_list); 1735 nfiles--; 1736 spin_unlock(&filehead_spin); 1737 fsetcred(fp, NULL); 1738 if (fp->f_nchandle.ncp) 1739 cache_drop(&fp->f_nchandle); 1740 kfree(fp, M_FILE); 1741 } 1742 1743 /* 1744 * called from init_main, initialize filedesc0 for proc0. 1745 */ 1746 void 1747 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask) 1748 { 1749 p0->p_fd = fdp0; 1750 p0->p_fdtol = NULL; 1751 fdp0->fd_refcnt = 1; 1752 fdp0->fd_cmask = cmask; 1753 fdp0->fd_files = fdp0->fd_builtin_files; 1754 fdp0->fd_nfiles = NDFILE; 1755 fdp0->fd_lastfile = -1; 1756 spin_init(&fdp0->fd_spin, "fdinitbootstrap"); 1757 } 1758 1759 /* 1760 * Build a new filedesc structure. 1761 */ 1762 struct filedesc * 1763 fdinit(struct proc *p) 1764 { 1765 struct filedesc *newfdp; 1766 struct filedesc *fdp = p->p_fd; 1767 1768 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO); 1769 spin_lock(&fdp->fd_spin); 1770 if (fdp->fd_cdir) { 1771 newfdp->fd_cdir = fdp->fd_cdir; 1772 vref(newfdp->fd_cdir); 1773 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1774 } 1775 1776 /* 1777 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of 1778 * proc0, but should unconditionally exist in other processes. 1779 */ 1780 if (fdp->fd_rdir) { 1781 newfdp->fd_rdir = fdp->fd_rdir; 1782 vref(newfdp->fd_rdir); 1783 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1784 } 1785 if (fdp->fd_jdir) { 1786 newfdp->fd_jdir = fdp->fd_jdir; 1787 vref(newfdp->fd_jdir); 1788 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1789 } 1790 spin_unlock(&fdp->fd_spin); 1791 1792 /* Create the file descriptor table. */ 1793 newfdp->fd_refcnt = 1; 1794 newfdp->fd_cmask = cmask; 1795 newfdp->fd_files = newfdp->fd_builtin_files; 1796 newfdp->fd_nfiles = NDFILE; 1797 newfdp->fd_lastfile = -1; 1798 spin_init(&newfdp->fd_spin, "fdinit"); 1799 1800 return (newfdp); 1801 } 1802 1803 /* 1804 * Share a filedesc structure. 1805 */ 1806 struct filedesc * 1807 fdshare(struct proc *p) 1808 { 1809 struct filedesc *fdp; 1810 1811 fdp = p->p_fd; 1812 spin_lock(&fdp->fd_spin); 1813 fdp->fd_refcnt++; 1814 spin_unlock(&fdp->fd_spin); 1815 return (fdp); 1816 } 1817 1818 /* 1819 * Copy a filedesc structure. 1820 */ 1821 int 1822 fdcopy(struct proc *p, struct filedesc **fpp) 1823 { 1824 struct filedesc *fdp = p->p_fd; 1825 struct filedesc *newfdp; 1826 struct fdnode *fdnode; 1827 int i; 1828 int ni; 1829 1830 /* 1831 * Certain daemons might not have file descriptors. 1832 */ 1833 if (fdp == NULL) 1834 return (0); 1835 1836 /* 1837 * Allocate the new filedesc and fd_files[] array. This can race 1838 * with operations by other threads on the fdp so we have to be 1839 * careful. 1840 */ 1841 newfdp = kmalloc(sizeof(struct filedesc), 1842 M_FILEDESC, M_WAITOK | M_ZERO | M_NULLOK); 1843 if (newfdp == NULL) { 1844 *fpp = NULL; 1845 return (-1); 1846 } 1847 again: 1848 spin_lock(&fdp->fd_spin); 1849 if (fdp->fd_lastfile < NDFILE) { 1850 newfdp->fd_files = newfdp->fd_builtin_files; 1851 i = NDFILE; 1852 } else { 1853 /* 1854 * We have to allocate (N^2-1) entries for our in-place 1855 * binary tree. Allow the table to shrink. 1856 */ 1857 i = fdp->fd_nfiles; 1858 ni = (i - 1) / 2; 1859 while (ni > fdp->fd_lastfile && ni > NDFILE) { 1860 i = ni; 1861 ni = (i - 1) / 2; 1862 } 1863 spin_unlock(&fdp->fd_spin); 1864 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode), 1865 M_FILEDESC, M_WAITOK | M_ZERO); 1866 1867 /* 1868 * Check for race, retry 1869 */ 1870 spin_lock(&fdp->fd_spin); 1871 if (i <= fdp->fd_lastfile) { 1872 spin_unlock(&fdp->fd_spin); 1873 kfree(newfdp->fd_files, M_FILEDESC); 1874 goto again; 1875 } 1876 } 1877 1878 /* 1879 * Dup the remaining fields. vref() and cache_hold() can be 1880 * safely called while holding the read spinlock on fdp. 1881 * 1882 * The read spinlock on fdp is still being held. 1883 * 1884 * NOTE: vref and cache_hold calls for the case where the vnode 1885 * or cache entry already has at least one ref may be called 1886 * while holding spin locks. 