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 1.78 2007/02/18 16:15:23 corecode 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> 99 100 #include <vm/vm.h> 101 #include <vm/vm_extern.h> 102 103 #include <sys/thread2.h> 104 #include <sys/file2.h> 105 #include <sys/spinlock2.h> 106 107 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd); 108 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr); 109 static struct file *funsetfd_locked (struct filedesc *fdp, int fd); 110 static int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp); 111 static void ffree(struct file *fp); 112 113 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 114 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 115 "file desc to leader structures"); 116 MALLOC_DEFINE(M_FILE, "file", "Open file structure"); 117 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 118 119 static d_open_t fdopen; 120 #define NUMFDESC 64 121 122 #define CDEV_MAJOR 22 123 static struct dev_ops fildesc_ops = { 124 { "FD", CDEV_MAJOR, 0 }, 125 .d_open = fdopen, 126 }; 127 128 static int badfo_readwrite (struct file *fp, struct uio *uio, 129 struct ucred *cred, int flags); 130 static int badfo_ioctl (struct file *fp, u_long com, caddr_t data, 131 struct ucred *cred); 132 static int badfo_poll (struct file *fp, int events, struct ucred *cred); 133 static int badfo_kqfilter (struct file *fp, struct knote *kn); 134 static int badfo_stat (struct file *fp, struct stat *sb, struct ucred *cred); 135 static int badfo_close (struct file *fp); 136 static int badfo_shutdown (struct file *fp, int how); 137 138 /* 139 * Descriptor management. 140 */ 141 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead); 142 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin); 143 static int nfiles; /* actual number of open files */ 144 extern int cmask; 145 146 /* 147 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared. 148 * 149 * MPSAFE - must be called with fdp->fd_spin exclusively held 150 */ 151 static __inline 152 void 153 fdfixup_locked(struct filedesc *fdp, int fd) 154 { 155 if (fd < fdp->fd_freefile) { 156 fdp->fd_freefile = fd; 157 } 158 while (fdp->fd_lastfile >= 0 && 159 fdp->fd_files[fdp->fd_lastfile].fp == NULL && 160 fdp->fd_files[fdp->fd_lastfile].reserved == 0 161 ) { 162 --fdp->fd_lastfile; 163 } 164 } 165 166 /* 167 * System calls on descriptors. 168 * 169 * MPSAFE 170 */ 171 int 172 sys_getdtablesize(struct getdtablesize_args *uap) 173 { 174 struct proc *p = curproc; 175 struct plimit *limit = p->p_limit; 176 177 spin_lock_rd(&limit->p_spin); 178 uap->sysmsg_result = 179 min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 180 spin_unlock_rd(&limit->p_spin); 181 return (0); 182 } 183 184 /* 185 * Duplicate a file descriptor to a particular value. 186 * 187 * note: keep in mind that a potential race condition exists when closing 188 * descriptors from a shared descriptor table (via rfork). 189 * 190 * MPSAFE 191 */ 192 int 193 sys_dup2(struct dup2_args *uap) 194 { 195 int error; 196 197 error = kern_dup(DUP_FIXED, uap->from, uap->to, uap->sysmsg_fds); 198 199 return (error); 200 } 201 202 /* 203 * Duplicate a file descriptor. 204 * 205 * MPSAFE 206 */ 207 int 208 sys_dup(struct dup_args *uap) 209 { 210 int error; 211 212 error = kern_dup(DUP_VARIABLE, uap->fd, 0, uap->sysmsg_fds); 213 214 return (error); 215 } 216 217 /* 218 * MPALMOSTSAFE - acquires mplock for fp operations 219 */ 220 int 221 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred) 222 { 223 struct thread *td = curthread; 224 struct proc *p = td->td_proc; 225 struct file *fp; 226 struct vnode *vp; 227 u_int newmin; 228 u_int oflags; 229 int tmp, error, flg = F_POSIX; 230 231 KKASSERT(p); 232 233 /* 234 * Operations on file descriptors that do not require a file pointer. 235 */ 236 switch (cmd) { 237 case F_GETFD: 238 error = fgetfdflags(p->p_fd, fd, &tmp); 239 if (error == 0) 240 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0; 241 return (error); 242 243 case F_SETFD: 244 if (dat->fc_cloexec & FD_CLOEXEC) 245 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE); 246 else 247 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE); 248 return (error); 249 case F_DUPFD: 250 newmin = dat->fc_fd; 251 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd); 252 return (error); 253 default: 254 break; 255 } 256 257 /* 258 * Operations on file pointers 259 */ 260 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 261 return (EBADF); 262 263 get_mplock(); 264 switch (cmd) { 265 case F_GETFL: 266 dat->fc_flags = OFLAGS(fp->f_flag); 267 error = 0; 268 break; 269 270 case F_SETFL: 271 oflags = fp->f_flag & FCNTLFLAGS; 272 fp->f_flag &= ~FCNTLFLAGS; 273 fp->f_flag |= FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 274 error = 0; 275 if ((fp->f_flag ^ oflags) & FASYNC) { 276 tmp = fp->f_flag & FASYNC; 277 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred); 278 } 279 if (error) 280 fp->f_flag = (fp->f_flag & ~FCNTLFLAGS) | oflags; 281 break; 282 283 case F_GETOWN: 284 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, cred); 285 break; 286 287 case F_SETOWN: 288 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, cred); 289 break; 290 291 case F_SETLKW: 292 flg |= F_WAIT; 293 /* Fall into F_SETLK */ 294 295 case F_SETLK: 296 if (fp->f_type != DTYPE_VNODE) { 297 error = EBADF; 298 break; 299 } 300 vp = (struct vnode *)fp->f_data; 301 302 /* 303 * copyin/lockop may block 304 */ 305 if (dat->fc_flock.l_whence == SEEK_CUR) 306 dat->fc_flock.l_start += fp->f_offset; 307 308 switch (dat->fc_flock.l_type) { 309 case F_RDLCK: 310 if ((fp->f_flag & FREAD) == 0) { 311 error = EBADF; 312 break; 313 } 314 p->p_leader->p_flag |= P_ADVLOCK; 315 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 316 &dat->fc_flock, flg); 317 break; 318 case F_WRLCK: 319 if ((fp->f_flag & FWRITE) == 0) { 320 error = EBADF; 321 break; 322 } 323 p->p_leader->p_flag |= P_ADVLOCK; 324 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 325 &dat->fc_flock, flg); 326 break; 327 case F_UNLCK: 328 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 329 &dat->fc_flock, F_POSIX); 330 break; 331 default: 332 error = EINVAL; 333 break; 334 } 335 336 /* 337 * It is possible to race a close() on the descriptor while 338 * we were blocked getting the lock. If this occurs the 339 * close might not have caught the lock. 340 */ 341 if (checkfpclosed(p->p_fd, fd, fp)) { 342 dat->fc_flock.l_whence = SEEK_SET; 343 dat->fc_flock.l_start = 0; 344 dat->fc_flock.l_len = 0; 345 dat->fc_flock.l_type = F_UNLCK; 346 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 347 F_UNLCK, &dat->fc_flock, F_POSIX); 348 } 349 break; 350 351 case F_GETLK: 352 if (fp->f_type != DTYPE_VNODE) { 353 error = EBADF; 354 break; 355 } 356 vp = (struct vnode *)fp->f_data; 357 /* 358 * copyin/lockop may block 359 */ 360 if (dat->fc_flock.l_type != F_RDLCK && 361 dat->fc_flock.l_type != F_WRLCK && 362 dat->fc_flock.l_type != F_UNLCK) { 363 error = EINVAL; 364 break; 365 } 366 if (dat->fc_flock.l_whence == SEEK_CUR) 367 dat->fc_flock.l_start += fp->f_offset; 368 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 369 &dat->fc_flock, F_POSIX); 370 break; 371 default: 372 error = EINVAL; 373 break; 374 } 375 rel_mplock(); 376 377 fdrop(fp); 378 return (error); 379 } 380 381 /* 382 * The file control system call. 383 * 384 * MPSAFE 385 */ 386 int 387 sys_fcntl(struct fcntl_args *uap) 388 { 389 union fcntl_dat dat; 390 int error; 391 392 switch (uap->cmd) { 393 case F_DUPFD: 394 dat.fc_fd = uap->arg; 395 break; 396 case F_SETFD: 397 dat.fc_cloexec = uap->arg; 398 break; 399 case F_SETFL: 400 dat.fc_flags = uap->arg; 401 break; 402 case F_SETOWN: 403 dat.fc_owner = uap->arg; 404 break; 405 case F_SETLKW: 406 case F_SETLK: 407 case F_GETLK: 408 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 409 sizeof(struct flock)); 410 if (error) 411 return (error); 412 break; 413 } 414 415 error = kern_fcntl(uap->fd, uap->cmd, &dat, curproc->p_ucred); 416 417 if (error == 0) { 418 switch (uap->cmd) { 419 case F_DUPFD: 420 uap->sysmsg_result = dat.fc_fd; 421 break; 422 case F_GETFD: 423 uap->sysmsg_result = dat.fc_cloexec; 424 break; 425 case F_GETFL: 426 uap->sysmsg_result = dat.fc_flags; 427 break; 428 case F_GETOWN: 429 uap->sysmsg_result = dat.fc_owner; 430 case F_GETLK: 431 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 432 sizeof(struct flock)); 433 break; 434 } 435 } 436 437 return (error); 438 } 439 440 /* 441 * Common code for dup, dup2, and fcntl(F_DUPFD). 