1 /* 2 * Copyright (c) 2005 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Jeffrey Hsu. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * 35 * Copyright (c) 1982, 1986, 1989, 1991, 1993 36 * The Regents of the University of California. All rights reserved. 37 * (c) UNIX System Laboratories, Inc. 38 * All or some portions of this file are derived from material licensed 39 * to the University of California by American Telephone and Telegraph 40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 41 * the permission of UNIX System Laboratories, Inc. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. Neither the name of the University nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 * 67 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 68 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $ 69 */ 70 71 #include "opt_compat.h" 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/malloc.h> 75 #include <sys/sysproto.h> 76 #include <sys/conf.h> 77 #include <sys/device.h> 78 #include <sys/file.h> 79 #include <sys/filedesc.h> 80 #include <sys/kernel.h> 81 #include <sys/sysctl.h> 82 #include <sys/vnode.h> 83 #include <sys/proc.h> 84 #include <sys/nlookup.h> 85 #include <sys/stat.h> 86 #include <sys/filio.h> 87 #include <sys/fcntl.h> 88 #include <sys/unistd.h> 89 #include <sys/resourcevar.h> 90 #include <sys/event.h> 91 #include <sys/kern_syscall.h> 92 #include <sys/kcore.h> 93 #include <sys/kinfo.h> 94 #include <sys/un.h> 95 #include <sys/objcache.h> 96 97 #include <vm/vm.h> 98 #include <vm/vm_extern.h> 99 100 #include <sys/thread2.h> 101 #include <sys/file2.h> 102 #include <sys/spinlock2.h> 103 104 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd); 105 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr); 106 static struct file *funsetfd_locked (struct filedesc *fdp, int fd); 107 static void ffree(struct file *fp); 108 109 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 110 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 111 "file desc to leader structures"); 112 MALLOC_DEFINE(M_FILE, "file", "Open file structure"); 113 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 114 115 static struct krate krate_uidinfo = { .freq = 1 }; 116 117 static d_open_t fdopen; 118 #define NUMFDESC 64 119 120 #define CDEV_MAJOR 22 121 static struct dev_ops fildesc_ops = { 122 { "FD", 0, 0 }, 123 .d_open = fdopen, 124 }; 125 126 /* 127 * Descriptor management. 128 */ 129 #ifndef NFILELIST_HEADS 130 #define NFILELIST_HEADS 257 /* primary number */ 131 #endif 132 133 struct filelist_head { 134 struct spinlock spin; 135 struct filelist list; 136 } __cachealign; 137 138 static struct filelist_head filelist_heads[NFILELIST_HEADS]; 139 140 static int nfiles; /* actual number of open files */ 141 extern int cmask; 142 143 struct lwkt_token revoke_token = LWKT_TOKEN_INITIALIZER(revoke_token); 144 145 static struct objcache *file_objcache; 146 147 static struct objcache_malloc_args file_malloc_args = { 148 .objsize = sizeof(struct file), 149 .mtype = M_FILE 150 }; 151 152 /* 153 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared. 154 * 155 * must be called with fdp->fd_spin exclusively held 156 */ 157 static __inline 158 void 159 fdfixup_locked(struct filedesc *fdp, int fd) 160 { 161 if (fd < fdp->fd_freefile) { 162 fdp->fd_freefile = fd; 163 } 164 while (fdp->fd_lastfile >= 0 && 165 fdp->fd_files[fdp->fd_lastfile].fp == NULL && 166 fdp->fd_files[fdp->fd_lastfile].reserved == 0 167 ) { 168 --fdp->fd_lastfile; 169 } 170 } 171 172 static __inline struct filelist_head * 173 fp2filelist(const struct file *fp) 174 { 175 u_int i; 176 177 i = (u_int)(uintptr_t)fp % NFILELIST_HEADS; 178 return &filelist_heads[i]; 179 } 180 181 /* 182 * System calls on descriptors. 183 */ 184 int 185 sys_getdtablesize(struct getdtablesize_args *uap) 186 { 187 struct proc *p = curproc; 188 struct plimit *limit = p->p_limit; 189 int dtsize; 190 191 spin_lock(&limit->p_spin); 192 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 193 dtsize = INT_MAX; 194 else 195 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur; 196 spin_unlock(&limit->p_spin); 197 198 if (dtsize > maxfilesperproc) 199 dtsize = maxfilesperproc; 200 if (dtsize < minfilesperproc) 201 dtsize = minfilesperproc; 202 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser) 203 dtsize = maxfilesperuser; 204 uap->sysmsg_result = dtsize; 205 return (0); 206 } 207 208 /* 209 * Duplicate a file descriptor to a particular value. 210 * 211 * note: keep in mind that a potential race condition exists when closing 212 * descriptors from a shared descriptor table (via rfork). 213 */ 214 int 215 sys_dup2(struct dup2_args *uap) 216 { 217 int error; 218 int fd = 0; 219 220 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd); 221 uap->sysmsg_fds[0] = fd; 222 223 return (error); 224 } 225 226 /* 227 * Duplicate a file descriptor. 228 */ 229 int 230 sys_dup(struct dup_args *uap) 231 { 232 int error; 233 int fd = 0; 234 235 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd); 236 uap->sysmsg_fds[0] = fd; 237 238 return (error); 239 } 240 241 /* 242 * MPALMOSTSAFE - acquires mplock for fp operations 243 */ 244 int 245 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred) 246 { 247 struct thread *td = curthread; 248 struct proc *p = td->td_proc; 249 struct file *fp; 250 struct vnode *vp; 251 u_int newmin; 252 u_int oflags; 253 u_int nflags; 254 int tmp, error, flg = F_POSIX; 255 256 KKASSERT(p); 257 258 /* 259 * Operations on file descriptors that do not require a file pointer. 260 */ 261 switch (cmd) { 262 case F_GETFD: 263 error = fgetfdflags(p->p_fd, fd, &tmp); 264 if (error == 0) 265 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0; 266 return (error); 267 268 case F_SETFD: 269 if (dat->fc_cloexec & FD_CLOEXEC) 270 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE); 271 else 272 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE); 273 return (error); 274 case F_DUPFD: 275 newmin = dat->fc_fd; 276 error = kern_dup(DUP_VARIABLE | DUP_FCNTL, fd, newmin, 277 &dat->fc_fd); 278 return (error); 279 case F_DUPFD_CLOEXEC: 280 newmin = dat->fc_fd; 281 error = kern_dup(DUP_VARIABLE | DUP_CLOEXEC | DUP_FCNTL, 282 fd, newmin, &dat->fc_fd); 283 return (error); 284 case F_DUP2FD: 285 newmin = dat->fc_fd; 286 error = kern_dup(DUP_FIXED, fd, newmin, &dat->fc_fd); 287 return (error); 288 case F_DUP2FD_CLOEXEC: 289 newmin = dat->fc_fd; 290 error = kern_dup(DUP_FIXED | DUP_CLOEXEC, fd, newmin, 291 &dat->fc_fd); 292 return (error); 293 default: 294 break; 295 } 296 297 /* 298 * Operations on file pointers 299 */ 300 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 301 return (EBADF); 302 303 switch (cmd) { 304 case F_GETFL: 305 dat->fc_flags = OFLAGS(fp->f_flag); 306 error = 0; 307 break; 308 309 case F_SETFL: 310 oflags = fp->f_flag; 311 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; 312 nflags |= oflags & ~FCNTLFLAGS; 313 314 error = 0; 315 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY)) 316 error = EINVAL; 317 if (error == 0 && ((nflags ^ oflags) & FASYNC)) { 318 tmp = nflags & FASYNC; 319 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, 320 cred, NULL); 321 } 322 323 /* 324 * If no error, must be atomically set. 325 */ 326 while (error == 0) { 327 oflags = fp->f_flag; 328 cpu_ccfence(); 329 nflags = (oflags & ~FCNTLFLAGS) | (nflags & FCNTLFLAGS); 330 if (atomic_cmpset_int(&fp->f_flag, oflags, nflags)) 331 break; 332 cpu_pause(); 333 } 334 break; 335 336 case F_GETOWN: 337 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, 338 cred, NULL); 339 break; 340 341 case F_SETOWN: 342 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, 343 cred, NULL); 344 break; 345 346 case F_SETLKW: 347 flg |= F_WAIT; 348 /* Fall into F_SETLK */ 349 350 case F_SETLK: 351 if (fp->f_type != DTYPE_VNODE) { 352 error = EBADF; 353 break; 354 } 355 vp = (struct vnode *)fp->f_data; 356 357 /* 358 * copyin/lockop may block 359 */ 360 if (dat->fc_flock.l_whence == SEEK_CUR) 361 dat->fc_flock.l_start += fp->f_offset; 362 363 switch (dat->fc_flock.l_type) { 364 case F_RDLCK: 365 if ((fp->f_flag & FREAD) == 0) { 366 error = EBADF; 367 break; 368 } 369 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) { 370 lwkt_gettoken(&p->p_leader->p_token); 371 p->p_leader->p_flags |= P_ADVLOCK; 372 lwkt_reltoken(&p->p_leader->p_token); 373 } 374 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 375 &dat->fc_flock, flg); 376 break; 377 case F_WRLCK: 378 if ((fp->f_flag & FWRITE) == 0) { 379 error = EBADF; 380 break; 381 } 382 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) { 383 lwkt_gettoken(&p->p_leader->p_token); 384 p->p_leader->p_flags |= P_ADVLOCK; 385 lwkt_reltoken(&p->p_leader->p_token); 386 } 387 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 388 &dat->fc_flock, flg); 389 break; 390 case F_UNLCK: 391 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 392 &dat->fc_flock, F_POSIX); 393 break; 394 default: 395 error = EINVAL; 396 break; 397 } 398 399 /* 400 * It is possible to race a close() on the descriptor while 401 * we were blocked getting the lock. If this occurs the 402 * close might not have caught the lock. 403 */ 404 if (checkfdclosed(p->p_fd, fd, fp)) { 405 dat->fc_flock.l_whence = SEEK_SET; 406 dat->fc_flock.l_start = 0; 407 dat->fc_flock.l_len = 0; 408 dat->fc_flock.l_type = F_UNLCK; 409 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 410 F_UNLCK, &dat->fc_flock, F_POSIX); 411 } 412 break; 413 414 case F_GETLK: 415 if (fp->f_type != DTYPE_VNODE) { 416 error = EBADF; 417 break; 418 } 419 vp = (struct vnode *)fp->f_data; 420 /* 421 * copyin/lockop may block 422 */ 423 if (dat->fc_flock.l_type != F_RDLCK && 