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