442 * 443 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells 444 * kern_dup() to destructively dup over an existing file descriptor if new 445 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest 446 * unused file descriptor that is greater than or equal to new. 447 * 448 * MPSAFE 449 */ 450 int 451 kern_dup(enum dup_type type, int old, int new, int *res) 452 { 453 struct thread *td = curthread; 454 struct proc *p = td->td_proc; 455 struct filedesc *fdp = p->p_fd; 456 struct file *fp; 457 struct file *delfp; 458 int oldflags; 459 int holdleaders; 460 int error, newfd; 461 462 /* 463 * Verify that we have a valid descriptor to dup from and 464 * possibly to dup to. 465 */ 466 retry: 467 spin_lock_wr(&fdp->fd_spin); 468 if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 469 new >= maxfilesperproc) { 470 spin_unlock_wr(&fdp->fd_spin); 471 return (EINVAL); 472 } 473 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) { 474 spin_unlock_wr(&fdp->fd_spin); 475 return (EBADF); 476 } 477 if (type == DUP_FIXED && old == new) { 478 *res = new; 479 spin_unlock_wr(&fdp->fd_spin); 480 return (0); 481 } 482 fp = fdp->fd_files[old].fp; 483 oldflags = fdp->fd_files[old].fileflags; 484 fhold(fp); /* MPSAFE - can be called with a spinlock held */ 485 486 /* 487 * Allocate a new descriptor if DUP_VARIABLE, or expand the table 488 * if the requested descriptor is beyond the current table size. 489 * 490 * This can block. Retry if the source descriptor no longer matches 491 * or if our expectation in the expansion case races. 492 * 493 * If we are not expanding or allocating a new decriptor, then reset 494 * the target descriptor to a reserved state so we have a uniform 495 * setup for the next code block. 496 */ 497 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) { 498 spin_unlock_wr(&fdp->fd_spin); 499 error = fdalloc(p, new, &newfd); 500 spin_lock_wr(&fdp->fd_spin); 501 if (error) { 502 spin_unlock_wr(&fdp->fd_spin); 503 fdrop(fp); 504 return (error); 505 } 506 /* 507 * Check for ripout 508 */ 509 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) { 510 fsetfd_locked(fdp, NULL, newfd); 511 spin_unlock_wr(&fdp->fd_spin); 512 fdrop(fp); 513 goto retry; 514 } 515 /* 516 * Check for expansion race 517 */ 518 if (type != DUP_VARIABLE && new != newfd) { 519 fsetfd_locked(fdp, NULL, newfd); 520 spin_unlock_wr(&fdp->fd_spin); 521 fdrop(fp); 522 goto retry; 523 } 524 /* 525 * Check for ripout, newfd reused old (this case probably 526 * can't occur). 527 */ 528 if (old == newfd) { 529 fsetfd_locked(fdp, NULL, newfd); 530 spin_unlock_wr(&fdp->fd_spin); 531 fdrop(fp); 532 goto retry; 533 } 534 new = newfd; 535 delfp = NULL; 536 } else { 537 if (fdp->fd_files[new].reserved) { 538 spin_unlock_wr(&fdp->fd_spin); 539 fdrop(fp); 540 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new); 541 tsleep(fdp, 0, "fdres", hz); 542 goto retry; 543 } 544 545 /* 546 * If the target descriptor was never allocated we have 547 * to allocate it. If it was we have to clean out the 548 * old descriptor. delfp inherits the ref from the 549 * descriptor table. 550 */ 551 delfp = fdp->fd_files[new].fp; 552 fdp->fd_files[new].fp = NULL; 553 fdp->fd_files[new].reserved = 1; 554 if (delfp == NULL) { 555 fdreserve_locked(fdp, new, 1); 556 if (new > fdp->fd_lastfile) 557 fdp->fd_lastfile = new; 558 } 559 560 } 561 562 /* 563 * NOTE: still holding an exclusive spinlock 564 */ 565 566 /* 567 * If a descriptor is being overwritten we may hve to tell 568 * fdfree() to sleep to ensure that all relevant process 569 * leaders can be traversed in closef(). 570 */ 571 if (delfp != NULL && p->p_fdtol != NULL) { 572 fdp->fd_holdleaderscount++; 573 holdleaders = 1; 574 } else { 575 holdleaders = 0; 576 } 577 KASSERT(delfp == NULL || type == DUP_FIXED, 578 ("dup() picked an open file")); 579 580 /* 581 * Duplicate the source descriptor, update lastfile. If the new 582 * descriptor was not allocated and we aren't replacing an existing 583 * descriptor we have to mark the descriptor as being in use. 584 * 585 * The fd_files[] array inherits fp's hold reference. 586 */ 587 fsetfd_locked(fdp, fp, new); 588 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE; 589 spin_unlock_wr(&fdp->fd_spin); 590 fdrop(fp); 591 *res = new; 592 593 /* 594 * If we dup'd over a valid file, we now own the reference to it 595 * and must dispose of it using closef() semantics (as if a 596 * close() were performed on it). 597 */ 598 if (delfp) { 599 (void)closef(delfp, td); 600 if (holdleaders) { 601 spin_lock_wr(&fdp->fd_spin); 602 fdp->fd_holdleaderscount--; 603 if (fdp->fd_holdleaderscount == 0 && 604 fdp->fd_holdleaderswakeup != 0) { 605 fdp->fd_holdleaderswakeup = 0; 606 spin_unlock_wr(&fdp->fd_spin); 607 wakeup(&fdp->fd_holdleaderscount); 608 } else { 609 spin_unlock_wr(&fdp->fd_spin); 610 } 611 } 612 } 613 return (0); 614 } 615 616 /* 617 * If sigio is on the list associated with a process or process group, 618 * disable signalling from the device, remove sigio from the list and 619 * free sigio. 620 */ 621 void 622 funsetown(struct sigio *sigio) 623 { 624 if (sigio == NULL) 625 return; 626 crit_enter(); 627 *(sigio->sio_myref) = NULL; 628 crit_exit(); 629 if (sigio->sio_pgid < 0) { 630 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 631 sigio, sio_pgsigio); 632 } else /* if ((*sigiop)->sio_pgid > 0) */ { 633 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 634 sigio, sio_pgsigio); 635 } 636 crfree(sigio->sio_ucred); 637 kfree(sigio, M_SIGIO); 638 } 639 640 /* Free a list of sigio structures. */ 641 void 642 funsetownlst(struct sigiolst *sigiolst) 643 { 644 struct sigio *sigio; 645 646 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) 647 funsetown(sigio); 648 } 649 650 /* 651 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 652 * 653 * After permission checking, add a sigio structure to the sigio list for 654 * the process or process group. 655 */ 656 int 657 fsetown(pid_t pgid, struct sigio **sigiop) 658 { 659 struct proc *proc; 660 struct pgrp *pgrp; 661 struct sigio *sigio; 662 663 if (pgid == 0) { 664 funsetown(*sigiop); 665 return (0); 666 } 667 if (pgid > 0) { 668 proc = pfind(pgid); 669 if (proc == NULL) 670 return (ESRCH); 671 672 /* 673 * Policy - Don't allow a process to FSETOWN a process 674 * in another session. 675 * 676 * Remove this test to allow maximum flexibility or 677 * restrict FSETOWN to the current process or process 678 * group for maximum safety. 679 */ 680 if (proc->p_session != curproc->p_session) 681 return (EPERM); 682 683 pgrp = NULL; 684 } else /* if (pgid < 0) */ { 685 pgrp = pgfind(-pgid); 686 if (pgrp == NULL) 687 return (ESRCH); 688 689 /* 690 * Policy - Don't allow a process to FSETOWN a process 691 * in another session. 692 * 693 * Remove this test to allow maximum flexibility or 694 * restrict FSETOWN to the current process or process 695 * group for maximum safety. 