424 dat->fc_flock.l_type != F_WRLCK && 425 dat->fc_flock.l_type != F_UNLCK) { 426 error = EINVAL; 427 break; 428 } 429 if (dat->fc_flock.l_whence == SEEK_CUR) 430 dat->fc_flock.l_start += fp->f_offset; 431 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, 432 &dat->fc_flock, F_POSIX); 433 break; 434 default: 435 error = EINVAL; 436 break; 437 } 438 439 fdrop(fp); 440 return (error); 441 } 442 443 /* 444 * The file control system call. 445 */ 446 int 447 sys_fcntl(struct fcntl_args *uap) 448 { 449 union fcntl_dat dat; 450 int error; 451 452 switch (uap->cmd) { 453 case F_DUPFD: 454 case F_DUP2FD: 455 case F_DUPFD_CLOEXEC: 456 case F_DUP2FD_CLOEXEC: 457 dat.fc_fd = uap->arg; 458 break; 459 case F_SETFD: 460 dat.fc_cloexec = uap->arg; 461 break; 462 case F_SETFL: 463 dat.fc_flags = uap->arg; 464 break; 465 case F_SETOWN: 466 dat.fc_owner = uap->arg; 467 break; 468 case F_SETLKW: 469 case F_SETLK: 470 case F_GETLK: 471 error = copyin((caddr_t)uap->arg, &dat.fc_flock, 472 sizeof(struct flock)); 473 if (error) 474 return (error); 475 break; 476 } 477 478 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred); 479 480 if (error == 0) { 481 switch (uap->cmd) { 482 case F_DUPFD: 483 case F_DUP2FD: 484 case F_DUPFD_CLOEXEC: 485 case F_DUP2FD_CLOEXEC: 486 uap->sysmsg_result = dat.fc_fd; 487 break; 488 case F_GETFD: 489 uap->sysmsg_result = dat.fc_cloexec; 490 break; 491 case F_GETFL: 492 uap->sysmsg_result = dat.fc_flags; 493 break; 494 case F_GETOWN: 495 uap->sysmsg_result = dat.fc_owner; 496 break; 497 case F_GETLK: 498 error = copyout(&dat.fc_flock, (caddr_t)uap->arg, 499 sizeof(struct flock)); 500 break; 501 } 502 } 503 504 return (error); 505 } 506 507 /* 508 * Common code for dup, dup2, and fcntl(F_DUPFD). 509 * 510 * There are four type flags: DUP_FCNTL, DUP_FIXED, DUP_VARIABLE, and 511 * DUP_CLOEXEC. 512 * 513 * DUP_FCNTL is for handling EINVAL vs. EBADF differences between 514 * fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC and dup2() (per POSIX). 515 * The next two flags are mutually exclusive, and the fourth is optional. 516 * DUP_FIXED tells kern_dup() to destructively dup over an existing file 517 * descriptor if "new" is already open. DUP_VARIABLE tells kern_dup() 518 * to find the lowest unused file descriptor that is greater than or 519 * equal to "new". DUP_CLOEXEC, which works with either of the first 520 * two flags, sets the close-on-exec flag on the "new" file descriptor. 521 */ 522 int 523 kern_dup(int flags, int old, int new, int *res) 524 { 525 struct thread *td = curthread; 526 struct proc *p = td->td_proc; 527 struct filedesc *fdp = p->p_fd; 528 struct file *fp; 529 struct file *delfp; 530 int oldflags; 531 int holdleaders; 532 int dtsize; 533 int error, newfd; 534 535 /* 536 * Verify that we have a valid descriptor to dup from and 537 * possibly to dup to. When the new descriptor is out of 538 * bounds, fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC must 539 * return EINVAL, while dup2() returns EBADF in 540 * this case. 541 * 542 * NOTE: maxfilesperuser is not applicable to dup() 543 */ 544 retry: 545 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 546 dtsize = INT_MAX; 547 else 548 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 549 if (dtsize > maxfilesperproc) 550 dtsize = maxfilesperproc; 551 if (dtsize < minfilesperproc) 552 dtsize = minfilesperproc; 553 554 if (new < 0 || new > dtsize) 555 return (flags & DUP_FCNTL ? EINVAL : EBADF); 556 557 spin_lock(&fdp->fd_spin); 558 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) { 559 spin_unlock(&fdp->fd_spin); 560 return (EBADF); 561 } 562 if ((flags & DUP_FIXED) && old == new) { 563 *res = new; 564 if (flags & DUP_CLOEXEC) 565 fdp->fd_files[new].fileflags |= UF_EXCLOSE; 566 spin_unlock(&fdp->fd_spin); 567 return (0); 568 } 569 fp = fdp->fd_files[old].fp; 570 oldflags = fdp->fd_files[old].fileflags; 571 fhold(fp); 572 573 /* 574 * Allocate a new descriptor if DUP_VARIABLE, or expand the table 575 * if the requested descriptor is beyond the current table size. 576 * 577 * This can block. Retry if the source descriptor no longer matches 578 * or if our expectation in the expansion case races. 579 * 580 * If we are not expanding or allocating a new decriptor, then reset 581 * the target descriptor to a reserved state so we have a uniform 582 * setup for the next code block. 583 */ 584 if ((flags & DUP_VARIABLE) || new >= fdp->fd_nfiles) { 585 spin_unlock(&fdp->fd_spin); 586 error = fdalloc(p, new, &newfd); 587 spin_lock(&fdp->fd_spin); 588 if (error) { 589 spin_unlock(&fdp->fd_spin); 590 fdrop(fp); 591 return (error); 592 } 593 /* 594 * Check for ripout 595 */ 596 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) { 597 fsetfd_locked(fdp, NULL, newfd); 598 spin_unlock(&fdp->fd_spin); 599 fdrop(fp); 600 goto retry; 601 } 602 /* 603 * Check for expansion race 604 */ 605 if ((flags & DUP_VARIABLE) == 0 && new != newfd) { 606 fsetfd_locked(fdp, NULL, newfd); 607 spin_unlock(&fdp->fd_spin); 608 fdrop(fp); 609 goto retry; 610 } 611 /* 612 * Check for ripout, newfd reused old (this case probably 613 * can't occur). 614 */ 615 if (old == newfd) { 616 fsetfd_locked(fdp, NULL, newfd); 617 spin_unlock(&fdp->fd_spin); 618 fdrop(fp); 619 goto retry; 620 } 621 new = newfd; 622 delfp = NULL; 623 } else { 624 if (fdp->fd_files[new].reserved) { 625 spin_unlock(&fdp->fd_spin); 626 fdrop(fp); 627 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new); 628 tsleep(fdp, 0, "fdres", hz); 629 goto retry; 630 } 631 632 /* 633 * If the target descriptor was never allocated we have 634 * to allocate it. If it was we have to clean out the 635 * old descriptor. delfp inherits the ref from the 636 * descriptor table. 637 */ 638 delfp = fdp->fd_files[new].fp; 639 fdp->fd_files[new].fp = NULL; 640 fdp->fd_files[new].reserved = 1; 641 if (delfp == NULL) { 642 fdreserve_locked(fdp, new, 1); 643 if (new > fdp->fd_lastfile) 644 fdp->fd_lastfile = new; 645 } 646 647 } 648 649 /* 650 * NOTE: still holding an exclusive spinlock 651 */ 652 653 /* 654 * If a descriptor is being overwritten we may hve to tell 655 * fdfree() to sleep to ensure that all relevant process 656 * leaders can be traversed in closef(). 657 */ 658 if (delfp != NULL && p->p_fdtol != NULL) { 659 fdp->fd_holdleaderscount++; 660 holdleaders = 1; 661 } else { 662 holdleaders = 0; 663 } 664 KASSERT(delfp == NULL || (flags & DUP_FIXED), 665 ("dup() picked an open file")); 666 667 /* 668 * Duplicate the source descriptor, update lastfile. If the new 669 * descriptor was not allocated and we aren't replacing an existing 670 * descriptor we have to mark the descriptor as being in use. 671 * 672 * The fd_files[] array inherits fp's hold reference. 673 */ 674 fsetfd_locked(fdp, fp, new); 675 if ((flags & DUP_CLOEXEC) != 0) 676 fdp->fd_files[new].fileflags = oldflags | UF_EXCLOSE; 677 else 678 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE; 679 spin_unlock(&fdp->fd_spin); 680 fdrop(fp); 681 *res = new; 682 683 /* 684 * If we dup'd over a valid file, we now own the reference to it 685 * and must dispose of it using closef() semantics (as if a 686 * close() were performed on it). 687 */ 688 if (delfp) { 689 if (SLIST_FIRST(&delfp->f_klist)) 690 knote_fdclose(delfp, fdp, new); 691 closef(delfp, p); 692 if (holdleaders) { 693 spin_lock(&fdp->fd_spin); 694 fdp->fd_holdleaderscount--; 695 if (fdp->fd_holdleaderscount == 0 && 696 fdp->fd_holdleaderswakeup != 0) { 697 fdp->fd_holdleaderswakeup = 0; 698 spin_unlock(&fdp->fd_spin); 699 wakeup(&fdp->fd_holdleaderscount); 700 } else { 701 spin_unlock(&fdp->fd_spin); 702 } 703 } 704 } 705 return (0); 706 } 707 708 /* 709 * If sigio is on the list associated with a process or process group, 710 * disable signalling from the device, remove sigio from the list and 711 * free sigio. 712 */ 713 void 714 funsetown(struct sigio **sigiop) 715 { 716 struct pgrp *pgrp; 717 struct proc *p; 718 struct sigio *sigio; 719 720 if ((sigio = *sigiop) != NULL) { 721 lwkt_gettoken(&sigio_token); /* protect sigio */ 722 KKASSERT(sigiop == sigio->sio_myref); 723 sigio = *sigiop; 724 *sigiop = NULL; 725 lwkt_reltoken(&sigio_token); 726 } 727 if (sigio == NULL) 728 return; 729 730 if (sigio->sio_pgid < 0) { 731 pgrp = sigio->sio_pgrp; 732 sigio->sio_pgrp = NULL; 733 lwkt_gettoken(&pgrp->pg_token); 734 SLIST_REMOVE(&pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio); 735 lwkt_reltoken(&pgrp->pg_token); 736 pgrel(pgrp); 737 } else /* if ((*sigiop)->sio_pgid > 0) */ { 738 p = sigio->sio_proc; 739 sigio->sio_proc = NULL; 740 PHOLD(p); 741 lwkt_gettoken(&p->p_token); 742 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); 743 lwkt_reltoken(&p->p_token); 744 PRELE(p); 745 } 746 crfree(sigio->sio_ucred); 747 sigio->sio_ucred = NULL; 748 kfree(sigio, M_SIGIO); 749 } 750 751 /* 752 * Free a list of sigio structures. Caller is responsible for ensuring 753 * that the list is MPSAFE. 754 */ 755 void 756 funsetownlst(struct sigiolst *sigiolst) 757 { 758 struct sigio *sigio; 759 760 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) 761 funsetown(sigio->sio_myref); 762 } 763 764 /* 765 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 766 * 767 * After permission checking, add a sigio structure to the sigio list for 768 * the process or process group. 769 */ 770 int 771 fsetown(pid_t pgid, struct sigio **sigiop) 772 { 773 struct proc *proc = NULL; 774 struct pgrp *pgrp = NULL; 775 struct sigio *sigio; 776 int error; 777 778 if (pgid == 0) { 779 funsetown(sigiop); 780 return (0); 781 } 782 783 if (pgid > 0) { 784 proc = pfind(pgid); 785 if (proc == NULL) { 786 error = ESRCH; 787 goto done; 788 } 789 790 /* 791 * Policy - Don't allow a process to FSETOWN a process 792 * in another session. 