kern_descrip.c revision 285356
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: head/sys/kern/kern_descrip.c 285356 2015-07-10 11:01:30Z mjg $"); 39 40#include "opt_capsicum.h" 41#include "opt_compat.h" 42#include "opt_ddb.h" 43#include "opt_ktrace.h" 44 45#include <sys/param.h> 46#include <sys/systm.h> 47 48#include <sys/capsicum.h> 49#include <sys/conf.h> 50#include <sys/fcntl.h> 51#include <sys/file.h> 52#include <sys/filedesc.h> 53#include <sys/filio.h> 54#include <sys/jail.h> 55#include <sys/kernel.h> 56#include <sys/limits.h> 57#include <sys/lock.h> 58#include <sys/malloc.h> 59#include <sys/mount.h> 60#include <sys/mutex.h> 61#include <sys/namei.h> 62#include <sys/selinfo.h> 63#include <sys/priv.h> 64#include <sys/proc.h> 65#include <sys/protosw.h> 66#include <sys/racct.h> 67#include <sys/resourcevar.h> 68#include <sys/sbuf.h> 69#include <sys/signalvar.h> 70#include <sys/socketvar.h> 71#include <sys/stat.h> 72#include <sys/sx.h> 73#include <sys/syscallsubr.h> 74#include <sys/sysctl.h> 75#include <sys/sysproto.h> 76#include <sys/unistd.h> 77#include <sys/user.h> 78#include <sys/vnode.h> 79#ifdef KTRACE 80#include <sys/ktrace.h> 81#endif 82 83#include <net/vnet.h> 84 85#include <security/audit/audit.h> 86 87#include <vm/uma.h> 88#include <vm/vm.h> 89 90#include <ddb/ddb.h> 91 92static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 93static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 94 "file desc to leader structures"); 95static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 96MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 97 98MALLOC_DECLARE(M_FADVISE); 99 100static uma_zone_t file_zone; 101static uma_zone_t filedesc0_zone; 102 103static int closefp(struct filedesc *fdp, int fd, struct file *fp, 104 struct thread *td, int holdleaders); 105static int fd_first_free(struct filedesc *fdp, int low, int size); 106static int fd_last_used(struct filedesc *fdp, int size); 107static void fdgrowtable(struct filedesc *fdp, int nfd); 108static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 109static void fdunused(struct filedesc *fdp, int fd); 110static void fdused(struct filedesc *fdp, int fd); 111static int getmaxfd(struct thread *td); 112 113/* 114 * Each process has: 115 * 116 * - An array of open file descriptors (fd_ofiles) 117 * - An array of file flags (fd_ofileflags) 118 * - A bitmap recording which descriptors are in use (fd_map) 119 * 120 * A process starts out with NDFILE descriptors. The value of NDFILE has 121 * been selected based the historical limit of 20 open files, and an 122 * assumption that the majority of processes, especially short-lived 123 * processes like shells, will never need more. 124 * 125 * If this initial allocation is exhausted, a larger descriptor table and 126 * map are allocated dynamically, and the pointers in the process's struct 127 * filedesc are updated to point to those. This is repeated every time 128 * the process runs out of file descriptors (provided it hasn't hit its 129 * resource limit). 130 * 131 * Since threads may hold references to individual descriptor table 132 * entries, the tables are never freed. Instead, they are placed on a 133 * linked list and freed only when the struct filedesc is released. 134 */ 135#define NDFILE 20 136#define NDSLOTSIZE sizeof(NDSLOTTYPE) 137#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 138#define NDSLOT(x) ((x) / NDENTRIES) 139#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 140#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 141 142/* 143 * SLIST entry used to keep track of ofiles which must be reclaimed when 144 * the process exits. 145 */ 146struct freetable { 147 struct fdescenttbl *ft_table; 148 SLIST_ENTRY(freetable) ft_next; 149}; 150 151/* 152 * Initial allocation: a filedesc structure + the head of SLIST used to 153 * keep track of old ofiles + enough space for NDFILE descriptors. 154 */ 155 156struct fdescenttbl0 { 157 int fdt_nfiles; 158 struct filedescent fdt_ofiles[NDFILE]; 159}; 160 161struct filedesc0 { 162 struct filedesc fd_fd; 163 SLIST_HEAD(, freetable) fd_free; 164 struct fdescenttbl0 fd_dfiles; 165 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 166}; 167 168/* 169 * Descriptor management. 170 */ 171volatile int openfiles; /* actual number of open files */ 172struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 173void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 174 175/* 176 * If low >= size, just return low. Otherwise find the first zero bit in the 177 * given bitmap, starting at low and not exceeding size - 1. Return size if 178 * not found. 179 */ 180static int 181fd_first_free(struct filedesc *fdp, int low, int size) 182{ 183 NDSLOTTYPE *map = fdp->fd_map; 184 NDSLOTTYPE mask; 185 int off, maxoff; 186 187 if (low >= size) 188 return (low); 189 190 off = NDSLOT(low); 191 if (low % NDENTRIES) { 192 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 193 if ((mask &= ~map[off]) != 0UL) 194 return (off * NDENTRIES + ffsl(mask) - 1); 195 ++off; 196 } 197 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 198 if (map[off] != ~0UL) 199 return (off * NDENTRIES + ffsl(~map[off]) - 1); 200 return (size); 201} 202 203/* 204 * Find the highest non-zero bit in the given bitmap, starting at 0 and 205 * not exceeding size - 1. Return -1 if not found. 206 */ 207static int 208fd_last_used(struct filedesc *fdp, int size) 209{ 210 NDSLOTTYPE *map = fdp->fd_map; 211 NDSLOTTYPE mask; 212 int off, minoff; 213 214 off = NDSLOT(size); 215 if (size % NDENTRIES) { 216 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 217 if ((mask &= map[off]) != 0) 218 return (off * NDENTRIES + flsl(mask) - 1); 219 --off; 220 } 221 for (minoff = NDSLOT(0); off >= minoff; --off) 222 if (map[off] != 0) 223 return (off * NDENTRIES + flsl(map[off]) - 1); 224 return (-1); 225} 226 227#ifdef INVARIANTS 228static int 229fdisused(struct filedesc *fdp, int fd) 230{ 231 232 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 233 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 234 235 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 236} 237#endif 238 239/* 240 * Mark a file descriptor as used. 241 */ 242static void 243fdused_init(struct filedesc *fdp, int fd) 244{ 245 246 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 247 248 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 249} 250 251static void 252fdused(struct filedesc *fdp, int fd) 253{ 254 255 FILEDESC_XLOCK_ASSERT(fdp); 256 257 fdused_init(fdp, fd); 258 if (fd > fdp->fd_lastfile) 259 fdp->fd_lastfile = fd; 260 if (fd == fdp->fd_freefile) 261 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 262} 263 264/* 265 * Mark a file descriptor as unused. 266 */ 267static void 268fdunused(struct filedesc *fdp, int fd) 269{ 270 271 FILEDESC_XLOCK_ASSERT(fdp); 272 273 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 274 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 275 ("fd=%d is still in use", fd)); 276 277 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 278 if (fd < fdp->fd_freefile) 279 fdp->fd_freefile = fd; 280 if (fd == fdp->fd_lastfile) 281 fdp->fd_lastfile = fd_last_used(fdp, fd); 282} 283 284/* 285 * Free a file descriptor. 286 * 287 * Avoid some work if fdp is about to be destroyed. 288 */ 289static inline void 290fdefree_last(struct filedescent *fde) 291{ 292 293 filecaps_free(&fde->fde_caps); 294} 295 296static inline void 297fdfree(struct filedesc *fdp, int fd) 298{ 299 struct filedescent *fde; 300 301 fde = &fdp->fd_ofiles[fd]; 302#ifdef CAPABILITIES 303 seq_write_begin(&fde->fde_seq); 304#endif 305 fdefree_last(fde); 306 fde->fde_file = NULL; 307 fdunused(fdp, fd); 308#ifdef CAPABILITIES 309 seq_write_end(&fde->fde_seq); 310#endif 311} 312 313/* 314 * System calls on descriptors. 315 */ 316#ifndef _SYS_SYSPROTO_H_ 317struct getdtablesize_args { 318 int dummy; 319}; 320#endif 321/* ARGSUSED */ 322int 323sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 324{ 325#ifdef RACCT 326 uint64_t lim; 327#endif 328 329 td->td_retval[0] = 330 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc); 331#ifdef RACCT 332 PROC_LOCK(td->td_proc); 333 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 334 PROC_UNLOCK(td->td_proc); 335 if (lim < td->td_retval[0]) 336 td->td_retval[0] = lim; 337#endif 338 return (0); 339} 340 341/* 342 * Duplicate a file descriptor to a particular value. 343 * 344 * Note: keep in mind that a potential race condition exists when closing 345 * descriptors from a shared descriptor table (via rfork). 346 */ 347#ifndef _SYS_SYSPROTO_H_ 348struct dup2_args { 349 u_int from; 350 u_int to; 351}; 352#endif 353/* ARGSUSED */ 354int 355sys_dup2(struct thread *td, struct dup2_args *uap) 356{ 357 358 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); 359} 360 361/* 362 * Duplicate a file descriptor. 363 */ 364#ifndef _SYS_SYSPROTO_H_ 365struct dup_args { 366 u_int fd; 367}; 368#endif 369/* ARGSUSED */ 370int 371sys_dup(struct thread *td, struct dup_args *uap) 372{ 373 374 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); 375} 376 377/* 378 * The file control system call. 379 */ 380#ifndef _SYS_SYSPROTO_H_ 381struct fcntl_args { 382 int fd; 383 int cmd; 384 long arg; 385}; 386#endif 387/* ARGSUSED */ 388int 389sys_fcntl(struct thread *td, struct fcntl_args *uap) 390{ 391 392 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 393} 394 395int 396kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) 397{ 398 struct flock fl; 399 struct __oflock ofl; 400 intptr_t arg1; 401 int error, newcmd; 402 403 error = 0; 404 newcmd = cmd; 405 switch (cmd) { 406 case F_OGETLK: 407 case F_OSETLK: 408 case F_OSETLKW: 409 /* 410 * Convert old flock structure to new. 411 */ 412 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 413 fl.l_start = ofl.l_start; 414 fl.l_len = ofl.l_len; 415 fl.l_pid = ofl.l_pid; 416 fl.l_type = ofl.l_type; 417 fl.l_whence = ofl.l_whence; 418 fl.l_sysid = 0; 419 420 switch (cmd) { 421 case F_OGETLK: 422 newcmd = F_GETLK; 423 break; 424 case F_OSETLK: 425 newcmd = F_SETLK; 426 break; 427 case F_OSETLKW: 428 newcmd = F_SETLKW; 429 break; 430 } 431 arg1 = (intptr_t)&fl; 432 break; 433 case F_GETLK: 434 case F_SETLK: 435 case F_SETLKW: 436 case F_SETLK_REMOTE: 437 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 438 arg1 = (intptr_t)&fl; 439 break; 440 default: 441 arg1 = arg; 442 break; 443 } 444 if (error) 445 return (error); 446 error = kern_fcntl(td, fd, newcmd, arg1); 447 if (error) 448 return (error); 449 if (cmd == F_OGETLK) { 450 ofl.l_start = fl.l_start; 451 ofl.l_len = fl.l_len; 452 ofl.l_pid = fl.l_pid; 453 ofl.l_type = fl.l_type; 454 ofl.l_whence = fl.l_whence; 455 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 456 } else if (cmd == F_GETLK) { 457 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 458 } 459 return (error); 460} 461 462int 463kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 464{ 465 struct filedesc *fdp; 466 struct flock *flp; 467 struct file *fp, *fp2; 468 struct filedescent *fde; 469 struct proc *p; 470 struct vnode *vp; 471 cap_rights_t rights; 472 int error, flg, tmp; 473 uint64_t bsize; 474 off_t foffset; 475 476 error = 0; 477 flg = F_POSIX; 478 p = td->td_proc; 479 fdp = p->p_fd; 480 481 switch (cmd) { 482 case F_DUPFD: 483 tmp = arg; 484 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); 485 break; 486 487 case F_DUPFD_CLOEXEC: 488 tmp = arg; 489 error = kern_dup(td, FDDUP_FCNTL, FDDUP_CLOEXEC, fd, tmp); 490 break; 491 492 case F_DUP2FD: 493 tmp = arg; 494 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); 495 break; 496 497 case F_DUP2FD_CLOEXEC: 498 tmp = arg; 499 error = kern_dup(td, FDDUP_FIXED, FDDUP_CLOEXEC, fd, tmp); 500 break; 501 502 case F_GETFD: 503 FILEDESC_SLOCK(fdp); 504 if (fget_locked(fdp, fd) == NULL) { 505 FILEDESC_SUNLOCK(fdp); 506 error = EBADF; 507 break; 508 } 509 fde = &fdp->fd_ofiles[fd]; 510 td->td_retval[0] = 511 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 512 FILEDESC_SUNLOCK(fdp); 513 break; 514 515 case F_SETFD: 516 FILEDESC_XLOCK(fdp); 517 if (fget_locked(fdp, fd) == NULL) { 518 FILEDESC_XUNLOCK(fdp); 519 error = EBADF; 520 break; 521 } 522 fde = &fdp->fd_ofiles[fd]; 523 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 524 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 