696 */ 697 if (pgrp->pg_session != curproc->p_session) 698 return (EPERM); 699 700 proc = NULL; 701 } 702 funsetown(*sigiop); 703 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 704 if (pgid > 0) { 705 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 706 sigio->sio_proc = proc; 707 } else { 708 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 709 sigio->sio_pgrp = pgrp; 710 } 711 sigio->sio_pgid = pgid; 712 sigio->sio_ucred = crhold(curproc->p_ucred); 713 /* It would be convenient if p_ruid was in ucred. */ 714 sigio->sio_ruid = curproc->p_ucred->cr_ruid; 715 sigio->sio_myref = sigiop; 716 crit_enter(); 717 *sigiop = sigio; 718 crit_exit(); 719 return (0); 720 } 721 722 /* 723 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 724 */ 725 pid_t 726 fgetown(struct sigio *sigio) 727 { 728 return (sigio != NULL ? sigio->sio_pgid : 0); 729 } 730 731 /* 732 * Close many file descriptors. 733 * 734 * MPSAFE 735 */ 736 int 737 sys_closefrom(struct closefrom_args *uap) 738 { 739 return(kern_closefrom(uap->fd)); 740 } 741 742 /* 743 * Close all file descriptors greater then or equal to fd 744 * 745 * MPSAFE 746 */ 747 int 748 kern_closefrom(int fd) 749 { 750 struct thread *td = curthread; 751 struct proc *p = td->td_proc; 752 struct filedesc *fdp; 753 754 KKASSERT(p); 755 fdp = p->p_fd; 756 757 if (fd < 0) 758 return (EINVAL); 759 760 /* 761 * NOTE: This function will skip unassociated descriptors and 762 * reserved descriptors that have not yet been assigned. 763 * fd_lastfile can change as a side effect of kern_close(). 764 */ 765 spin_lock_wr(&fdp->fd_spin); 766 while (fd <= fdp->fd_lastfile) { 767 if (fdp->fd_files[fd].fp != NULL) { 768 spin_unlock_wr(&fdp->fd_spin); 769 /* ok if this races another close */ 770 if (kern_close(fd) == EINTR) 771 return (EINTR); 772 spin_lock_wr(&fdp->fd_spin); 773 } 774 ++fd; 775 } 776 spin_unlock_wr(&fdp->fd_spin); 777 return (0); 778 } 779 780 /* 781 * Close a file descriptor. 782 * 783 * MPSAFE 784 */ 785 int 786 sys_close(struct close_args *uap) 787 { 788 return(kern_close(uap->fd)); 789 } 790 791 /* 792 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls 793 */ 794 int 795 kern_close(int fd) 796 { 797 struct thread *td = curthread; 798 struct proc *p = td->td_proc; 799 struct filedesc *fdp; 800 struct file *fp; 801 int error; 802 int holdleaders; 803 804 KKASSERT(p); 805 fdp = p->p_fd; 806 807 spin_lock_wr(&fdp->fd_spin); 808 if ((fp = funsetfd_locked(fdp, fd)) == NULL) { 809 spin_unlock_wr(&fdp->fd_spin); 810 return (EBADF); 811 } 812 holdleaders = 0; 813 if (p->p_fdtol != NULL) { 814 /* 815 * Ask fdfree() to sleep to ensure that all relevant 816 * process leaders can be traversed in closef(). 817 */ 818 fdp->fd_holdleaderscount++; 819 holdleaders = 1; 820 } 821 822 /* 823 * we now hold the fp reference that used to be owned by the descriptor 824 * array. 825 */ 826 spin_unlock_wr(&fdp->fd_spin); 827 if (fd < fdp->fd_knlistsize) { 828 get_mplock(); 829 if (fd < fdp->fd_knlistsize) 830 knote_fdclose(p, fd); 831 rel_mplock(); 832 } 833 error = closef(fp, td); 834 if (holdleaders) { 835 spin_lock_wr(&fdp->fd_spin); 836 fdp->fd_holdleaderscount--; 837 if (fdp->fd_holdleaderscount == 0 && 838 fdp->fd_holdleaderswakeup != 0) { 839 fdp->fd_holdleaderswakeup = 0; 840 spin_unlock_wr(&fdp->fd_spin); 841 wakeup(&fdp->fd_holdleaderscount); 842 } else { 843 spin_unlock_wr(&fdp->fd_spin); 844 } 845 } 846 return (error); 847 } 848 849 /* 850 * shutdown_args(int fd, int how) 851 */ 852 int 853 kern_shutdown(int fd, int how) 854 { 855 struct thread *td = curthread; 856 struct proc *p = td->td_proc; 857 struct file *fp; 858 int error; 859 860 KKASSERT(p); 861 862 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 863 return (EBADF); 864 error = fo_shutdown(fp, how); 865 fdrop(fp); 866 867 return (error); 868 } 869 870 int 871 sys_shutdown(struct shutdown_args *uap) 872 { 873 int error; 874 875 error = kern_shutdown(uap->s, uap->how); 876 877 return (error); 878 } 879 880 int 881 kern_fstat(int fd, struct stat *ub) 882 { 883 struct thread *td = curthread; 884 struct proc *p = td->td_proc; 885 struct file *fp; 886 int error; 887 888 KKASSERT(p); 889 890 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 891 return (EBADF); 892 error = fo_stat(fp, ub, p->p_ucred); 893 fdrop(fp); 894 895 return (error); 896 } 897 898 /* 899 * Return status information about a file descriptor. 900 */ 901 int 902 sys_fstat(struct fstat_args *uap) 903 { 904 struct stat st; 905 int error; 906 907 error = kern_fstat(uap->fd, &st); 908 909 if (error == 0) 910 error = copyout(&st, uap->sb, sizeof(st)); 911 return (error); 912 } 913 914 /* 915 * Return pathconf information about a file descriptor. 916 */ 917 /* ARGSUSED */ 918 int 919 sys_fpathconf(struct fpathconf_args *uap) 920 { 921 struct thread *td = curthread; 922 struct proc *p = td->td_proc; 923 struct file *fp; 924 struct vnode *vp; 925 int error = 0; 926 927 KKASSERT(p); 928 929 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 930 return (EBADF); 931 932 switch (fp->f_type) { 933 case DTYPE_PIPE: 934 case DTYPE_SOCKET: 935 if (uap->name != _PC_PIPE_BUF) { 936 error = EINVAL; 937 } else { 938 uap->sysmsg_result = PIPE_BUF; 939 error = 0; 940 } 941 break; 942 case DTYPE_FIFO: 943 case DTYPE_VNODE: 944 vp = (struct vnode *)fp->f_data; 945 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds); 946 break; 947 default: 948 error = EOPNOTSUPP; 949 break; 950 } 951 fdrop(fp); 952 return(error); 953 } 954 955 static int fdexpand; 956 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, ""); 957 958 /* 959 * Grow the file table so it can hold through descriptor (want). 960 * 961 * The fdp's spinlock must be held exclusively on entry and may be held 962 * exclusively on return. The spinlock may be cycled by the routine. 963 * 964 * MPSAFE 965 */ 966 static void 967 fdgrow_locked(struct filedesc *fdp, int want) 968 { 969 struct fdnode *newfiles; 970 struct fdnode *oldfiles; 971 int nf, extra; 972 973 nf = fdp->fd_nfiles; 974 do { 975 /* nf has to be of the form 2^n - 1 */ 976 nf = 2 * nf + 1; 977 } while (nf <= want); 978 979 spin_unlock_wr(&fdp->fd_spin); 980 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK); 981 spin_lock_wr(&fdp->fd_spin); 982 983 /* 984 * We could have raced another extend while we were not holding 985 * the spinlock. 986 */ 987 if (fdp->fd_nfiles >= nf) { 988 spin_unlock_wr(&fdp->fd_spin); 989 kfree(newfiles, M_FILEDESC); 990 spin_lock_wr(&fdp->fd_spin); 991 return; 992 } 993 /* 994 * Copy the existing ofile and ofileflags arrays 995 * and zero the new portion of each array. 996 */ 997 extra = nf - fdp->fd_nfiles; 998 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode)); 999 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode)); 1000 1001 oldfiles = fdp->fd_files; 1002 fdp->fd_files = newfiles; 1003 fdp->fd_nfiles = nf; 1004 1005 if (oldfiles != fdp->fd_builtin_files) { 1006 spin_unlock_wr(&fdp->fd_spin); 1007 kfree(oldfiles, M_FILEDESC); 1008 spin_lock_wr(&fdp->fd_spin); 1009 } 1010 fdexpand++; 1011 } 1012 1013 /* 1014 * Number of nodes in right subtree, including the root. 1015 */ 1016 static __inline int 1017 right_subtree_size(int n) 1018 { 1019 return (n ^ (n | (n + 1))); 1020 } 1021 1022 /* 1023 * Bigger ancestor. 1024 */ 1025 static __inline int 1026 right_ancestor(int n) 1027 { 1028 return (n | (n + 1)); 1029 } 1030 1031 /* 1032 * Smaller ancestor. 1033 */ 1034 static __inline int 1035 left_ancestor(int n) 1036 { 1037 return ((n & (n + 1)) - 1); 1038 } 1039 1040 /* 1041 * Traverse the in-place binary tree buttom-up adjusting the allocation 1042 * count so scans can determine where free descriptors are located. 1043 * 1044 * MPSAFE - caller must be holding an exclusive spinlock on fdp 1045 */ 1046 static 1047 void 1048 fdreserve_locked(struct filedesc *fdp, int fd, int incr) 1049 { 1050 while (fd >= 0) { 1051 fdp->fd_files[fd].allocated += incr; 1052 KKASSERT(fdp->fd_files[fd].allocated >= 0); 1053 fd = left_ancestor(fd); 1054 } 1055 } 1056 1057 /* 1058 * Reserve a file descriptor for the process. If no error occurs, the 1059 * caller MUST at some point call fsetfd() or assign a file pointer 1060 * or dispose of the reservation. 