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