793 * 794 * Remove this test to allow maximum flexibility or 795 * restrict FSETOWN to the current process or process 796 * group for maximum safety. 797 */ 798 if (proc->p_session != curproc->p_session) { 799 error = EPERM; 800 goto done; 801 } 802 } else /* if (pgid < 0) */ { 803 pgrp = pgfind(-pgid); 804 if (pgrp == NULL) { 805 error = ESRCH; 806 goto done; 807 } 808 809 /* 810 * Policy - Don't allow a process to FSETOWN a process 811 * in another session. 812 * 813 * Remove this test to allow maximum flexibility or 814 * restrict FSETOWN to the current process or process 815 * group for maximum safety. 816 */ 817 if (pgrp->pg_session != curproc->p_session) { 818 error = EPERM; 819 goto done; 820 } 821 } 822 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK | M_ZERO); 823 if (pgid > 0) { 824 KKASSERT(pgrp == NULL); 825 lwkt_gettoken(&proc->p_token); 826 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 827 sigio->sio_proc = proc; 828 lwkt_reltoken(&proc->p_token); 829 } else { 830 KKASSERT(proc == NULL); 831 lwkt_gettoken(&pgrp->pg_token); 832 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 833 sigio->sio_pgrp = pgrp; 834 lwkt_reltoken(&pgrp->pg_token); 835 pgrp = NULL; 836 } 837 sigio->sio_pgid = pgid; 838 sigio->sio_ucred = crhold(curthread->td_ucred); 839 /* It would be convenient if p_ruid was in ucred. */ 840 sigio->sio_ruid = sigio->sio_ucred->cr_ruid; 841 sigio->sio_myref = sigiop; 842 843 lwkt_gettoken(&sigio_token); 844 while (*sigiop) 845 funsetown(sigiop); 846 *sigiop = sigio; 847 lwkt_reltoken(&sigio_token); 848 error = 0; 849 done: 850 if (pgrp) 851 pgrel(pgrp); 852 if (proc) 853 PRELE(proc); 854 return (error); 855 } 856 857 /* 858 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 859 */ 860 pid_t 861 fgetown(struct sigio **sigiop) 862 { 863 struct sigio *sigio; 864 pid_t own; 865 866 lwkt_gettoken_shared(&sigio_token); 867 sigio = *sigiop; 868 own = (sigio != NULL ? sigio->sio_pgid : 0); 869 lwkt_reltoken(&sigio_token); 870 871 return (own); 872 } 873 874 /* 875 * Close many file descriptors. 876 */ 877 int 878 sys_closefrom(struct closefrom_args *uap) 879 { 880 return(kern_closefrom(uap->fd)); 881 } 882 883 /* 884 * Close all file descriptors greater then or equal to fd 885 */ 886 int 887 kern_closefrom(int fd) 888 { 889 struct thread *td = curthread; 890 struct proc *p = td->td_proc; 891 struct filedesc *fdp; 892 893 KKASSERT(p); 894 fdp = p->p_fd; 895 896 if (fd < 0) 897 return (EINVAL); 898 899 /* 900 * NOTE: This function will skip unassociated descriptors and 901 * reserved descriptors that have not yet been assigned. 902 * fd_lastfile can change as a side effect of kern_close(). 903 */ 904 spin_lock(&fdp->fd_spin); 905 while (fd <= fdp->fd_lastfile) { 906 if (fdp->fd_files[fd].fp != NULL) { 907 spin_unlock(&fdp->fd_spin); 908 /* ok if this races another close */ 909 if (kern_close(fd) == EINTR) 910 return (EINTR); 911 spin_lock(&fdp->fd_spin); 912 } 913 ++fd; 914 } 915 spin_unlock(&fdp->fd_spin); 916 return (0); 917 } 918 919 /* 920 * Close a file descriptor. 921 */ 922 int 923 sys_close(struct close_args *uap) 924 { 925 return(kern_close(uap->fd)); 926 } 927 928 /* 929 * close() helper 930 */ 931 int 932 kern_close(int fd) 933 { 934 struct thread *td = curthread; 935 struct proc *p = td->td_proc; 936 struct filedesc *fdp; 937 struct file *fp; 938 int error; 939 int holdleaders; 940 941 KKASSERT(p); 942 fdp = p->p_fd; 943 944 spin_lock(&fdp->fd_spin); 945 if ((fp = funsetfd_locked(fdp, fd)) == NULL) { 946 spin_unlock(&fdp->fd_spin); 947 return (EBADF); 948 } 949 holdleaders = 0; 950 if (p->p_fdtol != NULL) { 951 /* 952 * Ask fdfree() to sleep to ensure that all relevant 953 * process leaders can be traversed in closef(). 954 */ 955 fdp->fd_holdleaderscount++; 956 holdleaders = 1; 957 } 958 959 /* 960 * we now hold the fp reference that used to be owned by the descriptor 961 * array. 962 */ 963 spin_unlock(&fdp->fd_spin); 964 if (SLIST_FIRST(&fp->f_klist)) 965 knote_fdclose(fp, fdp, fd); 966 error = closef(fp, p); 967 if (holdleaders) { 968 spin_lock(&fdp->fd_spin); 969 fdp->fd_holdleaderscount--; 970 if (fdp->fd_holdleaderscount == 0 && 971 fdp->fd_holdleaderswakeup != 0) { 972 fdp->fd_holdleaderswakeup = 0; 973 spin_unlock(&fdp->fd_spin); 974 wakeup(&fdp->fd_holdleaderscount); 975 } else { 976 spin_unlock(&fdp->fd_spin); 977 } 978 } 979 return (error); 980 } 981 982 /* 983 * shutdown_args(int fd, int how) 984 */ 985 int 986 kern_shutdown(int fd, int how) 987 { 988 struct thread *td = curthread; 989 struct proc *p = td->td_proc; 990 struct file *fp; 991 int error; 992 993 KKASSERT(p); 994 995 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 996 return (EBADF); 997 error = fo_shutdown(fp, how); 998 fdrop(fp); 999 1000 return (error); 1001 } 1002 1003 /* 1004 * MPALMOSTSAFE 1005 */ 1006 int 1007 sys_shutdown(struct shutdown_args *uap) 1008 { 1009 int error; 1010 1011 error = kern_shutdown(uap->s, uap->how); 1012 1013 return (error); 1014 } 1015 1016 /* 1017 * fstat() helper 1018 */ 1019 int 1020 kern_fstat(int fd, struct stat *ub) 1021 { 1022 struct thread *td = curthread; 1023 struct proc *p = td->td_proc; 1024 struct file *fp; 1025 int error; 1026 1027 KKASSERT(p); 1028 1029 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) 1030 return (EBADF); 1031 error = fo_stat(fp, ub, td->td_ucred); 1032 fdrop(fp); 1033 1034 return (error); 1035 } 1036 1037 /* 1038 * Return status information about a file descriptor. 1039 */ 1040 int 1041 sys_fstat(struct fstat_args *uap) 1042 { 1043 struct stat st; 1044 int error; 1045 1046 error = kern_fstat(uap->fd, &st); 1047 1048 if (error == 0) 1049 error = copyout(&st, uap->sb, sizeof(st)); 1050 return (error); 1051 } 1052 1053 /* 1054 * Return pathconf information about a file descriptor. 1055 * 1056 * MPALMOSTSAFE 1057 */ 1058 int 1059 sys_fpathconf(struct fpathconf_args *uap) 1060 { 1061 struct thread *td = curthread; 1062 struct proc *p = td->td_proc; 1063 struct file *fp; 1064 struct vnode *vp; 1065 int error = 0; 1066 1067 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 1068 return (EBADF); 1069 1070 switch (fp->f_type) { 1071 case DTYPE_PIPE: 1072 case DTYPE_SOCKET: 1073 if (uap->name != _PC_PIPE_BUF) { 1074 error = EINVAL; 1075 } else { 1076 uap->sysmsg_result = PIPE_BUF; 1077 error = 0; 1078 } 1079 break; 1080 case DTYPE_FIFO: 1081 case DTYPE_VNODE: 1082 vp = (struct vnode *)fp->f_data; 1083 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg); 1084 break; 1085 default: 1086 error = EOPNOTSUPP; 1087 break; 1088 } 1089 fdrop(fp); 1090 return(error); 1091 } 1092 1093 static int fdexpand; 1094 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, 1095 "Number of times a file table has been expanded"); 1096 1097 /* 1098 * Grow the file table so it can hold through descriptor (want). 1099 * 1100 * The fdp's spinlock must be held exclusively on entry and may be held 1101 * exclusively on return. The spinlock may be cycled by the routine. 1102 */ 1103 static void 1104 fdgrow_locked(struct filedesc *fdp, int want) 1105 { 1106 struct fdnode *newfiles; 1107 struct fdnode *oldfiles; 1108 int nf, extra; 1109 1110 nf = fdp->fd_nfiles; 1111 do { 1112 /* nf has to be of the form 2^n - 1 */ 1113 nf = 2 * nf + 1; 1114 } while (nf <= want); 1115 1116 spin_unlock(&fdp->fd_spin); 1117 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK); 1118 spin_lock(&fdp->fd_spin); 1119 1120 /* 1121 * We could have raced another extend while we were not holding 1122 * the spinlock. 1123 */ 1124 if (fdp->fd_nfiles >= nf) { 1125 spin_unlock(&fdp->fd_spin); 1126 kfree(newfiles, M_FILEDESC); 1127 spin_lock(&fdp->fd_spin); 1128 return; 1129 } 1130 /* 1131 * Copy the existing ofile and ofileflags arrays 1132 * and zero the new portion of each array. 1133 */ 1134 extra = nf - fdp->fd_nfiles; 1135 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode)); 1136 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode)); 1137 1138 oldfiles = fdp->fd_files; 1139 fdp->fd_files = newfiles; 1140 fdp->fd_nfiles = nf; 1141 1142 if (oldfiles != fdp->fd_builtin_files) { 1143 spin_unlock(&fdp->fd_spin); 1144 kfree(oldfiles, M_FILEDESC); 1145 spin_lock(&fdp->fd_spin); 1146 } 1147 fdexpand++; 1148 } 1149 1150 /* 1151 * Number of nodes in right subtree, including the root. 1152 */ 1153 static __inline int 1154 right_subtree_size(int n) 1155 { 1156 return (n ^ (n | (n + 1))); 1157 } 1158 1159 /* 1160 * Bigger ancestor. 1161 */ 1162 static __inline int 1163 right_ancestor(int n) 1164 { 1165 return (n | (n + 1)); 1166 } 1167 1168 /* 1169 * Smaller ancestor. 1170 */ 1171 static __inline int 1172 left_ancestor(int n) 1173 { 1174 return ((n & (n + 1)) - 1); 1175 } 1176 1177 /* 1178 * Traverse the in-place binary tree buttom-up adjusting the allocation 1179 * count so scans can determine where free descriptors are located. 1180 * 1181 * caller must be holding an exclusive spinlock on fdp 1182 */ 1183 static 1184 void 1185 fdreserve_locked(struct filedesc *fdp, int fd, int incr) 1186 { 1187 while (fd >= 0) { 1188 fdp->fd_files[fd].allocated += incr; 1189 KKASSERT(fdp->fd_files[fd].allocated >= 0); 1190 fd = left_ancestor(fd); 1191 } 1192 } 1193 1194 /* 1195 * Reserve a file descriptor for the process. If no error occurs, the 1196 * caller MUST at some point call fsetfd() or assign a file pointer 1197 * or dispose of the reservation. 