1061 * 1062 * MPSAFE 1063 */ 1064 int 1065 fdalloc(struct proc *p, int want, int *result) 1066 { 1067 struct filedesc *fdp = p->p_fd; 1068 int fd, rsize, rsum, node, lim; 1069 1070 spin_lock_rd(&p->p_limit->p_spin); 1071 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 1072 spin_unlock_rd(&p->p_limit->p_spin); 1073 if (want >= lim) 1074 return (EMFILE); 1075 spin_lock_wr(&fdp->fd_spin); 1076 if (want >= fdp->fd_nfiles) 1077 fdgrow_locked(fdp, want); 1078 1079 /* 1080 * Search for a free descriptor starting at the higher 1081 * of want or fd_freefile. If that fails, consider 1082 * expanding the ofile array. 1083 * 1084 * NOTE! the 'allocated' field is a cumulative recursive allocation 1085 * count. If we happen to see a value of 0 then we can shortcut 1086 * our search. Otherwise we run through through the tree going 1087 * down branches we know have free descriptor(s) until we hit a 1088 * leaf node. The leaf node will be free but will not necessarily 1089 * have an allocated field of 0. 1090 */ 1091 retry: 1092 /* move up the tree looking for a subtree with a free node */ 1093 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim); 1094 fd = right_ancestor(fd)) { 1095 if (fdp->fd_files[fd].allocated == 0) 1096 goto found; 1097 1098 rsize = right_subtree_size(fd); 1099 if (fdp->fd_files[fd].allocated == rsize) 1100 continue; /* right subtree full */ 1101 1102 /* 1103 * Free fd is in the right subtree of the tree rooted at fd. 1104 * Call that subtree R. Look for the smallest (leftmost) 1105 * subtree of R with an unallocated fd: continue moving 1106 * down the left branch until encountering a full left 1107 * subtree, then move to the right. 1108 */ 1109 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) { 1110 node = fd + rsize; 1111 rsum += fdp->fd_files[node].allocated; 1112 if (fdp->fd_files[fd].allocated == rsum + rsize) { 1113 fd = node; /* move to the right */ 1114 if (fdp->fd_files[node].allocated == 0) 1115 goto found; 1116 rsum = 0; 1117 } 1118 } 1119 goto found; 1120 } 1121 1122 /* 1123 * No space in current array. Expand? 1124 */ 1125 if (fdp->fd_nfiles >= lim) { 1126 spin_unlock_wr(&fdp->fd_spin); 1127 return (EMFILE); 1128 } 1129 fdgrow_locked(fdp, want); 1130 goto retry; 1131 1132 found: 1133 KKASSERT(fd < fdp->fd_nfiles); 1134 if (fd > fdp->fd_lastfile) 1135 fdp->fd_lastfile = fd; 1136 if (want <= fdp->fd_freefile) 1137 fdp->fd_freefile = fd; 1138 *result = fd; 1139 KKASSERT(fdp->fd_files[fd].fp == NULL); 1140 KKASSERT(fdp->fd_files[fd].reserved == 0); 1141 fdp->fd_files[fd].fileflags = 0; 1142 fdp->fd_files[fd].reserved = 1; 1143 fdreserve_locked(fdp, fd, 1); 1144 spin_unlock_wr(&fdp->fd_spin); 1145 return (0); 1146 } 1147 1148 /* 1149 * Check to see whether n user file descriptors 1150 * are available to the process p. 1151 * 1152 * MPSAFE 1153 */ 1154 int 1155 fdavail(struct proc *p, int n) 1156 { 1157 struct filedesc *fdp = p->p_fd; 1158 struct fdnode *fdnode; 1159 int i, lim, last; 1160 1161 spin_lock_rd(&p->p_limit->p_spin); 1162 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 1163 spin_unlock_rd(&p->p_limit->p_spin); 1164 1165 spin_lock_rd(&fdp->fd_spin); 1166 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) { 1167 spin_unlock_rd(&fdp->fd_spin); 1168 return (1); 1169 } 1170 last = min(fdp->fd_nfiles, lim); 1171 fdnode = &fdp->fd_files[fdp->fd_freefile]; 1172 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) { 1173 if (fdnode->fp == NULL && --n <= 0) { 1174 spin_unlock_rd(&fdp->fd_spin); 1175 return (1); 1176 } 1177 } 1178 spin_unlock_rd(&fdp->fd_spin); 1179 return (0); 1180 } 1181 1182 /* 1183 * falloc: 1184 * Create a new open file structure and reserve a file decriptor 1185 * for the process that refers to it. 1186 * 1187 * Root creds are checked using p, or assumed if p is NULL. If 1188 * resultfd is non-NULL then p must also be non-NULL. No file 1189 * descriptor is reserved if resultfd is NULL. 1190 * 1191 * A file pointer with a refcount of 1 is returned. Note that the 1192 * file pointer is NOT associated with the descriptor. If falloc 1193 * returns success, fsetfd() MUST be called to either associate the 1194 * file pointer or clear the reservation. 1195 * 1196 * MPSAFE 1197 */ 1198 int 1199 falloc(struct proc *p, struct file **resultfp, int *resultfd) 1200 { 1201 static struct timeval lastfail; 1202 static int curfail; 1203 struct file *fp; 1204 int error; 1205 1206 fp = NULL; 1207 1208 /* 1209 * Handle filetable full issues and root overfill. 1210 */ 1211 if (nfiles >= maxfiles - maxfilesrootres && 1212 ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) { 1213 if (ppsratecheck(&lastfail, &curfail, 1)) { 1214 kprintf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n", 1215 (p ? p->p_ucred->cr_ruid : -1)); 1216 } 1217 error = ENFILE; 1218 goto done; 1219 } 1220 1221 /* 1222 * Allocate a new file descriptor. 1223 */ 1224 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO); 1225 spin_init(&fp->f_spin); 1226 fp->f_count = 1; 1227 fp->f_ops = &badfileops; 1228 fp->f_seqcount = 1; 1229 if (p) 1230 fp->f_cred = crhold(p->p_ucred); 1231 else 1232 fp->f_cred = crhold(proc0.p_ucred); 1233 spin_lock_wr(&filehead_spin); 1234 nfiles++; 1235 LIST_INSERT_HEAD(&filehead, fp, f_list); 1236 spin_unlock_wr(&filehead_spin); 1237 if (resultfd) { 1238 if ((error = fdalloc(p, 0, resultfd)) != 0) { 1239 fdrop(fp); 1240 fp = NULL; 1241 } 1242 } else { 1243 error = 0; 1244 } 1245 done: 1246 *resultfp = fp; 1247 return (error); 1248 } 1249 1250 /* 1251 * MPSAFE 1252 */ 1253 static 1254 int 1255 checkfpclosed(struct filedesc *fdp, int fd, struct file *fp) 1256 { 1257 int error; 1258 1259 spin_lock_rd(&fdp->fd_spin); 1260 if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp) 1261 error = EBADF; 1262 else 1263 error = 0; 1264 spin_unlock_rd(&fdp->fd_spin); 1265 return (error); 1266 } 1267 1268 /* 1269 * Associate a file pointer with a previously reserved file descriptor. 1270 * This function always succeeds. 1271 * 1272 * If fp is NULL, the file descriptor is returned to the pool. 1273 */ 1274 1275 /* 1276 * MPSAFE (exclusive spinlock must be held on call) 1277 */ 1278 static void 1279 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd) 1280 { 1281 KKASSERT((unsigned)fd < fdp->fd_nfiles); 1282 KKASSERT(fdp->fd_files[fd].reserved != 0); 1283 if (fp) { 1284 fhold(fp); 1285 fdp->fd_files[fd].fp = fp; 1286 fdp->fd_files[fd].reserved = 0; 1287 if (fp->f_type == DTYPE_KQUEUE) { 1288 if (fdp->fd_knlistsize < 0) 1289 fdp->fd_knlistsize = 0; 1290 } 1291 } else { 1292 fdp->fd_files[fd].reserved = 0; 1293 fdreserve_locked(fdp, fd, -1); 1294 fdfixup_locked(fdp, fd); 1295 } 1296 } 1297 1298 /* 1299 * MPSAFE 1300 */ 1301 void 1302 fsetfd(struct proc *p, struct file *fp, int fd) 1303 { 1304 struct filedesc *fdp = p->p_fd; 1305 1306 spin_lock_wr(&fdp->fd_spin); 1307 fsetfd_locked(fdp, fp, fd); 1308 spin_unlock_wr(&fdp->fd_spin); 1309 } 1310 1311 /* 1312 * MPSAFE (exclusive spinlock must be held on call) 1313 */ 1314 static 1315 struct file * 1316 funsetfd_locked(struct filedesc *fdp, int fd) 1317 { 1318 struct file *fp; 1319 1320 if ((unsigned)fd >= fdp->fd_nfiles) 1321 return (NULL); 1322 if ((fp = fdp->fd_files[fd].fp) == NULL) 1323 return (NULL); 1324 fdp->fd_files[fd].fp = NULL; 1325 fdp->fd_files[fd].fileflags = 0; 1326 1327 fdreserve_locked(fdp, fd, -1); 1328 fdfixup_locked(fdp, fd); 1329 return(fp); 1330 } 1331 1332 /* 1333 * MPSAFE 1334 */ 1335 int 1336 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp) 1337 { 1338 int error; 1339 1340 spin_lock_rd(&fdp->fd_spin); 1341 if (((u_int)fd) >= fdp->fd_nfiles) { 1342 error = EBADF; 1343 } else if (fdp->fd_files[fd].fp == NULL) { 1344 error = EBADF; 1345 } else { 1346 *flagsp = fdp->fd_files[fd].fileflags; 1347 error = 0; 1348 } 1349 spin_unlock_rd(&fdp->fd_spin); 1350 return (error); 1351 } 1352 1353 /* 1354 * MPSAFE 1355 */ 1356 int 1357 fsetfdflags(struct filedesc *fdp, int fd, int add_flags) 1358 { 1359 int error; 1360 1361 spin_lock_wr(&fdp->fd_spin); 1362 if (((u_int)fd) >= fdp->fd_nfiles) { 1363 error = EBADF; 1364 } else if (fdp->fd_files[fd].fp == NULL) { 1365 error = EBADF; 1366 } else { 1367 fdp->fd_files[fd].fileflags |= add_flags; 1368 error = 0; 1369 } 1370 spin_unlock_wr(&fdp->fd_spin); 1371 return (error); 1372 } 1373 1374 /* 1375 * MPSAFE 1376 */ 1377 int 1378 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags) 1379 { 1380 int error; 1381 1382 spin_lock_wr(&fdp->fd_spin); 1383 if (((u_int)fd) >= fdp->fd_nfiles) { 1384 error = EBADF; 1385 } else if (fdp->fd_files[fd].fp == NULL) { 1386 error = EBADF; 1387 } else { 1388 fdp->fd_files[fd].fileflags &= ~rem_flags; 1389 error = 0; 1390 } 1391 spin_unlock_wr(&fdp->fd_spin); 1392 return (error); 1393 } 1394 1395 void 1396 fsetcred(struct file *fp, struct ucred *cr) 1397 { 1398 crhold(cr); 1399 crfree(fp->f_cred); 1400 fp->f_cred = cr; 1401 } 1402 1403 /* 1404 * Free a file descriptor. 