1887 */ 1888 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) { 1889 vref(newfdp->fd_cdir); 1890 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1891 } 1892 /* 1893 * We must check for fd_rdir here, at least for now because 1894 * the init process is created before we have access to the 1895 * rootvode to take a reference to it. 1896 */ 1897 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) { 1898 vref(newfdp->fd_rdir); 1899 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1900 } 1901 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) { 1902 vref(newfdp->fd_jdir); 1903 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1904 } 1905 newfdp->fd_refcnt = 1; 1906 newfdp->fd_nfiles = i; 1907 newfdp->fd_lastfile = fdp->fd_lastfile; 1908 newfdp->fd_freefile = fdp->fd_freefile; 1909 newfdp->fd_cmask = fdp->fd_cmask; 1910 spin_init(&newfdp->fd_spin, "fdcopy"); 1911 1912 /* 1913 * Copy the descriptor table through (i). This also copies the 1914 * allocation state. Then go through and ref the file pointers 1915 * and clean up any KQ descriptors. 1916 * 1917 * kq descriptors cannot be copied. Since we haven't ref'd the 1918 * copied files yet we can ignore the return value from funsetfd(). 1919 * 1920 * The read spinlock on fdp is still being held. 1921 */ 1922 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode)); 1923 for (i = 0 ; i < newfdp->fd_nfiles; ++i) { 1924 fdnode = &newfdp->fd_files[i]; 1925 if (fdnode->reserved) { 1926 fdreserve_locked(newfdp, i, -1); 1927 fdnode->reserved = 0; 1928 fdfixup_locked(newfdp, i); 1929 } else if (fdnode->fp) { 1930 if (fdnode->fp->f_type == DTYPE_KQUEUE) { 1931 (void)funsetfd_locked(newfdp, i); 1932 } else { 1933 fhold(fdnode->fp); 1934 } 1935 } 1936 } 1937 spin_unlock(&fdp->fd_spin); 1938 *fpp = newfdp; 1939 return (0); 1940 } 1941 1942 /* 1943 * Release a filedesc structure. 1944 * 1945 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE) 1946 */ 1947 void 1948 fdfree(struct proc *p, struct filedesc *repl) 1949 { 1950 struct filedesc *fdp; 1951 struct fdnode *fdnode; 1952 int i; 1953 struct filedesc_to_leader *fdtol; 1954 struct file *fp; 1955 struct vnode *vp; 1956 struct flock lf; 1957 1958 /* 1959 * Certain daemons might not have file descriptors. 1960 */ 1961 fdp = p->p_fd; 1962 if (fdp == NULL) { 1963 p->p_fd = repl; 1964 return; 1965 } 1966 1967 /* 1968 * Severe messing around to follow. 1969 */ 1970 spin_lock(&fdp->fd_spin); 1971 1972 /* Check for special need to clear POSIX style locks */ 1973 fdtol = p->p_fdtol; 1974 if (fdtol != NULL) { 1975 KASSERT(fdtol->fdl_refcount > 0, 1976 ("filedesc_to_refcount botch: fdl_refcount=%d", 1977 fdtol->fdl_refcount)); 1978 if (fdtol->fdl_refcount == 1 && 1979 (p->p_leader->p_flags & P_ADVLOCK) != 0) { 1980 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1981 fdnode = &fdp->fd_files[i]; 1982 if (fdnode->fp == NULL || 1983 fdnode->fp->f_type != DTYPE_VNODE) { 1984 continue; 1985 } 1986 fp = fdnode->fp; 1987 fhold(fp); 1988 spin_unlock(&fdp->fd_spin); 1989 1990 lf.l_whence = SEEK_SET; 1991 lf.l_start = 0; 1992 lf.l_len = 0; 1993 lf.l_type = F_UNLCK; 1994 vp = (struct vnode *)fp->f_data; 1995 (void) VOP_ADVLOCK(vp, 1996 (caddr_t)p->p_leader, 1997 F_UNLCK, 1998 &lf, 1999 F_POSIX); 2000 fdrop(fp); 2001 spin_lock(&fdp->fd_spin); 2002 } 2003 } 2004 retry: 2005 if (fdtol->fdl_refcount == 1) { 2006 if (fdp->fd_holdleaderscount > 0 && 2007 (p->p_leader->p_flags & P_ADVLOCK) != 0) { 2008 /* 2009 * close() or do_dup() has cleared a reference 2010 * in a shared file descriptor table. 2011 */ 2012 fdp->fd_holdleaderswakeup = 1; 2013 ssleep(&fdp->fd_holdleaderscount, 2014 &fdp->fd_spin, 0, "fdlhold", 0); 2015 goto retry; 2016 } 2017 if (fdtol->fdl_holdcount > 0) { 2018 /* 2019 * Ensure that fdtol->fdl_leader 2020 * remains valid in closef(). 