1198 */ 1199 int 1200 fdalloc(struct proc *p, int want, int *result) 1201 { 1202 struct filedesc *fdp = p->p_fd; 1203 struct uidinfo *uip; 1204 int fd, rsize, rsum, node, lim; 1205 1206 /* 1207 * Check dtable size limit 1208 */ 1209 *result = -1; /* avoid gcc warnings */ 1210 spin_lock(&p->p_limit->p_spin); 1211 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 1212 lim = INT_MAX; 1213 else 1214 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 1215 spin_unlock(&p->p_limit->p_spin); 1216 1217 if (lim > maxfilesperproc) 1218 lim = maxfilesperproc; 1219 if (lim < minfilesperproc) 1220 lim = minfilesperproc; 1221 if (want >= lim) 1222 return (EMFILE); 1223 1224 /* 1225 * Check that the user has not run out of descriptors (non-root only). 1226 * As a safety measure the dtable is allowed to have at least 1227 * minfilesperproc open fds regardless of the maxfilesperuser limit. 1228 */ 1229 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) { 1230 uip = p->p_ucred->cr_uidinfo; 1231 if (uip->ui_openfiles > maxfilesperuser) { 1232 krateprintf(&krate_uidinfo, 1233 "Warning: user %d pid %d (%s) ran out of " 1234 "file descriptors (%d/%d)\n", 1235 p->p_ucred->cr_uid, (int)p->p_pid, 1236 p->p_comm, 1237 uip->ui_openfiles, maxfilesperuser); 1238 return(ENFILE); 1239 } 1240 } 1241 1242 /* 1243 * Grow the dtable if necessary 1244 */ 1245 spin_lock(&fdp->fd_spin); 1246 if (want >= fdp->fd_nfiles) 1247 fdgrow_locked(fdp, want); 1248 1249 /* 1250 * Search for a free descriptor starting at the higher 1251 * of want or fd_freefile. If that fails, consider 1252 * expanding the ofile array. 1253 * 1254 * NOTE! the 'allocated' field is a cumulative recursive allocation 1255 * count. If we happen to see a value of 0 then we can shortcut 1256 * our search. Otherwise we run through through the tree going 1257 * down branches we know have free descriptor(s) until we hit a 1258 * leaf node. The leaf node will be free but will not necessarily 1259 * have an allocated field of 0. 1260 */ 1261 retry: 1262 /* move up the tree looking for a subtree with a free node */ 1263 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim); 1264 fd = right_ancestor(fd)) { 1265 if (fdp->fd_files[fd].allocated == 0) 1266 goto found; 1267 1268 rsize = right_subtree_size(fd); 1269 if (fdp->fd_files[fd].allocated == rsize) 1270 continue; /* right subtree full */ 1271 1272 /* 1273 * Free fd is in the right subtree of the tree rooted at fd. 1274 * Call that subtree R. Look for the smallest (leftmost) 1275 * subtree of R with an unallocated fd: continue moving 1276 * down the left branch until encountering a full left 1277 * subtree, then move to the right. 1278 */ 1279 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) { 1280 node = fd + rsize; 1281 rsum += fdp->fd_files[node].allocated; 1282 if (fdp->fd_files[fd].allocated == rsum + rsize) { 1283 fd = node; /* move to the right */ 1284 if (fdp->fd_files[node].allocated == 0) 1285 goto found; 1286 rsum = 0; 1287 } 1288 } 1289 goto found; 1290 } 1291 1292 /* 1293 * No space in current array. Expand? 1294 */ 1295 if (fdp->fd_nfiles >= lim) { 1296 spin_unlock(&fdp->fd_spin); 1297 return (EMFILE); 1298 } 1299 fdgrow_locked(fdp, want); 1300 goto retry; 1301 1302 found: 1303 KKASSERT(fd < fdp->fd_nfiles); 1304 if (fd > fdp->fd_lastfile) 1305 fdp->fd_lastfile = fd; 1306 if (want <= fdp->fd_freefile) 1307 fdp->fd_freefile = fd; 1308 *result = fd; 1309 KKASSERT(fdp->fd_files[fd].fp == NULL); 1310 KKASSERT(fdp->fd_files[fd].reserved == 0); 1311 fdp->fd_files[fd].fileflags = 0; 1312 fdp->fd_files[fd].reserved = 1; 1313 fdreserve_locked(fdp, fd, 1); 1314 spin_unlock(&fdp->fd_spin); 1315 return (0); 1316 } 1317 1318 /* 1319 * Check to see whether n user file descriptors 1320 * are available to the process p. 1321 */ 1322 int 1323 fdavail(struct proc *p, int n) 1324 { 1325 struct filedesc *fdp = p->p_fd; 1326 struct fdnode *fdnode; 1327 int i, lim, last; 1328 1329 spin_lock(&p->p_limit->p_spin); 1330 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) 1331 lim = INT_MAX; 1332 else 1333 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; 1334 spin_unlock(&p->p_limit->p_spin); 1335 1336 if (lim > maxfilesperproc) 1337 lim = maxfilesperproc; 1338 if (lim < minfilesperproc) 1339 lim = minfilesperproc; 1340 1341 spin_lock(&fdp->fd_spin); 1342 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) { 1343 spin_unlock(&fdp->fd_spin); 1344 return (1); 1345 } 1346 last = min(fdp->fd_nfiles, lim); 1347 fdnode = &fdp->fd_files[fdp->fd_freefile]; 1348 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) { 1349 if (fdnode->fp == NULL && --n <= 0) { 1350 spin_unlock(&fdp->fd_spin); 1351 return (1); 1352 } 1353 } 1354 spin_unlock(&fdp->fd_spin); 1355 return (0); 1356 } 1357 1358 /* 1359 * Revoke open descriptors referencing (f_data, f_type) 1360 * 1361 * Any revoke executed within a prison is only able to 1362 * revoke descriptors for processes within that prison. 1363 * 1364 * Returns 0 on success or an error code. 1365 */ 1366 struct fdrevoke_info { 1367 void *data; 1368 short type; 1369 short unused; 1370 int found; 1371 struct ucred *cred; 1372 struct file *nfp; 1373 }; 1374 1375 static int fdrevoke_check_callback(struct file *fp, void *vinfo); 1376 static int fdrevoke_proc_callback(struct proc *p, void *vinfo); 1377 1378 int 1379 fdrevoke(void *f_data, short f_type, struct ucred *cred) 1380 { 1381 struct fdrevoke_info info; 1382 int error; 1383 1384 bzero(&info, sizeof(info)); 1385 info.data = f_data; 1386 info.type = f_type; 1387 info.cred = cred; 1388 error = falloc(NULL, &info.nfp, NULL); 1389 if (error) 1390 return (error); 1391 1392 /* 1393 * Scan the file pointer table once. dups do not dup file pointers, 1394 * only descriptors, so there is no leak. Set FREVOKED on the fps 1395 * being revoked. 1396 * 1397 * Any fps sent over unix-domain sockets will be revoked by the 1398 * socket code checking for FREVOKED when the fps are externialized. 1399 * revoke_token is used to make sure that fps marked FREVOKED and 1400 * externalized will be picked up by the following allproc_scan(). 1401 */ 1402 lwkt_gettoken(&revoke_token); 1403 allfiles_scan_exclusive(fdrevoke_check_callback, &info); 1404 lwkt_reltoken(&revoke_token); 1405 1406 /* 1407 * If any fps were marked track down the related descriptors 1408 * and close them. Any dup()s at this point will notice 1409 * the FREVOKED already set in the fp and do the right thing. 1410 */ 1411 if (info.found) 1412 allproc_scan(fdrevoke_proc_callback, &info); 1413 fdrop(info.nfp); 1414 return(0); 1415 } 1416 1417 /* 1418 * Locate matching file pointers directly. 1419 * 1420 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls! 1421 */ 1422 static int 1423 fdrevoke_check_callback(struct file *fp, void *vinfo) 1424 { 1425 struct fdrevoke_info *info = vinfo; 1426 1427 /* 1428 * File pointers already flagged for revokation are skipped. 1429 */ 1430 if (fp->f_flag & FREVOKED) 1431 return(0); 1432 1433 /* 1434 * If revoking from a prison file pointers created outside of 1435 * that prison, or file pointers without creds, cannot be revoked. 1436 */ 1437 if (info->cred->cr_prison && 1438 (fp->f_cred == NULL || 1439 info->cred->cr_prison != fp->f_cred->cr_prison)) { 1440 return(0); 1441 } 1442 1443 /* 1444 * If the file pointer matches then mark it for revocation. The 1445 * flag is currently only used by unp_revoke_gc(). 1446 * 1447 * info->found is a heuristic and can race in a SMP environment. 1448 */ 1449 if (info->data == fp->f_data && info->type == fp->f_type) { 1450 atomic_set_int(&fp->f_flag, FREVOKED); 1451 info->found = 1; 1452 } 1453 return(0); 1454 } 1455 1456 /* 1457 * Locate matching file pointers via process descriptor tables. 1458 */ 1459 static int 1460 fdrevoke_proc_callback(struct proc *p, void *vinfo) 1461 { 1462 struct fdrevoke_info *info = vinfo; 1463 struct filedesc *fdp; 1464 struct file *fp; 1465 int n; 1466 1467 if (p->p_stat == SIDL || p->p_stat == SZOMB) 1468 return(0); 1469 if (info->cred->cr_prison && 1470 info->cred->cr_prison != p->p_ucred->cr_prison) { 1471 return(0); 1472 } 1473 1474 /* 1475 * If the controlling terminal of the process matches the 1476 * vnode being revoked we clear the controlling terminal. 1477 * 1478 * The normal spec_close() may not catch this because it 1479 * uses curproc instead of p. 1480 */ 1481 if (p->p_session && info->type == DTYPE_VNODE && 1482 info->data == p->p_session->s_ttyvp) { 1483 p->p_session->s_ttyvp = NULL; 1484 vrele(info->data); 1485 } 1486 1487 /* 1488 * Softref the fdp to prevent it from being destroyed 1489 */ 1490 spin_lock(&p->p_spin); 1491 if ((fdp = p->p_fd) == NULL) { 1492 spin_unlock(&p->p_spin); 1493 return(0); 1494 } 1495 atomic_add_int(&fdp->fd_softrefs, 1); 1496 spin_unlock(&p->p_spin); 1497 1498 /* 1499 * Locate and close any matching file descriptors. 1500 */ 1501 spin_lock(&fdp->fd_spin); 1502 for (n = 0; n < fdp->fd_nfiles; ++n) { 1503 if ((fp = fdp->fd_files[n].fp) == NULL) 1504 continue; 1505 if (fp->f_flag & FREVOKED) { 1506 fhold(info->nfp); 1507 fdp->fd_files[n].fp = info->nfp; 1508 spin_unlock(&fdp->fd_spin); 1509 knote_fdclose(fp, fdp, n); /* XXX */ 1510 closef(fp, p); 1511 spin_lock(&fdp->fd_spin); 1512 } 1513 } 1514 spin_unlock(&fdp->fd_spin); 1515 atomic_subtract_int(&fdp->fd_softrefs, 1); 1516 return(0); 1517 } 1518 1519 /* 1520 * falloc: 1521 * Create a new open file structure and reserve a file decriptor 1522 * for the process that refers to it. 1523 * 1524 * Root creds are checked using lp, or assumed if lp is NULL. If 1525 * resultfd is non-NULL then lp must also be non-NULL. No file 1526 * descriptor is reserved (and no process context is needed) if 1527 * resultfd is NULL. 1528 * 1529 * A file pointer with a refcount of 1 is returned. Note that the 1530 * file pointer is NOT associated with the descriptor. If falloc 1531 * returns success, fsetfd() MUST be called to either associate the 1532 * file pointer or clear the reservation. 