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 struct lwkt_token revoke_token = LWKT_TOKEN_INITIALIZER(revoke_token); 134 135 /* 136 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared. 137 * 138 * must be called with fdp->fd_spin exclusively held 139 */ 140 static __inline 141 void 142 fdfixup_locked(struct filedesc *fdp, int fd) 143 { 144 if (fd < fdp->fd_freefile) { 145 fdp->fd_freefile = fd; 146 } 147 while (fdp->fd_lastfile >= 0 && 148 fdp->fd_files[fdp->fd_lastfile].fp == NULL && 149 fdp->fd_files[fdp->fd_lastfile].reserved == 0 150 ) { 151 --fdp->fd_lastfile; 152 } 153 } 154 155 /* 156 * System calls on descriptors. 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 int 189 sys_dup2(struct dup2_args *uap) 190 { 191 int error; 192 int fd = 0; 193 194 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd); 195 uap->sysmsg_fds[0] = fd; 196 197 return (error); 198 } 199 200 /* 201 * Duplicate a file descriptor. 202 */ 203 int 204 sys_dup(struct dup_args *uap) 205 { 206 int error; 207 int fd = 0; 208 209 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd); 210 uap->sysmsg_fds[0] = fd; 211 212 return (error); 213 } 214 215 /* 216 * MPALMOSTSAFE - acquires mplock for fp operations 217 */ 218 int 219 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred) 220 { 221 struct thread *td = curthread; 222 struct proc *p = td->td_proc; 223 struct file *fp; 224 struct vnode *vp; 225 u_int newmin; 226 u_int oflags; 227 u_int nflags; 228 int tmp, error, flg = F_POSIX; 229 230 KKASSERT(p); 231 232 /* 233 * Operations on file descriptors that do not require a file pointer. 234 */ 235 switch (cmd) { 236 case F_GETFD: 237 error = fgetfdflags(p->p_fd, fd, &tmp); 238 if (error == 0) 239 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0; 240 return (error); 241 242 case F_SETFD: 243 if (dat->fc_cloexec & FD_CLOEXEC) 244 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE); 245 else 246 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE); 247 return (error); 248 case F_DUPFD: 249 newmin = dat->fc_fd; 250 error = kern_dup(DUP_VARIABLE | DUP_FCNTL, fd, newmin, 251 &dat->fc_fd); 252 return (error); 253 case F_DUPFD_CLOEXEC: 254 newmin = dat->fc_fd; 255 error = kern_dup(DUP_VARIABLE | DUP_CLOEXEC | DUP_FCNTL, 256 fd, newmin, &dat->fc_fd); 257 return (error); 258 case F_DUP2FD: 259 newmin = dat->fc_fd; 260 error = kern_dup(DUP_FIXED, fd, newmin, &dat->fc_fd); 261 return (error); 262 case F_DUP2FD_CLOEXEC: 263 newmin = dat->fc_fd; 264 error = kern_dup(DUP_FIXED | DUP_CLOEXEC, fd, newmin, 265 &dat->fc_fd); 266 return (error); 267 default: 268 break; 269 } 270 271 /* 272 * Operations on file pointers 273 */ 274 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 275 return (EBADF); 276 277 switch (cmd) { 278 case F_GETFL: 279 dat->fc_flags = OFLAGS(fp->f_flag); 280 error = 0; 281 break; 282 283 case F_SETFL: 284 oflags = fp->f_flag; 285 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 286 nflags |= oflags & ~FCNTLFLAGS; 287 288 error = 0; 289 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY)) 290 error = EINVAL; 291 if (error == 0 && ((nflags ^ oflags) & FASYNC)) { 292 tmp = nflags & FASYNC; 293 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, 294 cred, NULL); 295 } 296 297 /* 298 * If no error, must be atomically set. 299 */ 300 while (error == 0) { 301 oflags = fp->f_flag; 302 cpu_ccfence(); 303 nflags = (oflags & ~FCNTLFLAGS) | (nflags & FCNTLFLAGS); 304 if (atomic_cmpset_int(&fp->f_flag, oflags, nflags)) 305 break; 306 cpu_pause(); 307 } 308 break; 309 310 case F_GETOWN: 311 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, 312 cred, NULL); 313 break; 314 315 case F_SETOWN: 316 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, 317 cred, NULL); 318 break; 319 320 case F_SETLKW: 321 flg |= F_WAIT; 322 /* Fall into F_SETLK */ 323 324 case F_SETLK: 325 if (fp->f_type != DTYPE_VNODE) { 326 error = EBADF; 327 break; 328 } 329 vp = (struct vnode *)fp->f_data; 330 331 /* 332 * copyin/lockop may block 333 */ 334 if (dat->fc_flock.l_whence == SEEK_CUR) 335 dat->fc_flock.l_start += fp->f_offset; 336 337 switch (dat->fc_flock.l_type) { 338 case F_RDLCK: 339 if ((fp->f_flag & FREAD) == 0) { 340 error = EBADF; 341 break; 342 } 343 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) { 344 lwkt_gettoken(&p->p_leader->p_token); 345 p->p_leader->p_flags |= P_ADVLOCK; 346 lwkt_reltoken(&p->p_leader->p_token); 347 } 348 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 349 &dat->fc_flock, flg); 350 break; 351 case F_WRLCK: 352 if ((fp->f_flag & FWRITE) == 0) { 353 error = EBADF; 354 break; 355 } 356 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) { 357 lwkt_gettoken(&p->p_leader->p_token); 358 p->p_leader->p_flags |= P_ADVLOCK; 359 lwkt_reltoken(&p->p_leader->p_token); 360 } 361 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 362 &dat->fc_flock, flg); 363 break; 364 case F_UNLCK: 365 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 366 &dat->fc_flock, F_POSIX); 367 break; 368 default: 369 error = EINVAL; 370 break; 371 } 372 373 /* 374 * It is possible to race a close() on the descriptor while 375 * we were blocked getting the lock. If this occurs the 376 * close might not have caught the lock. 377 */ 378 if (checkfdclosed(p->p_fd, fd, fp)) { 379 dat->fc_flock.l_whence = SEEK_SET; 380 dat->fc_flock.l_start = 0; 381 dat->fc_flock.l_len = 0; 382 dat->fc_flock.l_type = F_UNLCK; 383 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 384 F_UNLCK, &dat->fc_flock, F_POSIX); 385 } 386 break; 387 388 case F_GETLK: 389 if (fp->f_type != DTYPE_VNODE) { 390 error = EBADF; 391 break; 392 } 393 vp = (struct vnode *)fp->f_data; 394 /* 395 * copyin/lockop may block 396 */ 397 if (dat->fc_flock.l_type != F_RDLCK && 398 dat->fc_flock.l_type != F_WRLCK && 399 dat->fc_flock.l_type != F_UNLCK) { 400 error = EINVAL; 401 break; 402 } 403 if (dat->fc_flock.l_whence == SEEK_CUR) 404 dat->fc_flock.l_start += fp->f_offset; 405 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 406 &dat->fc_flock, F_POSIX); 407 break; 408 default: 409 error = EINVAL; 410 break; 411 } 412 413 fdrop(fp); 414 return (error); 415 } 416 417 /* 418 * The file control system call. 419 */ 420 int 421 sys_fcntl(struct fcntl_args *uap) 422 { 423 union fcntl_dat dat; 424 int error; 425 426 switch (uap->cmd) { 427 case F_DUPFD: 428 case F_DUP2FD: 429 case F_DUPFD_CLOEXEC: 430 case F_DUP2FD_CLOEXEC: 431 dat.fc_fd = uap->arg; 432 break; 433 case F_SETFD: 434 dat.fc_cloexec = uap->arg; 435 break; 436 case F_SETFL: 437 dat.fc_flags = uap->arg; 438 break; 439 case F_SETOWN: 440 dat.fc_owner = uap->arg; 441 break; 442 case F_SETLKW: 443 case F_SETLK: 444 case F_GETLK: 445 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 446 sizeof(struct flock)); 447 if (error) 448 return (error); 449 break; 450 } 451 452 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred); 453 454 if (error == 0) { 455 switch (uap->cmd) { 456 case F_DUPFD: 457 case F_DUP2FD: 458 case F_DUPFD_CLOEXEC: 459 case F_DUP2FD_CLOEXEC: 460 uap->sysmsg_result = dat.fc_fd; 461 break; 462 case F_GETFD: 463 uap->sysmsg_result = dat.fc_cloexec; 464 break; 465 case F_GETFL: 466 uap->sysmsg_result = dat.fc_flags; 467 break; 468 case F_GETOWN: 469 uap->sysmsg_result = dat.fc_owner; 470 break; 471 case F_GETLK: 472 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 473 sizeof(struct flock)); 474 break; 475 } 476 } 477 478 return (error); 479 } 480 481 /* 482 * Common code for dup, dup2, and fcntl(F_DUPFD). 483 * 484 * There are four type flags: DUP_FCNTL, DUP_FIXED, DUP_VARIABLE, and 485 * DUP_CLOEXEC. 486 * 487 * DUP_FCNTL is for handling EINVAL vs. EBADF differences between 488 * fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC and dup2() (per POSIX). 489 * The next two flags are mutually exclusive, and the fourth is optional. 490 * DUP_FIXED tells kern_dup() to destructively dup over an existing file 491 * descriptor if "new" is already open. DUP_VARIABLE tells kern_dup() 492 * to find the lowest unused file descriptor that is greater than or 493 * equal to "new". DUP_CLOEXEC, which works with either of the first 494 * two flags, sets the close-on-exec flag on the "new" file descriptor. 