2021 */ 2022 fdtol->fdl_wakeup = 1; 2023 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0); 2024 goto retry; 2025 } 2026 } 2027 fdtol->fdl_refcount--; 2028 if (fdtol->fdl_refcount == 0 && 2029 fdtol->fdl_holdcount == 0) { 2030 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2031 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2032 } else { 2033 fdtol = NULL; 2034 } 2035 p->p_fdtol = NULL; 2036 if (fdtol != NULL) { 2037 spin_unlock(&fdp->fd_spin); 2038 kfree(fdtol, M_FILEDESC_TO_LEADER); 2039 spin_lock(&fdp->fd_spin); 2040 } 2041 } 2042 if (--fdp->fd_refcnt > 0) { 2043 spin_unlock(&fdp->fd_spin); 2044 spin_lock(&p->p_spin); 2045 p->p_fd = repl; 2046 spin_unlock(&p->p_spin); 2047 return; 2048 } 2049 2050 /* 2051 * Even though we are the last reference to the structure allproc 2052 * scans may still reference the structure. Maintain proper 2053 * locks until we can replace p->p_fd. 2054 * 2055 * Also note that kqueue's closef still needs to reference the 2056 * fdp via p->p_fd, so we have to close the descriptors before 2057 * we replace p->p_fd. 2058 */ 2059 for (i = 0; i <= fdp->fd_lastfile; ++i) { 2060 if (fdp->fd_files[i].fp) { 2061 fp = funsetfd_locked(fdp, i); 2062 if (fp) { 2063 spin_unlock(&fdp->fd_spin); 2064 if (SLIST_FIRST(&fp->f_klist)) 2065 knote_fdclose(fp, fdp, i); 2066 closef(fp, p); 2067 spin_lock(&fdp->fd_spin); 2068 } 2069 } 2070 } 2071 spin_unlock(&fdp->fd_spin); 2072 2073 /* 2074 * Interlock against an allproc scan operations (typically frevoke). 2075 */ 2076 spin_lock(&p->p_spin); 2077 p->p_fd = repl; 2078 spin_unlock(&p->p_spin); 2079 2080 /* 2081 * Wait for any softrefs to go away. This race rarely occurs so 2082 * we can use a non-critical-path style poll/sleep loop. The 2083 * race only occurs against allproc scans. 2084 * 2085 * No new softrefs can occur with the fdp disconnected from the 2086 * process. 2087 */ 2088 if (fdp->fd_softrefs) { 2089 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid); 2090 while (fdp->fd_softrefs) 2091 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1); 2092 } 2093 2094 if (fdp->fd_files != fdp->fd_builtin_files) 2095 kfree(fdp->fd_files, M_FILEDESC); 2096 if (fdp->fd_cdir) { 2097 cache_drop(&fdp->fd_ncdir); 2098 vrele(fdp->fd_cdir); 2099 } 2100 if (fdp->fd_rdir) { 2101 cache_drop(&fdp->fd_nrdir); 2102 vrele(fdp->fd_rdir); 2103 } 2104 if (fdp->fd_jdir) { 2105 cache_drop(&fdp->fd_njdir); 2106 vrele(fdp->fd_jdir); 2107 } 2108 kfree(fdp, M_FILEDESC); 2109 } 2110 2111 /* 2112 * Retrieve and reference the file pointer associated with a descriptor. 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 int 2142 holdsock(struct filedesc *fdp, int fd, struct file **fpp) 2143 { 2144 struct file *fp; 2145 int error; 2146 2147 spin_lock_shared(&fdp->fd_spin); 2148 if ((unsigned)fd >= fdp->fd_nfiles) { 2149 error = EBADF; 2150 fp = NULL; 2151 goto done; 2152 } 2153 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2154 error = EBADF; 2155 goto done; 2156 } 2157 if (fp->f_type != DTYPE_SOCKET) { 2158 error = ENOTSOCK; 2159 goto done; 2160 } 2161 fhold(fp); 2162 error = 0; 2163 done: 2164 spin_unlock_shared(&fdp->fd_spin); 2165 *fpp = fp; 2166 return (error); 2167 } 2168 2169 /* 2170 * Convert a user file descriptor to a held file pointer. 2171 */ 2172 int 2173 holdvnode(struct filedesc *fdp, int fd, struct file **fpp) 2174 { 2175 struct file *fp; 2176 int error; 2177 2178 spin_lock_shared(&fdp->fd_spin); 2179 if ((unsigned)fd >= fdp->fd_nfiles) { 2180 error = EBADF; 2181 fp = NULL; 2182 goto done; 2183 } 2184 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2185 error = EBADF; 2186 goto done; 2187 } 2188 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 2189 fp = NULL; 2190 error = EINVAL; 2191 goto done; 2192 } 2193 fhold(fp); 2194 error = 0; 2195 done: 2196 spin_unlock_shared(&fdp->fd_spin); 2197 *fpp = fp; 2198 return (error); 2199 } 2200 2201 /* 2202 * For setugid programs, we don't want to people to use that setugidness 2203 * to generate error messages which write to a file which otherwise would 2204 * otherwise be off-limits to the process. 