1 /* 2 * Copyright (c) 2005 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Jeffrey Hsu. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * 35 * Copyright (c) 1982, 1986, 1989, 1991, 1993 36 * The Regents of the University of California. All rights reserved. 37 * (c) UNIX System Laboratories, Inc. 38 * All or some portions of this file are derived from material licensed 39 * to the University of California by American Telephone and Telegraph 40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 41 * the permission of UNIX System Laboratories, Inc. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed by the University of 54 * California, Berkeley and its contributors. 55 * 4. Neither the name of the University nor the names of its contributors 56 * may be used to endorse or promote products derived from this software 57 * without specific prior written permission. 58 * 59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 69 * SUCH DAMAGE. 70 * 71 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 72 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $ 73 * $DragonFly: src/sys/kern/kern_descrip.c,v 1.79 2008/08/31 13:18:28 aggelos Exp $ 74 */ 75 76 #include "opt_compat.h" 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/malloc.h> 80 #include <sys/sysproto.h> 81 #include <sys/conf.h> 82 #include <sys/device.h> 83 #include <sys/file.h> 84 #include <sys/filedesc.h> 85 #include <sys/kernel.h> 86 #include <sys/sysctl.h> 87 #include <sys/vnode.h> 88 #include <sys/proc.h> 89 #include <sys/nlookup.h> 90 #include <sys/file.h> 91 #include <sys/stat.h> 92 #include <sys/filio.h> 93 #include <sys/fcntl.h> 94 #include <sys/unistd.h> 95 #include <sys/resourcevar.h> 96 #include <sys/event.h> 97 #include <sys/kern_syscall.h> 98 #include <sys/kcore.h> 99 #include <sys/kinfo.h> 100 #include <sys/un.h> 101 102 #include <vm/vm.h> 103 #include <vm/vm_extern.h> 104 105 #include <sys/thread2.h> 106 #include <sys/file2.h> 107 #include <sys/spinlock2.h> 108 #include <sys/mplock2.h> 109 110 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd); 111 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr); 112 static struct file *funsetfd_locked (struct filedesc *fdp, int fd); 113 static int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp); 114 static void ffree(struct file *fp); 115 116 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 117 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 118 "file desc to leader structures"); 119 MALLOC_DEFINE(M_FILE, "file", "Open file structure"); 120 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 121 122 static d_open_t fdopen; 123 #define NUMFDESC 64 124 125 #define CDEV_MAJOR 22 126 static struct dev_ops fildesc_ops = { 127 { "FD", CDEV_MAJOR, 0 }, 128 .d_open = fdopen, 129 }; 130 131 /* 132 * Descriptor management. 133 */ 134 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead); 135 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin); 136 static int nfiles; /* actual number of open files */ 137 extern int cmask; 138 139 /* 140 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared. 141 * 142 * MPSAFE - must be called with fdp->fd_spin exclusively held 143 */ 144 static __inline 145 void 146 fdfixup_locked(struct filedesc *fdp, int fd) 147 { 148 if (fd < fdp->fd_freefile) { 149 fdp->fd_freefile = fd; 150 } 151 while (fdp->fd_lastfile >= 0 && 152 fdp->fd_files[fdp->fd_lastfile].fp == NULL && 153 fdp->fd_files[fdp->fd_lastfile].reserved == 0 154 ) { 155 --fdp->fd_lastfile; 156 } 157 } 158 159 /* 160 * System calls on descriptors. 161 * 162 * MPSAFE 163 */ 164 int 165 sys_getdtablesize(struct getdtablesize_args *uap) 166 { 167 struct proc *p = curproc; 168 struct plimit *limit = p->p_limit; 169 170 spin_lock_rd(&limit->p_spin); 171 uap->sysmsg_result = 172 min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 173 spin_unlock_rd(&limit->p_spin); 174 return (0); 175 } 176 177 /* 178 * Duplicate a file descriptor to a particular value. 179 * 180 * note: keep in mind that a potential race condition exists when closing 181 * descriptors from a shared descriptor table (via rfork). 182 * 183 * MPSAFE 184 */ 185 int 186 sys_dup2(struct dup2_args *uap) 187 { 188 int error; 189 int fd = 0; 190 191 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd); 192 uap->sysmsg_fds[0] = fd; 193 194 return (error); 195 } 196 197 /* 198 * Duplicate a file descriptor. 199 * 200 * MPSAFE 201 */ 202 int 203 sys_dup(struct dup_args *uap) 204 { 205 int error; 206 int fd = 0; 207 208 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd); 209 uap->sysmsg_fds[0] = fd; 210 211 return (error); 212 } 213 214 /* 215 * MPALMOSTSAFE - acquires mplock for fp operations 216 */ 217 int 218 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred) 219 { 220 struct thread *td = curthread; 221 struct proc *p = td->td_proc; 222 struct file *fp; 223 struct vnode *vp; 224 u_int newmin; 225 u_int oflags; 226 u_int nflags; 227 int tmp, error, flg = F_POSIX; 228 229 KKASSERT(p); 230 231 /* 232 * Operations on file descriptors that do not require a file pointer. 233 */ 234 switch (cmd) { 235 case F_GETFD: 236 error = fgetfdflags(p->p_fd, fd, &tmp); 237 if (error == 0) 238 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0; 239 return (error); 240 241 case F_SETFD: 242 if (dat->fc_cloexec & FD_CLOEXEC) 243 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE); 244 else 245 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE); 246 return (error); 247 case F_DUPFD: 248 newmin = dat->fc_fd; 249 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd); 250 return (error); 251 default: 252 break; 253 } 254 255 /* 256 * Operations on file pointers 257 */ 258 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 259 return (EBADF); 260 261 get_mplock(); 262 switch (cmd) { 263 case F_GETFL: 264 dat->fc_flags = OFLAGS(fp->f_flag); 265 error = 0; 266 break; 267 268 case F_SETFL: 269 oflags = fp->f_flag; 270 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 271 nflags |= oflags & ~FCNTLFLAGS; 272 273 error = 0; 274 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY)) 275 error = EINVAL; 276 if (error == 0 && ((nflags ^ oflags) & FASYNC)) { 277 tmp = nflags & FASYNC; 278 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, 279 cred, NULL); 280 } 281 if (error == 0) 282 fp->f_flag = nflags; 283 break; 284 285 case F_GETOWN: 286 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, 287 cred, NULL); 288 break; 289 290 case F_SETOWN: 291 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, 292 cred, NULL); 293 break; 294 295 case F_SETLKW: 296 flg |= F_WAIT; 297 /* Fall into F_SETLK */ 298 299 case F_SETLK: 300 if (fp->f_type != DTYPE_VNODE) { 301 error = EBADF; 302 break; 303 } 304 vp = (struct vnode *)fp->f_data; 305 306 /* 307 * copyin/lockop may block 308 */ 309 if (dat->fc_flock.l_whence == SEEK_CUR) 310 dat->fc_flock.l_start += fp->f_offset; 311 312 switch (dat->fc_flock.l_type) { 313 case F_RDLCK: 314 if ((fp->f_flag & FREAD) == 0) { 315 error = EBADF; 316 break; 317 } 318 p->p_leader->p_flag |= P_ADVLOCK; 319 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 320 &dat->fc_flock, flg); 321 break; 322 case F_WRLCK: 323 if ((fp->f_flag & FWRITE) == 0) { 324 error = EBADF; 325 break; 326 } 327 p->p_leader->p_flag |= P_ADVLOCK; 328 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 329 &dat->fc_flock, flg); 330 break; 331 case F_UNLCK: 332 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 333 &dat->fc_flock, F_POSIX); 334 break; 335 default: 336 error = EINVAL; 337 break; 338 } 339 340 /* 341 * It is possible to race a close() on the descriptor while 342 * we were blocked getting the lock. If this occurs the 343 * close might not have caught the lock. 344 */ 345 if (checkfpclosed(p->p_fd, fd, fp)) { 346 dat->fc_flock.l_whence = SEEK_SET; 347 dat->fc_flock.l_start = 0; 348 dat->fc_flock.l_len = 0; 349 dat->fc_flock.l_type = F_UNLCK; 350 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 351 F_UNLCK, &dat->fc_flock, F_POSIX); 352 } 353 break; 354 355 case F_GETLK: 356 if (fp->f_type != DTYPE_VNODE) { 357 error = EBADF; 358 break; 359 } 360 vp = (struct vnode *)fp->f_data; 361 /* 362 * copyin/lockop may block 363 */ 364 if (dat->fc_flock.l_type != F_RDLCK && 365 dat->fc_flock.l_type != F_WRLCK && 366 dat->fc_flock.l_type != F_UNLCK) { 367 error = EINVAL; 368 break; 369 } 370 if (dat->fc_flock.l_whence == SEEK_CUR) 371 dat->fc_flock.l_start += fp->f_offset; 372 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 373 &dat->fc_flock, F_POSIX); 374 break; 375 default: 376 error = EINVAL; 377 break; 378 } 379 rel_mplock(); 380 381 fdrop(fp); 382 return (error); 383 } 384 385 /* 386 * The file control system call. 387 * 388 * MPSAFE 389 */ 390 int 391 sys_fcntl(struct fcntl_args *uap) 392 { 393 union fcntl_dat dat; 394 int error; 395 396 switch (uap->cmd) { 397 case F_DUPFD: 398 dat.fc_fd = uap->arg; 399 break; 400 case F_SETFD: 401 dat.fc_cloexec = uap->arg; 402 break; 403 case F_SETFL: 404 dat.fc_flags = uap->arg; 405 break; 406 case F_SETOWN: 407 dat.fc_owner = uap->arg; 408 break; 409 case F_SETLKW: 410 case F_SETLK: 411 case F_GETLK: 412 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 413 sizeof(struct flock)); 414 if (error) 415 return (error); 416 break; 417 } 418 419 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred); 420 421 if (error == 0) { 422 switch (uap->cmd) { 423 case F_DUPFD: 424 uap->sysmsg_result = dat.fc_fd; 425 break; 426 case F_GETFD: 427 uap->sysmsg_result = dat.fc_cloexec; 428 break; 429 case F_GETFL: 430 uap->sysmsg_result = dat.fc_flags; 431 break; 432 case F_GETOWN: 433 uap->sysmsg_result = dat.fc_owner; 434 case F_GETLK: 435 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 436 sizeof(struct flock)); 437 break; 438 } 439 } 440 441 return (error); 442 } 443 444 /* 445 * Common code for dup, dup2, and fcntl(F_DUPFD). 