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