495 */ 496 int 497 kern_dup(int flags, int old, int new, int *res) 498 { 499 struct thread *td = curthread; 500 struct proc *p = td->td_proc; 501 struct filedesc *fdp = p->p_fd; 502 struct file *fp; 503 struct file *delfp; 504 int oldflags; 505 int holdleaders; 506 int dtsize; 507 int error, newfd; 508 509 /* 510 * Verify that we have a valid descriptor to dup from and 511 * possibly to dup to. When the new descriptor is out of 512 * bounds, fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC must 513 * return EINVAL, while dup2() returns EBADF in 514 * this case. 515 * 516 * NOTE: maxfilesperuser is not applicable to dup() 517 */ 518 retry: 519 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 520 dtsize = INT_MAX; 521 else 522 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 523 if (dtsize > maxfilesperproc) 524 dtsize = maxfilesperproc; 525 if (dtsize < minfilesperproc) 526 dtsize = minfilesperproc; 527 528 if (new < 0 || new > dtsize) 529 return (flags & DUP_FCNTL ? EINVAL : EBADF); 530 531 spin_lock(&fdp->fd_spin); 532 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) { 533 spin_unlock(&fdp->fd_spin); 534 return (EBADF); 535 } 536 if ((flags & DUP_FIXED) && old == new) { 537 *res = new; 538 if (flags & DUP_CLOEXEC) 539 fdp->fd_files[new].fileflags |= UF_EXCLOSE; 540 spin_unlock(&fdp->fd_spin); 541 return (0); 542 } 543 fp = fdp->fd_files[old].fp; 544 oldflags = fdp->fd_files[old].fileflags; 545 fhold(fp); 546 547 /* 548 * Allocate a new descriptor if DUP_VARIABLE, or expand the table 549 * if the requested descriptor is beyond the current table size. 550 * 551 * This can block. Retry if the source descriptor no longer matches 552 * or if our expectation in the expansion case races. 553 * 554 * If we are not expanding or allocating a new decriptor, then reset 555 * the target descriptor to a reserved state so we have a uniform 556 * setup for the next code block. 557 */ 558 if ((flags & DUP_VARIABLE) || new >= fdp->fd_nfiles) { 559 spin_unlock(&fdp->fd_spin); 560 error = fdalloc(p, new, &newfd); 561 spin_lock(&fdp->fd_spin); 562 if (error) { 563 spin_unlock(&fdp->fd_spin); 564 fdrop(fp); 565 return (error); 566 } 567 /* 568 * Check for ripout 569 */ 570 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) { 571 fsetfd_locked(fdp, NULL, newfd); 572 spin_unlock(&fdp->fd_spin); 573 fdrop(fp); 574 goto retry; 575 } 576 /* 577 * Check for expansion race 578 */ 579 if ((flags & DUP_VARIABLE) == 0 && new != newfd) { 580 fsetfd_locked(fdp, NULL, newfd); 581 spin_unlock(&fdp->fd_spin); 582 fdrop(fp); 583 goto retry; 584 } 585 /* 586 * Check for ripout, newfd reused old (this case probably 587 * can't occur). 588 */ 589 if (old == newfd) { 590 fsetfd_locked(fdp, NULL, newfd); 591 spin_unlock(&fdp->fd_spin); 592 fdrop(fp); 593 goto retry; 594 } 595 new = newfd; 596 delfp = NULL; 597 } else { 598 if (fdp->fd_files[new].reserved) { 599 spin_unlock(&fdp->fd_spin); 600 fdrop(fp); 601 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new); 602 tsleep(fdp, 0, "fdres", hz); 603 goto retry; 604 } 605 606 /* 607 * If the target descriptor was never allocated we have 608 * to allocate it. If it was we have to clean out the 609 * old descriptor. delfp inherits the ref from the 610 * descriptor table. 611 */ 612 delfp = fdp->fd_files[new].fp; 613 fdp->fd_files[new].fp = NULL; 614 fdp->fd_files[new].reserved = 1; 615 if (delfp == NULL) { 616 fdreserve_locked(fdp, new, 1); 617 if (new > fdp->fd_lastfile) 618 fdp->fd_lastfile = new; 619 } 620 621 } 622 623 /* 624 * NOTE: still holding an exclusive spinlock 625 */ 626 627 /* 628 * If a descriptor is being overwritten we may hve to tell 629 * fdfree() to sleep to ensure that all relevant process 630 * leaders can be traversed in closef(). 631 */ 632 if (delfp != NULL && p->p_fdtol != NULL) { 633 fdp->fd_holdleaderscount++; 634 holdleaders = 1; 635 } else { 636 holdleaders = 0; 637 } 638 KASSERT(delfp == NULL || (flags & DUP_FIXED), 639 ("dup() picked an open file")); 640 641 /* 642 * Duplicate the source descriptor, update lastfile. If the new 643 * descriptor was not allocated and we aren't replacing an existing 644 * descriptor we have to mark the descriptor as being in use. 645 * 646 * The fd_files[] array inherits fp's hold reference. 647 */ 648 fsetfd_locked(fdp, fp, new); 649 if ((flags & DUP_CLOEXEC) != 0) 650 fdp->fd_files[new].fileflags = oldflags | UF_EXCLOSE; 651 else 652 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE; 653 spin_unlock(&fdp->fd_spin); 654 fdrop(fp); 655 *res = new; 656 657 /* 658 * If we dup'd over a valid file, we now own the reference to it 659 * and must dispose of it using closef() semantics (as if a 660 * close() were performed on it). 661 */ 662 if (delfp) { 663 if (SLIST_FIRST(&delfp->f_klist)) 664 knote_fdclose(delfp, fdp, new); 665 closef(delfp, p); 666 if (holdleaders) { 667 spin_lock(&fdp->fd_spin); 668 fdp->fd_holdleaderscount--; 669 if (fdp->fd_holdleaderscount == 0 && 670 fdp->fd_holdleaderswakeup != 0) { 671 fdp->fd_holdleaderswakeup = 0; 672 spin_unlock(&fdp->fd_spin); 673 wakeup(&fdp->fd_holdleaderscount); 674 } else { 675 spin_unlock(&fdp->fd_spin); 676 } 677 } 678 } 679 return (0); 680 } 681 682 /* 683 * If sigio is on the list associated with a process or process group, 684 * disable signalling from the device, remove sigio from the list and 685 * free sigio. 686 */ 687 void 688 funsetown(struct sigio **sigiop) 689 { 690 struct pgrp *pgrp; 691 struct proc *p; 692 struct sigio *sigio; 693 694 if ((sigio = *sigiop) != NULL) { 695 lwkt_gettoken(&sigio_token); /* protect sigio */ 696 KKASSERT(sigiop == sigio->sio_myref); 697 sigio = *sigiop; 698 *sigiop = NULL; 699 lwkt_reltoken(&sigio_token); 700 } 701 if (sigio == NULL) 702 return; 703 704 if (sigio->sio_pgid < 0) { 705 pgrp = sigio->sio_pgrp; 706 sigio->sio_pgrp = NULL; 707 lwkt_gettoken(&pgrp->pg_token); 708 SLIST_REMOVE(&pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio); 709 lwkt_reltoken(&pgrp->pg_token); 710 pgrel(pgrp); 711 } else /* if ((*sigiop)->sio_pgid > 0) */ { 712 p = sigio->sio_proc; 713 sigio->sio_proc = NULL; 714 PHOLD(p); 715 lwkt_gettoken(&p->p_token); 716 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); 717 lwkt_reltoken(&p->p_token); 718 PRELE(p); 719 } 720 crfree(sigio->sio_ucred); 721 sigio->sio_ucred = NULL; 722 kfree(sigio, M_SIGIO); 723 } 724 725 /* 726 * Free a list of sigio structures. Caller is responsible for ensuring 727 * that the list is MPSAFE. 728 */ 729 void 730 funsetownlst(struct sigiolst *sigiolst) 731 { 732 struct sigio *sigio; 733 734 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) 735 funsetown(sigio->sio_myref); 736 } 737 738 /* 739 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 740 * 741 * After permission checking, add a sigio structure to the sigio list for 742 * the process or process group. 743 */ 744 int 745 fsetown(pid_t pgid, struct sigio **sigiop) 746 { 747 struct proc *proc = NULL; 748 struct pgrp *pgrp = NULL; 749 struct sigio *sigio; 750 int error; 751 752 if (pgid == 0) { 753 funsetown(sigiop); 754 return (0); 755 } 756 757 if (pgid > 0) { 758 proc = pfind(pgid); 759 if (proc == NULL) { 760 error = ESRCH; 761 goto done; 762 } 763 764 /* 765 * Policy - Don't allow a process to FSETOWN a process 766 * in another session. 767 * 768 * Remove this test to allow maximum flexibility or 769 * restrict FSETOWN to the current process or process 770 * group for maximum safety. 771 */ 772 if (proc->p_session != curproc->p_session) { 773 error = EPERM; 774 goto done; 775 } 776 } else /* if (pgid < 0) */ { 777 pgrp = pgfind(-pgid); 778 if (pgrp == NULL) { 779 error = ESRCH; 780 goto done; 781 } 782 783 /* 784 * Policy - Don't allow a process to FSETOWN a process 785 * in another session. 786 * 787 * Remove this test to allow maximum flexibility or 788 * restrict FSETOWN to the current process or process 789 * group for maximum safety. 790 */ 791 if (pgrp->pg_session != curproc->p_session) { 792 error = EPERM; 793 goto done; 794 } 795 } 796 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK | M_ZERO); 797 if (pgid > 0) { 798 KKASSERT(pgrp == NULL); 799 lwkt_gettoken(&proc->p_token); 800 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 801 sigio->sio_proc = proc; 802 lwkt_reltoken(&proc->p_token); 803 } else { 804 KKASSERT(proc == NULL); 805 lwkt_gettoken(&pgrp->pg_token); 806 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 807 sigio->sio_pgrp = pgrp; 808 lwkt_reltoken(&pgrp->pg_token); 809 pgrp = NULL; 810 } 811 sigio->sio_pgid = pgid; 812 sigio->sio_ucred = crhold(curthread->td_ucred); 813 /* It would be convenient if p_ruid was in ucred. */ 814 sigio->sio_ruid = sigio->sio_ucred->cr_ruid; 815 sigio->sio_myref = sigiop; 816 817 lwkt_gettoken(&sigio_token); 818 while (*sigiop) 819 funsetown(sigiop); 820 *sigiop = sigio; 821 lwkt_reltoken(&sigio_token); 822 error = 0; 823 done: 824 if (pgrp) 825 pgrel(pgrp); 826 if (proc) 827 PRELE(proc); 828 return (error); 829 } 830 831 /* 832 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 833 */ 834 pid_t 835 fgetown(struct sigio **sigiop) 836 { 837 struct sigio *sigio; 838 pid_t own; 839 840 lwkt_gettoken_shared(&sigio_token); 841 sigio = *sigiop; 842 own = (sigio != NULL ? sigio->sio_pgid : 0); 843 lwkt_reltoken(&sigio_token); 844 845 return (own); 846 } 847 848 /* 849 * Close many file descriptors. 850 */ 851 int 852 sys_closefrom(struct closefrom_args *uap) 853 { 854 return(kern_closefrom(uap->fd)); 855 } 856 857 /* 858 * Close all file descriptors greater then or equal to fd 859 */ 860 int 861 kern_closefrom(int fd) 862 { 863 struct thread *td = curthread; 864 struct proc *p = td->td_proc; 865 struct filedesc *fdp; 866 867 KKASSERT(p); 868 fdp = p->p_fd; 869 870 if (fd < 0) 871 return (EINVAL); 872 873 /* 874 * NOTE: This function will skip unassociated descriptors and 875 * reserved descriptors that have not yet been assigned. 