446 * 447 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells 448 * kern_dup() to destructively dup over an existing file descriptor if new 449 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest 450 * unused file descriptor that is greater than or equal to new. 451 * 452 * MPSAFE 453 */ 454 int 455 kern_dup(enum dup_type type, int old, int new, int *res) 456 { 457 struct thread *td = curthread; 458 struct proc *p = td->td_proc; 459 struct filedesc *fdp = p->p_fd; 460 struct file *fp; 461 struct file *delfp; 462 int oldflags; 463 int holdleaders; 464 int error, newfd; 465 466 /* 467 * Verify that we have a valid descriptor to dup from and 468 * possibly to dup to. 469 */ 470 retry: 471 spin_lock_wr(&fdp->fd_spin); 472 if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 473 new >= maxfilesperproc) { 474 spin_unlock_wr(&fdp->fd_spin); 475 return (EINVAL); 476 } 477 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) { 478 spin_unlock_wr(&fdp->fd_spin); 479 return (EBADF); 480 } 481 if (type == DUP_FIXED && old == new) { 482 *res = new; 483 spin_unlock_wr(&fdp->fd_spin); 484 return (0); 485 } 486 fp = fdp->fd_files[old].fp; 487 oldflags = fdp->fd_files[old].fileflags; 488 fhold(fp); /* MPSAFE - can be called with a spinlock held */ 489 490 /* 491 * Allocate a new descriptor if DUP_VARIABLE, or expand the table 492 * if the requested descriptor is beyond the current table size. 493 * 494 * This can block. Retry if the source descriptor no longer matches 495 * or if our expectation in the expansion case races. 496 * 497 * If we are not expanding or allocating a new decriptor, then reset 498 * the target descriptor to a reserved state so we have a uniform 499 * setup for the next code block. 500 */ 501 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) { 502 spin_unlock_wr(&fdp->fd_spin); 503 error = fdalloc(p, new, &newfd); 504 spin_lock_wr(&fdp->fd_spin); 505 if (error) { 506 spin_unlock_wr(&fdp->fd_spin); 507 fdrop(fp); 508 return (error); 509 } 510 /* 511 * Check for ripout 512 */ 513 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) { 514 fsetfd_locked(fdp, NULL, newfd); 515 spin_unlock_wr(&fdp->fd_spin); 516 fdrop(fp); 517 goto retry; 518 } 519 /* 520 * Check for expansion race 521 */ 522 if (type != DUP_VARIABLE && new != newfd) { 523 fsetfd_locked(fdp, NULL, newfd); 524 spin_unlock_wr(&fdp->fd_spin); 525 fdrop(fp); 526 goto retry; 527 } 528 /* 529 * Check for ripout, newfd reused old (this case probably 530 * can't occur). 531 */ 532 if (old == newfd) { 533 fsetfd_locked(fdp, NULL, newfd); 534 spin_unlock_wr(&fdp->fd_spin); 535 fdrop(fp); 536 goto retry; 537 } 538 new = newfd; 539 delfp = NULL; 540 } else { 541 if (fdp->fd_files[new].reserved) { 542 spin_unlock_wr(&fdp->fd_spin); 543 fdrop(fp); 544 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new); 545 tsleep(fdp, 0, "fdres", hz); 546 goto retry; 547 } 548 549 /* 550 * If the target descriptor was never allocated we have 551 * to allocate it. If it was we have to clean out the 552 * old descriptor. delfp inherits the ref from the 553 * descriptor table. 554 */ 555 delfp = fdp->fd_files[new].fp; 556 fdp->fd_files[new].fp = NULL; 557 fdp->fd_files[new].reserved = 1; 558 if (delfp == NULL) { 559 fdreserve_locked(fdp, new, 1); 560 if (new > fdp->fd_lastfile) 561 fdp->fd_lastfile = new; 562 } 563 564 } 565 566 /* 567 * NOTE: still holding an exclusive spinlock 568 */ 569 570 /* 571 * If a descriptor is being overwritten we may hve to tell 572 * fdfree() to sleep to ensure that all relevant process 573 * leaders can be traversed in closef(). 574 */ 575 if (delfp != NULL && p->p_fdtol != NULL) { 576 fdp->fd_holdleaderscount++; 577 holdleaders = 1; 578 } else { 579 holdleaders = 0; 580 } 581 KASSERT(delfp == NULL || type == DUP_FIXED, 582 ("dup() picked an open file")); 583 584 /* 585 * Duplicate the source descriptor, update lastfile. If the new 586 * descriptor was not allocated and we aren't replacing an existing 587 * descriptor we have to mark the descriptor as being in use. 588 * 589 * The fd_files[] array inherits fp's hold reference. 590 */ 591 fsetfd_locked(fdp, fp, new); 592 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE; 593 spin_unlock_wr(&fdp->fd_spin); 594 fdrop(fp); 595 *res = new; 596 597 /* 598 * If we dup'd over a valid file, we now own the reference to it 599 * and must dispose of it using closef() semantics (as if a 600 * close() were performed on it). 601 */ 602 if (delfp) { 603 closef(delfp, p); 604 if (holdleaders) { 605 spin_lock_wr(&fdp->fd_spin); 606 fdp->fd_holdleaderscount--; 607 if (fdp->fd_holdleaderscount == 0 && 608 fdp->fd_holdleaderswakeup != 0) { 609 fdp->fd_holdleaderswakeup = 0; 610 spin_unlock_wr(&fdp->fd_spin); 611 wakeup(&fdp->fd_holdleaderscount); 612 } else { 613 spin_unlock_wr(&fdp->fd_spin); 614 } 615 } 616 } 617 return (0); 618 } 619 620 /* 621 * If sigio is on the list associated with a process or process group, 622 * disable signalling from the device, remove sigio from the list and 623 * free sigio. 624 */ 625 void 626 funsetown(struct sigio *sigio) 627 { 628 if (sigio == NULL) 629 return; 630 crit_enter(); 631 *(sigio->sio_myref) = NULL; 632 crit_exit(); 633 if (sigio->sio_pgid < 0) { 634 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 635 sigio, sio_pgsigio); 636 } else /* if ((*sigiop)->sio_pgid > 0) */ { 637 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 638 sigio, sio_pgsigio); 639 } 640 crfree(sigio->sio_ucred); 641 kfree(sigio, M_SIGIO); 642 } 643 644 /* Free a list of sigio structures. */ 645 void 646 funsetownlst(struct sigiolst *sigiolst) 647 { 648 struct sigio *sigio; 649 650 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) 651 funsetown(sigio); 652 } 653 654 /* 655 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 656 * 657 * After permission checking, add a sigio structure to the sigio list for 658 * the process or process group. 659 */ 660 int 661 fsetown(pid_t pgid, struct sigio **sigiop) 662 { 663 struct proc *proc; 664 struct pgrp *pgrp; 665 struct sigio *sigio; 666 667 if (pgid == 0) { 668 funsetown(*sigiop); 669 return (0); 670 } 671 if (pgid > 0) { 672 proc = pfind(pgid); 673 if (proc == NULL) 674 return (ESRCH); 675 676 /* 677 * Policy - Don't allow a process to FSETOWN a process 678 * in another session. 679 * 680 * Remove this test to allow maximum flexibility or 681 * restrict FSETOWN to the current process or process 682 * group for maximum safety. 683 */ 684 if (proc->p_session != curproc->p_session) 685 return (EPERM); 686 687 pgrp = NULL; 688 } else /* if (pgid < 0) */ { 689 pgrp = pgfind(-pgid); 690 if (pgrp == NULL) 691 return (ESRCH); 692 693 /* 694 * Policy - Don't allow a process to FSETOWN a process 695 * in another session. 696 * 697 * Remove this test to allow maximum flexibility or 698 * restrict FSETOWN to the current process or process 699 * group for maximum safety. 700 */ 701 if (pgrp->pg_session != curproc->p_session) 702 return (EPERM); 703 704 proc = NULL; 705 } 706 funsetown(*sigiop); 707 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 708 if (pgid > 0) { 709 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 710 sigio->sio_proc = proc; 711 } else { 712 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 713 sigio->sio_pgrp = pgrp; 714 } 715 sigio->sio_pgid = pgid; 716 sigio->sio_ucred = crhold(curthread->td_ucred); 717 /* It would be convenient if p_ruid was in ucred. */ 718 sigio->sio_ruid = sigio->sio_ucred->cr_ruid; 719 sigio->sio_myref = sigiop; 720 crit_enter(); 721 *sigiop = sigio; 722 crit_exit(); 723 return (0); 724 } 725 726 /* 727 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 728 */ 729 pid_t 730 fgetown(struct sigio *sigio) 731 { 732 return (sigio != NULL ? sigio->sio_pgid : 0); 733 } 734 735 /* 736 * Close many file descriptors. 