1405 */ 1406 static 1407 void 1408 ffree(struct file *fp) 1409 { 1410 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 1411 spin_lock_wr(&filehead_spin); 1412 LIST_REMOVE(fp, f_list); 1413 nfiles--; 1414 spin_unlock_wr(&filehead_spin); 1415 crfree(fp->f_cred); 1416 if (fp->f_nchandle.ncp) 1417 cache_drop(&fp->f_nchandle); 1418 kfree(fp, M_FILE); 1419 } 1420 1421 /* 1422 * called from init_main, initialize filedesc0 for proc0. 1423 */ 1424 void 1425 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask) 1426 { 1427 p0->p_fd = fdp0; 1428 p0->p_fdtol = NULL; 1429 fdp0->fd_refcnt = 1; 1430 fdp0->fd_cmask = cmask; 1431 fdp0->fd_files = fdp0->fd_builtin_files; 1432 fdp0->fd_nfiles = NDFILE; 1433 fdp0->fd_lastfile = -1; 1434 spin_init(&fdp0->fd_spin); 1435 } 1436 1437 /* 1438 * Build a new filedesc structure. 1439 * 1440 * NOT MPSAFE (vref) 1441 */ 1442 struct filedesc * 1443 fdinit(struct proc *p) 1444 { 1445 struct filedesc *newfdp; 1446 struct filedesc *fdp = p->p_fd; 1447 1448 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO); 1449 spin_lock_rd(&fdp->fd_spin); 1450 if (fdp->fd_cdir) { 1451 newfdp->fd_cdir = fdp->fd_cdir; 1452 vref(newfdp->fd_cdir); 1453 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1454 } 1455 1456 /* 1457 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of 1458 * proc0, but should unconditionally exist in other processes. 1459 */ 1460 if (fdp->fd_rdir) { 1461 newfdp->fd_rdir = fdp->fd_rdir; 1462 vref(newfdp->fd_rdir); 1463 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1464 } 1465 if (fdp->fd_jdir) { 1466 newfdp->fd_jdir = fdp->fd_jdir; 1467 vref(newfdp->fd_jdir); 1468 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1469 } 1470 spin_unlock_rd(&fdp->fd_spin); 1471 1472 /* Create the file descriptor table. */ 1473 newfdp->fd_refcnt = 1; 1474 newfdp->fd_cmask = cmask; 1475 newfdp->fd_files = newfdp->fd_builtin_files; 1476 newfdp->fd_nfiles = NDFILE; 1477 newfdp->fd_knlistsize = -1; 1478 newfdp->fd_lastfile = -1; 1479 spin_init(&newfdp->fd_spin); 1480 1481 return (newfdp); 1482 } 1483 1484 /* 1485 * Share a filedesc structure. 1486 * 1487 * MPSAFE 1488 */ 1489 struct filedesc * 1490 fdshare(struct proc *p) 1491 { 1492 struct filedesc *fdp; 1493 1494 fdp = p->p_fd; 1495 spin_lock_wr(&fdp->fd_spin); 1496 fdp->fd_refcnt++; 1497 spin_unlock_wr(&fdp->fd_spin); 1498 return (fdp); 1499 } 1500 1501 /* 1502 * Copy a filedesc structure. 1503 * 1504 * MPSAFE 1505 */ 1506 struct filedesc * 1507 fdcopy(struct proc *p) 1508 { 1509 struct filedesc *fdp = p->p_fd; 1510 struct filedesc *newfdp; 1511 struct fdnode *fdnode; 1512 int i; 1513 int ni; 1514 1515 /* 1516 * Certain daemons might not have file descriptors. 1517 */ 1518 if (fdp == NULL) 1519 return (NULL); 1520 1521 /* 1522 * Allocate the new filedesc and fd_files[] array. This can race 1523 * with operations by other threads on the fdp so we have to be 1524 * careful. 1525 */ 1526 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO); 1527 again: 1528 spin_lock_rd(&fdp->fd_spin); 1529 if (fdp->fd_lastfile < NDFILE) { 1530 newfdp->fd_files = newfdp->fd_builtin_files; 1531 i = NDFILE; 1532 } else { 1533 /* 1534 * We have to allocate (N^2-1) entries for our in-place 1535 * binary tree. Allow the table to shrink. 1536 */ 1537 i = fdp->fd_nfiles; 1538 ni = (i - 1) / 2; 1539 while (ni > fdp->fd_lastfile && ni > NDFILE) { 1540 i = ni; 1541 ni = (i - 1) / 2; 1542 } 1543 spin_unlock_rd(&fdp->fd_spin); 1544 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode), 1545 M_FILEDESC, M_WAITOK | M_ZERO); 1546 1547 /* 1548 * Check for race, retry 1549 */ 1550 spin_lock_rd(&fdp->fd_spin); 1551 if (i <= fdp->fd_lastfile) { 1552 spin_unlock_rd(&fdp->fd_spin); 1553 kfree(newfdp->fd_files, M_FILEDESC); 1554 goto again; 1555 } 1556 } 1557 1558 /* 1559 * Dup the remaining fields. vref() and cache_hold() can be 1560 * safely called while holding the read spinlock on fdp. 1561 * 1562 * The read spinlock on fdp is still being held. 1563 * 1564 * NOTE: vref and cache_hold calls for the case where the vnode 1565 * or cache entry already has at least one ref may be called 1566 * while holding spin locks. 1567 */ 1568 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) { 1569 vref(newfdp->fd_cdir); 1570 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1571 } 1572 /* 1573 * We must check for fd_rdir here, at least for now because 1574 * the init process is created before we have access to the 1575 * rootvode to take a reference to it. 1576 */ 1577 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) { 1578 vref(newfdp->fd_rdir); 1579 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1580 } 1581 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) { 1582 vref(newfdp->fd_jdir); 1583 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1584 } 1585 newfdp->fd_refcnt = 1; 1586 newfdp->fd_nfiles = i; 1587 newfdp->fd_lastfile = fdp->fd_lastfile; 1588 newfdp->fd_freefile = fdp->fd_freefile; 1589 newfdp->fd_cmask = fdp->fd_cmask; 1590 newfdp->fd_knlist = NULL; 1591 newfdp->fd_knlistsize = -1; 1592 newfdp->fd_knhash = NULL; 1593 newfdp->fd_knhashmask = 0; 1594 spin_init(&newfdp->fd_spin); 1595 1596 /* 1597 * Copy the descriptor table through (i). This also copies the 1598 * allocation state. Then go through and ref the file pointers 1599 * and clean up any KQ descriptors. 1600 * 1601 * kq descriptors cannot be copied. Since we haven't ref'd the 1602 * copied files yet we can ignore the return value from funsetfd(). 1603 * 1604 * The read spinlock on fdp is still being held. 1605 */ 1606 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode)); 1607 for (i = 0 ; i < newfdp->fd_nfiles; ++i) { 1608 fdnode = &newfdp->fd_files[i]; 1609 if (fdnode->reserved) { 1610 fdreserve_locked(newfdp, i, -1); 1611 fdnode->reserved = 0; 1612 fdfixup_locked(newfdp, i); 1613 } else if (fdnode->fp) { 1614 if (fdnode->fp->f_type == DTYPE_KQUEUE) { 1615 (void)funsetfd_locked(newfdp, i); 1616 } else { 1617 fhold(fdnode->fp); 1618 } 1619 } 1620 } 1621 spin_unlock_rd(&fdp->fd_spin); 1622 return (newfdp); 1623 } 1624 1625 /* 1626 * Release a filedesc structure. 1627 * 1628 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE) 1629 */ 1630 void 1631 fdfree(struct proc *p) 1632 { 1633 /* Take any thread of p */ 1634 struct thread *td = FIRST_LWP_IN_PROC(p)->lwp_thread; 1635 struct filedesc *fdp = p->p_fd; 1636 struct fdnode *fdnode; 1637 int i; 1638 struct filedesc_to_leader *fdtol; 1639 struct file *fp; 1640 struct vnode *vp; 1641 struct flock lf; 1642 1643 /* Certain daemons might not have file descriptors. */ 1644 if (fdp == NULL) 1645 return; 1646 1647 /* 1648 * Severe messing around to follow 1649 */ 1650 spin_lock_wr(&fdp->fd_spin); 1651 1652 /* Check for special need to clear POSIX style locks */ 1653 fdtol = p->p_fdtol; 1654 if (fdtol != NULL) { 1655 KASSERT(fdtol->fdl_refcount > 0, 1656 ("filedesc_to_refcount botch: fdl_refcount=%d", 1657 fdtol->fdl_refcount)); 1658 if (fdtol->fdl_refcount == 1 && 1659 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1660 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1661 fdnode = &fdp->fd_files[i]; 1662 if (fdnode->fp == NULL || 1663 fdnode->fp->f_type != DTYPE_VNODE) { 1664 continue; 1665 } 1666 fp = fdnode->fp; 1667 fhold(fp); 1668 spin_unlock_wr(&fdp->fd_spin); 1669 1670 lf.l_whence = SEEK_SET; 1671 lf.l_start = 0; 1672 lf.l_len = 0; 1673 lf.l_type = F_UNLCK; 1674 vp = (struct vnode *)fp->f_data; 1675 (void) VOP_ADVLOCK(vp, 1676 (caddr_t)p->p_leader, 1677 F_UNLCK, 1678 &lf, 1679 F_POSIX); 1680 fdrop(fp); 1681 spin_lock_wr(&fdp->fd_spin); 1682 } 1683 } 1684 retry: 1685 if (fdtol->fdl_refcount == 1) { 1686 if (fdp->fd_holdleaderscount > 0 && 1687 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1688 /* 1689 * close() or do_dup() has cleared a reference 1690 * in a shared file descriptor table. 