876 * fd_lastfile can change as a side effect of kern_close(). 877 */ 878 spin_lock(&fdp->fd_spin); 879 while (fd <= fdp->fd_lastfile) { 880 if (fdp->fd_files[fd].fp != NULL) { 881 spin_unlock(&fdp->fd_spin); 882 /* ok if this races another close */ 883 if (kern_close(fd) == EINTR) 884 return (EINTR); 885 spin_lock(&fdp->fd_spin); 886 } 887 ++fd; 888 } 889 spin_unlock(&fdp->fd_spin); 890 return (0); 891 } 892 893 /* 894 * Close a file descriptor. 895 */ 896 int 897 sys_close(struct close_args *uap) 898 { 899 return(kern_close(uap->fd)); 900 } 901 902 /* 903 * close() helper 904 */ 905 int 906 kern_close(int fd) 907 { 908 struct thread *td = curthread; 909 struct proc *p = td->td_proc; 910 struct filedesc *fdp; 911 struct file *fp; 912 int error; 913 int holdleaders; 914 915 KKASSERT(p); 916 fdp = p->p_fd; 917 918 spin_lock(&fdp->fd_spin); 919 if ((fp = funsetfd_locked(fdp, fd)) == NULL) { 920 spin_unlock(&fdp->fd_spin); 921 return (EBADF); 922 } 923 holdleaders = 0; 924 if (p->p_fdtol != NULL) { 925 /* 926 * Ask fdfree() to sleep to ensure that all relevant 927 * process leaders can be traversed in closef(). 928 */ 929 fdp->fd_holdleaderscount++; 930 holdleaders = 1; 931 } 932 933 /* 934 * we now hold the fp reference that used to be owned by the descriptor 935 * array. 936 */ 937 spin_unlock(&fdp->fd_spin); 938 if (SLIST_FIRST(&fp->f_klist)) 939 knote_fdclose(fp, fdp, fd); 940 error = closef(fp, p); 941 if (holdleaders) { 942 spin_lock(&fdp->fd_spin); 943 fdp->fd_holdleaderscount--; 944 if (fdp->fd_holdleaderscount == 0 && 945 fdp->fd_holdleaderswakeup != 0) { 946 fdp->fd_holdleaderswakeup = 0; 947 spin_unlock(&fdp->fd_spin); 948 wakeup(&fdp->fd_holdleaderscount); 949 } else { 950 spin_unlock(&fdp->fd_spin); 951 } 952 } 953 return (error); 954 } 955 956 /* 957 * shutdown_args(int fd, int how) 958 */ 959 int 960 kern_shutdown(int fd, int how) 961 { 962 struct thread *td = curthread; 963 struct proc *p = td->td_proc; 964 struct file *fp; 965 int error; 966 967 KKASSERT(p); 968 969 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 970 return (EBADF); 971 error = fo_shutdown(fp, how); 972 fdrop(fp); 973 974 return (error); 975 } 976 977 /* 978 * MPALMOSTSAFE 979 */ 980 int 981 sys_shutdown(struct shutdown_args *uap) 982 { 983 int error; 984 985 error = kern_shutdown(uap->s, uap->how); 986 987 return (error); 988 } 989 990 /* 991 * fstat() helper 992 */ 993 int 994 kern_fstat(int fd, struct stat *ub) 995 { 996 struct thread *td = curthread; 997 struct proc *p = td->td_proc; 998 struct file *fp; 999 int error; 1000 1001 KKASSERT(p); 1002 1003 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 1004 return (EBADF); 1005 error = fo_stat(fp, ub, td->td_ucred); 1006 fdrop(fp); 1007 1008 return (error); 1009 } 1010 1011 /* 1012 * Return status information about a file descriptor. 1013 */ 1014 int 1015 sys_fstat(struct fstat_args *uap) 1016 { 1017 struct stat st; 1018 int error; 1019 1020 error = kern_fstat(uap->fd, &st); 1021 1022 if (error == 0) 1023 error = copyout(&st, uap->sb, sizeof(st)); 1024 return (error); 1025 } 1026 1027 /* 1028 * Return pathconf information about a file descriptor. 1029 * 1030 * MPALMOSTSAFE 1031 */ 1032 int 1033 sys_fpathconf(struct fpathconf_args *uap) 1034 { 1035 struct thread *td = curthread; 1036 struct proc *p = td->td_proc; 1037 struct file *fp; 1038 struct vnode *vp; 1039 int error = 0; 1040 1041 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 1042 return (EBADF); 1043 1044 switch (fp->f_type) { 1045 case DTYPE_PIPE: 1046 case DTYPE_SOCKET: 1047 if (uap->name != _PC_PIPE_BUF) { 1048 error = EINVAL; 1049 } else { 1050 uap->sysmsg_result = PIPE_BUF; 1051 error = 0; 1052 } 1053 break; 1054 case DTYPE_FIFO: 1055 case DTYPE_VNODE: 1056 vp = (struct vnode *)fp->f_data; 1057 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg); 1058 break; 1059 default: 1060 error = EOPNOTSUPP; 1061 break; 1062 } 1063 fdrop(fp); 1064 return(error); 1065 } 1066 1067 static int fdexpand; 1068 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, 1069 "Number of times a file table has been expanded"); 1070 1071 /* 1072 * Grow the file table so it can hold through descriptor (want). 1073 * 1074 * The fdp's spinlock must be held exclusively on entry and may be held 1075 * exclusively on return. The spinlock may be cycled by the routine. 1076 */ 1077 static void 1078 fdgrow_locked(struct filedesc *fdp, int want) 1079 { 1080 struct fdnode *newfiles; 1081 struct fdnode *oldfiles; 1082 int nf, extra; 1083 1084 nf = fdp->fd_nfiles; 1085 do { 1086 /* nf has to be of the form 2^n - 1 */ 1087 nf = 2 * nf + 1; 1088 } while (nf <= want); 1089 1090 spin_unlock(&fdp->fd_spin); 1091 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK); 1092 spin_lock(&fdp->fd_spin); 1093 1094 /* 1095 * We could have raced another extend while we were not holding 1096 * the spinlock. 1097 */ 1098 if (fdp->fd_nfiles >= nf) { 1099 spin_unlock(&fdp->fd_spin); 1100 kfree(newfiles, M_FILEDESC); 1101 spin_lock(&fdp->fd_spin); 1102 return; 1103 } 1104 /* 1105 * Copy the existing ofile and ofileflags arrays 1106 * and zero the new portion of each array. 1107 */ 1108 extra = nf - fdp->fd_nfiles; 1109 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode)); 1110 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode)); 1111 1112 oldfiles = fdp->fd_files; 1113 fdp->fd_files = newfiles; 1114 fdp->fd_nfiles = nf; 1115 1116 if (oldfiles != fdp->fd_builtin_files) { 1117 spin_unlock(&fdp->fd_spin); 1118 kfree(oldfiles, M_FILEDESC); 1119 spin_lock(&fdp->fd_spin); 1120 } 1121 fdexpand++; 1122 } 1123 1124 /* 1125 * Number of nodes in right subtree, including the root. 1126 */ 1127 static __inline int 1128 right_subtree_size(int n) 1129 { 1130 return (n ^ (n | (n + 1))); 1131 } 1132 1133 /* 1134 * Bigger ancestor. 1135 */ 1136 static __inline int 1137 right_ancestor(int n) 1138 { 1139 return (n | (n + 1)); 1140 } 1141 1142 /* 1143 * Smaller ancestor. 1144 */ 1145 static __inline int 1146 left_ancestor(int n) 1147 { 1148 return ((n & (n + 1)) - 1); 1149 } 1150 1151 /* 1152 * Traverse the in-place binary tree buttom-up adjusting the allocation 1153 * count so scans can determine where free descriptors are located. 1154 * 1155 * caller must be holding an exclusive spinlock on fdp 1156 */ 1157 static 1158 void 1159 fdreserve_locked(struct filedesc *fdp, int fd, int incr) 1160 { 1161 while (fd >= 0) { 1162 fdp->fd_files[fd].allocated += incr; 1163 KKASSERT(fdp->fd_files[fd].allocated >= 0); 1164 fd = left_ancestor(fd); 1165 } 1166 } 1167 1168 /* 1169 * Reserve a file descriptor for the process. If no error occurs, the 1170 * caller MUST at some point call fsetfd() or assign a file pointer 1171 * or dispose of the reservation. 1172 */ 1173 int 1174 fdalloc(struct proc *p, int want, int *result) 1175 { 1176 struct filedesc *fdp = p->p_fd; 1177 struct uidinfo *uip; 1178 int fd, rsize, rsum, node, lim; 1179 1180 /* 1181 * Check dtable size limit 1182 */ 1183 spin_lock(&p->p_limit->p_spin); 1184 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 1185 lim = INT_MAX; 1186 else 1187 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 1188 spin_unlock(&p->p_limit->p_spin); 1189 1190 if (lim > maxfilesperproc) 1191 lim = maxfilesperproc; 1192 if (lim < minfilesperproc) 1193 lim = minfilesperproc; 1194 if (want >= lim) 1195 return (EMFILE); 1196 1197 /* 1198 * Check that the user has not run out of descriptors (non-root only). 1199 * As a safety measure the dtable is allowed to have at least 1200 * minfilesperproc open fds regardless of the maxfilesperuser limit. 1201 */ 1202 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) { 1203 uip = p->p_ucred->cr_uidinfo; 1204 if (uip->ui_openfiles > maxfilesperuser) { 1205 krateprintf(&krate_uidinfo, 1206 "Warning: user %d pid %d (%s) ran out of " 1207 "file descriptors (%d/%d)\n", 1208 p->p_ucred->cr_uid, (int)p->p_pid, 1209 p->p_comm, 1210 uip->ui_openfiles, maxfilesperuser); 1211 return(ENFILE); 1212 } 1213 } 1214 1215 /* 1216 * Grow the dtable if necessary 1217 */ 1218 spin_lock(&fdp->fd_spin); 1219 if (want >= fdp->fd_nfiles) 1220 fdgrow_locked(fdp, want); 1221 1222 /* 1223 * Search for a free descriptor starting at the higher 1224 * of want or fd_freefile. If that fails, consider 1225 * expanding the ofile array. 1226 * 1227 * NOTE! the 'allocated' field is a cumulative recursive allocation 1228 * count. If we happen to see a value of 0 then we can shortcut 1229 * our search. Otherwise we run through through the tree going 1230 * down branches we know have free descriptor(s) until we hit a 1231 * leaf node. The leaf node will be free but will not necessarily 1232 * have an allocated field of 0. 