525 FILEDESC_XUNLOCK(fdp); 526 break; 527 528 case F_GETFL: 529 error = fget_fcntl(td, fd, 530 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp); 531 if (error != 0) 532 break; 533 td->td_retval[0] = OFLAGS(fp->f_flag); 534 fdrop(fp, td); 535 break; 536 537 case F_SETFL: 538 error = fget_fcntl(td, fd, 539 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp); 540 if (error != 0) 541 break; 542 do { 543 tmp = flg = fp->f_flag; 544 tmp &= ~FCNTLFLAGS; 545 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 546 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 547 tmp = fp->f_flag & FNONBLOCK; 548 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 549 if (error != 0) { 550 fdrop(fp, td); 551 break; 552 } 553 tmp = fp->f_flag & FASYNC; 554 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 555 if (error == 0) { 556 fdrop(fp, td); 557 break; 558 } 559 atomic_clear_int(&fp->f_flag, FNONBLOCK); 560 tmp = 0; 561 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 562 fdrop(fp, td); 563 break; 564 565 case F_GETOWN: 566 error = fget_fcntl(td, fd, 567 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp); 568 if (error != 0) 569 break; 570 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 571 if (error == 0) 572 td->td_retval[0] = tmp; 573 fdrop(fp, td); 574 break; 575 576 case F_SETOWN: 577 error = fget_fcntl(td, fd, 578 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp); 579 if (error != 0) 580 break; 581 tmp = arg; 582 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 583 fdrop(fp, td); 584 break; 585 586 case F_SETLK_REMOTE: 587 error = priv_check(td, PRIV_NFS_LOCKD); 588 if (error) 589 return (error); 590 flg = F_REMOTE; 591 goto do_setlk; 592 593 case F_SETLKW: 594 flg |= F_WAIT; 595 /* FALLTHROUGH F_SETLK */ 596 597 case F_SETLK: 598 do_setlk: 599 cap_rights_init(&rights, CAP_FLOCK); 600 error = fget_unlocked(fdp, fd, &rights, &fp, NULL); 601 if (error != 0) 602 break; 603 if (fp->f_type != DTYPE_VNODE) { 604 error = EBADF; 605 fdrop(fp, td); 606 break; 607 } 608 609 flp = (struct flock *)arg; 610 if (flp->l_whence == SEEK_CUR) { 611 foffset = foffset_get(fp); 612 if (foffset < 0 || 613 (flp->l_start > 0 && 614 foffset > OFF_MAX - flp->l_start)) { 615 error = EOVERFLOW; 616 fdrop(fp, td); 617 break; 618 } 619 flp->l_start += foffset; 620 } 621 622 vp = fp->f_vnode; 623 switch (flp->l_type) { 624 case F_RDLCK: 625 if ((fp->f_flag & FREAD) == 0) { 626 error = EBADF; 627 break; 628 } 629 PROC_LOCK(p->p_leader); 630 p->p_leader->p_flag |= P_ADVLOCK; 631 PROC_UNLOCK(p->p_leader); 632 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 633 flp, flg); 634 break; 635 case F_WRLCK: 636 if ((fp->f_flag & FWRITE) == 0) { 637 error = EBADF; 638 break; 639 } 640 PROC_LOCK(p->p_leader); 641 p->p_leader->p_flag |= P_ADVLOCK; 642 PROC_UNLOCK(p->p_leader); 643 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 644 flp, flg); 645 break; 646 case F_UNLCK: 647 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 648 flp, flg); 649 break; 650 case F_UNLCKSYS: 651 /* 652 * Temporary api for testing remote lock 653 * infrastructure. 654 */ 655 if (flg != F_REMOTE) { 656 error = EINVAL; 657 break; 658 } 659 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 660 F_UNLCKSYS, flp, flg); 661 break; 662 default: 663 error = EINVAL; 664 break; 665 } 666 if (error != 0 || flp->l_type == F_UNLCK || 667 flp->l_type == F_UNLCKSYS) { 668 fdrop(fp, td); 669 break; 670 } 671 672 /* 673 * Check for a race with close. 674 * 675 * The vnode is now advisory locked (or unlocked, but this case 676 * is not really important) as the caller requested. 677 * We had to drop the filedesc lock, so we need to recheck if 678 * the descriptor is still valid, because if it was closed 679 * in the meantime we need to remove advisory lock from the 680 * vnode - close on any descriptor leading to an advisory 681 * locked vnode, removes that lock. 682 * We will return 0 on purpose in that case, as the result of 683 * successful advisory lock might have been externally visible 684 * already. This is fine - effectively we pretend to the caller 685 * that the closing thread was a bit slower and that the 686 * advisory lock succeeded before the close. 687 */ 688 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL); 689 if (error != 0) { 690 fdrop(fp, td); 691 break; 692 } 693 if (fp != fp2) { 694 flp->l_whence = SEEK_SET; 695 flp->l_start = 0; 696 flp->l_len = 0; 697 flp->l_type = F_UNLCK; 698 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 699 F_UNLCK, flp, F_POSIX); 700 } 701 fdrop(fp, td); 702 fdrop(fp2, td); 703 break; 704 705 case F_GETLK: 706 error = fget_unlocked(fdp, fd, 707 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL); 708 if (error != 0) 709 break; 710 if (fp->f_type != DTYPE_VNODE) { 711 error = EBADF; 712 fdrop(fp, td); 713 break; 714 } 715 flp = (struct flock *)arg; 716 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 717 flp->l_type != F_UNLCK) { 718 error = EINVAL; 719 fdrop(fp, td); 720 break; 721 } 722 if (flp->l_whence == SEEK_CUR) { 723 foffset = foffset_get(fp); 724 if ((flp->l_start > 0 && 725 foffset > OFF_MAX - flp->l_start) || 726 (flp->l_start < 0 && 727 foffset < OFF_MIN - flp->l_start)) { 728 error = EOVERFLOW; 729 fdrop(fp, td); 730 break; 731 } 732 flp->l_start += foffset; 733 } 734 vp = fp->f_vnode; 735 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 736 F_POSIX); 737 fdrop(fp, td); 738 break; 739 740 case F_RDAHEAD: 741 arg = arg ? 128 * 1024: 0; 742 /* FALLTHROUGH */ 743 case F_READAHEAD: 744 error = fget_unlocked(fdp, fd, 745 cap_rights_init(&rights), &fp, NULL); 746 if (error != 0) 747 break; 748 if (fp->f_type != DTYPE_VNODE) { 749 fdrop(fp, td); 750 error = EBADF; 751 break; 752 } 753 vp = fp->f_vnode; 754 /* 755 * Exclusive lock synchronizes against f_seqcount reads and 756 * writes in sequential_heuristic(). 757 */ 758 error = vn_lock(vp, LK_EXCLUSIVE); 759 if (error != 0) { 760 fdrop(fp, td); 761 break; 762 } 763 if (arg >= 0) { 764 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 765 fp->f_seqcount = (arg + bsize - 1) / bsize; 766 atomic_set_int(&fp->f_flag, FRDAHEAD); 767 } else { 768 atomic_clear_int(&fp->f_flag, FRDAHEAD); 769 } 770 VOP_UNLOCK(vp, 0); 771 fdrop(fp, td); 772 break; 773 774 default: 775 error = EINVAL; 776 break; 777 } 778 return (error); 779} 780 781static int 782getmaxfd(struct thread *td) 783{ 784 785 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 786} 787 788/* 789 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 790 */ 791int 792kern_dup(struct thread *td, u_int mode, int flags, int old, int new) 793{ 794 struct filedesc *fdp; 795 struct filedescent *oldfde, *newfde; 796 struct proc *p; 797 struct file *fp; 798 struct file *delfp; 799 int error, maxfd; 800 801 p = td->td_proc; 802 fdp = p->p_fd; 803 804 MPASS((flags & ~(FDDUP_CLOEXEC)) == 0); 805 MPASS(mode < FDDUP_LASTMODE); 806 807 /* 808 * Verify we have a valid descriptor to dup from and possibly to 809 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 810 * return EINVAL when the new descriptor is out of bounds. 811 */ 812 if (old < 0) 813 return (EBADF); 814 if (new < 0) 815 return (flags & FDDUP_FCNTL ? EINVAL : EBADF); 816 maxfd = getmaxfd(td); 817 if (new >= maxfd) 818 return (flags & FDDUP_FCNTL ? EINVAL : EBADF); 819 820 FILEDESC_XLOCK(fdp); 821 if (fget_locked(fdp, old) == NULL) { 822 FILEDESC_XUNLOCK(fdp); 823 return (EBADF); 824 } 825 oldfde = &fdp->fd_ofiles[old]; 826 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) { 827 td->td_retval[0] = new; 828 if (flags & FDDUP_CLOEXEC) 829 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 830 FILEDESC_XUNLOCK(fdp); 831 return (0); 832 } 833 fp = oldfde->fde_file; 834 fhold(fp); 835 836 /* 837 * If the caller specified a file descriptor, make sure the file 838 * table is large enough to hold it, and grab it. Otherwise, just 839 * allocate a new descriptor the usual way. 840 */ 841 switch (mode) { 842 case FDDUP_NORMAL: 843 case FDDUP_FCNTL: 844 if ((error = fdalloc(td, new, &new)) != 0) { 845 FILEDESC_XUNLOCK(fdp); 846 fdrop(fp, td); 847 return (error); 848 } 849 newfde = &fdp->fd_ofiles[new]; 850 break; 851 case FDDUP_MUSTREPLACE: 852 /* Target file descriptor must exist. */ 853 if (new >= fdp->fd_nfiles || 854 fdp->fd_ofiles[new].fde_file == NULL) { 855 FILEDESC_XUNLOCK(fdp); 856 fdrop(fp, td); 857 return (EBADF); 858 } 859 newfde = &fdp->fd_ofiles[new]; 860 break; 861 case FDDUP_FIXED: 862 if (new >= fdp->fd_nfiles) { 863 /* 864 * The resource limits are here instead of e.g. 865 * fdalloc(), because the file descriptor table may be 866 * shared between processes, so we can't really use 867 * racct_add()/racct_sub(). Instead of counting the 868 * number of actually allocated descriptors, just put 869 * the limit on the size of the file descriptor table. 870 */ 871#ifdef RACCT 872 if (racct_enable) { 873 PROC_LOCK(p); 874 error = racct_set(p, RACCT_NOFILE, new + 1); 875 PROC_UNLOCK(p); 876 if (error != 0) { 877 FILEDESC_XUNLOCK(fdp); 878 fdrop(fp, td); 879 return (EMFILE); 880 } 881 } 882#endif 883 fdgrowtable_exp(fdp, new + 1); 884 oldfde = &fdp->fd_ofiles[old]; 885 } 886 newfde = &fdp->fd_ofiles[new]; 887 if (newfde->fde_file == NULL) 888 fdused(fdp, new); 889 break; 890#ifdef INVARIANTS 891 default: 892 newfde = NULL; /* silence the compiler */ 893 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 894#endif 895 } 896 897 KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); 898 KASSERT(old != new, ("new fd is same as old")); 899 900 delfp = newfde->fde_file; 901 902 /* 903 * Duplicate the source descriptor. 904 */ 905#ifdef CAPABILITIES 906 seq_write_begin(&newfde->fde_seq); 907#endif 908 filecaps_free(&newfde->fde_caps); 909 memcpy(newfde, oldfde, fde_change_size); 910 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 911 if ((flags & FDDUP_CLOEXEC) != 0) 912 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 913 else 914 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 915#ifdef CAPABILITIES 916 seq_write_end(&newfde->fde_seq); 917#endif 918 td->td_retval[0] = new; 919 920 if (delfp != NULL) { 921 (void) closefp(fdp, new, delfp, td, 1); 922 /* closefp() drops the FILEDESC lock for us. */ 923 } else { 924 FILEDESC_XUNLOCK(fdp); 925 } 926 927 return (0); 928} 929 930/* 931 * If sigio is on the list associated with a process or process group, 932 * disable signalling from the device, remove sigio from the list and 933 * free sigio. 934 */ 935void 936funsetown(struct sigio **sigiop) 937{ 938 struct sigio *sigio; 939 940 SIGIO_LOCK(); 941 sigio = *sigiop; 942 if (sigio == NULL) { 943 SIGIO_UNLOCK(); 944 return; 945 } 946 *(sigio->sio_myref) = NULL; 947 if ((sigio)->sio_pgid < 0) { 948 struct pgrp *pg = (sigio)->sio_pgrp; 949 PGRP_LOCK(pg); 950 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 951 sigio, sio_pgsigio); 952 PGRP_UNLOCK(pg); 953 } else { 954 struct proc *p = (sigio)->sio_proc; 955 PROC_LOCK(p); 956 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 957 sigio, sio_pgsigio); 958 PROC_UNLOCK(p); 959 } 960 SIGIO_UNLOCK(); 961 crfree(sigio->sio_ucred); 962 free(sigio, M_SIGIO); 963} 964 965/* 966 * Free a list of sigio structures. 967 * We only need to lock the SIGIO_LOCK because we have made ourselves 968 * inaccessible to callers of fsetown and therefore do not need to lock 969 * the proc or pgrp struct for the list manipulation. 970 */ 971void 972funsetownlst(struct sigiolst *sigiolst) 973{ 974 struct proc *p; 975 struct pgrp *pg; 976 struct sigio *sigio; 977 978 sigio = SLIST_FIRST(sigiolst); 979 if (sigio == NULL) 980 return; 981 p = NULL; 982 pg = NULL; 983 984 /* 985 * Every entry of the list should belong 986 * to a single proc or pgrp. 987 */ 988 if (sigio->sio_pgid < 0) { 989 pg = sigio->sio_pgrp; 990 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 991 } else /* if (sigio->sio_pgid > 0) */ { 992 p = sigio->sio_proc; 993 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 994 } 995 996 SIGIO_LOCK(); 997 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 998 *(sigio->sio_myref) = NULL; 999 if (pg != NULL) { 1000 KASSERT(sigio->sio_pgid < 0, 1001 ("Proc sigio in pgrp sigio list")); 1002 KASSERT(sigio->sio_pgrp == pg, 1003 ("Bogus pgrp in sigio list")); 1004 PGRP_LOCK(pg); 1005 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 1006 sio_pgsigio); 1007 PGRP_UNLOCK(pg); 1008 } else /* if (p != NULL) */ { 1009 KASSERT(sigio->sio_pgid > 0, 1010 ("Pgrp sigio in proc sigio list")); 1011 KASSERT(sigio->sio_proc == p, 1012 ("Bogus proc in sigio list")); 1013 PROC_LOCK(p); 1014 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1015 sio_pgsigio); 1016 PROC_UNLOCK(p); 1017 } 1018 SIGIO_UNLOCK(); 1019 crfree(sigio->sio_ucred); 1020 free(sigio, M_SIGIO); 1021 SIGIO_LOCK(); 1022 } 1023 SIGIO_UNLOCK(); 1024} 1025 1026/* 1027 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1028 * 1029 * After permission checking, add a sigio structure to the sigio list for 1030 * the process or process group. 