737 * 738 * MPSAFE 739 */ 740 int 741 sys_closefrom(struct closefrom_args *uap) 742 { 743 return(kern_closefrom(uap->fd)); 744 } 745 746 /* 747 * Close all file descriptors greater then or equal to fd 748 * 749 * MPSAFE 750 */ 751 int 752 kern_closefrom(int fd) 753 { 754 struct thread *td = curthread; 755 struct proc *p = td->td_proc; 756 struct filedesc *fdp; 757 758 KKASSERT(p); 759 fdp = p->p_fd; 760 761 if (fd < 0) 762 return (EINVAL); 763 764 /* 765 * NOTE: This function will skip unassociated descriptors and 766 * reserved descriptors that have not yet been assigned. 767 * fd_lastfile can change as a side effect of kern_close(). 768 */ 769 spin_lock_wr(&fdp->fd_spin); 770 while (fd <= fdp->fd_lastfile) { 771 if (fdp->fd_files[fd].fp != NULL) { 772 spin_unlock_wr(&fdp->fd_spin); 773 /* ok if this races another close */ 774 if (kern_close(fd) == EINTR) 775 return (EINTR); 776 spin_lock_wr(&fdp->fd_spin); 777 } 778 ++fd; 779 } 780 spin_unlock_wr(&fdp->fd_spin); 781 return (0); 782 } 783 784 /* 785 * Close a file descriptor. 786 * 787 * MPSAFE 788 */ 789 int 790 sys_close(struct close_args *uap) 791 { 792 return(kern_close(uap->fd)); 793 } 794 795 /* 796 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls 797 */ 798 int 799 kern_close(int fd) 800 { 801 struct thread *td = curthread; 802 struct proc *p = td->td_proc; 803 struct filedesc *fdp; 804 struct file *fp; 805 int error; 806 int holdleaders; 807 808 KKASSERT(p); 809 fdp = p->p_fd; 810 811 spin_lock_wr(&fdp->fd_spin); 812 if ((fp = funsetfd_locked(fdp, fd)) == NULL) { 813 spin_unlock_wr(&fdp->fd_spin); 814 return (EBADF); 815 } 816 holdleaders = 0; 817 if (p->p_fdtol != NULL) { 818 /* 819 * Ask fdfree() to sleep to ensure that all relevant 820 * process leaders can be traversed in closef(). 821 */ 822 fdp->fd_holdleaderscount++; 823 holdleaders = 1; 824 } 825 826 /* 827 * we now hold the fp reference that used to be owned by the descriptor 828 * array. 829 */ 830 spin_unlock_wr(&fdp->fd_spin); 831 if (SLIST_FIRST(&fp->f_klist)) { 832 get_mplock(); 833 knote_fdclose(fp, fdp, fd); 834 rel_mplock(); 835 } 836 error = closef(fp, p); 837 if (holdleaders) { 838 spin_lock_wr(&fdp->fd_spin); 839 fdp->fd_holdleaderscount--; 840 if (fdp->fd_holdleaderscount == 0 && 841 fdp->fd_holdleaderswakeup != 0) { 842 fdp->fd_holdleaderswakeup = 0; 843 spin_unlock_wr(&fdp->fd_spin); 844 wakeup(&fdp->fd_holdleaderscount); 845 } else { 846 spin_unlock_wr(&fdp->fd_spin); 847 } 848 } 849 return (error); 850 } 851 852 /* 853 * shutdown_args(int fd, int how) 854 */ 855 int 856 kern_shutdown(int fd, int how) 857 { 858 struct thread *td = curthread; 859 struct proc *p = td->td_proc; 860 struct file *fp; 861 int error; 862 863 KKASSERT(p); 864 865 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 866 return (EBADF); 867 error = fo_shutdown(fp, how); 868 fdrop(fp); 869 870 return (error); 871 } 872 873 /* 874 * MPALMOSTSAFE 875 */ 876 int 877 sys_shutdown(struct shutdown_args *uap) 878 { 879 int error; 880 881 get_mplock(); 882 error = kern_shutdown(uap->s, uap->how); 883 rel_mplock(); 884 885 return (error); 886 } 887 888 /* 889 * MPSAFE 890 */ 891 int 892 kern_fstat(int fd, struct stat *ub) 893 { 894 struct thread *td = curthread; 895 struct proc *p = td->td_proc; 896 struct file *fp; 897 int error; 898 899 KKASSERT(p); 900 901 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 902 return (EBADF); 903 error = fo_stat(fp, ub, td->td_ucred); 904 fdrop(fp); 905 906 return (error); 907 } 908 909 /* 910 * Return status information about a file descriptor. 911 * 912 * MPSAFE 913 */ 914 int 915 sys_fstat(struct fstat_args *uap) 916 { 917 struct stat st; 918 int error; 919 920 error = kern_fstat(uap->fd, &st); 921 922 if (error == 0) 923 error = copyout(&st, uap->sb, sizeof(st)); 924 return (error); 925 } 926 927 /* 928 * Return pathconf information about a file descriptor. 929 * 930 * MPALMOSTSAFE 931 */ 932 int 933 sys_fpathconf(struct fpathconf_args *uap) 934 { 935 struct thread *td = curthread; 936 struct proc *p = td->td_proc; 937 struct file *fp; 938 struct vnode *vp; 939 int error = 0; 940 941 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 942 return (EBADF); 943 944 switch (fp->f_type) { 945 case DTYPE_PIPE: 946 case DTYPE_SOCKET: 947 if (uap->name != _PC_PIPE_BUF) { 948 error = EINVAL; 949 } else { 950 uap->sysmsg_result = PIPE_BUF; 951 error = 0; 952 } 953 break; 954 case DTYPE_FIFO: 955 case DTYPE_VNODE: 956 vp = (struct vnode *)fp->f_data; 957 get_mplock(); 958 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg); 959 rel_mplock(); 960 break; 961 default: 962 error = EOPNOTSUPP; 963 break; 964 } 965 fdrop(fp); 966 return(error); 967 } 968 969 static int fdexpand; 970 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, ""); 971 972 /* 973 * Grow the file table so it can hold through descriptor (want). 974 * 975 * The fdp's spinlock must be held exclusively on entry and may be held 976 * exclusively on return. The spinlock may be cycled by the routine. 977 * 978 * MPSAFE 979 */ 980 static void 981 fdgrow_locked(struct filedesc *fdp, int want) 982 { 983 struct fdnode *newfiles; 984 struct fdnode *oldfiles; 985 int nf, extra; 986 987 nf = fdp->fd_nfiles; 988 do { 989 /* nf has to be of the form 2^n - 1 */ 990 nf = 2 * nf + 1; 991 } while (nf <= want); 992 993 spin_unlock_wr(&fdp->fd_spin); 994 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK); 995 spin_lock_wr(&fdp->fd_spin); 996 997 /* 998 * We could have raced another extend while we were not holding 999 * the spinlock. 1000 */ 1001 if (fdp->fd_nfiles >= nf) { 1002 spin_unlock_wr(&fdp->fd_spin); 1003 kfree(newfiles, M_FILEDESC); 1004 spin_lock_wr(&fdp->fd_spin); 1005 return; 1006 } 1007 /* 1008 * Copy the existing ofile and ofileflags arrays 1009 * and zero the new portion of each array. 1010 */ 1011 extra = nf - fdp->fd_nfiles; 1012 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode)); 1013 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode)); 1014 1015 oldfiles = fdp->fd_files; 1016 fdp->fd_files = newfiles; 1017 fdp->fd_nfiles = nf; 1018 1019 if (oldfiles != fdp->fd_builtin_files) { 1020 spin_unlock_wr(&fdp->fd_spin); 1021 kfree(oldfiles, M_FILEDESC); 1022 spin_lock_wr(&fdp->fd_spin); 1023 } 1024 fdexpand++; 1025 } 1026 1027 /* 1028 * Number of nodes in right subtree, including the root. 1029 */ 1030 static __inline int 1031 right_subtree_size(int n) 1032 { 1033 return (n ^ (n | (n + 1))); 1034 } 1035 1036 /* 1037 * Bigger ancestor. 1038 */ 1039 static __inline int 1040 right_ancestor(int n) 1041 { 1042 return (n | (n + 1)); 1043 } 1044 1045 /* 1046 * Smaller ancestor. 1047 */ 1048 static __inline int 1049 left_ancestor(int n) 1050 { 1051 return ((n & (n + 1)) - 1); 1052 } 1053 1054 /* 1055 * Traverse the in-place binary tree buttom-up adjusting the allocation 1056 * count so scans can determine where free descriptors are located. 1057 * 1058 * MPSAFE - caller must be holding an exclusive spinlock on fdp 1059 */ 1060 static 1061 void 1062 fdreserve_locked(struct filedesc *fdp, int fd, int incr) 1063 { 1064 while (fd >= 0) { 1065 fdp->fd_files[fd].allocated += incr; 1066 KKASSERT(fdp->fd_files[fd].allocated >= 0); 1067 fd = left_ancestor(fd); 1068 } 1069 } 1070 1071 /* 1072 * Reserve a file descriptor for the process. If no error occurs, the 1073 * caller MUST at some point call fsetfd() or assign a file pointer 1074 * or dispose of the reservation. 1075 * 1076 * MPSAFE 1077 */ 1078 int 1079 fdalloc(struct proc *p, int want, int *result) 1080 { 1081 struct filedesc *fdp = p->p_fd; 1082 int fd, rsize, rsum, node, lim; 1083 1084 spin_lock_rd(&p->p_limit->p_spin); 1085 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 1086 spin_unlock_rd(&p->p_limit->p_spin); 1087 if (want >= lim) 1088 return (EMFILE); 1089 spin_lock_wr(&fdp->fd_spin); 1090 if (want >= fdp->fd_nfiles) 1091 fdgrow_locked(fdp, want); 1092 1093 /* 1094 * Search for a free descriptor starting at the higher 1095 * of want or fd_freefile. If that fails, consider 1096 * expanding the ofile array. 1097 * 1098 * NOTE! the 'allocated' field is a cumulative recursive allocation 1099 * count. If we happen to see a value of 0 then we can shortcut 1100 * our search. Otherwise we run through through the tree going 1101 * down branches we know have free descriptor(s) until we hit a 1102 * leaf node. The leaf node will be free but will not necessarily 1103 * have an allocated field of 0. 