1233 */ 1234 retry: 1235 /* move up the tree looking for a subtree with a free node */ 1236 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim); 1237 fd = right_ancestor(fd)) { 1238 if (fdp->fd_files[fd].allocated == 0) 1239 goto found; 1240 1241 rsize = right_subtree_size(fd); 1242 if (fdp->fd_files[fd].allocated == rsize) 1243 continue; /* right subtree full */ 1244 1245 /* 1246 * Free fd is in the right subtree of the tree rooted at fd. 1247 * Call that subtree R. Look for the smallest (leftmost) 1248 * subtree of R with an unallocated fd: continue moving 1249 * down the left branch until encountering a full left 1250 * subtree, then move to the right. 1251 */ 1252 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) { 1253 node = fd + rsize; 1254 rsum += fdp->fd_files[node].allocated; 1255 if (fdp->fd_files[fd].allocated == rsum + rsize) { 1256 fd = node; /* move to the right */ 1257 if (fdp->fd_files[node].allocated == 0) 1258 goto found; 1259 rsum = 0; 1260 } 1261 } 1262 goto found; 1263 } 1264 1265 /* 1266 * No space in current array. Expand? 1267 */ 1268 if (fdp->fd_nfiles >= lim) { 1269 spin_unlock(&fdp->fd_spin); 1270 return (EMFILE); 1271 } 1272 fdgrow_locked(fdp, want); 1273 goto retry; 1274 1275 found: 1276 KKASSERT(fd < fdp->fd_nfiles); 1277 if (fd > fdp->fd_lastfile) 1278 fdp->fd_lastfile = fd; 1279 if (want <= fdp->fd_freefile) 1280 fdp->fd_freefile = fd; 1281 *result = fd; 1282 KKASSERT(fdp->fd_files[fd].fp == NULL); 1283 KKASSERT(fdp->fd_files[fd].reserved == 0); 1284 fdp->fd_files[fd].fileflags = 0; 1285 fdp->fd_files[fd].reserved = 1; 1286 fdreserve_locked(fdp, fd, 1); 1287 spin_unlock(&fdp->fd_spin); 1288 return (0); 1289 } 1290 1291 /* 1292 * Check to see whether n user file descriptors 1293 * are available to the process p. 1294 */ 1295 int 1296 fdavail(struct proc *p, int n) 1297 { 1298 struct filedesc *fdp = p->p_fd; 1299 struct fdnode *fdnode; 1300 int i, lim, last; 1301 1302 spin_lock(&p->p_limit->p_spin); 1303 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 1304 lim = INT_MAX; 1305 else 1306 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 1307 spin_unlock(&p->p_limit->p_spin); 1308 1309 if (lim > maxfilesperproc) 1310 lim = maxfilesperproc; 1311 if (lim < minfilesperproc) 1312 lim = minfilesperproc; 1313 1314 spin_lock(&fdp->fd_spin); 1315 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) { 1316 spin_unlock(&fdp->fd_spin); 1317 return (1); 1318 } 1319 last = min(fdp->fd_nfiles, lim); 1320 fdnode = &fdp->fd_files[fdp->fd_freefile]; 1321 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) { 1322 if (fdnode->fp == NULL && --n <= 0) { 1323 spin_unlock(&fdp->fd_spin); 1324 return (1); 1325 } 1326 } 1327 spin_unlock(&fdp->fd_spin); 1328 return (0); 1329 } 1330 1331 /* 1332 * Revoke open descriptors referencing (f_data, f_type) 1333 * 1334 * Any revoke executed within a prison is only able to 1335 * revoke descriptors for processes within that prison. 1336 * 1337 * Returns 0 on success or an error code. 1338 */ 1339 struct fdrevoke_info { 1340 void *data; 1341 short type; 1342 short unused; 1343 int found; 1344 struct ucred *cred; 1345 struct file *nfp; 1346 }; 1347 1348 static int fdrevoke_check_callback(struct file *fp, void *vinfo); 1349 static int fdrevoke_proc_callback(struct proc *p, void *vinfo); 1350 1351 int 1352 fdrevoke(void *f_data, short f_type, struct ucred *cred) 1353 { 1354 struct fdrevoke_info info; 1355 int error; 1356 1357 bzero(&info, sizeof(info)); 1358 info.data = f_data; 1359 info.type = f_type; 1360 info.cred = cred; 1361 error = falloc(NULL, &info.nfp, NULL); 1362 if (error) 1363 return (error); 1364 1365 /* 1366 * Scan the file pointer table once. dups do not dup file pointers, 1367 * only descriptors, so there is no leak. Set FREVOKED on the fps 1368 * being revoked. 1369 * 1370 * Any fps sent over unix-domain sockets will be revoked by the 1371 * socket code checking for FREVOKED when the fps are externialized. 1372 * revoke_token is used to make sure that fps marked FREVOKED and 1373 * externalized will be picked up by the following allproc_scan(). 1374 */ 1375 lwkt_gettoken(&revoke_token); 1376 allfiles_scan_exclusive(fdrevoke_check_callback, &info); 1377 lwkt_reltoken(&revoke_token); 1378 1379 /* 1380 * If any fps were marked track down the related descriptors 1381 * and close them. Any dup()s at this point will notice 1382 * the FREVOKED already set in the fp and do the right thing. 1383 */ 1384 if (info.found) 1385 allproc_scan(fdrevoke_proc_callback, &info); 1386 fdrop(info.nfp); 1387 return(0); 1388 } 1389 1390 /* 1391 * Locate matching file pointers directly. 1392 * 1393 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls! 1394 */ 1395 static int 1396 fdrevoke_check_callback(struct file *fp, void *vinfo) 1397 { 1398 struct fdrevoke_info *info = vinfo; 1399 1400 /* 1401 * File pointers already flagged for revokation are skipped. 1402 */ 1403 if (fp->f_flag & FREVOKED) 1404 return(0); 1405 1406 /* 1407 * If revoking from a prison file pointers created outside of 1408 * that prison, or file pointers without creds, cannot be revoked. 1409 */ 1410 if (info->cred->cr_prison && 1411 (fp->f_cred == NULL || 1412 info->cred->cr_prison != fp->f_cred->cr_prison)) { 1413 return(0); 1414 } 1415 1416 /* 1417 * If the file pointer matches then mark it for revocation. The 1418 * flag is currently only used by unp_revoke_gc(). 1419 * 1420 * info->found is a heuristic and can race in a SMP environment. 1421 */ 1422 if (info->data == fp->f_data && info->type == fp->f_type) { 1423 atomic_set_int(&fp->f_flag, FREVOKED); 1424 info->found = 1; 1425 } 1426 return(0); 1427 } 1428 1429 /* 1430 * Locate matching file pointers via process descriptor tables. 1431 */ 1432 static int 1433 fdrevoke_proc_callback(struct proc *p, void *vinfo) 1434 { 1435 struct fdrevoke_info *info = vinfo; 1436 struct filedesc *fdp; 1437 struct file *fp; 1438 int n; 1439 1440 if (p->p_stat == SIDL || p->p_stat == SZOMB) 1441 return(0); 1442 if (info->cred->cr_prison && 1443 info->cred->cr_prison != p->p_ucred->cr_prison) { 1444 return(0); 1445 } 1446 1447 /* 1448 * If the controlling terminal of the process matches the 1449 * vnode being revoked we clear the controlling terminal. 1450 * 1451 * The normal spec_close() may not catch this because it 1452 * uses curproc instead of p. 1453 */ 1454 if (p->p_session && info->type == DTYPE_VNODE && 1455 info->data == p->p_session->s_ttyvp) { 1456 p->p_session->s_ttyvp = NULL; 1457 vrele(info->data); 1458 } 1459 1460 /* 1461 * Softref the fdp to prevent it from being destroyed 1462 */ 1463 spin_lock(&p->p_spin); 1464 if ((fdp = p->p_fd) == NULL) { 1465 spin_unlock(&p->p_spin); 1466 return(0); 1467 } 1468 atomic_add_int(&fdp->fd_softrefs, 1); 1469 spin_unlock(&p->p_spin); 1470 1471 /* 1472 * Locate and close any matching file descriptors. 1473 */ 1474 spin_lock(&fdp->fd_spin); 1475 for (n = 0; n < fdp->fd_nfiles; ++n) { 1476 if ((fp = fdp->fd_files[n].fp) == NULL) 1477 continue; 1478 if (fp->f_flag & FREVOKED) { 1479 fhold(info->nfp); 1480 fdp->fd_files[n].fp = info->nfp; 1481 spin_unlock(&fdp->fd_spin); 1482 knote_fdclose(fp, fdp, n); /* XXX */ 1483 closef(fp, p); 1484 spin_lock(&fdp->fd_spin); 1485 } 1486 } 1487 spin_unlock(&fdp->fd_spin); 1488 atomic_subtract_int(&fdp->fd_softrefs, 1); 1489 return(0); 1490 } 1491 1492 /* 1493 * falloc: 1494 * Create a new open file structure and reserve a file decriptor 1495 * for the process that refers to it. 1496 * 1497 * Root creds are checked using lp, or assumed if lp is NULL. If 1498 * resultfd is non-NULL then lp must also be non-NULL. No file 1499 * descriptor is reserved (and no process context is needed) if 1500 * resultfd is NULL. 1501 * 1502 * A file pointer with a refcount of 1 is returned. Note that the 1503 * file pointer is NOT associated with the descriptor. If falloc 1504 * returns success, fsetfd() MUST be called to either associate the 1505 * file pointer or clear the reservation. 1506 */ 1507 int 1508 falloc(struct lwp *lp, struct file **resultfp, int *resultfd) 1509 { 1510 static struct timeval lastfail; 1511 static int curfail; 1512 struct file *fp; 1513 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred; 1514 int error; 1515 1516 fp = NULL; 1517 1518 /* 1519 * Handle filetable full issues and root overfill. 1520 */ 1521 if (nfiles >= maxfiles - maxfilesrootres && 1522 (cred->cr_ruid != 0 || nfiles >= maxfiles)) { 1523 if (ppsratecheck(&lastfail, &curfail, 1)) { 1524 kprintf("kern.maxfiles limit exceeded by uid %d, " 1525 "please see tuning(7).\n", 1526 cred->cr_ruid); 1527 } 1528 error = ENFILE; 1529 goto done; 1530 } 1531 1532 /* 1533 * Allocate a new file descriptor. 