1031 */ 1032int 1033fsetown(pid_t pgid, struct sigio **sigiop) 1034{ 1035 struct proc *proc; 1036 struct pgrp *pgrp; 1037 struct sigio *sigio; 1038 int ret; 1039 1040 if (pgid == 0) { 1041 funsetown(sigiop); 1042 return (0); 1043 } 1044 1045 ret = 0; 1046 1047 /* Allocate and fill in the new sigio out of locks. */ 1048 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1049 sigio->sio_pgid = pgid; 1050 sigio->sio_ucred = crhold(curthread->td_ucred); 1051 sigio->sio_myref = sigiop; 1052 1053 sx_slock(&proctree_lock); 1054 if (pgid > 0) { 1055 proc = pfind(pgid); 1056 if (proc == NULL) { 1057 ret = ESRCH; 1058 goto fail; 1059 } 1060 1061 /* 1062 * Policy - Don't allow a process to FSETOWN a process 1063 * in another session. 1064 * 1065 * Remove this test to allow maximum flexibility or 1066 * restrict FSETOWN to the current process or process 1067 * group for maximum safety. 1068 */ 1069 PROC_UNLOCK(proc); 1070 if (proc->p_session != curthread->td_proc->p_session) { 1071 ret = EPERM; 1072 goto fail; 1073 } 1074 1075 pgrp = NULL; 1076 } else /* if (pgid < 0) */ { 1077 pgrp = pgfind(-pgid); 1078 if (pgrp == NULL) { 1079 ret = ESRCH; 1080 goto fail; 1081 } 1082 PGRP_UNLOCK(pgrp); 1083 1084 /* 1085 * Policy - Don't allow a process to FSETOWN a process 1086 * in another session. 1087 * 1088 * Remove this test to allow maximum flexibility or 1089 * restrict FSETOWN to the current process or process 1090 * group for maximum safety. 1091 */ 1092 if (pgrp->pg_session != curthread->td_proc->p_session) { 1093 ret = EPERM; 1094 goto fail; 1095 } 1096 1097 proc = NULL; 1098 } 1099 funsetown(sigiop); 1100 if (pgid > 0) { 1101 PROC_LOCK(proc); 1102 /* 1103 * Since funsetownlst() is called without the proctree 1104 * locked, we need to check for P_WEXIT. 1105 * XXX: is ESRCH correct? 1106 */ 1107 if ((proc->p_flag & P_WEXIT) != 0) { 1108 PROC_UNLOCK(proc); 1109 ret = ESRCH; 1110 goto fail; 1111 } 1112 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1113 sigio->sio_proc = proc; 1114 PROC_UNLOCK(proc); 1115 } else { 1116 PGRP_LOCK(pgrp); 1117 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1118 sigio->sio_pgrp = pgrp; 1119 PGRP_UNLOCK(pgrp); 1120 } 1121 sx_sunlock(&proctree_lock); 1122 SIGIO_LOCK(); 1123 *sigiop = sigio; 1124 SIGIO_UNLOCK(); 1125 return (0); 1126 1127fail: 1128 sx_sunlock(&proctree_lock); 1129 crfree(sigio->sio_ucred); 1130 free(sigio, M_SIGIO); 1131 return (ret); 1132} 1133 1134/* 1135 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1136 */ 1137pid_t 1138fgetown(sigiop) 1139 struct sigio **sigiop; 1140{ 1141 pid_t pgid; 1142 1143 SIGIO_LOCK(); 1144 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1145 SIGIO_UNLOCK(); 1146 return (pgid); 1147} 1148 1149/* 1150 * Function drops the filedesc lock on return. 1151 */ 1152static int 1153closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1154 int holdleaders) 1155{ 1156 int error; 1157 1158 FILEDESC_XLOCK_ASSERT(fdp); 1159 1160 if (holdleaders) { 1161 if (td->td_proc->p_fdtol != NULL) { 1162 /* 1163 * Ask fdfree() to sleep to ensure that all relevant 1164 * process leaders can be traversed in closef(). 1165 */ 1166 fdp->fd_holdleaderscount++; 1167 } else { 1168 holdleaders = 0; 1169 } 1170 } 1171 1172 /* 1173 * We now hold the fp reference that used to be owned by the 1174 * descriptor array. We have to unlock the FILEDESC *AFTER* 1175 * knote_fdclose to prevent a race of the fd getting opened, a knote 1176 * added, and deleteing a knote for the new fd. 1177 */ 1178 knote_fdclose(td, fd); 1179 1180 /* 1181 * We need to notify mqueue if the object is of type mqueue. 1182 */ 1183 if (fp->f_type == DTYPE_MQUEUE) 1184 mq_fdclose(td, fd, fp); 1185 FILEDESC_XUNLOCK(fdp); 1186 1187 error = closef(fp, td); 1188 if (holdleaders) { 1189 FILEDESC_XLOCK(fdp); 1190 fdp->fd_holdleaderscount--; 1191 if (fdp->fd_holdleaderscount == 0 && 1192 fdp->fd_holdleaderswakeup != 0) { 1193 fdp->fd_holdleaderswakeup = 0; 1194 wakeup(&fdp->fd_holdleaderscount); 1195 } 1196 FILEDESC_XUNLOCK(fdp); 1197 } 1198 return (error); 1199} 1200 1201/* 1202 * Close a file descriptor. 1203 */ 1204#ifndef _SYS_SYSPROTO_H_ 1205struct close_args { 1206 int fd; 1207}; 1208#endif 1209/* ARGSUSED */ 1210int 1211sys_close(struct thread *td, struct close_args *uap) 1212{ 1213 1214 return (kern_close(td, uap->fd)); 1215} 1216 1217int 1218kern_close(struct thread *td, int fd) 1219{ 1220 struct filedesc *fdp; 1221 struct file *fp; 1222 1223 fdp = td->td_proc->p_fd; 1224 1225 AUDIT_SYSCLOSE(td, fd); 1226 1227 FILEDESC_XLOCK(fdp); 1228 if ((fp = fget_locked(fdp, fd)) == NULL) { 1229 FILEDESC_XUNLOCK(fdp); 1230 return (EBADF); 1231 } 1232 fdfree(fdp, fd); 1233 1234 /* closefp() drops the FILEDESC lock for us. */ 1235 return (closefp(fdp, fd, fp, td, 1)); 1236} 1237 1238/* 1239 * Close open file descriptors. 1240 */ 1241#ifndef _SYS_SYSPROTO_H_ 1242struct closefrom_args { 1243 int lowfd; 1244}; 1245#endif 1246/* ARGSUSED */ 1247int 1248sys_closefrom(struct thread *td, struct closefrom_args *uap) 1249{ 1250 struct filedesc *fdp; 1251 int fd; 1252 1253 fdp = td->td_proc->p_fd; 1254 AUDIT_ARG_FD(uap->lowfd); 1255 1256 /* 1257 * Treat negative starting file descriptor values identical to 1258 * closefrom(0) which closes all files. 1259 */ 1260 if (uap->lowfd < 0) 1261 uap->lowfd = 0; 1262 FILEDESC_SLOCK(fdp); 1263 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1264 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1265 FILEDESC_SUNLOCK(fdp); 1266 (void)kern_close(td, fd); 1267 FILEDESC_SLOCK(fdp); 1268 } 1269 } 1270 FILEDESC_SUNLOCK(fdp); 1271 return (0); 1272} 1273 1274#if defined(COMPAT_43) 1275/* 1276 * Return status information about a file descriptor. 1277 */ 1278#ifndef _SYS_SYSPROTO_H_ 1279struct ofstat_args { 1280 int fd; 1281 struct ostat *sb; 1282}; 1283#endif 1284/* ARGSUSED */ 1285int 1286ofstat(struct thread *td, struct ofstat_args *uap) 1287{ 1288 struct ostat oub; 1289 struct stat ub; 1290 int error; 1291 1292 error = kern_fstat(td, uap->fd, &ub); 1293 if (error == 0) { 1294 cvtstat(&ub, &oub); 1295 error = copyout(&oub, uap->sb, sizeof(oub)); 1296 } 1297 return (error); 1298} 1299#endif /* COMPAT_43 */ 1300 1301/* 1302 * Return status information about a file descriptor. 1303 */ 1304#ifndef _SYS_SYSPROTO_H_ 1305struct fstat_args { 1306 int fd; 1307 struct stat *sb; 1308}; 1309#endif 1310/* ARGSUSED */ 1311int 1312sys_fstat(struct thread *td, struct fstat_args *uap) 1313{ 1314 struct stat ub; 1315 int error; 1316 1317 error = kern_fstat(td, uap->fd, &ub); 1318 if (error == 0) 1319 error = copyout(&ub, uap->sb, sizeof(ub)); 1320 return (error); 1321} 1322 1323int 1324kern_fstat(struct thread *td, int fd, struct stat *sbp) 1325{ 1326 struct file *fp; 1327 cap_rights_t rights; 1328 int error; 1329 1330 AUDIT_ARG_FD(fd); 1331 1332 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1333 if (error != 0) 1334 return (error); 1335 1336 AUDIT_ARG_FILE(td->td_proc, fp); 1337 1338 error = fo_stat(fp, sbp, td->td_ucred, td); 1339 fdrop(fp, td); 1340#ifdef KTRACE 1341 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1342 ktrstat(sbp); 1343#endif 1344 return (error); 1345} 1346 1347/* 1348 * Return status information about a file descriptor. 1349 */ 1350#ifndef _SYS_SYSPROTO_H_ 1351struct nfstat_args { 1352 int fd; 1353 struct nstat *sb; 1354}; 1355#endif 1356/* ARGSUSED */ 1357int 1358sys_nfstat(struct thread *td, struct nfstat_args *uap) 1359{ 1360 struct nstat nub; 1361 struct stat ub; 1362 int error; 1363 1364 error = kern_fstat(td, uap->fd, &ub); 1365 if (error == 0) { 1366 cvtnstat(&ub, &nub); 1367 error = copyout(&nub, uap->sb, sizeof(nub)); 1368 } 1369 return (error); 1370} 1371 1372/* 1373 * Return pathconf information about a file descriptor. 1374 */ 1375#ifndef _SYS_SYSPROTO_H_ 1376struct fpathconf_args { 1377 int fd; 1378 int name; 1379}; 1380#endif 1381/* ARGSUSED */ 1382int 1383sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1384{ 1385 struct file *fp; 1386 struct vnode *vp; 1387 cap_rights_t rights; 1388 int error; 1389 1390 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1391 if (error != 0) 1392 return (error); 1393 1394 /* If asynchronous I/O is available, it works for all descriptors. */ 1395 if (uap->name == _PC_ASYNC_IO) { 1396 td->td_retval[0] = async_io_version; 1397 goto out; 1398 } 1399 vp = fp->f_vnode; 1400 if (vp != NULL) { 1401 vn_lock(vp, LK_SHARED | LK_RETRY); 1402 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1403 VOP_UNLOCK(vp, 0); 1404 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1405 if (uap->name != _PC_PIPE_BUF) { 1406 error = EINVAL; 1407 } else { 1408 td->td_retval[0] = PIPE_BUF; 1409 error = 0; 1410 } 1411 } else { 1412 error = EOPNOTSUPP; 1413 } 1414out: 1415 fdrop(fp, td); 1416 return (error); 1417} 1418 1419/* 1420 * Initialize filecaps structure. 1421 */ 1422void 1423filecaps_init(struct filecaps *fcaps) 1424{ 1425 1426 bzero(fcaps, sizeof(*fcaps)); 1427 fcaps->fc_nioctls = -1; 1428} 1429 1430/* 1431 * Copy filecaps structure allocating memory for ioctls array if needed. 1432 */ 1433void 1434filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1435{ 1436 size_t size; 1437 1438 *dst = *src; 1439 if (src->fc_ioctls != NULL) { 1440 KASSERT(src->fc_nioctls > 0, 1441 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1442 1443 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1444 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1445 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1446 } 1447} 1448 1449/* 1450 * Move filecaps structure to the new place and clear the old place. 1451 */ 1452void 1453filecaps_move(struct filecaps *src, struct filecaps *dst) 1454{ 1455 1456 *dst = *src; 1457 bzero(src, sizeof(*src)); 1458} 1459 1460/* 1461 * Fill the given filecaps structure with full rights. 1462 */ 1463static void 1464filecaps_fill(struct filecaps *fcaps) 1465{ 1466 1467 CAP_ALL(&fcaps->fc_rights); 1468 fcaps->fc_ioctls = NULL; 1469 fcaps->fc_nioctls = -1; 1470 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1471} 1472 1473/* 1474 * Free memory allocated within filecaps structure. 1475 */ 1476void 1477filecaps_free(struct filecaps *fcaps) 1478{ 1479 1480 free(fcaps->fc_ioctls, M_FILECAPS); 1481 bzero(fcaps, sizeof(*fcaps)); 1482} 1483 1484/* 1485 * Validate the given filecaps structure. 1486 */ 1487static void 1488filecaps_validate(const struct filecaps *fcaps, const char *func) 1489{ 1490 1491 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1492 ("%s: invalid rights", func)); 1493 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1494 ("%s: invalid fcntls", func)); 1495 KASSERT(fcaps->fc_fcntls == 0 || 1496 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1497 ("%s: fcntls without CAP_FCNTL", func)); 1498 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1499 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1500 ("%s: invalid ioctls", func)); 1501 KASSERT(fcaps->fc_nioctls == 0 || 1502 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1503 ("%s: ioctls without CAP_IOCTL", func)); 1504} 1505 1506static void 1507fdgrowtable_exp(struct filedesc *fdp, int nfd) 1508{ 1509 int nfd1; 1510 1511 FILEDESC_XLOCK_ASSERT(fdp); 1512 1513 nfd1 = fdp->fd_nfiles * 2; 1514 if (nfd1 < nfd) 1515 nfd1 = nfd; 1516 fdgrowtable(fdp, nfd1); 1517} 1518 1519/* 1520 * Grow the file table to accomodate (at least) nfd descriptors. 