1691 */ 1692 fdp->fd_holdleaderswakeup = 1; 1693 msleep(&fdp->fd_holdleaderscount, 1694 &fdp->fd_spin, 0, "fdlhold", 0); 1695 goto retry; 1696 } 1697 if (fdtol->fdl_holdcount > 0) { 1698 /* 1699 * Ensure that fdtol->fdl_leader 1700 * remains valid in closef(). 1701 */ 1702 fdtol->fdl_wakeup = 1; 1703 msleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0); 1704 goto retry; 1705 } 1706 } 1707 fdtol->fdl_refcount--; 1708 if (fdtol->fdl_refcount == 0 && 1709 fdtol->fdl_holdcount == 0) { 1710 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1711 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1712 } else { 1713 fdtol = NULL; 1714 } 1715 p->p_fdtol = NULL; 1716 if (fdtol != NULL) { 1717 spin_unlock_wr(&fdp->fd_spin); 1718 kfree(fdtol, M_FILEDESC_TO_LEADER); 1719 spin_lock_wr(&fdp->fd_spin); 1720 } 1721 } 1722 if (--fdp->fd_refcnt > 0) { 1723 spin_unlock_wr(&fdp->fd_spin); 1724 return; 1725 } 1726 spin_unlock_wr(&fdp->fd_spin); 1727 1728 /* 1729 * we are the last reference to the structure, we can 1730 * safely assume it will not change out from under us. 1731 */ 1732 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1733 if (fdp->fd_files[i].fp) 1734 closef(fdp->fd_files[i].fp, td); 1735 } 1736 if (fdp->fd_files != fdp->fd_builtin_files) 1737 kfree(fdp->fd_files, M_FILEDESC); 1738 if (fdp->fd_cdir) { 1739 cache_drop(&fdp->fd_ncdir); 1740 vrele(fdp->fd_cdir); 1741 } 1742 if (fdp->fd_rdir) { 1743 cache_drop(&fdp->fd_nrdir); 1744 vrele(fdp->fd_rdir); 1745 } 1746 if (fdp->fd_jdir) { 1747 cache_drop(&fdp->fd_njdir); 1748 vrele(fdp->fd_jdir); 1749 } 1750 if (fdp->fd_knlist) 1751 kfree(fdp->fd_knlist, M_KQUEUE); 1752 if (fdp->fd_knhash) 1753 kfree(fdp->fd_knhash, M_KQUEUE); 1754 kfree(fdp, M_FILEDESC); 1755 } 1756 1757 /* 1758 * Retrieve and reference the file pointer associated with a descriptor. 1759 * 1760 * MPSAFE 1761 */ 1762 struct file * 1763 holdfp(struct filedesc *fdp, int fd, int flag) 1764 { 1765 struct file* fp; 1766 1767 spin_lock_rd(&fdp->fd_spin); 1768 if (((u_int)fd) >= fdp->fd_nfiles) { 1769 fp = NULL; 1770 goto done; 1771 } 1772 if ((fp = fdp->fd_files[fd].fp) == NULL) 1773 goto done; 1774 if ((fp->f_flag & flag) == 0 && flag != -1) { 1775 fp = NULL; 1776 goto done; 1777 } 1778 fhold(fp); 1779 done: 1780 spin_unlock_rd(&fdp->fd_spin); 1781 return (fp); 1782 } 1783 1784 /* 1785 * holdsock() - load the struct file pointer associated 1786 * with a socket into *fpp. If an error occurs, non-zero 1787 * will be returned and *fpp will be set to NULL. 1788 * 1789 * MPSAFE 1790 */ 1791 int 1792 holdsock(struct filedesc *fdp, int fd, struct file **fpp) 1793 { 1794 struct file *fp; 1795 int error; 1796 1797 spin_lock_rd(&fdp->fd_spin); 1798 if ((unsigned)fd >= fdp->fd_nfiles) { 1799 error = EBADF; 1800 fp = NULL; 1801 goto done; 1802 } 1803 if ((fp = fdp->fd_files[fd].fp) == NULL) { 1804 error = EBADF; 1805 goto done; 1806 } 1807 if (fp->f_type != DTYPE_SOCKET) { 1808 error = ENOTSOCK; 1809 goto done; 1810 } 1811 fhold(fp); 1812 error = 0; 1813 done: 1814 spin_unlock_rd(&fdp->fd_spin); 1815 *fpp = fp; 1816 return (error); 1817 } 1818 1819 /* 1820 * Convert a user file descriptor to a held file pointer. 1821 * 1822 * MPSAFE 1823 */ 1824 int 1825 holdvnode(struct filedesc *fdp, int fd, struct file **fpp) 1826 { 1827 struct file *fp; 1828 int error; 1829 1830 spin_lock_rd(&fdp->fd_spin); 1831 if ((unsigned)fd >= fdp->fd_nfiles) { 1832 error = EBADF; 1833 fp = NULL; 1834 goto done; 1835 } 1836 if ((fp = fdp->fd_files[fd].fp) == NULL) { 1837 error = EBADF; 1838 goto done; 1839 } 1840 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 1841 error = EINVAL; 1842 goto done; 1843 } 1844 fhold(fp); 1845 error = 0; 1846 done: 1847 spin_unlock_rd(&fdp->fd_spin); 1848 *fpp = fp; 1849 return (error); 1850 } 1851 1852 /* 1853 * For setugid programs, we don't want to people to use that setugidness 1854 * to generate error messages which write to a file which otherwise would 1855 * otherwise be off-limits to the process. 1856 * 1857 * This is a gross hack to plug the hole. A better solution would involve 1858 * a special vop or other form of generalized access control mechanism. We 1859 * go ahead and just reject all procfs file systems accesses as dangerous. 1860 * 1861 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1862 * sufficient. We also don't for check setugidness since we know we are. 1863 */ 1864 static int 1865 is_unsafe(struct file *fp) 1866 { 1867 if (fp->f_type == DTYPE_VNODE && 1868 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 1869 return (1); 1870 return (0); 1871 } 1872 1873 /* 1874 * Make this setguid thing safe, if at all possible. 1875 * 1876 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 1877 */ 1878 void 1879 setugidsafety(struct proc *p) 1880 { 1881 /* Take any thread of p */ 1882 struct thread *td = FIRST_LWP_IN_PROC(p)->lwp_thread; 1883 struct filedesc *fdp = p->p_fd; 1884 int i; 1885 1886 /* Certain daemons might not have file descriptors. */ 1887 if (fdp == NULL) 1888 return; 1889 1890 /* 1891 * note: fdp->fd_files may be reallocated out from under us while 1892 * we are blocked in a close. Be careful! 1893 */ 1894 for (i = 0; i <= fdp->fd_lastfile; i++) { 1895 if (i > 2) 1896 break; 1897 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) { 1898 struct file *fp; 1899 1900 if (i < fdp->fd_knlistsize) 1901 knote_fdclose(p, i); 1902 /* 1903 * NULL-out descriptor prior to close to avoid 1904 * a race while close blocks. 1905 */ 1906 if ((fp = funsetfd_locked(fdp, i)) != NULL) 1907 closef(fp, td); 1908 } 1909 } 1910 } 1911 1912 /* 1913 * Close any files on exec? 1914 * 1915 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 1916 */ 1917 void 1918 fdcloseexec(struct proc *p) 1919 { 1920 /* Take any thread of p */ 1921 struct thread *td = FIRST_LWP_IN_PROC(p)->lwp_thread; 1922 struct filedesc *fdp = p->p_fd; 1923 int i; 1924 1925 /* Certain daemons might not have file descriptors. */ 1926 if (fdp == NULL) 1927 return; 1928 1929 /* 1930 * We cannot cache fd_files since operations may block and rip 1931 * them out from under us. 1932 */ 1933 for (i = 0; i <= fdp->fd_lastfile; i++) { 1934 if (fdp->fd_files[i].fp != NULL && 1935 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) { 1936 struct file *fp; 1937 1938 if (i < fdp->fd_knlistsize) 1939 knote_fdclose(p, i); 1940 /* 1941 * NULL-out descriptor prior to close to avoid 1942 * a race while close blocks. 