1533 */ 1534 int 1535 falloc(struct lwp *lp, struct file **resultfp, int *resultfd) 1536 { 1537 static struct timeval lastfail; 1538 static int curfail; 1539 struct filelist_head *head; 1540 struct file *fp; 1541 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred; 1542 int error; 1543 1544 fp = NULL; 1545 1546 /* 1547 * Handle filetable full issues and root overfill. 1548 */ 1549 if (nfiles >= maxfiles - maxfilesrootres && 1550 (cred->cr_ruid != 0 || nfiles >= maxfiles)) { 1551 if (ppsratecheck(&lastfail, &curfail, 1)) { 1552 kprintf("kern.maxfiles limit exceeded by uid %d, " 1553 "please see tuning(7).\n", 1554 cred->cr_ruid); 1555 } 1556 error = ENFILE; 1557 goto done; 1558 } 1559 1560 /* 1561 * Allocate a new file descriptor. 1562 */ 1563 fp = objcache_get(file_objcache, M_WAITOK); 1564 bzero(fp, sizeof(*fp)); 1565 spin_init(&fp->f_spin, "falloc"); 1566 SLIST_INIT(&fp->f_klist); 1567 fp->f_count = 1; 1568 fp->f_ops = &badfileops; 1569 fp->f_seqcount = 1; 1570 fsetcred(fp, cred); 1571 atomic_add_int(&nfiles, 1); 1572 1573 head = fp2filelist(fp); 1574 spin_lock(&head->spin); 1575 LIST_INSERT_HEAD(&head->list, fp, f_list); 1576 spin_unlock(&head->spin); 1577 1578 if (resultfd) { 1579 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) { 1580 fdrop(fp); 1581 fp = NULL; 1582 } 1583 } else { 1584 error = 0; 1585 } 1586 done: 1587 *resultfp = fp; 1588 return (error); 1589 } 1590 1591 /* 1592 * Check for races against a file descriptor by determining that the 1593 * file pointer is still associated with the specified file descriptor, 1594 * and a close is not currently in progress. 1595 */ 1596 int 1597 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp) 1598 { 1599 int error; 1600 1601 spin_lock_shared(&fdp->fd_spin); 1602 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp) 1603 error = EBADF; 1604 else 1605 error = 0; 1606 spin_unlock_shared(&fdp->fd_spin); 1607 return (error); 1608 } 1609 1610 /* 1611 * Associate a file pointer with a previously reserved file descriptor. 1612 * This function always succeeds. 1613 * 1614 * If fp is NULL, the file descriptor is returned to the pool. 1615 */ 1616 1617 /* 1618 * (exclusive spinlock must be held on call) 1619 */ 1620 static void 1621 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd) 1622 { 1623 KKASSERT((unsigned)fd < fdp->fd_nfiles); 1624 KKASSERT(fdp->fd_files[fd].reserved != 0); 1625 if (fp) { 1626 fhold(fp); 1627 fdp->fd_files[fd].fp = fp; 1628 fdp->fd_files[fd].reserved = 0; 1629 } else { 1630 fdp->fd_files[fd].reserved = 0; 1631 fdreserve_locked(fdp, fd, -1); 1632 fdfixup_locked(fdp, fd); 1633 } 1634 } 1635 1636 void 1637 fsetfd(struct filedesc *fdp, struct file *fp, int fd) 1638 { 1639 spin_lock(&fdp->fd_spin); 1640 fsetfd_locked(fdp, fp, fd); 1641 spin_unlock(&fdp->fd_spin); 1642 } 1643 1644 /* 1645 * (exclusive spinlock must be held on call) 1646 */ 1647 static 1648 struct file * 1649 funsetfd_locked(struct filedesc *fdp, int fd) 1650 { 1651 struct file *fp; 1652 1653 if ((unsigned)fd >= fdp->fd_nfiles) 1654 return (NULL); 1655 if ((fp = fdp->fd_files[fd].fp) == NULL) 1656 return (NULL); 1657 fdp->fd_files[fd].fp = NULL; 1658 fdp->fd_files[fd].fileflags = 0; 1659 1660 fdreserve_locked(fdp, fd, -1); 1661 fdfixup_locked(fdp, fd); 1662 return(fp); 1663 } 1664 1665 /* 1666 * WARNING: May not be called before initial fsetfd(). 1667 */ 1668 int 1669 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp) 1670 { 1671 int error; 1672 1673 spin_lock(&fdp->fd_spin); 1674 if (((u_int)fd) >= fdp->fd_nfiles) { 1675 error = EBADF; 1676 } else if (fdp->fd_files[fd].fp == NULL) { 1677 error = EBADF; 1678 } else { 1679 *flagsp = fdp->fd_files[fd].fileflags; 1680 error = 0; 1681 } 1682 spin_unlock(&fdp->fd_spin); 1683 return (error); 1684 } 1685 1686 /* 1687 * WARNING: May not be called before initial fsetfd(). 1688 */ 1689 int 1690 fsetfdflags(struct filedesc *fdp, int fd, int add_flags) 1691 { 1692 int error; 1693 1694 spin_lock(&fdp->fd_spin); 1695 if (((u_int)fd) >= fdp->fd_nfiles) { 1696 error = EBADF; 1697 } else if (fdp->fd_files[fd].fp == NULL) { 1698 error = EBADF; 1699 } else { 1700 fdp->fd_files[fd].fileflags |= add_flags; 1701 error = 0; 1702 } 1703 spin_unlock(&fdp->fd_spin); 1704 return (error); 1705 } 1706 1707 /* 1708 * WARNING: May not be called before initial fsetfd(). 1709 */ 1710 int 1711 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags) 1712 { 1713 int error; 1714 1715 spin_lock(&fdp->fd_spin); 1716 if (((u_int)fd) >= fdp->fd_nfiles) { 1717 error = EBADF; 1718 } else if (fdp->fd_files[fd].fp == NULL) { 1719 error = EBADF; 1720 } else { 1721 fdp->fd_files[fd].fileflags &= ~rem_flags; 1722 error = 0; 1723 } 1724 spin_unlock(&fdp->fd_spin); 1725 return (error); 1726 } 1727 1728 /* 1729 * Set/Change/Clear the creds for a fp and synchronize the uidinfo. 1730 */ 1731 void 1732 fsetcred(struct file *fp, struct ucred *ncr) 1733 { 1734 struct ucred *ocr; 1735 struct uidinfo *uip; 1736 1737 ocr = fp->f_cred; 1738 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) { 1739 if (ocr) { 1740 uip = ocr->cr_uidinfo; 1741 atomic_add_int(&uip->ui_openfiles, -1); 1742 } 1743 if (ncr) { 1744 uip = ncr->cr_uidinfo; 1745 atomic_add_int(&uip->ui_openfiles, 1); 1746 } 1747 } 1748 if (ncr) 1749 crhold(ncr); 1750 fp->f_cred = ncr; 1751 if (ocr) 1752 crfree(ocr); 1753 } 1754 1755 /* 1756 * Free a file descriptor. 1757 */ 1758 static 1759 void 1760 ffree(struct file *fp) 1761 { 1762 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); 1763 fsetcred(fp, NULL); 1764 if (fp->f_nchandle.ncp) 1765 cache_drop(&fp->f_nchandle); 1766 objcache_put(file_objcache, fp); 1767 } 1768 1769 /* 1770 * called from init_main, initialize filedesc0 for proc0. 1771 */ 1772 void 1773 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask) 1774 { 1775 p0->p_fd = fdp0; 1776 p0->p_fdtol = NULL; 1777 fdp0->fd_refcnt = 1; 1778 fdp0->fd_cmask = cmask; 1779 fdp0->fd_files = fdp0->fd_builtin_files; 1780 fdp0->fd_nfiles = NDFILE; 1781 fdp0->fd_lastfile = -1; 1782 spin_init(&fdp0->fd_spin, "fdinitbootstrap"); 1783 } 1784 1785 /* 1786 * Build a new filedesc structure. 1787 */ 1788 struct filedesc * 1789 fdinit(struct proc *p) 1790 { 1791 struct filedesc *newfdp; 1792 struct filedesc *fdp = p->p_fd; 1793 1794 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO); 1795 spin_lock(&fdp->fd_spin); 1796 if (fdp->fd_cdir) { 1797 newfdp->fd_cdir = fdp->fd_cdir; 1798 vref(newfdp->fd_cdir); 1799 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1800 } 1801 1802 /* 1803 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of 1804 * proc0, but should unconditionally exist in other processes. 1805 */ 1806 if (fdp->fd_rdir) { 1807 newfdp->fd_rdir = fdp->fd_rdir; 1808 vref(newfdp->fd_rdir); 1809 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1810 } 1811 if (fdp->fd_jdir) { 1812 newfdp->fd_jdir = fdp->fd_jdir; 1813 vref(newfdp->fd_jdir); 1814 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1815 } 1816 spin_unlock(&fdp->fd_spin); 1817 1818 /* Create the file descriptor table. */ 1819 newfdp->fd_refcnt = 1; 1820 newfdp->fd_cmask = cmask; 1821 newfdp->fd_files = newfdp->fd_builtin_files; 1822 newfdp->fd_nfiles = NDFILE; 1823 newfdp->fd_lastfile = -1; 1824 spin_init(&newfdp->fd_spin, "fdinit"); 1825 1826 return (newfdp); 1827 } 1828 1829 /* 1830 * Share a filedesc structure. 1831 */ 1832 struct filedesc * 1833 fdshare(struct proc *p) 1834 { 1835 struct filedesc *fdp; 1836 1837 fdp = p->p_fd; 1838 spin_lock(&fdp->fd_spin); 1839 fdp->fd_refcnt++; 1840 spin_unlock(&fdp->fd_spin); 1841 return (fdp); 1842 } 1843 1844 /* 1845 * Copy a filedesc structure. 1846 */ 1847 int 1848 fdcopy(struct proc *p, struct filedesc **fpp) 1849 { 1850 struct filedesc *fdp = p->p_fd; 1851 struct filedesc *newfdp; 1852 struct fdnode *fdnode; 1853 int i; 1854 int ni; 1855 1856 /* 1857 * Certain daemons might not have file descriptors. 1858 */ 1859 if (fdp == NULL) 1860 return (0); 1861 1862 /* 1863 * Allocate the new filedesc and fd_files[] array. This can race 1864 * with operations by other threads on the fdp so we have to be 1865 * careful. 1866 */ 1867 newfdp = kmalloc(sizeof(struct filedesc), 1868 M_FILEDESC, M_WAITOK | M_ZERO | M_NULLOK); 1869 if (newfdp == NULL) { 1870 *fpp = NULL; 1871 return (-1); 1872 } 1873 again: 1874 spin_lock(&fdp->fd_spin); 1875 if (fdp->fd_lastfile < NDFILE) { 1876 newfdp->fd_files = newfdp->fd_builtin_files; 1877 i = NDFILE; 1878 } else { 1879 /* 1880 * We have to allocate (N^2-1) entries for our in-place 1881 * binary tree. Allow the table to shrink. 1882 */ 1883 i = fdp->fd_nfiles; 1884 ni = (i - 1) / 2; 1885 while (ni > fdp->fd_lastfile && ni > NDFILE) { 1886 i = ni; 1887 ni = (i - 1) / 2; 1888 } 1889 spin_unlock(&fdp->fd_spin); 1890 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode), 1891 M_FILEDESC, M_WAITOK | M_ZERO); 1892 1893 /* 1894 * Check for race, retry 1895 */ 1896 spin_lock(&fdp->fd_spin); 1897 if (i <= fdp->fd_lastfile) { 1898 spin_unlock(&fdp->fd_spin); 1899 kfree(newfdp->fd_files, M_FILEDESC); 1900 goto again; 1901 } 1902 } 1903 1904 /* 1905 * Dup the remaining fields. vref() and cache_hold() can be 1906 * safely called while holding the read spinlock on fdp. 1907 * 1908 * The read spinlock on fdp is still being held. 1909 * 1910 * NOTE: vref and cache_hold calls for the case where the vnode 1911 * or cache entry already has at least one ref may be called 1912 * while holding spin locks. 