1534 */ 1535 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO); 1536 spin_init(&fp->f_spin, "falloc"); 1537 SLIST_INIT(&fp->f_klist); 1538 fp->f_count = 1; 1539 fp->f_ops = &badfileops; 1540 fp->f_seqcount = 1; 1541 fsetcred(fp, cred); 1542 spin_lock(&filehead_spin); 1543 nfiles++; 1544 LIST_INSERT_HEAD(&filehead, fp, f_list); 1545 spin_unlock(&filehead_spin); 1546 if (resultfd) { 1547 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) { 1548 fdrop(fp); 1549 fp = NULL; 1550 } 1551 } else { 1552 error = 0; 1553 } 1554 done: 1555 *resultfp = fp; 1556 return (error); 1557 } 1558 1559 /* 1560 * Check for races against a file descriptor by determining that the 1561 * file pointer is still associated with the specified file descriptor, 1562 * and a close is not currently in progress. 1563 */ 1564 int 1565 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp) 1566 { 1567 int error; 1568 1569 spin_lock_shared(&fdp->fd_spin); 1570 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp) 1571 error = EBADF; 1572 else 1573 error = 0; 1574 spin_unlock_shared(&fdp->fd_spin); 1575 return (error); 1576 } 1577 1578 /* 1579 * Associate a file pointer with a previously reserved file descriptor. 1580 * This function always succeeds. 1581 * 1582 * If fp is NULL, the file descriptor is returned to the pool. 1583 */ 1584 1585 /* 1586 * (exclusive spinlock must be held on call) 1587 */ 1588 static void 1589 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd) 1590 { 1591 KKASSERT((unsigned)fd < fdp->fd_nfiles); 1592 KKASSERT(fdp->fd_files[fd].reserved != 0); 1593 if (fp) { 1594 fhold(fp); 1595 fdp->fd_files[fd].fp = fp; 1596 fdp->fd_files[fd].reserved = 0; 1597 } else { 1598 fdp->fd_files[fd].reserved = 0; 1599 fdreserve_locked(fdp, fd, -1); 1600 fdfixup_locked(fdp, fd); 1601 } 1602 } 1603 1604 void 1605 fsetfd(struct filedesc *fdp, struct file *fp, int fd) 1606 { 1607 spin_lock(&fdp->fd_spin); 1608 fsetfd_locked(fdp, fp, fd); 1609 spin_unlock(&fdp->fd_spin); 1610 } 1611 1612 /* 1613 * (exclusive spinlock must be held on call) 1614 */ 1615 static 1616 struct file * 1617 funsetfd_locked(struct filedesc *fdp, int fd) 1618 { 1619 struct file *fp; 1620 1621 if ((unsigned)fd >= fdp->fd_nfiles) 1622 return (NULL); 1623 if ((fp = fdp->fd_files[fd].fp) == NULL) 1624 return (NULL); 1625 fdp->fd_files[fd].fp = NULL; 1626 fdp->fd_files[fd].fileflags = 0; 1627 1628 fdreserve_locked(fdp, fd, -1); 1629 fdfixup_locked(fdp, fd); 1630 return(fp); 1631 } 1632 1633 /* 1634 * WARNING: May not be called before initial fsetfd(). 1635 */ 1636 int 1637 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp) 1638 { 1639 int error; 1640 1641 spin_lock(&fdp->fd_spin); 1642 if (((u_int)fd) >= fdp->fd_nfiles) { 1643 error = EBADF; 1644 } else if (fdp->fd_files[fd].fp == NULL) { 1645 error = EBADF; 1646 } else { 1647 *flagsp = fdp->fd_files[fd].fileflags; 1648 error = 0; 1649 } 1650 spin_unlock(&fdp->fd_spin); 1651 return (error); 1652 } 1653 1654 /* 1655 * WARNING: May not be called before initial fsetfd(). 1656 */ 1657 int 1658 fsetfdflags(struct filedesc *fdp, int fd, int add_flags) 1659 { 1660 int error; 1661 1662 spin_lock(&fdp->fd_spin); 1663 if (((u_int)fd) >= fdp->fd_nfiles) { 1664 error = EBADF; 1665 } else if (fdp->fd_files[fd].fp == NULL) { 1666 error = EBADF; 1667 } else { 1668 fdp->fd_files[fd].fileflags |= add_flags; 1669 error = 0; 1670 } 1671 spin_unlock(&fdp->fd_spin); 1672 return (error); 1673 } 1674 1675 /* 1676 * WARNING: May not be called before initial fsetfd(). 1677 */ 1678 int 1679 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags) 1680 { 1681 int error; 1682 1683 spin_lock(&fdp->fd_spin); 1684 if (((u_int)fd) >= fdp->fd_nfiles) { 1685 error = EBADF; 1686 } else if (fdp->fd_files[fd].fp == NULL) { 1687 error = EBADF; 1688 } else { 1689 fdp->fd_files[fd].fileflags &= ~rem_flags; 1690 error = 0; 1691 } 1692 spin_unlock(&fdp->fd_spin); 1693 return (error); 1694 } 1695 1696 /* 1697 * Set/Change/Clear the creds for a fp and synchronize the uidinfo. 1698 */ 1699 void 1700 fsetcred(struct file *fp, struct ucred *ncr) 1701 { 1702 struct ucred *ocr; 1703 struct uidinfo *uip; 1704 1705 ocr = fp->f_cred; 1706 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) { 1707 if (ocr) { 1708 uip = ocr->cr_uidinfo; 1709 atomic_add_int(&uip->ui_openfiles, -1); 1710 } 1711 if (ncr) { 1712 uip = ncr->cr_uidinfo; 1713 atomic_add_int(&uip->ui_openfiles, 1); 1714 } 1715 } 1716 if (ncr) 1717 crhold(ncr); 1718 fp->f_cred = ncr; 1719 if (ocr) 1720 crfree(ocr); 1721 } 1722 1723 /* 1724 * Free a file descriptor. 1725 */ 1726 static 1727 void 1728 ffree(struct file *fp) 1729 { 1730 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 1731 fsetcred(fp, NULL); 1732 if (fp->f_nchandle.ncp) 1733 cache_drop(&fp->f_nchandle); 1734 kfree(fp, M_FILE); 1735 } 1736 1737 /* 1738 * called from init_main, initialize filedesc0 for proc0. 1739 */ 1740 void 1741 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask) 1742 { 1743 p0->p_fd = fdp0; 1744 p0->p_fdtol = NULL; 1745 fdp0->fd_refcnt = 1; 1746 fdp0->fd_cmask = cmask; 1747 fdp0->fd_files = fdp0->fd_builtin_files; 1748 fdp0->fd_nfiles = NDFILE; 1749 fdp0->fd_lastfile = -1; 1750 spin_init(&fdp0->fd_spin, "fdinitbootstrap"); 1751 } 1752 1753 /* 1754 * Build a new filedesc structure. 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, "fdinit"); 1793 1794 return (newfdp); 1795 } 1796 1797 /* 1798 * Share a filedesc structure. 1799 */ 1800 struct filedesc * 1801 fdshare(struct proc *p) 1802 { 1803 struct filedesc *fdp; 1804 1805 fdp = p->p_fd; 1806 spin_lock(&fdp->fd_spin); 1807 fdp->fd_refcnt++; 1808 spin_unlock(&fdp->fd_spin); 1809 return (fdp); 1810 } 1811 1812 /* 1813 * Copy a filedesc structure. 1814 */ 1815 int 1816 fdcopy(struct proc *p, struct filedesc **fpp) 1817 { 1818 struct filedesc *fdp = p->p_fd; 1819 struct filedesc *newfdp; 1820 struct fdnode *fdnode; 1821 int i; 1822 int ni; 1823 1824 /* 1825 * Certain daemons might not have file descriptors. 1826 */ 1827 if (fdp == NULL) 1828 return (0); 1829 1830 /* 1831 * Allocate the new filedesc and fd_files[] array. This can race 1832 * with operations by other threads on the fdp so we have to be 1833 * careful. 1834 */ 1835 newfdp = kmalloc(sizeof(struct filedesc), 1836 M_FILEDESC, M_WAITOK | M_ZERO | M_NULLOK); 1837 if (newfdp == NULL) { 1838 *fpp = NULL; 1839 return (-1); 1840 } 1841 again: 1842 spin_lock(&fdp->fd_spin); 1843 if (fdp->fd_lastfile < NDFILE) { 1844 newfdp->fd_files = newfdp->fd_builtin_files; 1845 i = NDFILE; 1846 } else { 1847 /* 1848 * We have to allocate (N^2-1) entries for our in-place 1849 * binary tree. Allow the table to shrink. 1850 */ 1851 i = fdp->fd_nfiles; 1852 ni = (i - 1) / 2; 1853 while (ni > fdp->fd_lastfile && ni > NDFILE) { 1854 i = ni; 1855 ni = (i - 1) / 2; 1856 } 1857 spin_unlock(&fdp->fd_spin); 1858 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode), 1859 M_FILEDESC, M_WAITOK | M_ZERO); 1860 1861 /* 1862 * Check for race, retry 1863 */ 1864 spin_lock(&fdp->fd_spin); 1865 if (i <= fdp->fd_lastfile) { 1866 spin_unlock(&fdp->fd_spin); 1867 kfree(newfdp->fd_files, M_FILEDESC); 1868 goto again; 1869 } 1870 } 1871 1872 /* 1873 * Dup the remaining fields. vref() and cache_hold() can be 1874 * safely called while holding the read spinlock on fdp. 1875 * 1876 * The read spinlock on fdp is still being held. 1877 * 1878 * NOTE: vref and cache_hold calls for the case where the vnode 1879 * or cache entry already has at least one ref may be called 1880 * while holding spin locks. 1881 */ 1882 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) { 1883 vref(newfdp->fd_cdir); 1884 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1885 } 1886 /* 1887 * We must check for fd_rdir here, at least for now because 1888 * the init process is created before we have access to the 1889 * rootvode to take a reference to it. 1890 */ 1891 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) { 1892 vref(newfdp->fd_rdir); 1893 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1894 } 1895 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) { 1896 vref(newfdp->fd_jdir); 1897 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1898 } 1899 newfdp->fd_refcnt = 1; 1900 newfdp->fd_nfiles = i; 1901 newfdp->fd_lastfile = fdp->fd_lastfile; 1902 newfdp->fd_freefile = fdp->fd_freefile; 1903 newfdp->fd_cmask = fdp->fd_cmask; 1904 spin_init(&newfdp->fd_spin, "fdcopy"); 1905 1906 /* 1907 * Copy the descriptor table through (i). This also copies the 1908 * allocation state. Then go through and ref the file pointers 1909 * and clean up any KQ descriptors. 1910 * 1911 * kq descriptors cannot be copied. Since we haven't ref'd the 1912 * copied files yet we can ignore the return value from funsetfd(). 1913 * 1914 * The read spinlock on fdp is still being held. 1915 */ 1916 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode)); 1917 for (i = 0 ; i < newfdp->fd_nfiles; ++i) { 1918 fdnode = &newfdp->fd_files[i]; 1919 if (fdnode->reserved) { 1920 fdreserve_locked(newfdp, i, -1); 1921 fdnode->reserved = 0; 1922 fdfixup_locked(newfdp, i); 1923 } else if (fdnode->fp) { 1924 if (fdnode->fp->f_type == DTYPE_KQUEUE) { 1925 (void)funsetfd_locked(newfdp, i); 1926 } else { 1927 fhold(fdnode->fp); 1928 } 1929 } 1930 } 1931 spin_unlock(&fdp->fd_spin); 1932 *fpp = newfdp; 1933 return (0); 1934 } 1935 1936 /* 1937 * Release a filedesc structure. 