1521 */ 1522static void 1523fdgrowtable(struct filedesc *fdp, int nfd) 1524{ 1525 struct filedesc0 *fdp0; 1526 struct freetable *ft; 1527 struct fdescenttbl *ntable; 1528 struct fdescenttbl *otable; 1529 int nnfiles, onfiles; 1530 NDSLOTTYPE *nmap, *omap; 1531 1532 /* 1533 * If lastfile is -1 this struct filedesc was just allocated and we are 1534 * growing it to accomodate for the one we are going to copy from. There 1535 * is no need to have a lock on this one as it's not visible to anyone. 1536 */ 1537 if (fdp->fd_lastfile != -1) 1538 FILEDESC_XLOCK_ASSERT(fdp); 1539 1540 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1541 1542 /* save old values */ 1543 onfiles = fdp->fd_nfiles; 1544 otable = fdp->fd_files; 1545 omap = fdp->fd_map; 1546 1547 /* compute the size of the new table */ 1548 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1549 if (nnfiles <= onfiles) 1550 /* the table is already large enough */ 1551 return; 1552 1553 /* 1554 * Allocate a new table. We need enough space for the number of 1555 * entries, file entries themselves and the struct freetable we will use 1556 * when we decommission the table and place it on the freelist. 1557 * We place the struct freetable in the middle so we don't have 1558 * to worry about padding. 1559 */ 1560 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1561 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1562 sizeof(struct freetable), 1563 M_FILEDESC, M_ZERO | M_WAITOK); 1564 /* copy the old data */ 1565 ntable->fdt_nfiles = nnfiles; 1566 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1567 onfiles * sizeof(ntable->fdt_ofiles[0])); 1568 1569 /* 1570 * Allocate a new map only if the old is not large enough. It will 1571 * grow at a slower rate than the table as it can map more 1572 * entries than the table can hold. 1573 */ 1574 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1575 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1576 M_ZERO | M_WAITOK); 1577 /* copy over the old data and update the pointer */ 1578 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1579 fdp->fd_map = nmap; 1580 } 1581 1582 /* 1583 * Make sure that ntable is correctly initialized before we replace 1584 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1585 * data. 1586 */ 1587 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1588 1589 /* 1590 * Do not free the old file table, as some threads may still 1591 * reference entries within it. Instead, place it on a freelist 1592 * which will be processed when the struct filedesc is released. 1593 * 1594 * Note that if onfiles == NDFILE, we're dealing with the original 1595 * static allocation contained within (struct filedesc0 *)fdp, 1596 * which must not be freed. 1597 */ 1598 if (onfiles > NDFILE) { 1599 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1600 fdp0 = (struct filedesc0 *)fdp; 1601 ft->ft_table = otable; 1602 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1603 } 1604 /* 1605 * The map does not have the same possibility of threads still 1606 * holding references to it. So always free it as long as it 1607 * does not reference the original static allocation. 1608 */ 1609 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1610 free(omap, M_FILEDESC); 1611} 1612 1613/* 1614 * Allocate a file descriptor for the process. 1615 */ 1616int 1617fdalloc(struct thread *td, int minfd, int *result) 1618{ 1619 struct proc *p = td->td_proc; 1620 struct filedesc *fdp = p->p_fd; 1621 int fd, maxfd, allocfd; 1622#ifdef RACCT 1623 int error; 1624#endif 1625 1626 FILEDESC_XLOCK_ASSERT(fdp); 1627 1628 if (fdp->fd_freefile > minfd) 1629 minfd = fdp->fd_freefile; 1630 1631 maxfd = getmaxfd(td); 1632 1633 /* 1634 * Search the bitmap for a free descriptor starting at minfd. 1635 * If none is found, grow the file table. 1636 */ 1637 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1638 if (fd >= maxfd) 1639 return (EMFILE); 1640 if (fd >= fdp->fd_nfiles) { 1641 allocfd = min(fd * 2, maxfd); 1642#ifdef RACCT 1643 if (racct_enable) { 1644 PROC_LOCK(p); 1645 error = racct_set(p, RACCT_NOFILE, allocfd); 1646 PROC_UNLOCK(p); 1647 if (error != 0) 1648 return (EMFILE); 1649 } 1650#endif 1651 /* 1652 * fd is already equal to first free descriptor >= minfd, so 1653 * we only need to grow the table and we are done. 1654 */ 1655 fdgrowtable_exp(fdp, allocfd); 1656 } 1657 1658 /* 1659 * Perform some sanity checks, then mark the file descriptor as 1660 * used and return it to the caller. 1661 */ 1662 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1663 ("invalid descriptor %d", fd)); 1664 KASSERT(!fdisused(fdp, fd), 1665 ("fd_first_free() returned non-free descriptor")); 1666 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1667 ("file descriptor isn't free")); 1668 fdused(fdp, fd); 1669 *result = fd; 1670 return (0); 1671} 1672 1673/* 1674 * Allocate n file descriptors for the process. 1675 */ 1676int 1677fdallocn(struct thread *td, int minfd, int *fds, int n) 1678{ 1679 struct proc *p = td->td_proc; 1680 struct filedesc *fdp = p->p_fd; 1681 int i; 1682 1683 FILEDESC_XLOCK_ASSERT(fdp); 1684 1685 for (i = 0; i < n; i++) 1686 if (fdalloc(td, 0, &fds[i]) != 0) 1687 break; 1688 1689 if (i < n) { 1690 for (i--; i >= 0; i--) 1691 fdunused(fdp, fds[i]); 1692 return (EMFILE); 1693 } 1694 1695 return (0); 1696} 1697 1698/* 1699 * Create a new open file structure and allocate a file decriptor for the 1700 * process that refers to it. We add one reference to the file for the 1701 * descriptor table and one reference for resultfp. This is to prevent us 1702 * being preempted and the entry in the descriptor table closed after we 1703 * release the FILEDESC lock. 1704 */ 1705int 1706falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1707{ 1708 struct file *fp; 1709 int error, fd; 1710 1711 error = falloc_noinstall(td, &fp); 1712 if (error) 1713 return (error); /* no reference held on error */ 1714 1715 error = finstall(td, fp, &fd, flags, NULL); 1716 if (error) { 1717 fdrop(fp, td); /* one reference (fp only) */ 1718 return (error); 1719 } 1720 1721 if (resultfp != NULL) 1722 *resultfp = fp; /* copy out result */ 1723 else 1724 fdrop(fp, td); /* release local reference */ 1725 1726 if (resultfd != NULL) 1727 *resultfd = fd; 1728 1729 return (0); 1730} 1731 1732/* 1733 * Create a new open file structure without allocating a file descriptor. 1734 */ 1735int 1736falloc_noinstall(struct thread *td, struct file **resultfp) 1737{ 1738 struct file *fp; 1739 int maxuserfiles = maxfiles - (maxfiles / 20); 1740 static struct timeval lastfail; 1741 static int curfail; 1742 1743 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1744 1745 if ((openfiles >= maxuserfiles && 1746 priv_check(td, PRIV_MAXFILES) != 0) || 1747 openfiles >= maxfiles) { 1748 if (ppsratecheck(&lastfail, &curfail, 1)) { 1749 printf("kern.maxfiles limit exceeded by uid %i, " 1750 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1751 } 1752 return (ENFILE); 1753 } 1754 atomic_add_int(&openfiles, 1); 1755 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1756 refcount_init(&fp->f_count, 1); 1757 fp->f_cred = crhold(td->td_ucred); 1758 fp->f_ops = &badfileops; 1759 *resultfp = fp; 1760 return (0); 1761} 1762 1763/* 1764 * Install a file in a file descriptor table. 1765 */ 1766void 1767_finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 1768 struct filecaps *fcaps) 1769{ 1770 struct filedescent *fde; 1771 1772 MPASS(fp != NULL); 1773 if (fcaps != NULL) 1774 filecaps_validate(fcaps, __func__); 1775 FILEDESC_XLOCK_ASSERT(fdp); 1776 1777 fde = &fdp->fd_ofiles[fd]; 1778#ifdef CAPABILITIES 1779 seq_write_begin(&fde->fde_seq); 1780#endif 1781 fde->fde_file = fp; 1782 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 1783 if (fcaps != NULL) 1784 filecaps_move(fcaps, &fde->fde_caps); 1785 else 1786 filecaps_fill(&fde->fde_caps); 1787#ifdef CAPABILITIES 1788 seq_write_end(&fde->fde_seq); 1789#endif 1790} 1791 1792int 1793finstall(struct thread *td, struct file *fp, int *fd, int flags, 1794 struct filecaps *fcaps) 1795{ 1796 struct filedesc *fdp = td->td_proc->p_fd; 1797 int error; 1798 1799 MPASS(fd != NULL); 1800 1801 FILEDESC_XLOCK(fdp); 1802 if ((error = fdalloc(td, 0, fd))) { 1803 FILEDESC_XUNLOCK(fdp); 1804 return (error); 1805 } 1806 fhold(fp); 1807 _finstall(fdp, fp, *fd, flags, fcaps); 1808 FILEDESC_XUNLOCK(fdp); 1809 return (0); 1810} 1811 1812/* 1813 * Build a new filedesc structure from another. 1814 * Copy the current, root, and jail root vnode references. 1815 * 1816 * If fdp is not NULL, return with it shared locked. 1817 */ 1818struct filedesc * 1819fdinit(struct filedesc *fdp, bool prepfiles) 1820{ 1821 struct filedesc0 *newfdp0; 1822 struct filedesc *newfdp; 1823 1824 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 1825 newfdp = &newfdp0->fd_fd; 1826 1827 /* Create the file descriptor table. */ 1828 FILEDESC_LOCK_INIT(newfdp); 1829 refcount_init(&newfdp->fd_refcnt, 1); 1830 refcount_init(&newfdp->fd_holdcnt, 1); 1831 newfdp->fd_cmask = CMASK; 1832 newfdp->fd_map = newfdp0->fd_dmap; 1833 newfdp->fd_lastfile = -1; 1834 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 1835 newfdp->fd_files->fdt_nfiles = NDFILE; 1836 1837 if (fdp == NULL) 1838 return (newfdp); 1839 1840 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) 1841 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1842 1843 FILEDESC_SLOCK(fdp); 1844 newfdp->fd_cdir = fdp->fd_cdir; 1845 if (newfdp->fd_cdir) 1846 VREF(newfdp->fd_cdir); 1847 newfdp->fd_rdir = fdp->fd_rdir; 1848 if (newfdp->fd_rdir) 1849 VREF(newfdp->fd_rdir); 1850 newfdp->fd_jdir = fdp->fd_jdir; 1851 if (newfdp->fd_jdir) 1852 VREF(newfdp->fd_jdir); 1853 1854 if (!prepfiles) { 1855 FILEDESC_SUNLOCK(fdp); 1856 } else { 1857 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1858 FILEDESC_SUNLOCK(fdp); 1859 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1860 FILEDESC_SLOCK(fdp); 1861 } 1862 } 1863 1864 return (newfdp); 1865} 1866 1867static struct filedesc * 1868fdhold(struct proc *p) 1869{ 1870 struct filedesc *fdp; 1871 1872 PROC_LOCK_ASSERT(p, MA_OWNED); 1873 fdp = p->p_fd; 1874 if (fdp != NULL) 1875 refcount_acquire(&fdp->fd_holdcnt); 1876 return (fdp); 1877} 1878 1879static void 1880fddrop(struct filedesc *fdp) 1881{ 1882 1883 if (fdp->fd_holdcnt > 1) { 1884 if (refcount_release(&fdp->fd_holdcnt) == 0) 1885 return; 1886 } 1887 1888 FILEDESC_LOCK_DESTROY(fdp); 1889 uma_zfree(filedesc0_zone, fdp); 1890} 1891 1892/* 1893 * Share a filedesc structure. 1894 */ 1895struct filedesc * 1896fdshare(struct filedesc *fdp) 1897{ 1898 1899 refcount_acquire(&fdp->fd_refcnt); 1900 return (fdp); 1901} 1902 1903/* 1904 * Unshare a filedesc structure, if necessary by making a copy 1905 */ 1906void 1907fdunshare(struct thread *td) 1908{ 1909 struct filedesc *tmp; 1910 struct proc *p = td->td_proc; 1911 1912 if (p->p_fd->fd_refcnt == 1) 1913 return; 1914 1915 tmp = fdcopy(p->p_fd); 1916 fdescfree(td); 1917 p->p_fd = tmp; 1918} 1919 1920/* 1921 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1922 * this is to ease callers, not catch errors. 1923 */ 1924struct filedesc * 1925fdcopy(struct filedesc *fdp) 1926{ 1927 struct filedesc *newfdp; 1928 struct filedescent *nfde, *ofde; 1929 int i; 1930 1931 MPASS(fdp != NULL); 1932 1933 newfdp = fdinit(fdp, true); 1934 /* copy all passable descriptors (i.e. not kqueue) */ 1935 newfdp->fd_freefile = -1; 1936 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1937 ofde = &fdp->fd_ofiles[i]; 1938 if (ofde->fde_file == NULL || 1939 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 1940 if (newfdp->fd_freefile == -1) 1941 newfdp->fd_freefile = i; 1942 continue; 1943 } 1944 nfde = &newfdp->fd_ofiles[i]; 1945 *nfde = *ofde; 1946 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1947 fhold(nfde->fde_file); 1948 fdused_init(newfdp, i); 1949 newfdp->fd_lastfile = i; 1950 } 1951 if (newfdp->fd_freefile == -1) 1952 newfdp->fd_freefile = i; 1953 newfdp->fd_cmask = fdp->fd_cmask; 1954 FILEDESC_SUNLOCK(fdp); 1955 return (newfdp); 1956} 1957 1958/* 1959 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 1960 * one of processes using it exits) and the table used to be shared. 