2205 * 2206 * This is a gross hack to plug the hole. A better solution would involve 2207 * a special vop or other form of generalized access control mechanism. We 2208 * go ahead and just reject all procfs file systems accesses as dangerous. 2209 * 2210 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2211 * sufficient. We also don't for check setugidness since we know we are. 2212 */ 2213 static int 2214 is_unsafe(struct file *fp) 2215 { 2216 if (fp->f_type == DTYPE_VNODE && 2217 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 2218 return (1); 2219 return (0); 2220 } 2221 2222 /* 2223 * Make this setguid thing safe, if at all possible. 2224 * 2225 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2226 */ 2227 void 2228 setugidsafety(struct proc *p) 2229 { 2230 struct filedesc *fdp = p->p_fd; 2231 int i; 2232 2233 /* Certain daemons might not have file descriptors. */ 2234 if (fdp == NULL) 2235 return; 2236 2237 /* 2238 * note: fdp->fd_files may be reallocated out from under us while 2239 * we are blocked in a close. Be careful! 2240 */ 2241 for (i = 0; i <= fdp->fd_lastfile; i++) { 2242 if (i > 2) 2243 break; 2244 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) { 2245 struct file *fp; 2246 2247 /* 2248 * NULL-out descriptor prior to close to avoid 2249 * a race while close blocks. 2250 */ 2251 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2252 knote_fdclose(fp, fdp, i); 2253 closef(fp, p); 2254 } 2255 } 2256 } 2257 } 2258 2259 /* 2260 * Close any files on exec? 2261 * 2262 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2263 */ 2264 void 2265 fdcloseexec(struct proc *p) 2266 { 2267 struct filedesc *fdp = p->p_fd; 2268 int i; 2269 2270 /* Certain daemons might not have file descriptors. */ 2271 if (fdp == NULL) 2272 return; 2273 2274 /* 2275 * We cannot cache fd_files since operations may block and rip 2276 * them out from under us. 2277 */ 2278 for (i = 0; i <= fdp->fd_lastfile; i++) { 2279 if (fdp->fd_files[i].fp != NULL && 2280 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) { 2281 struct file *fp; 2282 2283 /* 2284 * NULL-out descriptor prior to close to avoid 2285 * a race while close blocks. 2286 */ 2287 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2288 knote_fdclose(fp, fdp, i); 2289 closef(fp, p); 2290 } 2291 } 2292 } 2293 } 2294 2295 /* 2296 * It is unsafe for set[ug]id processes to be started with file 2297 * descriptors 0..2 closed, as these descriptors are given implicit 2298 * significance in the Standard C library. fdcheckstd() will create a 2299 * descriptor referencing /dev/null for each of stdin, stdout, and 2300 * stderr that is not already open. 2301 * 2302 * NOT MPSAFE - calls falloc, vn_open, etc 2303 */ 2304 int 2305 fdcheckstd(struct lwp *lp) 2306 { 2307 struct nlookupdata nd; 2308 struct filedesc *fdp; 2309 struct file *fp; 2310 int retval; 2311 int i, error, flags, devnull; 2312 2313 fdp = lp->lwp_proc->p_fd; 2314 if (fdp == NULL) 2315 return (0); 2316 devnull = -1; 2317 error = 0; 2318 for (i = 0; i < 3; i++) { 2319 if (fdp->fd_files[i].fp != NULL) 2320 continue; 2321 if (devnull < 0) { 2322 if ((error = falloc(lp, &fp, &devnull)) != 0) 2323 break; 2324 2325 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, 2326 NLC_FOLLOW|NLC_LOCKVP); 2327 flags = FREAD | FWRITE; 2328 if (error == 0) 2329 error = vn_open(&nd, fp, flags, 0); 2330 if (error == 0) 2331 fsetfd(fdp, fp, devnull); 2332 else 2333 fsetfd(fdp, NULL, devnull); 2334 fdrop(fp); 2335 nlookup_done(&nd); 2336 if (error) 2337 break; 2338 KKASSERT(i == devnull); 2339 } else { 2340 error = kern_dup(DUP_FIXED, devnull, i, &retval); 2341 if (error != 0) 2342 break; 2343 } 2344 } 2345 return (error); 2346 } 2347 2348 /* 2349 * Internal form of close. 