1104 */ 1105 retry: 1106 /* move up the tree looking for a subtree with a free node */ 1107 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim); 1108 fd = right_ancestor(fd)) { 1109 if (fdp->fd_files[fd].allocated == 0) 1110 goto found; 1111 1112 rsize = right_subtree_size(fd); 1113 if (fdp->fd_files[fd].allocated == rsize) 1114 continue; /* right subtree full */ 1115 1116 /* 1117 * Free fd is in the right subtree of the tree rooted at fd. 1118 * Call that subtree R. Look for the smallest (leftmost) 1119 * subtree of R with an unallocated fd: continue moving 1120 * down the left branch until encountering a full left 1121 * subtree, then move to the right. 1122 */ 1123 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) { 1124 node = fd + rsize; 1125 rsum += fdp->fd_files[node].allocated; 1126 if (fdp->fd_files[fd].allocated == rsum + rsize) { 1127 fd = node; /* move to the right */ 1128 if (fdp->fd_files[node].allocated == 0) 1129 goto found; 1130 rsum = 0; 1131 } 1132 } 1133 goto found; 1134 } 1135 1136 /* 1137 * No space in current array. Expand? 1138 */ 1139 if (fdp->fd_nfiles >= lim) { 1140 spin_unlock_wr(&fdp->fd_spin); 1141 return (EMFILE); 1142 } 1143 fdgrow_locked(fdp, want); 1144 goto retry; 1145 1146 found: 1147 KKASSERT(fd < fdp->fd_nfiles); 1148 if (fd > fdp->fd_lastfile) 1149 fdp->fd_lastfile = fd; 1150 if (want <= fdp->fd_freefile) 1151 fdp->fd_freefile = fd; 1152 *result = fd; 1153 KKASSERT(fdp->fd_files[fd].fp == NULL); 1154 KKASSERT(fdp->fd_files[fd].reserved == 0); 1155 fdp->fd_files[fd].fileflags = 0; 1156 fdp->fd_files[fd].reserved = 1; 1157 fdreserve_locked(fdp, fd, 1); 1158 spin_unlock_wr(&fdp->fd_spin); 1159 return (0); 1160 } 1161 1162 /* 1163 * Check to see whether n user file descriptors 1164 * are available to the process p. 1165 * 1166 * MPSAFE 1167 */ 1168 int 1169 fdavail(struct proc *p, int n) 1170 { 1171 struct filedesc *fdp = p->p_fd; 1172 struct fdnode *fdnode; 1173 int i, lim, last; 1174 1175 spin_lock_rd(&p->p_limit->p_spin); 1176 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 1177 spin_unlock_rd(&p->p_limit->p_spin); 1178 1179 spin_lock_rd(&fdp->fd_spin); 1180 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) { 1181 spin_unlock_rd(&fdp->fd_spin); 1182 return (1); 1183 } 1184 last = min(fdp->fd_nfiles, lim); 1185 fdnode = &fdp->fd_files[fdp->fd_freefile]; 1186 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) { 1187 if (fdnode->fp == NULL && --n <= 0) { 1188 spin_unlock_rd(&fdp->fd_spin); 1189 return (1); 1190 } 1191 } 1192 spin_unlock_rd(&fdp->fd_spin); 1193 return (0); 1194 } 1195 1196 /* 1197 * Revoke open descriptors referencing (f_data, f_type) 1198 * 1199 * Any revoke executed within a prison is only able to 1200 * revoke descriptors for processes within that prison. 1201 * 1202 * Returns 0 on success or an error code. 1203 */ 1204 struct fdrevoke_info { 1205 void *data; 1206 short type; 1207 short unused; 1208 int count; 1209 int intransit; 1210 struct ucred *cred; 1211 struct file *nfp; 1212 }; 1213 1214 static int fdrevoke_check_callback(struct file *fp, void *vinfo); 1215 static int fdrevoke_proc_callback(struct proc *p, void *vinfo); 1216 1217 int 1218 fdrevoke(void *f_data, short f_type, struct ucred *cred) 1219 { 1220 struct fdrevoke_info info; 1221 int error; 1222 1223 bzero(&info, sizeof(info)); 1224 info.data = f_data; 1225 info.type = f_type; 1226 info.cred = cred; 1227 error = falloc(NULL, &info.nfp, NULL); 1228 if (error) 1229 return (error); 1230 1231 /* 1232 * Scan the file pointer table once. dups do not dup file pointers, 1233 * only descriptors, so there is no leak. Set FREVOKED on the fps 1234 * being revoked. 1235 */ 1236 allfiles_scan_exclusive(fdrevoke_check_callback, &info); 1237 1238 /* 1239 * If any fps were marked track down the related descriptors 1240 * and close them. Any dup()s at this point will notice 1241 * the FREVOKED already set in the fp and do the right thing. 1242 * 1243 * Any fps with non-zero msgcounts (aka sent over a unix-domain 1244 * socket) bumped the intransit counter and will require a 1245 * scan. Races against fps leaving the socket are closed by 1246 * the socket code checking for FREVOKED. 1247 */ 1248 if (info.count) 1249 allproc_scan(fdrevoke_proc_callback, &info); 1250 if (info.intransit) 1251 unp_revoke_gc(info.nfp); 1252 fdrop(info.nfp); 1253 return(0); 1254 } 1255 1256 /* 1257 * Locate matching file pointers directly. 1258 */ 1259 static int 1260 fdrevoke_check_callback(struct file *fp, void *vinfo) 1261 { 1262 struct fdrevoke_info *info = vinfo; 1263 1264 /* 1265 * File pointers already flagged for revokation are skipped. 1266 */ 1267 if (fp->f_flag & FREVOKED) 1268 return(0); 1269 1270 /* 1271 * If revoking from a prison file pointers created outside of 1272 * that prison, or file pointers without creds, cannot be revoked. 1273 */ 1274 if (info->cred->cr_prison && 1275 (fp->f_cred == NULL || 1276 info->cred->cr_prison != fp->f_cred->cr_prison)) { 1277 return(0); 1278 } 1279 1280 /* 1281 * If the file pointer matches then mark it for revocation. The 1282 * flag is currently only used by unp_revoke_gc(). 1283 * 1284 * info->count is a heuristic and can race in a SMP environment. 1285 */ 1286 if (info->data == fp->f_data && info->type == fp->f_type) { 1287 atomic_set_int(&fp->f_flag, FREVOKED); 1288 info->count += fp->f_count; 1289 if (fp->f_msgcount) 1290 ++info->intransit; 1291 } 1292 return(0); 1293 } 1294 1295 /* 1296 * Locate matching file pointers via process descriptor tables. 1297 */ 1298 static int 1299 fdrevoke_proc_callback(struct proc *p, void *vinfo) 1300 { 1301 struct fdrevoke_info *info = vinfo; 1302 struct filedesc *fdp; 1303 struct file *fp; 1304 int n; 1305 1306 if (p->p_stat == SIDL || p->p_stat == SZOMB) 1307 return(0); 1308 if (info->cred->cr_prison && 1309 info->cred->cr_prison != p->p_ucred->cr_prison) { 1310 return(0); 1311 } 1312 1313 /* 1314 * If the controlling terminal of the process matches the 1315 * vnode being revoked we clear the controlling terminal. 1316 * 1317 * The normal spec_close() may not catch this because it 1318 * uses curproc instead of p. 1319 */ 1320 if (p->p_session && info->type == DTYPE_VNODE && 1321 info->data == p->p_session->s_ttyvp) { 1322 p->p_session->s_ttyvp = NULL; 1323 vrele(info->data); 1324 } 1325 1326 /* 1327 * Softref the fdp to prevent it from being destroyed 1328 */ 1329 spin_lock_wr(&p->p_spin); 1330 if ((fdp = p->p_fd) == NULL) { 1331 spin_unlock_wr(&p->p_spin); 1332 return(0); 1333 } 1334 atomic_add_int(&fdp->fd_softrefs, 1); 1335 spin_unlock_wr(&p->p_spin); 1336 1337 /* 1338 * Locate and close any matching file descriptors. 1339 */ 1340 spin_lock_wr(&fdp->fd_spin); 1341 for (n = 0; n < fdp->fd_nfiles; ++n) { 1342 if ((fp = fdp->fd_files[n].fp) == NULL) 1343 continue; 1344 if (fp->f_flag & FREVOKED) { 1345 fhold(info->nfp); 1346 fdp->fd_files[n].fp = info->nfp; 1347 spin_unlock_wr(&fdp->fd_spin); 1348 knote_fdclose(fp, fdp, n); /* XXX */ 1349 closef(fp, p); 1350 spin_lock_wr(&fdp->fd_spin); 1351 --info->count; 1352 } 1353 } 1354 spin_unlock_wr(&fdp->fd_spin); 1355 atomic_subtract_int(&fdp->fd_softrefs, 1); 1356 return(0); 1357 } 1358 1359 /* 1360 * falloc: 1361 * Create a new open file structure and reserve a file decriptor 1362 * for the process that refers to it. 1363 * 1364 * Root creds are checked using lp, or assumed if lp is NULL. If 1365 * resultfd is non-NULL then lp must also be non-NULL. No file 1366 * descriptor is reserved (and no process context is needed) if 1367 * resultfd is NULL. 1368 * 1369 * A file pointer with a refcount of 1 is returned. Note that the 1370 * file pointer is NOT associated with the descriptor. If falloc 1371 * returns success, fsetfd() MUST be called to either associate the 1372 * file pointer or clear the reservation. 1373 * 1374 * MPSAFE 1375 */ 1376 int 1377 falloc(struct lwp *lp, struct file **resultfp, int *resultfd) 1378 { 1379 static struct timeval lastfail; 1380 static int curfail; 1381 struct file *fp; 1382 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred; 1383 int error; 1384 1385 fp = NULL; 1386 1387 /* 1388 * Handle filetable full issues and root overfill. 1389 */ 1390 if (nfiles >= maxfiles - maxfilesrootres && 1391 (cred->cr_ruid != 0 || nfiles >= maxfiles)) { 1392 if (ppsratecheck(&lastfail, &curfail, 1)) { 1393 kprintf("kern.maxfiles limit exceeded by uid %d, " 1394 "please see tuning(7).\n", 1395 cred->cr_ruid); 1396 } 1397 error = ENFILE; 1398 goto done; 1399 } 1400 1401 /* 1402 * Allocate a new file descriptor. 