1938 * 1939 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE) 1940 */ 1941 void 1942 fdfree(struct proc *p, struct filedesc *repl) 1943 { 1944 struct filedesc *fdp; 1945 struct fdnode *fdnode; 1946 int i; 1947 struct filedesc_to_leader *fdtol; 1948 struct file *fp; 1949 struct vnode *vp; 1950 struct flock lf; 1951 1952 /* 1953 * Certain daemons might not have file descriptors. 1954 */ 1955 fdp = p->p_fd; 1956 if (fdp == NULL) { 1957 p->p_fd = repl; 1958 return; 1959 } 1960 1961 /* 1962 * Severe messing around to follow. 1963 */ 1964 spin_lock(&fdp->fd_spin); 1965 1966 /* Check for special need to clear POSIX style locks */ 1967 fdtol = p->p_fdtol; 1968 if (fdtol != NULL) { 1969 KASSERT(fdtol->fdl_refcount > 0, 1970 ("filedesc_to_refcount botch: fdl_refcount=%d", 1971 fdtol->fdl_refcount)); 1972 if (fdtol->fdl_refcount == 1 && 1973 (p->p_leader->p_flags & P_ADVLOCK) != 0) { 1974 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1975 fdnode = &fdp->fd_files[i]; 1976 if (fdnode->fp == NULL || 1977 fdnode->fp->f_type != DTYPE_VNODE) { 1978 continue; 1979 } 1980 fp = fdnode->fp; 1981 fhold(fp); 1982 spin_unlock(&fdp->fd_spin); 1983 1984 lf.l_whence = SEEK_SET; 1985 lf.l_start = 0; 1986 lf.l_len = 0; 1987 lf.l_type = F_UNLCK; 1988 vp = (struct vnode *)fp->f_data; 1989 (void) VOP_ADVLOCK(vp, 1990 (caddr_t)p->p_leader, 1991 F_UNLCK, 1992 &lf, 1993 F_POSIX); 1994 fdrop(fp); 1995 spin_lock(&fdp->fd_spin); 1996 } 1997 } 1998 retry: 1999 if (fdtol->fdl_refcount == 1) { 2000 if (fdp->fd_holdleaderscount > 0 && 2001 (p->p_leader->p_flags & P_ADVLOCK) != 0) { 2002 /* 2003 * close() or do_dup() has cleared a reference 2004 * in a shared file descriptor table. 2005 */ 2006 fdp->fd_holdleaderswakeup = 1; 2007 ssleep(&fdp->fd_holdleaderscount, 2008 &fdp->fd_spin, 0, "fdlhold", 0); 2009 goto retry; 2010 } 2011 if (fdtol->fdl_holdcount > 0) { 2012 /* 2013 * Ensure that fdtol->fdl_leader 2014 * remains valid in closef(). 2015 */ 2016 fdtol->fdl_wakeup = 1; 2017 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0); 2018 goto retry; 2019 } 2020 } 2021 fdtol->fdl_refcount--; 2022 if (fdtol->fdl_refcount == 0 && 2023 fdtol->fdl_holdcount == 0) { 2024 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2025 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2026 } else { 2027 fdtol = NULL; 2028 } 2029 p->p_fdtol = NULL; 2030 if (fdtol != NULL) { 2031 spin_unlock(&fdp->fd_spin); 2032 kfree(fdtol, M_FILEDESC_TO_LEADER); 2033 spin_lock(&fdp->fd_spin); 2034 } 2035 } 2036 if (--fdp->fd_refcnt > 0) { 2037 spin_unlock(&fdp->fd_spin); 2038 spin_lock(&p->p_spin); 2039 p->p_fd = repl; 2040 spin_unlock(&p->p_spin); 2041 return; 2042 } 2043 2044 /* 2045 * Even though we are the last reference to the structure allproc 2046 * scans may still reference the structure. Maintain proper 2047 * locks until we can replace p->p_fd. 2048 * 2049 * Also note that kqueue's closef still needs to reference the 2050 * fdp via p->p_fd, so we have to close the descriptors before 2051 * we replace p->p_fd. 2052 */ 2053 for (i = 0; i <= fdp->fd_lastfile; ++i) { 2054 if (fdp->fd_files[i].fp) { 2055 fp = funsetfd_locked(fdp, i); 2056 if (fp) { 2057 spin_unlock(&fdp->fd_spin); 2058 if (SLIST_FIRST(&fp->f_klist)) 2059 knote_fdclose(fp, fdp, i); 2060 closef(fp, p); 2061 spin_lock(&fdp->fd_spin); 2062 } 2063 } 2064 } 2065 spin_unlock(&fdp->fd_spin); 2066 2067 /* 2068 * Interlock against an allproc scan operations (typically frevoke). 2069 */ 2070 spin_lock(&p->p_spin); 2071 p->p_fd = repl; 2072 spin_unlock(&p->p_spin); 2073 2074 /* 2075 * Wait for any softrefs to go away. This race rarely occurs so 2076 * we can use a non-critical-path style poll/sleep loop. The 2077 * race only occurs against allproc scans. 2078 * 2079 * No new softrefs can occur with the fdp disconnected from the 2080 * process. 2081 */ 2082 if (fdp->fd_softrefs) { 2083 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid); 2084 while (fdp->fd_softrefs) 2085 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1); 2086 } 2087 2088 if (fdp->fd_files != fdp->fd_builtin_files) 2089 kfree(fdp->fd_files, M_FILEDESC); 2090 if (fdp->fd_cdir) { 2091 cache_drop(&fdp->fd_ncdir); 2092 vrele(fdp->fd_cdir); 2093 } 2094 if (fdp->fd_rdir) { 2095 cache_drop(&fdp->fd_nrdir); 2096 vrele(fdp->fd_rdir); 2097 } 2098 if (fdp->fd_jdir) { 2099 cache_drop(&fdp->fd_njdir); 2100 vrele(fdp->fd_jdir); 2101 } 2102 kfree(fdp, M_FILEDESC); 2103 } 2104 2105 /* 2106 * Retrieve and reference the file pointer associated with a descriptor. 2107 */ 2108 struct file * 2109 holdfp(struct filedesc *fdp, int fd, int flag) 2110 { 2111 struct file* fp; 2112 2113 spin_lock_shared(&fdp->fd_spin); 2114 if (((u_int)fd) >= fdp->fd_nfiles) { 2115 fp = NULL; 2116 goto done; 2117 } 2118 if ((fp = fdp->fd_files[fd].fp) == NULL) 2119 goto done; 2120 if ((fp->f_flag & flag) == 0 && flag != -1) { 2121 fp = NULL; 2122 goto done; 2123 } 2124 fhold(fp); 2125 done: 2126 spin_unlock_shared(&fdp->fd_spin); 2127 return (fp); 2128 } 2129 2130 /* 2131 * holdsock() - load the struct file pointer associated 2132 * with a socket into *fpp. If an error occurs, non-zero 2133 * will be returned and *fpp will be set to NULL. 2134 */ 2135 int 2136 holdsock(struct filedesc *fdp, int fd, struct file **fpp) 2137 { 2138 struct file *fp; 2139 int error; 2140 2141 spin_lock_shared(&fdp->fd_spin); 2142 if ((unsigned)fd >= fdp->fd_nfiles) { 2143 error = EBADF; 2144 fp = NULL; 2145 goto done; 2146 } 2147 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2148 error = EBADF; 2149 goto done; 2150 } 2151 if (fp->f_type != DTYPE_SOCKET) { 2152 error = ENOTSOCK; 2153 goto done; 2154 } 2155 fhold(fp); 2156 error = 0; 2157 done: 2158 spin_unlock_shared(&fdp->fd_spin); 2159 *fpp = fp; 2160 return (error); 2161 } 2162 2163 /* 2164 * Convert a user file descriptor to a held file pointer. 2165 */ 2166 int 2167 holdvnode(struct filedesc *fdp, int fd, struct file **fpp) 2168 { 2169 struct file *fp; 2170 int error; 2171 2172 spin_lock_shared(&fdp->fd_spin); 2173 if ((unsigned)fd >= fdp->fd_nfiles) { 2174 error = EBADF; 2175 fp = NULL; 2176 goto done; 2177 } 2178 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2179 error = EBADF; 2180 goto done; 2181 } 2182 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 2183 fp = NULL; 2184 error = EINVAL; 2185 goto done; 2186 } 2187 fhold(fp); 2188 error = 0; 2189 done: 2190 spin_unlock_shared(&fdp->fd_spin); 2191 *fpp = fp; 2192 return (error); 2193 } 2194 2195 /* 2196 * For setugid programs, we don't want to people to use that setugidness 2197 * to generate error messages which write to a file which otherwise would 2198 * otherwise be off-limits to the process. 2199 * 2200 * This is a gross hack to plug the hole. A better solution would involve 2201 * a special vop or other form of generalized access control mechanism. We 2202 * go ahead and just reject all procfs file systems accesses as dangerous. 2203 * 2204 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2205 * sufficient. We also don't for check setugidness since we know we are. 2206 */ 2207 static int 2208 is_unsafe(struct file *fp) 2209 { 2210 if (fp->f_type == DTYPE_VNODE && 2211 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 2212 return (1); 2213 return (0); 2214 } 2215 2216 /* 2217 * Make this setguid thing safe, if at all possible. 2218 * 2219 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2220 */ 2221 void 2222 setugidsafety(struct proc *p) 2223 { 2224 struct filedesc *fdp = p->p_fd; 2225 int i; 2226 2227 /* Certain daemons might not have file descriptors. */ 2228 if (fdp == NULL) 2229 return; 2230 2231 /* 2232 * note: fdp->fd_files may be reallocated out from under us while 2233 * we are blocked in a close. Be careful! 2234 */ 2235 for (i = 0; i <= fdp->fd_lastfile; i++) { 2236 if (i > 2) 2237 break; 2238 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) { 2239 struct file *fp; 2240 2241 /* 2242 * NULL-out descriptor prior to close to avoid 2243 * a race while close blocks. 2244 */ 2245 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2246 knote_fdclose(fp, fdp, i); 2247 closef(fp, p); 2248 } 2249 } 2250 } 2251 } 2252 2253 /* 2254 * Close any files on exec? 2255 * 2256 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2257 */ 2258 void 2259 fdcloseexec(struct proc *p) 2260 { 2261 struct filedesc *fdp = p->p_fd; 2262 int i; 2263 2264 /* Certain daemons might not have file descriptors. */ 2265 if (fdp == NULL) 2266 return; 2267 2268 /* 2269 * We cannot cache fd_files since operations may block and rip 2270 * them out from under us. 2271 */ 2272 for (i = 0; i <= fdp->fd_lastfile; i++) { 2273 if (fdp->fd_files[i].fp != NULL && 2274 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) { 2275 struct file *fp; 2276 2277 /* 2278 * NULL-out descriptor prior to close to avoid 2279 * a race while close blocks. 