2350 * Decrement reference count on file structure. 2351 * Note: td and/or p may be NULL when closing a file 2352 * that was being passed in a message. 2353 * 2354 * MPALMOSTSAFE - acquires mplock for VOP operations 2355 */ 2356 int 2357 closef(struct file *fp, struct proc *p) 2358 { 2359 struct vnode *vp; 2360 struct flock lf; 2361 struct filedesc_to_leader *fdtol; 2362 2363 if (fp == NULL) 2364 return (0); 2365 2366 /* 2367 * POSIX record locking dictates that any close releases ALL 2368 * locks owned by this process. This is handled by setting 2369 * a flag in the unlock to free ONLY locks obeying POSIX 2370 * semantics, and not to free BSD-style file locks. 2371 * If the descriptor was in a message, POSIX-style locks 2372 * aren't passed with the descriptor. 2373 */ 2374 if (p != NULL && fp->f_type == DTYPE_VNODE && 2375 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2376 ) { 2377 if ((p->p_leader->p_flags & P_ADVLOCK) != 0) { 2378 lf.l_whence = SEEK_SET; 2379 lf.l_start = 0; 2380 lf.l_len = 0; 2381 lf.l_type = F_UNLCK; 2382 vp = (struct vnode *)fp->f_data; 2383 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 2384 &lf, F_POSIX); 2385 } 2386 fdtol = p->p_fdtol; 2387 if (fdtol != NULL) { 2388 lwkt_gettoken(&p->p_token); 2389 /* 2390 * Handle special case where file descriptor table 2391 * is shared between multiple process leaders. 2392 */ 2393 for (fdtol = fdtol->fdl_next; 2394 fdtol != p->p_fdtol; 2395 fdtol = fdtol->fdl_next) { 2396 if ((fdtol->fdl_leader->p_flags & 2397 P_ADVLOCK) == 0) 2398 continue; 2399 fdtol->fdl_holdcount++; 2400 lf.l_whence = SEEK_SET; 2401 lf.l_start = 0; 2402 lf.l_len = 0; 2403 lf.l_type = F_UNLCK; 2404 vp = (struct vnode *)fp->f_data; 2405 (void) VOP_ADVLOCK(vp, 2406 (caddr_t)fdtol->fdl_leader, 2407 F_UNLCK, &lf, F_POSIX); 2408 fdtol->fdl_holdcount--; 2409 if (fdtol->fdl_holdcount == 0 && 2410 fdtol->fdl_wakeup != 0) { 2411 fdtol->fdl_wakeup = 0; 2412 wakeup(fdtol); 2413 } 2414 } 2415 lwkt_reltoken(&p->p_token); 2416 } 2417 } 2418 return (fdrop(fp)); 2419 } 2420 2421 /* 2422 * fhold() can only be called if f_count is already at least 1 (i.e. the 2423 * caller of fhold() already has a reference to the file pointer in some 2424 * manner or other). 2425 * 2426 * f_count is not spin-locked. Instead, atomic ops are used for 2427 * incrementing, decrementing, and handling the 1->0 transition. 2428 */ 2429 void 2430 fhold(struct file *fp) 2431 { 2432 atomic_add_int(&fp->f_count, 1); 2433 } 2434 2435 /* 2436 * fdrop() - drop a reference to a descriptor 2437 * 2438 * MPALMOSTSAFE - acquires mplock for final close sequence 2439 */ 2440 int 2441 fdrop(struct file *fp) 2442 { 2443 struct flock lf; 2444 struct vnode *vp; 2445 int error; 2446 2447 /* 2448 * A combined fetch and subtract is needed to properly detect 2449 * 1->0 transitions, otherwise two cpus dropping from a ref 2450 * count of 2 might both try to run the 1->0 code. 2451 */ 2452 if (atomic_fetchadd_int(&fp->f_count, -1) > 1) 2453 return (0); 2454 2455 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL); 2456 2457 /* 2458 * The last reference has gone away, we own the fp structure free 2459 * and clear. 2460 */ 2461 if (fp->f_count < 0) 2462 panic("fdrop: count < 0"); 2463 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE && 2464 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2465 ) { 2466 lf.l_whence = SEEK_SET; 2467 lf.l_start = 0; 2468 lf.l_len = 0; 2469 lf.l_type = F_UNLCK; 2470 vp = (struct vnode *)fp->f_data; 2471 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2472 } 2473 if (fp->f_ops != &badfileops) 2474 error = fo_close(fp); 2475 else 2476 error = 0; 2477 ffree(fp); 2478 return (error); 2479 } 2480 2481 /* 2482 * Apply an advisory lock on a file descriptor. 