1403 */ 1404 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO); 1405 spin_init(&fp->f_spin); 1406 SLIST_INIT(&fp->f_klist); 1407 fp->f_count = 1; 1408 fp->f_ops = &badfileops; 1409 fp->f_seqcount = 1; 1410 fp->f_cred = crhold(cred); 1411 spin_lock_wr(&filehead_spin); 1412 nfiles++; 1413 LIST_INSERT_HEAD(&filehead, fp, f_list); 1414 spin_unlock_wr(&filehead_spin); 1415 if (resultfd) { 1416 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) { 1417 fdrop(fp); 1418 fp = NULL; 1419 } 1420 } else { 1421 error = 0; 1422 } 1423 done: 1424 *resultfp = fp; 1425 return (error); 1426 } 1427 1428 /* 1429 * MPSAFE 1430 */ 1431 static 1432 int 1433 checkfpclosed(struct filedesc *fdp, int fd, struct file *fp) 1434 { 1435 int error; 1436 1437 spin_lock_rd(&fdp->fd_spin); 1438 if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp) 1439 error = EBADF; 1440 else 1441 error = 0; 1442 spin_unlock_rd(&fdp->fd_spin); 1443 return (error); 1444 } 1445 1446 /* 1447 * Associate a file pointer with a previously reserved file descriptor. 1448 * This function always succeeds. 1449 * 1450 * If fp is NULL, the file descriptor is returned to the pool. 1451 */ 1452 1453 /* 1454 * MPSAFE (exclusive spinlock must be held on call) 1455 */ 1456 static void 1457 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd) 1458 { 1459 KKASSERT((unsigned)fd < fdp->fd_nfiles); 1460 KKASSERT(fdp->fd_files[fd].reserved != 0); 1461 if (fp) { 1462 fhold(fp); 1463 fdp->fd_files[fd].fp = fp; 1464 fdp->fd_files[fd].reserved = 0; 1465 } else { 1466 fdp->fd_files[fd].reserved = 0; 1467 fdreserve_locked(fdp, fd, -1); 1468 fdfixup_locked(fdp, fd); 1469 } 1470 } 1471 1472 /* 1473 * MPSAFE 1474 */ 1475 void 1476 fsetfd(struct filedesc *fdp, struct file *fp, int fd) 1477 { 1478 spin_lock_wr(&fdp->fd_spin); 1479 fsetfd_locked(fdp, fp, fd); 1480 spin_unlock_wr(&fdp->fd_spin); 1481 } 1482 1483 /* 1484 * MPSAFE (exclusive spinlock must be held on call) 1485 */ 1486 static 1487 struct file * 1488 funsetfd_locked(struct filedesc *fdp, int fd) 1489 { 1490 struct file *fp; 1491 1492 if ((unsigned)fd >= fdp->fd_nfiles) 1493 return (NULL); 1494 if ((fp = fdp->fd_files[fd].fp) == NULL) 1495 return (NULL); 1496 fdp->fd_files[fd].fp = NULL; 1497 fdp->fd_files[fd].fileflags = 0; 1498 1499 fdreserve_locked(fdp, fd, -1); 1500 fdfixup_locked(fdp, fd); 1501 return(fp); 1502 } 1503 1504 /* 1505 * MPSAFE 1506 */ 1507 int 1508 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp) 1509 { 1510 int error; 1511 1512 spin_lock_rd(&fdp->fd_spin); 1513 if (((u_int)fd) >= fdp->fd_nfiles) { 1514 error = EBADF; 1515 } else if (fdp->fd_files[fd].fp == NULL) { 1516 error = EBADF; 1517 } else { 1518 *flagsp = fdp->fd_files[fd].fileflags; 1519 error = 0; 1520 } 1521 spin_unlock_rd(&fdp->fd_spin); 1522 return (error); 1523 } 1524 1525 /* 1526 * MPSAFE 1527 */ 1528 int 1529 fsetfdflags(struct filedesc *fdp, int fd, int add_flags) 1530 { 1531 int error; 1532 1533 spin_lock_wr(&fdp->fd_spin); 1534 if (((u_int)fd) >= fdp->fd_nfiles) { 1535 error = EBADF; 1536 } else if (fdp->fd_files[fd].fp == NULL) { 1537 error = EBADF; 1538 } else { 1539 fdp->fd_files[fd].fileflags |= add_flags; 1540 error = 0; 1541 } 1542 spin_unlock_wr(&fdp->fd_spin); 1543 return (error); 1544 } 1545 1546 /* 1547 * MPSAFE 1548 */ 1549 int 1550 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags) 1551 { 1552 int error; 1553 1554 spin_lock_wr(&fdp->fd_spin); 1555 if (((u_int)fd) >= fdp->fd_nfiles) { 1556 error = EBADF; 1557 } else if (fdp->fd_files[fd].fp == NULL) { 1558 error = EBADF; 1559 } else { 1560 fdp->fd_files[fd].fileflags &= ~rem_flags; 1561 error = 0; 1562 } 1563 spin_unlock_wr(&fdp->fd_spin); 1564 return (error); 1565 } 1566 1567 void 1568 fsetcred(struct file *fp, struct ucred *cr) 1569 { 1570 crhold(cr); 1571 crfree(fp->f_cred); 1572 fp->f_cred = cr; 1573 } 1574 1575 /* 1576 * Free a file descriptor. 1577 */ 1578 static 1579 void 1580 ffree(struct file *fp) 1581 { 1582 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 1583 spin_lock_wr(&filehead_spin); 1584 LIST_REMOVE(fp, f_list); 1585 nfiles--; 1586 spin_unlock_wr(&filehead_spin); 1587 crfree(fp->f_cred); 1588 if (fp->f_nchandle.ncp) 1589 cache_drop(&fp->f_nchandle); 1590 kfree(fp, M_FILE); 1591 } 1592 1593 /* 1594 * called from init_main, initialize filedesc0 for proc0. 1595 */ 1596 void 1597 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask) 1598 { 1599 p0->p_fd = fdp0; 1600 p0->p_fdtol = NULL; 1601 fdp0->fd_refcnt = 1; 1602 fdp0->fd_cmask = cmask; 1603 fdp0->fd_files = fdp0->fd_builtin_files; 1604 fdp0->fd_nfiles = NDFILE; 1605 fdp0->fd_lastfile = -1; 1606 spin_init(&fdp0->fd_spin); 1607 } 1608 1609 /* 1610 * Build a new filedesc structure. 1611 * 1612 * NOT MPSAFE (vref) 1613 */ 1614 struct filedesc * 1615 fdinit(struct proc *p) 1616 { 1617 struct filedesc *newfdp; 1618 struct filedesc *fdp = p->p_fd; 1619 1620 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO); 1621 spin_lock_rd(&fdp->fd_spin); 1622 if (fdp->fd_cdir) { 1623 newfdp->fd_cdir = fdp->fd_cdir; 1624 vref(newfdp->fd_cdir); 1625 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1626 } 1627 1628 /* 1629 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of 1630 * proc0, but should unconditionally exist in other processes. 1631 */ 1632 if (fdp->fd_rdir) { 1633 newfdp->fd_rdir = fdp->fd_rdir; 1634 vref(newfdp->fd_rdir); 1635 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1636 } 1637 if (fdp->fd_jdir) { 1638 newfdp->fd_jdir = fdp->fd_jdir; 1639 vref(newfdp->fd_jdir); 1640 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1641 } 1642 spin_unlock_rd(&fdp->fd_spin); 1643 1644 /* Create the file descriptor table. */ 1645 newfdp->fd_refcnt = 1; 1646 newfdp->fd_cmask = cmask; 1647 newfdp->fd_files = newfdp->fd_builtin_files; 1648 newfdp->fd_nfiles = NDFILE; 1649 newfdp->fd_lastfile = -1; 1650 spin_init(&newfdp->fd_spin); 1651 1652 return (newfdp); 1653 } 1654 1655 /* 1656 * Share a filedesc structure. 1657 * 1658 * MPSAFE 1659 */ 1660 struct filedesc * 1661 fdshare(struct proc *p) 1662 { 1663 struct filedesc *fdp; 1664 1665 fdp = p->p_fd; 1666 spin_lock_wr(&fdp->fd_spin); 1667 fdp->fd_refcnt++; 1668 spin_unlock_wr(&fdp->fd_spin); 1669 return (fdp); 1670 } 1671 1672 /* 1673 * Copy a filedesc structure. 1674 * 1675 * MPSAFE 1676 */ 1677 struct filedesc * 1678 fdcopy(struct proc *p) 1679 { 1680 struct filedesc *fdp = p->p_fd; 1681 struct filedesc *newfdp; 1682 struct fdnode *fdnode; 1683 int i; 1684 int ni; 1685 1686 /* 1687 * Certain daemons might not have file descriptors. 1688 */ 1689 if (fdp == NULL) 1690 return (NULL); 1691 1692 /* 1693 * Allocate the new filedesc and fd_files[] array. This can race 1694 * with operations by other threads on the fdp so we have to be 1695 * careful. 1696 */ 1697 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO); 1698 again: 1699 spin_lock_rd(&fdp->fd_spin); 1700 if (fdp->fd_lastfile < NDFILE) { 1701 newfdp->fd_files = newfdp->fd_builtin_files; 1702 i = NDFILE; 1703 } else { 1704 /* 1705 * We have to allocate (N^2-1) entries for our in-place 1706 * binary tree. Allow the table to shrink. 1707 */ 1708 i = fdp->fd_nfiles; 1709 ni = (i - 1) / 2; 1710 while (ni > fdp->fd_lastfile && ni > NDFILE) { 1711 i = ni; 1712 ni = (i - 1) / 2; 1713 } 1714 spin_unlock_rd(&fdp->fd_spin); 1715 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode), 1716 M_FILEDESC, M_WAITOK | M_ZERO); 1717 1718 /* 1719 * Check for race, retry 1720 */ 1721 spin_lock_rd(&fdp->fd_spin); 1722 if (i <= fdp->fd_lastfile) { 1723 spin_unlock_rd(&fdp->fd_spin); 1724 kfree(newfdp->fd_files, M_FILEDESC); 1725 goto again; 1726 } 1727 } 1728 1729 /* 1730 * Dup the remaining fields. vref() and cache_hold() can be 1731 * safely called while holding the read spinlock on fdp. 1732 * 1733 * The read spinlock on fdp is still being held. 1734 * 1735 * NOTE: vref and cache_hold calls for the case where the vnode 1736 * or cache entry already has at least one ref may be called 1737 * while holding spin locks. 1738 */ 1739 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) { 1740 vref(newfdp->fd_cdir); 1741 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1742 } 1743 /* 1744 * We must check for fd_rdir here, at least for now because 1745 * the init process is created before we have access to the 1746 * rootvode to take a reference to it. 1747 */ 1748 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) { 1749 vref(newfdp->fd_rdir); 1750 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1751 } 1752 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) { 1753 vref(newfdp->fd_jdir); 1754 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1755 } 1756 newfdp->fd_refcnt = 1; 1757 newfdp->fd_nfiles = i; 1758 newfdp->fd_lastfile = fdp->fd_lastfile; 1759 newfdp->fd_freefile = fdp->fd_freefile; 1760 newfdp->fd_cmask = fdp->fd_cmask; 1761 spin_init(&newfdp->fd_spin); 1762 1763 /* 1764 * Copy the descriptor table through (i). This also copies the 1765 * allocation state. Then go through and ref the file pointers 1766 * and clean up any KQ descriptors. 1767 * 1768 * kq descriptors cannot be copied. Since we haven't ref'd the 1769 * copied files yet we can ignore the return value from funsetfd(). 