1961 */ 1962static void 1963fdclearlocks(struct thread *td) 1964{ 1965 struct filedesc *fdp; 1966 struct filedesc_to_leader *fdtol; 1967 struct flock lf; 1968 struct file *fp; 1969 struct proc *p; 1970 struct vnode *vp; 1971 int i; 1972 1973 p = td->td_proc; 1974 fdp = p->p_fd; 1975 fdtol = p->p_fdtol; 1976 MPASS(fdtol != NULL); 1977 1978 FILEDESC_XLOCK(fdp); 1979 KASSERT(fdtol->fdl_refcount > 0, 1980 ("filedesc_to_refcount botch: fdl_refcount=%d", 1981 fdtol->fdl_refcount)); 1982 if (fdtol->fdl_refcount == 1 && 1983 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1984 for (i = 0; i <= fdp->fd_lastfile; i++) { 1985 fp = fdp->fd_ofiles[i].fde_file; 1986 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1987 continue; 1988 fhold(fp); 1989 FILEDESC_XUNLOCK(fdp); 1990 lf.l_whence = SEEK_SET; 1991 lf.l_start = 0; 1992 lf.l_len = 0; 1993 lf.l_type = F_UNLCK; 1994 vp = fp->f_vnode; 1995 (void) VOP_ADVLOCK(vp, 1996 (caddr_t)p->p_leader, F_UNLCK, 1997 &lf, F_POSIX); 1998 FILEDESC_XLOCK(fdp); 1999 fdrop(fp, td); 2000 } 2001 } 2002retry: 2003 if (fdtol->fdl_refcount == 1) { 2004 if (fdp->fd_holdleaderscount > 0 && 2005 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2006 /* 2007 * close() or kern_dup() has cleared a reference 2008 * in a shared file descriptor table. 2009 */ 2010 fdp->fd_holdleaderswakeup = 1; 2011 sx_sleep(&fdp->fd_holdleaderscount, 2012 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2013 goto retry; 2014 } 2015 if (fdtol->fdl_holdcount > 0) { 2016 /* 2017 * Ensure that fdtol->fdl_leader remains 2018 * valid in closef(). 2019 */ 2020 fdtol->fdl_wakeup = 1; 2021 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2022 "fdlhold", 0); 2023 goto retry; 2024 } 2025 } 2026 fdtol->fdl_refcount--; 2027 if (fdtol->fdl_refcount == 0 && 2028 fdtol->fdl_holdcount == 0) { 2029 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2030 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2031 } else 2032 fdtol = NULL; 2033 p->p_fdtol = NULL; 2034 FILEDESC_XUNLOCK(fdp); 2035 if (fdtol != NULL) 2036 free(fdtol, M_FILEDESC_TO_LEADER); 2037} 2038 2039/* 2040 * Release a filedesc structure. 2041 */ 2042void 2043fdescfree(struct thread *td) 2044{ 2045 struct proc *p; 2046 struct filedesc0 *fdp0; 2047 struct filedesc *fdp; 2048 struct freetable *ft, *tft; 2049 struct filedescent *fde; 2050 struct file *fp; 2051 struct vnode *cdir, *jdir, *rdir; 2052 int i; 2053 2054 p = td->td_proc; 2055 fdp = p->p_fd; 2056 MPASS(fdp != NULL); 2057 2058#ifdef RACCT 2059 if (racct_enable) { 2060 PROC_LOCK(p); 2061 racct_set(p, RACCT_NOFILE, 0); 2062 PROC_UNLOCK(p); 2063 } 2064#endif 2065 2066 if (td->td_proc->p_fdtol != NULL) 2067 fdclearlocks(td); 2068 2069 PROC_LOCK(p); 2070 p->p_fd = NULL; 2071 PROC_UNLOCK(p); 2072 2073 if (refcount_release(&fdp->fd_refcnt) == 0) 2074 return; 2075 2076 FILEDESC_XLOCK(fdp); 2077 cdir = fdp->fd_cdir; 2078 fdp->fd_cdir = NULL; 2079 rdir = fdp->fd_rdir; 2080 fdp->fd_rdir = NULL; 2081 jdir = fdp->fd_jdir; 2082 fdp->fd_jdir = NULL; 2083 FILEDESC_XUNLOCK(fdp); 2084 2085 for (i = 0; i <= fdp->fd_lastfile; i++) { 2086 fde = &fdp->fd_ofiles[i]; 2087 fp = fde->fde_file; 2088 if (fp != NULL) { 2089 fdefree_last(fde); 2090 (void) closef(fp, td); 2091 } 2092 } 2093 2094 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2095 free(fdp->fd_map, M_FILEDESC); 2096 if (fdp->fd_nfiles > NDFILE) 2097 free(fdp->fd_files, M_FILEDESC); 2098 2099 fdp0 = (struct filedesc0 *)fdp; 2100 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2101 free(ft->ft_table, M_FILEDESC); 2102 2103 if (cdir != NULL) 2104 vrele(cdir); 2105 if (rdir != NULL) 2106 vrele(rdir); 2107 if (jdir != NULL) 2108 vrele(jdir); 2109 2110 fddrop(fdp); 2111} 2112 2113/* 2114 * For setugid programs, we don't want to people to use that setugidness 2115 * to generate error messages which write to a file which otherwise would 2116 * otherwise be off-limits to the process. We check for filesystems where 2117 * the vnode can change out from under us after execve (like [lin]procfs). 2118 * 2119 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2120 * sufficient. We also don't check for setugidness since we know we are. 2121 */ 2122static bool 2123is_unsafe(struct file *fp) 2124{ 2125 struct vnode *vp; 2126 2127 if (fp->f_type != DTYPE_VNODE) 2128 return (false); 2129 2130 vp = fp->f_vnode; 2131 return ((vp->v_vflag & VV_PROCDEP) != 0); 2132} 2133 2134/* 2135 * Make this setguid thing safe, if at all possible. 2136 */ 2137void 2138fdsetugidsafety(struct thread *td) 2139{ 2140 struct filedesc *fdp; 2141 struct file *fp; 2142 int i; 2143 2144 fdp = td->td_proc->p_fd; 2145 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2146 MPASS(fdp->fd_nfiles >= 3); 2147 for (i = 0; i <= 2; i++) { 2148 fp = fdp->fd_ofiles[i].fde_file; 2149 if (fp != NULL && is_unsafe(fp)) { 2150 FILEDESC_XLOCK(fdp); 2151 knote_fdclose(td, i); 2152 /* 2153 * NULL-out descriptor prior to close to avoid 2154 * a race while close blocks. 2155 */ 2156 fdfree(fdp, i); 2157 FILEDESC_XUNLOCK(fdp); 2158 (void) closef(fp, td); 2159 } 2160 } 2161} 2162 2163/* 2164 * If a specific file object occupies a specific file descriptor, close the 2165 * file descriptor entry and drop a reference on the file object. This is a 2166 * convenience function to handle a subsequent error in a function that calls 2167 * falloc() that handles the race that another thread might have closed the 2168 * file descriptor out from under the thread creating the file object. 2169 */ 2170void 2171fdclose(struct thread *td, struct file *fp, int idx) 2172{ 2173 struct filedesc *fdp = td->td_proc->p_fd; 2174 2175 FILEDESC_XLOCK(fdp); 2176 if (fdp->fd_ofiles[idx].fde_file == fp) { 2177 fdfree(fdp, idx); 2178 FILEDESC_XUNLOCK(fdp); 2179 fdrop(fp, td); 2180 } else 2181 FILEDESC_XUNLOCK(fdp); 2182} 2183 2184/* 2185 * Close any files on exec? 2186 */ 2187void 2188fdcloseexec(struct thread *td) 2189{ 2190 struct filedesc *fdp; 2191 struct filedescent *fde; 2192 struct file *fp; 2193 int i; 2194 2195 fdp = td->td_proc->p_fd; 2196 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2197 for (i = 0; i <= fdp->fd_lastfile; i++) { 2198 fde = &fdp->fd_ofiles[i]; 2199 fp = fde->fde_file; 2200 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2201 (fde->fde_flags & UF_EXCLOSE))) { 2202 FILEDESC_XLOCK(fdp); 2203 fdfree(fdp, i); 2204 (void) closefp(fdp, i, fp, td, 0); 2205 /* closefp() drops the FILEDESC lock. */ 2206 } 2207 } 2208} 2209 2210/* 2211 * It is unsafe for set[ug]id processes to be started with file 2212 * descriptors 0..2 closed, as these descriptors are given implicit 2213 * significance in the Standard C library. fdcheckstd() will create a 2214 * descriptor referencing /dev/null for each of stdin, stdout, and 2215 * stderr that is not already open. 2216 */ 2217int 2218fdcheckstd(struct thread *td) 2219{ 2220 struct filedesc *fdp; 2221 register_t save; 2222 int i, error, devnull; 2223 2224 fdp = td->td_proc->p_fd; 2225 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2226 MPASS(fdp->fd_nfiles >= 3); 2227 devnull = -1; 2228 for (i = 0; i <= 2; i++) { 2229 if (fdp->fd_ofiles[i].fde_file != NULL) 2230 continue; 2231 2232 save = td->td_retval[0]; 2233 if (devnull != -1) { 2234 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2235 } else { 2236 error = kern_openat(td, AT_FDCWD, "/dev/null", 2237 UIO_SYSSPACE, O_RDWR, 0); 2238 if (error == 0) { 2239 devnull = td->td_retval[0]; 2240 KASSERT(devnull == i, ("we didn't get our fd")); 2241 } 2242 } 2243 td->td_retval[0] = save; 2244 if (error != 0) 2245 return (error); 2246 } 2247 return (0); 2248} 2249 2250/* 2251 * Internal form of close. Decrement reference count on file structure. 2252 * Note: td may be NULL when closing a file that was being passed in a 2253 * message. 2254 * 2255 * XXXRW: Giant is not required for the caller, but often will be held; this 2256 * makes it moderately likely the Giant will be recursed in the VFS case. 2257 */ 2258int 2259closef(struct file *fp, struct thread *td) 2260{ 2261 struct vnode *vp; 2262 struct flock lf; 2263 struct filedesc_to_leader *fdtol; 2264 struct filedesc *fdp; 2265 2266 /* 2267 * POSIX record locking dictates that any close releases ALL 2268 * locks owned by this process. This is handled by setting 2269 * a flag in the unlock to free ONLY locks obeying POSIX 2270 * semantics, and not to free BSD-style file locks. 2271 * If the descriptor was in a message, POSIX-style locks 2272 * aren't passed with the descriptor, and the thread pointer 2273 * will be NULL. Callers should be careful only to pass a 2274 * NULL thread pointer when there really is no owning 2275 * context that might have locks, or the locks will be 2276 * leaked. 2277 */ 2278 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2279 vp = fp->f_vnode; 2280 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2281 lf.l_whence = SEEK_SET; 2282 lf.l_start = 0; 2283 lf.l_len = 0; 2284 lf.l_type = F_UNLCK; 2285 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2286 F_UNLCK, &lf, F_POSIX); 2287 } 2288 fdtol = td->td_proc->p_fdtol; 2289 if (fdtol != NULL) { 2290 /* 2291 * Handle special case where file descriptor table is 2292 * shared between multiple process leaders. 2293 */ 2294 fdp = td->td_proc->p_fd; 2295 FILEDESC_XLOCK(fdp); 2296 for (fdtol = fdtol->fdl_next; 2297 fdtol != td->td_proc->p_fdtol; 2298 fdtol = fdtol->fdl_next) { 2299 if ((fdtol->fdl_leader->p_flag & 2300 P_ADVLOCK) == 0) 2301 continue; 2302 fdtol->fdl_holdcount++; 2303 FILEDESC_XUNLOCK(fdp); 2304 lf.l_whence = SEEK_SET; 2305 lf.l_start = 0; 2306 lf.l_len = 0; 2307 lf.l_type = F_UNLCK; 2308 vp = fp->f_vnode; 2309 (void) VOP_ADVLOCK(vp, 2310 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2311 F_POSIX); 2312 FILEDESC_XLOCK(fdp); 2313 fdtol->fdl_holdcount--; 2314 if (fdtol->fdl_holdcount == 0 && 2315 fdtol->fdl_wakeup != 0) { 2316 fdtol->fdl_wakeup = 0; 2317 wakeup(fdtol); 2318 } 2319 } 2320 FILEDESC_XUNLOCK(fdp); 2321 } 2322 } 2323 return (fdrop(fp, td)); 2324} 2325 2326/* 2327 * Initialize the file pointer with the specified properties. 2328 * 2329 * The ops are set with release semantics to be certain that the flags, type, 2330 * and data are visible when ops is. This is to prevent ops methods from being 2331 * called with bad data. 2332 */ 2333void 2334finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2335{ 2336 fp->f_data = data; 2337 fp->f_flag = flag; 2338 fp->f_type = type; 2339 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2340} 2341 2342int 2343fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2344 struct file **fpp, seq_t *seqp) 2345{ 2346#ifdef CAPABILITIES 2347 struct filedescent *fde; 2348#endif 2349 struct fdescenttbl *fdt; 2350 struct file *fp; 2351 u_int count; 2352#ifdef CAPABILITIES 2353 seq_t seq; 2354 cap_rights_t haverights; 2355 int error; 2356#endif 2357 2358 fdt = fdp->fd_files; 2359 if ((u_int)fd >= fdt->fdt_nfiles) 2360 return (EBADF); 2361 /* 2362 * Fetch the descriptor locklessly. We avoid fdrop() races by 2363 * never raising a refcount above 0. To accomplish this we have 2364 * to use a cmpset loop rather than an atomic_add. The descriptor 2365 * must be re-verified once we acquire a reference to be certain 2366 * that the identity is still correct and we did not lose a race 2367 * due to preemption. 2368 */ 2369 for (;;) { 2370#ifdef CAPABILITIES 2371 seq = seq_read(fd_seq(fdt, fd)); 2372 fde = &fdt->fdt_ofiles[fd]; 2373 haverights = *cap_rights_fde(fde); 2374 fp = fde->fde_file; 2375 if (!seq_consistent(fd_seq(fdt, fd), seq)) { 2376 cpu_spinwait(); 2377 continue; 2378 } 2379#else 2380 fp = fdt->fdt_ofiles[fd].fde_file; 2381#endif 2382 if (fp == NULL) 2383 return (EBADF); 2384#ifdef CAPABILITIES 2385 error = cap_check(&haverights, needrightsp); 2386 if (error != 0) 2387 return (error); 2388#endif 2389 retry: 2390 count = fp->f_count; 2391 if (count == 0) { 2392 /* 2393 * Force a reload. Other thread could reallocate the 2394 * table before this fd was closed, so it possible that 2395 * there is a stale fp pointer in cached version. 2396 */ 2397 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); 2398 continue; 2399 } 2400 /* 2401 * Use an acquire barrier to force re-reading of fdt so it is 2402 * refreshed for verification. 2403 */ 2404 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0) 2405 goto retry; 2406 fdt = fdp->fd_files; 2407#ifdef CAPABILITIES 2408 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2409#else 2410 if (fp == fdt->fdt_ofiles[fd].fde_file) 2411#endif 2412 break; 2413 fdrop(fp, curthread); 2414 } 2415 *fpp = fp; 2416 if (seqp != NULL) { 2417#ifdef CAPABILITIES 2418 *seqp = seq; 2419#endif 2420 } 2421 return (0); 2422} 2423 2424/* 2425 * Extract the file pointer associated with the specified descriptor for the 2426 * current user process. 