2483 * 2484 * Just attempt to get a record lock of the requested type on 2485 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2486 * 2487 * MPALMOSTSAFE 2488 */ 2489 int 2490 sys_flock(struct flock_args *uap) 2491 { 2492 struct proc *p = curproc; 2493 struct file *fp; 2494 struct vnode *vp; 2495 struct flock lf; 2496 int error; 2497 2498 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 2499 return (EBADF); 2500 if (fp->f_type != DTYPE_VNODE) { 2501 error = EOPNOTSUPP; 2502 goto done; 2503 } 2504 vp = (struct vnode *)fp->f_data; 2505 lf.l_whence = SEEK_SET; 2506 lf.l_start = 0; 2507 lf.l_len = 0; 2508 if (uap->how & LOCK_UN) { 2509 lf.l_type = F_UNLCK; 2510 atomic_clear_int(&fp->f_flag, FHASLOCK); /* race ok */ 2511 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2512 goto done; 2513 } 2514 if (uap->how & LOCK_EX) 2515 lf.l_type = F_WRLCK; 2516 else if (uap->how & LOCK_SH) 2517 lf.l_type = F_RDLCK; 2518 else { 2519 error = EBADF; 2520 goto done; 2521 } 2522 if (uap->how & LOCK_NB) 2523 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0); 2524 else 2525 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT); 2526 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */ 2527 done: 2528 fdrop(fp); 2529 return (error); 2530 } 2531 2532 /* 2533 * File Descriptor pseudo-device driver (/dev/fd/). 2534 * 2535 * Opening minor device N dup()s the file (if any) connected to file 2536 * descriptor N belonging to the calling process. Note that this driver 2537 * consists of only the ``open()'' routine, because all subsequent 2538 * references to this file will be direct to the other driver. 2539 */ 2540 static int 2541 fdopen(struct dev_open_args *ap) 2542 { 2543 thread_t td = curthread; 2544 2545 KKASSERT(td->td_lwp != NULL); 2546 2547 /* 2548 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the 2549 * the file descriptor being sought for duplication. The error 2550 * return ensures that the vnode for this device will be released 2551 * by vn_open. Open will detect this special error and take the 2552 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2553 * will simply report the error. 2554 */ 2555 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev); 2556 return (ENODEV); 2557 } 2558 2559 /* 2560 * The caller has reserved the file descriptor dfd for us. On success we 2561 * must fsetfd() it. On failure the caller will clean it up. 2562 */ 2563 int 2564 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error) 2565 { 2566 struct file *wfp; 2567 struct file *xfp; 2568 int werror; 2569 2570 if ((wfp = holdfp(fdp, sfd, -1)) == NULL) 2571 return (EBADF); 2572 2573 /* 2574 * Close a revoke/dup race. Duping a descriptor marked as revoked 2575 * will dup a dummy descriptor instead of the real one. 2576 */ 2577 if (wfp->f_flag & FREVOKED) { 2578 kprintf("Warning: attempt to dup() a revoked descriptor\n"); 2579 fdrop(wfp); 2580 wfp = NULL; 2581 werror = falloc(NULL, &wfp, NULL); 2582 if (werror) 2583 return (werror); 2584 } 2585 2586 /* 2587 * There are two cases of interest here. 2588 * 2589 * For ENODEV simply dup sfd to file descriptor dfd and return. 2590 * 2591 * For ENXIO steal away the file structure from sfd and store it 2592 * dfd. sfd is effectively closed by this operation. 2593 * 2594 * Any other error code is just returned. 2595 */ 2596 switch (error) { 2597 case ENODEV: 2598 /* 2599 * Check that the mode the file is being opened for is a 2600 * subset of the mode of the existing descriptor. 