1913 */ 1914 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) { 1915 vref(newfdp->fd_cdir); 1916 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); 1917 } 1918 /* 1919 * We must check for fd_rdir here, at least for now because 1920 * the init process is created before we have access to the 1921 * rootvode to take a reference to it. 1922 */ 1923 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) { 1924 vref(newfdp->fd_rdir); 1925 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); 1926 } 1927 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) { 1928 vref(newfdp->fd_jdir); 1929 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); 1930 } 1931 newfdp->fd_refcnt = 1; 1932 newfdp->fd_nfiles = i; 1933 newfdp->fd_lastfile = fdp->fd_lastfile; 1934 newfdp->fd_freefile = fdp->fd_freefile; 1935 newfdp->fd_cmask = fdp->fd_cmask; 1936 spin_init(&newfdp->fd_spin, "fdcopy"); 1937 1938 /* 1939 * Copy the descriptor table through (i). This also copies the 1940 * allocation state. Then go through and ref the file pointers 1941 * and clean up any KQ descriptors. 1942 * 1943 * kq descriptors cannot be copied. Since we haven't ref'd the 1944 * copied files yet we can ignore the return value from funsetfd(). 1945 * 1946 * The read spinlock on fdp is still being held. 1947 */ 1948 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode)); 1949 for (i = 0 ; i < newfdp->fd_nfiles; ++i) { 1950 fdnode = &newfdp->fd_files[i]; 1951 if (fdnode->reserved) { 1952 fdreserve_locked(newfdp, i, -1); 1953 fdnode->reserved = 0; 1954 fdfixup_locked(newfdp, i); 1955 } else if (fdnode->fp) { 1956 if (fdnode->fp->f_type == DTYPE_KQUEUE) { 1957 (void)funsetfd_locked(newfdp, i); 1958 } else { 1959 fhold(fdnode->fp); 1960 } 1961 } 1962 } 1963 spin_unlock(&fdp->fd_spin); 1964 *fpp = newfdp; 1965 return (0); 1966 } 1967 1968 /* 1969 * Release a filedesc structure. 1970 * 1971 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE) 1972 */ 1973 void 1974 fdfree(struct proc *p, struct filedesc *repl) 1975 { 1976 struct filedesc *fdp; 1977 struct fdnode *fdnode; 1978 int i; 1979 struct filedesc_to_leader *fdtol; 1980 struct file *fp; 1981 struct vnode *vp; 1982 struct flock lf; 1983 1984 /* 1985 * Certain daemons might not have file descriptors. 1986 */ 1987 fdp = p->p_fd; 1988 if (fdp == NULL) { 1989 p->p_fd = repl; 1990 return; 1991 } 1992 1993 /* 1994 * Severe messing around to follow. 1995 */ 1996 spin_lock(&fdp->fd_spin); 1997 1998 /* Check for special need to clear POSIX style locks */ 1999 fdtol = p->p_fdtol; 2000 if (fdtol != NULL) { 2001 KASSERT(fdtol->fdl_refcount > 0, 2002 ("filedesc_to_refcount botch: fdl_refcount=%d", 2003 fdtol->fdl_refcount)); 2004 if (fdtol->fdl_refcount == 1 && 2005 (p->p_leader->p_flags & P_ADVLOCK) != 0) { 2006 for (i = 0; i <= fdp->fd_lastfile; ++i) { 2007 fdnode = &fdp->fd_files[i]; 2008 if (fdnode->fp == NULL || 2009 fdnode->fp->f_type != DTYPE_VNODE) { 2010 continue; 2011 } 2012 fp = fdnode->fp; 2013 fhold(fp); 2014 spin_unlock(&fdp->fd_spin); 2015 2016 lf.l_whence = SEEK_SET; 2017 lf.l_start = 0; 2018 lf.l_len = 0; 2019 lf.l_type = F_UNLCK; 2020 vp = (struct vnode *)fp->f_data; 2021 (void) VOP_ADVLOCK(vp, 2022 (caddr_t)p->p_leader, 2023 F_UNLCK, 2024 &lf, 2025 F_POSIX); 2026 fdrop(fp); 2027 spin_lock(&fdp->fd_spin); 2028 } 2029 } 2030 retry: 2031 if (fdtol->fdl_refcount == 1) { 2032 if (fdp->fd_holdleaderscount > 0 && 2033 (p->p_leader->p_flags & P_ADVLOCK) != 0) { 2034 /* 2035 * close() or do_dup() has cleared a reference 2036 * in a shared file descriptor table. 2037 */ 2038 fdp->fd_holdleaderswakeup = 1; 2039 ssleep(&fdp->fd_holdleaderscount, 2040 &fdp->fd_spin, 0, "fdlhold", 0); 2041 goto retry; 2042 } 2043 if (fdtol->fdl_holdcount > 0) { 2044 /* 2045 * Ensure that fdtol->fdl_leader 2046 * remains valid in closef(). 2047 */ 2048 fdtol->fdl_wakeup = 1; 2049 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0); 2050 goto retry; 2051 } 2052 } 2053 fdtol->fdl_refcount--; 2054 if (fdtol->fdl_refcount == 0 && 2055 fdtol->fdl_holdcount == 0) { 2056 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2057 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2058 } else { 2059 fdtol = NULL; 2060 } 2061 p->p_fdtol = NULL; 2062 if (fdtol != NULL) { 2063 spin_unlock(&fdp->fd_spin); 2064 kfree(fdtol, M_FILEDESC_TO_LEADER); 2065 spin_lock(&fdp->fd_spin); 2066 } 2067 } 2068 if (--fdp->fd_refcnt > 0) { 2069 spin_unlock(&fdp->fd_spin); 2070 spin_lock(&p->p_spin); 2071 p->p_fd = repl; 2072 spin_unlock(&p->p_spin); 2073 return; 2074 } 2075 2076 /* 2077 * Even though we are the last reference to the structure allproc 2078 * scans may still reference the structure. Maintain proper 2079 * locks until we can replace p->p_fd. 2080 * 2081 * Also note that kqueue's closef still needs to reference the 2082 * fdp via p->p_fd, so we have to close the descriptors before 2083 * we replace p->p_fd. 2084 */ 2085 for (i = 0; i <= fdp->fd_lastfile; ++i) { 2086 if (fdp->fd_files[i].fp) { 2087 fp = funsetfd_locked(fdp, i); 2088 if (fp) { 2089 spin_unlock(&fdp->fd_spin); 2090 if (SLIST_FIRST(&fp->f_klist)) 2091 knote_fdclose(fp, fdp, i); 2092 closef(fp, p); 2093 spin_lock(&fdp->fd_spin); 2094 } 2095 } 2096 } 2097 spin_unlock(&fdp->fd_spin); 2098 2099 /* 2100 * Interlock against an allproc scan operations (typically frevoke). 2101 */ 2102 spin_lock(&p->p_spin); 2103 p->p_fd = repl; 2104 spin_unlock(&p->p_spin); 2105 2106 /* 2107 * Wait for any softrefs to go away. This race rarely occurs so 2108 * we can use a non-critical-path style poll/sleep loop. The 2109 * race only occurs against allproc scans. 2110 * 2111 * No new softrefs can occur with the fdp disconnected from the 2112 * process. 2113 */ 2114 if (fdp->fd_softrefs) { 2115 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid); 2116 while (fdp->fd_softrefs) 2117 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1); 2118 } 2119 2120 if (fdp->fd_files != fdp->fd_builtin_files) 2121 kfree(fdp->fd_files, M_FILEDESC); 2122 if (fdp->fd_cdir) { 2123 cache_drop(&fdp->fd_ncdir); 2124 vrele(fdp->fd_cdir); 2125 } 2126 if (fdp->fd_rdir) { 2127 cache_drop(&fdp->fd_nrdir); 2128 vrele(fdp->fd_rdir); 2129 } 2130 if (fdp->fd_jdir) { 2131 cache_drop(&fdp->fd_njdir); 2132 vrele(fdp->fd_jdir); 2133 } 2134 kfree(fdp, M_FILEDESC); 2135 } 2136 2137 /* 2138 * Retrieve and reference the file pointer associated with a descriptor. 2139 */ 2140 struct file * 2141 holdfp(struct filedesc *fdp, int fd, int flag) 2142 { 2143 struct file* fp; 2144 2145 spin_lock_shared(&fdp->fd_spin); 2146 if (((u_int)fd) >= fdp->fd_nfiles) { 2147 fp = NULL; 2148 goto done; 2149 } 2150 if ((fp = fdp->fd_files[fd].fp) == NULL) 2151 goto done; 2152 if ((fp->f_flag & flag) == 0 && flag != -1) { 2153 fp = NULL; 2154 goto done; 2155 } 2156 fhold(fp); 2157 done: 2158 spin_unlock_shared(&fdp->fd_spin); 2159 return (fp); 2160 } 2161 2162 /* 2163 * holdsock() - load the struct file pointer associated 2164 * with a socket into *fpp. If an error occurs, non-zero 2165 * will be returned and *fpp will be set to NULL. 2166 */ 2167 int 2168 holdsock(struct filedesc *fdp, int fd, struct file **fpp) 2169 { 2170 struct file *fp; 2171 int error; 2172 2173 spin_lock_shared(&fdp->fd_spin); 2174 if ((unsigned)fd >= fdp->fd_nfiles) { 2175 error = EBADF; 2176 fp = NULL; 2177 goto done; 2178 } 2179 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2180 error = EBADF; 2181 goto done; 2182 } 2183 if (fp->f_type != DTYPE_SOCKET) { 2184 error = ENOTSOCK; 2185 goto done; 2186 } 2187 fhold(fp); 2188 error = 0; 2189 done: 2190 spin_unlock_shared(&fdp->fd_spin); 2191 *fpp = fp; 2192 return (error); 2193 } 2194 2195 /* 2196 * Convert a user file descriptor to a held file pointer. 2197 */ 2198 int 2199 holdvnode(struct filedesc *fdp, int fd, struct file **fpp) 2200 { 2201 struct file *fp; 2202 int error; 2203 2204 spin_lock_shared(&fdp->fd_spin); 2205 if ((unsigned)fd >= fdp->fd_nfiles) { 2206 error = EBADF; 2207 fp = NULL; 2208 goto done; 2209 } 2210 if ((fp = fdp->fd_files[fd].fp) == NULL) { 2211 error = EBADF; 2212 goto done; 2213 } 2214 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 2215 fp = NULL; 2216 error = EINVAL; 2217 goto done; 2218 } 2219 fhold(fp); 2220 error = 0; 2221 done: 2222 spin_unlock_shared(&fdp->fd_spin); 2223 *fpp = fp; 2224 return (error); 2225 } 2226 2227 /* 2228 * For setugid programs, we don't want to people to use that setugidness 2229 * to generate error messages which write to a file which otherwise would 2230 * otherwise be off-limits to the process. 2231 * 2232 * This is a gross hack to plug the hole. A better solution would involve 2233 * a special vop or other form of generalized access control mechanism. We 2234 * go ahead and just reject all procfs file systems accesses as dangerous. 2235 * 2236 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2237 * sufficient. We also don't for check setugidness since we know we are. 2238 */ 2239 static int 2240 is_unsafe(struct file *fp) 2241 { 2242 if (fp->f_type == DTYPE_VNODE && 2243 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) 2244 return (1); 2245 return (0); 2246 } 2247 2248 /* 2249 * Make this setguid thing safe, if at all possible. 2250 * 2251 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2252 */ 2253 void 2254 setugidsafety(struct proc *p) 2255 { 2256 struct filedesc *fdp = p->p_fd; 2257 int i; 2258 2259 /* Certain daemons might not have file descriptors. */ 2260 if (fdp == NULL) 2261 return; 2262 2263 /* 2264 * note: fdp->fd_files may be reallocated out from under us while 2265 * we are blocked in a close. Be careful! 2266 */ 2267 for (i = 0; i <= fdp->fd_lastfile; i++) { 2268 if (i > 2) 2269 break; 2270 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) { 2271 struct file *fp; 2272 2273 /* 2274 * NULL-out descriptor prior to close to avoid 2275 * a race while close blocks. 