2427 * 2428 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2429 * returned. 2430 * 2431 * File's rights will be checked against the capability rights mask. 2432 * 2433 * If an error occured the non-zero error is returned and *fpp is set to 2434 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2435 * responsible for fdrop(). 2436 */ 2437static __inline int 2438_fget(struct thread *td, int fd, struct file **fpp, int flags, 2439 cap_rights_t *needrightsp, seq_t *seqp) 2440{ 2441 struct filedesc *fdp; 2442 struct file *fp; 2443 int error; 2444 2445 *fpp = NULL; 2446 fdp = td->td_proc->p_fd; 2447 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp); 2448 if (error != 0) 2449 return (error); 2450 if (fp->f_ops == &badfileops) { 2451 fdrop(fp, td); 2452 return (EBADF); 2453 } 2454 2455 /* 2456 * FREAD and FWRITE failure return EBADF as per POSIX. 2457 */ 2458 error = 0; 2459 switch (flags) { 2460 case FREAD: 2461 case FWRITE: 2462 if ((fp->f_flag & flags) == 0) 2463 error = EBADF; 2464 break; 2465 case FEXEC: 2466 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2467 ((fp->f_flag & FWRITE) != 0)) 2468 error = EBADF; 2469 break; 2470 case 0: 2471 break; 2472 default: 2473 KASSERT(0, ("wrong flags")); 2474 } 2475 2476 if (error != 0) { 2477 fdrop(fp, td); 2478 return (error); 2479 } 2480 2481 *fpp = fp; 2482 return (0); 2483} 2484 2485int 2486fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2487{ 2488 2489 return (_fget(td, fd, fpp, 0, rightsp, NULL)); 2490} 2491 2492int 2493fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2494 struct file **fpp) 2495{ 2496 int error; 2497#ifndef CAPABILITIES 2498 error = _fget(td, fd, fpp, 0, rightsp, NULL); 2499 if (maxprotp != NULL) 2500 *maxprotp = VM_PROT_ALL; 2501#else 2502 struct filedesc *fdp = td->td_proc->p_fd; 2503 seq_t seq; 2504 2505 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 2506 for (;;) { 2507 error = _fget(td, fd, fpp, 0, rightsp, &seq); 2508 if (error != 0) 2509 return (error); 2510 /* 2511 * If requested, convert capability rights to access flags. 2512 */ 2513 if (maxprotp != NULL) 2514 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); 2515 if (!fd_modified(fdp, fd, seq)) 2516 break; 2517 fdrop(*fpp, td); 2518 } 2519#endif 2520 return (error); 2521} 2522 2523int 2524fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2525{ 2526 2527 return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2528} 2529 2530int 2531fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2532{ 2533 2534 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2535} 2536 2537int 2538fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 2539 struct file **fpp) 2540{ 2541 struct filedesc *fdp = td->td_proc->p_fd; 2542#ifndef CAPABILITIES 2543 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); 2544#else 2545 int error; 2546 seq_t seq; 2547 2548 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 2549 for (;;) { 2550 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); 2551 if (error != 0) 2552 return (error); 2553 error = cap_fcntl_check(fdp, fd, needfcntl); 2554 if (!fd_modified(fdp, fd, seq)) 2555 break; 2556 fdrop(*fpp, td); 2557 } 2558 if (error != 0) { 2559 fdrop(*fpp, td); 2560 *fpp = NULL; 2561 } 2562 return (error); 2563#endif 2564} 2565 2566/* 2567 * Like fget() but loads the underlying vnode, or returns an error if the 2568 * descriptor does not represent a vnode. Note that pipes use vnodes but 2569 * never have VM objects. The returned vnode will be vref()'d. 2570 * 2571 * XXX: what about the unused flags ? 2572 */ 2573static __inline int 2574_fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2575 struct vnode **vpp) 2576{ 2577 struct file *fp; 2578 int error; 2579 2580 *vpp = NULL; 2581 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2582 if (error != 0) 2583 return (error); 2584 if (fp->f_vnode == NULL) { 2585 error = EINVAL; 2586 } else { 2587 *vpp = fp->f_vnode; 2588 vref(*vpp); 2589 } 2590 fdrop(fp, td); 2591 2592 return (error); 2593} 2594 2595int 2596fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2597{ 2598 2599 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2600} 2601 2602int 2603fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2604 struct filecaps *havecaps, struct vnode **vpp) 2605{ 2606 struct filedesc *fdp; 2607 struct file *fp; 2608#ifdef CAPABILITIES 2609 int error; 2610#endif 2611 2612 fdp = td->td_proc->p_fd; 2613 fp = fget_locked(fdp, fd); 2614 if (fp == NULL || fp->f_ops == &badfileops) 2615 return (EBADF); 2616 2617#ifdef CAPABILITIES 2618 if (needrightsp != NULL) { 2619 error = cap_check(cap_rights(fdp, fd), needrightsp); 2620 if (error != 0) 2621 return (error); 2622 } 2623#endif 2624 2625 if (fp->f_vnode == NULL) 2626 return (EINVAL); 2627 2628 *vpp = fp->f_vnode; 2629 vref(*vpp); 2630 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2631 2632 return (0); 2633} 2634 2635int 2636fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2637{ 2638 2639 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2640} 2641 2642int 2643fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2644{ 2645 2646 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2647} 2648 2649#ifdef notyet 2650int 2651fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2652 struct vnode **vpp) 2653{ 2654 2655 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2656} 2657#endif 2658 2659/* 2660 * Like fget() but loads the underlying socket, or returns an error if the 2661 * descriptor does not represent a socket. 2662 * 2663 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2664 * in the future. 2665 * 2666 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2667 * on their file descriptor reference to prevent the socket from being free'd 2668 * during use. 2669 */ 2670int 2671fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2672 u_int *fflagp) 2673{ 2674 struct file *fp; 2675 int error; 2676 2677 *spp = NULL; 2678 if (fflagp != NULL) 2679 *fflagp = 0; 2680 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2681 return (error); 2682 if (fp->f_type != DTYPE_SOCKET) { 2683 error = ENOTSOCK; 2684 } else { 2685 *spp = fp->f_data; 2686 if (fflagp) 2687 *fflagp = fp->f_flag; 2688 SOCK_LOCK(*spp); 2689 soref(*spp); 2690 SOCK_UNLOCK(*spp); 2691 } 2692 fdrop(fp, td); 2693 2694 return (error); 2695} 2696 2697/* 2698 * Drop the reference count on the socket and XXX release the SX lock in the 2699 * future. The last reference closes the socket. 2700 * 2701 * Note: fputsock() is deprecated, see comment for fgetsock(). 2702 */ 2703void 2704fputsock(struct socket *so) 2705{ 2706 2707 ACCEPT_LOCK(); 2708 SOCK_LOCK(so); 2709 CURVNET_SET(so->so_vnet); 2710 sorele(so); 2711 CURVNET_RESTORE(); 2712} 2713 2714/* 2715 * Handle the last reference to a file being closed. 2716 */ 2717int 2718_fdrop(struct file *fp, struct thread *td) 2719{ 2720 int error; 2721 2722 if (fp->f_count != 0) 2723 panic("fdrop: count %d", fp->f_count); 2724 error = fo_close(fp, td); 2725 atomic_subtract_int(&openfiles, 1); 2726 crfree(fp->f_cred); 2727 free(fp->f_advice, M_FADVISE); 2728 uma_zfree(file_zone, fp); 2729 2730 return (error); 2731} 2732 2733/* 2734 * Apply an advisory lock on a file descriptor. 2735 * 2736 * Just attempt to get a record lock of the requested type on the entire file 2737 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2738 */ 2739#ifndef _SYS_SYSPROTO_H_ 2740struct flock_args { 2741 int fd; 2742 int how; 2743}; 2744#endif 2745/* ARGSUSED */ 2746int 2747sys_flock(struct thread *td, struct flock_args *uap) 2748{ 2749 struct file *fp; 2750 struct vnode *vp; 2751 struct flock lf; 2752 cap_rights_t rights; 2753 int error; 2754 2755 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2756 if (error != 0) 2757 return (error); 2758 if (fp->f_type != DTYPE_VNODE) { 2759 fdrop(fp, td); 2760 return (EOPNOTSUPP); 2761 } 2762 2763 vp = fp->f_vnode; 2764 lf.l_whence = SEEK_SET; 2765 lf.l_start = 0; 2766 lf.l_len = 0; 2767 if (uap->how & LOCK_UN) { 2768 lf.l_type = F_UNLCK; 2769 atomic_clear_int(&fp->f_flag, FHASLOCK); 2770 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2771 goto done2; 2772 } 2773 if (uap->how & LOCK_EX) 2774 lf.l_type = F_WRLCK; 2775 else if (uap->how & LOCK_SH) 2776 lf.l_type = F_RDLCK; 2777 else { 2778 error = EBADF; 2779 goto done2; 2780 } 2781 atomic_set_int(&fp->f_flag, FHASLOCK); 2782 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2783 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2784done2: 2785 fdrop(fp, td); 2786 return (error); 2787} 2788/* 2789 * Duplicate the specified descriptor to a free descriptor. 2790 */ 2791int 2792dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2793 int openerror, int *indxp) 2794{ 2795 struct filedescent *newfde, *oldfde; 2796 struct file *fp; 2797 int error, indx; 2798 2799 KASSERT(openerror == ENODEV || openerror == ENXIO, 2800 ("unexpected error %d in %s", openerror, __func__)); 2801 2802 /* 2803 * If the to-be-dup'd fd number is greater than the allowed number 2804 * of file descriptors, or the fd to be dup'd has already been 2805 * closed, then reject. 2806 */ 2807 FILEDESC_XLOCK(fdp); 2808 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2809 FILEDESC_XUNLOCK(fdp); 2810 return (EBADF); 2811 } 2812 2813 error = fdalloc(td, 0, &indx); 2814 if (error != 0) { 2815 FILEDESC_XUNLOCK(fdp); 2816 return (error); 2817 } 2818 2819 /* 2820 * There are two cases of interest here. 2821 * 2822 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2823 * 2824 * For ENXIO steal away the file structure from (dfd) and store it in 2825 * (indx). (dfd) is effectively closed by this operation. 2826 */ 2827 switch (openerror) { 2828 case ENODEV: 2829 /* 2830 * Check that the mode the file is being opened for is a 2831 * subset of the mode of the existing descriptor. 2832 */ 2833 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2834 fdunused(fdp, indx); 2835 FILEDESC_XUNLOCK(fdp); 2836 return (EACCES); 2837 } 2838 fhold(fp); 2839 newfde = &fdp->fd_ofiles[indx]; 2840 oldfde = &fdp->fd_ofiles[dfd]; 2841#ifdef CAPABILITIES 2842 seq_write_begin(&newfde->fde_seq); 2843#endif 2844 memcpy(newfde, oldfde, fde_change_size); 2845 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 2846#ifdef CAPABILITIES 2847 seq_write_end(&newfde->fde_seq); 2848#endif 2849 break; 2850 case ENXIO: 2851 /* 2852 * Steal away the file pointer from dfd and stuff it into indx. 2853 */ 2854 newfde = &fdp->fd_ofiles[indx]; 2855 oldfde = &fdp->fd_ofiles[dfd]; 2856#ifdef CAPABILITIES 2857 seq_write_begin(&newfde->fde_seq); 2858#endif 2859 memcpy(newfde, oldfde, fde_change_size); 2860 oldfde->fde_file = NULL; 2861 fdunused(fdp, dfd); 2862#ifdef CAPABILITIES 2863 seq_write_end(&newfde->fde_seq); 2864#endif 2865 break; 2866 } 2867 FILEDESC_XUNLOCK(fdp); 2868 *indxp = indx; 2869 return (0); 2870} 2871 2872/* 2873 * Scan all active processes and prisons to see if any of them have a current 2874 * or root directory of `olddp'. If so, replace them with the new mount point. 2875 */ 2876void 2877mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2878{ 2879 struct filedesc *fdp; 2880 struct prison *pr; 2881 struct proc *p; 2882 int nrele; 2883 2884 if (vrefcnt(olddp) == 1) 2885 return; 2886 nrele = 0; 2887 sx_slock(&allproc_lock); 2888 FOREACH_PROC_IN_SYSTEM(p) { 2889 PROC_LOCK(p); 2890 fdp = fdhold(p); 2891 PROC_UNLOCK(p); 2892 if (fdp == NULL) 2893 continue; 2894 FILEDESC_XLOCK(fdp); 2895 if (fdp->fd_cdir == olddp) { 2896 vref(newdp); 2897 fdp->fd_cdir = newdp; 2898 nrele++; 2899 } 2900 if (fdp->fd_rdir == olddp) { 2901 vref(newdp); 2902 fdp->fd_rdir = newdp; 2903 nrele++; 2904 } 2905 if (fdp->fd_jdir == olddp) { 2906 vref(newdp); 2907 fdp->fd_jdir = newdp; 2908 nrele++; 2909 } 2910 FILEDESC_XUNLOCK(fdp); 2911 fddrop(fdp); 2912 } 2913 sx_sunlock(&allproc_lock); 2914 if (rootvnode == olddp) { 2915 vref(newdp); 2916 rootvnode = newdp; 2917 nrele++; 2918 } 2919 mtx_lock(&prison0.pr_mtx); 2920 if (prison0.pr_root == olddp) { 2921 vref(newdp); 2922 prison0.pr_root = newdp; 2923 nrele++; 2924 } 2925 mtx_unlock(&prison0.pr_mtx); 2926 sx_slock(&allprison_lock); 2927 TAILQ_FOREACH(pr, &allprison, pr_list) { 2928 mtx_lock(&pr->pr_mtx); 2929 if (pr->pr_root == olddp) { 2930 vref(newdp); 2931 pr->pr_root = newdp; 2932 nrele++; 2933 } 2934 mtx_unlock(&pr->pr_mtx); 2935 } 2936 sx_sunlock(&allprison_lock); 2937 while (nrele--) 2938 vrele(olddp); 2939} 2940 2941struct filedesc_to_leader * 2942filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2943{ 2944 struct filedesc_to_leader *fdtol; 2945 2946 fdtol = malloc(sizeof(struct filedesc_to_leader), 2947 M_FILEDESC_TO_LEADER, M_WAITOK); 2948 fdtol->fdl_refcount = 1; 2949 fdtol->fdl_holdcount = 0; 2950 fdtol->fdl_wakeup = 0; 2951 fdtol->fdl_leader = leader; 2952 if (old != NULL) { 2953 FILEDESC_XLOCK(fdp); 2954 fdtol->fdl_next = old->fdl_next; 2955 fdtol->fdl_prev = old; 2956 old->fdl_next = fdtol; 2957 fdtol->fdl_next->fdl_prev = fdtol; 2958 FILEDESC_XUNLOCK(fdp); 2959 } else { 2960 fdtol->fdl_next = fdtol; 2961 fdtol->fdl_prev = fdtol; 2962 } 2963 return (fdtol); 2964} 2965 2966/* 2967 * Get file structures globally. 