2601 */ 2602 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2603 error = EACCES; 2604 break; 2605 } 2606 spin_lock(&fdp->fd_spin); 2607 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2608 fsetfd_locked(fdp, wfp, dfd); 2609 spin_unlock(&fdp->fd_spin); 2610 error = 0; 2611 break; 2612 case ENXIO: 2613 /* 2614 * Steal away the file pointer from dfd, and stuff it into indx. 2615 */ 2616 spin_lock(&fdp->fd_spin); 2617 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2618 fsetfd(fdp, wfp, dfd); 2619 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) { 2620 spin_unlock(&fdp->fd_spin); 2621 fdrop(xfp); 2622 } else { 2623 spin_unlock(&fdp->fd_spin); 2624 } 2625 error = 0; 2626 break; 2627 default: 2628 break; 2629 } 2630 fdrop(wfp); 2631 return (error); 2632 } 2633 2634 /* 2635 * NOT MPSAFE - I think these refer to a common file descriptor table 2636 * and we need to spinlock that to link fdtol in. 2637 */ 2638 struct filedesc_to_leader * 2639 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 2640 struct proc *leader) 2641 { 2642 struct filedesc_to_leader *fdtol; 2643 2644 fdtol = kmalloc(sizeof(struct filedesc_to_leader), 2645 M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO); 2646 fdtol->fdl_refcount = 1; 2647 fdtol->fdl_holdcount = 0; 2648 fdtol->fdl_wakeup = 0; 2649 fdtol->fdl_leader = leader; 2650 if (old != NULL) { 2651 fdtol->fdl_next = old->fdl_next; 2652 fdtol->fdl_prev = old; 2653 old->fdl_next = fdtol; 2654 fdtol->fdl_next->fdl_prev = fdtol; 2655 } else { 2656 fdtol->fdl_next = fdtol; 2657 fdtol->fdl_prev = fdtol; 2658 } 2659 return fdtol; 2660 } 2661 2662 /* 2663 * Scan all file pointers in the system. The callback is made with 2664 * the master list spinlock held exclusively. 2665 */ 2666 void 2667 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data) 2668 { 2669 struct file *fp; 2670 int res; 2671 2672 spin_lock(&filehead_spin); 2673 LIST_FOREACH(fp, &filehead, f_list) { 2674 res = callback(fp, data); 2675 if (res < 0) 2676 break; 2677 } 2678 spin_unlock(&filehead_spin); 2679 } 2680 2681 /* 2682 * Get file structures. 2683 * 2684 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe) 2685 */ 2686 2687 struct sysctl_kern_file_info { 2688 int count; 2689 int error; 2690 struct sysctl_req *req; 2691 }; 2692 2693 static int sysctl_kern_file_callback(struct proc *p, void *data); 2694 2695 static int 2696 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2697 { 2698 struct sysctl_kern_file_info info; 2699 2700 /* 2701 * Note: because the number of file descriptors is calculated 2702 * in different ways for sizing vs returning the data, 2703 * there is information leakage from the first loop. However, 2704 * it is of a similar order of magnitude to the leakage from 2705 * global system statistics such as kern.openfiles. 2706 * 2707 * When just doing a count, note that we cannot just count 2708 * the elements and add f_count via the filehead list because 2709 * threaded processes share their descriptor table and f_count might 2710 * still be '1' in that case. 2711 * 2712 * Since the SYSCTL op can block, we must hold the process to 2713 * prevent it being ripped out from under us either in the 2714 * file descriptor loop or in the greater LIST_FOREACH. The 2715 * process may be in varying states of disrepair. If the process 2716 * is in SZOMB we may have caught it just as it is being removed 2717 * from the allproc list, we must skip it in that case to maintain 2718 * an unbroken chain through the allproc list. 2719 */ 2720 info.count = 0; 2721 info.error = 0; 2722 info.req = req; 2723 allproc_scan(sysctl_kern_file_callback, &info); 2724 2725 /* 2726 * When just calculating the size, overestimate a bit to try to 2727 * prevent system activity from causing the buffer-fill call 2728 * to fail later on. 2729 */ 2730 if (req->oldptr == NULL) { 2731 info.count = (info.count + 16) + (info.count / 10); 2732 info.error = SYSCTL_OUT(req, NULL, 2733 info.count * sizeof(struct kinfo_file)); 2734 } 2735 return (info.error); 2736 } 2737 2738 static int 2739 sysctl_kern_file_callback(struct proc *p, void *data) 2740 { 2741 struct sysctl_kern_file_info *info = data; 2742 struct kinfo_file kf; 2743 struct filedesc *fdp; 2744 struct file *fp; 2745 uid_t uid; 2746 int n; 2747 2748 if (p->p_stat == SIDL || p->p_stat == SZOMB) 2749 return(0); 2750 if (!(PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)) 2751 return(0); 2752 2753 /* 2754 * Softref the fdp to prevent it from being destroyed 2755 */ 2756 spin_lock(&p->p_spin); 2757 if ((fdp = p->p_fd) == NULL) { 2758 spin_unlock(&p->p_spin); 2759 return(0); 2760 } 2761 atomic_add_int(&fdp->fd_softrefs, 1); 2762 spin_unlock(&p->p_spin); 2763 2764 /* 2765 * The fdp's own spinlock prevents the contents from being 2766 * modified. 2767 */ 2768 spin_lock_shared(&fdp->fd_spin); 2769 for (n = 0; n < fdp->fd_nfiles; ++n) { 2770 if ((fp = fdp->fd_files[n].fp) == NULL) 2771 continue; 2772 if (info->req->oldptr == NULL) { 2773 ++info->count; 2774 } else { 2775 uid = p->p_ucred ? p->p_ucred->cr_uid : -1; 2776 kcore_make_file(&kf, fp, p->p_pid, uid, n); 2777 spin_unlock_shared(&fdp->fd_spin); 2778 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf)); 2779 spin_lock_shared(&fdp->fd_spin); 2780 if (info->error) 2781 break; 2782 } 2783 } 2784 spin_unlock_shared(&fdp->fd_spin); 2785 atomic_subtract_int(&fdp->fd_softrefs, 1); 2786 if (info->error) 2787 return(-1); 2788 return(0); 2789 } 2790 2791 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2792 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 2793 2794 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW, 2795 &minfilesperproc, 0, "Minimum files allowed open per process"); 2796 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2797 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2798 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW, 2799 &maxfilesperuser, 0, "Maximum files allowed open per user"); 2800 2801 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2802 &maxfiles, 0, "Maximum number of files"); 2803 2804 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, 2805 &maxfilesrootres, 0, "Descriptors reserved for root use"); 2806 2807 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2808 &nfiles, 0, "System-wide number of open files"); 2809 2810 static void 2811 fildesc_drvinit(void *unused) 2812 { 2813 int fd; 2814 2815 for (fd = 0; fd < NUMFDESC; fd++) { 2816 make_dev(&fildesc_ops, fd, 2817 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 2818 } 2819 2820 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 2821 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 2822 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 2823 } 2824 2825 struct fileops badfileops = { 2826 .fo_read = badfo_readwrite, 2827 .fo_write = badfo_readwrite, 2828 .fo_ioctl = badfo_ioctl, 2829 .fo_kqfilter = badfo_kqfilter, 2830 .fo_stat = badfo_stat, 2831 .fo_close = badfo_close, 2832 .fo_shutdown = badfo_shutdown 2833 }; 2834 2835 int 2836 badfo_readwrite( 2837 struct file *fp, 2838 struct uio *uio, 2839 struct ucred *cred, 2840 int flags 2841 ) { 2842 return (EBADF); 2843 } 2844 2845 int 2846 badfo_ioctl(struct file *fp, u_long com, caddr_t data, 2847 struct ucred *cred, struct sysmsg *msgv) 2848 { 2849 return (EBADF); 2850 } 2851 2852 /* 2853 * Must return an error to prevent registration, typically 2854 * due to a revoked descriptor (file_filtops assigned). 2855 */ 2856 int 2857 badfo_kqfilter(struct file *fp, struct knote *kn) 2858 { 2859 return (EOPNOTSUPP); 2860 } 2861 2862 int 2863 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred) 2864 { 2865 return (EBADF); 2866 } 2867 2868 int 2869 badfo_close(struct file *fp) 2870 { 2871 return (EBADF); 2872 } 2873 2874 int 2875 badfo_shutdown(struct file *fp, int how) 2876 { 2877 return (EBADF); 2878 } 2879 2880 int 2881 nofo_shutdown(struct file *fp, int how) 2882 { 2883 return (EOPNOTSUPP); 2884 } 2885 2886 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE + CDEV_MAJOR, 2887 fildesc_drvinit,NULL); 2888