2280 */ 2281 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2282 knote_fdclose(fp, fdp, i); 2283 closef(fp, p); 2284 } 2285 } 2286 } 2287 } 2288 2289 /* 2290 * It is unsafe for set[ug]id processes to be started with file 2291 * descriptors 0..2 closed, as these descriptors are given implicit 2292 * significance in the Standard C library. fdcheckstd() will create a 2293 * descriptor referencing /dev/null for each of stdin, stdout, and 2294 * stderr that is not already open. 2295 * 2296 * NOT MPSAFE - calls falloc, vn_open, etc 2297 */ 2298 int 2299 fdcheckstd(struct lwp *lp) 2300 { 2301 struct nlookupdata nd; 2302 struct filedesc *fdp; 2303 struct file *fp; 2304 int retval; 2305 int i, error, flags, devnull; 2306 2307 fdp = lp->lwp_proc->p_fd; 2308 if (fdp == NULL) 2309 return (0); 2310 devnull = -1; 2311 error = 0; 2312 for (i = 0; i < 3; i++) { 2313 if (fdp->fd_files[i].fp != NULL) 2314 continue; 2315 if (devnull < 0) { 2316 if ((error = falloc(lp, &fp, &devnull)) != 0) 2317 break; 2318 2319 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, 2320 NLC_FOLLOW|NLC_LOCKVP); 2321 flags = FREAD | FWRITE; 2322 if (error == 0) 2323 error = vn_open(&nd, fp, flags, 0); 2324 if (error == 0) 2325 fsetfd(fdp, fp, devnull); 2326 else 2327 fsetfd(fdp, NULL, devnull); 2328 fdrop(fp); 2329 nlookup_done(&nd); 2330 if (error) 2331 break; 2332 KKASSERT(i == devnull); 2333 } else { 2334 error = kern_dup(DUP_FIXED, devnull, i, &retval); 2335 if (error != 0) 2336 break; 2337 } 2338 } 2339 return (error); 2340 } 2341 2342 /* 2343 * Internal form of close. 2344 * Decrement reference count on file structure. 2345 * Note: td and/or p may be NULL when closing a file 2346 * that was being passed in a message. 2347 * 2348 * MPALMOSTSAFE - acquires mplock for VOP operations 2349 */ 2350 int 2351 closef(struct file *fp, struct proc *p) 2352 { 2353 struct vnode *vp; 2354 struct flock lf; 2355 struct filedesc_to_leader *fdtol; 2356 2357 if (fp == NULL) 2358 return (0); 2359 2360 /* 2361 * POSIX record locking dictates that any close releases ALL 2362 * locks owned by this process. This is handled by setting 2363 * a flag in the unlock to free ONLY locks obeying POSIX 2364 * semantics, and not to free BSD-style file locks. 2365 * If the descriptor was in a message, POSIX-style locks 2366 * aren't passed with the descriptor. 2367 */ 2368 if (p != NULL && fp->f_type == DTYPE_VNODE && 2369 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2370 ) { 2371 if ((p->p_leader->p_flags & P_ADVLOCK) != 0) { 2372 lf.l_whence = SEEK_SET; 2373 lf.l_start = 0; 2374 lf.l_len = 0; 2375 lf.l_type = F_UNLCK; 2376 vp = (struct vnode *)fp->f_data; 2377 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 2378 &lf, F_POSIX); 2379 } 2380 fdtol = p->p_fdtol; 2381 if (fdtol != NULL) { 2382 lwkt_gettoken(&p->p_token); 2383 /* 2384 * Handle special case where file descriptor table 2385 * is shared between multiple process leaders. 2386 */ 2387 for (fdtol = fdtol->fdl_next; 2388 fdtol != p->p_fdtol; 2389 fdtol = fdtol->fdl_next) { 2390 if ((fdtol->fdl_leader->p_flags & 2391 P_ADVLOCK) == 0) 2392 continue; 2393 fdtol->fdl_holdcount++; 2394 lf.l_whence = SEEK_SET; 2395 lf.l_start = 0; 2396 lf.l_len = 0; 2397 lf.l_type = F_UNLCK; 2398 vp = (struct vnode *)fp->f_data; 2399 (void) VOP_ADVLOCK(vp, 2400 (caddr_t)fdtol->fdl_leader, 2401 F_UNLCK, &lf, F_POSIX); 2402 fdtol->fdl_holdcount--; 2403 if (fdtol->fdl_holdcount == 0 && 2404 fdtol->fdl_wakeup != 0) { 2405 fdtol->fdl_wakeup = 0; 2406 wakeup(fdtol); 2407 } 2408 } 2409 lwkt_reltoken(&p->p_token); 2410 } 2411 } 2412 return (fdrop(fp)); 2413 } 2414 2415 /* 2416 * fhold() can only be called if f_count is already at least 1 (i.e. the 2417 * caller of fhold() already has a reference to the file pointer in some 2418 * manner or other). 2419 * 2420 * Atomic ops are used for incrementing and decrementing f_count before 2421 * the 1->0 transition. f_count 1->0 transition is special, see the 2422 * comment in fdrop(). 2423 */ 2424 void 2425 fhold(struct file *fp) 2426 { 2427 /* 0->1 transition will never work */ 2428 KASSERT(fp->f_count > 0, ("fhold: invalid f_count %d", fp->f_count)); 2429 atomic_add_int(&fp->f_count, 1); 2430 } 2431 2432 /* 2433 * fdrop() - drop a reference to a descriptor 2434 */ 2435 int 2436 fdrop(struct file *fp) 2437 { 2438 struct flock lf; 2439 struct vnode *vp; 2440 int error, do_free = 0; 2441 2442 /* 2443 * NOTE: 2444 * Simple atomic_fetchadd_int(f_count, -1) here will cause use- 2445 * after-free or double free (due to f_count 0->1 transition), if 2446 * fhold() is called on the fps found through filehead iteration. 2447 */ 2448 for (;;) { 2449 int count = fp->f_count; 2450 2451 cpu_ccfence(); 2452 KASSERT(count > 0, ("fdrop: invalid f_count %d", count)); 2453 if (count == 1) { 2454 /* 2455 * About to drop the last reference, hold the 2456 * filehead spin lock and drop it, so that no 2457 * one could see this fp through filehead anymore, 2458 * let alone fhold() this fp. 2459 */ 2460 spin_lock(&filehead_spin); 2461 if (atomic_cmpset_int(&fp->f_count, count, 0)) { 2462 LIST_REMOVE(fp, f_list); 2463 nfiles--; 2464 spin_unlock(&filehead_spin); 2465 do_free = 1; /* free this fp */ 2466 break; 2467 } 2468 spin_unlock(&filehead_spin); 2469 /* retry */ 2470 } else if (atomic_cmpset_int(&fp->f_count, count, count - 1)) { 2471 break; 2472 } 2473 /* retry */ 2474 } 2475 if (!do_free) 2476 return (0); 2477 2478 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL); 2479 2480 /* 2481 * The last reference has gone away, we own the fp structure free 2482 * and clear. 2483 */ 2484 if (fp->f_count < 0) 2485 panic("fdrop: count < 0"); 2486 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE && 2487 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2488 ) { 2489 lf.l_whence = SEEK_SET; 2490 lf.l_start = 0; 2491 lf.l_len = 0; 2492 lf.l_type = F_UNLCK; 2493 vp = (struct vnode *)fp->f_data; 2494 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2495 } 2496 if (fp->f_ops != &badfileops) 2497 error = fo_close(fp); 2498 else 2499 error = 0; 2500 ffree(fp); 2501 return (error); 2502 } 2503 2504 /* 2505 * Apply an advisory lock on a file descriptor. 2506 * 2507 * Just attempt to get a record lock of the requested type on 2508 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2509 * 2510 * MPALMOSTSAFE 2511 */ 2512 int 2513 sys_flock(struct flock_args *uap) 2514 { 2515 struct proc *p = curproc; 2516 struct file *fp; 2517 struct vnode *vp; 2518 struct flock lf; 2519 int error; 2520 2521 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 2522 return (EBADF); 2523 if (fp->f_type != DTYPE_VNODE) { 2524 error = EOPNOTSUPP; 2525 goto done; 2526 } 2527 vp = (struct vnode *)fp->f_data; 2528 lf.l_whence = SEEK_SET; 2529 lf.l_start = 0; 2530 lf.l_len = 0; 2531 if (uap->how & LOCK_UN) { 2532 lf.l_type = F_UNLCK; 2533 atomic_clear_int(&fp->f_flag, FHASLOCK); /* race ok */ 2534 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2535 goto done; 2536 } 2537 if (uap->how & LOCK_EX) 2538 lf.l_type = F_WRLCK; 2539 else if (uap->how & LOCK_SH) 2540 lf.l_type = F_RDLCK; 2541 else { 2542 error = EBADF; 2543 goto done; 2544 } 2545 if (uap->how & LOCK_NB) 2546 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0); 2547 else 2548 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT); 2549 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */ 2550 done: 2551 fdrop(fp); 2552 return (error); 2553 } 2554 2555 /* 2556 * File Descriptor pseudo-device driver (/dev/fd/). 2557 * 2558 * Opening minor device N dup()s the file (if any) connected to file 2559 * descriptor N belonging to the calling process. Note that this driver 2560 * consists of only the ``open()'' routine, because all subsequent 2561 * references to this file will be direct to the other driver. 2562 */ 2563 static int 2564 fdopen(struct dev_open_args *ap) 2565 { 2566 thread_t td = curthread; 2567 2568 KKASSERT(td->td_lwp != NULL); 2569 2570 /* 2571 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the 2572 * the file descriptor being sought for duplication. The error 2573 * return ensures that the vnode for this device will be released 2574 * by vn_open. Open will detect this special error and take the 2575 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2576 * will simply report the error. 