1770 * 1771 * The read spinlock on fdp is still being held. 1772 */ 1773 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode)); 1774 for (i = 0 ; i < newfdp->fd_nfiles; ++i) { 1775 fdnode = &newfdp->fd_files[i]; 1776 if (fdnode->reserved) { 1777 fdreserve_locked(newfdp, i, -1); 1778 fdnode->reserved = 0; 1779 fdfixup_locked(newfdp, i); 1780 } else if (fdnode->fp) { 1781 if (fdnode->fp->f_type == DTYPE_KQUEUE) { 1782 (void)funsetfd_locked(newfdp, i); 1783 } else { 1784 fhold(fdnode->fp); 1785 } 1786 } 1787 } 1788 spin_unlock_rd(&fdp->fd_spin); 1789 return (newfdp); 1790 } 1791 1792 /* 1793 * Release a filedesc structure. 1794 * 1795 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE) 1796 */ 1797 void 1798 fdfree(struct proc *p, struct filedesc *repl) 1799 { 1800 struct filedesc *fdp; 1801 struct fdnode *fdnode; 1802 int i; 1803 struct filedesc_to_leader *fdtol; 1804 struct file *fp; 1805 struct vnode *vp; 1806 struct flock lf; 1807 1808 /* 1809 * Certain daemons might not have file descriptors. 1810 */ 1811 fdp = p->p_fd; 1812 if (fdp == NULL) { 1813 p->p_fd = repl; 1814 return; 1815 } 1816 1817 /* 1818 * Severe messing around to follow. 1819 */ 1820 spin_lock_wr(&fdp->fd_spin); 1821 1822 /* Check for special need to clear POSIX style locks */ 1823 fdtol = p->p_fdtol; 1824 if (fdtol != NULL) { 1825 KASSERT(fdtol->fdl_refcount > 0, 1826 ("filedesc_to_refcount botch: fdl_refcount=%d", 1827 fdtol->fdl_refcount)); 1828 if (fdtol->fdl_refcount == 1 && 1829 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1830 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1831 fdnode = &fdp->fd_files[i]; 1832 if (fdnode->fp == NULL || 1833 fdnode->fp->f_type != DTYPE_VNODE) { 1834 continue; 1835 } 1836 fp = fdnode->fp; 1837 fhold(fp); 1838 spin_unlock_wr(&fdp->fd_spin); 1839 1840 lf.l_whence = SEEK_SET; 1841 lf.l_start = 0; 1842 lf.l_len = 0; 1843 lf.l_type = F_UNLCK; 1844 vp = (struct vnode *)fp->f_data; 1845 (void) VOP_ADVLOCK(vp, 1846 (caddr_t)p->p_leader, 1847 F_UNLCK, 1848 &lf, 1849 F_POSIX); 1850 fdrop(fp); 1851 spin_lock_wr(&fdp->fd_spin); 1852 } 1853 } 1854 retry: 1855 if (fdtol->fdl_refcount == 1) { 1856 if (fdp->fd_holdleaderscount > 0 && 1857 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1858 /* 1859 * close() or do_dup() has cleared a reference 1860 * in a shared file descriptor table. 1861 */ 1862 fdp->fd_holdleaderswakeup = 1; 1863 ssleep(&fdp->fd_holdleaderscount, 1864 &fdp->fd_spin, 0, "fdlhold", 0); 1865 goto retry; 1866 } 1867 if (fdtol->fdl_holdcount > 0) { 1868 /* 1869 * Ensure that fdtol->fdl_leader 1870 * remains valid in closef(). 1871 */ 1872 fdtol->fdl_wakeup = 1; 1873 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0); 1874 goto retry; 1875 } 1876 } 1877 fdtol->fdl_refcount--; 1878 if (fdtol->fdl_refcount == 0 && 1879 fdtol->fdl_holdcount == 0) { 1880 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1881 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1882 } else { 1883 fdtol = NULL; 1884 } 1885 p->p_fdtol = NULL; 1886 if (fdtol != NULL) { 1887 spin_unlock_wr(&fdp->fd_spin); 1888 kfree(fdtol, M_FILEDESC_TO_LEADER); 1889 spin_lock_wr(&fdp->fd_spin); 1890 } 1891 } 1892 if (--fdp->fd_refcnt > 0) { 1893 spin_unlock_wr(&fdp->fd_spin); 1894 spin_lock_wr(&p->p_spin); 1895 p->p_fd = repl; 1896 spin_unlock_wr(&p->p_spin); 1897 return; 1898 } 1899 1900 /* 1901 * Even though we are the last reference to the structure allproc 1902 * scans may still reference the structure. Maintain proper 1903 * locks until we can replace p->p_fd. 1904 * 1905 * Also note that kqueue's closef still needs to reference the 1906 * fdp via p->p_fd, so we have to close the descriptors before 1907 * we replace p->p_fd. 1908 */ 1909 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1910 if (fdp->fd_files[i].fp) { 1911 fp = funsetfd_locked(fdp, i); 1912 if (fp) { 1913 spin_unlock_wr(&fdp->fd_spin); 1914 closef(fp, p); 1915 spin_lock_wr(&fdp->fd_spin); 1916 } 1917 } 1918 } 1919 spin_unlock_wr(&fdp->fd_spin); 1920 1921 /* 1922 * Interlock against an allproc scan operations (typically frevoke). 1923 */ 1924 spin_lock_wr(&p->p_spin); 1925 p->p_fd = repl; 1926 spin_unlock_wr(&p->p_spin); 1927 1928 /* 1929 * Wait for any softrefs to go away. This race rarely occurs so 1930 * we can use a non-critical-path style poll/sleep loop. The 1931 * race only occurs against allproc scans. 1932 * 1933 * No new softrefs can occur with the fdp disconnected from the 1934 * process. 1935 */ 1936 if (fdp->fd_softrefs) { 1937 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid); 1938 while (fdp->fd_softrefs) 1939 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1); 1940 } 1941 1942 if (fdp->fd_files != fdp->fd_builtin_files) 1943 kfree(fdp->fd_files, M_FILEDESC); 1944 if (fdp->fd_cdir) { 1945 cache_drop(&fdp->fd_ncdir); 1946 vrele(fdp->fd_cdir); 1947 } 1948 if (fdp->fd_rdir) { 1949 cache_drop(&fdp->fd_nrdir); 1950 vrele(fdp->fd_rdir); 1951 } 1952 if (fdp->fd_jdir) { 1953 cache_drop(&fdp->fd_njdir); 1954 vrele(fdp->fd_jdir); 1955 } 1956 kfree(fdp, M_FILEDESC); 1957 } 1958 1959 /* 1960 * Retrieve and reference the file pointer associated with a descriptor. 1961 * 1962 * MPSAFE 1963 */ 1964 struct file * 1965 holdfp(struct filedesc *fdp, int fd, int flag) 1966 { 1967 struct file* fp; 1968 1969 spin_lock_rd(&fdp->fd_spin); 1970 if (((u_int)fd) >= fdp->fd_nfiles) { 1971 fp = NULL; 1972 goto done; 1973 } 1974 if ((fp = fdp->fd_files[fd].fp) == NULL) 1975 goto done; 1976 if ((fp->f_flag & flag) == 0 && flag != -1) { 1977 fp = NULL; 1978 goto done; 1979 } 1980 fhold(fp); 1981 done: 1982 spin_unlock_rd(&fdp->fd_spin); 1983 return (fp); 1984 } 1985 1986 /* 1987 * holdsock() - load the struct file pointer associated 1988 * with a socket into *fpp. If an error occurs, non-zero 1989 * will be returned and *fpp will be set to NULL. 1990 * 1991 * MPSAFE 1992 */ 1993 int 1994 holdsock(struct filedesc *fdp, int fd, struct file **fpp) 1995 { 1996 struct file *fp; 1997 int error; 1998 1999 spin_lock_rd(&fdp->fd_spin); 2000 if ((unsigned)fd >= fdp->fd_nfiles) { 2001 error = EBADF; 2002 fp = NULL; 2003 goto done; 2004 } 2005 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2006 error = EBADF; 2007 goto done; 2008 } 2009 if (fp->f_type != DTYPE_SOCKET) { 2010 error = ENOTSOCK; 2011 goto done; 2012 } 2013 fhold(fp); 2014 error = 0; 2015 done: 2016 spin_unlock_rd(&fdp->fd_spin); 2017 *fpp = fp; 2018 return (error); 2019 } 2020 2021 /* 2022 * Convert a user file descriptor to a held file pointer. 2023 * 2024 * MPSAFE 2025 */ 2026 int 2027 holdvnode(struct filedesc *fdp, int fd, struct file **fpp) 2028 { 2029 struct file *fp; 2030 int error; 2031 2032 spin_lock_rd(&fdp->fd_spin); 2033 if ((unsigned)fd >= fdp->fd_nfiles) { 2034 error = EBADF; 2035 fp = NULL; 2036 goto done; 2037 } 2038 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2039 error = EBADF; 2040 goto done; 2041 } 2042 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 2043 fp = NULL; 2044 error = EINVAL; 2045 goto done; 2046 } 2047 fhold(fp); 2048 error = 0; 2049 done: 2050 spin_unlock_rd(&fdp->fd_spin); 2051 *fpp = fp; 2052 return (error); 2053 } 2054 2055 /* 2056 * For setugid programs, we don't want to people to use that setugidness 2057 * to generate error messages which write to a file which otherwise would 2058 * otherwise be off-limits to the process. 2059 * 2060 * This is a gross hack to plug the hole. A better solution would involve 2061 * a special vop or other form of generalized access control mechanism. We 2062 * go ahead and just reject all procfs file systems accesses as dangerous. 2063 * 2064 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2065 * sufficient. We also don't for check setugidness since we know we are. 2066 */ 2067 static int 2068 is_unsafe(struct file *fp) 2069 { 2070 if (fp->f_type == DTYPE_VNODE && 2071 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 2072 return (1); 2073 return (0); 2074 } 2075 2076 /* 2077 * Make this setguid thing safe, if at all possible. 2078 * 2079 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2080 */ 2081 void 2082 setugidsafety(struct proc *p) 2083 { 2084 struct filedesc *fdp = p->p_fd; 2085 int i; 2086 2087 /* Certain daemons might not have file descriptors. */ 2088 if (fdp == NULL) 2089 return; 2090 2091 /* 2092 * note: fdp->fd_files may be reallocated out from under us while 2093 * we are blocked in a close. Be careful! 2094 */ 2095 for (i = 0; i <= fdp->fd_lastfile; i++) { 2096 if (i > 2) 2097 break; 2098 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) { 2099 struct file *fp; 2100 2101 /* 2102 * NULL-out descriptor prior to close to avoid 2103 * a race while close blocks. 