2968 */ 2969static int 2970sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2971{ 2972 struct xfile xf; 2973 struct filedesc *fdp; 2974 struct file *fp; 2975 struct proc *p; 2976 int error, n; 2977 2978 error = sysctl_wire_old_buffer(req, 0); 2979 if (error != 0) 2980 return (error); 2981 if (req->oldptr == NULL) { 2982 n = 0; 2983 sx_slock(&allproc_lock); 2984 FOREACH_PROC_IN_SYSTEM(p) { 2985 PROC_LOCK(p); 2986 if (p->p_state == PRS_NEW) { 2987 PROC_UNLOCK(p); 2988 continue; 2989 } 2990 fdp = fdhold(p); 2991 PROC_UNLOCK(p); 2992 if (fdp == NULL) 2993 continue; 2994 /* overestimates sparse tables. */ 2995 if (fdp->fd_lastfile > 0) 2996 n += fdp->fd_lastfile; 2997 fddrop(fdp); 2998 } 2999 sx_sunlock(&allproc_lock); 3000 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 3001 } 3002 error = 0; 3003 bzero(&xf, sizeof(xf)); 3004 xf.xf_size = sizeof(xf); 3005 sx_slock(&allproc_lock); 3006 FOREACH_PROC_IN_SYSTEM(p) { 3007 PROC_LOCK(p); 3008 if (p->p_state == PRS_NEW) { 3009 PROC_UNLOCK(p); 3010 continue; 3011 } 3012 if (p_cansee(req->td, p) != 0) { 3013 PROC_UNLOCK(p); 3014 continue; 3015 } 3016 xf.xf_pid = p->p_pid; 3017 xf.xf_uid = p->p_ucred->cr_uid; 3018 fdp = fdhold(p); 3019 PROC_UNLOCK(p); 3020 if (fdp == NULL) 3021 continue; 3022 FILEDESC_SLOCK(fdp); 3023 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 3024 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3025 continue; 3026 xf.xf_fd = n; 3027 xf.xf_file = fp; 3028 xf.xf_data = fp->f_data; 3029 xf.xf_vnode = fp->f_vnode; 3030 xf.xf_type = fp->f_type; 3031 xf.xf_count = fp->f_count; 3032 xf.xf_msgcount = 0; 3033 xf.xf_offset = foffset_get(fp); 3034 xf.xf_flag = fp->f_flag; 3035 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 3036 if (error) 3037 break; 3038 } 3039 FILEDESC_SUNLOCK(fdp); 3040 fddrop(fdp); 3041 if (error) 3042 break; 3043 } 3044 sx_sunlock(&allproc_lock); 3045 return (error); 3046} 3047 3048SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 3049 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 3050 3051#ifdef KINFO_FILE_SIZE 3052CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3053#endif 3054 3055static int 3056xlate_fflags(int fflags) 3057{ 3058 static const struct { 3059 int fflag; 3060 int kf_fflag; 3061 } fflags_table[] = { 3062 { FAPPEND, KF_FLAG_APPEND }, 3063 { FASYNC, KF_FLAG_ASYNC }, 3064 { FFSYNC, KF_FLAG_FSYNC }, 3065 { FHASLOCK, KF_FLAG_HASLOCK }, 3066 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3067 { FREAD, KF_FLAG_READ }, 3068 { FWRITE, KF_FLAG_WRITE }, 3069 { O_CREAT, KF_FLAG_CREAT }, 3070 { O_DIRECT, KF_FLAG_DIRECT }, 3071 { O_EXCL, KF_FLAG_EXCL }, 3072 { O_EXEC, KF_FLAG_EXEC }, 3073 { O_EXLOCK, KF_FLAG_EXLOCK }, 3074 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3075 { O_SHLOCK, KF_FLAG_SHLOCK }, 3076 { O_TRUNC, KF_FLAG_TRUNC } 3077 }; 3078 unsigned int i; 3079 int kflags; 3080 3081 kflags = 0; 3082 for (i = 0; i < nitems(fflags_table); i++) 3083 if (fflags & fflags_table[i].fflag) 3084 kflags |= fflags_table[i].kf_fflag; 3085 return (kflags); 3086} 3087 3088/* Trim unused data from kf_path by truncating the structure size. */ 3089static void 3090pack_kinfo(struct kinfo_file *kif) 3091{ 3092 3093 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3094 strlen(kif->kf_path) + 1; 3095 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3096} 3097 3098static void 3099export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3100 struct kinfo_file *kif, struct filedesc *fdp) 3101{ 3102 int error; 3103 3104 bzero(kif, sizeof(*kif)); 3105 3106 /* Set a default type to allow for empty fill_kinfo() methods. */ 3107 kif->kf_type = KF_TYPE_UNKNOWN; 3108 kif->kf_flags = xlate_fflags(fp->f_flag); 3109 if (rightsp != NULL) 3110 kif->kf_cap_rights = *rightsp; 3111 else 3112 cap_rights_init(&kif->kf_cap_rights); 3113 kif->kf_fd = fd; 3114 kif->kf_ref_count = fp->f_count; 3115 kif->kf_offset = foffset_get(fp); 3116 3117 /* 3118 * This may drop the filedesc lock, so the 'fp' cannot be 3119 * accessed after this call. 3120 */ 3121 error = fo_fill_kinfo(fp, kif, fdp); 3122 if (error == 0) 3123 kif->kf_status |= KF_ATTR_VALID; 3124 pack_kinfo(kif); 3125} 3126 3127static void 3128export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3129 struct kinfo_file *kif) 3130{ 3131 int error; 3132 3133 bzero(kif, sizeof(*kif)); 3134 3135 kif->kf_type = KF_TYPE_VNODE; 3136 error = vn_fill_kinfo_vnode(vp, kif); 3137 if (error == 0) 3138 kif->kf_status |= KF_ATTR_VALID; 3139 kif->kf_flags = xlate_fflags(fflags); 3140 cap_rights_init(&kif->kf_cap_rights); 3141 kif->kf_fd = fd; 3142 kif->kf_ref_count = -1; 3143 kif->kf_offset = -1; 3144 pack_kinfo(kif); 3145 vrele(vp); 3146} 3147 3148struct export_fd_buf { 3149 struct filedesc *fdp; 3150 struct sbuf *sb; 3151 ssize_t remainder; 3152 struct kinfo_file kif; 3153}; 3154 3155static int 3156export_kinfo_to_sb(struct export_fd_buf *efbuf) 3157{ 3158 struct kinfo_file *kif; 3159 3160 kif = &efbuf->kif; 3161 if (efbuf->remainder != -1) { 3162 if (efbuf->remainder < kif->kf_structsize) { 3163 /* Terminate export. */ 3164 efbuf->remainder = 0; 3165 return (0); 3166 } 3167 efbuf->remainder -= kif->kf_structsize; 3168 } 3169 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3170} 3171 3172static int 3173export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3174 struct export_fd_buf *efbuf) 3175{ 3176 int error; 3177 3178 if (efbuf->remainder == 0) 3179 return (0); 3180 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp); 3181 FILEDESC_SUNLOCK(efbuf->fdp); 3182 error = export_kinfo_to_sb(efbuf); 3183 FILEDESC_SLOCK(efbuf->fdp); 3184 return (error); 3185} 3186 3187static int 3188export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3189 struct export_fd_buf *efbuf) 3190{ 3191 int error; 3192 3193 if (efbuf->remainder == 0) 3194 return (0); 3195 if (efbuf->fdp != NULL) 3196 FILEDESC_SUNLOCK(efbuf->fdp); 3197 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif); 3198 error = export_kinfo_to_sb(efbuf); 3199 if (efbuf->fdp != NULL) 3200 FILEDESC_SLOCK(efbuf->fdp); 3201 return (error); 3202} 3203 3204/* 3205 * Store a process file descriptor information to sbuf. 3206 * 3207 * Takes a locked proc as argument, and returns with the proc unlocked. 3208 */ 3209int 3210kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3211{ 3212 struct file *fp; 3213 struct filedesc *fdp; 3214 struct export_fd_buf *efbuf; 3215 struct vnode *cttyvp, *textvp, *tracevp; 3216 int error, i; 3217 cap_rights_t rights; 3218 3219 PROC_LOCK_ASSERT(p, MA_OWNED); 3220 3221 /* ktrace vnode */ 3222 tracevp = p->p_tracevp; 3223 if (tracevp != NULL) 3224 vref(tracevp); 3225 /* text vnode */ 3226 textvp = p->p_textvp; 3227 if (textvp != NULL) 3228 vref(textvp); 3229 /* Controlling tty. */ 3230 cttyvp = NULL; 3231 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3232 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3233 if (cttyvp != NULL) 3234 vref(cttyvp); 3235 } 3236 fdp = fdhold(p); 3237 PROC_UNLOCK(p); 3238 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3239 efbuf->fdp = NULL; 3240 efbuf->sb = sb; 3241 efbuf->remainder = maxlen; 3242 if (tracevp != NULL) 3243 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3244 efbuf); 3245 if (textvp != NULL) 3246 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3247 if (cttyvp != NULL) 3248 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3249 efbuf); 3250 error = 0; 3251 if (fdp == NULL) 3252 goto fail; 3253 efbuf->fdp = fdp; 3254 FILEDESC_SLOCK(fdp); 3255 /* working directory */ 3256 if (fdp->fd_cdir != NULL) { 3257 vref(fdp->fd_cdir); 3258 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3259 } 3260 /* root directory */ 3261 if (fdp->fd_rdir != NULL) { 3262 vref(fdp->fd_rdir); 3263 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3264 } 3265 /* jail directory */ 3266 if (fdp->fd_jdir != NULL) { 3267 vref(fdp->fd_jdir); 3268 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3269 } 3270 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3271 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3272 continue; 3273#ifdef CAPABILITIES 3274 rights = *cap_rights(fdp, i); 3275#else /* !CAPABILITIES */ 3276 cap_rights_init(&rights); 3277#endif 3278 /* 3279 * Create sysctl entry. It is OK to drop the filedesc 3280 * lock inside of export_file_to_sb() as we will 3281 * re-validate and re-evaluate its properties when the 3282 * loop continues. 3283 */ 3284 error = export_file_to_sb(fp, i, &rights, efbuf); 3285 if (error != 0 || efbuf->remainder == 0) 3286 break; 3287 } 3288 FILEDESC_SUNLOCK(fdp); 3289 fddrop(fdp); 3290fail: 3291 free(efbuf, M_TEMP); 3292 return (error); 3293} 3294 3295#define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3296 3297/* 3298 * Get per-process file descriptors for use by procstat(1), et al. 3299 */ 3300static int 3301sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3302{ 3303 struct sbuf sb; 3304 struct proc *p; 3305 ssize_t maxlen; 3306 int error, error2, *name; 3307 3308 name = (int *)arg1; 3309 3310 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3311 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3312 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3313 if (error != 0) { 3314 sbuf_delete(&sb); 3315 return (error); 3316 } 3317 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3318 error = kern_proc_filedesc_out(p, &sb, maxlen); 3319 error2 = sbuf_finish(&sb); 3320 sbuf_delete(&sb); 3321 return (error != 0 ? error : error2); 3322} 3323 3324#ifdef KINFO_OFILE_SIZE 3325CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3326#endif 3327 3328#ifdef COMPAT_FREEBSD7 3329static void 3330kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3331{ 3332 3333 okif->kf_structsize = sizeof(*okif); 3334 okif->kf_type = kif->kf_type; 3335 okif->kf_fd = kif->kf_fd; 3336 okif->kf_ref_count = kif->kf_ref_count; 3337 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3338 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3339 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3340 okif->kf_offset = kif->kf_offset; 3341 okif->kf_vnode_type = kif->kf_vnode_type; 3342 okif->kf_sock_domain = kif->kf_sock_domain; 3343 okif->kf_sock_type = kif->kf_sock_type; 3344 okif->kf_sock_protocol = kif->kf_sock_protocol; 3345 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3346 okif->kf_sa_local = kif->kf_sa_local; 3347 okif->kf_sa_peer = kif->kf_sa_peer; 3348} 3349 3350static int 3351export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3352 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3353{ 3354 int error; 3355 3356 vref(vp); 3357 FILEDESC_SUNLOCK(fdp); 3358 export_vnode_to_kinfo(vp, type, 0, kif); 3359 kinfo_to_okinfo(kif, okif); 3360 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3361 FILEDESC_SLOCK(fdp); 3362 return (error); 3363} 3364 3365/* 3366 * Get per-process file descriptors for use by procstat(1), et al. 3367 */ 3368static int 3369sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3370{ 3371 struct kinfo_ofile *okif; 3372 struct kinfo_file *kif; 3373 struct filedesc *fdp; 3374 int error, i, *name; 3375 struct file *fp; 3376 struct proc *p; 3377 3378 name = (int *)arg1; 3379 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3380 if (error != 0) 3381 return (error); 3382 fdp = fdhold(p); 3383 PROC_UNLOCK(p); 3384 if (fdp == NULL) 3385 return (ENOENT); 3386 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3387 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3388 FILEDESC_SLOCK(fdp); 3389 if (fdp->fd_cdir != NULL) 3390 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3391 okif, fdp, req); 3392 if (fdp->fd_rdir != NULL) 3393 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3394 okif, fdp, req); 3395 if (fdp->fd_jdir != NULL) 3396 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3397 okif, fdp, req); 3398 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3399 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3400 continue; 3401 export_file_to_kinfo(fp, i, NULL, kif, fdp); 3402 FILEDESC_SUNLOCK(fdp); 3403 kinfo_to_okinfo(kif, okif); 3404 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3405 FILEDESC_SLOCK(fdp); 3406 if (error) 3407 break; 3408 } 3409 FILEDESC_SUNLOCK(fdp); 3410 fddrop(fdp); 3411 free(kif, M_TEMP); 3412 free(okif, M_TEMP); 3413 return (0); 3414} 3415 3416static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3417 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3418 "Process ofiledesc entries"); 3419#endif /* COMPAT_FREEBSD7 */ 3420 3421int 3422vntype_to_kinfo(int vtype) 3423{ 3424 struct { 3425 int vtype; 3426 int kf_vtype; 3427 } vtypes_table[] = { 3428 { VBAD, KF_VTYPE_VBAD }, 3429 { VBLK, KF_VTYPE_VBLK }, 3430 { VCHR, KF_VTYPE_VCHR }, 3431 { VDIR, KF_VTYPE_VDIR }, 3432 { VFIFO, KF_VTYPE_VFIFO }, 3433 { VLNK, KF_VTYPE_VLNK }, 3434 { VNON, KF_VTYPE_VNON }, 3435 { VREG, KF_VTYPE_VREG }, 3436 { VSOCK, KF_VTYPE_VSOCK } 3437 }; 3438 unsigned int i; 3439 3440 /* 3441 * Perform vtype translation. 