2577 */ 2578 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev); 2579 return (ENODEV); 2580 } 2581 2582 /* 2583 * The caller has reserved the file descriptor dfd for us. On success we 2584 * must fsetfd() it. On failure the caller will clean it up. 2585 */ 2586 int 2587 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error) 2588 { 2589 struct file *wfp; 2590 struct file *xfp; 2591 int werror; 2592 2593 if ((wfp = holdfp(fdp, sfd, -1)) == NULL) 2594 return (EBADF); 2595 2596 /* 2597 * Close a revoke/dup race. Duping a descriptor marked as revoked 2598 * will dup a dummy descriptor instead of the real one. 2599 */ 2600 if (wfp->f_flag & FREVOKED) { 2601 kprintf("Warning: attempt to dup() a revoked descriptor\n"); 2602 fdrop(wfp); 2603 wfp = NULL; 2604 werror = falloc(NULL, &wfp, NULL); 2605 if (werror) 2606 return (werror); 2607 } 2608 2609 /* 2610 * There are two cases of interest here. 2611 * 2612 * For ENODEV simply dup sfd to file descriptor dfd and return. 2613 * 2614 * For ENXIO steal away the file structure from sfd and store it 2615 * dfd. sfd is effectively closed by this operation. 2616 * 2617 * Any other error code is just returned. 2618 */ 2619 switch (error) { 2620 case ENODEV: 2621 /* 2622 * Check that the mode the file is being opened for is a 2623 * subset of the mode of the existing descriptor. 2624 */ 2625 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2626 error = EACCES; 2627 break; 2628 } 2629 spin_lock(&fdp->fd_spin); 2630 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2631 fsetfd_locked(fdp, wfp, dfd); 2632 spin_unlock(&fdp->fd_spin); 2633 error = 0; 2634 break; 2635 case ENXIO: 2636 /* 2637 * Steal away the file pointer from dfd, and stuff it into indx. 2638 */ 2639 spin_lock(&fdp->fd_spin); 2640 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2641 fsetfd(fdp, wfp, dfd); 2642 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) { 2643 spin_unlock(&fdp->fd_spin); 2644 fdrop(xfp); 2645 } else { 2646 spin_unlock(&fdp->fd_spin); 2647 } 2648 error = 0; 2649 break; 2650 default: 2651 break; 2652 } 2653 fdrop(wfp); 2654 return (error); 2655 } 2656 2657 /* 2658 * NOT MPSAFE - I think these refer to a common file descriptor table 2659 * and we need to spinlock that to link fdtol in. 2660 */ 2661 struct filedesc_to_leader * 2662 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 2663 struct proc *leader) 2664 { 2665 struct filedesc_to_leader *fdtol; 2666 2667 fdtol = kmalloc(sizeof(struct filedesc_to_leader), 2668 M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO); 2669 fdtol->fdl_refcount = 1; 2670 fdtol->fdl_holdcount = 0; 2671 fdtol->fdl_wakeup = 0; 2672 fdtol->fdl_leader = leader; 2673 if (old != NULL) { 2674 fdtol->fdl_next = old->fdl_next; 2675 fdtol->fdl_prev = old; 2676 old->fdl_next = fdtol; 2677 fdtol->fdl_next->fdl_prev = fdtol; 2678 } else { 2679 fdtol->fdl_next = fdtol; 2680 fdtol->fdl_prev = fdtol; 2681 } 2682 return fdtol; 2683 } 2684 2685 /* 2686 * Scan all file pointers in the system. The callback is made with 2687 * the master list spinlock held exclusively. 2688 */ 2689 void 2690 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data) 2691 { 2692 struct file *fp; 2693 int res; 2694 2695 spin_lock(&filehead_spin); 2696 LIST_FOREACH(fp, &filehead, f_list) { 2697 res = callback(fp, data); 2698 if (res < 0) 2699 break; 2700 } 2701 spin_unlock(&filehead_spin); 2702 } 2703 2704 /* 2705 * Get file structures. 2706 * 2707 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe) 2708 */ 2709 2710 struct sysctl_kern_file_info { 2711 int count; 2712 int error; 2713 struct sysctl_req *req; 2714 }; 2715 2716 static int sysctl_kern_file_callback(struct proc *p, void *data); 2717 2718 static int 2719 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2720 { 2721 struct sysctl_kern_file_info info; 2722 2723 /* 2724 * Note: because the number of file descriptors is calculated 2725 * in different ways for sizing vs returning the data, 2726 * there is information leakage from the first loop. However, 2727 * it is of a similar order of magnitude to the leakage from 2728 * global system statistics such as kern.openfiles. 2729 * 2730 * When just doing a count, note that we cannot just count 2731 * the elements and add f_count via the filehead list because 2732 * threaded processes share their descriptor table and f_count might 2733 * still be '1' in that case. 2734 * 2735 * Since the SYSCTL op can block, we must hold the process to 2736 * prevent it being ripped out from under us either in the 2737 * file descriptor loop or in the greater LIST_FOREACH. The 2738 * process may be in varying states of disrepair. If the process 2739 * is in SZOMB we may have caught it just as it is being removed 2740 * from the allproc list, we must skip it in that case to maintain 2741 * an unbroken chain through the allproc list. 2742 */ 2743 info.count = 0; 2744 info.error = 0; 2745 info.req = req; 2746 allproc_scan(sysctl_kern_file_callback, &info); 2747 2748 /* 2749 * When just calculating the size, overestimate a bit to try to 2750 * prevent system activity from causing the buffer-fill call 2751 * to fail later on. 2752 */ 2753 if (req->oldptr == NULL) { 2754 info.count = (info.count + 16) + (info.count / 10); 2755 info.error = SYSCTL_OUT(req, NULL, 2756 info.count * sizeof(struct kinfo_file)); 2757 } 2758 return (info.error); 2759 } 2760 2761 static int 2762 sysctl_kern_file_callback(struct proc *p, void *data) 2763 { 2764 struct sysctl_kern_file_info *info = data; 2765 struct kinfo_file kf; 2766 struct filedesc *fdp; 2767 struct file *fp; 2768 uid_t uid; 2769 int n; 2770 2771 if (p->p_stat == SIDL || p->p_stat == SZOMB) 2772 return(0); 2773 if (!(PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)) 2774 return(0); 2775 2776 /* 2777 * Softref the fdp to prevent it from being destroyed 2778 */ 2779 spin_lock(&p->p_spin); 2780 if ((fdp = p->p_fd) == NULL) { 2781 spin_unlock(&p->p_spin); 2782 return(0); 2783 } 2784 atomic_add_int(&fdp->fd_softrefs, 1); 2785 spin_unlock(&p->p_spin); 2786 2787 /* 2788 * The fdp's own spinlock prevents the contents from being 2789 * modified. 2790 */ 2791 spin_lock_shared(&fdp->fd_spin); 2792 for (n = 0; n < fdp->fd_nfiles; ++n) { 2793 if ((fp = fdp->fd_files[n].fp) == NULL) 2794 continue; 2795 if (info->req->oldptr == NULL) { 2796 ++info->count; 2797 } else { 2798 uid = p->p_ucred ? p->p_ucred->cr_uid : -1; 2799 kcore_make_file(&kf, fp, p->p_pid, uid, n); 2800 spin_unlock_shared(&fdp->fd_spin); 2801 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf)); 2802 spin_lock_shared(&fdp->fd_spin); 2803 if (info->error) 2804 break; 2805 } 2806 } 2807 spin_unlock_shared(&fdp->fd_spin); 2808 atomic_subtract_int(&fdp->fd_softrefs, 1); 2809 if (info->error) 2810 return(-1); 2811 return(0); 2812 } 2813 2814 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2815 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 2816 2817 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW, 2818 &minfilesperproc, 0, "Minimum files allowed open per process"); 2819 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2820 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2821 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW, 2822 &maxfilesperuser, 0, "Maximum files allowed open per user"); 2823 2824 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2825 &maxfiles, 0, "Maximum number of files"); 2826 2827 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, 2828 &maxfilesrootres, 0, "Descriptors reserved for root use"); 2829 2830 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2831 &nfiles, 0, "System-wide number of open files"); 2832 2833 static void 2834 fildesc_drvinit(void *unused) 2835 { 2836 int fd; 2837 2838 for (fd = 0; fd < NUMFDESC; fd++) { 2839 make_dev(&fildesc_ops, fd, 2840 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 2841 } 2842 2843 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 2844 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 2845 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 2846 } 2847 2848 struct fileops badfileops = { 2849 .fo_read = badfo_readwrite, 2850 .fo_write = badfo_readwrite, 2851 .fo_ioctl = badfo_ioctl, 2852 .fo_kqfilter = badfo_kqfilter, 2853 .fo_stat = badfo_stat, 2854 .fo_close = badfo_close, 2855 .fo_shutdown = badfo_shutdown 2856 }; 2857 2858 int 2859 badfo_readwrite( 2860 struct file *fp, 2861 struct uio *uio, 2862 struct ucred *cred, 2863 int flags 2864 ) { 2865 return (EBADF); 2866 } 2867 2868 int 2869 badfo_ioctl(struct file *fp, u_long com, caddr_t data, 2870 struct ucred *cred, struct sysmsg *msgv) 2871 { 2872 return (EBADF); 2873 } 2874 2875 /* 2876 * Must return an error to prevent registration, typically 2877 * due to a revoked descriptor (file_filtops assigned). 2878 */ 2879 int 2880 badfo_kqfilter(struct file *fp, struct knote *kn) 2881 { 2882 return (EOPNOTSUPP); 2883 } 2884 2885 int 2886 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred) 2887 { 2888 return (EBADF); 2889 } 2890 2891 int 2892 badfo_close(struct file *fp) 2893 { 2894 return (EBADF); 2895 } 2896 2897 int 2898 badfo_shutdown(struct file *fp, int how) 2899 { 2900 return (EBADF); 2901 } 2902 2903 int 2904 nofo_shutdown(struct file *fp, int how) 2905 { 2906 return (EOPNOTSUPP); 2907 } 2908 2909 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE + CDEV_MAJOR, 2910 fildesc_drvinit,NULL); 2911