1943 */ 1944 if ((fp = funsetfd_locked(fdp, i)) != NULL) 1945 closef(fp, td); 1946 } 1947 } 1948 } 1949 1950 /* 1951 * It is unsafe for set[ug]id processes to be started with file 1952 * descriptors 0..2 closed, as these descriptors are given implicit 1953 * significance in the Standard C library. fdcheckstd() will create a 1954 * descriptor referencing /dev/null for each of stdin, stdout, and 1955 * stderr that is not already open. 1956 * 1957 * NOT MPSAFE - calls falloc, vn_open, etc 1958 */ 1959 int 1960 fdcheckstd(struct proc *p) 1961 { 1962 struct nlookupdata nd; 1963 struct filedesc *fdp; 1964 struct file *fp; 1965 register_t retval; 1966 int i, error, flags, devnull; 1967 1968 fdp = p->p_fd; 1969 if (fdp == NULL) 1970 return (0); 1971 devnull = -1; 1972 error = 0; 1973 for (i = 0; i < 3; i++) { 1974 if (fdp->fd_files[i].fp != NULL) 1975 continue; 1976 if (devnull < 0) { 1977 if ((error = falloc(p, &fp, &devnull)) != 0) 1978 break; 1979 1980 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, 1981 NLC_FOLLOW|NLC_LOCKVP); 1982 flags = FREAD | FWRITE; 1983 if (error == 0) 1984 error = vn_open(&nd, fp, flags, 0); 1985 if (error == 0) 1986 fsetfd(p, fp, devnull); 1987 else 1988 fsetfd(p, NULL, devnull); 1989 fdrop(fp); 1990 nlookup_done(&nd); 1991 if (error) 1992 break; 1993 KKASSERT(i == devnull); 1994 } else { 1995 error = kern_dup(DUP_FIXED, devnull, i, &retval); 1996 if (error != 0) 1997 break; 1998 } 1999 } 2000 return (error); 2001 } 2002 2003 /* 2004 * Internal form of close. 2005 * Decrement reference count on file structure. 2006 * Note: td and/or p may be NULL when closing a file 2007 * that was being passed in a message. 2008 * 2009 * MPALMOSTSAFE - acquires mplock for VOP operations 2010 */ 2011 int 2012 closef(struct file *fp, struct thread *td) 2013 { 2014 struct vnode *vp; 2015 struct flock lf; 2016 struct filedesc_to_leader *fdtol; 2017 struct proc *p; 2018 2019 if (fp == NULL) 2020 return (0); 2021 if (td == NULL) { 2022 td = curthread; 2023 p = NULL; /* allow no proc association */ 2024 } else { 2025 p = td->td_proc; /* can also be NULL */ 2026 } 2027 /* 2028 * POSIX record locking dictates that any close releases ALL 2029 * locks owned by this process. This is handled by setting 2030 * a flag in the unlock to free ONLY locks obeying POSIX 2031 * semantics, and not to free BSD-style file locks. 2032 * If the descriptor was in a message, POSIX-style locks 2033 * aren't passed with the descriptor. 2034 */ 2035 if (p != NULL && fp->f_type == DTYPE_VNODE && 2036 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2037 ) { 2038 get_mplock(); 2039 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) { 2040 lf.l_whence = SEEK_SET; 2041 lf.l_start = 0; 2042 lf.l_len = 0; 2043 lf.l_type = F_UNLCK; 2044 vp = (struct vnode *)fp->f_data; 2045 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 2046 &lf, F_POSIX); 2047 } 2048 fdtol = p->p_fdtol; 2049 if (fdtol != NULL) { 2050 /* 2051 * Handle special case where file descriptor table 2052 * is shared between multiple process leaders. 2053 */ 2054 for (fdtol = fdtol->fdl_next; 2055 fdtol != p->p_fdtol; 2056 fdtol = fdtol->fdl_next) { 2057 if ((fdtol->fdl_leader->p_flag & 2058 P_ADVLOCK) == 0) 2059 continue; 2060 fdtol->fdl_holdcount++; 2061 lf.l_whence = SEEK_SET; 2062 lf.l_start = 0; 2063 lf.l_len = 0; 2064 lf.l_type = F_UNLCK; 2065 vp = (struct vnode *)fp->f_data; 2066 (void) VOP_ADVLOCK(vp, 2067 (caddr_t)fdtol->fdl_leader, 2068 F_UNLCK, &lf, F_POSIX); 2069 fdtol->fdl_holdcount--; 2070 if (fdtol->fdl_holdcount == 0 && 2071 fdtol->fdl_wakeup != 0) { 2072 fdtol->fdl_wakeup = 0; 2073 wakeup(fdtol); 2074 } 2075 } 2076 } 2077 rel_mplock(); 2078 } 2079 return (fdrop(fp)); 2080 } 2081 2082 /* 2083 * MPSAFE 2084 * 2085 * fhold() can only be called if f_count is already at least 1 (i.e. the 2086 * caller of fhold() already has a reference to the file pointer in some 2087 * manner or other). 2088 * 2089 * This is a rare case where callers are allowed to hold spinlocks, so 2090 * we can't ourselves. Since we are not obtaining the fp spinlock, 2091 * we have to use an atomic lock to interlock against fdrop(). 2092 */ 2093 void 2094 fhold(struct file *fp) 2095 { 2096 atomic_add_int(&fp->f_count, 1); 2097 } 2098 2099 /* 2100 * A spinlock is required to handle 1->0 transitions on f_count. We have 2101 * to use atomic_sub_int so as not to race the atomic_add_int in fhold(). 2102 * 2103 * MPALMOSTSAFE - acquires mplock for final close sequence 2104 */ 2105 int 2106 fdrop(struct file *fp) 2107 { 2108 struct flock lf; 2109 struct vnode *vp; 2110 int error; 2111 2112 spin_lock_wr(&fp->f_spin); 2113 atomic_subtract_int(&fp->f_count, 1); 2114 if (fp->f_count > 0) { 2115 spin_unlock_wr(&fp->f_spin); 2116 return (0); 2117 } 2118 spin_unlock_wr(&fp->f_spin); 2119 2120 get_mplock(); 2121 2122 /* 2123 * The last reference has gone away, we own the fp structure free 2124 * and clear. 2125 */ 2126 if (fp->f_count < 0) 2127 panic("fdrop: count < 0"); 2128 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE && 2129 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2130 ) { 2131 lf.l_whence = SEEK_SET; 2132 lf.l_start = 0; 2133 lf.l_len = 0; 2134 lf.l_type = F_UNLCK; 2135 vp = (struct vnode *)fp->f_data; 2136 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2137 } 2138 if (fp->f_ops != &badfileops) 2139 error = fo_close(fp); 2140 else 2141 error = 0; 2142 ffree(fp); 2143 rel_mplock(); 2144 return (error); 2145 } 2146 2147 /* 2148 * Apply an advisory lock on a file descriptor. 2149 * 2150 * Just attempt to get a record lock of the requested type on 2151 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2152 */ 2153 int 2154 sys_flock(struct flock_args *uap) 2155 { 2156 struct proc *p = curproc; 2157 struct file *fp; 2158 struct vnode *vp; 2159 struct flock lf; 2160 int error; 2161 2162 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 2163 return (EBADF); 2164 if (fp->f_type != DTYPE_VNODE) { 2165 error = EOPNOTSUPP; 2166 goto done; 2167 } 2168 vp = (struct vnode *)fp->f_data; 2169 lf.l_whence = SEEK_SET; 2170 lf.l_start = 0; 2171 lf.l_len = 0; 2172 if (uap->how & LOCK_UN) { 2173 lf.l_type = F_UNLCK; 2174 fp->f_flag &= ~FHASLOCK; 2175 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2176 goto done; 2177 } 2178 if (uap->how & LOCK_EX) 2179 lf.l_type = F_WRLCK; 2180 else if (uap->how & LOCK_SH) 2181 lf.l_type = F_RDLCK; 2182 else { 2183 error = EBADF; 2184 goto done; 2185 } 2186 fp->f_flag |= FHASLOCK; 2187 if (uap->how & LOCK_NB) 2188 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0); 2189 else 2190 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT); 2191 done: 2192 fdrop(fp); 2193 return (error); 2194 } 2195 2196 /* 2197 * File Descriptor pseudo-device driver (/dev/fd/). 2198 * 2199 * Opening minor device N dup()s the file (if any) connected to file 2200 * descriptor N belonging to the calling process. Note that this driver 2201 * consists of only the ``open()'' routine, because all subsequent 2202 * references to this file will be direct to the other driver. 2203 */ 2204 /* ARGSUSED */ 2205 static int 2206 fdopen(struct dev_open_args *ap) 2207 { 2208 thread_t td = curthread; 2209 2210 KKASSERT(td->td_lwp != NULL); 2211 2212 /* 2213 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the 2214 * the file descriptor being sought for duplication. The error 2215 * return ensures that the vnode for this device will be released 2216 * by vn_open. Open will detect this special error and take the 2217 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2218 * will simply report the error. 2219 */ 2220 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev); 2221 return (ENODEV); 2222 } 2223 2224 /* 2225 * The caller has reserved the file descriptor dfd for us. On success we 2226 * must fsetfd() it. On failure the caller will clean it up. 2227 * 2228 * NOT MPSAFE - isn't getting spinlocks, possibly other things 2229 */ 2230 int 2231 dupfdopen(struct proc *p, int dfd, int sfd, int mode, int error) 2232 { 2233 struct filedesc *fdp = p->p_fd; 2234 struct file *wfp; 2235 struct file *xfp; 2236 2237 if ((wfp = holdfp(fdp, sfd, -1)) == NULL) 2238 return (EBADF); 2239 2240 /* 2241 * There are two cases of interest here. 2242 * 2243 * For ENODEV simply dup sfd to file descriptor dfd and return. 2244 * 2245 * For ENXIO steal away the file structure from sfd and store it 2246 * dfd. sfd is effectively closed by this operation. 2247 * 2248 * Any other error code is just returned. 