2276 */ 2277 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2278 knote_fdclose(fp, fdp, i); 2279 closef(fp, p); 2280 } 2281 } 2282 } 2283 } 2284 2285 /* 2286 * Close any files on exec? 2287 * 2288 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() 2289 */ 2290 void 2291 fdcloseexec(struct proc *p) 2292 { 2293 struct filedesc *fdp = p->p_fd; 2294 int i; 2295 2296 /* Certain daemons might not have file descriptors. */ 2297 if (fdp == NULL) 2298 return; 2299 2300 /* 2301 * We cannot cache fd_files since operations may block and rip 2302 * them out from under us. 2303 */ 2304 for (i = 0; i <= fdp->fd_lastfile; i++) { 2305 if (fdp->fd_files[i].fp != NULL && 2306 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) { 2307 struct file *fp; 2308 2309 /* 2310 * NULL-out descriptor prior to close to avoid 2311 * a race while close blocks. 2312 */ 2313 if ((fp = funsetfd_locked(fdp, i)) != NULL) { 2314 knote_fdclose(fp, fdp, i); 2315 closef(fp, p); 2316 } 2317 } 2318 } 2319 } 2320 2321 /* 2322 * It is unsafe for set[ug]id processes to be started with file 2323 * descriptors 0..2 closed, as these descriptors are given implicit 2324 * significance in the Standard C library. fdcheckstd() will create a 2325 * descriptor referencing /dev/null for each of stdin, stdout, and 2326 * stderr that is not already open. 2327 * 2328 * NOT MPSAFE - calls falloc, vn_open, etc 2329 */ 2330 int 2331 fdcheckstd(struct lwp *lp) 2332 { 2333 struct nlookupdata nd; 2334 struct filedesc *fdp; 2335 struct file *fp; 2336 int retval; 2337 int i, error, flags, devnull; 2338 2339 fdp = lp->lwp_proc->p_fd; 2340 if (fdp == NULL) 2341 return (0); 2342 devnull = -1; 2343 error = 0; 2344 for (i = 0; i < 3; i++) { 2345 if (fdp->fd_files[i].fp != NULL) 2346 continue; 2347 if (devnull < 0) { 2348 if ((error = falloc(lp, &fp, &devnull)) != 0) 2349 break; 2350 2351 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, 2352 NLC_FOLLOW|NLC_LOCKVP); 2353 flags = FREAD | FWRITE; 2354 if (error == 0) 2355 error = vn_open(&nd, fp, flags, 0); 2356 if (error == 0) 2357 fsetfd(fdp, fp, devnull); 2358 else 2359 fsetfd(fdp, NULL, devnull); 2360 fdrop(fp); 2361 nlookup_done(&nd); 2362 if (error) 2363 break; 2364 KKASSERT(i == devnull); 2365 } else { 2366 error = kern_dup(DUP_FIXED, devnull, i, &retval); 2367 if (error != 0) 2368 break; 2369 } 2370 } 2371 return (error); 2372 } 2373 2374 /* 2375 * Internal form of close. 2376 * Decrement reference count on file structure. 2377 * Note: td and/or p may be NULL when closing a file 2378 * that was being passed in a message. 2379 * 2380 * MPALMOSTSAFE - acquires mplock for VOP operations 2381 */ 2382 int 2383 closef(struct file *fp, struct proc *p) 2384 { 2385 struct vnode *vp; 2386 struct flock lf; 2387 struct filedesc_to_leader *fdtol; 2388 2389 if (fp == NULL) 2390 return (0); 2391 2392 /* 2393 * POSIX record locking dictates that any close releases ALL 2394 * locks owned by this process. This is handled by setting 2395 * a flag in the unlock to free ONLY locks obeying POSIX 2396 * semantics, and not to free BSD-style file locks. 2397 * If the descriptor was in a message, POSIX-style locks 2398 * aren't passed with the descriptor. 2399 */ 2400 if (p != NULL && fp->f_type == DTYPE_VNODE && 2401 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2402 ) { 2403 if ((p->p_leader->p_flags & P_ADVLOCK) != 0) { 2404 lf.l_whence = SEEK_SET; 2405 lf.l_start = 0; 2406 lf.l_len = 0; 2407 lf.l_type = F_UNLCK; 2408 vp = (struct vnode *)fp->f_data; 2409 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 2410 &lf, F_POSIX); 2411 } 2412 fdtol = p->p_fdtol; 2413 if (fdtol != NULL) { 2414 lwkt_gettoken(&p->p_token); 2415 /* 2416 * Handle special case where file descriptor table 2417 * is shared between multiple process leaders. 2418 */ 2419 for (fdtol = fdtol->fdl_next; 2420 fdtol != p->p_fdtol; 2421 fdtol = fdtol->fdl_next) { 2422 if ((fdtol->fdl_leader->p_flags & 2423 P_ADVLOCK) == 0) 2424 continue; 2425 fdtol->fdl_holdcount++; 2426 lf.l_whence = SEEK_SET; 2427 lf.l_start = 0; 2428 lf.l_len = 0; 2429 lf.l_type = F_UNLCK; 2430 vp = (struct vnode *)fp->f_data; 2431 (void) VOP_ADVLOCK(vp, 2432 (caddr_t)fdtol->fdl_leader, 2433 F_UNLCK, &lf, F_POSIX); 2434 fdtol->fdl_holdcount--; 2435 if (fdtol->fdl_holdcount == 0 && 2436 fdtol->fdl_wakeup != 0) { 2437 fdtol->fdl_wakeup = 0; 2438 wakeup(fdtol); 2439 } 2440 } 2441 lwkt_reltoken(&p->p_token); 2442 } 2443 } 2444 return (fdrop(fp)); 2445 } 2446 2447 /* 2448 * fhold() can only be called if f_count is already at least 1 (i.e. the 2449 * caller of fhold() already has a reference to the file pointer in some 2450 * manner or other). 2451 * 2452 * Atomic ops are used for incrementing and decrementing f_count before 2453 * the 1->0 transition. f_count 1->0 transition is special, see the 2454 * comment in fdrop(). 2455 */ 2456 void 2457 fhold(struct file *fp) 2458 { 2459 /* 0->1 transition will never work */ 2460 KASSERT(fp->f_count > 0, ("fhold: invalid f_count %d", fp->f_count)); 2461 atomic_add_int(&fp->f_count, 1); 2462 } 2463 2464 /* 2465 * fdrop() - drop a reference to a descriptor 2466 */ 2467 int 2468 fdrop(struct file *fp) 2469 { 2470 struct flock lf; 2471 struct vnode *vp; 2472 int error, do_free = 0; 2473 2474 /* 2475 * NOTE: 2476 * Simple atomic_fetchadd_int(f_count, -1) here will cause use- 2477 * after-free or double free (due to f_count 0->1 transition), if 2478 * fhold() is called on the fps found through filehead iteration. 2479 */ 2480 for (;;) { 2481 int count = fp->f_count; 2482 2483 cpu_ccfence(); 2484 KASSERT(count > 0, ("fdrop: invalid f_count %d", count)); 2485 if (count == 1) { 2486 struct filelist_head *head = fp2filelist(fp); 2487 2488 /* 2489 * About to drop the last reference, hold the 2490 * filehead spin lock and drop it, so that no 2491 * one could see this fp through filehead anymore, 2492 * let alone fhold() this fp. 2493 */ 2494 spin_lock(&head->spin); 2495 if (atomic_cmpset_int(&fp->f_count, count, 0)) { 2496 LIST_REMOVE(fp, f_list); 2497 spin_unlock(&head->spin); 2498 atomic_subtract_int(&nfiles, 1); 2499 do_free = 1; /* free this fp */ 2500 break; 2501 } 2502 spin_unlock(&head->spin); 2503 /* retry */ 2504 } else if (atomic_cmpset_int(&fp->f_count, count, count - 1)) { 2505 break; 2506 } 2507 /* retry */ 2508 } 2509 if (!do_free) 2510 return (0); 2511 2512 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL); 2513 2514 /* 2515 * The last reference has gone away, we own the fp structure free 2516 * and clear. 2517 */ 2518 if (fp->f_count < 0) 2519 panic("fdrop: count < 0"); 2520 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE && 2521 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) 2522 ) { 2523 lf.l_whence = SEEK_SET; 2524 lf.l_start = 0; 2525 lf.l_len = 0; 2526 lf.l_type = F_UNLCK; 2527 vp = (struct vnode *)fp->f_data; 2528 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2529 } 2530 if (fp->f_ops != &badfileops) 2531 error = fo_close(fp); 2532 else 2533 error = 0; 2534 ffree(fp); 2535 return (error); 2536 } 2537 2538 /* 2539 * Apply an advisory lock on a file descriptor. 2540 * 2541 * Just attempt to get a record lock of the requested type on 2542 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2543 * 2544 * MPALMOSTSAFE 2545 */ 2546 int 2547 sys_flock(struct flock_args *uap) 2548 { 2549 struct proc *p = curproc; 2550 struct file *fp; 2551 struct vnode *vp; 2552 struct flock lf; 2553 int error; 2554 2555 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) 2556 return (EBADF); 2557 if (fp->f_type != DTYPE_VNODE) { 2558 error = EOPNOTSUPP; 2559 goto done; 2560 } 2561 vp = (struct vnode *)fp->f_data; 2562 lf.l_whence = SEEK_SET; 2563 lf.l_start = 0; 2564 lf.l_len = 0; 2565 if (uap->how & LOCK_UN) { 2566 lf.l_type = F_UNLCK; 2567 atomic_clear_int(&fp->f_flag, FHASLOCK); /* race ok */ 2568 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); 2569 goto done; 2570 } 2571 if (uap->how & LOCK_EX) 2572 lf.l_type = F_WRLCK; 2573 else if (uap->how & LOCK_SH) 2574 lf.l_type = F_RDLCK; 2575 else { 2576 error = EBADF; 2577 goto done; 2578 } 2579 if (uap->how & LOCK_NB) 2580 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0); 2581 else 2582 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT); 2583 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */ 2584 done: 2585 fdrop(fp); 2586 return (error); 2587 } 2588 2589 /* 2590 * File Descriptor pseudo-device driver (/dev/fd/). 2591 * 2592 * Opening minor device N dup()s the file (if any) connected to file 2593 * descriptor N belonging to the calling process. Note that this driver 2594 * consists of only the ``open()'' routine, because all subsequent 2595 * references to this file will be direct to the other driver. 2596 */ 2597 static int 2598 fdopen(struct dev_open_args *ap) 2599 { 2600 thread_t td = curthread; 2601 2602 KKASSERT(td->td_lwp != NULL); 2603 2604 /* 2605 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the 2606 * the file descriptor being sought for duplication. The error 2607 * return ensures that the vnode for this device will be released 2608 * by vn_open. Open will detect this special error and take the 2609 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2610 * will simply report the error. 2611 */ 2612 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev); 2613 return (ENODEV); 2614 } 2615 2616 /* 2617 * The caller has reserved the file descriptor dfd for us. On success we 2618 * must fsetfd() it. On failure the caller will clean it up. 2619 */ 2620 int 2621 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error) 2622 { 2623 struct file *wfp; 2624 struct file *xfp; 2625 int werror; 2626 2627 if ((wfp = holdfp(fdp, sfd, -1)) == NULL) 2628 return (EBADF); 2629 2630 /* 2631 * Close a revoke/dup race. Duping a descriptor marked as revoked 2632 * will dup a dummy descriptor instead of the real one. 2633 */ 2634 if (wfp->f_flag & FREVOKED) { 2635 kprintf("Warning: attempt to dup() a revoked descriptor\n"); 2636 fdrop(wfp); 2637 wfp = NULL; 2638 werror = falloc(NULL, &wfp, NULL); 2639 if (werror) 2640 return (werror); 2641 } 2642 2643 /* 2644 * There are two cases of interest here. 2645 * 2646 * For ENODEV simply dup sfd to file descriptor dfd and return. 