2104 */ 2105 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2106 knote_fdclose(fp, fdp, i); 2107 closef(fp, p); 2108 } 2109 } 2110 } 2111 } 2112 2113 /* 2114 * Close any files on exec? 2115 * 2116 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2117 */ 2118 void 2119 fdcloseexec(struct proc *p) 2120 { 2121 struct filedesc *fdp = p->p_fd; 2122 int i; 2123 2124 /* Certain daemons might not have file descriptors. */ 2125 if (fdp == NULL) 2126 return; 2127 2128 /* 2129 * We cannot cache fd_files since operations may block and rip 2130 * them out from under us. 2131 */ 2132 for (i = 0; i <= fdp->fd_lastfile; i++) { 2133 if (fdp->fd_files[i].fp != NULL && 2134 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) { 2135 struct file *fp; 2136 2137 /* 2138 * NULL-out descriptor prior to close to avoid 2139 * a race while close blocks. 2140 */ 2141 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2142 knote_fdclose(fp, fdp, i); 2143 closef(fp, p); 2144 } 2145 } 2146 } 2147 } 2148 2149 /* 2150 * It is unsafe for set[ug]id processes to be started with file 2151 * descriptors 0..2 closed, as these descriptors are given implicit 2152 * significance in the Standard C library. fdcheckstd() will create a 2153 * descriptor referencing /dev/null for each of stdin, stdout, and 2154 * stderr that is not already open. 2155 * 2156 * NOT MPSAFE - calls falloc, vn_open, etc 2157 */ 2158 int 2159 fdcheckstd(struct lwp *lp) 2160 { 2161 struct nlookupdata nd; 2162 struct filedesc *fdp; 2163 struct file *fp; 2164 int retval; 2165 int i, error, flags, devnull; 2166 2167 fdp = lp->lwp_proc->p_fd; 2168 if (fdp == NULL) 2169 return (0); 2170 devnull = -1; 2171 error = 0; 2172 for (i = 0; i < 3; i++) { 2173 if (fdp->fd_files[i].fp != NULL) 2174 continue; 2175 if (devnull < 0) { 2176 if ((error = falloc(lp, &fp, &devnull)) != 0) 2177 break; 2178 2179 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, 2180 NLC_FOLLOW|NLC_LOCKVP); 2181 flags = FREAD | FWRITE; 2182 if (error == 0) 2183 error = vn_open(&nd, fp, flags, 0); 2184 if (error == 0) 2185 fsetfd(fdp, fp, devnull); 2186 else 2187 fsetfd(fdp, NULL, devnull); 2188 fdrop(fp); 2189 nlookup_done(&nd); 2190 if (error) 2191 break; 2192 KKASSERT(i == devnull); 2193 } else { 2194 error = kern_dup(DUP_FIXED, devnull, i, &retval); 2195 if (error != 0) 2196 break; 2197 } 2198 } 2199 return (error); 2200 } 2201 2202 /* 2203 * Internal form of close. 2204 * Decrement reference count on file structure. 2205 * Note: td and/or p may be NULL when closing a file 2206 * that was being passed in a message. 2207 * 2208 * MPALMOSTSAFE - acquires mplock for VOP operations 2209 */ 2210 int 2211 closef(struct file *fp, struct proc *p) 2212 { 2213 struct vnode *vp; 2214 struct flock lf; 2215 struct filedesc_to_leader *fdtol; 2216 2217 if (fp == NULL) 2218 return (0); 2219 2220 /* 2221 * POSIX record locking dictates that any close releases ALL 2222 * locks owned by this process. This is handled by setting 2223 * a flag in the unlock to free ONLY locks obeying POSIX 2224 * semantics, and not to free BSD-style file locks. 2225 * If the descriptor was in a message, POSIX-style locks 2226 * aren't passed with the descriptor. 2227 */ 2228 if (p != NULL && fp->f_type == DTYPE_VNODE && 2229 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2230 ) { 2231 get_mplock(); 2232 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) { 2233 lf.l_whence = SEEK_SET; 2234 lf.l_start = 0; 2235 lf.l_len = 0; 2236 lf.l_type = F_UNLCK; 2237 vp = (struct vnode *)fp->f_data; 2238 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 2239 &lf, F_POSIX); 2240 } 2241 fdtol = p->p_fdtol; 2242 if (fdtol != NULL) { 2243 /* 2244 * Handle special case where file descriptor table 2245 * is shared between multiple process leaders. 2246 */ 2247 for (fdtol = fdtol->fdl_next; 2248 fdtol != p->p_fdtol; 2249 fdtol = fdtol->fdl_next) { 2250 if ((fdtol->fdl_leader->p_flag & 2251 P_ADVLOCK) == 0) 2252 continue; 2253 fdtol->fdl_holdcount++; 2254 lf.l_whence = SEEK_SET; 2255 lf.l_start = 0; 2256 lf.l_len = 0; 2257 lf.l_type = F_UNLCK; 2258 vp = (struct vnode *)fp->f_data; 2259 (void) VOP_ADVLOCK(vp, 2260 (caddr_t)fdtol->fdl_leader, 2261 F_UNLCK, &lf, F_POSIX); 2262 fdtol->fdl_holdcount--; 2263 if (fdtol->fdl_holdcount == 0 && 2264 fdtol->fdl_wakeup != 0) { 2265 fdtol->fdl_wakeup = 0; 2266 wakeup(fdtol); 2267 } 2268 } 2269 } 2270 rel_mplock(); 2271 } 2272 return (fdrop(fp)); 2273 } 2274 2275 /* 2276 * MPSAFE 2277 * 2278 * fhold() can only be called if f_count is already at least 1 (i.e. the 2279 * caller of fhold() already has a reference to the file pointer in some 2280 * manner or other). 2281 * 2282 * f_count is not spin-locked. Instead, atomic ops are used for 2283 * incrementing, decrementing, and handling the 1->0 transition. 2284 */ 2285 void 2286 fhold(struct file *fp) 2287 { 2288 atomic_add_int(&fp->f_count, 1); 2289 } 2290 2291 /* 2292 * fdrop() - drop a reference to a descriptor 2293 * 2294 * MPALMOSTSAFE - acquires mplock for final close sequence 2295 */ 2296 int 2297 fdrop(struct file *fp) 2298 { 2299 struct flock lf; 2300 struct vnode *vp; 2301 int error; 2302 2303 /* 2304 * A combined fetch and subtract is needed to properly detect 2305 * 1->0 transitions, otherwise two cpus dropping from a ref 2306 * count of 2 might both try to run the 1->0 code. 2307 */ 2308 if (atomic_fetchadd_int(&fp->f_count, -1) > 1) 2309 return (0); 2310 2311 get_mplock(); 2312 2313 /* 2314 * The last reference has gone away, we own the fp structure free 2315 * and clear. 2316 */ 2317 if (fp->f_count < 0) 2318 panic("fdrop: count < 0"); 2319 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE && 2320 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2321 ) { 2322 lf.l_whence = SEEK_SET; 2323 lf.l_start = 0; 2324 lf.l_len = 0; 2325 lf.l_type = F_UNLCK; 2326 vp = (struct vnode *)fp->f_data; 2327 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2328 } 2329 if (fp->f_ops != &badfileops) 2330 error = fo_close(fp); 2331 else 2332 error = 0; 2333 ffree(fp); 2334 rel_mplock(); 2335 return (error); 2336 } 2337 2338 /* 2339 * Apply an advisory lock on a file descriptor. 2340 * 2341 * Just attempt to get a record lock of the requested type on 2342 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2343 * 2344 * MPALMOSTSAFE 2345 */ 2346 int 2347 sys_flock(struct flock_args *uap) 2348 { 2349 struct proc *p = curproc; 2350 struct file *fp; 2351 struct vnode *vp; 2352 struct flock lf; 2353 int error; 2354 2355 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 2356 return (EBADF); 2357 get_mplock(); 2358 if (fp->f_type != DTYPE_VNODE) { 2359 error = EOPNOTSUPP; 2360 goto done; 2361 } 2362 vp = (struct vnode *)fp->f_data; 2363 lf.l_whence = SEEK_SET; 2364 lf.l_start = 0; 2365 lf.l_len = 0; 2366 if (uap->how & LOCK_UN) { 2367 lf.l_type = F_UNLCK; 2368 fp->f_flag &= ~FHASLOCK; 2369 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2370 goto done; 2371 } 2372 if (uap->how & LOCK_EX) 2373 lf.l_type = F_WRLCK; 2374 else if (uap->how & LOCK_SH) 2375 lf.l_type = F_RDLCK; 2376 else { 2377 error = EBADF; 2378 goto done; 2379 } 2380 fp->f_flag |= FHASLOCK; 2381 if (uap->how & LOCK_NB) 2382 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0); 2383 else 2384 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT); 2385 done: 2386 rel_mplock(); 2387 fdrop(fp); 2388 return (error); 2389 } 2390 2391 /* 2392 * File Descriptor pseudo-device driver (/dev/fd/). 2393 * 2394 * Opening minor device N dup()s the file (if any) connected to file 2395 * descriptor N belonging to the calling process. Note that this driver 2396 * consists of only the ``open()'' routine, because all subsequent 2397 * references to this file will be direct to the other driver. 2398 */ 2399 static int 2400 fdopen(struct dev_open_args *ap) 2401 { 2402 thread_t td = curthread; 2403 2404 KKASSERT(td->td_lwp != NULL); 2405 2406 /* 2407 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the 2408 * the file descriptor being sought for duplication. The error 2409 * return ensures that the vnode for this device will be released 2410 * by vn_open. Open will detect this special error and take the 2411 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2412 * will simply report the error. 2413 */ 2414 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev); 2415 return (ENODEV); 2416 } 2417 2418 /* 2419 * The caller has reserved the file descriptor dfd for us. On success we 2420 * must fsetfd() it. On failure the caller will clean it up. 2421 * 2422 * MPSAFE 2423 */ 2424 int 2425 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error) 2426 { 2427 struct file *wfp; 2428 struct file *xfp; 2429 int werror; 2430 2431 if ((wfp = holdfp(fdp, sfd, -1)) == NULL) 2432 return (EBADF); 2433 2434 /* 2435 * Close a revoke/dup race. Duping a descriptor marked as revoked 2436 * will dup a dummy descriptor instead of the real one. 2437 */ 2438 if (wfp->f_flag & FREVOKED) { 2439 kprintf("Warning: attempt to dup() a revoked descriptor\n"); 2440 fdrop(wfp); 2441 wfp = NULL; 2442 werror = falloc(NULL, &wfp, NULL); 2443 if (werror) 2444 return (werror); 2445 } 2446 2447 /* 2448 * There are two cases of interest here. 2449 * 2450 * For ENODEV simply dup sfd to file descriptor dfd and return. 