2249 */ 2250 switch (error) { 2251 case ENODEV: 2252 /* 2253 * Check that the mode the file is being opened for is a 2254 * subset of the mode of the existing descriptor. 2255 */ 2256 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) 2257 return (EACCES); 2258 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2259 fsetfd(p, wfp, dfd); 2260 error = 0; 2261 break; 2262 case ENXIO: 2263 /* 2264 * Steal away the file pointer from dfd, and stuff it into indx. 2265 */ 2266 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2267 fsetfd(p, wfp, dfd); 2268 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) 2269 fdrop(xfp); 2270 KKASSERT(xfp == wfp); /* XXX MP RACE */ 2271 error = 0; 2272 break; 2273 default: 2274 break; 2275 } 2276 fdrop(wfp); 2277 return (error); 2278 } 2279 2280 /* 2281 * NOT MPSAFE - I think these refer to a common file descriptor table 2282 * and we need to spinlock that to link fdtol in. 2283 */ 2284 struct filedesc_to_leader * 2285 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 2286 struct proc *leader) 2287 { 2288 struct filedesc_to_leader *fdtol; 2289 2290 fdtol = kmalloc(sizeof(struct filedesc_to_leader), 2291 M_FILEDESC_TO_LEADER, M_WAITOK); 2292 fdtol->fdl_refcount = 1; 2293 fdtol->fdl_holdcount = 0; 2294 fdtol->fdl_wakeup = 0; 2295 fdtol->fdl_leader = leader; 2296 if (old != NULL) { 2297 fdtol->fdl_next = old->fdl_next; 2298 fdtol->fdl_prev = old; 2299 old->fdl_next = fdtol; 2300 fdtol->fdl_next->fdl_prev = fdtol; 2301 } else { 2302 fdtol->fdl_next = fdtol; 2303 fdtol->fdl_prev = fdtol; 2304 } 2305 return fdtol; 2306 } 2307 2308 /* 2309 * Scan all file pointers in the system. The callback is made with 2310 * both the master list spinlock held and the fp spinlock held, 2311 * both exclusively. 2312 * 2313 * MPSAFE 2314 * 2315 * WARNING: both the filehead spinlock and the file pointer spinlock are 2316 * held exclusively when the callback is made. The file pointer is not 2317 * referenced. 2318 */ 2319 void 2320 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data) 2321 { 2322 struct file *fp; 2323 int res; 2324 2325 spin_lock_wr(&filehead_spin); 2326 LIST_FOREACH(fp, &filehead, f_list) { 2327 spin_lock_wr(&fp->f_spin); 2328 res = callback(fp, data); 2329 spin_unlock_wr(&fp->f_spin); 2330 if (res < 0) 2331 break; 2332 } 2333 spin_unlock_wr(&filehead_spin); 2334 } 2335 2336 /* 2337 * Get file structures. 2338 * 2339 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe) 2340 */ 2341 2342 struct sysctl_kern_file_info { 2343 int count; 2344 int error; 2345 struct sysctl_req *req; 2346 }; 2347 2348 static int sysctl_kern_file_callback(struct proc *p, void *data); 2349 2350 static int 2351 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2352 { 2353 struct sysctl_kern_file_info info; 2354 2355 /* 2356 * Note: because the number of file descriptors is calculated 2357 * in different ways for sizing vs returning the data, 2358 * there is information leakage from the first loop. However, 2359 * it is of a similar order of magnitude to the leakage from 2360 * global system statistics such as kern.openfiles. 2361 * 2362 * When just doing a count, note that we cannot just count 2363 * the elements and add f_count via the filehead list because 2364 * threaded processes share their descriptor table and f_count might 2365 * still be '1' in that case. 2366 * 2367 * Since the SYSCTL op can block, we must hold the process to 2368 * prevent it being ripped out from under us either in the 2369 * file descriptor loop or in the greater LIST_FOREACH. The 2370 * process may be in varying states of disrepair. If the process 2371 * is in SZOMB we may have caught it just as it is being removed 2372 * from the allproc list, we must skip it in that case to maintain 2373 * an unbroken chain through the allproc list. 2374 */ 2375 info.count = 0; 2376 info.error = 0; 2377 info.req = req; 2378 allproc_scan(sysctl_kern_file_callback, &info); 2379 2380 /* 2381 * When just calculating the size, overestimate a bit to try to 2382 * prevent system activity from causing the buffer-fill call 2383 * to fail later on. 2384 */ 2385 if (req->oldptr == NULL) { 2386 info.count = (info.count + 16) + (info.count / 10); 2387 info.error = SYSCTL_OUT(req, NULL, 2388 info.count * sizeof(struct kinfo_file)); 2389 } 2390 return (info.error); 2391 } 2392 2393 static int 2394 sysctl_kern_file_callback(struct proc *p, void *data) 2395 { 2396 struct sysctl_kern_file_info *info = data; 2397 struct kinfo_file kf; 2398 struct filedesc *fdp; 2399 struct file *fp; 2400 uid_t uid; 2401 int n; 2402 2403 if (p->p_stat == SIDL || p->p_stat == SZOMB) 2404 return(0); 2405 if (!PRISON_CHECK(info->req->td->td_proc->p_ucred, p->p_ucred) != 0) 2406 return(0); 2407 if ((fdp = p->p_fd) == NULL) 2408 return(0); 2409 spin_lock_rd(&fdp->fd_spin); 2410 for (n = 0; n < fdp->fd_nfiles; ++n) { 2411 if ((fp = fdp->fd_files[n].fp) == NULL) 2412 continue; 2413 if (info->req->oldptr == NULL) { 2414 ++info->count; 2415 } else { 2416 uid = p->p_ucred ? p->p_ucred->cr_uid : -1; 2417 kcore_make_file(&kf, fp, p->p_pid, uid, n); 2418 spin_unlock_rd(&fdp->fd_spin); 2419 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf)); 2420 spin_lock_rd(&fdp->fd_spin); 2421 if (info->error) 2422 break; 2423 } 2424 } 2425 spin_unlock_rd(&fdp->fd_spin); 2426 if (info->error) 2427 return(-1); 2428 return(0); 2429 } 2430 2431 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2432 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 2433 2434 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2435 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2436 2437 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2438 &maxfiles, 0, "Maximum number of files"); 2439 2440 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, 2441 &maxfilesrootres, 0, "Descriptors reserved for root use"); 2442 2443 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2444 &nfiles, 0, "System-wide number of open files"); 2445 2446 static void 2447 fildesc_drvinit(void *unused) 2448 { 2449 int fd; 2450 2451 dev_ops_add(&fildesc_ops, 0, 0); 2452 for (fd = 0; fd < NUMFDESC; fd++) { 2453 make_dev(&fildesc_ops, fd, 2454 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 2455 } 2456 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 2457 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 2458 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 2459 } 2460 2461 /* 2462 * MPSAFE 2463 */ 2464 struct fileops badfileops = { 2465 .fo_read = badfo_readwrite, 2466 .fo_write = badfo_readwrite, 2467 .fo_ioctl = badfo_ioctl, 2468 .fo_poll = badfo_poll, 2469 .fo_kqfilter = badfo_kqfilter, 2470 .fo_stat = badfo_stat, 2471 .fo_close = badfo_close, 2472 .fo_shutdown = badfo_shutdown 2473 }; 2474 2475 /* 2476 * MPSAFE 2477 */ 2478 static int 2479 badfo_readwrite( 2480 struct file *fp, 2481 struct uio *uio, 2482 struct ucred *cred, 2483 int flags 2484 ) { 2485 return (EBADF); 2486 } 2487 2488 /* 2489 * MPSAFE 2490 */ 2491 static int 2492 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *cred) 2493 { 2494 return (EBADF); 2495 } 2496 2497 /* 2498 * MPSAFE 2499 */ 2500 static int 2501 badfo_poll(struct file *fp, int events, struct ucred *cred) 2502 { 2503 return (0); 2504 } 2505 2506 /* 2507 * MPSAFE 2508 */ 2509 static int 2510 badfo_kqfilter(struct file *fp, struct knote *kn) 2511 { 2512 return (0); 2513 } 2514 2515 static int 2516 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred) 2517 { 2518 return (EBADF); 2519 } 2520 2521 /* 2522 * MPSAFE 2523 */ 2524 static int 2525 badfo_close(struct file *fp) 2526 { 2527 return (EBADF); 2528 } 2529 2530 /* 2531 * MPSAFE 2532 */ 2533 static int 2534 badfo_shutdown(struct file *fp, int how) 2535 { 2536 return (EBADF); 2537 } 2538 2539 /* 2540 * MPSAFE 2541 */ 2542 int 2543 nofo_shutdown(struct file *fp, int how) 2544 { 2545 return (EOPNOTSUPP); 2546 } 2547 2548 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, 2549 fildesc_drvinit,NULL) 2550