2647 * 2648 * For ENXIO steal away the file structure from sfd and store it 2649 * dfd. sfd is effectively closed by this operation. 2650 * 2651 * Any other error code is just returned. 2652 */ 2653 switch (error) { 2654 case ENODEV: 2655 /* 2656 * Check that the mode the file is being opened for is a 2657 * subset of the mode of the existing descriptor. 2658 */ 2659 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2660 error = EACCES; 2661 break; 2662 } 2663 spin_lock(&fdp->fd_spin); 2664 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2665 fsetfd_locked(fdp, wfp, dfd); 2666 spin_unlock(&fdp->fd_spin); 2667 error = 0; 2668 break; 2669 case ENXIO: 2670 /* 2671 * Steal away the file pointer from dfd, and stuff it into indx. 2672 */ 2673 spin_lock(&fdp->fd_spin); 2674 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; 2675 fsetfd(fdp, wfp, dfd); 2676 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) { 2677 spin_unlock(&fdp->fd_spin); 2678 fdrop(xfp); 2679 } else { 2680 spin_unlock(&fdp->fd_spin); 2681 } 2682 error = 0; 2683 break; 2684 default: 2685 break; 2686 } 2687 fdrop(wfp); 2688 return (error); 2689 } 2690 2691 /* 2692 * NOT MPSAFE - I think these refer to a common file descriptor table 2693 * and we need to spinlock that to link fdtol in. 2694 */ 2695 struct filedesc_to_leader * 2696 filedesc_to_leader_alloc(struct filedesc_to_leader *old, 2697 struct proc *leader) 2698 { 2699 struct filedesc_to_leader *fdtol; 2700 2701 fdtol = kmalloc(sizeof(struct filedesc_to_leader), 2702 M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO); 2703 fdtol->fdl_refcount = 1; 2704 fdtol->fdl_holdcount = 0; 2705 fdtol->fdl_wakeup = 0; 2706 fdtol->fdl_leader = leader; 2707 if (old != NULL) { 2708 fdtol->fdl_next = old->fdl_next; 2709 fdtol->fdl_prev = old; 2710 old->fdl_next = fdtol; 2711 fdtol->fdl_next->fdl_prev = fdtol; 2712 } else { 2713 fdtol->fdl_next = fdtol; 2714 fdtol->fdl_prev = fdtol; 2715 } 2716 return fdtol; 2717 } 2718 2719 /* 2720 * Scan all file pointers in the system. The callback is made with 2721 * the master list spinlock held exclusively. 2722 */ 2723 void 2724 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data) 2725 { 2726 int i; 2727 2728 for (i = 0; i < NFILELIST_HEADS; ++i) { 2729 struct filelist_head *head = &filelist_heads[i]; 2730 struct file *fp; 2731 2732 spin_lock(&head->spin); 2733 LIST_FOREACH(fp, &head->list, f_list) { 2734 int res; 2735 2736 res = callback(fp, data); 2737 if (res < 0) 2738 break; 2739 } 2740 spin_unlock(&head->spin); 2741 } 2742 } 2743 2744 /* 2745 * Get file structures. 2746 * 2747 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe) 2748 */ 2749 2750 struct sysctl_kern_file_info { 2751 int count; 2752 int error; 2753 struct sysctl_req *req; 2754 }; 2755 2756 static int sysctl_kern_file_callback(struct proc *p, void *data); 2757 2758 static int 2759 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2760 { 2761 struct sysctl_kern_file_info info; 2762 2763 /* 2764 * Note: because the number of file descriptors is calculated 2765 * in different ways for sizing vs returning the data, 2766 * there is information leakage from the first loop. However, 2767 * it is of a similar order of magnitude to the leakage from 2768 * global system statistics such as kern.openfiles. 2769 * 2770 * When just doing a count, note that we cannot just count 2771 * the elements and add f_count via the filehead list because 2772 * threaded processes share their descriptor table and f_count might 2773 * still be '1' in that case. 2774 * 2775 * Since the SYSCTL op can block, we must hold the process to 2776 * prevent it being ripped out from under us either in the 2777 * file descriptor loop or in the greater LIST_FOREACH. The 2778 * process may be in varying states of disrepair. If the process 2779 * is in SZOMB we may have caught it just as it is being removed 2780 * from the allproc list, we must skip it in that case to maintain 2781 * an unbroken chain through the allproc list. 2782 */ 2783 info.count = 0; 2784 info.error = 0; 2785 info.req = req; 2786 allproc_scan(sysctl_kern_file_callback, &info); 2787 2788 /* 2789 * When just calculating the size, overestimate a bit to try to 2790 * prevent system activity from causing the buffer-fill call 2791 * to fail later on. 2792 */ 2793 if (req->oldptr == NULL) { 2794 info.count = (info.count + 16) + (info.count / 10); 2795 info.error = SYSCTL_OUT(req, NULL, 2796 info.count * sizeof(struct kinfo_file)); 2797 } 2798 return (info.error); 2799 } 2800 2801 static int 2802 sysctl_kern_file_callback(struct proc *p, void *data) 2803 { 2804 struct sysctl_kern_file_info *info = data; 2805 struct kinfo_file kf; 2806 struct filedesc *fdp; 2807 struct file *fp; 2808 uid_t uid; 2809 int n; 2810 2811 if (p->p_stat == SIDL || p->p_stat == SZOMB) 2812 return(0); 2813 if (!(PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)) 2814 return(0); 2815 2816 /* 2817 * Softref the fdp to prevent it from being destroyed 2818 */ 2819 spin_lock(&p->p_spin); 2820 if ((fdp = p->p_fd) == NULL) { 2821 spin_unlock(&p->p_spin); 2822 return(0); 2823 } 2824 atomic_add_int(&fdp->fd_softrefs, 1); 2825 spin_unlock(&p->p_spin); 2826 2827 /* 2828 * The fdp's own spinlock prevents the contents from being 2829 * modified. 2830 */ 2831 spin_lock_shared(&fdp->fd_spin); 2832 for (n = 0; n < fdp->fd_nfiles; ++n) { 2833 if ((fp = fdp->fd_files[n].fp) == NULL) 2834 continue; 2835 if (info->req->oldptr == NULL) { 2836 ++info->count; 2837 } else { 2838 uid = p->p_ucred ? p->p_ucred->cr_uid : -1; 2839 kcore_make_file(&kf, fp, p->p_pid, uid, n); 2840 spin_unlock_shared(&fdp->fd_spin); 2841 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf)); 2842 spin_lock_shared(&fdp->fd_spin); 2843 if (info->error) 2844 break; 2845 } 2846 } 2847 spin_unlock_shared(&fdp->fd_spin); 2848 atomic_subtract_int(&fdp->fd_softrefs, 1); 2849 if (info->error) 2850 return(-1); 2851 return(0); 2852 } 2853 2854 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2855 0, 0, sysctl_kern_file, "S,file", "Entire file table"); 2856 2857 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW, 2858 &minfilesperproc, 0, "Minimum files allowed open per process"); 2859 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2860 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2861 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW, 2862 &maxfilesperuser, 0, "Maximum files allowed open per user"); 2863 2864 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2865 &maxfiles, 0, "Maximum number of files"); 2866 2867 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, 2868 &maxfilesrootres, 0, "Descriptors reserved for root use"); 2869 2870 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2871 &nfiles, 0, "System-wide number of open files"); 2872 2873 static void 2874 fildesc_drvinit(void *unused) 2875 { 2876 int fd; 2877 2878 for (fd = 0; fd < NUMFDESC; fd++) { 2879 make_dev(&fildesc_ops, fd, 2880 UID_BIN, GID_BIN, 0666, "fd/%d", fd); 2881 } 2882 2883 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); 2884 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); 2885 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); 2886 } 2887 2888 struct fileops badfileops = { 2889 .fo_read = badfo_readwrite, 2890 .fo_write = badfo_readwrite, 2891 .fo_ioctl = badfo_ioctl, 2892 .fo_kqfilter = badfo_kqfilter, 2893 .fo_stat = badfo_stat, 2894 .fo_close = badfo_close, 2895 .fo_shutdown = badfo_shutdown 2896 }; 2897 2898 int 2899 badfo_readwrite( 2900 struct file *fp, 2901 struct uio *uio, 2902 struct ucred *cred, 2903 int flags 2904 ) { 2905 return (EBADF); 2906 } 2907 2908 int 2909 badfo_ioctl(struct file *fp, u_long com, caddr_t data, 2910 struct ucred *cred, struct sysmsg *msgv) 2911 { 2912 return (EBADF); 2913 } 2914 2915 /* 2916 * Must return an error to prevent registration, typically 2917 * due to a revoked descriptor (file_filtops assigned). 2918 */ 2919 int 2920 badfo_kqfilter(struct file *fp, struct knote *kn) 2921 { 2922 return (EOPNOTSUPP); 2923 } 2924 2925 int 2926 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred) 2927 { 2928 return (EBADF); 2929 } 2930 2931 int 2932 badfo_close(struct file *fp) 2933 { 2934 return (EBADF); 2935 } 2936 2937 int 2938 badfo_shutdown(struct file *fp, int how) 2939 { 2940 return (EBADF); 2941 } 2942 2943 int 2944 nofo_shutdown(struct file *fp, int how) 2945 { 2946 return (EOPNOTSUPP); 2947 } 2948 2949 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE + CDEV_MAJOR, 2950 fildesc_drvinit,NULL); 2951 2952 static void 2953 filelist_heads_init(void *arg __unused) 2954 { 2955 int i; 2956 2957 for (i = 0; i < NFILELIST_HEADS; ++i) { 2958 struct filelist_head *head = &filelist_heads[i]; 2959 2960 spin_init(&head->spin, "filehead_spin"); 2961 LIST_INIT(&head->list); 2962 } 2963 } 2964 2965 SYSINIT(filelistheads, SI_BOOT1_LOCK, SI_ORDER_ANY, 2966 filelist_heads_init, NULL); 2967 2968 static void 2969 file_objcache_init(void *dummy __unused) 2970 { 2971 file_objcache = objcache_create("file", maxfiles, maxfiles / 8, 2972 NULL, NULL, NULL, /* TODO: ctor/dtor */ 2973 objcache_malloc_alloc, objcache_malloc_free, &file_malloc_args); 2974 } 2975 SYSINIT(fpobjcache, SI_BOOT2_POST_SMP, SI_ORDER_ANY, file_objcache_init, NULL); 2976