2451 * 2452 * For ENXIO steal away the file structure from sfd and store it 2453 * dfd. sfd is effectively closed by this operation. 2454 * 2455 * Any other error code is just returned. 2456 */ 2457 switch (error) { 2458 case ENODEV: 2459 /* 2460 * Check that the mode the file is being opened for is a 2461 * subset of the mode of the existing descriptor. 2462 */ 2463 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2464 error = EACCES; 2465 break; 2466 } 2467 spin_lock_wr(&fdp->fd_spin); 2468 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2469 fsetfd_locked(fdp, wfp, dfd); 2470 spin_unlock_wr(&fdp->fd_spin); 2471 error = 0; 2472 break; 2473 case ENXIO: 2474 /* 2475 * Steal away the file pointer from dfd, and stuff it into indx. 2476 */ 2477 spin_lock_wr(&fdp->fd_spin); 2478 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2479 fsetfd(fdp, wfp, dfd); 2480 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) { 2481 spin_unlock_wr(&fdp->fd_spin); 2482 fdrop(xfp); 2483 } else { 2484 spin_unlock_wr(&fdp->fd_spin); 2485 } 2486 error = 0; 2487 break; 2488 default: 2489 break; 2490 } 2491 fdrop(wfp); 2492 return (error); 2493 } 2494 2495 /* 2496 * NOT MPSAFE - I think these refer to a common file descriptor table 2497 * and we need to spinlock that to link fdtol in. 2498 */ 2499 struct filedesc_to_leader * 2500 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 2501 struct proc *leader) 2502 { 2503 struct filedesc_to_leader *fdtol; 2504 2505 fdtol = kmalloc(sizeof(struct filedesc_to_leader), 2506 M_FILEDESC_TO_LEADER, M_WAITOK); 2507 fdtol->fdl_refcount = 1; 2508 fdtol->fdl_holdcount = 0; 2509 fdtol->fdl_wakeup = 0; 2510 fdtol->fdl_leader = leader; 2511 if (old != NULL) { 2512 fdtol->fdl_next = old->fdl_next; 2513 fdtol->fdl_prev = old; 2514 old->fdl_next = fdtol; 2515 fdtol->fdl_next->fdl_prev = fdtol; 2516 } else { 2517 fdtol->fdl_next = fdtol; 2518 fdtol->fdl_prev = fdtol; 2519 } 2520 return fdtol; 2521 } 2522 2523 /* 2524 * Scan all file pointers in the system. The callback is made with 2525 * the master list spinlock held exclusively. 2526 * 2527 * MPSAFE 2528 */ 2529 void 2530 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data) 2531 { 2532 struct file *fp; 2533 int res; 2534 2535 spin_lock_wr(&filehead_spin); 2536 LIST_FOREACH(fp, &filehead, f_list) { 2537 res = callback(fp, data); 2538 if (res < 0) 2539 break; 2540 } 2541 spin_unlock_wr(&filehead_spin); 2542 } 2543 2544 /* 2545 * Get file structures. 2546 * 2547 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe) 2548 */ 2549 2550 struct sysctl_kern_file_info { 2551 int count; 2552 int error; 2553 struct sysctl_req *req; 2554 }; 2555 2556 static int sysctl_kern_file_callback(struct proc *p, void *data); 2557 2558 static int 2559 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2560 { 2561 struct sysctl_kern_file_info info; 2562 2563 /* 2564 * Note: because the number of file descriptors is calculated 2565 * in different ways for sizing vs returning the data, 2566 * there is information leakage from the first loop. However, 2567 * it is of a similar order of magnitude to the leakage from 2568 * global system statistics such as kern.openfiles. 2569 * 2570 * When just doing a count, note that we cannot just count 2571 * the elements and add f_count via the filehead list because 2572 * threaded processes share their descriptor table and f_count might 2573 * still be '1' in that case. 2574 * 2575 * Since the SYSCTL op can block, we must hold the process to 2576 * prevent it being ripped out from under us either in the 2577 * file descriptor loop or in the greater LIST_FOREACH. The 2578 * process may be in varying states of disrepair. If the process 2579 * is in SZOMB we may have caught it just as it is being removed 2580 * from the allproc list, we must skip it in that case to maintain 2581 * an unbroken chain through the allproc list. 2582 */ 2583 info.count = 0; 2584 info.error = 0; 2585 info.req = req; 2586 allproc_scan(sysctl_kern_file_callback, &info); 2587 2588 /* 2589 * When just calculating the size, overestimate a bit to try to 2590 * prevent system activity from causing the buffer-fill call 2591 * to fail later on. 2592 */ 2593 if (req->oldptr == NULL) { 2594 info.count = (info.count + 16) + (info.count / 10); 2595 info.error = SYSCTL_OUT(req, NULL, 2596 info.count * sizeof(struct kinfo_file)); 2597 } 2598 return (info.error); 2599 } 2600 2601 static int 2602 sysctl_kern_file_callback(struct proc *p, void *data) 2603 { 2604 struct sysctl_kern_file_info *info = data; 2605 struct kinfo_file kf; 2606 struct filedesc *fdp; 2607 struct file *fp; 2608 uid_t uid; 2609 int n; 2610 2611 if (p->p_stat == SIDL || p->p_stat == SZOMB) 2612 return(0); 2613 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0) 2614 return(0); 2615 2616 /* 2617 * Softref the fdp to prevent it from being destroyed 2618 */ 2619 spin_lock_wr(&p->p_spin); 2620 if ((fdp = p->p_fd) == NULL) { 2621 spin_unlock_wr(&p->p_spin); 2622 return(0); 2623 } 2624 atomic_add_int(&fdp->fd_softrefs, 1); 2625 spin_unlock_wr(&p->p_spin); 2626 2627 /* 2628 * The fdp's own spinlock prevents the contents from being 2629 * modified. 2630 */ 2631 spin_lock_rd(&fdp->fd_spin); 2632 for (n = 0; n < fdp->fd_nfiles; ++n) { 2633 if ((fp = fdp->fd_files[n].fp) == NULL) 2634 continue; 2635 if (info->req->oldptr == NULL) { 2636 ++info->count; 2637 } else { 2638 uid = p->p_ucred ? p->p_ucred->cr_uid : -1; 2639 kcore_make_file(&kf, fp, p->p_pid, uid, n); 2640 spin_unlock_rd(&fdp->fd_spin); 2641 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf)); 2642 spin_lock_rd(&fdp->fd_spin); 2643 if (info->error) 2644 break; 2645 } 2646 } 2647 spin_unlock_rd(&fdp->fd_spin); 2648 atomic_subtract_int(&fdp->fd_softrefs, 1); 2649 if (info->error) 2650 return(-1); 2651 return(0); 2652 } 2653 2654 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2655 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 2656 2657 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2658 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2659 2660 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2661 &maxfiles, 0, "Maximum number of files"); 2662 2663 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, 2664 &maxfilesrootres, 0, "Descriptors reserved for root use"); 2665 2666 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2667 &nfiles, 0, "System-wide number of open files"); 2668 2669 static void 2670 fildesc_drvinit(void *unused) 2671 { 2672 int fd; 2673 2674 for (fd = 0; fd < NUMFDESC; fd++) { 2675 make_dev(&fildesc_ops, fd, 2676 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 2677 } 2678 2679 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 2680 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 2681 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 2682 } 2683 2684 /* 2685 * MPSAFE 2686 */ 2687 struct fileops badfileops = { 2688 .fo_read = badfo_readwrite, 2689 .fo_write = badfo_readwrite, 2690 .fo_ioctl = badfo_ioctl, 2691 .fo_poll = badfo_poll, 2692 .fo_kqfilter = badfo_kqfilter, 2693 .fo_stat = badfo_stat, 2694 .fo_close = badfo_close, 2695 .fo_shutdown = badfo_shutdown 2696 }; 2697 2698 /* 2699 * MPSAFE 2700 */ 2701 int 2702 badfo_readwrite( 2703 struct file *fp, 2704 struct uio *uio, 2705 struct ucred *cred, 2706 int flags 2707 ) { 2708 return (EBADF); 2709 } 2710 2711 /* 2712 * MPSAFE 2713 */ 2714 int 2715 badfo_ioctl(struct file *fp, u_long com, caddr_t data, 2716 struct ucred *cred, struct sysmsg *msgv) 2717 { 2718 return (EBADF); 2719 } 2720 2721 /* 2722 * MPSAFE 2723 */ 2724 int 2725 badfo_poll(struct file *fp, int events, struct ucred *cred) 2726 { 2727 return (0); 2728 } 2729 2730 /* 2731 * MPSAFE 2732 */ 2733 int 2734 badfo_kqfilter(struct file *fp, struct knote *kn) 2735 { 2736 return (0); 2737 } 2738 2739 int 2740 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred) 2741 { 2742 return (EBADF); 2743 } 2744 2745 /* 2746 * MPSAFE 2747 */ 2748 int 2749 badfo_close(struct file *fp) 2750 { 2751 return (EBADF); 2752 } 2753 2754 /* 2755 * MPSAFE 2756 */ 2757 int 2758 badfo_shutdown(struct file *fp, int how) 2759 { 2760 return (EBADF); 2761 } 2762 2763 /* 2764 * MPSAFE 2765 */ 2766 int 2767 nofo_shutdown(struct file *fp, int how) 2768 { 2769 return (EOPNOTSUPP); 2770 } 2771 2772 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, 2773 fildesc_drvinit,NULL) 2774