3442 */ 3443 for (i = 0; i < nitems(vtypes_table); i++) 3444 if (vtypes_table[i].vtype == vtype) 3445 return (vtypes_table[i].kf_vtype); 3446 3447 return (KF_VTYPE_UNKNOWN); 3448} 3449 3450static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3451 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3452 "Process filedesc entries"); 3453 3454/* 3455 * Store a process current working directory information to sbuf. 3456 * 3457 * Takes a locked proc as argument, and returns with the proc unlocked. 3458 */ 3459int 3460kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3461{ 3462 struct filedesc *fdp; 3463 struct export_fd_buf *efbuf; 3464 int error; 3465 3466 PROC_LOCK_ASSERT(p, MA_OWNED); 3467 3468 fdp = fdhold(p); 3469 PROC_UNLOCK(p); 3470 if (fdp == NULL) 3471 return (EINVAL); 3472 3473 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3474 efbuf->fdp = fdp; 3475 efbuf->sb = sb; 3476 efbuf->remainder = maxlen; 3477 3478 FILEDESC_SLOCK(fdp); 3479 if (fdp->fd_cdir == NULL) 3480 error = EINVAL; 3481 else { 3482 vref(fdp->fd_cdir); 3483 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, 3484 FREAD, efbuf); 3485 } 3486 FILEDESC_SUNLOCK(fdp); 3487 fddrop(fdp); 3488 free(efbuf, M_TEMP); 3489 return (error); 3490} 3491 3492/* 3493 * Get per-process current working directory. 3494 */ 3495static int 3496sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 3497{ 3498 struct sbuf sb; 3499 struct proc *p; 3500 ssize_t maxlen; 3501 int error, error2, *name; 3502 3503 name = (int *)arg1; 3504 3505 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 3506 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3507 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3508 if (error != 0) { 3509 sbuf_delete(&sb); 3510 return (error); 3511 } 3512 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3513 error = kern_proc_cwd_out(p, &sb, maxlen); 3514 error2 = sbuf_finish(&sb); 3515 sbuf_delete(&sb); 3516 return (error != 0 ? error : error2); 3517} 3518 3519static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 3520 sysctl_kern_proc_cwd, "Process current working directory"); 3521 3522#ifdef DDB 3523/* 3524 * For the purposes of debugging, generate a human-readable string for the 3525 * file type. 3526 */ 3527static const char * 3528file_type_to_name(short type) 3529{ 3530 3531 switch (type) { 3532 case 0: 3533 return ("zero"); 3534 case DTYPE_VNODE: 3535 return ("vnod"); 3536 case DTYPE_SOCKET: 3537 return ("sock"); 3538 case DTYPE_PIPE: 3539 return ("pipe"); 3540 case DTYPE_FIFO: 3541 return ("fifo"); 3542 case DTYPE_KQUEUE: 3543 return ("kque"); 3544 case DTYPE_CRYPTO: 3545 return ("crpt"); 3546 case DTYPE_MQUEUE: 3547 return ("mque"); 3548 case DTYPE_SHM: 3549 return ("shm"); 3550 case DTYPE_SEM: 3551 return ("ksem"); 3552 default: 3553 return ("unkn"); 3554 } 3555} 3556 3557/* 3558 * For the purposes of debugging, identify a process (if any, perhaps one of 3559 * many) that references the passed file in its file descriptor array. Return 3560 * NULL if none. 3561 */ 3562static struct proc * 3563file_to_first_proc(struct file *fp) 3564{ 3565 struct filedesc *fdp; 3566 struct proc *p; 3567 int n; 3568 3569 FOREACH_PROC_IN_SYSTEM(p) { 3570 if (p->p_state == PRS_NEW) 3571 continue; 3572 fdp = p->p_fd; 3573 if (fdp == NULL) 3574 continue; 3575 for (n = 0; n <= fdp->fd_lastfile; n++) { 3576 if (fp == fdp->fd_ofiles[n].fde_file) 3577 return (p); 3578 } 3579 } 3580 return (NULL); 3581} 3582 3583static void 3584db_print_file(struct file *fp, int header) 3585{ 3586 struct proc *p; 3587 3588 if (header) 3589 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3590 "File", "Type", "Data", "Flag", "GCFl", "Count", 3591 "MCount", "Vnode", "FPID", "FCmd"); 3592 p = file_to_first_proc(fp); 3593 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3594 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3595 0, fp->f_count, 0, fp->f_vnode, 3596 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3597} 3598 3599DB_SHOW_COMMAND(file, db_show_file) 3600{ 3601 struct file *fp; 3602 3603 if (!have_addr) { 3604 db_printf("usage: show file <addr>\n"); 3605 return; 3606 } 3607 fp = (struct file *)addr; 3608 db_print_file(fp, 1); 3609} 3610 3611DB_SHOW_COMMAND(files, db_show_files) 3612{ 3613 struct filedesc *fdp; 3614 struct file *fp; 3615 struct proc *p; 3616 int header; 3617 int n; 3618 3619 header = 1; 3620 FOREACH_PROC_IN_SYSTEM(p) { 3621 if (p->p_state == PRS_NEW) 3622 continue; 3623 if ((fdp = p->p_fd) == NULL) 3624 continue; 3625 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3626 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3627 continue; 3628 db_print_file(fp, header); 3629 header = 0; 3630 } 3631 } 3632} 3633#endif 3634 3635SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3636 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3637 3638SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3639 &maxfiles, 0, "Maximum number of files"); 3640 3641SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3642 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3643 3644/* ARGSUSED*/ 3645static void 3646filelistinit(void *dummy) 3647{ 3648 3649 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3650 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3651 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 3652 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 3653 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3654} 3655SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3656 3657/*-------------------------------------------------------------------*/ 3658 3659static int 3660badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3661 int flags, struct thread *td) 3662{ 3663 3664 return (EBADF); 3665} 3666 3667static int 3668badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3669 struct thread *td) 3670{ 3671 3672 return (EINVAL); 3673} 3674 3675static int 3676badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3677 struct thread *td) 3678{ 3679 3680 return (EBADF); 3681} 3682 3683static int 3684badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3685 struct thread *td) 3686{ 3687 3688 return (0); 3689} 3690 3691static int 3692badfo_kqfilter(struct file *fp, struct knote *kn) 3693{ 3694 3695 return (EBADF); 3696} 3697 3698static int 3699badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3700 struct thread *td) 3701{ 3702 3703 return (EBADF); 3704} 3705 3706static int 3707badfo_close(struct file *fp, struct thread *td) 3708{ 3709 3710 return (0); 3711} 3712 3713static int 3714badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3715 struct thread *td) 3716{ 3717 3718 return (EBADF); 3719} 3720 3721static int 3722badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3723 struct thread *td) 3724{ 3725 3726 return (EBADF); 3727} 3728 3729static int 3730badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3731 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3732 int kflags, struct thread *td) 3733{ 3734 3735 return (EBADF); 3736} 3737 3738static int 3739badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 3740{ 3741 3742 return (0); 3743} 3744 3745struct fileops badfileops = { 3746 .fo_read = badfo_readwrite, 3747 .fo_write = badfo_readwrite, 3748 .fo_truncate = badfo_truncate, 3749 .fo_ioctl = badfo_ioctl, 3750 .fo_poll = badfo_poll, 3751 .fo_kqfilter = badfo_kqfilter, 3752 .fo_stat = badfo_stat, 3753 .fo_close = badfo_close, 3754 .fo_chmod = badfo_chmod, 3755 .fo_chown = badfo_chown, 3756 .fo_sendfile = badfo_sendfile, 3757 .fo_fill_kinfo = badfo_fill_kinfo, 3758}; 3759 3760int 3761invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 3762 int flags, struct thread *td) 3763{ 3764 3765 return (EOPNOTSUPP); 3766} 3767 3768int 3769invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3770 struct thread *td) 3771{ 3772 3773 return (EINVAL); 3774} 3775 3776int 3777invfo_ioctl(struct file *fp, u_long com, void *data, 3778 struct ucred *active_cred, struct thread *td) 3779{ 3780 3781 return (ENOTTY); 3782} 3783 3784int 3785invfo_poll(struct file *fp, int events, struct ucred *active_cred, 3786 struct thread *td) 3787{ 3788 3789 return (poll_no_poll(events)); 3790} 3791 3792int 3793invfo_kqfilter(struct file *fp, struct knote *kn) 3794{ 3795 3796 return (EINVAL); 3797} 3798 3799int 3800invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3801 struct thread *td) 3802{ 3803 3804 return (EINVAL); 3805} 3806 3807int 3808invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3809 struct thread *td) 3810{ 3811 3812 return (EINVAL); 3813} 3814 3815int 3816invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3817 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3818 int kflags, struct thread *td) 3819{ 3820 3821 return (EINVAL); 3822} 3823 3824/*-------------------------------------------------------------------*/ 3825 3826/* 3827 * File Descriptor pseudo-device driver (/dev/fd/). 3828 * 3829 * Opening minor device N dup()s the file (if any) connected to file 3830 * descriptor N belonging to the calling process. Note that this driver 3831 * consists of only the ``open()'' routine, because all subsequent 3832 * references to this file will be direct to the other driver. 3833 * 3834 * XXX: we could give this one a cloning event handler if necessary. 3835 */ 3836 3837/* ARGSUSED */ 3838static int 3839fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3840{ 3841 3842 /* 3843 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3844 * the file descriptor being sought for duplication. The error 3845 * return ensures that the vnode for this device will be released 3846 * by vn_open. Open will detect this special error and take the 3847 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3848 * will simply report the error. 3849 */ 3850 td->td_dupfd = dev2unit(dev); 3851 return (ENODEV); 3852} 3853 3854static struct cdevsw fildesc_cdevsw = { 3855 .d_version = D_VERSION, 3856 .d_open = fdopen, 3857 .d_name = "FD", 3858}; 3859 3860static void 3861fildesc_drvinit(void *unused) 3862{ 3863 struct cdev *dev; 3864 3865 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3866 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3867 make_dev_alias(dev, "stdin"); 3868 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3869 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3870 make_dev_alias(dev, "stdout"); 3871 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 3872 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3873 make_dev_alias(dev, "stderr"); 3874} 3875 3876SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3877